EE611 Deterministic Systems Jordan From, Functions of Square Matrices, Singular Value Decomposition Kevin D. Donohue Electrical and Computer Engineering University of Kentucky Eigenvalues and Eigenvectors A complex (or real) number λ is an eigenvalue of an n by n matrix A iff ∃ a nonzero vector eigenvector x ∋ A x= x A− I x=0 In order for the nullity of A− I to be greater than 0: ∣A− I∣=0 The polynomial resulting from the above determinant operation is call the characteristic polynomial of A and is denoted as: =∣ I−A∣ Examples Find the eigenvectors and values the following matrices: [ A= −4 6 −1 1 [ ] −3 0 1 A= 0.25 −2.75 −0.25 0.25 0.25 −2.25 [ A= 4 13 −2 −6 ] [ ] −2.75 0.25 0.75 A= 0 −3 0 0.25 0.25 −2.25 ] Jordan-Form Representations A Jordan form matrix consist of a diagonal matrix (in the case of distinct eigenvalues) or block diagonal and triangular forms (in the case of repeated eigenvalues.) [ [ ] [ ] ] [ ] 1 0 0 0 0 2 0 0 A= 0 0 3 0 0 0 0 4 1 0 0 0 0 2 0 0 A= 0 0 2 1 0 0 0 2 1 0 0 0 0 2 1 0 A= 0 0 2 1 0 0 0 2 1 1 0 0 0 1 0 0 A= 0 0 2 1 0 0 0 2 Distinct Eigenvalue Example If all eigenvalues are distinct then eigenvectors associated with each eigenvalue are l.i. and can be used form the Q matrix for a similarity transformation to make the matrix diagonal. Find the P and Q matrices to diagonalize the matrix below: [ A= −4 6 −1 1 ] Repeated Eigenvalues If eigenvalues for an n by n matrix are repeated, then the existence of n l.i. eigenvectors is not guaranteed. The generalized eigenvectors can be used in cases where n l.i. eigenvectors cannot be found. A generalized eigenvector of grade n for eigenvalue λ satisfies the following relationships n A− I v=0 A− I n−1 v≠0 Generalized eigenvectors lead to the block-diagonal and triangular Jordan forms. Repeated Eigenvalue Examples Find the P and Q matrices to put the matrices below in Jordan form: [ ] [ ] −3 0 1 A= 0.25 −2.75 −0.25 0.25 0.25 −2.25 −2.75 0.25 0.75 A= 0 −3 0 0.25 0.25 −2.25 Companion-Form Matrices Verify that companion-form matrices have a characteristic equation of =∣ I−A∣=4 1 32 2 3 4 [ 0 1 A= 0 0 0 0 1 0 0 0 0 1 − 4 −3 − 2 −1 ] [ 0 1 0 0 0 1 0 A= 0 0 0 0 1 −4 −3 −2 −1 ] Square Matrix Polynomials If an N by N matrix is diagonal, raising it to a power is simply raising the diagonal elements to the same power and can be decompose into N polynomial equations. If matrix is block diagonal, then powers can be applied to each block and decomposed along the blocks. Apply a similarity transformation to a matrix polynomial and show that Q−1 f A=Q f A Q−1 given A=Q A Minimal Polynomials Given the characteristic polynomial of NM by N matrix: n =∣ I−A∣= N 1 N −1 ... N =∏ − i i=1 where M is the number of diagonal blocks corresponding to an ni multiplicity of eigenvalues. i Define the index of λi, denoted as ni , as the largest order of all Jordan blocks associated with λi. Then the minimal polynomial is M given by: n =∏ −i i i=1 It can be shown that ψ (λ) is the smallest order polynomial such that A=0 Cayley-Hamilton Theorem Given the characteristic polynomial of N by N matrix A: =∣ I−A∣=N 1 N −1 ... N Then A=A N 1 A N−1... N I=0 By noting that minimal polynomial ψ (λ) can be factor out of characteristics polynomial, the above theorem is established. This implies that −A N =1 A N−1 2 A N−2 ... N I and in general any polynomial of A can be expressed in terms of an N linear combination of {I, A, ... , An-1} f A= N −1 A N−1 ...1 A0 I Functions in Terms of Polynomials Given f(λ) and an N by N matrix A with characteristic polynomial: M ni =∏ −i i=1 where M is the number of diagonal blocks corresponding to an ni multiplicity of eigenvalues. Define N-1 order polynomial: N −1 h= N −1 ...1 0 f(Α) can be evaluated with h(Α) if the β coefficients are resolved with the following set of equations: f l i =h l i for l=0,1,...ni −1 and i=1,2,... , M Functions in Terms of Polynomials Given find [ 2 A= 0 −1 −3 ] A10 [ −3 0 1 Given A= 0.25 −2.75 −0.25 0.25 0.25 −2.25 Find exp−2 A ] Singular Values Given m by n real matrix H and M = H'H. Since M is symmetric and semi-positive definite, its eigenvalues are real and positive. Let r be the number of nonzero eigenvalues of M and rank them: 2 1 2 2 2 r ≥ ≥.....≥ 0=...= the λ values are called the singular values of H Given [ A= −1 0 1 2 −1 0 ] Find the singular values of A and A' 2 n Singular Value Decomposition Every m by n matrix H can be decomposed into: H=R S Q ' with R'R = RR' = Im , Q'Q = QQ' = In , and m by n diagonal matrix S with singular values of H on the diagonal. Given [ −3 0 1 A= 1 −1 0 10 10 −2 ] Use the SVD to take the inverse of A