06. Eigenvalue Problem (1)

  • Upload
    edmond

  • View
    215

  • Download
    0

Embed Size (px)

Citation preview

  • 8/18/2019 06. Eigenvalue Problem (1)

    1/33

    The Eigenvalue Problem

  • 8/18/2019 06. Eigenvalue Problem (1)

    2/33

    Homogeneous Systems

    •  A homogeneous linear system

     A x  = 0•  If det( A)!0, the unique solution is the trivial

    solution x  = 0.•  If det( A)=0, there exist nontrivial solutions. 

  • 8/18/2019 06. Eigenvalue Problem (1)

    3/33

    Example 1

    Find the nontrivial solutions to the

    homogeneous system

    045

    02

    02

    321

    321

    321

    =++

    =++

    =!

    +

     x x x

     x x x

     x x x

  • 8/18/2019 06. Eigenvalue Problem (1)

    4/33

    Solution 1

    The system reduces to:

    Letting x3=t , the solution can be expressed as

     x 1+  x 

    2  = 0

    !

     x 2 +  x 

    3 = 0

     x 1+ 2 x 

    2 !  x 

    3 = 0

    2 x 1

    +  x 2

     +  x 3

     = 0

    5 x 1+ 4 x 

    2 +  x 

    3 = 0

    t  x

    t  x

    t  x

    =

    =

    !

    =

    3

    2

    1

    !

    "

    !#

    $

    !

    %

    !&

    '(=

    1

    1

    1

    t  X 

  • 8/18/2019 06. Eigenvalue Problem (1)

    5/33

    Applications

    • 

    Quantum mechanics, geology, mathematics,image processing, etc.

    • 

    MechanicsPrincipal stresses and the orientation of theprincipal planes

    • 

    Structural Dynamics

    Natural frequencies and mode shapes ofstructures

  • 8/18/2019 06. Eigenvalue Problem (1)

    6/33

    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 !  for which there exists a

    nonzero vector x  such that

    Eigenvalue

    (characteristic value)

    Eigenvector

    (characteristic vector) 

     x  x  A   ! =

  • 8/18/2019 06. Eigenvalue Problem (1)

    7/33

    Solution Methods

    •  Analytical Solution

     –  Solution of the characteristic equation

    •  Numerical Solution

     –  Power method

     –  Inverse power method

     –  Jacobi’s method, QR method

  • 8/18/2019 06. Eigenvalue Problem (1)

    8/33

    Analytical Method

    • 

    The eigenvalues of A are the solutions of the characteristic equation 

    •  When the determinant is expanded, it becomesa polynomial of degree n, called the

    characteristic polynomial .

    ( )   0 x  I  A   =! " 

    ( )   0det

    21

    22221

    11211

    =

    !

    !

    !

    =!

    nnnn

    n

    n

    aaa

    aaa

    aaa

    !

    "#""

    !

    !

     I  A

  • 8/18/2019 06. Eigenvalue Problem (1)

    9/33

    Eigenvalues

    If A is an n x n real matrix, then its n eigenvalues! 1, ! 2, !, ! n. are the real and complex roots of

    the characteristic polynomial

    Properties:

    ( ) ( ) I  A p   ! !    "= det

    ( )   !!==

    ==

    n

    n

    i

    iiatrace

    11

    "  A   ( )   !=

    =

    n

    1

    det   "  A

    ( ) ( ) ( )( ) ( )nn p   ! ! ! ! ! ! ! 

      """"

    =

      !211

  • 8/18/2019 06. Eigenvalue Problem (1)

    10/33

    Eigenvector

    If !  is an eigenvalue of A and the nonzero vector v

    has the property that

    Then v is called the eigenvector  of A

    corresponding to the eigenvalue ! . v is obtainedby solving the homogeneous system ( A - !   I ) v =

    0.

    vv A   ! =

  • 8/18/2019 06. Eigenvalue Problem (1)

    11/33

    Analytical Method

    Hand computations used to solve the eigenvalue

    problem when the dimension n is small.

    1. 

    Find the coefficients of the characteristicpolynomial p(! )= det( A - !   I )

    2. 

    Find its roots (eigenvalues).

    3.  Find the nonzero solutions (eigenvectors) of

    the homogeneous linear system ( A - !   I ) v = 0. 

  • 8/18/2019 06. Eigenvalue Problem (1)

    12/33

    Example 2

    Find the eigenpairs !  j , v j  for the matrix

    !

    !!

    "

    #

    $

    $$

    %

    &

    '

    '''

    =

    310

    121

    013

     A

  • 8/18/2019 06. Eigenvalue Problem (1)

    13/33

    Solution 2

    Eigenvalues. The characteristic equation det( A - !   I )=0 is

    Therefore, the three eigenvalues are

    ! 1 = 1, ! 2  = 3, and ! 3 = 4.

    012198

    310

    121

    013

    23 =+!

    +!

    =

    !!

    !!!

    !!

    " " " 

    ( )( )( )   0431   =!!!!   " " " 

  • 8/18/2019 06. Eigenvalue Problem (1)

    14/33

    Solution 2

    Eigenvectors.

    For ! 1 = 1,

    For ! 2  = 3,

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    (((

    )

    *

    +++

    ,

    -

    ...

    .

    0

    0

    0

    210

    111

    012

    3

    2

    1

     x

     x

     x

    !"

    !#

    $

    !%

    !&

    '

    =

    1

    2

    1

    1  av

    02

    02

    32

    21

    =+!

    =!

     x x

     x x

    a x

    a x

    a x

    =

    =

    =

    3

    1

    2  2

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    (((

    )

    *

    +++

    ,

    -

    .

    ...

    .

    0

    0

    0

    010

    111

    010

    3

    2

    1

     x

     x

     x

    !"

    !#

    $

    !%

    !&

    '

    (

    =

    1

    0

    1

    2  bv

    0

    0

    2

    31

    =

    =+

     x

     x x

    b x

    b x

     x

    !=

    =

    =

    3

    1

    2  0

  • 8/18/2019 06. Eigenvalue Problem (1)

    15/33

    Solution 2

    Eigenvectors.

    For ! 3 = 4,

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    (((

    )

    *

    +++

    ,

    -

    ..

    ...

    ..

    0

    0

    0

    110

    121

    011

    3

    2

    1

     x

     x

     x

    !"

    !#

    $

    !%

    !&

    '

    (=

    1

    1

    1

    2  cv

    0

    0

    32

    21

    =+

    =+

     x x

     x x

    c x

    c x

    c x

    =

    !=

    =

    1

    2

    3

  • 8/18/2019 06. Eigenvalue Problem (1)

    16/33

    Numerical Methods

    1.  Power Method

    2. 

    Shifted-inverse Power Method

  • 8/18/2019 06. Eigenvalue Problem (1)

    17/33

    Power Method

     Assume that the n x n matrix A has n distinct

    eigenvalues ! 1, ! 2, !, ! n and that they are

    ordered in decreasing magnitude; that is,

    |! 1| > |! 2| " |! 3| " ! " |! n|

    eigenvector v1 corresponding to ! 1 is called adominant eigenvector .

  • 8/18/2019 06. Eigenvalue Problem (1)

    18/33

    Power Method

    If x 0 is chosen appropriately, then the sequences{ x k  = [ x1

    (k )  x2(k ) !  xn

    (k ) ]’} and {ck } generated

    recursively by

     yk  = A  x kand

     x k +1 = (1/ck +1) yk ,

    where

    ck +1 = max{| xi

    (k )|

    will converge to the dominant eigenvector v1 andeigenvalue ! 1, respectively.

  • 8/18/2019 06. Eigenvalue Problem (1)

    19/33

    Example 3

    Use the power method to find the dominant

    eigenpair for the matrix

    !!!

    "

    #

    $$$

    %

    &

    ''

    '''

    =

    10264

    7172

    5110

     A

  • 8/18/2019 06. Eigenvalue Problem (1)

    20/33

    Solution 3

    Start with x 0 = {1 1 1}T and use to generate the

    sequence of vectors { x k } and constants {c

    k }.

    The first iteration produces

    113

    2

    2

    1

    1

    12

    12

    8

    6

    1

    1

    1

    10264

    7172

    5110

     x c=

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    (

    ((

    )

    *

    +

    ++

    ,

    -

    ..

    ..

    .

  • 8/18/2019 06. Eigenvalue Problem (1)

    21/33

    Solution 3

    The second iteration produces

    Iteration generates the sequence

    The sequence converges to v = {2/5 3/5 1}T, ! =4.

    228

    5

    16

    7

    316

    3

    10

    3

    7

    3

    2

    2

    1

    13

    16

    110264

    7172

    5110

     x c=

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    =

    !"

    !#

    $

    !%

    !&

    '

    (((

    )

    *

    +++

    ,

    -

    ..

    ..

    .

    !,

    119

    78,

    19

    38,

    12

    9,

    13

    16,

    1

    1278

    47

    52

    21

    38

    23

    76

    31

    18

    11

    12

    5

    8

    5

    16

    7

    3

    2

    2

    1

    !"

    !#

    $

    !%

    !&

    '

    !"

    !#

    $

    !%

    !&

    '

    !"

    !#

    $

    !%

    !&

    '

    !"

    !#

    $

    !%

    !&

    '

    !"

    !#

    $

    !%

    !&

    '

  • 8/18/2019 06. Eigenvalue Problem (1)

    22/33

    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

  • 8/18/2019 06. Eigenvalue Problem (1)

    23/33

    Shifted-inverse Power Method

    Shifting Eigenvalues. Suppose that ! , v  is an

    eigenpair of A. If "  is any constant, then !  - #, v

    is an eigenpair of the matrix A - "  I. 

    Inverse Eigenvalues. Suppose that ! , v  is an

    eigenpair of A. If ! !0, then 1 / ! , v is an

    eigenpair of the matrix A-1. 

    Suppose that ! , v  is an eigenpair of A. If " !! , then

    1 / (!  - #), v is an eigenpair of the matrix ( A-# I )-1. 

  • 8/18/2019 06. Eigenvalue Problem (1)

    24/33

    Shifted-inverse Power Method

     A constant "  can be chosen so that µ 1 = 1/(! i - #)is the dominant eigenvalue of ( A - # I )-1.

    Furthermore, if x 0 is chosen appropriately, then

    the sequences { x k  = [ x

    1

    (k )  x2

    (k ) !  xn

    (k ) ]’} and {ck }

    generated recursively by 

     yk  = ( A - "  I )-1  x 

    k

    and

     x k +1 = (1/ck +1) yk ,

    where

    ck +1

    = max{| xi(k )|}

  • 8/18/2019 06. Eigenvalue Problem (1)

    25/33

    Shifted-inverse Power Method

    will converge to the dominant eigenpair of the

    matrix ( A - # I )-1. Finally, the corresponding

    eigenvalue for the matrix A is given by the

    calculation 

    ! µ 

    "    +=

    1

    1

      j

  • 8/18/2019 06. Eigenvalue Problem (1)

    26/33

    Example 4

    Employ the shifted-inverse power method to find theeigenpairs of the matrix

    Note: The eigenvalues of A are ! 1 = 4,! 2 = 2 and ! 3 = 1,

    and select and appropriate " and starting vector for

    each case

    !!!

    "

    #

    $$$

    %

    &

    ''

    ''

    '

    =

    10264

    7172

    5110

     A

  • 8/18/2019 06. Eigenvalue Problem (1)

    27/33

    Solution 4

    Case (1): for the eigenvalue ! 1 = 4, we select" = 4.2 and the starting vector x 

    0 = {1 1 1}T. First

    form the matrix ( A  – 4.2 I )-1, iteration produces

    0

    1

    1

    1

    2.14264

    78.122

    5112.4

    0  x  y   ==

    !!

    !!!!

    "#$

    %&'

    ()

    *+,

    -

    11

    1

    6078431373.0

    4117647059.0

    18181818.23

    18181818.23

    09090909.14

    545454545.9

    0  x  y   c=!=

    !

    !

    !

    =

    "#$

    %&'

    "#$

    %&'

  • 8/18/2019 06. Eigenvalue Problem (1)

    28/33

    Solution 4

  • 8/18/2019 06. Eigenvalue Problem (1)

    29/33

    Solution 4

    the sequence converges to µ1 = -5 which is

    the dominant eigenvalue of ( A  – 4.2 I )-1 ,

    and xk  converges to v1 = [2/5, 3/5,1]

    T.

    the eigenvalue ! 1 of A is given by

    ! 1 = (1/ µ

    1 ) + " = 1/(-5) + 4.2 = -0.2+4.2 = 4

  • 8/18/2019 06. Eigenvalue Problem (1)

    30/33

    Solution 4

    Case (2): for the eigenvalue ! 2 = 2, we select " =

    2.1 and the starting vector x 0 = {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

    ! 2 = (1/ µ2 ) + " = (1/-10) + 2.1 = -0.1+2.1 =2 

    v1 = [1/4, 1/2,1]T.

  • 8/18/2019 06. Eigenvalue Problem (1)

    31/33

    Solution 4

  • 8/18/2019 06. Eigenvalue Problem (1)

    32/33

    Solution 4

    Case (3): for the eigenvalue ! 3 = 1, we select " =

    0.875 and the starting vector x 0 = {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

    ! 2 = (1/ µ1 ) + " = (1/8) + 0.875 = 0.125+0.875 = 1 

    v3 = [1/2, 1/2,1]T.

  • 8/18/2019 06. Eigenvalue Problem (1)

    33/33

    Solution 4