109
Random Walks On Graphs: Theory & Applications Random Walks On Graphs: Theory & Applications Amir Daneshgar Department of Mathematical Sciences Sharif University of Technology August 23, 2004

Random Walks On Graphs: Theory & Applicationsmath.sharif.ir/.../daneshgar/Lectures/RandomWalks2004.pdf · 2019. 7. 14. · Random Walks On Graphs: Theory & Applications Introduction

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

  • Random Walks On Graphs: Theory & Applications

    Random Walks On Graphs:Theory & Applications

    Amir Daneshgar

    Department of Mathematical Sciences

    Sharif University of Technology

    August 23, 2004

  • Random Walks On Graphs: Theory & Applications

    Introduction

    Basic idea

    I G and H are two geometric objects, we are going to compare.

  • Random Walks On Graphs: Theory & Applications

    Introduction

    Basic idea

    I PG and PH are diffusions on G and H, respectively, related toeach other through the map σ : G −→ H.

  • Random Walks On Graphs: Theory & Applications

    Introduction

    Basic idea

    I σ has some nice local properties.

    e.g. does not decrease local density.

  • Random Walks On Graphs: Theory & Applications

    Introduction

    Basic idea

    I If these condition are nice enough, we may deduce someproperties of these diffusions that are related to somegeometric properties of the objects.

    e.g. PH is faster than PG .

  • Random Walks On Graphs: Theory & Applications

    Introduction

    The main setup

    I For a geometric object G, the most natural operator relatedto its geometry is the Laplacian.

    I Laplacian appears naturally in different diffusion processes.

    I Usually the rate of convergence of the deffusion is controlledby the eigenvalues and the eigenfunctions of the Laplacian.

    I Eigenfunctions of the natural Laplacian are closely related tothe symmetries of the object.

  • Random Walks On Graphs: Theory & Applications

    Introduction

    The main setup

    I For a geometric object G, the most natural operator relatedto its geometry is the Laplacian.

    I Laplacian appears naturally in different diffusion processes.

    I Usually the rate of convergence of the deffusion is controlledby the eigenvalues and the eigenfunctions of the Laplacian.

    I Eigenfunctions of the natural Laplacian are closely related tothe symmetries of the object.

  • Random Walks On Graphs: Theory & Applications

    Introduction

    The main setup

    I For a geometric object G, the most natural operator relatedto its geometry is the Laplacian.

    I Laplacian appears naturally in different diffusion processes.

    I Usually the rate of convergence of the deffusion is controlledby the eigenvalues and the eigenfunctions of the Laplacian.

    I Eigenfunctions of the natural Laplacian are closely related tothe symmetries of the object.

  • Random Walks On Graphs: Theory & Applications

    Introduction

    The main setup

    I For a geometric object G, the most natural operator relatedto its geometry is the Laplacian.

    I Laplacian appears naturally in different diffusion processes.

    I Usually the rate of convergence of the deffusion is controlledby the eigenvalues and the eigenfunctions of the Laplacian.

    I Eigenfunctions of the natural Laplacian are closely related tothe symmetries of the object.

  • Random Walks On Graphs: Theory & Applications

    Outline

    Outline

    The main setupNatural random walks on graphsRandom walks and LaplacianSummary of part I

    Graph homomorphisms and random walksSome important problems in combinatoricsA couple of resultsSome applications

    Epilogue

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    GRAPH: a geometric object

    I A graph is a discrete geometric object with a set of vertices(denoting the space) and a set of edges (denoting theconnections).

    I In this talk we concentrate on simple graphs.

    I But everything can be extended to the general case.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    GRAPH: a geometric object

    I A graph is a discrete geometric object with a set of vertices(denoting the space) and a set of edges (denoting theconnections).

    I In this talk we concentrate on simple graphs.

    I But everything can be extended to the general case.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    GRAPH: a geometric object

    I A graph is a discrete geometric object with a set of vertices(denoting the space) and a set of edges (denoting theconnections).

    I In this talk we concentrate on simple graphs.

    I But everything can be extended to the general case.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Natural random walks on graphs

    The natural random walk on a graph

    I The natural random walk on a graph G is defined as,

    KG(u, v) =

    { 1d

    +

    G(u)

    u → v

    0 u 6→ v,

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    The Laplacian

    I

    K =

    0 12 0

    12

    13 0

    13

    13

    0 12 012

    13

    13

    13 0

    ∆ = I−K =

    1 −12 0 −12

    −13 1 −13 −

    13

    0 −12 1 −12

    −13 −13 −

    13 1

    .

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Evolution of a random walk

    I If xn is the probability distribution at stage n, then

    xn = xn−1KG ⇒ xn = x0KnG.

    How can one compute the nth power KnG?

    I If the chain is ergodic then the following limit exists and doesnot depend on the initial distribution, x0 ,

    limn

    x0KnG

    = π.

    I The stationary distribution π is in the left-kernel of theLaplacian (dually as a measure), i.e.

    π = πKG ⇔ π∆ = π(I −KG) = 0.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Evolution of a random walk

    I If xn is the probability distribution at stage n, then

    xn = xn−1KG ⇒ xn = x0KnG.

    How can one compute the nth power KnG?

    I If the chain is ergodic then the following limit exists and doesnot depend on the initial distribution, x0 ,

    limn

    x0KnG

    = π.

    I The stationary distribution π is in the left-kernel of theLaplacian (dually as a measure), i.e.

    π = πKG ⇔ π∆ = π(I −KG) = 0.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Evolution of a random walk

    I If xn is the probability distribution at stage n, then

    xn = xn−1KG ⇒ xn = x0KnG.

    How can one compute the nth power KnG?

    I If the chain is ergodic then the following limit exists and doesnot depend on the initial distribution, x0 ,

    limn

    x0KnG

    = π.

    I The stationary distribution π is in the left-kernel of theLaplacian (dually as a measure), i.e.

    π = πKG ⇔ π∆ = π(I −KG) = 0.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Evolution of a random walk

    I If xn is the probability distribution at stage n, then

    xn = xn−1KG ⇒ xn = x0KnG.

    How can one compute the nth power KnG?

    I If the chain is ergodic then the following limit exists and doesnot depend on the initial distribution, x0 ,

    limn

    x0KnG

    = π.

    I The stationary distribution π is in the left-kernel of theLaplacian (dually as a measure), i.e.

    π = πKG ⇔ π∆ = π(I −KG) = 0.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Fourier transform and diagonalization

    I There is a relationship between the solution sets of thefollowing equations,

    π = πKG ↔ h = KG h.

    I Such a function h is called harmonic.

    I If one can diagonalize the kernel KG as KG = PΛP−1, then

    in the new coordinates changed by P , everything is simplifiedas,

    ∆ = I − Λ & (π∆ = 0 ⇔ ∆h = 0).

    I How can we find such a diagonalization?

    I Fourier transform is available when there are enoughsymmetry in G.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Fourier transform and diagonalization

    I There is a relationship between the solution sets of thefollowing equations,

    π = πKG ↔ h = KG h.

    I Such a function h is called harmonic.

    I If one can diagonalize the kernel KG as KG = PΛP−1, then

    in the new coordinates changed by P , everything is simplifiedas,

    ∆ = I − Λ & (π∆ = 0 ⇔ ∆h = 0).

    I How can we find such a diagonalization?

    I Fourier transform is available when there are enoughsymmetry in G.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Fourier transform and diagonalization

    I There is a relationship between the solution sets of thefollowing equations,

    π = πKG ↔ h = KG h.

    I Such a function h is called harmonic.

    I If one can diagonalize the kernel KG as KG = PΛP−1, then

    in the new coordinates changed by P , everything is simplifiedas,

    ∆ = I − Λ & (π∆ = 0 ⇔ ∆h = 0).

    I How can we find such a diagonalization?

    I Fourier transform is available when there are enoughsymmetry in G.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Fourier transform and diagonalization

    I There is a relationship between the solution sets of thefollowing equations,

    π = πKG ↔ h = KG h.

    I Such a function h is called harmonic.

    I If one can diagonalize the kernel KG as KG = PΛP−1, then

    in the new coordinates changed by P , everything is simplifiedas,

    ∆ = I − Λ & (π∆ = 0 ⇔ ∆h = 0).

    I How can we find such a diagonalization?

    I Fourier transform is available when there are enoughsymmetry in G.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Random walks and Laplacian

    Fourier transform and diagonalization

    I There is a relationship between the solution sets of thefollowing equations,

    π = πKG ↔ h = KG h.

    I Such a function h is called harmonic.

    I If one can diagonalize the kernel KG as KG = PΛP−1, then

    in the new coordinates changed by P , everything is simplifiedas,

    ∆ = I − Λ & (π∆ = 0 ⇔ ∆h = 0).

    I How can we find such a diagonalization?

    I Fourier transform is available when there are enoughsymmetry in G.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Summary of part I

    Summing up

    I Laplacian is an important self-adjoint operator that is closelyrelated to the geometry of its space.

    I The eigenvalues of the Laplacian control the rate ofconvergence of some natural diffusions on the space.

    I There is a good knowledge about the eigenspace of theLaplacian when the space admits enough symmetry.

    I It is possible to compare the rates of convergence of twocoupled diffusions when the coupling is through a nearlycontinuous mapping.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Summary of part I

    Summing up

    I Laplacian is an important self-adjoint operator that is closelyrelated to the geometry of its space.

    I The eigenvalues of the Laplacian control the rate ofconvergence of some natural diffusions on the space.

    I There is a good knowledge about the eigenspace of theLaplacian when the space admits enough symmetry.

    I It is possible to compare the rates of convergence of twocoupled diffusions when the coupling is through a nearlycontinuous mapping.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Summary of part I

    Summing up

    I Laplacian is an important self-adjoint operator that is closelyrelated to the geometry of its space.

    I The eigenvalues of the Laplacian control the rate ofconvergence of some natural diffusions on the space.

    I There is a good knowledge about the eigenspace of theLaplacian when the space admits enough symmetry.

    I It is possible to compare the rates of convergence of twocoupled diffusions when the coupling is through a nearlycontinuous mapping.

  • Random Walks On Graphs: Theory & Applications

    The main setup

    Summary of part I

    Summing up

    I Laplacian is an important self-adjoint operator that is closelyrelated to the geometry of its space.

    I The eigenvalues of the Laplacian control the rate ofconvergence of some natural diffusions on the space.

    I There is a good knowledge about the eigenspace of theLaplacian when the space admits enough symmetry.

    I It is possible to compare the rates of convergence of twocoupled diffusions when the coupling is through a nearlycontinuous mapping.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A graph homomorphism

    I A graph homomorphism σ from a graph G to a graph H is amap σ : V (G) −→ V (H) such that uv ∈ E(G) impliesσ(u)σ(v) ∈ E(H).

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A graph homomorphism

    I A graph homomorphism σ from a graph G to a graph H is amap σ : V (G) −→ V (H) such that uv ∈ E(G) impliesσ(u)σ(v) ∈ E(H).

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A question

    I Does there exist a homomorphism from the Petersen graph tothe triangle K3?

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A question

    I Does there exist a homomorphism from the Petersen graph tothe triangle K3?

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Graph colouring

    I Homomorphisms to Kn is equivalent to colouring the verticesof the graph by n colours such that the terminal ends of eachedge have different colours.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Graph colouring

    I Homomorphisms to Kn is equivalent to colouring the verticesof the graph by n colours such that the terminal ends of eachedge have different colours.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Another question!

    I Does there exist a homomorphism from the Petersen graph tothe 5-cycle C5?

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Another question!

    I Does there exist a homomorphism from the Petersen graph tothe 5-cycle C5?

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.I The circular colouring problem.I The fractional colouring problem.I The graph partitioning problem, specially, existence results in

    design theory.I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.I The circular colouring problem.I The fractional colouring problem.I The graph partitioning problem, specially, existence results in

    design theory.I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.

    I The circular colouring problem.I The fractional colouring problem.I The graph partitioning problem, specially, existence results in

    design theory.I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.I The circular colouring problem.

    I The fractional colouring problem.I The graph partitioning problem, specially, existence results in

    design theory.I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.I The circular colouring problem.I The fractional colouring problem.

    I The graph partitioning problem, specially, existence results indesign theory.

    I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.I The circular colouring problem.I The fractional colouring problem.I The graph partitioning problem, specially, existence results in

    design theory.

    I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Graph homomorphisms and combinatorics

    I Graph homomorphisms are natural maps in the category ofgraphs.

    I Many different concepts in combinatorics are related to thehomomorphism problem, e.g.

    I The ordinary colouring problem.I The circular colouring problem.I The fractional colouring problem.I The graph partitioning problem, specially, existence results in

    design theory.I The Hamiltonicity problem.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some important problems in combinatorics

    Algorithmic considerations

    I The following problem is NP-complete(P. Hell & J. Nesetril 1990).

    Problem: HCOL.Constant: A graph H.

    Given: A graph G.Question: Does there exist a homomorphism σ : G −→ H?

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A couple of results

    Some definitions and notations I

    I Notations Hom(G, H), Homv(G, H) and Home(G, H)denote the sets of ordinary, onto-vertices and onto-edgeshomomorphisms from G to H, respectively.

    I Inverse image parameters: If σ ∈ Hom(G, H) andX, Y ⊆ V (G) we define,

    I E(X, Y ) , {uv ∈ E(G) | u ∈ X & v ∈ Y },I Mσ , min

    x,y∈V (H){|E(σ−1(x), σ−1(y))| | E(σ−1(x), σ−1(y)) 6= ∅},

    I Mσ

    , maxx,y∈V (H)

    |E(σ−1(x), σ−1(y))|,

    I Sσ , minx∈V (H)

    |σ−1(x)|,

    I Sσ

    , maxx∈V (H)

    |σ−1(x)|.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A couple of results

    Some definitions and notations I

    I Notations Hom(G, H), Homv(G, H) and Home(G, H)denote the sets of ordinary, onto-vertices and onto-edgeshomomorphisms from G to H, respectively.

    I Inverse image parameters: If σ ∈ Hom(G, H) andX, Y ⊆ V (G) we define,

    I E(X, Y ) , {uv ∈ E(G) | u ∈ X & v ∈ Y },I Mσ , min

    x,y∈V (H){|E(σ−1(x), σ−1(y))| | E(σ−1(x), σ−1(y)) 6= ∅},

    I Mσ

    , maxx,y∈V (H)

    |E(σ−1(x), σ−1(y))|,

    I Sσ , minx∈V (H)

    |σ−1(x)|,

    I Sσ

    , maxx∈V (H)

    |σ−1(x)|.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A couple of results

    Some definitions and notations II

    I For a graph G we define AG to be the adjacency matrix of Gand we let DG be the diagonal matrix whose diagonal entrycorresponding to the vertex v, is dv the degree of v. Then thecombinatorial Laplacian of G is defined as ∆G , DG −AG .

    I We order the eigenvalues of ∆G as follows,

    λG

    1≤ λG

    2≤ · · · ≤ λG

    n,

    I We consider the following random walk,

    K(u, v) =

    1∆ u ∼ v & u 6= v

    ∆−du∆ u = v

    0 otherwise.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A couple of results

    Some definitions and notations II

    I For a graph G we define AG to be the adjacency matrix of Gand we let DG be the diagonal matrix whose diagonal entrycorresponding to the vertex v, is dv the degree of v. Then thecombinatorial Laplacian of G is defined as ∆G , DG −AG .

    I We order the eigenvalues of ∆G as follows,

    λG

    1≤ λG

    2≤ · · · ≤ λG

    n,

    I We consider the following random walk,

    K(u, v) =

    1∆ u ∼ v & u 6= v

    ∆−du∆ u = v

    0 otherwise.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A couple of results

    Some definitions and notations II

    I For a graph G we define AG to be the adjacency matrix of Gand we let DG be the diagonal matrix whose diagonal entrycorresponding to the vertex v, is dv the degree of v. Then thecombinatorial Laplacian of G is defined as ∆G , DG −AG .

    I We order the eigenvalues of ∆G as follows,

    λG

    1≤ λG

    2≤ · · · ≤ λG

    n,

    I We consider the following random walk,

    K(u, v) =

    1∆ u ∼ v & u 6= v

    ∆−du∆ u = v

    0 otherwise.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    A couple of results

    A comparison theorem(A. Daneshgar & H. Hajiabolhassan 2002)

    I Let G and H be two graphs with |V (G)| = n and|V (H)| = m.a ) If σ ∈ Homv(G, H), then for all 1 ≤ k ≤ m,

    λG

    k≤ M

    σ

    Sσλ

    H

    k.

    b ) If σ ∈ Home(G, H), then for all 1 ≤ k ≤ m,

    λG

    n−m+k≥ MσSσ

    λH

    k.

    c ) If σ ∈ Hom(G, H) and H is both vertex and edge transitivethen,

    λG

    n≥ 2|E(G)|

    n∆H

    λH

    m.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    hamiltonicity of the Petersen graph!

    I If a graph G with n vertices is Hamiltonian then there exists ahomomorphism such as σ ∈ Homv(Cn , G) such that

    Sσ = Sσ = Mσ

    = 1

    .

    I Considering λP

    5and Part (a) show that Homv(C10 , P ) = ∅.

    I Hence, Petersen graph is not Hamiltonian.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    hamiltonicity of the Petersen graph!

    I If a graph G with n vertices is Hamiltonian then there exists ahomomorphism such as σ ∈ Homv(Cn , G) such that

    Sσ = Sσ = Mσ

    = 1

    .

    I Considering λP

    5and Part (a) show that Homv(C10 , P ) = ∅.

    I Hence, Petersen graph is not Hamiltonian.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    hamiltonicity of the Petersen graph!

    I If a graph G with n vertices is Hamiltonian then there exists ahomomorphism such as σ ∈ Homv(Cn , G) such that

    Sσ = Sσ = Mσ

    = 1

    .

    I Considering λP

    5and Part (a) show that Homv(C10 , P ) = ∅.

    I Hence, Petersen graph is not Hamiltonian.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    hamiltonicity of the Petersen graph!

    I If a graph G with n vertices is Hamiltonian then there exists ahomomorphism such as σ ∈ Homv(Cn , G) such that

    Sσ = Sσ = Mσ

    = 1

    .

    I Considering λP

    5and Part (a) show that Homv(C10 , P ) = ∅.

    I Hence, Petersen graph is not Hamiltonian.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Fisher’s inequality!

    I Let λH denote the graph on V (H) such that each edge of Hhas multiplicity λ. It is well-known that a 2− (v, k, λ) designon the set V = {1, 2, . . . , v} with b blocks can be consideredas a decomposition of λKv by b copies of Kk .

    I In other words, this is equivalent to considering the existenceof a homomorphism σ ∈ Home(∪bi=1Kk ,Kv) for whichMσ = M

    σ= λ. It is easy to see that for such a

    homomorphism one has r = Sσ = Sσ

    = bkv .I To prove Fisher’s inequality, assume that b < v. Then by Part

    (a) for the bth eigenvalue we should have k ≤ λr v. But sincer(k − 1) = λ(v − 1) holds for any 2-design, we should haveλ ≥ r and consequently, v = k.

    I Hence, v > k ⇒ b ≥ v.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Fisher’s inequality!

    I Let λH denote the graph on V (H) such that each edge of Hhas multiplicity λ. It is well-known that a 2− (v, k, λ) designon the set V = {1, 2, . . . , v} with b blocks can be consideredas a decomposition of λKv by b copies of Kk .

    I In other words, this is equivalent to considering the existenceof a homomorphism σ ∈ Home(∪bi=1Kk ,Kv) for whichMσ = M

    σ= λ. It is easy to see that for such a

    homomorphism one has r = Sσ = Sσ

    = bkv .I To prove Fisher’s inequality, assume that b < v. Then by Part

    (a) for the bth eigenvalue we should have k ≤ λr v. But sincer(k − 1) = λ(v − 1) holds for any 2-design, we should haveλ ≥ r and consequently, v = k.

    I Hence, v > k ⇒ b ≥ v.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Fisher’s inequality!

    I Let λH denote the graph on V (H) such that each edge of Hhas multiplicity λ. It is well-known that a 2− (v, k, λ) designon the set V = {1, 2, . . . , v} with b blocks can be consideredas a decomposition of λKv by b copies of Kk .

    I In other words, this is equivalent to considering the existenceof a homomorphism σ ∈ Home(∪bi=1Kk ,Kv) for whichMσ = M

    σ= λ. It is easy to see that for such a

    homomorphism one has r = Sσ = Sσ

    = bkv .

    I To prove Fisher’s inequality, assume that b < v. Then by Part(a) for the bth eigenvalue we should have k ≤ λr v. But sincer(k − 1) = λ(v − 1) holds for any 2-design, we should haveλ ≥ r and consequently, v = k.

    I Hence, v > k ⇒ b ≥ v.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Fisher’s inequality!

    I Let λH denote the graph on V (H) such that each edge of Hhas multiplicity λ. It is well-known that a 2− (v, k, λ) designon the set V = {1, 2, . . . , v} with b blocks can be consideredas a decomposition of λKv by b copies of Kk .

    I In other words, this is equivalent to considering the existenceof a homomorphism σ ∈ Home(∪bi=1Kk ,Kv) for whichMσ = M

    σ= λ. It is easy to see that for such a

    homomorphism one has r = Sσ = Sσ

    = bkv .I To prove Fisher’s inequality, assume that b < v. Then by Part

    (a) for the bth eigenvalue we should have k ≤ λr v. But sincer(k − 1) = λ(v − 1) holds for any 2-design, we should haveλ ≥ r and consequently, v = k.

    I Hence, v > k ⇒ b ≥ v.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Fisher’s inequality!

    I Let λH denote the graph on V (H) such that each edge of Hhas multiplicity λ. It is well-known that a 2− (v, k, λ) designon the set V = {1, 2, . . . , v} with b blocks can be consideredas a decomposition of λKv by b copies of Kk .

    I In other words, this is equivalent to considering the existenceof a homomorphism σ ∈ Home(∪bi=1Kk ,Kv) for whichMσ = M

    σ= λ. It is easy to see that for such a

    homomorphism one has r = Sσ = Sσ

    = bkv .I To prove Fisher’s inequality, assume that b < v. Then by Part

    (a) for the bth eigenvalue we should have k ≤ λr v. But sincer(k − 1) = λ(v − 1) holds for any 2-design, we should haveλ ≥ r and consequently, v = k.

    I Hence, v > k ⇒ b ≥ v.

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Petersen and C5!

    I Let us consider a homomorphism σ ∈ Hom(P,C5). First, notethat since P contains two disjoint copies of C5 , we candeduce that Sσ = 2.

    I Since C5 is critical, the homomorphism should be inHome(P,C5), and it is easy to see that Mσ ≥ 3.

    I Applying Part (b) for k = 3 shows that we should have

    1.5 ≤ MσSσ

    ≤ 1.38,

    which is a contradiction.

    I Hence,

    Hom(P,C5) = Home(P,C5) = ∅

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Petersen and C5!

    I Let us consider a homomorphism σ ∈ Hom(P,C5). First, notethat since P contains two disjoint copies of C5 , we candeduce that Sσ = 2.

    I Since C5 is critical, the homomorphism should be inHome(P,C5), and it is easy to see that Mσ ≥ 3.

    I Applying Part (b) for k = 3 shows that we should have

    1.5 ≤ MσSσ

    ≤ 1.38,

    which is a contradiction.

    I Hence,

    Hom(P,C5) = Home(P,C5) = ∅

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Petersen and C5!

    I Let us consider a homomorphism σ ∈ Hom(P,C5). First, notethat since P contains two disjoint copies of C5 , we candeduce that Sσ = 2.

    I Since C5 is critical, the homomorphism should be inHome(P,C5), and it is easy to see that Mσ ≥ 3.

    I Applying Part (b) for k = 3 shows that we should have

    1.5 ≤ MσSσ

    ≤ 1.38,

    which is a contradiction.

    I Hence,

    Hom(P,C5) = Home(P,C5) = ∅

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Petersen and C5!

    I Let us consider a homomorphism σ ∈ Hom(P,C5). First, notethat since P contains two disjoint copies of C5 , we candeduce that Sσ = 2.

    I Since C5 is critical, the homomorphism should be inHome(P,C5), and it is easy to see that Mσ ≥ 3.

    I Applying Part (b) for k = 3 shows that we should have

    1.5 ≤ MσSσ

    ≤ 1.38,

    which is a contradiction.

    I Hence,

    Hom(P,C5) = Home(P,C5) = ∅

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    Petersen and C5!

    I Let us consider a homomorphism σ ∈ Hom(P,C5). First, notethat since P contains two disjoint copies of C5 , we candeduce that Sσ = 2.

    I Since C5 is critical, the homomorphism should be inHome(P,C5), and it is easy to see that Mσ ≥ 3.

    I Applying Part (b) for k = 3 shows that we should have

    1.5 ≤ MσSσ

    ≤ 1.38,

    which is a contradiction.

    I Hence,

    Hom(P,C5) = Home(P,C5) = ∅

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    A spectral bound for the fractionalchromatic number!

    I Note that the automorphism group of KG(m,n) actstransitively on both vertices and edges.

    I Also, the smallest and largest eigenvalues of the adjacencymatrix of the Kneser graph KG(m,n) are −

    (m−n−1

    n−1)

    and(m−n

    n

    ), respectively.

    I Let χf(G) = mn be the fractional chromatic number of the

    graph G and also let |V (G)| = ν and d = 2|E(G)|ν .I Applying Part (c) yields,

    χf(G) ≥

    λG

    ν

    λGν− d

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    A spectral bound for the fractionalchromatic number!

    I Note that the automorphism group of KG(m,n) actstransitively on both vertices and edges.

    I Also, the smallest and largest eigenvalues of the adjacencymatrix of the Kneser graph KG(m,n) are −

    (m−n−1

    n−1)

    and(m−n

    n

    ), respectively.

    I Let χf(G) = mn be the fractional chromatic number of the

    graph G and also let |V (G)| = ν and d = 2|E(G)|ν .I Applying Part (c) yields,

    χf(G) ≥

    λG

    ν

    λGν− d

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    A spectral bound for the fractionalchromatic number!

    I Note that the automorphism group of KG(m,n) actstransitively on both vertices and edges.

    I Also, the smallest and largest eigenvalues of the adjacencymatrix of the Kneser graph KG(m,n) are −

    (m−n−1

    n−1)

    and(m−n

    n

    ), respectively.

    I Let χf(G) = mn be the fractional chromatic number of the

    graph G and also let |V (G)| = ν and d = 2|E(G)|ν .I Applying Part (c) yields,

    χf(G) ≥

    λG

    ν

    λGν− d

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    A spectral bound for the fractionalchromatic number!

    I Note that the automorphism group of KG(m,n) actstransitively on both vertices and edges.

    I Also, the smallest and largest eigenvalues of the adjacencymatrix of the Kneser graph KG(m,n) are −

    (m−n−1

    n−1)

    and(m−n

    n

    ), respectively.

    I Let χf(G) = mn be the fractional chromatic number of the

    graph G and also let |V (G)| = ν and d = 2|E(G)|ν .

    I Applying Part (c) yields,

    χf(G) ≥

    λG

    ν

    λGν− d

    .

  • Random Walks On Graphs: Theory & Applications

    Graph homomorphisms and random walks

    Some applications

    A spectral bound for the fractionalchromatic number!

    I Note that the automorphism group of KG(m,n) actstransitively on both vertices and edges.

    I Also, the smallest and largest eigenvalues of the adjacencymatrix of the Kneser graph KG(m,n) are −

    (m−n−1

    n−1)

    and(m−n

    n

    ), respectively.

    I Let χf(G) = mn be the fractional chromatic number of the

    graph G and also let |V (G)| = ν and d = 2|E(G)|ν .I Applying Part (c) yields,

    χf(G) ≥

    λG

    ν

    λGν− d

    .

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some connections

    The main setup we applied in this talk has many other variants,e.g.

    I (Conservation of energy v.s. Probability) Potential theory andPhysics.

    I (Rate of convergence and mixing times v.s. Heat kernel)Analysis and Probability theory.

    I (Geometry of homogeneous spaces) Graph theory, Numbertheory, Hyperbolic geometry.Specially, harmonic analysis on reductive groups.

    I (Group presentations) Hyperbolic geometry, Combinatorics,3-manifolds.

    I (Expanders and rapid mixing) Computer science,Cryptography, Graph theory, Communication theory.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some connections

    The main setup we applied in this talk has many other variants,e.g.

    I (Conservation of energy v.s. Probability) Potential theory andPhysics.

    I (Rate of convergence and mixing times v.s. Heat kernel)Analysis and Probability theory.

    I (Geometry of homogeneous spaces) Graph theory, Numbertheory, Hyperbolic geometry.Specially, harmonic analysis on reductive groups.

    I (Group presentations) Hyperbolic geometry, Combinatorics,3-manifolds.

    I (Expanders and rapid mixing) Computer science,Cryptography, Graph theory, Communication theory.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some connections

    The main setup we applied in this talk has many other variants,e.g.

    I (Conservation of energy v.s. Probability) Potential theory andPhysics.

    I (Rate of convergence and mixing times v.s. Heat kernel)Analysis and Probability theory.

    I (Geometry of homogeneous spaces) Graph theory, Numbertheory, Hyperbolic geometry.Specially, harmonic analysis on reductive groups.

    I (Group presentations) Hyperbolic geometry, Combinatorics,3-manifolds.

    I (Expanders and rapid mixing) Computer science,Cryptography, Graph theory, Communication theory.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some connections

    The main setup we applied in this talk has many other variants,e.g.

    I (Conservation of energy v.s. Probability) Potential theory andPhysics.

    I (Rate of convergence and mixing times v.s. Heat kernel)Analysis and Probability theory.

    I (Geometry of homogeneous spaces) Graph theory, Numbertheory, Hyperbolic geometry.Specially, harmonic analysis on reductive groups.

    I (Group presentations) Hyperbolic geometry, Combinatorics,3-manifolds.

    I (Expanders and rapid mixing) Computer science,Cryptography, Graph theory, Communication theory.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some connections

    The main setup we applied in this talk has many other variants,e.g.

    I (Conservation of energy v.s. Probability) Potential theory andPhysics.

    I (Rate of convergence and mixing times v.s. Heat kernel)Analysis and Probability theory.

    I (Geometry of homogeneous spaces) Graph theory, Numbertheory, Hyperbolic geometry.Specially, harmonic analysis on reductive groups.

    I (Group presentations) Hyperbolic geometry, Combinatorics,3-manifolds.

    I (Expanders and rapid mixing) Computer science,Cryptography, Graph theory, Communication theory.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some connections

    The main setup we applied in this talk has many other variants,e.g.

    I (Conservation of energy v.s. Probability) Potential theory andPhysics.

    I (Rate of convergence and mixing times v.s. Heat kernel)Analysis and Probability theory.

    I (Geometry of homogeneous spaces) Graph theory, Numbertheory, Hyperbolic geometry.Specially, harmonic analysis on reductive groups.

    I (Group presentations) Hyperbolic geometry, Combinatorics,3-manifolds.

    I (Expanders and rapid mixing) Computer science,Cryptography, Graph theory, Communication theory.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some questions!

    I (Homomorphism transformations) Develop transformationmethods that makes the inequalities more applicable.

    I (Infinite enumerable case) Develop limit theorems such thatone can use inequalities asymptotically for infinite graphs.

    I (Group representations) Consider some well known classes ofgroups for which the character table is known and study thecorresponding colouring numbers.

    I (Nodal domains) Study nodal domains and the correspondingoptimization problems.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some questions!

    I (Homomorphism transformations) Develop transformationmethods that makes the inequalities more applicable.

    I (Infinite enumerable case) Develop limit theorems such thatone can use inequalities asymptotically for infinite graphs.

    I (Group representations) Consider some well known classes ofgroups for which the character table is known and study thecorresponding colouring numbers.

    I (Nodal domains) Study nodal domains and the correspondingoptimization problems.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some questions!

    I (Homomorphism transformations) Develop transformationmethods that makes the inequalities more applicable.

    I (Infinite enumerable case) Develop limit theorems such thatone can use inequalities asymptotically for infinite graphs.

    I (Group representations) Consider some well known classes ofgroups for which the character table is known and study thecorresponding colouring numbers.

    I (Nodal domains) Study nodal domains and the correspondingoptimization problems.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some questions!

    I (Homomorphism transformations) Develop transformationmethods that makes the inequalities more applicable.

    I (Infinite enumerable case) Develop limit theorems such thatone can use inequalities asymptotically for infinite graphs.

    I (Group representations) Consider some well known classes ofgroups for which the character table is known and study thecorresponding colouring numbers.

    I (Nodal domains) Study nodal domains and the correspondingoptimization problems.

  • Random Walks On Graphs: Theory & Applications

    Epilogue

    Some questions!

    I (Homomorphism transformations) Develop transformationmethods that makes the inequalities more applicable.

    I (Infinite enumerable case) Develop limit theorems such thatone can use inequalities asymptotically for infinite graphs.

    I (Group representations) Consider some well known classes ofgroups for which the character table is known and study thecorresponding colouring numbers.

    I (Nodal domains) Study nodal domains and the correspondingoptimization problems.

  • Random Walks On Graphs: Theory & Applications

    The End

    The End.

    Thank You!

    IntroductionOutlineThe main setupNatural random walks on graphsRandom walks and LaplacianSummary of part I

    Graph homomorphisms and random walksSome important problems in combinatoricsA couple of resultsSome applications

    EpilogueThe End