Class 20: Spectral Graph Theory

Class 20: Spectral Graph Theory

2/23/2017 Ma/CS 6b Class 20: Spectral Graph Theory By Adam Sheffer Recall: Parity of a Permutation 푆푛 – the set of permutations of 1,2, … , 푛 . A permutation 휎 ∈ 푆푛 is even if it can be written as a composition of an even number of transpositions. Otherwise, 휎 is odd. 1, 휎 is even 푠푔푛 휎 = ቊ −1, 휎 is odd 1 2/23/2017 Determinants 푆푛 – the set of permutations of 1,2, … , 푛 . 푀 – an 푛 × 푛 matrix. 푚푖푗 – 푗’th cell in 푖’th row of 푀. The determinant of 푀 is det 푀 = 푀 = ෍ 푚1,휎(1) ⋯ 푚푛,휎 푛 ⋅ 푠푔푛 휎 휎∈푆푛 Simple Example det 푀 = 푀 = ෍ 푚1,휎(1) ⋯ 푚푛,휎 푛 ⋅ 푠푔푛 휎 휎∈푆푛 푎 푏 = 푎푑 − 푏푐. 푐 푑 2 2/23/2017 Eigenvalues and Eigenvectors 퐴 – an 푛 × 푛 matrix of real numbers. The eigenvalues of 퐴 are the numbers 휆 such that 퐴푥 = 휆푥 for some nonzero vector 푥 ∈ ℝ푛. ◦ A vector 푥 that satisfies 퐴푥 = 휆푥 for some eigenvalue 휆 is an eigenvector of 퐴. 2 1 Example. 퐴 = . 1 2 ◦ Eigenvalue 휆1 = 3 with eigenvectors of the form 푎, 푎 . ◦ Eigenvalue 휆2 = 1 with eigenvalues of the form 푎, −푎 . Characteristic Polynomial 퐴 – an 푛 × 푛 matrix of real numbers. The characteristic polynomial of 퐴 is 휙 퐴; 휆 = det 휆퐼 − 퐴 . ◦ The roots of 휙 퐴; 휆 are exactly the eigenvalues of 퐴. ◦ The algebraic multiplicity of an eigenvalue 휆푖 is the multiplicity of 휆푖 in 휙 퐴; 휆 . ◦ (Do not confuse with the geometric multiplicity, which is the dimension of the space of eigenvectors associated with 휆). 3 2/23/2017 Example: Multiplicities 2 0 0 0 퐴 = 1 2 0 0 0 1 3 0 0 0 1 3 휆 − 2 0 0 0 det 휆퐼 − 퐴 = det −1 휆 − 2 0 0 0 −1 휆 − 3 0 0 0 −1 휆 − 3 = 휆 − 2 2 휆 − 3 2. So we have the eigenvalues 휆1 = 2 and 휆2 = 3, each with (algebraic) multiplicity 2. Recall: Adjacency Matrix Consider a graph 퐺 = 푉, 퐸 . ◦ We order the vertices as 푉 = 푣1, 푣2, … , 푣푛 . ◦ The adjacency matrix of 퐺 is a symmetric 푛 × 푛 matrix 퐴. The cell 퐴푖푗 contains the number of edges between 푣푖 and 푣푗. 푣1 0 1 0 0 1 1 0 1 0 1 푣 푣 퐴 = 0 1 0 1 0 5 2 0 0 1 0 1 1 1 0 1 0 푣4 푣3 4 2/23/2017 The Spectrum of a Graph Consider a graph 퐺 = 푉, 퐸 and let 퐴 be the adjacency matrix of 퐺. ◦ The eigenvalues of 퐺 are the eigenvalues of 퐴. ◦ The characteristic polynomial 휙 퐺; 휆 is the characteristic polynomial of 퐴. ◦ The spectrum of 퐺 is 휆 , … , 휆 푠푝푒푐 퐺 = 1 푡 , 푚1, … , 푚푡 where 휆1, … , 휆푡 are the distinct eigenvalues of 퐴 and 푚푖 is the multiplicity of 휆푖. Example: Spectrum 푣1 푣2 0 1 0 1 퐴 = 1 0 1 0 0 1 0 1 1 0 1 0 푣4 푣3 휆 −1 0 −1 det 휆퐼 − 퐴 = det −1 휆 −1 0 0 −1 휆 −1 −1 0 −1 휆 = 휆2 휆 − 2 휆 + 2 . 0 2 −2 푠푝푒푐 퐶 = 4 2 1 1 5 2/23/2017 Spectral Graph Theory Spectral graph theory is the study of properties of a graph in relationship to the characteristic polynomial, eigenvalues, and eigenvectors of matrices associated with the graph, such as its adjacency matrix. ◦ Beyond being useful in graph theory, it is also used in research in quantum chemistry. Slight Change of Notation Instead of multiplicities, let 휆1, … , 휆푛 be the not necessarily distinct eigenvalues of 퐺. 2 −1 For example, if the spectrum is , 2 2 we write 휆1 = 휆2 = 2 and 휆3 = 휆4 = −1 (instead of 휆1 = 2, 푚1 = 2, 휆2 = −1, 푚2 = 2). 6 2/23/2017 The Spectral Theorem Every adjacency matrix 퐴 is symmetric and real. Theorem. Any real symmetric 푛 × 푛 matrix has real eigenvalues and 푛 orthonormal eigenvectors. ◦ The algebraic and geometric multiplicities are the same in this case. 푛 ◦ We have 휙 퐴; 휆 = ς푖=1 휆 − 휆푖 . ◦ The multiplicity of an eigenvalue 휆 is 푛 − 푟푎푛푘 휆퐼 − 퐴 . Changing the Diagonal 퐴 – an 푛 × 푛 real symmetric matrix. 푐 – a real constant. Adding 푐 to each cell in the diagonal of 퐴 increases each of the eigenvalues of 퐴 by 푐. ◦ Since 훼 + 푐 is a root of det 휆퐼 − (푐퐼 + 퐴) if and only if 훼 is a root of det(휆퐼 − 퐴). 7 2/23/2017 The Spectrum of a Complete Graph Problem. What is the spectrum of 퐾푛? ◦ 1푛×푛 – the all 1’s 푛 × 푛 matrix. ◦ The adjacency matrix is 퐴 = 1푛×푛 − 퐼. ◦ What is the spectrum of 1푛×푛? 푛 0 . 1 푛 − 1 (Eigenvectors 1,1, … , 1 , 1, −1,0, … , 0 , etc.) ◦ Using the argument from the previous slide, we have 푛 − 1 −1 푠푝푒푐 퐾 = . 푛 1 푛 − 1 Trace 퐴 – 푛 × 푛 adjacency matrix of a graph 퐺. 푛 The trace of 퐴 is 푡푟푎푐푒 퐴 = σ푖=1 퐴푖푖. In 휙 퐴; 휆 = det 휆퐼 − 퐴 , the coefficient of 휆푛−1 is −푡푟푎푐푒 퐴 . 푛 Since 휙 퐴; 휆 = ς푖=1 휆 − 휆푖 , we have 푛 푡푟푎푐푒 퐴 = σ푖=1 휆푖 Since 퐺 is simple, we get 푛 푡푟푎푐푒 퐴 = ෍ 퐴푖푖 = 0. 푖=1 푛 ◦ Thus, we have σ푖=1 휆푖 = 0. 8 2/23/2017 Complete Bipartite Graphs Problem. What is the spectrum of 퐾푚,푛? 0 1 ◦ We have 퐴 = 푚×푚 푚×푛 . 1푛×푚 0푛×푛 ◦ We have 푟푎푛푘 퐴 = 2 so the multiplicity of 휆 = 0 is 푚 + 푛 − 푟푎푛푘 0 ⋅ 퐼 − 퐴 = 푚 + 푛 − 2. ◦ Denote the remaining eigenvalues as 휆2 and 휆3. We have 0 = 푡푟푎푐푒 퐴 = 휆2 + 휆3. Searching for 휆2 We know that the spectrum of 퐾푚,푛 is 0 휆 −휆 2 2 . 푚 + 푛 − 2 1 1 ◦ We thus have 푚+푛 푚+푛−2 2 2 휙 퐾푚,푛; 휆 = ෑ 휆 − 휆푖 = 휆 휆 − 휆2 푖=1 0 1 ◦ Recall that 퐴 = 푚×푚 푚×푛 . 1푛×푚 0푛×푛 ◦ Thus the coefficient of 휆푚+푛−2 in det 휆퐼 − 퐴 is − σ푖≠푗 퐴푖푗퐴푗푖 = −푚푛. 9 2/23/2017 Finding 휆2 푚+푛 푚+푛 2 푚+푛−2 휙 퐺; 휆 = ෑ 휆 − 휆푖 = 휆 − 휆2휆 . 푖=1 The coefficient of 휆푚+푛−2 in det 휆퐼 − 퐴 is − σ푖≠푗 퐴푖푗퐴푗푖 = −푚푛. So the spectrum of 퐾푚,푛 is 0 푚푛 − 푚푛 . 푚 + 푛 − 2 1 1 Which Famous American Serial Killer is Related to Mathematics Charles Manson The Unabomber Ted Bundy 10 2/23/2017 Recall: Minors A minor of a matrix 퐴 is the determinant of a square sub-matrix, cut down from 퐴 by removing one or more of its rows or columns. A 푘 × 푘 minor is the determinant of a 푘 × 푘 submatrix of 퐴. The Coefficient of 휆푛−2 퐴 – 푛 × 푛 adjacency matrix of a (simple) graph 퐺 = (푉, 퐸). ◦ The leading coefficient of 휙 퐺; 휆 is 1. ◦ The coefficient of 휆푛−1 is −푡푟푎푐푒 퐴 = 0. What is the coefficient of 휆푛−2? ◦ As in the example of 퐾푚,푛, it is − σ푖≠푗 퐴푖푗퐴푗푖 (sum of the 2 × 2 minors with the same row and column indices). ◦ Such a minor is -1 if 푣푖, 푣푗 ∈ 퐸, and 0 otherwise. ◦ Therefore, the coefficient of 휆푛−2 is −|퐸|. 11 2/23/2017 퐴 – 푛 × 푛 adjacency matrix of a graph 퐺. What is the coefficient of 휆푛−3 in 휙 퐺; 휆 ? ◦ It is the sum of the 3 × 3 minors with the same row and column indices). ◦ Such a 3 × 3 minor is nonzero iff for the corresponding indices 푖, 푗, 푘 we have that 푣푖, 푣푗, 푣푘 form a cycle of length three. Then 0 푒푖푗 푒푖푘 0 −1 −1 det 푒푖푗 0 푒푗푘 = det −1 0 −1 푒푖푘 푒푗푘 0 −1 −1 0 = −2. ◦ Thus, the coefficient of 휆푛−3 is −2 times the number of copies of 퐾3 in 퐺. A Generalization 퐴 – 푛 × 푛 adjacency matrix of a graph 퐺. The coefficient of 휆푛−푖 in 휙 퐺; 휆 is the sum of the 푖 × 푖 minors of 퐴 with the same row and column indices (times −1 푖). 12 2/23/2017 Powers of 퐴 퐴 – 푛 × 푛 adjacency matrix of a graph 퐺. Recall that cell 푖, 푗 of 퐴푘 is the number of paths of length 푘 between 푣푖 and 푣푗. If 휆 is an eigenvalue of 퐴, what do we know about 퐴푘? ◦ 푣 – an eigenvector of 휆. ◦ 퐴푘푣 = 휆퐴푘−1푣 = 휆2퐴푘−2푣 = ⋯ = 휆푘푣. ◦ So 휆푘 is an eigenvalue of 퐴푘, with corresponding eigenvector 푣. Adding Isolated Vertices 퐴 – 푛 × 푛 adjacency matrix of a graph 퐺. What happens to the spectrum of 퐺 when adding an isolated vertex? ◦ We add a row and a column of zeros to 퐴. ◦ Recall that the multiplicity of an eigenvalue 휆 is 푛 − 푟푎푛푘 휆퐼 − 퐴 . ◦ Specifically, the multiplicity of the eigenvalue 0 is 푛 − 푟푎푛푘(퐴). ◦ Since adding the new row and column does not change 푟푎푛푘(퐴), this increases the multiplicity of 0 by 1. 13 2/23/2017 Eigenvalues of Bipartite Graphs Lemma. Let 퐺 be a bipartite graph and let 휆 be a nonzero eigenvalue of 퐺 with multiplicity 푚. Then −휆 is also an eigenvalue of 퐺 with multiplicity 푚. 푣1 푣2 푣4 푣3 Example: Spectrum 푣1 푣2 0 1 0 1 퐴 = 1 0 1 0 0 1 0 1 1 0 1 0 푣4 푣3 휆 −1 0 −1 det 휆퐼 − 퐴 = det −1 휆 −1 0 0 −1 휆 −1 −1 0 −1 휆 = 휆2 휆 − 2 휆 + 2 .

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    17 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us