0% found this document useful (0 votes)
133 views35 pages

Eigenvalue Problem PDF

The document discusses the eigenvalue problem and methods for solving it. It begins by defining a homogeneous linear system and how the existence of nontrivial solutions depends on the determinant. It then provides an example of finding nontrivial solutions. Next, it defines eigenvalues and eigenvectors, and describes analytical and numerical methods for solving the eigenvalue problem including the power method. It concludes by providing an example of using the power method.

Uploaded by

Mikhail Tabucal
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
133 views35 pages

Eigenvalue Problem PDF

The document discusses the eigenvalue problem and methods for solving it. It begins by defining a homogeneous linear system and how the existence of nontrivial solutions depends on the determinant. It then provides an example of finding nontrivial solutions. Next, it defines eigenvalues and eigenvectors, and describes analytical and numerical methods for solving the eigenvalue problem including the power method. It concludes by providing an example of using the power method.

Uploaded by

Mikhail Tabucal
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 35

The Eigenvalue Problem

Homogeneous Systems
•  A homogeneous linear system

Ax=0
•  If det(A)≠0, the unique solution is the trivial
solution x = 0.
•  If det(A)=0, there exist nontrivial solutions.
Example 1
Find the nontrivial solutions to the
homogeneous system

x1 + 2 x2 − x3 = 0
2 x1 + x2 + x3 = 0
5 x1 + 4 x2 + x3 = 0
Solution 1
The system reduces to:
x1 + 2x2 − x3 = 0
x1 + x2 =0
2x1 + x2 + x3 = 0
− x 2 + x3 = 0
5x1 + 4x2 + x3 = 0
Letting x3=t, the solution can be expressed as
x1 = −t ⎧− 1⎫
x2 = t ⎪ ⎪
X = t ⎨ 1 ⎬
x3 = t ⎪ 1 ⎪
⎩ ⎭
Applications
•  Quantum mechanics, geology, mathematics,
image processing, etc.

•  Mechanics
Principal stresses and the orientation of the
principal planes

•  Structural Dynamics
Natural frequencies and mode shapes of
structures
The Eigenvalue Problem
Let A be a square matrix of dimension n x n, and
let x be a vector of dimension n. The problem is
to find scalars λ for which there exists a
nonzero vector x such that

Ax = λ x
Eigenvector Eigenvalue
(characteristic vector) (characteristic value)
Solution Methods
•  Analytical Solution
–  Solution of the characteristic equation

•  Numerical Solution
–  Power method
–  Inverse power method
–  Jacobi’s method, QR method
Analytical Method
•  The eigenvalues of A are the solutions of the
(
characteristic equation A − λI x = 0 )
a11 − λ a12 ! a1n
a21 a22 − λ ! a2 n
det ( A − λI ) = =0
" " # "
an1 an 2 ! ann − λ

•  When the determinant is expanded, it becomes


a polynomial of degree n, called the
characteristic polynomial.
Eigenvalues
If A is an n x n real matrix, then its n eigenvalues
λ1, λ2, …, λn. are the real and complex roots of
the characteristic polynomial

p(λ ) = det ( A − λI )
Properties:
n
p(λ ) = (− 1) (λ − λ1 )(λ − λ2 )!(λ − λn )
n n n
trace( A) = ∑a = ∑λii k det ( A) = ∏λ k
i =1 k =1 k =1
Eigenvector
If λ is an eigenvalue of A and the nonzero vector v
has the property that

Av = λ v
Then v is called the eigenvector of A
corresponding to the eigenvalue λ. v is obtained
by solving the homogeneous system (A - λ I) v =
0.
Analytical Method
Hand computations used to solve the eigenvalue
problem when the dimension n is small.

1.  Find the coefficients of the characteristic


polynomial p(λ)= det(A - λ I)

2.  Find its roots (eigenvalues).

3.  Find the nonzero solutions (eigenvectors) of


the homogeneous linear system (A - λ I) v = 0.
Linear Dependence
Definition 1. The vectors v1, v2, …, vr are said to
be linearly independent if

c1v1 + c2v2 + … + cnvn = 0

implies that c1=0, c2=0, …, cn=0. The vectors


are linearly dependent if there exists a set of
numbers {c1, c2, …, cn} not all zero, such that the
equation above holds.
Theorems
Theorem 1. For each distinct eigenvalue λ there
exists at least one eigenvector v corresponding
to λ. If λ has multiplicity r, then there exists at
most r linearly independent eigenvectors v1, v2,
…, vr that correspond to λ.

Theorem 2. Suppose that A is a square matrix and


λ1, λ2, …, λk are distinct eigenvalues of A, with
associated eigenvectors v1, v2, …, vk,
respectively; then {v1, v2, …, vk} is a set of
linearly independent vectors.
Example 2
Find the eigenpairs λj, vj for the matrix

⎡ 3 − 1 0 ⎤
⎢
A = ⎢− 1 2 − 1⎥⎥
⎢⎣ 0 − 1 3 ⎥⎦
Solution 2
Eigenvalues. The characteristic equation det(A - λ I)=0 is

3−λ −1 0
3 2
−1 2−λ − 1 = −λ + 8λ − 19λ + 12 = 0
0 −1 3 − λ − (λ − 1)(λ − 3)(λ − 4) = 0

Therefore, the three eigenvalues are

λ1 = 1, λ2 = 3, and λ3 = 4.
Solution 2
Eigenvectors.
For λ1 = 1,
⎡ 2 − 1 0 ⎤ ⎧ x1 ⎫ ⎧0⎫ x2 = 2a ⎧1 ⎫
⎢− 12 x11− x2− 1⎥ ⎪ x ⎪==0⎪0⎪ ⎪ ⎪
⎢ ⎥ ⎨ 2 ⎬ ⎨ ⎬ x1 = a v1 = a ⎨2⎬
− x + 2x = 0
⎢⎣ 0 − 1 22 ⎥⎦ ⎪⎩ x33⎪⎭ ⎪⎩0⎪⎭ x3 = a ⎪1 ⎪
⎩ ⎭
For λ2 = 3,
⎡ 0 − 1 0 ⎤ ⎧ x1 ⎫ ⎧0⎫ x2 = 0 ⎧ 1 ⎫
⎢− 1 x1 +⎥x⎪3 =⎪0 ⎪ ⎪ ⎪ ⎪
⎢ − 1 − 1⎥ ⎨ x2 ⎬ = ⎨0⎬ x1 = b v 2 = b ⎨ 0 ⎬
x2 ⎪ = ⎪0 ⎪ ⎪ ⎪− 1⎪
⎢⎣ 0 − 1 0 ⎦ ⎩ x3 ⎭ ⎩0⎭
⎥ x3 = − b ⎩ ⎭
Solution 2
Eigenvectors.
For λ3 = 4,
⎡− 1 − 1 0 ⎤ ⎧ x1 ⎫ ⎧0⎫ x3 = c ⎧ 1 ⎫
⎢− 1 x1 + x2 ⎥ ⎪ = ⎪0 ⎪ ⎪ ⎪ ⎪
⎢ − 2 − 1⎥ ⎨ x2 ⎬ = ⎨0⎬ x2 = −c v 2 = c ⎨− 1⎬
x2 + ⎪x3 =⎪0 ⎪ ⎪ ⎪ 1 ⎪
⎢⎣ 0 − 1 − 1⎥⎦ ⎩ x3 ⎭ ⎩0⎭ x1 = c ⎩ ⎭
Numerical Methods

1.  Power Method


2.  Shifted-inverse Power Method
Power Method
Assume that the n x n matrix A has n distinct
eigenvalues λ1, λ2, …, λn and that they are
ordered in decreasing magnitude; that is,

|λ1| > |λ2| ≥ |λ3| ≥ … ≥ |λn|

eigenvector v1 corresponding to λ1 is called a


dominant eigenvector.
Power Method
If x0 is chosen appropriately, then the sequences
{xk = [x1(k) x2(k) … xn(k) ]’} and {ck} generated
recursively by
yk = A xk
and
xk+1 = (1/ck+1) yk,
where
ck+1 = max{|xi(k)|
will converge to the dominant eigenvector v1 and
eigenvalue λ1, respectively.
Example 3
Use the power method to find the dominant
eigenpair for the matrix

⎡ 0 11 − 5 ⎤
⎢
A = ⎢− 2 17 − 7 ⎥⎥
⎢⎣− 4 26 − 10⎥⎦
Solution 3
Start with x0 = {1 1 1}T and use to generate the
sequence of vectors {xk} and constants {ck}.
The first iteration produces

⎡ 0 11 − 5 ⎤ ⎧1⎫ ⎧ 6 ⎫ ⎧ 12 ⎫
⎢− 2 17 − 7 ⎥ ⎪1⎪ = ⎪ 8 ⎪ = 12⎪ 2 ⎪ = c x
⎢ ⎥ ⎨ ⎬ ⎨ ⎬ ⎨ 3 ⎬ 1 1
⎢⎣− 4 26 − 10⎥⎦ ⎪⎩1⎪⎭ ⎪⎩12⎪⎭ ⎪1 ⎪
⎩ ⎭
Solution 3
The second iteration produces
⎡ 0 11 − 5 ⎤ ⎧ 12 ⎫ ⎧ 73 ⎫ ⎧167 ⎫
⎢− 2 17 − 7 ⎥ ⎪ 2 ⎪ = ⎪10 ⎪ = 16 ⎪ 5 ⎪ = c x
⎢ ⎥ ⎨ 3 ⎬ ⎨ 3 ⎬ 3 ⎨ 8 ⎬ 2 2
⎢⎣− 4 26 − 10⎥⎦ ⎪⎩1 ⎪⎭ ⎪⎩163 ⎪⎭ ⎪ 1 ⎪
⎩ ⎭
Iteration generates the sequence
⎧ 12 ⎫ ⎧167 ⎫ ⎧125 ⎫ 31
⎧ 76 ⎫ 21
⎧ 52 ⎫
⎪ 2 ⎪ 16 ⎪ 5 ⎪ 9 ⎪ 11 ⎪ 38 ⎪ 23 ⎪ 78 ⎪ 47 ⎪
12⎨ 3 ⎬, ⎨ 8 ⎬, ⎨18 ⎬, ⎨ 38 ⎬, ⎨ 78 ⎬,!
⎪1 ⎪ 3 ⎪ 1 ⎪ 2 ⎪ ⎪ 9 ⎪ ⎪ 19 ⎪ ⎪
⎩ ⎭ ⎩ ⎭ ⎩ 1 ⎭ ⎩ 1 ⎭ ⎩ 1 ⎭
The sequence converges to v = {2/5 3/5 1}T, λ=4.
C-Implementation
Vector PowerMethod(
Matrix A, /* nxn matrix A */
Vector x0, /* nx1 starting vector */
double e, /* tolerance */
int N, /* maximum iterations */
double * ev /* eigenvalue */
);

Returns: The dominant eigenvector


Shifted-inverse Power Method
Shifting Eigenvalues. Suppose that λ, v is an
eigenpair of A. If α is any constant, then λ - α, v
is an eigenpair of the matrix A - αI.

Inverse Eigenvalues. Suppose that λ, v is an


eigenpair of A. If λ≠0, then 1/λ, v is an
eigenpair of the matrix A-1.

Suppose that λ, v is an eigenpair of A. If α≠λ, then


1/(λ - α), v is an eigenpair of the matrix (A-αI)-1.
Shifted-inverse Power Method
A constant α can be chosen so that µ1 = 1/(λi - α)
is the dominant eigenvalue of (A - αI)-1.
Furthermore, if x0 is chosen appropriately, then
the sequences {xk = [x1(k) x2(k) … xn(k) ]’} and {ck}
generated recursively by
yk = (A - αI)-1 xk
and
xk+1 = (1/ck+1) yk,
where
ck+1 = max{|xi(k)|}
Shifted-inverse Power Method
will converge to the dominant eigenpair of the
matrix (A - αI)-1. Finally, the corresponding
eigenvalue for the matrix A is given by the
calculation
1
λj = +α
µ1
Example 4
Employ the shifted-inverse power method to find the
eigenpairs of the matrix

⎡ 0 11 − 5 ⎤
A = ⎢⎢− 2 17 − 7 ⎥⎥
⎢⎣− 4 26 − 10⎥⎦

Note: The eigenvalues of A are λ1 = 4,λ2 = 2 and λ3 = 1,


and select and appropriate α and starting vector for
each case
Solution 4
Case (1): for the eigenvalue λ1 = 4, we select
α = 4.2 and the starting vector x0 = {1 1 1}T. First
form the matrix (A – 4.2I)-1, iteration produces

⎡−4.2 11 −5 ⎤ ⎧1⎫
⎢ − 2 12.8 −7 ⎥ y 0 = ⎨1⎬ = x 0
⎣ − 4 26 −14.2 ⎦ ⎩1⎭
⎧−9.545454545 ⎫ ⎧0.4117647059⎫
y0 = ⎨ −14.09090909 ⎬ = −23.18181818 ⎨0.6078431373⎬ = c1 x1
⎩− 23.18181818⎭ ⎩ 1 ⎭
Solution 4
Solution 4
the sequence converges to µ1 = -5 which is
the dominant eigenvalue of (A – 4.2I)-1 ,
and xk converges to v1 = [2/5, 3/5,1]T.

the eigenvalue λ1 of A is given by


λ1 = (1/ µ1 ) + α = 1/(-5) + 4.2 = -0.2+4.2 = 4
Solution 4
Case (2): for the eigenvalue λ2 = 2, we select α =
2.1 and the starting vector x0 = {1 1 1}T then form
the matrix (A – 2.1I)-1.

the sequence converges to µ1 = -10 and the


eigenpair of the matrix is

λ2 = (1/ µ2 ) + α = (1/-10) + 2.1 = -0.1+2.1 =2


v1 = [1/4, 1/2,1]T.
Solution 4
Solution 4
Case (3): for the eigenvalue λ3 = 1, we select α =
0.875 and the starting vector x0 = {0 1 1}T then
form the matrix (A – 0.875I)-1.

the sequence converges to µ1 = 8 and the


eigenpair of the matrix is

λ2 = (1/ µ1 ) + α = (1/8) + 0.875 = 0.125+0.875 = 1


v3 = [1/2, 1/2,1]T.
Solution 4

You might also like