Sharif Univ. of Tech. EE. Dept.
Generalized Eigenvalue Problem A l i & Applications Analysis A li ti
Mohammad M. Alem-Karladani
Summer 2008
Outline
Introduction Applications Definitions
Infinite I fi it Eigenvalue Ei l Singularity & Regularity Symmetry
Symmetric Definite Quadratic Eigenvalue Problem Solution Method
Introduction
Standard Eigenvalue Problem
Ax = x
x0
det ( A I ) = 0
Generalized Eigenvalue Problem
Ax = Bx
x0
det ( A B ) = 0
Applications
Mechanical Vibration
o
Newtons Laws
k1
it y = x e 1 1 it y x e = 2 2
d 2 y1 (k1 + k2 ) y1 + k2 y2 = m1 2 d dt 2 d y2 k2 y1 k2 y2 = m2 dt 2
m1 y1 k2
k1 + k2 k 2
k2 x1 2 m1 = k2 x2 0
0 x1 m2 x2
Kx = Mx
o
m2
Stiffness Matrix K , Mass Matrix
y2
Applications
Electrical Oscillation
o
Kirchhoffs Laws
v1 C1
L1
v2 C2 L2
dv1 1 = v v dt C ( ) 2 1 1 L1 dt 1 (v v )dt 1 v dt = C dv2 1 2 2 2 L L dt 1 2
1 d 2 v1 (v2 v1 ) = C1 2 dt L1 d 2 v2 1 1 L (v1 v2 ) L v2 = C2 dt 2 1 2
it v = x e 1 1 it v x e = 2 2
L11 1 L1
1 L1 0 x1 x1 2 C1 = x 1 1 x 0 C L1 + L2 2 2 2
Ax = Bx
Definitions
Generalized Eigenvalue Problem for a Matrix Pair
( Ann , Bnn )
Ax = Bx
Eigenvalue of the Pair
( A, B) :
x0
Eigenvector of the Pair ( A, B ) : S d d Ei Standard Eigenvalue l Problem: P bl
x
( A, B) : det ( A B ) = 0
B=I
Characteristic Equation of the Pair Matrix Pencil:
A B
Infinity of Eigenvalue
Characteristic Equation det ( A B ) or less. 1 1 0 0 Exp. A = B=
0 1 0 1
= 0 is of degree n
1 d t ( A B ) = (1 ) = 0 1 = 1 x1 = det 1
Infinite Eigenvalue
1 x2 = 0
0 = Bx =
Ax
2 =
Theorem:
is an e.v. of ( A, B) iff det( B) = 0
d t( B) 0 det( det( d t( A B) = 0 det( d t( B 1 A I ) = 0
Singularity vs. Regularity
x 0 , A x = B x = 0 ( A B ) x = 0 det( ( A B) 0
Singular The pair ( A, B ) is called Regular
det ( A B ) 0
Otherwise
Theorem: If the null spaces of A and B have nontrivial intersection, then ( A, B ) is a singular pair. C Counterexample l to the h reverse:
0 1 0 0 0 1 B= 0 0 0 0 ) N ( A) = span p ( 1 0 1 ) N ( B) = span p ( 0 0
) ( , x = 1 0
1 0 0 0 0 0 A= 0 0 1
N ( A) N ( B) = {0}
1 det ( A B ) = 0 0 0 0
0 0 1
Symmetric GEVP
Eigenvalues are real. Standard EVP: AT = A Eigenvectors are orthonormal Generalized EVP: Is symmetry of A and B is sufficient? Counterexample: 1 2 1 1 A= B= 2 3 1 0.5 2+ 1 + 2 = + +1 = 0 det ( A B) = 0 2 + 3 + / 2
The product of two symmetric matrices is not necessarily symmetric. Ax = Bx B 1 Ax = x
Symmetric Definite GEVP
Definition: A is symmetric and B is positive definite. Theorem: In Symmetric Definite GEVP, GEVP Eigenvalues are real. Eigenvectors are B-orthonormal
Cholesky Decomposition
Proof: B is P.D.
B = LLT
Ax = Bx Ax = LLT x L1 A( LT ) 1 LT x = LT x C = L1 A( LT ) 1 , y = LT x Cy = y
CT = C
T T T T T T yT y = ( L x ) L x = x LL x = x i j i j i j i Bx j = ij
Quadratic Eigenvalue Problem
In many applications: Steady-state solution:
d 2y dt 2
+D
dy + Ky = 0 dt
y = x e t
Quadratic Eigenvalue Problem: ( 2 M + D + K )x = 0 Standard Nonsymmetric Eigenvalue Problem:
M 1 D M 1 K x x = 0 x x I
Generalized Symmetric Eigenvalue Problem: D K x M 0 x K 0 x = 0 x K
Solution Method
Definition: Two pairs ( A, B ) and ( A, B ) are strictly unitarily equivalent q if there exist unitary y matrices U and V such that:
A = UAV
B = UBV
Corollary: Two strictly unitarily equivalent pairs have the same characteristic equations. det ( A B ) = 0 det ( A B ) = 0 Generalized Schur Theorem: For every pair ( A, B ) there exits unitary matrices Q , Z and upper triangular matrices T , S such that:
A = QTZ
B = QSZ
det (T S ) =
(t
i =1
ii
sii ) = 0
tii i = sii
