28
Computing Humbert Surfaces David Gruenewald [email protected] The University of Sydney AustMS Conference 2007

[email protected] The University of [email protected] The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Computing Humbert Surfaces

David [email protected]

The University of Sydney

AustMS Conference 2007

Page 2: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Outline

Humbert surfacesDefinitionsAlgebraic models

Computing Humbert surfacesDegree formulaPower seriesLinear algebraRuntime analysisExample

Humbert intersectionsQuaternion ordersComputing Shimura curvesExample

Page 3: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

The Siegel upper half plane

DefinitionThe Siegel upper half plane of degree g is

Hg = {τ ∈ Matg×g(C) | tτ = τ , Im (τ) > 0} .

I Each τ ∈ Hg corresponds to a PPAV Aτ/C with periodmatrix (τ Ig) ∈ Matg×2g(C).

I Aτ∼= Aτ ′ ⇔ ∃M =

(a bc d

)∈ Sp2g(Z) such that

τ ′ = M · τ := (aτ + b)(cτ + d)−1.

I Ag = Sp2g(Z)\Hg is a moduli space for dimension g PPAV’s.

I dimAg = 12g(g + 1). In particular, dimA2 = 3 and A2 is

called the Siegel modular threefold.

Page 4: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Extra endomorphisms

Let A be a PPAS (g = 2). Then End(A) is an order inEnd(A)⊗Q which isomorphic to one of the following algebras:

(0) quartic CM field

(1) indefinite quaternion algebra over Q(2) real quadratic field

(3) Q

The irreducible components of the corresponding moduli spaces inA2 which have “extra endomorphisms” are known as

(0) CM points

(1) Shimura curves

(2) Humbert surfaces

Page 5: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Humbert’s equation

Humbert showed that any

(τ1 τ2

τ2 τ3

)∈ A2 satisfying the equation

kτ1 + `τ2 − τ3 = 0

defines a Humbert surface H∆ of discriminant ∆ = 4k + ` > 0.

Example

H1 = Sp4(Z)\{(

τ1 τ3

τ3 τ3

)}= Sp4(Z)\

{(τ1 00 τ3

)}, the set of

abelian varieties which split as a product of elliptic curves.

Task: Find “useful” algebraic models for H∆.

Page 6: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Algebraic models

I Torelli says that the map C 7→ Jac(C) defines a birationalmap between M2, the moduli space of genus 2 curves and A2

(In fact M2∼= A2 −H1).

I The function field of A2 (and hence M2) is C(j1, j2, j3)where ji are the absolute Igusa invariants.

I There exists an irreducible polynomial H∆(j1, j2, j3) whosezero set is the Humbert surface of discriminant ∆.

Unfortunately, working with ji is impractical (enormous degrees,giant coefficients).Solution: add some level structure.

Page 7: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Algebraic modelsRunge’s model

Runge uses level Γ∗(2, 4) structure, with four theta functions:

θ

[a

(0, 0)

](2τ), a ∈ Z2/2Z2

where

θ

[m′

m′′

](τ) =

∑x∈Z2

e2πi

(12(x+m′

2)·τ ·t(x+m′

2)+(x+m′

2)·t(m′′

2))

are classical theta functions of half integral characteristicsdetermined by values m′,m′′ ∈ Z2/2Z2.

Page 8: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Algebraic modelsRosenhain model

We use level Γ(2)-structure with three functions

λ1(τ) =(

θ[ 0 00 0 ] θ[ 0 0

1 0 ]θ[ 0 0

1 1 ] θ[ 0 00 1 ]

)2

,

λ2(τ) =(

θ[ 0 01 0 ] θ[ 1 1

0 0 ]θ[ 0 0

0 1 ] θ[ 1 11 1 ]

)2

,

λ3(τ) =(

θ[ 0 00 0 ] θ[ 1 1

0 0 ]θ[ 0 0

1 1 ] θ[ 1 11 1 ]

)2

called Rosenhain invariants. These generate the function field ofA2(2) = Γ(2)\H2.

Page 9: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Relation to genus 2 curves

I Given a genus 2 curve C : y2 =∏6

i=1(x− ui) we can sendthree of the ui to 0, 1,∞ via a fractional linear transformationto get an isomorphic curve with a Rosenhain model

y2 = x(x− 1)(x− t1)(x− t2)(x− t3) .

The ti are called Rosenhain invariants.

I (0, 1,∞, t1, t2, t3) determines an ordering of the Weierstrasspoints and a level 2 structure on Jac(C) (∈ A2(2)).

I Let M2(2) denote the moduli space of genus 2 curvestogether with a full level 2 structure. Points of M2(2) aregiven by triples (t1, t2, t3) where ti 6= tj , 0, 1 for all i, j.

I The forgetful morphism M2(2) →M2 is a Galois covering ofdegree 720 = |S6| where S6 acts on the Weierstrass 6-tuple bypermutations, followed by renormalising the first three to(0, 1,∞).

Page 10: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Runge’s method

Let φ : A′ → A2 be a finite cover of A2. Then

φ−1H∆ =⋃finite

H(i)∆ .

Given functions {fi(τ)}i=1,...,n generating the function field of A′,

compute H(i)∆ (f1, . . . , fn) as follows:

1. Calculate the degree of the Humbert components H(i)∆ (given

by a formula).

2. Compute power series representations of the fi(τ) restrictedto H∆ ⊂ H2.

3. Solve H(i)∆ (f1, . . . , fn) = 0 in the power series ring (truncated

series with large precision) using linear algebra.

We shall consider the level 2 covering A2(2) → A2.

Page 11: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Step 1 - degree formula

Fortunately much arithmetic-geometric information is known aboutHumbert surfaces (van der Geer ’82). The number of Humbertcomponents in A2(2) is

m(∆) =

10 if ∆ ≡ 1 mod 815 if ∆ ≡ 0 mod 46 if ∆ ≡ 5 mod 8

(see Besser ’98).

Page 12: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

The degree of any Humbert component H(i)∆ in A2(2) is given by a

recursive formula

a∆ =∑x>0

v(∆/x2)m(∆/x2) deg(H

(i)∆/x2

)where

v(x) =

1/2 if x = 11 if x ≥ 2, x ≡ 0, 1 mod 4.

0 otherwise

Moreover, a∆ is the coefficient of a certain modular form of weight5/2 for the group Γ0(4), which fortunately has a more elementarydescription due to a formula of Siegel:

a∆ − 24∑x∈Z

σ1

(∆− x2

4

)=

{12∆− 2 if ∆ = �

0 otherwise

Page 13: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Here are the degrees for small discriminants:∆ 1 4 5 8 9 12 13 16 17 20 21 24

deg(H(i)∆ ) 2 4 8 8 24 16 40 32 48 32 80 48

actual deg 1 2 8 8 16 16 40 24 48 32 80 48

Remarks

I When ∆ ≡ 0 (mod 4) we have

mRunge(∆) = 4×mRosenhain(∆)

⇒ degRunge(∆) =14× degRosenhain(∆) .

I In reality (after computing these equations) the actual degrees

of H(i)n2 (λj) are less than what the formula produces. For

example

H1 : {ei − ej = 0, i 6= j} ∪ {ei = 0} ∪ {ei − 1 = 0} .

Page 14: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Step 2 - power series

Write ∆ = 4k + ` where ` is either 0 or 1, and k is uniquelydetermined. The Humbert surface of discriminant ∆ can bedefined by the set

H∆ = Sp4(Z)\{(

τ1 τ2

τ2 kτ1 + `τ2

)∈ H2

}.

Restrict θ[

a bc d

]to H∆ to get a Laurent series

θ[

a bc d

](τ) =

∑(x1,x2)∈Z2

eπi(x1c+x2d)r(2x1+a)2+k(2x2+b)2q2(2x1+a)(2x2+b)+`(2x2+b)2

where r = e2πiτ1/8 and q = e2πiτ2/8.

Page 15: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Unfortunately q has negative exponents. Substitute r = pq to get∑(x1,x2)∈Z2

(−1)x1c+x2dp(2x1+a)2+k(2x2+b)2q(2x1+a+2x2+b)2+(k+`−1)(2x2+b)2

which is a power series with integer coefficients.Using this representation one can compute the restriction ofλ1, λ2, λ3 to a Humbert surface as elements of Z[[p, q]]/(pN , qN )fairly easily.

Page 16: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Step 3 - linear algebra

Let d = deg(H(i)∆ ). To find the algebraic relation H

(i)∆ :

I Compute all monomials of degree ≤ d in the variablese1, e2, e3.

I Substitute ei = λi(p, q) ∈ Z[[p, q]]/(pN , qN ) in eachmonomial.

I Use linear algebra to find linear dependencies between thepower series monomials pmqn (compute null space of a bigmatrix).

Page 17: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

I With high enough precision there will be exactly one linearrelation between the monomials ei. This produces the

polynomial relation H(i)∆ (e1, e2, e3) = 0 which defines a

Humbert component.

I Once one component has been determined, the others caneasily be found by looking at the Rosenhain (S6) orbit of acomponent.

These other components will turn out to be useful when we look atShimura curves.

Page 18: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Runtime analysis

I There are:I

(d+33

)= O(d3) monomials to be evaluated

I O(N2) coefficients of evaluated power series expressions ofprecision N .

I Runtime cost is dominated by the nullspace calculation:O(d6N2) ≥ O(d9) to find a unique solution.

I Symmetries of the equation (arising from the fixed group ofthe humbert component) can be exploited to reduce thematrix size by a constant factor, giving a speedup by aconstant factor.

I Not overly efficient, but least it’s only a one time calculation..

Page 19: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Example

We calculate a component of H5:

λ1 = 1 + 16p4q8 + O(p12q12)λ2 = 1 + 4q4 + 8q8 − 8p4q4 − 24p4q8 + 4p8q8 + 48p8q8 + O(p12q12)λ3 = 1 + 4q4 + 8q8 + 8p4q4 + 40p4q8 + 4p8q8 + 48p8q8 + O(p12q12)

Using power series with precision 65, we compute the Humbertcomponent

e22e

23 − 2e

22e

33 + e

22e

43 + 2e1e2e

33 − 2e1e2e

43 − 2e1e

22e3 − 2e1e

22e

23 + 4e1e

22e

33 + 2e1e

32e3

−2e1e32e

33 + e

21e

43 − 2e

21e2e

33 + e

21e

22 + 4e

21e

22e3 − 4e

21e

22e

23 − 2e

21e

32 − 2e

21e

32e3

+4e21e

32e

23 + e

21e

42 − 2e

21e

42e3 + e

21e

42e

23 − 2e

31e

33 − 2e

31e2e3 + 4e

31e2e

23 + 2e

31e2e

33

−2e31e

22e

23 + 2e

31e

32e3 − 2e

31e

32e

23 + e

41e

23 − 2e

41e2e

23 + e

41e

22e

23

Page 20: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Application: Computing Shimura Curves

Page 21: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Quaternion ordersbare essentials

Let R be an order in an indefinite Q-quaternion algebra A.

I R is a QM-order if R = End(X) for some abelian surface X.

I Any x ∈ A satisfies x2 − tx + n = 0 where t, n are thereduced trace, norm respectively.

I ∆(x) = t(x)2 − 4n(x) defines a discriminant form

∆(x, y) =12(∆(x + y)−∆(x)−∆(y)) .

I The discriminant d(x1, . . . , x4) of a module generated byx1, . . . , x4 is defined to be the positive square root of

d(x1, . . . , x4)2 = −det(t(xixj)) .

Page 22: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Theorem (Runge ’99)

1. Any QM-order can be written as R = Z + Zα + Zβ + Zαβsuch that the discriminant matrix

S∆ =(

∆(α) ∆(α, β)∆(α, β) ∆(β)

)is positive definite. The discriminant of R equals det(S∆)/4.

2. A change of basis corresponds to changing the discriminantmatrix to tgS∆g for some g ∈ GL2(Z). ⇒ can assumediscriminant matrix is reduced.

3. If two orders have the same discriminant matrix which isprimitive (gcd(entries)=1) then the corresponding Shimuracurves are isomorphic.

Page 23: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Theorem (Hashimoto ’95, Runge ’99)

Let O = Z[ω] be a quadratic order of discriminant ∆. Let S∆ be adiscriminant matrix of a QM order R. The following areequivalent:

1. ∆ is primitively represented by S∆.

2. There exists an embedding O ↪→ R such that R ∩Q(ω) = O.

3. A Shimura curve C with QM order R is contained in H∆.

If we work in a finite cover, we have

C(h) ⊂ H(i)∆(α) ∩H

(j)∆(β)

if and only if we can write

tgS∆g =(

∆(α) ∗∗ ∆(β)

)for some g ∈ GL2(Z).

Page 24: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Example

H5 ∩H8 contains four Shimura curves CS :

Discriminant matrix QM-order discriminantS = det(S)/4(

5 00 8

)10(

5 22 8

)9(

5 44 8

)∼

(5 11 5

)6(

5 66 8

)∼

(1 00 4

)4

I This intersection was first computed by Hashimoto andMurabayashi (1995).

Page 25: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Computing intersections

I For intersections of Humbert components H∆i(e1, e2, e3) wecan find plane affine models simply by taking resultants withrespect to e1.

I As we are working with coordinates in M2(2) = A2(2)−H1,we will not be able to compute any Shimura curves in H1.

I S6 acts on Humbert components, hence acts on theirintersections producing isomorphic curves.

I Take one curve from each S6-orbit. Each of theseintersections is a component of a Shimura curve CS for somediscriminant matrix S.

Page 26: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Example

In our H5 ∩H8 example, there are three non-equivalentintersections:

C1 : a genus 1 curveC2 : a genus 3 hyperelliptic curveC3 : a genus 3 non-hyperelliptic curve

and the Shimura curves in M2(2) are

C( 5 00 8 ), C( 5 2

2 8 ) and C( 5 11 5 )

so there is a one-one correspondence between the Ci and the CS ,to be determined.

Page 27: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

Look at other Humbert intersections with “related” discriminants.Write D(a, b) for the set of discriminant matrices of QM-orders ofShimura curves in Ha ∩Hb. We have

D(5, 5) = { ( 5 11 5 ), ( 4 2

2 5 )}D(4, 5) = { ( 4 0

0 5 ), ( 4 22 5 )} ∪ {( 1 0

0 4 )}D(5, 9) = {( 5 1

1 9 ) , ( 5 22 8 ), ( 4 0

0 5 )} ∪ {( 1 00 4 )}

D(5, 8) = {( 5 00 8 ) , ( 5 2

2 8 ), ( 5 11 5 )} ∪ {( 1 0

0 4 )}

I Since D(4, 5) ∩ D(5, 5) = {( 4 22 5 )} we can identify the

corresponding curve. Hence we also know ( 5 11 5 ) (and ( 4 0

0 5 )).I Similarly ( 5 2

2 8 ) can be matched byD(5, 8) ∩ D(5, 9) = {( 5 2

2 8 )}.

Page 28: davidg@maths.usyd.edu.au The University of Sydneydavidg@maths.usyd.edu.au The University of Sydney AustMS Conference 2007 Outline Humbert surfaces Definitions Algebraic models Computing

In the end we find that:

C( 5 22 8 ) : the genus 1 curve

C( 5 11 5 ) : the genus 3 hyperelliptic curve

C( 5 00 8 ) : the genus 3 non-hyperelliptic curve

Thanks for listening!