+ All Categories
Home > Documents > solution of linear equations system

solution of linear equations system

Date post: 02-Jun-2018
Category:
Upload: attique-javaid
View: 228 times
Download: 0 times
Share this document with a friend

of 33

Transcript
  • 8/10/2019 solution of linear equations system

    1/33

    SOLUTION OF LINEAR EQUATION SYSTEMS

    DR. AJMAL SHAH, PIEAS

    Where Ais the square sparse coefficient matrix

    is a vector(or column matrix) containing the

    variable values at the grid nodes

    Qis the vector containing the terms on the right-

    hand side

    QA

    In CFD we always get the sparse matrices

    (matrices with many zero elements) afterdiscretization of partial differential equations

  • 8/10/2019 solution of linear equations system

    2/33

    SOLUTION OF LINEAR EQUATION SYSTEMS

    DR. AJMAL SHAH, PIEAS

    All the non-zero coefficients lie on the main

    diagonal and the two neighboring diagonals for

    1Dcase

    The next two neighboring diagonals are added for2D case and even the next two diagonals for 3D

    case

    All other coefficients are zero

    This structure allows the use of efficient iterative

    solvers

    QA

  • 8/10/2019 solution of linear equations system

    3/33

    For 1Dand 2Dthe coefficient matrix A, vector and

    Qvector will have the following shape

    QA

    SOLUTION OF LINEAR EQUATION SYSTEMS

    DR. AJMAL SHAH, PIEAS

    5

    4

    3

    2

    1

    5

    4

    3

    2

    1

    *0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    00

    Q

    QQ

    Q

    Q

    A

    A

    A

    AA

    AA

    A

    A

    A

    A

    A

    A

    p

    E

    W

    p

    E

    W

    p

    E

    W

    p

    E

    W

    p

    5

    4

    3

    2

    1

    5

    4

    3

    2

    1

    *

    0

    00

    00

    0

    Q

    Q

    Q

    Q

    Q

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    A

    p

    N

    E

    S

    p

    N

    E

    W

    S

    p

    N

    E

    W

    S

    p

    N

    W

    S

    p

    QAAA EEWWPP QAAAAA EENNSSWWPP

    QAA k kkPP

  • 8/10/2019 solution of linear equations system

    4/33

    Direct Methods

    Gauss Elimination

    LU Decomposition

    Tri-diagonal SystemsCyclic Reduction

    DR. AJMAL SHAH, PIEAS

    SOLUTION OF LINEAR EQUATION SYSTEMS

    Iterative Methods

    Jacobi Iteration

    Gauss-Seidel Iteration

    Successive OverRelaxation (S.O.R)

  • 8/10/2019 solution of linear equations system

    5/33

    An analytical method to solve simultaneous linear

    system of equations of the form [A][X]=[C]

    The solution obtained is exact

    Its basis is the systematic reduction of large

    systems of equations to smaller onesWe use full-matrix notation, because it is the basic

    method and most of the methods are based on it

    It has two major steps

    1. Forward Elimination2. Back Substitution

    DR. AJMAL SHAH, PIEAS

    GAUSS ELIMINATION METHOD

  • 8/10/2019 solution of linear equations system

    6/33

    2.2792.177

    8.106

    1121441864

    1525

    3

    2

    1

    xx

    x

    The goal of forward elimination is to transform the

    coefficient matrix into an upper triangular matrix

    735.0

    21.96

    8.106

    7.000

    56.18.40

    1525

    3

    2

    1

    x

    x

    x

    DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

  • 8/10/2019 solution of linear equations system

    7/33DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

    112144

    1864

    1525

    A

    11214456.18.40

    1525

    56.212;56.225

    64

    RowRow

    76.48.160

    56.18.40

    1525

    76.513;76.525

    144

    RowRow

    ExampleConsider a matrix Aof 3*3

    Perform the forward elimination

    step

    7.000

    56.18.40

    1525

    5.323;5.38.4

    8.16

    RowRow

  • 8/10/2019 solution of linear equations system

    8/33

    Solve each equation starting from the last equation

    Example of a system of 3equations

    735.0

    21.96

    8.106

    7.00056.18.40

    1525

    3

    2

    1

    xx

    x

    DR. AJMAL SHAH, PIEAS

    BACK SUBSTITUTION

    735.07.0 3x

    21.9656.18.4 32 xx

  • 8/10/2019 solution of linear equations system

    9/33

    We have to make a21

    in equation 2equal to zero

    Divide Equation 1by a11and multiply by a21

    ).. .( 1131321211111

    21 bxaxaxaxaa

    ann

    1

    11

    211

    11

    21212

    11

    21121 ... b

    a

    axa

    a

    axa

    a

    axa nn

    DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

    nnnnnnn

    n

    n

    n

    b

    b

    b

    b

    x

    x

    x

    x

    aaaa

    aaaa

    aaaa

    aaaa

    3

    2

    1

    3

    2

    1

    321

    3333231

    2232221

    1131211

  • 8/10/2019 solution of linear equations system

    10/33

    )...( 111

    21

    1

    11

    21

    212

    11

    21

    121 ba

    a

    xaa

    a

    xaa

    a

    xa nn

    1

    11

    2121

    11

    212212

    11

    2122 .. . b

    a

    abxa

    a

    aaxa

    a

    aa nnn

    '

    2

    '

    22

    '

    22 ... bxaxa nn

    22323222121 ... bxaxaxaxa nn

    Subtract the result from Equation 2, and replace

    equation2 with this new equation

    DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

    nnnnnnn

    n

    n

    n

    b

    b

    b

    b

    x

    x

    x

    x

    aaaa

    aaaa

    aaaa

    aaaa

    3

    2

    1

    3

    2

    1

    321

    3333231

    2232221

    1131211

  • 8/10/2019 solution of linear equations system

    11/33

    Repeat this procedure for the remaining equations toreduce the set of equations as

    11313212111 ... bxaxaxaxa nn '

    2

    '

    23

    '

    232

    '

    22 ... bxaxaxa nn '

    3

    '

    33

    '

    332

    '

    32 ... bxaxaxa nn

    ''

    3

    '

    32

    '

    2 ... nnnnnn bxaxaxa

    . . .

    . . .

    . . .

    DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

  • 8/10/2019 solution of linear equations system

    12/33

    Repeat the same procedure for the2

    nd term of Equation

    3 and so on

    11313212111 ... bxaxaxaxa nn

    '

    2

    '

    23

    '

    232

    '

    22 ... bxaxaxa nn "

    3

    "

    33

    "

    33 ... bxaxa nn

    ""

    3

    "

    3 ... nnnnn bxaxa

    . .

    . .

    . .

    DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

  • 8/10/2019 solution of linear equations system

    13/33

    At the end of (n-1)Forward Elimination steps, thesystem of equations will look like

    '

    2

    '

    23

    '

    232

    '

    22 ... bxaxaxa nn

    33333 ... bxaxa nn

    nnnn

    bxa

    . .

    . .

    . .

    11313212111 ... bxaxaxaxa nn

    DR. AJMAL SHAH, PIEAS

    FORWARD ELIMINATION

  • 8/10/2019 solution of linear equations system

    14/33

    n

    n

    '

    n

    n

    nn

    nnnn

    n

    '

    n

    ''

    n

    b

    b

    b

    b

    b

    x

    x

    x

    x

    x

    a

    aa

    aa

    aaa

    aaaa

    1

    3

    2

    1

    1

    3

    2

    1

    )1()1)(1(

    333

    22322

    1131211

    00

    0

    0

    0

    0

    0

    00

    0

    DR. AJMAL SHAH, PIEAS

    MATRIX FORM AT THE END OF FORWARD ELIMINATION

  • 8/10/2019 solution of linear equations system

    15/33

    '

    2

    '

    23

    '

    232

    '

    22 ... bxaxaxa nn

    "

    3

    "

    33

    "

    33 ... bxaxa nn

    nnnn bxa

    . .. .

    . .

    11313212111 ... bxaxaxaxa nn

    DR. AJMAL SHAH, PIEAS

    BACK SUBSTITUTION

    "'

    1

    '

    1,

    ' ... iiiniiiiii bxaxaxa

    . .

    . .

    . .

  • 8/10/2019 solution of linear equations system

    16/33

    1,...,1for1

    ni

    a

    xab

    xii

    n

    ijjiji

    i

    nn

    nn

    abx

    1,...,1for... ,22,11,

    ni

    a

    xaxaxabx

    ii

    nniiiiiiii

    i

    DR. AJMAL SHAH, PIEAS

    BACK SUBSTITUTION

    ''

    1

    '

    1,

    ' ... iiiniiiiii bxaxaxa

    nnnn bxa Calculating x

    nfrom

    last row (equation)

    The ithrow (equation) is

  • 8/10/2019 solution of linear equations system

    17/33

    The operation count for gaussian elimination The number of operations required in the forward

    elimination is proportional to n3/3 arithmetic

    operations

    The number of operations required in the backsubstitution is proportional to n2/2 arithmetic

    operations

    The high cost of gauss elimination providesincentive to search for more efficient special

    solvers for matrices such as the sparseones arising

    from the discretization of differential equations

    DR. AJMAL SHAH, PIEAS

    OPERATION COUNT

  • 8/10/2019 solution of linear equations system

    18/33

    A number of variations on gauss elimination have been

    proposedOne variant of value to CFDis LUdecomposition

    Any matrix A can be factored into the product of lower

    (L)and upper (U)triangular matrices

    The diagonal elements of L, lii, or U, uii, are kept

    equal to 1

    QA

    LUA

    QLUA

    YU

    DR. AJMAL SHAH, PIEAS

    LU DECOMPOSITION

    QLY

  • 8/10/2019 solution of linear equations system

    19/33

    LU factorization can be performed without knowing the

    vector Q

    If many systems involving the same matrix are to be

    solved, considerable savings can be obtained

    U is the same as the coefficient matrix at the end of the

    forward elimination step

    L is obtained using the multipliers that were used in the

    forward elimination process

    DR. AJMAL SHAH, PIEAS

    LU DECOMPOSITION

    33

    2322

    131211

    3231

    21

    00

    0

    1

    01001

    .

    u

    uuuuu

    ULA

  • 8/10/2019 solution of linear equations system

    20/33DR. AJMAL SHAH, PIEAS

    LU DECOMPOSITION

    112144

    1864

    1525

    A

    11214456.18.40

    1525

    56.212;56.225

    64

    RowRow

    76.48.160

    56.18.40

    1525

    76.513;76.525

    144

    RowRow

    ExampleConsider a matrix Aof 3*3

    Perform the forward elimination

    step

    7.000

    56.18.40

    1525

    5.323;5.38.4

    8.16

    RowRow

  • 8/10/2019 solution of linear equations system

    21/33DR. AJMAL SHAH, PIEAS

    LU DECOMPOSITION

    Using the multipliers used during the Forward

    Elimination Procedure to find L

    7.000

    56.18.40

    1525

    U

    1

    01

    001

    3231

    21

    L 112144

    1864

    1525

    A

    56.225

    64

    11

    2121

    a

    a 76.5

    25

    144

    11

    3131

    a

    a 5.3

    8.4

    8.16

    22

    32

    32

    a

    a

    15.376.5

    0156.2001

    L

    7.000

    56.18.401525

    15.376.5

    0156.2001

    .UL ?

    QA LUA

    QLUA YUQLY ;

  • 8/10/2019 solution of linear equations system

    22/33

    ii

    i

    Ei

    i

    Pi

    i

    W QAAA 11

    For 1Dproblems the resulting algebraic equations

    have an especially simple structureThey constitute a tri-diagonalsystem

    DR. AJMAL SHAH, PIEAS

    TRI-DIAGONAL SYSTEMS

    n

    n

    i

    n

    n

    i

    nn

    nn

    nn

    nnnn

    iiiiii

    Q

    Q

    Q

    Q

    Q

    Q

    A

    A

    A

    AA

    AAA

    AAA

    AAA

    AA

    1

    3

    2

    1

    1

    3

    2

    1

    ,

    ,1

    1,

    1,1,1

    1,,1,

    343332

    232221

    1211

    *

    0.........0.........

    00.

    0.

    .....

    .0

    .0

    ...

    .0

    00..

    00

    00

    00

    ..

    ....

    ....

    ....

    ....

    0

    0

    00

  • 8/10/2019 solution of linear equations system

    23/33

    1

    1

    i

    Ei

    P

    i

    Wi

    P

    i

    P AA

    AAA

    DR. AJMAL SHAH, PIEAS

    TRI-DIAGONAL SYSTEMS

    Only one element needs to be eliminated from each

    row during the forward elimination process

    When the algorithm has reached the ith row, onlyAPneeds to be modified; the new value is:

    n

    n

    i

    n

    n

    i

    nn

    nn

    nn

    nnnn

    iiiiii

    Q

    Q

    Q

    Q

    Q

    Q

    AA

    AAA

    AAA

    AAAAAA

    AA

    1

    3

    2

    1

    1

    3

    2

    1

    ,

    ,1

    1,

    1,1,1

    1,,1,

    343332

    232221

    1211

    *

    0.........

    0.........

    00.0.

    ....

    .

    .0

    .0

    ...

    .0

    00.0

    0000

    00

    ..

    ....

    ....

    ....

    ....

    00

    00

    ii

    ii

    ii

    iiii AA

    AAA ,1

    1,1

    1,

    ,,

  • 8/10/2019 solution of linear equations system

    24/33

    The forcing term is also modified:

    This tri-diagonal solution method is sometimes called

    the Thomas Algorithm or the Tri-diagonal Matrix

    Algorithm (TDMA)

    1

    *

    1*

    i

    P

    i

    i

    Wii

    A

    QAQQ

    *

    1 ii

    i

    Ei

    i

    P QAA

    DR. AJMAL SHAH, PIEAS

    TRI-DIAGONAL SYSTEMS

    n

    n

    i

    n

    n

    i

    nn

    nnnn

    iiii

    Q

    Q

    Q

    Q

    Q

    Q

    A

    AA

    AA

    AA

    AA

    AA

    1

    3

    2

    1

    1

    3

    2

    1

    ,

    ,11,1

    1,,

    3433

    2322

    1211

    *

    0.........0.........

    000.

    00.

    ....0.

    .0

    .0

    ...

    .0

    0

    0

    0

    0

    00.0

    00

    00

    00

    ....

    00

    00

    00

    n

    P

    nn

    A

    Q *

    i

    P

    i

    i

    Eii

    A

    AQ 1*

  • 8/10/2019 solution of linear equations system

    25/33

    A Fortran code requires 8 executable lines forTDMA method

    The number of operations is proportional to n,

    the number of unknowns, rather than the n3of full

    matrix Gauss elimination

    The cost per unknown is independent of the n

    Thus, Very economical method

    But applicable to tridiagonal matrices only

    DR. AJMAL SHAH, PIEAS

    TRI-DIAGONAL SYSTEMS

  • 8/10/2019 solution of linear equations system

    26/33

    Jacobi Iteration

    Gauss-Seidel Iteration

    Successive Over Relaxation (S.O.R) SORis a method used to accelerate the

    convergence

    Gauss-Seidel Iteration is a special case of SOR

    method

    DR. AJMAL SHAH, PIEAS

    DIFFERENT ITERATIVE METHODS

  • 8/10/2019 solution of linear equations system

    27/33

    The cost of direct methods is very high

    The discretization error is usually much largerthan the solver accuracy

    Therefore no need of solving equations so

    accurately This leaves an opening for iterative methods

    They must be used for non linear systems butare more economical for linear systems as

    well

    DR. AJMAL SHAH, PIEAS

    DIFFERENT ITERATIVE METHODS

  • 8/10/2019 solution of linear equations system

    28/33

    The truncation error in the discretization procedure is

    proportional to grid spacing, x So as the number of grid points is increased and xis reduced,

    the error in the numerical solution would decrease and the

    agreement between the numerical and exact solutions would

    get better

    DR. AJMAL SHAH, PIEAS

    GRID CONVERGENCE

  • 8/10/2019 solution of linear equations system

    29/33

    nnnnnn

    nn

    nn

    bxaxaxa

    bxaxaxabxaxaxa

    2211

    22222121

    11212111

    0

    0

    2

    01

    0

    n

    x

    x

    x

    x

    )(1 0

    1

    0

    2121

    11

    1

    1 nnxaxaba

    x

    )(1 0

    11

    0

    22

    0

    11

    1

    nnnnnnnn

    n xaxaxaba

    x

    )(1 0203230121222

    12 nnxaxaxab

    ax

    DR. AJMAL SHAH, PIEAS

    JACOBI ITERATION

    1

    1 1

    1 1 i

    j

    n

    ij

    k

    jij

    k

    jiji

    ii

    k

    i xaxaba

    x

  • 8/10/2019 solution of linear equations system

    30/33

    nnnnnn

    nn

    nn

    bxaxaxa

    bxaxaxa

    bxaxaxa

    2211

    22222121

    11212111

    0

    0

    2

    01

    0

    nx

    x

    x

    x

    )(1 0

    2

    0

    323

    1

    1212

    22

    1

    2 nnxaxaxaba

    x

    Use the latest

    update

    DR. AJMAL SHAH, PIEAS

    GAUSS-SEIDEL (GS) ITERATION

    )(1 0

    1

    0

    2121

    11

    1

    1 nnxaxaba

    x

    )(1 1

    11

    1

    22

    1

    11

    1

    nnnnnnnn

    n xaxaxaba

    x

    1

    1 1

    11 1 i

    j

    n

    ij

    k

    jij

    k

    jiji

    ii

    k

    i xaxaba

    x

  • 8/10/2019 solution of linear equations system

    31/33

    Gauss-seidel iteration converges more rapidlythan the jacobi iteration since it uses the latest

    updates

    But there are some cases thatjacobiiteration does

    converge but gauss-seideldoes not

    To accelerate the gauss-seidel method even

    further, successive over relaxation method can be

    used

    DR. AJMAL SHAH, PIEAS

    COMPARISON

  • 8/10/2019 solution of linear equations system

    32/33

    1

    1 1

    11

    1)1(

    i

    j

    n

    ij

    kjijkjiji

    ii

    kiki xaxabaxx

    1

  • 8/10/2019 solution of linear equations system

    33/33

    The operation count for Gaussian Elimination orLU Decomposition was O (n3) (order of n3)

    For iterative methods, the number of scalar

    multiplications is O (n2

    )at each iteration If the total number of iterations required for

    convergence is much less than n, then iterative

    methods are more efficient than direct methods

    Also iterative methods are well suited for sparsematrices

    OPERATION COUNT


Recommended