101
Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems nt work with Shang-Hua Teng (Boston Universi

Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Embed Size (px)

Citation preview

Page 1: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Daniel A. Spielman

MIT

Fast, Randomized Algorithms forPartitioning, Sparsification, andthe Solution of Linear Systems

Joint work with Shang-Hua Teng (Boston University)

Page 2: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Nearly-Linear Time Algorithms forGraph Partitioning, Graph Sparsification, and Solving Linear Systems S-Teng ‘04

Papers

The Mixing Rate of Markov Chains, An Isoperimetric Inequality, and Computing The Volume Lovasz-Simonovits ‘93

The Eigenvalues of Random Symmetric Matrices Füredi-Komlós ‘81

Page 3: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Preconditioning to solve linear system

Overview

Find B that approximates A, such that solving is easy

Three techniques: Augmented Spanning Trees [Vaidya ’90]

Sparsification: Approximate graph by sparse graph

Partitioning: Runtime proportional to #nodes removed

Page 4: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

I. Linear system solvers

II. Sparsification using partitioning and random sampling

III. Graph partitioning by truncated random walks

Outline

Eigenvalues of random graphs

Combinatorial and Spectral Graph Theory

Page 5: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Diagonally Dominant Matrices

Just solve for

Solve wherediags of A at least sum of abs of others in row

1 2 3

Ex: Laplacian matrices of weighted graphs = negative weight from i to j diagonal = weighted degree

Page 6: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Complexity of Solving Ax = bA positive semi-definite

General Direct Methods:

Gaussian Elimination (Cholesky)

Conjugate Gradient

Fast matrix inversion

n is dimensionm is number non-zeros

Page 7: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Complexity of Solving Ax = bA positive semi-definite, structured

Planar: Nested dissection [Lipton-Rose-Tarjan ’79]

Tree: like path, work up from leaves

Path: forward and backward elimination

Express A = L LT

Page 8: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Preconditioned Conjugate Gradient

Omit for rest of talk

Iterative Methods

Find easy B that approximates A

Solve in time

Qualityof approximation

Time to solveBy = c

Page 9: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Iteratively solve in time

Main Result

For symmetric, diagonally dominant A

General:

Planar:

Page 10: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Vaidya’s Subgraph Preconditioners

Precondition A by the Laplacian of a subgraph, B

1 2

34

1 2

34

A B

Page 11: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Vaidya ’90 Relate to graph embedding cong/dil

use MSTaugmented MST

History

Gremban, Miller ‘96 Steiner vertices, many details

Bern, Boman, Chen, Gilbert, Hendrickson, Nguyen, Toledo All the details, algebraic approach

Joshi ’97, Reif ‘98 Recursive, multi-level approach

planar:

planar:

Page 12: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Maggs, Miller, Parekh, Ravi, Woo ‘02after preprocessing

History

Boman, Hendrickson ‘01 Low-stretch spanning trees

Clustering, Partitioning, Sparsification, Recursion Lower-stretch spanning trees

S-Teng ‘03 Augmented low-stretch trees

Page 13: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

if A-B is positive semi-definite,that is,

if A is positive semi-definite

For A Laplacian of graph with edges E

The relative condition number:

Page 14: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Bounding

For B subgraph of A,

and so

Page 15: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Fundamental Inequality

18

1

2 3

8 7

45

6

Page 16: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

18

1

2 3

8 7

45

6

Fundamental Inequality

Page 17: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Application to eigenvalues of graphs

Example: For complete graph on n nodes, all non-zero eigs = n

= 0 for Laplacian matrices, and

-7 -5 -3 -1 1 3 5 7

For path,

Page 18: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lower bound on

[Gauttery-Leighton-Miller ‘97]

Page 19: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lower bound on

So

And

Page 20: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Preconditioning with a Spanning Tree B

A B

Every edge of A not in B has unique path in B

Page 21: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

When B is a Tree

Page 22: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Low-Stretch Spanning Trees

where

Theorem (Alon-Karp-Peleg-West ’91)

Theorem (Elkin-Emek-S-Teng ’04)

Theorem (Boman-Hendrickson ’01)

Page 23: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

B a spanning tree plus s edges, in total

Vaidya’s Augmented Spanning Trees

Page 24: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Adding Edges to a Tree

Page 25: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Adding Edges to a Tree

Partition tree into t sub-trees, balancing stretch

Page 26: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Adding Edges to a Tree

Partition tree into t sub-trees, balancing stretchFor sub-trees connected in A, add one such bridge edge, carefully

(and in blue)

Page 27: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Adding Edges to a Tree

Theorem:

in general

if planar

t sub-trees

improve to

Page 28: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

All graphs can be well-approximated by a sparse graph

SparsificationFeder-Motwani ’91, Benczur-Karger ’96

D. Eppstein, Z. Galil, G.F. Italiano, and T. Spencer. ‘93 D. Eppstein, Z. Galil, G.F. Italiano, and A. Nissenzweig ’97

Page 29: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

SparsificationBenczur-Karger ’96: Can find sparse subgraph H s.t.

We need

(edges are subset, but different weights)

H has edges

Page 30: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Example: Complete Graph

If A is Laplacian of Kn, all non-zero eigenvalues are n

If B is Laplacian of Ramanujan expander all non-zero eigenvalues satisfy

And so

Page 31: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Example: Dumbell

If B does not contain middle edge,

Kn Kn

A

Page 32: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Example: Grid plus edge

1111

• Random sampling not sufficient.• Cut approximation not sufficient

= (m-1)2 + k(m-1)

(m-1)2

Page 33: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Conductance

Cut = partition of vertices

Conductance of S

S

Conductance of G

Page 34: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Conductance

S S

S

Page 35: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Conductance and Sparsification

If conductance high (expander) can precondition by random sampling

If conductance low can partition graph by removing few edges

Decomposition: Partition of vertex set remove few edges graph on each partition has high conductance

Page 36: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Graph Decomposition

Lemma: Exists partition of vertices

Each Vi has large At most half the edges cross partition

sample these edges

recurse on these edges

Alg:

Page 37: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Graph Decomposition Exists

Each Vi has At most half edges cross

Lemma: Exists partition of vertices

Proof: Let S be largest set s.t.

If

then

Page 38: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Graph Decomposition Exists

Proof: Let S be largest set s.t.

If

then

S

Page 39: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Graph Decomposition Exists

Proof: Let S be largest set s.t.

If

then

S

If S smallonly recurse in S

If S big, V-S not too big

S V - S

Bounded recursion depth

Page 40: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Sparsification from Graph Decomposition

sample these edges

recurse on these edges

Alg:

Theorem: Exists B with #edges <

Need to find the partition in nearly-linear time

Page 41: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Sparsification

Thm: For all A, find B edges

Thm: Given G, find subgraph H s.t.

in time

#edges(H) <

General solve time:

Page 42: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Cheeger’s Inequality[Sinclair-Jerrum ‘89]:

For Laplacian L, and D: diagonal matrix with = degree of node i

Page 43: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Useful if is big

Random Sampling

Given a graph will randomly sample to get

So that is small,

where is diagonal matrix of degrees in

Page 44: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Useful if is big

If

and

Then, for all x orthogonal to (mutual) nullspace

Page 45: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

But, don’t want D in there

D has no impact:

So

implies

Page 46: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Work with adjacency matrix

No difference, because

= weight of edge from i to j, 0 if i = j

Page 47: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Random Sampling Rule

Choose param governing sparsity of

Keep edge (i,j) with prob

If keep edge, raise weight by

Page 48: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Random Sampling Rule

guarantees

And,

guarantees expect at most edges in

Page 49: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Random Sampling Theorem

Theorem

Useful for

From now on, all edges have weight 1

Will prove in unweighted case.

Page 50: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Analysis by Trace

Trace = sum of diagonal entries = sum of eigenvalues

For even k

Page 51: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Analysis by Trace

Main Lemma:

Proof of Theorem:

Markov:

k-th root

Page 52: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Expected Trace

21

3

4

Page 53: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Expected Trace

Most terms zero because

So, sum is zero unless each edge appears at least twice.

Will code such walks to count them.

Page 54: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Coding walks

For sequence where

S = {i : edge not used before, either way}

the index of the time edge was used before, either way

Page 55: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Coding walks: Example

2

1

3 4

step: 0, 1, 2, 3, 4, 5, 6, 7, 8vert: 1, 2, 3, 4, 2, 3, 4, 2, 1

S = {1, 2, 3, 4}

1 2 2 3 3 4 4 2

: 5 2 6 3 7 4 8 1

Now, do a few slides pointing out what it meas

Page 56: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Coding walks: Example

2

1

3 4

step: 0, 1, 2, 3, 4, 5, 6, 7, 8vert: 1, 2, 3, 4, 2, 3, 4, 2, 1

S = {1, 2, 3, 4}

1 2 2 3 3 4 4 2

: 5 2 6 3 7 4 8 1

Now, do a few slides pointing out what it meas

Page 57: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

S = {1, 2, 3, 4}

1 2 2 3 3 4 4 2

Coding walks: Example

2

1

3 4

step: 0, 1, 2, 3, 4, 5, 6, 7, 8vert: 1, 2, 3, 4, 2, 3, 4, 2, 1

: 5 2 6 3 7 4 8 1

Now, do a few slides pointing out what it meas

Page 58: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

S = {1, 2, 3, 4}

1 2 2 3 3 4 4 2

Coding walks: Example

2

1

3 4

step: 0, 1, 2, 3, 4, 5, 6, 7, 8vert: 1, 2, 3, 4, 2, 3, 4, 2, 1

: 5 2 6 3 7 4 8 1

Now, do a few slides pointing out what it meas

Page 59: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

: 5 2 6 3 7 4 8 1

S = {1, 2, 3, 4}

1 2 2 3 3 4 4 2

Coding walks: Example

2

1

3 4

step: 0, 1, 2, 3, 4, 5, 6, 7, 8vert: 1, 2, 3, 4, 2, 3, 4, 2, 1

Now, do a few slides pointing out what it meas

Page 60: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Valid s

For each i in S is a neighbor of with a probability of being chosen < 1

That is, can be non-zero

For each i not in S can take edge indicated by

That is

Page 61: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Expected Trace by from Code

Page 62: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Expected Trace from Code

Are ways to choose given

Page 63: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Expected Trace from Code

Are ways to choose given

Page 64: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Random Sampling Theorem

Theorem

Useful for

Random sampling sparsifies graphs of high conductance

Page 65: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Graph Partitioning Algorithms

SDP/LP: too slow

Spectral: one cut quickly, but can be unbalanced many runs

Multilevel (Chaco/Metis): can’t analyze, miss small sparse cuts

New Alg: Based on truncated random walks. Approximates optimal balance,

Can be use to decompose

Run time

Page 66: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lazy Random Walk

At each step: stay put with prob ½ otherwise, move to neighbor according to weight

Diffusing probability mass: keep ½ for self, distribute rest among neighbors

Page 67: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lazy Random WalkDiffusing probability mass: keep ½ for self, distribute rest among neighbors

1 0

0

0

Or, with self-loops, distribute evenly over edges

1

3

2

2

Page 68: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lazy Random WalkDiffusing probability mass: keep ½ for self, distribute rest among neighbors

1/2 1/2

0

0

1

3

2

2

Or, with self-loops, distribute evenly over edges

Page 69: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lazy Random WalkDiffusing probability mass: keep ½ for self, distribute rest among neighbors

1/3 1/2

1/12

1/12

1

3

2

2

Or, with self-loops, distribute evenly over edges

Page 70: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lazy Random WalkDiffusing probability mass: keep ½ for self, distribute rest among neighbors

1/4 11/24

7/48

7/48

1

3

2

2

Or, with self-loops, distribute evenly over edges

Page 71: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lazy Random WalkDiffusing probability mass: keep ½ for self, distribute rest among neighbors

1/8 3/8

2/8

2/8

1

3

2

2

In limit, prob mass ~ degree

Page 72: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Why lazy?

Otherwise, might be no limit

1

1

0

0

Page 73: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Why so lazy to keep ½ mass?

Diagonally dominant matrices

weight of edge from i to j

degree of node i, if i = j

prob go from i to j

Page 74: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Rate of convergence

Cheeger’s inequality: related to conductance

If low conductance, slow convergence

If start uniform on S, prob leave at each step =

After steps, at least ¾ of mass still in S

Page 75: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lovasz-Simonovits Theorem

If slow convergence, then low conductance

And, can find the cut from highest probability nodes

.137

.135

.134

.129 .112

.094

Page 76: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lovasz-Simonovits Theorem

From now on, every node has same degree, d

For all vertex sets S, and all t · T

Where

k verts with most prob at step t

Page 77: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Lovasz-Simonovits Theorem

If start from node in set S with small conductance will output a set of small conductance

Extension: mostly contained in S

k verts with most prob at step t

Page 78: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Speed of Lovasz-Simonovits

Want run-time proportional to #vertices removedLocal clustering on massive graph

If want cut of conductancecan run for steps.

Problem: most vertices can have non-zero mass when cut is found

Page 79: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Speeding up Lovasz-Simonovits

Round all small entries of to zero

Algorithm Nibble input: start vertex v, target set size k start: step: all values < map to zero

Theorem: if v in conductance < size k set output a set with conductance < mostly overlapping

Page 80: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

The Potential Function

For integer k

Linear in between these points

Page 81: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Concave: slopes decrease

For

Page 82: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Easy Inequality

Page 83: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Fancy Inequality

Page 84: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Fancy Inequality

Page 85: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Chords with little progress

Page 86: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Dominating curve, makes progress

Page 87: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Dominating curve, makes progress

Page 88: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Proof of Easy Inequality

time t-1

time t

Order vertices by probability mass

If top k at time t-1 only connect to top k at time t get equality

Page 89: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Proof of Easy Inequality

time t-1

time t

Order vertices by probability mass

If top k at time t-1 only connect to top k at time t get equality

Otherwise, some mass leaves, and get inequality

Page 90: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

External edges from Self Loops

Lemma: For every set S, and every set R of same size At least frac of edges from S don’t hit R

3

3

3

3

3

3

S) = 1/2

S R

Tight example:

Page 91: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

External edges from Self Loops

Lemma: For every set S, and every set R of same size At least S) frac of edges from S don’t hit R

Proof:

S R

When R=S, is by definition.Each edge in R-S can absorb d edges from S.But each vertex of S-R has d self-loops that do not go to R.

Page 92: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Proof of Fancy Inequality

time t-1

time t

topin out

It(k) = top + in · top + (in + out)/2 = top/2 + (top + in + out)/2

Page 93: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Local Clustering

Theorem: If S is set of conductance < v is random vertex of S Then output set of conductance < mostly in S, in time proportional to size of output.

Page 94: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Local Clustering

Can it be done for all conductances???

Theorem: If S is set of conductance < v is random vertex of S Then output set of conductance < mostly in S, in time proportional to size of output.

Page 95: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Experimental code: ClusTree

1. Cluster, crudely

2. Make trees in clusters

3. Add edges between trees, optimally

No recursion on reduced matrices

Page 96: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

ImplementationClusTree: in java

timing not included could increase total time by 20%

PCGCholeskydroptol Inc CholVaidya

TAUCS Chen, Rotkin, Toledo

Orderings: amd, genmmd, Metis, RCM

Intel Xeon 3.06GHz, 512k L2 Cache, 1M L3 Cache

Page 97: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

2D grid, Neumann bdry

run to residual error 10-8

0 0.5 1 1.5 2

x 106

0

20

40

60

80

100

120

num nodes

seco

nd

s

2d Grid, Neumann

pcg/ic/genmmddirect/genmmdVaidyaclusTree

Page 98: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

0 0.5 1 1.5 2

x 106

0

20

40

60

80

100

120

num nodes

seco

nd

s

2d Grid, Neumann

pcg/ic/genmmddirect/genmmdVaidyaclusTree

0 0.5 1 1.5 2

x 106

0

20

40

60

80

100

num nodes

seco

nd

s

2d Grid, Dirichlet

direct/genmmdVaidyapcg/mic/rcmclusTree

2d Grid Dirichlet 2D grid, Neumann

Impact of boundary conditions

Page 99: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

0 5 10 15

x 105

0

20

40

60

80

100

120

num nodes

seco

nd

s

2d Unstructured Delaunay Mesh, Neumann

Vaidyadirect/Metispcg/ic/genmmdclusTree

2D Unstructured Delaunay Mesh

0 5 10 15

x 105

0

20

40

60

80

100

120

num nodes

seco

nd

s

2d Unstructured Delaunay Mesh, Dirichlet

Vaidyadirect/Metispcg/mic/genmmdclusTree

Dirichlet Neumann

Page 100: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

Future Work

Practical Local Clustering

More Sparsification

Other physical problems:

Cheeger’s inequalty

Implications for combinatorics,and Spectral Hypergraph Theory

Page 101: Daniel A. Spielman MIT Fast, Randomized Algorithms for Partitioning, Sparsification, and the Solution of Linear Systems Joint work with Shang-Hua Teng

“Nearly-Linear Time Algorithms for Graph Paritioning, Sparsification, and Solving Linear Systems” (STOC ’04, Arxiv)

To learn more

Will be split into two papers:numerical and combinatorial

My lecture notes forSpectral Graph Theory and its Applications