# hw5solns - 15-750 — Graduate Algorithms — Spring...

This preview shows pages 1–3. Sign up to view the full content.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 15-750 — Graduate Algorithms — Spring 2009Miller and Dinitz and TangwongsanAssignment 5 Solutions1Something about Laplacian(25 pts.)The Laplacian of a graph encodes a number of interesting properties of the graph. In this problem,we will examine some (random) facts about graph Laplacians.a)Positive Semi-definiteness.Ann×nreal symmetric matrixAispositive semi-definiteif forallx∈Rn,x>Ax≥0. We know a lot of nice properties of such matrices from the theory oflinear algebra. For instance, it is known that the following statements are equivalent: 1) alleigenvalues ofAare non-negative, 2) for allx∈Rn,x>Ax≥0, and 3) there exists a matrixBsuch thatBTB=A. Indeed, an example of positive semi-definite matrices is the Laplacianof a undirected graph.Using any of the equivalent definitions above, prove that ifG= (V,E) is a simple undirectedgraph with weights given byw:V×V→R+∪ {}, thenL(G) is positive semi-definite. Weadopt the convention thatw(i,j) = 0 if (i,j)6∈E.Solution:LetA= [ai,j]n×n=L(G), andx∈Rnbe given. Consider that thei-th row ofAx,denoted by(Ax)i, is∑nj=1ai,j·xj, sox>Ax=nXi=1xi·(Ax)i=nXi=1nXj=1ai,jxixj.Now sinceA=L(G)is symmetric andai,i=∑j6=iw(i,j) =∑j6=iw(j,i), we havex>Ax=nXi=1ai,ix2i+ 2Xi<jai,jxixj=nXi=1Xj6=iw(i,j)x2i-2Xi<jw(i,j)xixj=Xi<jw(i,j)(x2i-2xixj+x2j) =Xi<jw(i,j)(xi-xj)2≥,which proves thatL(G)is positive semi-definite.b)Max-Cut.In theMax-Cutproblem, we are given a weighted undirected graphG=(V,E,w), and we are interested in finding a partition ofVintoSandS=V\Ssuchthat∑(i,j)∈S×Sw(i,j) is maximized. This corresponds to maximizing the number of edgescrossing the cut in the unweighted version. LetMax-Cut(G) denote the value of the best cutonG. That is,Max-Cut(G) := maxS⊆VX(i,j)∈S×Sw(i,j).Assume that all edge weights are non-negative. Show thatMax-Cut(G) =maxx∈{-1,1}n14x>L(G)x.15-750 HW 5 Solutions2Solution:For a cut(S,V\S), defineval(S,V\S) =X(i,j)∈S×(V\S)w(i,j).In part (a), we established thatx>L(G)x=∑i<jw(i,j)(xi-xj)2for anyx∈Rn. Specifically,ifx∈ {-1,1}n, we havex>L(G)x= 4Xi<jxi6=xjw(i,j),which suggests the following natural bijection between cuts and±1vectors: thei-th componentof the vectorx∈ {-1,1}nis1if and only ifibelongs to the setSin the cut(S,V\S). As such,ifxis the vector corresponding to(S,V\S), then14x>L(G)x=Xi<jxi6=xjw(i,j) =val(S,V\S)Since every cut has a corresponding±1vector and vice versa, we conclude1that Max-Cut(G) =maxx∈{-1,1}n14x>L(G)x.2Randomizeds-tConnectivity(25 pts.)Suppose we’re given an undirected graphG= (V,E) and two nodessandtinG, and we want todetermine if there is a path connectingsandt. As we have seen in class, this is easily accomplishedinO(|V|+|E|) using DFS or BFS. The space usage of such algorithms, however, isO(|V|)....
View Full Document

## This document was uploaded on 11/03/2009.

### Page1 / 5

hw5solns - 15-750 — Graduate Algorithms — Spring...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online