102
Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility Lecture I Gunther Uhlmann University of Washington, CMM (Chile), HKUST (Hing Kong) & University of Helsinki Valparaiso, Chile, August 2015

Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

  • Upload
    others

  • View
    12

  • Download
    1

Embed Size (px)

Citation preview

Page 1: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Summer School on Mathematical Physics

Inverse Problems Visibility and Invisibility

Lecture I

Gunther Uhlmann

University of Washington CMM (Chile)

HKUST (Hing Kong) amp University of Helsinki

Valparaiso Chile August 2015

Inverse Boundary Problems

Can one determine the internal properties of a medium by making

measurements outside the medium (non-invasive)

X-ray tomography (CAT-scans)

Problem Can we recover the density from attenuation of X-rays

1

Radon (1917) n = 2

f(x) = Unknown function

Idetector = eminusintL fIsource

Rf(s θ) = g(s θ) =int〈xθ〉=s

f(x)dH =intLf

f(x) =1

4π2pv

intS1dθint d

dsg(s θ)ds

〈x θ〉 minus s

2

LINEAR (No Scattering)

X-ray tomography (CT)

PET MRI

3

NONLINEAR (Scattering)

Ultrasound

Electrical

Impedance

Tomography

(EIT)

4

Hybrid Methods

Superposition of 2 images each obtained with a single wave

One single wave in sensitive only to a given contrast

Ultrasound to bulk compressibility

Photoacoustic

ImagingOptical wave to dielectric permittivity

Thermoacoustic

Imaging

LF Electromagnetic wave to electrical

impedance conductivity

5

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 2: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Inverse Boundary Problems

Can one determine the internal properties of a medium by making

measurements outside the medium (non-invasive)

X-ray tomography (CAT-scans)

Problem Can we recover the density from attenuation of X-rays

1

Radon (1917) n = 2

f(x) = Unknown function

Idetector = eminusintL fIsource

Rf(s θ) = g(s θ) =int〈xθ〉=s

f(x)dH =intLf

f(x) =1

4π2pv

intS1dθint d

dsg(s θ)ds

〈x θ〉 minus s

2

LINEAR (No Scattering)

X-ray tomography (CT)

PET MRI

3

NONLINEAR (Scattering)

Ultrasound

Electrical

Impedance

Tomography

(EIT)

4

Hybrid Methods

Superposition of 2 images each obtained with a single wave

One single wave in sensitive only to a given contrast

Ultrasound to bulk compressibility

Photoacoustic

ImagingOptical wave to dielectric permittivity

Thermoacoustic

Imaging

LF Electromagnetic wave to electrical

impedance conductivity

5

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 3: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Radon (1917) n = 2

f(x) = Unknown function

Idetector = eminusintL fIsource

Rf(s θ) = g(s θ) =int〈xθ〉=s

f(x)dH =intLf

f(x) =1

4π2pv

intS1dθint d

dsg(s θ)ds

〈x θ〉 minus s

2

LINEAR (No Scattering)

X-ray tomography (CT)

PET MRI

3

NONLINEAR (Scattering)

Ultrasound

Electrical

Impedance

Tomography

(EIT)

4

Hybrid Methods

Superposition of 2 images each obtained with a single wave

One single wave in sensitive only to a given contrast

Ultrasound to bulk compressibility

Photoacoustic

ImagingOptical wave to dielectric permittivity

Thermoacoustic

Imaging

LF Electromagnetic wave to electrical

impedance conductivity

5

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 4: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

LINEAR (No Scattering)

X-ray tomography (CT)

PET MRI

3

NONLINEAR (Scattering)

Ultrasound

Electrical

Impedance

Tomography

(EIT)

4

Hybrid Methods

Superposition of 2 images each obtained with a single wave

One single wave in sensitive only to a given contrast

Ultrasound to bulk compressibility

Photoacoustic

ImagingOptical wave to dielectric permittivity

Thermoacoustic

Imaging

LF Electromagnetic wave to electrical

impedance conductivity

5

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 5: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

NONLINEAR (Scattering)

Ultrasound

Electrical

Impedance

Tomography

(EIT)

4

Hybrid Methods

Superposition of 2 images each obtained with a single wave

One single wave in sensitive only to a given contrast

Ultrasound to bulk compressibility

Photoacoustic

ImagingOptical wave to dielectric permittivity

Thermoacoustic

Imaging

LF Electromagnetic wave to electrical

impedance conductivity

5

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 6: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Hybrid Methods

Superposition of 2 images each obtained with a single wave

One single wave in sensitive only to a given contrast

Ultrasound to bulk compressibility

Photoacoustic

ImagingOptical wave to dielectric permittivity

Thermoacoustic

Imaging

LF Electromagnetic wave to electrical

impedance conductivity

5

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 7: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Photoacoustic Tomography

Photoacoustic Effect The sound of light

Picture from Economist

(The sound of light)

Graham Bell When

rapid pulses of light are

incident on a sample of

matter they can be ab-

sorbed and the resulting

energy will then be radi-

ated as heat This heat

causes detectable sound

waves due to pressure

variation in the surround-

ing medium

6

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 8: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Thermoacoustic Tomography

Wikipedia

7

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 9: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

(Loading Melanoma3DMovieavi)

Lihong Wang (Washington U)

8

Melanoma3DMovieavi
Media File (videoavi)

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 10: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Mathematical Model

First Step in PAT and TAT is to reconstruct H(x) from u(x t)|partΩtimes(0T )

where u solves

(part2t minus c2(x)∆)u = 0 on Rn times R+

u|t=0 = βH(x)

parttu|t=0 = 0

Second Step in PAT and TAT is to reconstruct the optical or

electrical properties from H(x) (internal measurements)

9

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 11: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CALDERONrsquoS PROBLEM and EIT

Ω sub Rn

(n = 23)

Can one determine the electrical conductivity of Ω γ(x) by making

voltage and current measurements at the boundary

(Calderon Geophysical prospection)

Early breast cancer detection

Normal breast tissue 03 mhoCancerous breast tumor 20 mho

10

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 12: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

11

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 13: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

REMINISCENCIA DE MI VIDA MATEMATICA

Speech at Universidad Autonoma de Madrid accepting the lsquoDoctor

Honoris Causarsquo

My work at ldquoYacimientos Petroliferos Fiscalesrdquo (YPF) was

very interesting but I was not well treated otherwise I would

have stayed there

12

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 14: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

(Loading imagesrawlongmpg)

Mark Nelson httpnelsonbeckmanillinoisedu

13

rawlongmpg
Media File (videompeg)

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 15: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

(Loading imageselectrompg)

Mark Nelson httpnelsonbeckmanillinoisedu

14

electrompg
Media File (videompeg)

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 16: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

15

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 17: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

16

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 18: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

17

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 19: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Combine with mammography for early detection

RPI Group (D Isaacson)

Sensitivity = predicted to have cancer

Total that have cancertimes 100

Specificity = predicted NOT to have cancer

Total that do NOT have cancertimes 100

Results for Equivocal Mammograms (N = 273)

Mamm T-Scanalone Adjunctive

Sensitivity 60 82(Biopsy ps=50)Specificity 41 57(Biopsy neg=223)

18

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 20: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Other Applications

- Non-destructive testing (corrosion cracks)

- Seepage of groundwater pollutants

- Medical Imaging (EIT)

Tissue Conductivity (mho)Blood 67Liver 28

Cardiac muscle 63 (longitudinal)23 (transversal)

Grey matter 35White matter 15

Lung 10 (expiration)04 (inspiration)

19

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 21: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

ACT3 imaging blood as it leaves the heart (blue) and fills the lungs (red) during

systole

20

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 22: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

(Loading DBarPerfMovie1avi)

Thanks to D Issacson

21

DBarPerfMovie1avi
Media File (videoavi)

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 23: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CALDERONrsquoS PROBLEM (EIT)

Consider a body Ω sub Rn An electrical potential u(x) causes the

current

I(x) = γ(x)nablau

The conductivity γ(x) can be isotropic that is scalar or anisotropic

that is a matrix valued function If the current has no sources or

sinks we have

div(γ(x)nablau) = 0 in Ω

22

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 24: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

γ(x) = conductivity

f = voltage potential at partΩ

Current flux at partΩ = (ν middot γnablau)∣∣∣partΩ

were ν is the unit outer normal

Information is encoded in

map

Λγ(f) = ν middot γnablau∣∣∣partΩ

EIT (Calderonrsquos inverse problem)

Does Λγ determine γ

Λγ = Dirichlet-to-Neumann map

23

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 25: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

div(γ(x)nablau(x)) = 0

u∣∣∣partΩ

= f

Dirichlet Integral

Qγ(f) =int

Ωγ(x)|nablau(x)|2dx

Qγ(f g) =int

Ωγ(x)nablau middot nablavdx

div(γ(x)nablav(x)) = 0

v∣∣∣partΩ

= g

24

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 26: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

div(γnablau) = 0u|partΩ = f

div(γnablav) = 0

v|partΩ = gΛγ(f) = γ

partu

partν

∣∣∣∣partΩ

Qγ(f g) =int

Ωγnablau middot nablavdx

A P Calderon On an inverse boundary value problem in Seminar on Nu-

merical Analysis and its Applications to Continuum Physics RIo de Janeiro 1980

Qγ(f) =int

Ωγ|nablau(x)|2dx =

intpartΩ

Λγ(f)fdS

25

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 27: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Linearization

limεrarr0+

Qγ+εh(f g)minusQγ(f g)

ε=int

Ωhnablau middot nablavdx

Case γ = 1 div(γnablau) = ∆u = 0

Linearized Problem Suppose we knowintΩhnablau middot nablavdx forall∆u = ∆v = 0

Can we recover h

26

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 28: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Linearized problem at γ = 1intΩhnablau middot nablavdx data forall∆u = ∆v = 0

Can we recover h

u = exmiddotρ

v = eminusxmiddotρ ρ isin Cn ρ middot ρ = 0

ρ =η minus iξ

2 ρ middot ρ = 0 hArr |η| = |ξ| η middot ξ = 0

|ξ|2int

Ωheminusixmiddotξdx known

we can recover χΩh(ξ) therefore h on Ω

27

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 29: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem (Kohn-Vogelius 1984)

Assume γ isin Cinfin(Ω) From Λγ we can determine partαγ∣∣∣partΩ

forallα

Proof (Sylvester-U Lee-U)

Λγ is a pseudodifferential operator of order 1 (Calderon)

partΩ = xn = 0 locally

Coordinates x = (xprime xn) xprime isin Rnminus1

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

28

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 30: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Λγf(xprime) =inteixprimemiddotξprimeλγ(xprime ξprime)f(ξprime)dξprime

λγ(xprime ξprime) = γ(0 xprime)|ξprime|+ a0(xprime ξprime) + middot middot middot+ aj(xprime ξprime) + middot middot middot

with aj(xprime ξprime) pos homogeneous of degree minusj in ξprime

aj(xprime λξprime) = λminusjaj(x

prime ξprime) λ gt 0

Result From aj we can determine partjγpartνj

∣∣∣∣xn=0

29

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 31: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem n ge 3 (Sylvester-U 1987)

γ isin C2(Ω) 0 lt C1 le γ(x) le C2 on Ω

Λγ1 = Λγ2 rArr γ1 = γ2

bull Extended to γ isin C32(Ω) (Paivarinta-Panchenko-U Brown-Torres

2003)

bull γ isin C1+ε(Ω) γ conormal (Greenleaf-Lassas-U 2003)

bull γ isin C1(Ω) (Haberman-Tataru 2012)

Complex-Geometrical Optics Solutions (CGO)

bull Reconstruction A Nachman (1988)

bull Stability G Alessandrini (1988)

bull Numerical Methods (D Issacson J Muller S Siltanen)

30

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 32: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Reduction to Schrodinger equation

div(γnablaw) = 0

u =radicγw

Then the equation is transformed into

(∆minus q)u = 0 q =∆radicγ

radicγ

(∆ =

nsumi=1

part2

partx2i

)

(∆minus q)u = 0

u∣∣∣partΩ

= f

Define Λq(f) =partu

partν

∣∣∣partΩ

ν = unit-outer normal to partΩ

31

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 33: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

IDENTITY

intΩ

(q1 minus q2)u1u2 =intpartΩ

((Λq1 minus Λq2)u1

∣∣∣partΩ

)u2

∣∣∣partΩdS

(∆minus qi)ui = 0

If Λγ1 = Λγ2 rArr Λq1 = Λq2 andintΩ

(q1 minus q2)u1u2 = 0

GOAL Find MANY solutions of (∆minus qi)ui = 0

32

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 34: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CGO SOLUTIONS

Calderon Let ρ isin Cn ρ middot ρ = 0

ρ = η + ik η k isin Rn |η| = |k| η middot k = 0

u = exmiddotρ = exmiddotηeixmiddotk

∆u = 0 u =

exponentially decreasing x middot η lt 0

oscillating x middot η = 0

exponentially increasing x middot η gt 0

33

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 35: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

COMPLEX GEOMETRICAL OPTICS

(Sylvester-U) n ge 2 q isin Linfin(Ω)

Let ρ isin Cn (ρ = η + ik η k isin Rn) such that ρ middot ρ = 0

(|η| = |k| η middot k = 0)

Then for |ρ| sufficiently large we can find solutions of

(∆minus q)wρ = 0 on Ω

of the form

wρ = exmiddotρ(1 + Ψq(x ρ))

with Ψq rarr 0 in Ω as |ρ| rarr infin

34

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 36: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Proof Λq1 = Λq2 rArr q1 = q2intΩ

(q1 minus q2)u1u2 = 0

u1 = exmiddotρ1(1 + Ψq1(x ρ1)) u2 = exmiddotρ2(1 + Ψq2(x ρ2))

ρ1 middot ρ1 = ρ2 middot ρ2 = 0 ρ1 = η + i(k + l)ρ2 = minusη + i(k minus l)

η middot k = η middot l = l middot k = 0 |η|2 = |k|2 + |l|2

intΩ

(q1 minus q2)e2ixmiddotk(1 + Ψq1 + Ψq2 + Ψq1Ψq2) = 0

Letting |l| rarr infinint

Ω(q1 minus q2)e2ixmiddotk = 0 forallk =rArr q1 = q2

35

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 37: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

APPLICATIONS

n ge 3 (∆minus q) = 0 Λq determines q

bull EIT Λγ determines γ

bull Optical Tomography (Diffusion Approximation)

iωU minusnabla middotD(x)nablaU + σa(x)U = 0 in Ω

U= Density of photons D=Diffusion Coefficient σa(x)= optical

absorption

RESULT bull If ω 6= 0 we can recover both D(x) and σa(x)bull If ω = 0 we can recover either D(x) or σa(x)

36

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 38: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

OTHER APPLICATIONS (Fixed energy)

bull Optics (∆ minus k2n(x))u = 0 n(x) isotropic index of refraction

(q(x) = k2n(x))

bull Acoustic div( 1ρ(x)nablap) + ω2κ(x)p = 0 ρ density κ compress-

ibility (need two frequencies ω)

bull Inverse quantum scattering at fixed energy (∆minus qminusλ2)u = 0

q potential

bull Maxwellrsquos Equation (Isotropic)

(Ola-Somersalo) Reduction to (∆minusQ) Q an 8times 8 matrix

bull Quantitative Photoacoustic Tomography

(Bal-U)

37

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 39: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

PARTIAL DATA PROBLEM

Suppose we measure

Λγ(f)|Γ suppf sube Γprime

Γ Γprime open subsets of partΩ

Can one recover γ

Important case Γ = Γprime

38

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 40: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

EXTENSION OF CGO SOLUTIONS

u = exmiddotρ(1 + Ψq(x ρ))

ρ isin Cn ρ middot ρ = 0

(Not helpful for localizing)

Kenig-Sjostrand-U (2007)

u = eτ(ϕ(x)+iψ(x))(a(x) +R(x τ))

τ isin R ϕ ψ real-valued R(x τ)rarr 0 as τ rarrinfin

ϕ limiting Carleman weight

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

Example ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

39

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 41: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CGO SOLUTIONS

u = eτ(ϕ(x)+iψ(x))(a0(x) +R(x τ))

R(x τ)τrarrinfinminusrarr 0 in Ω

ϕ(x) = ln |xminus x0|

Complex Spherical Waves

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

40

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 42: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Earlier result (Bukhgeim-U 2002)

n ge 3 Let ξ isin Snminus1 = x isin Rn |x| = 1 We define

partΩplusmn =

x isin partΩ 〈ν ξ〉 gt 0

lt 0

(ν is the unit outer normal) Let δ gt 0

partΩ+δ(ξ) = x isin partΩ 〈ν ξ〉 gt δpartΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Theorem (Bukhgeim-U) Suppose we know

Λq(f)|partΩminusδ(ξ)supp f sube partΩ

Then we can recover q

41

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 43: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Carleman estimate ξ isin Snminus1

Let q isin Linfin(Ω) u isin C2(Ω) u|partΩ = 0 For τ ge τ0

τ2int

Ω|eminusτ〈xξ〉u|2 dx+ τ

intpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x)

leC

intΩ|eminusτ〈xξ〉(∆minus q)u|2dxminus τ

intpartΩminus〈ξ ν(x)〉|eminusτ〈xξ〉

partu

partν|2dS

42

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 44: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Remarks ∆ρu = eminusxmiddotρ∆(exmiddotρu)

bull Carleman estimate for domains with boundary for

∆ρ = ∆ + 2ρ middot nabla

bull Weight is linear 〈x ξ〉

Corollary u = 0 on partΩ partupartν |partΩminus = 0

τintpartΩ+

〈x ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|eminusτ〈xξ〉(∆minus q)u|2dx

We need 〈ν(x) ξ〉 ge δ gt 0

43

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 45: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Bukhgeim-U (n ge 3)

Λq1(f)|partΩminusδ(ξ)= Λq2(f)|partΩminusδ(ξ)

forallf =rArr q1 = q2

partΩminusδ(ξ) = x isin partΩ 〈ν ξ〉 lt δ

Sketch of proof

Choose u2 = exmiddotρ2(1 + Ψq2(x ρ2))

solution of (∆minus q2)u2 = 0

ρ2 = τξ + i(k + l)

|k|2 + |l|2 = τ2

Let u1 be such that u1|partΩ = u2|partΩpartu1

partν|partΩminusδ(ξ)

=partu2

partν|partΩminusδ(ξ)

44

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 46: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

u2 = exmiddotρ2(1 + Ψq2(x ρ2))

(∆minus q2)u2 = 0 ρ2 = τξ + i(k + l)

u1|partΩ = u2|partΩpartu1

partν1|partΩminusδ(ξ)

=partu2

partν1|partΩminusδ(ξ)

u = u1 minus u2 q = q1 minus q2

v1 = exmiddotρ1(1 + Ψq1(x ρ1)) ρ1 = minusτξ + i(k minus l)

solution of (∆minus q1)v1 = 0

intΩqu2v1dx =

intpartΩ

partu

partνv1dS

Note that u|partΩ = 0 partupartν |partΩminusδ(ξ)

= 0

45

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 47: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

q = q1 minus q2

(lowast)int

Ωqu2v1dx =

intpartΩ+δ(ξ)

partu

partνv1dS

u2 = exmiddotρ1(1 + Ψq1)v1 = exmiddotρ2(1 + Ψq2)

Fix k isin Rn ρ1 + ρ2 = 2ik

Carleman estimate

τintpartΩ+

〈ξ ν(x)〉|eminusτ〈xξ〉partu

partν|2dS(x) le C

intΩ|(∆minus q1)ueminusτ〈xξ〉|2dS(x)

Need 〈x ξ〉 ge δ gt 0 |RHS| le C as |ρ| rarr infin LHSrarrint

Ωqe2ixmiddotkdx

46

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 48: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

We getint

Ωe2ixmiddotkq(x)dx = 0

k perp ξ But we can move ξ a

little bit

ξ isin Snminus1

partΩminusδ(ξ) = x isin partΩ 〈ν(x) ξ〉 lt δ

We obtain χΩq(minus2k) = 0 in an open cone =rArr q = 0

Carleman estimate =rArr control of partupartν

∣∣∣partΩ+δ

(with appropriate linear

weights) (Stability estimates Heck-Wang)

47

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 49: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem (Kenig-Sjostrand-U) Ω strictly convex

Λq1

∣∣∣Γ

= Λq2

∣∣∣Γ Γ sube partΩ Γ arbitrary

rArr q1 = q2

uτ = eτ(ϕ+iψ)aτ ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Eikonal nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|ψ(x) = d( xminusx0

|xminusx0| ω) ω isin Snminus1 smooth

for x isin Ω

Transport (nablaϕ+ inablaψ) middot nablaaτ = 0

(Cauchy-Riemann equation in plane generated by nablaϕnablaψ)

48

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 50: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

ϕ(x) = ln |xminus x0| x0 isinch(Ω)

Carleman Estimates

u|partΩ = partupartν |partΩminus = 0 partΩplusmn = x isin partΩnablaϕ middot ν

gtlt 0

intpartΩ+

lt nablaϕ ν gt |eminusτϕ(x)partu

partν|2ds le

C

τ

intΩ|(∆minus q)ueminusτϕ(x)|2ds

This gives control of partupartν |partΩ+δ

partΩ+δ = x isin partΩnablaϕ middot ν ge δ

49

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 51: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

More general CGO solutions

uτ = eτ(ϕ+iψ)aτ

τ 0 τ = 1h (semicl) ϕψ real-valued

bull ϕ is a limiting Carleman weight

eϕhh2(minus∆ + q)eminus

ϕh

has semiclassical principal symbol

Pϕ(x ξ) = ξ2 minus (nablaϕ)2 + 2inablaϕ middot ξ

Hormanderrsquos condition

Re Pϕ Im Pϕ le 0 on Pϕ = 0

We need ϕminusϕ to be phase of solutions

LCW Re Pϕ Im Pϕ = 0

nablaϕ 6= 0 in an open neighborhood of Ω

50

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 52: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CGO solutions uh = e1h(ϕ+iψ)ah

bull ϕ LCW ϕ real-valued

RePϕ ImPϕ = 0 on Pϕ = 0

nablaϕ 6= 0 on an open neighborhood of Ω

Examples (Dos Santos Ferreira-Kenig-Salo-U 2009)

(a) ϕ(x) = x middot ξ ξ isin Rn |ξ| = 1

(b) ϕ(x) = a ln |xminus x0|+ b (a b constants) x0 isinch(Ω)

(c) ϕ(x) =a〈xminus x0 ξ〉|xminus x0|2

+ b ξ isin Rn

(d) ϕ(x) = aarctan 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(e) ϕ(x) = aarctanh 2〈xminusx0ξ〉|xminusx0|2minus|ξ|2

+ b

(f) n = 2 ϕ is a harmonic function

51

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 53: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Instead of intΩe2ixmiddotkq(x)dx = 0

k perp ξ (ξ isin Snminus1) as in Bukhgeim-U argument we getintΩeiλf(x) q(x) a1a2dx = 0

λ any real number a1 a2 6= 0 f(x) real-analytic a1 a2 real analytic

Analytic microlocal analysis =rArr q = 0 (like inversion of real-analytic Radon transform)

52

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 54: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Linearization (Analog of Calderon)

Theorem (Dos Santos Ferreira Kenig Sjostrand-U)

intΩhuv = 0

Γ sube partΩ Γ open

∆u = ∆v = 0 u v isin Cinfin(Ω)

supp u|partΩ supp v|partΩ sube Γ

rArr h = 0

53

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 55: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Complex Spherical Waves

uτ = eτ(ϕ+iψ)aτ

ϕ(x) = ln |xminus x0| x0 isin ch(Ω)

Also used to determine inclusions obstacles etc

a) Conductivity Ide-Isozaki-Nakata-Siltanen-U

b) Helmholtz Nakamura-Yosida

c) Elasticity J-N Wang-U

d) 2D Systems J-N Wang-U

e) Maxwell T Zhou

54

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 56: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Complex Spherical Waves

(Loading reconperfect1mpg)

55

reconperfect1mpg
Media File (videompeg)

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 57: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

The Two Dimensional Case

Theorem (n = 2) Let γj isin C2(Ω) j = 12

Assume Λγ1 = Λγ2 Then γ1 = γ2

bull Nachman (1996)

bull Brown-U (1997) Improved to γj Lipschitz

bull Astala-Paivarinta (2006) Improved to γj isin Linfin(Ω)

Recall

div(γnablau) = 0 γ isin Linfin(Ω)u|partΩ = f

Qγ(f) =int

Ωγ|nablau|2dx = 〈Λγf f〉L2(partΩ)

56

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 58: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

This follows from more general result

Theorem (n = 2 Bukhgeim 2008) Let qj isin Linfin(Ω) j = 12

Assume Λq1 = Λq2 Then q1 = q2

Recall

(∆minus q)u = 0u|partΩ = f

Λq(f) =partu

partν

∣∣∣∣∣partΩ

with ν-unit outer normal

57

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 59: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Λq1 = Λq2 rArr q1 = q2

Sketch of proof New class of CGO solutions

u1(z τ) = eτz2 (

1 + r1(z τ))

u2(z τ) = eminusτ z2 (

1 + r2(z τ)) τ 1

solve (∆minus qj)uj = 0 with rj(z τ)rarr 0 on Ω sufficiently fast

Notation z = x1 + ix2

Remark z2 = x21 minus x

22 + 2ix1x2 = ϕ+ iψ

nablaϕ middot nablaψ = 0 |nablaϕ| = |nablaψ|

ϕ harmonic ψ conjugate harmonic

59

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 60: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Λq1 = Λq2 rArrint

Ω(q1 minus q2)u1u2dx = 0

(∆minus qj)uj = 0

u1 = eτz2 (

1 + r1(z τ)) u2 = eminusτ z

2 (1 + r2(z τ)

)Substitutingint

Ω(q1 minus q2)e4iτx1x2(1 + r1 + r2 + r1r2)dx = 0

Letting τ rarrinfin and using stationary phase

(q1 minus q2)(0) = 0

Changing z to z minus z0 we get

(q1 minus q2)(z0) = 0

60

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 61: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Partial data

Let Γ sube partΩ Γ open

Let qj isin C1+ε(Ω) ε gt 0 j = 12

Theorem (Imanuvilov-U-Yamamoto 2010) n=2 Assume

Λq1(f)∣∣∣Γ

= Λq2(f)∣∣∣Γ

forall f suppf sube Γ Then

q1 = q2

bull Riemann Surfaces Guillarmou-Tzou (2011)

61

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 62: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Partial data Γ0 = partΩminus Γ

Construct CGO solutions

∆uj minus qjuj = 0 in Ωuj|Γ0

= 0

In this case intΩ

(q1 minus q2)u1u2dx = 0

if Λq1(f)|Γ = Λq2(f)|Γ suppf sube Γ

62

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 63: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

intΩ

(q1 minus q2)u1u2 = 0

uj|Γ0= uj|partΩminusΓ = 0

u1(x) = eτΦ(z)(a(z) +a0(z)

τ) + eτΦ(z)(a(z) +

a1(z)

τ) + eτϕR

(1)τ

u2(x) = eminusτΦ(z)(a(z) +b0(z)

τ) + eminusτΦ(z)(a(z) +

b1(z)

τ) + eτϕR

(2)τ

Φ = ϕ+ iψ holomorphic

63

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 64: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

u1 = Re eτΦ(z)(a(z) + middot middot middot ) u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

Φ(z) = ϕ+ iψ holomorphic

uj|partΩminusΓ = 0

p isin Ω Φ has non-degenerate critical point at p (Morse function)

parta = 0 Re a|partΩminusΓ = 0

a = 0 at other critical points

Stationary phase in intΩ

(q1 minus q2)u1u2 = 0

64

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 65: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

u1 = Re eτΦ(z)(a(z) + middot middot middot )u2 = Re eminusτΦ(z)(a(z) + middot middot middot )

uj|partΩminusΓ = 0

Φ(z) Morse function with non-degenerate critical point at pintΩ

(q1 minus q2)u1u2 = 0

Stationary phase

=rArr (q1 minus q2)(p) = 0

65

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 66: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Corollary Obstacle Problem

Ω D sub R2 smooth boundary

such that D sub Ω

V sub partΩ open set

Let qj isin C2+α(ΩD) for some α gt 0 j = 12

Cqj =

(u|V partνu|V ) (∆minus qj)u = 0 in ΩD

supp u|partΩ sub V u isin H1(ΩD)

Then Cq1 = Cq2 =rArr q1 = q2

66

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 67: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Carleman Estimate With Degenerate Weights

Lemma 1

Let partΩminus Γ = x isin partΩ ν middot nablaϕ = 0 Then for τ sufficiently large existsolution of

∆uminus qu = f in Ωu|partΩminusΓ = g

such that

ueminusτϕL2(Ω) le C(|τ |minus12feminusτϕL2(Ω) + geminusτϕL2(ΩminusΓ)

)

67

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 68: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Phase Function

Lemma 2 (Vekua) Given points x1 xN in Ω and constants bi i =

12 C0j C1j C2j j = 1 N there exists an open and dense set

Θ sube C2(partΩminus Γ)times C2(partΩminus Γ)times C3N

solution of

φ+ iψ holomorphic in Ω

(φ ψ)|partΩminusΓ = (b1 b2)(φ+ iψ)(xj) = C0jpartpartz(φ+ iψ)(xj) = C1jpart2

partz2(φ+ iψ)(xj) = C2j

68

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 69: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CGO Solutions (n = 2)

partminus1z g(z) = minus

1

π

intΩ

g(ξ1 ξ2)

ξ1 + iξ2 minus zdξ1dξ2 partminus1

z g = partminus1z g

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

RΦτg = eτ(Φ(z)minusΦ(z))partminus1z (geτ(Φ(z)minusΦ(z)))

H = non-degenerate critical points of Φ

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 70: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

CGO Solutions (n = 2)

∆u1 minus q1u1 = 0 in Ω

u1

∣∣∣partΩΓ

= 0

Let Φ be a holomorphic Morse function such that ImΦ = 0 on

partΩΓ Let Φ = ϕ+ iψ

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

69

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 71: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

u1(x) = eτΦ(z)(a(z) + a0(z)τ) + eτΦz(a(z) + a0(z)τ)

+ eτϕu11 + eτϕu12

Choice of a a0 a1

a a0 a1 isin C2(Ω) partza = partza0 = partza1 equiv 0

Re a∣∣∣partΩΓ

= 0 a = partza = 0 on Hcap partΩ

(a0(z) + a1(z))∣∣∣partΩΓ

=partminus1z (aq1)minusM1(z)

4partzΦ

+partminus1z (a(z)q1)minusM3(z)

4partzΦ

70

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 72: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

The polynomials M1(z) and M3(z) satisfy

partjz(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

partjz

(partminus1z (aq1)minusM3(z)

)= 0 z isin H j = 012

Let ei i = 12 be smooth e1 + e2 = 1 on Ω with e1 = 0 in a

neighborhood of Hminus partΩ and e2 = 1 in a neighborhood of partΩ

71

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 73: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Remainder term

Choice of u11

u11 = minus1

4eiτψRΦτ

(e1

(partminus1z (aq1)minusM1(z)

))minus

1

4eminusiτψRΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

minuseiτψ

τ

e2

(partminus1z (aq1)minusM1(z)

)4partzΦ

minuseiτψ

τ

e2

(partminus1z (a(z)q1)minusM3(z)

)4partzΦ

72

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 74: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Other remainder term

Find u12 such that

∆(u12eτϕ)minus q1u12e

τϕ minus q1u11eτϕ + h1e

τϕ in Ω

u12

∣∣∣partΩΓ

=1

4RΦτ

(e1

(partminus1z (a(z)q1)minusM1(z)

))+

1

4RΦminusτ

(e1

(partminus1z (a(z))q1 minusM3(z)

))

u12L2(Ω) = o

(1

τ

) τ rarrinfin

73

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 75: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Here

h1 = eiτψ∆

e2

(partminus1z (a(z)q1)minusM1(z)

)4τpartzΦ

+eminusiτψ∆

e2

(partminus1z (a(z)q1)minusM3(z)

)4τpartzΦ

minusa0q1

τeiτψ minus

a1q1

τeminusiτψ

74

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 76: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Similarly

∆v minus q2v = 0 in Ω v∣∣∣partΩΓ

= 0

Construct solution v of the form

v(x) = eminusτΦ(z)(a(z) + b0(z)τ

)+ eminusτΦ(z)

(a(z) + b0(z)τ

)+eminusτϕv11 + eminusτϕv12

75

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 77: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Main Term

R =int

Ω(q1 minus q2)(a(a0 + b0) + a(a1 + b1))dx

+1

4

intΩ

(q1 minus q2)

a partminus1z (aq2)minusM2(z)

partzΦ+ a

partminus1z (aq2)minusM4(z)

partzΦ

dx+

1

4

intΩ

(q1 minus q2)

a partminus1z (aq1)minusM1(z)

partzΦ+ a

partminus1z (aq1)minusM3(z)

partzΦ

dx

76

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 78: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Proof of Uniqueness for Partial Data

bull Take geometiric optics solution u1 to

∆u1 minus q1u1 = 0 u1

∣∣∣partΩΓ

= 0

bull u2 ∆u2 minus q2u2 = 0 u2

∣∣∣partΩ

= u1

∣∣∣partΩ

DN maps are equal rArr nablau2 = nablau1 on Γ

u = u1 minus u2 rArr ∆uminus q2u = (q1 minus q2)u2

u∣∣∣partΩ

= 0partu

partν

∣∣∣∣∣Γ

= 0

bull Take complex geometric optics solution v to

∆v minus q2v = 0 v∣∣∣partΩΓ

= 0

77

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 79: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

0 =int

Ωv(∆uminus q2u)dx = minus

intΩ

(q1 minus q2)vu1dx

Stationary phase + estimates for u12 rArr

2lsum

k=1

π((q1 minus q2)|a|2)(xk)Re ei2τ Im Φ(xk)

|(det Im Φprimeprime)(xk)|12

+R = o(1)

as τ rarrinfin

[left side] = almost perodic function in τ

Bohrrsquos theorem inplies [left side] = 0 for all τ

78

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 80: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Phase function

We can choose Φ such that

Im Φ(xk) 6= Im Φ(xj) j 6= k

Let a(xk) 6= 0 Then stationary phase implies

q1(xk) = q2(xk)

79

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 81: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Partial Data for Second Order Elliptic Equations (n = 2)

(ImanuvilovndashUndashYamamoto 2011)

∆g +A(z)part

partz+B(z)

part

partz+ q z = x1 + ix2

g = (gij) positive definite symmetric matrix

∆gu =1radic

det(g)

nsumij=1

part

partxi(radic

det(g)gijpartu

partxj) gij = (gij)

minus1

Includes

bull Anisotropic Calderonrsquos Problem

bull Magnetic Schrodinger Equation

bull Convection terms

80

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 82: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Anisotropic case

Cardiac muscle 63 mho (longitudinal)23 mho (transversal)

γ = (γij)

conductivity

positive-definite symmetric

matrix

Ω sube RnΩ bounded Under assumptions of no sources or sinks of

current the potential u satisfies

div(γnablau) = 0

nsumij=1

part

partxi

(γij

partu

partxi

)=0 in Ω

u∣∣partΩ

=f

()

f = voltage potential at boundary

Isotropic γij(x) = α(x)δij δij =

1 i = j

0 i 6= j

81

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 83: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

ν =(ν1 middot middot middot νn

)is the unit outer normal to partΩ

Λγ(f) is the induced current flux at partΩ

Λγ is the voltage to current map or Dirichlet - to - Neumann map

82

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 84: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

nsumij=1

part

partxi

γij partupartxj

=0 in Ω

u∣∣∣partΩ

=f

()

Λγ(f) =nsum

ij=1

νiγijpartu

partxj

∣∣∣∣∣∣partΩ

EIT Can we recover γ in Ω from Λγ

83

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 85: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

div(γnablau) = 0

u∣∣∣partΩ

= fΛγ(f) =

nsumij=1

γijνipartu

partxj

∣∣∣partΩ

Λγ rArr γ

Answer No Λψlowastγ = Λγ

where ψ Ωrarr Ω change of variables

ψ|partΩ = Identity

ψlowastγ =

(Dψ)T γ Dψ|detDψ|

ψminus1

v = u ψminus1

84

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 86: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem (ImanuvilovndashUndashYamamoto 2011) Ω sub R2 Γ sub partΩ Γ

open γk = (γijk ) isin Cinfin(Ω) k = 12 positive definite symmetric

Assume

Λγ1(f)|Γ = Λγ2(f)|Γ forallf suppf sub Γ

Then existF Ωrarr Ω Cinfin diffeomorphism F |Γ = Identity such that

Flowastγ1 = γ2

Full Data (Γ = partΩ)

bull γk isin C2(Ω) Nachman (1996)

bull γk Lipschitz SunndashU (2001)

bull γk isin Linfin(Ω) AstalandashLassasndashPavarinta (2006)

85

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 87: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

DIRICHLET-TO-NEUMANN MAP (Lee-U 1989)

(M g) compact Riemannian manifold with boundary∆g Laplace-Beltrami operator g = (gij) pos def symmetric matrix

∆gu =1

radicdet g

nsumij=1

part

partxi

radicdet g gijpartu

partxj

(gij) = (gij)minus1

∆gu = 0 on M

u∣∣∣partM

= f

Conductivity

γij =radic

det g gij

Λg(f) =nsum

ij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

ν = (ν1 middot middot middot νn) unit-outer normal

86

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 88: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

∆gu = 0

u∣∣∣partM

= f

Λg(f) =partu

partνg=

nsumij=1

νjgijpartu

partxi

radicdet g

∣∣∣∣∣partM

current flux at partM

Inverse-problem (EIT)

Can we recover g from Λg

Λg = Dirichlet-to-Neumann map or voltage to current map

87

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 89: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

ANOTHER MOTIVATION (STRING THEORY)

HOLOGRAPHY

Dirichlet-to-Neumann map is the ldquoboundary-2pt functionrdquo

Inverse problem Can we recover (M g) (bulk) from boundary-2pt function

M Parrati and R Rabadan Boundary rigidity and holography JHEP

0401 (2004) 034

88

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 90: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

∆gu = 0

u∣∣∣partM

= fΛg(f) =

partu

partνg

∣∣∣∣∣partM

Λg rArr g

Answer No Λψlowastg = Λg where

ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity and

ψlowastg =(Dψ g (Dψ)T

) ψ

89

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 91: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Show Λψlowastg = Λg ψ M rarrM diffeomorphism ψ∣∣∣partM

= Identity

Qg(f) =sumijintM gij partupartxi

partupartxj

radicdet gdx

Qg(f) = minusintpartM

Λg(f)fdS

Qg hArr Λgv = u ψ ∆ψlowastgv = 0

Qψlowastg = Qg rArr Λψlowastg = Λg

90

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 92: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem (n ge 3) (Lassas-U 2001 Lassas-Taylor-U 2003) (M gi) i =

12 real-analytic connected compact Riemannian manifolds with

boundary Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity so that

g1 = ψlowastg2

In fact one can determine topology of M as well (only need to know

Λg partM)

91

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 93: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem (Guillarmou-Sa Barreto 2009) (M gi) i = 12 are com-

pact Riemannian manifolds with boundary that are Einstein As-

sume

Λg1 = Λg2

Then existψ M rarrM diffeomorphism ψ|partM = Identity such that

g1 = ψlowastg2

Note Einstein manifolds with boundary are real analytic in the

interior

92

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 94: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Theorem (n = 2)(Lassas-U 2001)

(M gi) i = 12 connected Riemannian manifold with boundary

Let Γ sube partM Γ open Assume

Λg1(f)|Γ = Λg2(f)|Γ forallf f supported in Γ

Then existψ M rarrM diffeomorphism ψ∣∣∣Γ

= Identity and

β gt 0 β∣∣∣Γ

= 1 so that

g1 = βψlowastg2

In fact one can determine topology of M as well

93

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 95: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Moding Out the Diffeomorphism Group

Some conformal class Λβg = Λg β isin Cinfin(M)

=rArr β = 1

More general problem

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

Inverse Problem Does Λg determines q

94

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 96: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

(∆g minus q)u = 0 Λg(f) = partupartνg|partM Λg rarr q

Theorem (n=2) (Guillarmou-Tzou 2009)

YES

Earlier results

bull R2 q small (Sylvester-U 1986)

bull R2 q generic (Sun-U 2001)

bull R2 q = ∆radicλradicλ γ gt 0 (Nachmann 1996)

bull Riemannian surfaces q = ∆radicλradicλ γ gt 0 (Henkin-

Michel 2008)

bull q isin Linfin (Bukhgeim 2008)

95

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 97: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

MODING OUT GROUP OF DIFFEOMORPHISM

(n ge 3)

(∆g minus q)u = 0 q isin Cinfin(M)u|partM = f

Λg(f) = partupartνg|partM

(lowast) g(x1 xprime) = c(x)

(1 00 g0(xprime)

) c gt 0

Theorem (Dos Santos-Kenig-Salo-U) Assume that there is a global

coordinate system so that () is true In addition g0 is simple

Then Λg determines uniquely q

Simple No conjugate points and strictly convex

96

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 98: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

g(x1 xprime) = c(x)

(1 00 g0(xprime)

) xprime isin Rnminus1

Examples

(a) g(x) conformal to Euclidean metric (Sylvester-U 1987)

(b) g(x) conformal to hyperbolic metric (Isozaki 2004)

(c) g(x) conformal to metric on sphere (minus a point)

97

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 99: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Non-uniqueness for EIT (Invisibility)

Motivation (Greenleaf-Lassas-U MRL 2003)

When bridge connecting the two parts

of the manifold gets narrower the

boundary measurements give less infor-

mation about isolated area

When we realize the manifold in Euclidean space we should obtain

conductivities whose boundary measurements give no information

about certain parts of the domain

98

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 100: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Greenleaf-Lassas-U (2003 MRL)

Let Ω = B(02) sub R3D = B(01)

where B(0 r) = x isin R3 |x| lt r

F Ω 0 rarr Ω D

F (x) = (|x|2

+ 1)x

|x|

F diffeomorphism F∣∣∣partΩ

= Identity

99

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 101: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Letg = identity metric in B(02)g = (Fminus1)lowastg on B(02) B(01)σ = conductivity associated to g

In spherical coordinates (r φ θ)rarr (r sin θ cosφ r sin θ sinφ r cos θ)

σ =

2(r minus 1)2 sin θ 0 00 2 sin θ 00 0 2(sin θ)minus1

Let g be the metric in B(02) (positive definite in B(01)) st g = g

in B(02) B(01) Then

Theorem (Greenleaf-Lassas-U 2003)

Λg = Λg

100

Based on work of Greenleaf-Lassas-U MRL 2003

101

Page 102: Summer School on Mathematical Physics Inverse Problems ...smp2015/Gunther_Uhlmann_lecture_1.pdf · Summer School on Mathematical Physics Inverse Problems: Visibility and Invisibility

Based on work of Greenleaf-Lassas-U MRL 2003

101