U.C. Berkeley CS278: Computational Complexity Professor Luca Trevisan 2/21/2008. Notes for Lecture 8

Similar documents
Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

U.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016

Lecture 10: May 6, 2013

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

Notes on Space-Bounded Complexity

U.C. Berkeley CS294: Beyond Worst-Case Analysis Handout 6 Luca Trevisan September 12, 2017

U.C. Berkeley CS278: Computational Complexity Professor Luca Trevisan 9/6/2004. Notes for Lecture 3

Finding Dense Subgraphs in G(n, 1/2)

Stanford University Graph Partitioning and Expanders Handout 3 Luca Trevisan May 8, 2013

Stanford University CS254: Computational Complexity Notes 7 Luca Trevisan January 29, Notes for Lecture 7

Notes on Space-Bounded Complexity

Lecture Space-Bounded Derandomization

Computing Correlated Equilibria in Multi-Player Games

Lecture 4: Constant Time SVD Approximation

Min Cut, Fast Cut, Polynomial Identities

Calculation of time complexity (3%)

Problem Set 9 Solutions

Eigenvalues of Random Graphs

Lecture 17: Lee-Sidford Barrier

Spectral Clustering. Shannon Quinn

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Bézier curves. Michael S. Floater. September 10, These notes provide an introduction to Bézier curves. i=0

Random Walks on Digraphs

Lecture 2: Gram-Schmidt Vectors and the LLL Algorithm

arxiv: v1 [quant-ph] 6 Sep 2007

Lecture 4: Universal Hash Functions/Streaming Cont d

Singular Value Decomposition: Theory and Applications

Math 261 Exercise sheet 2

Non-negative Matrices and Distributed Control

More metrics on cartesian products

Edge Isoperimetric Inequalities

Turing Machines (intro)

Feature Selection: Part 1

A 2D Bounded Linear Program (H,c) 2D Linear Programming

NP-Completeness : Proofs

10-701/ Machine Learning, Fall 2005 Homework 3

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 12 10/21/2013. Martingale Concentration Inequalities and Applications

6.842 Randomness and Computation February 18, Lecture 4

763622S ADVANCED QUANTUM MECHANICS Solution Set 1 Spring c n a n. c n 2 = 1.

Lecture 12: Discrete Laplacian

Lecture 4: November 17, Part 1 Single Buffer Management

NOTES ON SIMPLIFICATION OF MATRICES

Randić Energy and Randić Estrada Index of a Graph

Google PageRank with Stochastic Matrix

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence

18.1 Introduction and Recap

arxiv: v2 [cs.ds] 1 Feb 2017

1 The Mistake Bound Model

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

Volume 18 Figure 1. Notation 1. Notation 2. Observation 1. Remark 1. Remark 2. Remark 3. Remark 4. Remark 5. Remark 6. Theorem A [2]. Theorem B [2].

Errors for Linear Systems

Discrete Mathematics. Laplacian spectral characterization of some graphs obtained by product operation

Dynamic Programming. Preview. Dynamic Programming. Dynamic Programming. Dynamic Programming (Example: Fibonacci Sequence)

CALCULUS CLASSROOM CAPSULES

1 Matrix representations of canonical matrices

Norms, Condition Numbers, Eigenvalues and Eigenvectors

Assortment Optimization under MNL

Statistical Mechanics and Combinatorics : Lecture III

O-line Temporary Tasks Assignment. Abstract. In this paper we consider the temporary tasks assignment

Notes on Frequency Estimation in Data Streams

COS 511: Theoretical Machine Learning. Lecturer: Rob Schapire Lecture # 15 Scribe: Jieming Mao April 1, 2013

Finding Primitive Roots Pseudo-Deterministically

A property of the elementary symmetric functions

Dynamic Systems on Graphs

Case A. P k = Ni ( 2L i k 1 ) + (# big cells) 10d 2 P k.

LECTURE 9 CANONICAL CORRELATION ANALYSIS

The Minimum Universal Cost Flow in an Infeasible Flow Network

COS 521: Advanced Algorithms Game Theory and Linear Programming

princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 3: Large deviations bounds and applications Lecturer: Sanjeev Arora

Affine transformations and convexity

Salmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2

2.3 Nilpotent endomorphisms

Randomness and Computation

arxiv: v1 [cs.gt] 14 Mar 2019

Lecture Randomized Load Balancing strategies and their analysis. Probability concepts include, counting, the union bound, and Chernoff bounds.

= z 20 z n. (k 20) + 4 z k = 4

1 GSW Iterative Techniques for y = Ax

Bezier curves. Michael S. Floater. August 25, These notes provide an introduction to Bezier curves. i=0

Difference Equations

However, since P is a symmetric idempotent matrix, of P are either 0 or 1 [Eigen-values

Example: (13320, 22140) =? Solution #1: The divisors of are 1, 2, 3, 4, 5, 6, 9, 10, 12, 15, 18, 20, 27, 30, 36, 41,

C/CS/Phy191 Problem Set 3 Solutions Out: Oct 1, 2008., where ( 00. ), so the overall state of the system is ) ( ( ( ( 00 ± 11 ), Φ ± = 1

(k,?)-sandwich Problems: why not ask for special kinds of bread?

8.6 The Complex Number System

Maximizing the number of nonnegative subsets

Outline and Reading. Dynamic Programming. Dynamic Programming revealed. Computing Fibonacci. The General Dynamic Programming Technique

ρ some λ THE INVERSE POWER METHOD (or INVERSE ITERATION) , for , or (more usually) to

THE ARIMOTO-BLAHUT ALGORITHM FOR COMPUTATION OF CHANNEL CAPACITY. William A. Pearlman. References: S. Arimoto - IEEE Trans. Inform. Thy., Jan.

Section 8.3 Polar Form of Complex Numbers

MATH Homework #2

Homework Assignment 3 Due in class, Thursday October 15

Lecture Notes on Linear Regression

Bernoulli Numbers and Polynomials

HMMT February 2016 February 20, 2016

MATH Sensitivity of Eigenvalue Problems

Note on quantum counting classes

The Geometry of Logit and Probit

Foundations of Arithmetic

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Transcription:

U.C. Berkeley CS278: Computatonal Complexty Handout N8 Professor Luca Trevsan 2/21/2008 Notes for Lecture 8 1 Undrected Connectvty In the undrected s t connectvty problem (abbrevated ST-UCONN) we are gven an undrected graph G = (V, E) and two vertces s, t V, and the queston s whether that s a path between s and t n G. ST-UCONN s complete for the class SL of decson problems that are solvable by symmetrc non-determnstc machnes that use O(log n) space. A non-determnstc machne s symmetrc f whenever t can make a transton from a global state s to a global state s then the transton from s to s s also possble. The proof of SL-completeness of ST- UCONN s dentcal to the proof of NL-completeness of ST-CONN except for the addtonal observaton that the transton graph of a symmetrc machne s undrected. Rengold [Re05] has gven a log-space algorthm for ST-UCONN, thus showng that SL = L. Ths was a great breakthrough, and we wll spend a good amount of tme studyng t and ts context. 2 Randomzed Log-space We now wsh to ntroduce randomzed space-bounded Turng machne. For smplcty, we wll only ntroduce randomzed machnes for solvng decson problems. In addton to a read-only nput tape and a read/wrte work tape, such machnes also have a read-only random tape to whch they have one-way access, meanng that the head on that tape can only more, say, left-to-rght. For every fxed nput and fxed content of the random tape, the machne s completely determnstc, and ether accepts or rejects. For a Turng machne M, an nput x and a content r of the random tape, we denote by M(r, x) the outcome of the computaton. We say that a decson problem L belongs to the class RL (for randomzed log-space) f there s a probablstc Turng machne M that uses O(log n) space on nputs of length n and such that For every content of the random tape and for every nput x, M halts. For every x L, Pr r [M(r, x) accepts ] 1/2 For every x L, Pr r [M(r, x) accepts ] = 0. Notce that the frst property mples that M always runs n polynomal tme. It s easy to observe that any constant bgger than 0 and smaller than 1 could be equvalently used nstead of 1/2 n the defnton above. It also follows from the defnton that L RL NL. The followng result shows that, ndeed, L SL RL NL. Theorem 1 The problem ST-UCONN s n RL. 1

The algorthm s very smple. Gven an undrected graph G = (V, E) and two vertces s, t, t performs a random walk of length 100 n 3 startng from s. If t s never reached, the algorthm rejects. nput: G = (V, E), s, t v s for 1 to 100 E V pck at random a neghbor w of v f w = t then halt and accept v w reject The analyss of the algorthm s based on the fact that f we start a random walk from a vertex s of an undrected vertex G, then each vertex n the connected component of s s lkely to be vsted at least once after O( V E ) steps. As we develop spectral graph theory n the next few lectures, we wll see the proof of the weaker bound O( E 2 ). 3 Egenvalues and expanders We now embark on the study of graph expanson and algebrac graph theory. Wthn the next lecture or two we wll: () know about the equvalence of edge expanson and egenvalue gap, () understand spectral parttonng, () know how to effcently construct a famly of bounded-degree expanders. We wll then return to the queston of the space complexty of ST-UCONN and see how Rengold s algorthm works by reducng ST-UCONN n arbtrary graphs to ST-UCONN n bounded-degree expanders, the latter problem havng an easy log-space solutons. We begn wth the defnton of (normalzed) edge expanson. All graphs that we wll talk about from now on wll be regular. Defnton 1 (Edge-expanson of a graph) Let G = (V, E) be a d-regular, then we defne the normalzed edge expanson of G as h(g) := edges(s, V S) mn S V /2 d S In what follows, we consder G = (V, E) to be a gven d-regular graph and A R V V ts adjacency matrx, that s A(u, v) := number of edges between u and v (1) We denote by M := 1 da the random walk transton matrx of G. The ntuton for ths defnton s that f we take a one-step random walk startng at vertex u, then M(u, v) s the probablty that we reach vertex v. Defnton 2 (Egenvalues and egenvectors) If M C n n, λ C, x C n and xm = λx then λ s an egenvalue of M and x s an egenvector of M for the egenvalue λ. 2

Example 1 Let M be the transton matrx of a regular graph. Then (1, 1,, 1) M = (1, 1,, 1). Therefore, the vector (1, 1,, 1) s an egenvector of M wth correspondng egenvalue 1. Generally, xm = λx x(m λi) = 0 det(m λi) = 0. det(m λi) s a polynomal n λ over C of degree n, and t has n roots (countng multplctes). Therefore, λ s an egenvalue of M ff t s a root of det(m λi) and so, countng multplctes, M has n egenvalues. Theorem 2 If M R n n s symmetrc then the followng propertes hold: 1. all n egenvalues λ 1,, λ n are real 2. one can fnd an orthogonal set of egenvectors x 1,, x n such that x has correspondng egenvalue λ and x x j for j. We note that a multple of an egenvector s also an egenvector and therefore we can assume w.l.o.g. that all the x have length one. From now on we fx the conventon that f we denote by λ 1,..., λ n the egenvalues of M, then λ 1 λ 2 λ n. There are several equvalent characterzatons of the egenvalues of M; the followng one wll be useful. Lemma 3 Let M R n n be symmetrc. Then Proof: λ 1 = max x R n, x =1 xmxt = max x R n xx T (2) (a) Assume λ 1 λ 2, λ n. Then x 1 Mx T 1 = λ 1x 1 x T 1 = λ 1 therefore, max x R n, x =1{ } λ 1. (b) Conversely, let x be any vector of length one, x R n, x = 1. Let x = a 1 x 1 + a 2 x 2 + + a n x n. =,j x()x(j)m(, j) = ( a x )M( a x ) T = ( λ a x )( a x j ) T = j Therefore max x R n, x =1{ } λ 1. λ a 2 max λ a 2 = λ 1 3

We can also prove that λ 2 = For (a) use x = x 2, and conclude max x R n,x x 1 λ xx T 2. For (b) take any x R n, x x 1. Let x = a 2 x 2 + + a n x n. Then max x R n,x x 1 xx T (3) xx T = n =2 λ a 2 n =2 a2 λ 2. A smlar argument shows that max{ λ 2,..., λ n } = max x x 1 xx T (4) We now know enough to characterze the largest egenvalue of the adjacency matrx of a regular graph. Theorem 4 Let G be a regular graph and M ts adjacency matrx. Let λ 1 λ n be ts egenvalues. Then λ 1 = 1. Proof: Trvally, λ 1 1 because 1 s an egenvalue. Let x R n, x = 1, xm = λ 1 x 0 u,v M(u, v)(x(u) x(v)) 2 = 2 v x(v) 2 2 u,v x(u)x(v)m(u, v) Snce d λ 1 and d λ 1 t follows d = λ 1. = 2xx T 2 = 2 2λ 1 1 λ 1 What about λ 2? An mportant theme n ths theory s that the dfference 1 λ 2 characterzes the expanson of the graph. The followng s a smple specal case whch s a good warm-up example. Clam 5 Let G be a regular graph and M ts adjacency matrx. egenvalues. Then λ 2 = 1 f and only f the graph s dsconnected. Let λ 1 λ n ts Proof: Choose x 1 = 1 n (1, 1, 1) and x 2 another egenvector orthogonal to x 1. x 2 should be (x 2 (1),, x 2 (n)) wth x 2() = 0. Therefore, some entres should be postve and some others should be negatve; n partcular, the entres of x 2 are not all equal. 0 u,v M(u, v)(x 2 (u) x 2 (v)) 2 = 2 2λ 2 = 0 Therefore, for x 2 any two adjacent vertces must have dentcal labels and, snce the labels are not all equal, the graph has to be dsconnected. 4

Conversely, f the graph s dsconnected then let S and V S be a partton of the graph that s crossed by no edge. Let p = S V S V, q = V. Assgn { q f v S x(v) = p f v / S Frst, observe that x (1, 1,, 1) snce v x(v) = q S p V S = qpn pqn = 0. Second, look at xm = (dq, dq,, dq, pd, pd,, pd) = dx. }{{}}{{} S V S Therefore, f the graph s dsconnected we have λ 2 = 1. References The defnton of SL s due to Lews and Papadmtrou [LP82]. Pror to Rengold s algorthm [Re05], an algorthm for ST-UCONN was known runnng n polynomal tme and O(log 2 n) space (but the polynomal had very hgh degree), due to Nsan [Ns94]. There was also an algorthm that has O(log 4/3 n) space complexty and superpolynomal tme complexty, due to Armon, Ta-Shma, Nsan and Wgderson [ATSWZ00], mprovng on a prevous algorthm by Nsan, Szemeredy and Wgderson [NSW92]. The best known determnstc smulaton of RL uses O((log n) 3/2 ) space, and s due to Saks and Zhou [SZ95]. References [ATSWZ00] Roy Armon, Amnon Ta-Shma, Av Wgderson, and Shyu Zhou. An o(log(n) 4/3 ) space algorthm for (s, t) connectvty n undrected graphs. Journal of the ACM, 47(2):294 311, 2000. 5 [LP82] Harry R. Lews and Chrstos H. Papadmtrou. Symmetrc space-bounded computaton. Theoretcal Computer Scence, 19:161 187, 1982. 5 [Ns94] N. Nsan. RL SC. Computatonal Complexty, 4(1), 1994. 5 [NSW92] N. Nsan, E. Szemered, and A. Wgderson. Undrected connectvty n O(log 1.5 n) space. In Proceedngs of the 33rd IEEE Symposum on Foundatons of Computer Scence, pages 24 29, 1992. 5 [Re05] [SZ95] Omer Rengold. Undrected ST-connectvty n log-space. In Proceedngs of the 37th ACM Symposum on Theory of Computng, pages 376 385, 2005. 1, 5 M. Saks and S. Zhou. RSPACE(S) DSPACE(S 3/2 ). In Proceedngs of the 36th IEEE Symposum on Foundatons of Computer Scence, pages 344 353, 1995. 5 5