Eigenvalues of a Graph Scott Grayson Adjacency Matrix source: http://www.stoimen.com/blog/2012/08/31/computer-algorithms-graphs-and-their-representation/ Properties of an Adjancency Matrix • Symmetric o • n eigenvalues corresponding to n eigenvectors Zero Trace (sum of the diagonal) o sum of all eigenvalues equals the trace o :. sum of all eigenvalues is zero Eigenvalues of a Graph A = Images: wolframalpha and wikipedia Eigenvalues of a Graph A = To find eigenvalues, solve for k: det( A - k*I ) = 0 *where I is the Identity Images: wolframalpha and wikipedia Eigenvalues of a Graph A = Characteristic polynomial: To find eigenvalues, solve for k: det( A - k*I ) = 0 Eigenvalues: k = -1, -1, 2 *where I is the Identity Images: wolframalpha and wikipedia More on EigenValues of A • • The term “spectra” is used to describe the eigenvalues, eigenvectors and characteristic polynomial of the graph Non isomorphic graphs with the same spectra are called “co-spectral” o Co-spectral Trees are common Co Spectral Trees Example • • These trees are non-isomorphic, but co-spectral. o Characteristic polynomial: “As n -> infinity, almost no trees are uniquely determines by their spectra” Images: “Introduction to Graph Theory” by West Laplacian Matrix • L=D-A o L is the Laplacian matrix o A is the adjacency matrix o D is the degree matrix diagonal matrix containing the degree of each vertex Image: Wikipedia Properties of the Laplacian Spectrum • • • Eigenvalues will range between zero and 2 The smallest eigenvalue of L is zero If G is connected, the eigenvalue zero has multiplicity 1 o • if multiplicity > 1 this tells us how many connected components the graph has If the largest eigenvalue is 2, G has a bipartite component Part of a Lecture by Luca Trevisan http://youtu.be/iu6EX9Xt3gA?t=7m53s Applications • • Minimization for other graph problems o ex. coloring Examining connectivity in networks o Google PageRank algorithm o Recommendations (music, movies friends) PageRank • • • • • Developed in 1996 by Larry Page and Sergey Brin at Stanford old method: “text ranking” PageRank attempts to model a person randomly clicking links Viewed as an eigenvalue problem Adjacency matrix for links between web pages o Values between 0 and 1 PageRank R = PageRank vector M = adjacency matrix d = damping factor N = number of websites • • Requires multiple passes o recursive o some links are more important than others Damping factor o about 85% of links are self links History • 1980 “Spectra of Graphs” by Cvetković, Doob, and Sachs o • o 2nd edition in 1988 3rd edition in 1995 Some other research came from the quantum chemistry field References • • • • • • Brouwer, Andries E., and Willem H. Haemers. "The Spectra of Graphs." N.p., n.d. Web. 2 Apr. 2014. <http://www.win.tue.nl/~aeb/2WF02/spectra.pdf>. Chung, Fan. "Eigenvalues and the Laplacian of a Graph." N.p., n.d. Web. <http://www.math.ucsd.edu/~fan/research/cb/ch1.pdf>. Fox, Jacob. "Spectral Graph Theory." N.p., n.d. Web. <http://math.mit.edu/~fox/MAT307-lecture18.pdf>. "Lecture #3: PageRank Algorithm - The Mathematics of Google Search." PageRank Algorithm. N.p., n.d. Web. 02 Apr. 2014. <http://www.math.cornell.edu/~mec/Winter2009/RalucaRemus/Lecture3/lecture3.html>. Lovasz, Laszlo. "Eigenvalues of Graphs." N.p., n.d. Web. 2 Apr. 2014. <http://www.cs.elte.hu/~lovasz/eigenvalsx.pdf>. Spielman, Daniel. "The Laplacian." N.p., n.d. Web. <http://www.cs.yale.edu/homes/spielman/561/2009/lect0209.pdf>. • West, Douglas Brent. Introduction to Graph Theory. Upper Saddle River, NJ: Prentice Hall, 2001. Print. • Wilf, H. S. "Eigenvalues of a Graph and Its Chromatic Number." N.p., n.d. Web. HW 1 Find the eigenvalues of the Laplacian of this graph: Image: Wikipedia HW 2 Prove or disprove: • If k vertices have identical neighborhoods. Then zero is an eigenvalue with multiplicity at least k-1 * this question refers to the eigenvalues of the adjacency matrix. Not Laplacian