Download - Matrix
MATRIX METHODSSYSTEMS OF LINEAR EQUATIONS
ENGR 351 Numerical Methods for EngineersSouthern Illinois University CarbondaleCollege of EngineeringDr. L.R. ChevalierDr. B.A. DeVantier
Copyright© 2003 by Lizette R. Chevalier and Bruce A. DeVantier
Permission is granted to students at Southern Illinois University at Carbondaleto make one copy of this material for use in the class ENGR 351, NumericalMethods for Engineers. No other permission is granted.
All other rights are reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means,electronic, mechanical, photocopying, recording, or otherwise, withoutthe prior written permission of the copyright owner.
Systems of Linear Algebraic EquationsSpecific Study Objectives
• Understand the graphic interpretation of ill-conditioned systems and how it relates to the determinant
• Be familiar with terminology: forward elimination, back substitution, pivot equations and pivot coefficient
• Apply matrix inversion to evaluate stimulus-response computations in engineering
• Understand why the Gauss-Seidel method is particularly well-suited for large sparse systems of equations
• Know how to assess diagonal dominance of a system of equations and how it relates to whether the system can be solved with the Gauss-Seidel method
Specific Study Objectives
• Understand the rationale behind relaxation and how to apply this technique
Specific Study Objectives
How to represent a system of linear equations as a matrix
[A]{x} = {c}
where {x} and {c} are both column vectors
44.0
67.0
01.0
5.03.01.0
9.115.0
152.03.0
}{}{
44.05.03.01.0
67.09.15.0
01.052.03.0
3
2
1
321
321
321
x
x
x
CXA
xxx
xxx
xxx
How to represent a system of linear equations as a matrix
Practical application
• Consider a problem in structural engineering
• Find the forces and reactions associated with a statically determinant truss
hinge: transmits bothvertical and horizontalforces at the surface
roller: transmitsvertical forces
30
90
60
1000 kg
30
90
60
F1
H2
V2 V3
2
3
1
FREE BODY DIAGRAM F
F
H
v
0
0
F2
F3
Node 1 F1,V
F1,H
F3F1
6030
F F F F
F F F F
F F
F F
H H
V V
0 30 60
0 30 60
30 60 0
30 60 1000
1 3 1
1 3 1
1 3
1 3
cos cos
sin sin
cos cos
sin sin
,
,
F H F F
F V F
H
V
0 30
0 30
2 2 1
2 1
cos
sin
Node 2
F2
F1
30
H2V2
F F F
F F V
H
V
0 60
0 60
3 2
3 3
cos
sin
Node 3
F2
F3
60
V3
060sin
060cos
030sin
030cos
100060sin30sin
060cos30cos
33
23
12
122
31
31
VF
FF
FV
FFH
FF
FF
SIX EQUATIONSSIX UNKNOWNS
F1 F2 F3 H2 V2 V3
1
2
3
4
5
6
-cos30 0 cos60 0 0 0
-sin30 0 -sin60 0 0 0 cos30 1 0 1 0 0
sin30 0 0 0 1 0
0 -1 -cos60 0 0 0
0 0 sin60 0 0 1
0
-1000
0
0
0
0
Do some book keeping
This is the basis for your matrices and the equation[A]{x}={c}
0 866 0 0 5 0 0 0
0 5 0 0 866 0 0 0
0 866 1 0 1 0 0
0 5 0 0 0 1 0
0 1 0 5 0 0 0
0 0 0 866 0 0 1
0
1000
0
0
0
0
1
2
3
2
2
3
. .
. .
.
.
.
.
F
F
F
H
V
V
System of Linear Equations
• We have focused our last lectures on finding a value of x that satisfied a single equation• f(x) = 0
• Now we will deal with the case of determining the values of x1, x2, .....xn, that simultaneously satisfy a set of equations
System of Linear Equations• Simultaneous equations
• f1(x1, x2, .....xn) = 0
• f2(x1, x2, .....xn) = 0 • .............• fn(x1, x2, .....xn) = 0
• Methods will be for linear equations• a11x1 + a12x2 +...... a1nxn =c1
• a21x1 + a22x2 +...... a2nxn =c2
• ..........
• an1x1 + an2x2 +...... annxn =cn
Mathematical BackgroundMatrix Notation
• a horizontal set of elements is called a row• a vertical set is called a column• first subscript refers to the row number• second subscript refers to column number
A
a a a a
a a a a
a a a a
n
n
m m m mn
11 12 13 1
21 22 23 2
1 2 3
...
...
. . . .
...
mnmmm
n
n
aaaa
aaaa
aaaa
A
...
....
...
...
321
2232221
1131211
This matrix has m rows an n column.
It has the dimensions m by n (m x n)
mnmmm
n
n
aaaa
aaaa
aaaa
A
...
....
...
...
321
2232221
1131211
This matrix has m rows and n column.
It has the dimensions m by n (m x n)
notesubscript
A
a a a a
a a a a
a a a a
n
n
m m m mn
11 12 13 1
21 22 23 2
1 2 3
...
...
. . . .
...
row 2
column 3Note the consistentscheme with subscriptsdenoting row,column
Row vector: m=1
Column vector: n=1 Square matrix: m = n
B b b bn 1 2 .......
C
c
c
cm
1
2
.
.
A
a a a
a a a
a a a
11 12 13
21 22 23
31 32 33
Types of Matrices
• Symmetric matrix• Diagonal matrix• Identity matrix• Inverse of a matrix• Transpose of a matrix• Upper triangular matrix• Lower triangular matrix• Banded matrix
Definitions
Symmetric Matrix
aij = aji for all i’s and j’s
A
5 1 2
1 3 7
2 7 8Does a23 = a32 ?
Yes. Check the other elementson your own.
Diagonal Matrix
A square matrix where all elements off the main diagonal are zero
A
a
a
a
a
11
22
33
44
0 0 0
0 0 0
0 0 0
0 0 0
Identity Matrix
A diagonal matrix where all elements on the main diagonal are equal to 1
A
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
The symbol [I] is used to denote the identify matrix.
Inverse of [A]
IAAAA 11
Transpose of [A]
A
a a a
a a a
a a a
t
m
m
n n mn
11 21 1
12 22 2
1 2
. . .
. . .
. . . . . .
. . . . . .
. . . . . .
. . .
Upper Triangle Matrix
Elements below the main diagonal are zero
A
a a a
a a
a
11 12 13
22 23
33
0
0 0
Lower Triangular Matrix
All elements above the main diagonal are zero
A
5 0 0
1 3 0
2 7 8
Banded Matrix
All elements are zero with the exception of a band centered on the main diagonal
A
a a
a a a
a a a
a a
11 12
21 22 23
32 33 34
43 44
0 0
0
0
0 0
Matrix Operating Rules
• Addition/subtraction• add/subtract corresponding terms• aij + bij = cij
• Addition/subtraction are commutative• [A] + [B] = [B] + [A]
• Addition/subtraction are associative• [A] + ([B]+[C]) = ([A] +[B]) + [C]
Matrix Operating Rules• Multiplication of a matrix [A] by a
scalar g is obtained by multiplying every element of [A] by g
B g A
ga ga ga
ga ga ga
ga ga ga
n
n
m m mn
11 12 1
21 22 2
1 2
. . .
. . .
. . . . . .
. . . . . .
. . . . . .
. . .
Matrix Operating Rules
• The product of two matrices is represented as [C] = [A][B]
• n = column dimensions of [A]• n = row dimensions of [B]
c a bij ik kjk
N
1
[A] m x n [B] n x k = [C] m x k
interior dimensionsmust be equal
exterior dimensions conform to dimension of resulting matrix
Simple way to check whether matrix multiplication is possible
Recall the equation presented for matrix multiplication• The product of two matrices is
represented as [C] = [A][B]
• n = column dimensions of [A]• n = row dimensions of [B]
c a bij ik kjk
N
1
ExampleDetermine [C] given [A][B] = [C]
203
123
142
320
241
231
B
A
Matrix multiplication• If the dimensions are suitable,
matrix multiplication is associative• ([A][B])[C] = [A]([B][C])
• If the dimensions are suitable, matrix multiplication is distributive• ([A] + [B])[C] = [A][C] + [B][C]
• Multiplication is generally not commutative• [A][B] is not equal to [B][A]
Determinants
Denoted as det A or lAl
for a 2 x 2 matrix
bcaddc
ba
bcaddc
ba
Determinants
254
329
132
For a 3 x 3
254
329
132
254
329
132
+ - +
516
234
971
Problem
Determine the determinant of the matrix.
Properties of Determinants
• det A = det AT
• If all entries of any row or column is zero, then det A = 0
• If two rows or two columns are identical, then det A = 0
• Note: determinants can be calculated using mdeterm function in Excel
Excel Demonstration
• Excel treats matrices as arrays• To obtain the results of
multiplication, addition, and inverse operations, you hit control-shift-enter as opposed to enter.
• The resulting matrix cannot be altered…let’s see an example using Excel in class
matrix.xls
Matrix Methods
• Cramer’s Rule• Gauss elimination• Matrix inversion • Gauss Seidel/Jacobi
A
a a a
a a a
a a a
11 12 13
21 22 23
31 32 33
Graphical Method2 equations, 2 unknowns
a x a x c
a x a x c
xa
ax
c
a
xa
ax
c
a
11 1 12 2 1
21 1 22 2 2
211
121
1
12
221
221
2
22
x2
x1
( x1, x2 )
3 2 18
3
29
1 2
2 1
x x
x x
x2
x1
3
2
9
x x
x x
1 2
2 1
2 2
1
21
x2
x1
2
1
1
3 2 18
2 2
3
29
1
21
1 2
1 2
2 1
2 1
x x
x x
x x
x x
x2
x1
( 4 , 3 )
3
2
2
1
9
1
Check: 3(4) + 2(3) = 12 + 6 = 18
Special Cases
• No solution• Infinite solution• Ill-conditioned
x2
x1
( x1, x2 )
a) No solution - same slope f(x)
xb) infinite solution f(x)
x
-1/2 x1 + x2 = 1-x1 +2x2 = 2
c) ill conditionedso close that the points ofintersection are difficult todetect visually
f(x)
x
• If the determinant is zero, the slopes are identical
a x a x c
a x a x c11 1 12 2 1
21 1 22 2 2
Rearrange these equations so that we have an alternative version in the form of a straight line:
i.e. x2 = (slope) x1 + intercept
Ill Conditioned Systems
xaa
xca
xaa
xca
211
121
1
12
221
221
2
22
If the slopes are nearly equal (ill-conditioned)
aa
aa
a a a a
a a a a
11
12
21
22
11 22 21 12
11 22 21 12 0
a a
a aA11 12
21 22
det
Isn’t this the determinant?
Ill Conditioned Systems
If the determinant is zero the slopes are equal.This can mean:
- no solution- infinite number of solutions
If the determinant is close to zero, the system is illconditioned.
So it seems that we should use check the determinant of a system before any further calculations are done.
Let’s try an example.
Ill Conditioned Systems
Example
Determine whether the following matrix is ill-conditioned.
12
22
5.22.19
7.42.37
2
1
x
x
37 2 4 7
19 2 2 537 2 2 5 4 7 19 2
2 76
. .
. .. . . .
.
What does this tell us? Is this close to zero? Hard to say.
If we scale the matrix first, i.e. divide by the largesta value in each row, we can get a better sense of things.
Solution
-80
-60
-40
-20
0
0 5 10 15
x
y
This is further justifiedwhen we consider a graphof the two functions.
Clearly the slopes arenearly equal
1 0126
1 01300 004
.
..
Solution
Another Check
• Scale the matrix of coefficients, [A], so that the largest element in each row is 1. If there are elements of [A]-1 that are several orders of magnitude greater than one, it is likely that the system is ill-conditioned.
• Multiply the inverse by the original coefficient matrix. If the results are not close to the identity matrix, the system is ill-conditioned.
• Invert the inverted matrix. If it is not close to the original coefficient matrix, the system is ill-conditioned.
We will consider how to obtain an inverted matrix later.
Cramer’s Rule
• Not efficient for solving large numbers of linear equations
• Useful for explaining some inherent problems associated with solving linear equations.
bxA
b
b
b
x
x
x
aaa
aaa
aaa
3
2
1
3
2
1
333231
232221
131211
Cramer’s Rule
xA
b a a
b a a
b a a1
1 12 13
2 22 23
3 32 33
1
to solve forxi - place {b} inthe ith column
Cramer’s Rule
to solve forxi - place {b} inthe ith column
33231
22221
11211
3
33331
23221
13111
2
33323
23222
13121
1
1
11
baa
baa
baa
Ax
aba
aba
aba
Ax
aab
aab
aab
Ax
EXAMPLE Use of Cramer’s Rule
2 3 5
5
2 3
1 1
5
5
1 2
1 2
1
2
x x
x x
x
x
2 3
1 1
5
5
2 1 3 1 2 3 5
1
5
5 3
5 1
1
55 1 3 5
20
54
1
5
2 5
1 5
1
52 5 5 1
5
51
1
2
1
2
x
x
A
x
x
Solution
Elimination of Unknowns( algebraic approach)
2112221111121
1212122111121
112222121
211212111
2222121
1212111
caxaaxaa
SUBTRACTcaxaaxaa
acxaxa
acxaxa
cxaxa
cxaxa
21122211
2121221
11222112
2111212
1122112112222111
2112221111121
1212122111121
aaaa
cacax
aaaa
cacax
acacxaaxaa
caxaaxaa
SUBTRACTcaxaaxaa
NOTE: same result asCramer’s Rule
Elimination of Unknowns( algebraic approach)
Gauss Elimination
• One of the earliest methods developed for solving simultaneous equations
• Important algorithm in use today• Involves combining equations in
order to eliminate unknowns and create an upper triangular matrix
• Progressively back substitute to find each unknown
Two Phases of Gauss Elimination
a a a c
a a a c
a a a c
a a a c
a a c
a c
11 12 13 1
21 22 23 2
31 32 33 3
11 12 13 1
22 23 2
33 3
0
0 0
|
|
|
|
|
|
' ' '
' ' ' '
ForwardElimination
Note: the prime indicatesthe number of times the element has changed from the original value.
Two Phases of Gauss Elimination
11
31321211
'22
3123
'2
2
''33
''3
3
''3
''33
'2
'23
'22
1131211
|00
|0
|
a
xaxacx
a
xacx
a
cx
ca
caa
caaa
Back substitution
Rules
• Any equation can be multiplied (or divided) by a nonzero scalar
• Any equation can be added to (or subtracted from) another equation
• The positions of any two equations in the set can be interchanged.
EXAMPLE
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
Perform Gauss Elimination of the following matrix.
Solution
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
Multiply the first equation by a21/ a11 = 4/2 = 2
Note: a11 is called the pivot element
2624 321 xxx
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
2624 321 xxx
a21 / a11 = 4/2 = 2
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
3952
1744
2624
321
321
321
xxx
xxx
xxx
a21 / a11 = 4/2 = 2
Subtract the revised first equation from the second equation
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
a21 / a11 = 4/2 = 2
3952
1744
2624
321
321
321
xxx
xxx
xxx
4 4 4 2 7 6 1 2
0 2 11 2 3
1 2 3
x x x
x x x
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
a21 / a11 = 4/2 = 2
Subtract the revised first equation from the second equation
3952
1744
2624
321
321
321
xxx
xxx
xxx
4 4 4 2 7 6 1 2
0 2 11 2 3
1 2 3
x x x
x x x
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
a21 / a11 = 4/2 = 2
Subtract the revised first equation from the second equation
3952
1744
2624
321
321
321
xxx
xxx
xxx
3952
120
132
321
321
321
xxx
xxx
xxx
NEWMATRIX
4 4 4 2 7 6 1 2
0 2 11 2 3
1 2 3
x x x
x x x
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
a21 / a11 = 4/2 = 2
Subtract the revised first equation from the second equation
3952
1744
2624
321
321
321
xxx
xxx
xxx
3952
120
132
321
321
321
xxx
xxx
xxx
NOW LET’SGET A ZEROHERE
Multiply equation 1 by a31/a11 = 2/2 = 1and subtract from equation 3
2 2 5 1 9 3 3 1
0 4 6 21 2 3
1 2 3
x x x
x x x
2 3 1
4 4 7 1
2 5 9 3
1 2 3
1 2 3
1 2 3
x x x
x x x
x x x
Solution
2 3 1
4 4 7 1
2 5 9 3
2 3 1
2 1
4 6 2
1 2 3
1 2 3
1 2 3
1 2 3
2 3
2 3
x x x
x x x
x x x
x x x
x x
x x
Following the same rationale, subtract the 3rd equation from the first equation
Continue thecomputation by multiplying the second equationby a32’/a22’ = 4/2 =2
Subtract the third equation of the newmatrix
Solution
2 3 1
2 1
4 6 2
2 3 1
2 1
4 4
1 2 3
2 3
2 3
1 2 3
2 3
3
x x x
x x
x x
x x x
x x
x
THIS DERIVATION OFAN UPPER TRIANGULAR MATRIXIS CALLED THE FORWARDELIMINATION PROCESS
Solution
From the system we immediately calculate:
x3
4
41
Continue to back substitute
2 3 1
2 1
4 4
1 2 3
2 3
3
x x x
x x
x
x
x
2
1
1 1
21
1 3 1
2
1
2
THIS SERIES OFSTEPS IS THEBACK SUBSTITUTION
Solution
Pitfalls of the Elimination Method
• Division by zero• Round off errors
• magnitude of the pivot element is small compared to other elements
• Ill conditioned systems
Pivoting
• Partial pivoting• rows are switched so that the pivot element is
not zero• rows are switched so that the largest element
is the pivot element
• Complete pivoting• columns as well as rows are searched for the
largest element and switched• rarely used because switching columns
changes the order of the x’s adding unjustified complexity to the computer program
For example
Pivoting is used here to avoid division by zero
2 3 8
4 6 7 3
2 6 5
2 3
1 2 3
1 2 3
x x
x x x
x x x
4 6 7 3
2 3 8
2 6 5
1 2 3
2 3
1 2 3
x x x
x x
x x x
Another Improvement: Scaling
• Minimizes round-off errors for cases where some of the equations in a system have much larger coefficients than others
• In engineering practice, this is often due to the widely different units used in the development of the simultaneous equations
• As long as each equation is consistent, the system will be technically correct and solvable
Use Gauss Elimination to solve the following setof linear equations. Employ partial pivoting when necessary.
3 13 50
2 6 45
4 8 4
2 3
1 2 3
1 3
x x
x x x
x x
Example (solution in notes)
3 13 50
2 6 45
4 8 4
2 3
1 2 3
1 3
x x
x x x
x x
First write in matrix form, employing short hand presented in class.
0 3 13 50
2 6 1 45
4 0 8 4
We will clearly run intoproblems of divisionby zero.
Use partial pivoting
Solution
0 3 13 50
2 6 1 45
4 0 8 4
Pivot with equationwith largest an1
501330
45162
4804
4804
45162
501330
501330
43360
4804
501330
45162
4804
4804
45162
501330
Begin developingupper triangular matrix
4 0 8 4
0 6 3 43
0 3 13 50
4 0 8 4
0 6 3 43
0 0 14 5 28 5
28 5
14 51 966
43 3 1 966
68149
4 8 1 966
42 931
3 8149 13 1 966 50
3 2
1
. .
.
..
..
..
. .
x x
x
CHECK
okay...end ofproblem
Gauss-Jordan
• Variation of Gauss elimination
• Primary motive for introducing this method is that it provides a simple and convenient method for computing the matrix inverse.
• When an unknown is eliminated, it is eliminated from all other equations, rather than just the subsequent one
• All rows are normalized by dividing them by their pivot elements
• Elimination step results in an identity matrix rather than an UT matrix
A
a a a
a a
a
11 12 13
22 23
33
0
0 0 A
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
Gauss-Jordan
Graphical depiction of Gauss-Jordan
a a a c
a a a c
a a a c
c
c
c
n
n
n
11 12 13 1
21 22 23 2
31 32 33 3
2
3
1 0 0
0 1 0
0 0 1
1
|
|
|
|
|
|
' ' '
' ' ' '
1 0 0
0 1 0
0 0 1
1
1
2
3
1
2 2
3 3
|
|
|
c
c
c
x c
x c
x c
n
n
n
n
n
n
a a a c
a a a c
a a a c
c
c
c
n
n
n
11 12 13 1
21 22 23 2
31 32 33 3
2
3
1 0 0
0 1 0
0 0 1
1
|
|
|
|
|
|
' ' '
' ' ' '
Graphical depiction of Gauss-Jordan
Matrix Inversion• [A] [A] -1 = [A]-1 [A] = I• One application of the inverse is to
solve several systems differing only by {c}• [A]{x} = {c}• [A]-1[A] {x} = [A]-1{c}• [I]{x}={x}= [A]-1{c}
• One quick method to compute the inverse is to augment [A] with [I] instead of {c}
Graphical Depiction of the Gauss-Jordan Method with Matrix Inversion
A I
a a a
a a a
a a a
a a a
a a a
a a a
I A
11 12 13
21 22 23
31 32 33
111
121
131
211
221
231
311
321
331
1
1 0 0
0 1 0
0 0 1
1 0 0
0 1 0
0 0 1
Note: the superscript“-1” denotes thatthe original valueshave been convertedto the matrix inverse,not 1/aij
WHEN IS THE INVERSE MATRIX USEFUL?
CONSIDER STIMULUS-RESPONSE CALCULATIONS THAT ARE SO COMMON IN ENGINEERING.
Stimulus-Response Computations• Conservation Laws
massforceheatmomentum
• We considered the conservation of force in the earlier example of a truss
• [A]{x}={c}• [interactions]{response}={stimuli}• Superposition
• if a system subject to several different stimuli, the response can be computed individually and the results summed to obtain a total response
• Proportionality• multiplying the stimuli by a quantity results
in the response to those stimuli being multiplied by the same quantity
• These concepts are inherent in the scaling of terms during the inversion of the matrix
Stimulus-Response Computations
Example
Given the following, determine {x} for the two different loads {c}
174
321
413
362
1121
T
T
c
c
A
cAx
Solution
174
321
413
362
1121
T
T
c
c
A
cAx
{c}T = {1 2 3}x1 = (2)(1) + (-1)(2) + (1)(3) = 3x2 = (-2)(1) + (6)(2) + (3)(3) = 19x3 = (-3)(1) + (1)(2) + (-4)(3) = -13
{c} T= {4 -7 1)x1 = (2)(4) + (-1)(-7) + (1)(1)=16x2 = (-2)(4) + (6)(-7) + (3)(1) = -47x3 = (-3)(4) + (1)(-7) + (-4)(1) = -23
Gauss Seidel Method
• An iterative approach• Continue until we converge within some pre-
specified tolerance of error• Round off is no longer an issue, since you control
the level of error that is acceptable• Fundamentally different from Gauss elimination.
This is an approximate, iterative method particularly good for large number of equations
Gauss-Seidel Method
• If the diagonal elements are all nonzero, the first equation can be solved for x1
• Solve the second equation for x2, etc.x
c a x a x a x
an n
11 12 2 13 3 1
11
To assure that you understand this, write the equation for x2
xc a x a x a x
a
xc a x a x a x
a
xc a x a x a x
a
xc a x a x a x
a
n n
n n
n n
nn n n nn n
nn
11 12 2 13 3 1
11
22 21 1 23 3 2
22
33 31 1 32 2 3
33
1 1 3 2 1 1
Gauss-Seidel Method
• Start the solution process by guessing values of x
• A simple way to obtain initial guesses is to assume that they are all zero
• Calculate new values of xi starting with• x1 = c1/a11
• Progressively substitute through the equations
• Repeat until tolerance is reached
x c a x a x a
x c a x a x a
x c a x a x a
x c a a a ca x
x c a x a a x
x c a x a x a x
1 1 12 2 13 3 11
2 2 21 1 23 3 22
3 3 31 1 32 2 33
1 1 12 13 111
111
2 2 21 1 23 22 2
3 3 31 1 32 2 33 3
0 0
0
/
/
/
/ '
' / '
' ' / '
Gauss-Seidel Method
Example
2 3 1 2
4 1 2 2
3 2 1 1
Given the following augmented matrix, complete one iteration of the Gauss Seidel method.
2 3 1 2
4 1 2 2
3 2 1 1
x c a a a ca x
x
x c a x a a x
x
x c a x a x a x
x
1 1 12 13 111
111
1
2 2 21 1 23 22 2
2
3 3 31 1 32 2 33 3
3
0 0
2 3 0 1 0
2
2
21
0
2 4 1 2 0
1
2 4
16
1 3 1 2 6
1
1 3 12
110
/ '
' / '
' ' / '
GAUSS SEIDEL
Jacobi Iteration
• Iterative like Gauss Seidel• Gauss-Seidel immediately uses the
value of xi in the next equation to predict x i+1
• Jacobi calculates all new values of xi’s to calculate a set of new xi values
FIRST ITERATION
x c a x a x a x c a x a x a
x c a x a x a x c a x a x a
x c a x a x a x c a x a x a
SECOND ITERATION
x c a x a x a x c a x a x a
x c a x a x a x c a x
1 1 12 2 13 3 11 1 1 12 2 13 3 11
2 2 21 1 23 3 22 2 2 21 1 23 3 22
3 3 31 1 32 2 33 3 3 31 1 32 2 33
1 1 12 2 13 3 11 1 1 12 2 13 3 11
2 2 21 1 23 3 22 2 2 21 1
/ /
/ /
/ /
/ /
/
a x a
x c a x a x a x c a x a x a
23 3 22
3 3 31 1 32 2 33 3 3 31 1 32 2 33
/
/ /
Graphical depiction of difference between Gauss-Seidel and Jacobi
2 3 1 2
4 1 2 2
3 2 1 1
Note: We worked the Gauss Seidel method earlier
Given the following augmented matrix, complete one iteration of the Gauss Seidel method and the Jacobi method.
Example
Gauss-Seidel Methodconvergence criterion
a iij
ij
ij s
x x
x,
1
100
as in previous iterative procedures in finding the roots,we consider the present and previous estimates.
As with the open methods we studied previously with onepoint iterations
1. The method can diverge2. May converge very slowly
Class question:where do theseformulas come from?
Convergence criteria for two linear equations
u x xc
a
a
ax
v x xc
a
a
ax
consider the partial derivatives of u and v
u
x
u
x
a
a
v
x
a
a
v
x
1 21
11
12
112
1 22
22
21
222
1 2
12
11
1
21
22 2
0
0
,
,
Convergence criteria for two linear equations cont.
u
x
v
x
u
y
v
y
1
1
Criteria for convergencewhere presented earlierin class materialfor nonlinear equations.
Noting that x = x1 andy = x2
Substituting the previous equation:
Convergence criteria for two linear equations cont.
a
a
a
a21
22
12
11
1 1
This is stating that the absolute values of the slopes mustbe less than unity to ensure convergence.
Extended to n equations:
a a where j n excluding j iii ij 1,
Convergence criteria for two linear equations cont.
a a where j n excluding j iii ij 1,
This condition is sufficient but not necessary; for convergence.
When met, the matrix is said to be diagonally dominant.
Diagonal Dominance
4
3
9
x
x
x
9.05.01.0
4.08.02.0
4.02.01
3
2
1
To determine whether a matrix is diagonally dominant you need to evaluate the values on the diagonal.
Diagonal Dominance
Now, check to see if these numbers satisfy the following rule for each row (note: each row represents a unique equation).a a where j n excluding j iii ij 1,
4
3
9
x
x
x
9.05.01.0
4.08.02.0
4.02.01
3
2
1
x2
x1
Review the conceptsof divergence andconvergence by graphicallyillustrating Gauss-Seidelfor two linear equations
u x x
v x x
:
:
11 13 286
11 9 991 2
1 2
x1
Note: we are convergingon the solution
v x x
u x x
:
:
11 9 99
11 13 2861 2
1 2
CONVERGENCE
x2
x1
Change the order ofthe equations: i.e. changedirection of initial estimates
u x x
v x x
:
:
11 13 286
11 9 991 2
1 2
DIVERGENCE
x2
Improvement of Convergence Using Relaxation
This is a modification that will enhance slow convergence.
After each new value of x is computed, calculate a new valuebased on a weighted average of the present and previousiteration.
x x xinew
inew
iold 1
Improvement of Convergence Using Relaxation
• if = 1unmodified• if 0 < < 1 underrelaxation
• nonconvergent systems may converge• hasten convergence by dampening out
oscillations
• if 1<< 2 overrelaxation• extra weight is placed on the present value• assumption that new value is moving to the
correct solution by too slowly
x x xinew
inew
iold 1