Harjadi Gunawan @2012

Gauss Elimination

1
 Introduction

Roots of a single equation: f ( x)  0

A general set of
equations:
 f 1 ( x1 , x 2 ,  x n )  0

- n equations,
 f 2 ( x1 , x 2 ,  x n )  0
- n unknowns. 
 
 f n ( x1 , x 2 ,  x n )  0

2
Linear Algebraic Equations
a11 x1  a12 x2    a1n xn  b1

a 21 x1  a22 x2    a2 n xn  b2

 
a n1 x1  a n 2 x2    a nn xn  bn

Nonlinear Equations
a11 x1  a12 x1 x2    a1n ( xn ) 5  b1

a21 ( x1 ) 3  a 22 e x2    a 2 n ( x2 ) 3 / xn  b2

 
a x  a x    a x  b
 n1 1 n2 2 nn n n
3
 Review of Matrices
 a 11 a 12  a 1m 
a a 2 m 
a 22  2nd row Elements are indicated by a
[A]  
21 ij
     
 
 a n1 a n2  a nm  n  m
row column

mth column

Row vector: Column vector:

 c1 
[R ]   r1 r2  rn  1  n c 
[C ]   
2

 
Square matrix:  
c m  m  1
- [A]nxm is a square matrix if n=m.
- A system of n equations with n unknonws has a square
coefficient matrix. 4
 Review of Matrices
• Main (principle) diagonal:
[A]nxn consists of elements aii ; i=1,...,n

• Symmetric matrix:
If aij = aji  [A]nxn is a symmetric matrix

• Diagonal matrix:
[A]nxn is diagonal if aij = 0 for all i=1,...,n ; j=1,...,n
and ij

• Identity matrix:
[A]nxn is an identity matrix if it is diagonal with a ii=1
i=1,...,n . Shown as [I] 5
 Review of Matrices
• Upper triangular matrix:

[A]nxn is upper triangular if aij=0 i=1,...,n ; j=1,...,n and i>j

• Lower triangular matrix:

[A]nxn is lower triangular if aij=0 i=1,...,n ; j=1,...,n and i<j

• Inverse of a matrix:

[A]-1 is the inverse of [A]nxn if [A]-1[A] = [I]

• Transpose of a matrix:

[B] is the transpose of [A]nxn if bij=aji Shown as [A] or [A]T

6
 Special Types of Square Matrices

5 1 2 16   a 11  1 
1 3 7 39     
[ A]   [D ]   a 22  [I ]  
1 
2 7 9 6     1 
     
16 39 6 88  a nn   1
Symmetric Diagonal Identity
 a11 a12  a1n   a11 
   
a a a a
[ A]   22  2n  [ A ]   21 22 
      
   
 a nn   a n1   a nn 
Upper Triangular Lower Triangular

7
 Review of Matrices
• Matrix multiplication:

r
Note: [A][B]  [B][A]
cij   aik bkj
k 1
8
 Review of Matrices

• Augmented matrix: is a special way of showing two

matrices together.
a11 a12 
For example A  a 21 a 22  augmented with the
b1   a 11 a 12 b1 
column vector B    is 
b 2   a 21 a 22 b 2 

• Determinant of a matrix:
A single number. Determinant of [A] is shown as |A|.

9
Part 3- Objectives

10
 Solving Small Numbers of Equations

There are many ways to solve a system of linear

equations:
• Graphical method
• Cramer’s rule For n ≤ 3
• Method of elimination
• Numerical methods for solving larger number of
linear equations:
- Gauss elimination
- LU decompositions and matrix inversion
11
 1. Graphical method
• For two equations (n = 2):
a11 x1  a12 x2  b1
a21 x1  a22 x2  b2
• Solve both equations for x2: the intersection of the lines presents
the solution.
 a11  b1
x2    x1   x2  (slope)x1  intercept
 a12  a12
 a21  b2
x2   
 x1 
 a22  a22
• For n = 3, each equation will be a plane on a 3D coordinate system.
Solution is the point where these planes intersect.
• For n > 3, graphical solution is not practical.
12
 Graphical Method -Example
• Solve:
3 x1  2 x2  18
 x1  2 x2  2

• Plot x2 vs. x1, the

intersection of the
lines presents the
solution.

13
 Graphical Method

No solution Infinite solution ill condition

(sensitive to round-off
errors)
14
 Graphical Method
Mathematically
• Coefficient matrices of (a) & (b) are singular. There
is no unique solution for these systems.
Determinants of the coefficient matrices are zero
and these matrices can not be inverted.

• Coefficient matrix of (c) is almost singular. Its

inverse is difficult to take. This system has a unique
solution, which is not easy to determine
numerically because of its extreme sensitivity to
round-off errors.

15
 2.Determinants and Cramer’s Rule
 Determinant can be illustrated for a set of three
equations:

 A . x    B 
 Where [A] is the coefficient matrix:

 a 11 a 12 a 13 

 A    a 21 a 22 a 23 

 a 31 a 32 a 33 

16
 Cramer’s Rule
a11 a12 a13 b1 a12 a13
D  a21 a22 a23 b2 a22 a23
a31 a32 a33 b3 a32 a33
x1 
a22 a23 D
D11   a22 a33  a32 a23
a32 a33 a11 b1 a13
a21 a23 a21 b2 a23
D12   a21 a33  a31 a23 a31 b3 a33
a31 a33 x2 
a21 a22 D
D13   a21 a32  a31 a22
a31 a32 a11 a12 b1
a21 a22 b2
a22 a23 a21 a23 a21 a22 a31 a32 b3
D  a11  a12  a13 x3 
a32 a33 a31 a33 a31 a32 D
17
 Cramer’s Rule
• For a singular system D = 0  Solution can not
be obtained.

• For large systems Cramer’s rule is not practical

because calculating determinants is costly.

18
19
20
 3. Method of Elimination
• The basic strategy is to successively solve one of
the equations of the set for one of the
unknowns and to eliminate that variable from
the remaining equations by substitution.

• The elimination of unknowns can be extended to

systems with more than two or three equations.
However, the method becomes extremely
tedious to solve by hand.

21
 Elimination of Unknowns Method
2.5x1 + 6.2x2 = 3.0
Given a 2x2 set of equations:
4.8x1 - 8.6x2 = 5.5

• Multiply the 1st eqn by 8.6 21.50x1 + 53.32x2=25.8

and the 2nd eqn by 6.2  29.76x1 – 53.32x2=34.1
• Add these equations  51.26 x1 + 0 x2 = 59.9

• Solve for x1 : x1 = 59.9/51.26 = 1.168552478

• Using the 1st eqn solve for x2 :
x2 =(3.0–2.5*1.168552478)/6.2 = 0.01268045242
• Check if these satisfy the 2nd eqn:
4.8*1.168552478–8.6*0.01268045242 = 5.500000004

(Difference is due to the round-off errors).

22
 Naive Gauss Elimination Method

• It is a formalized way of the previous elimination

technique to large sets of equations by developing a
systematic scheme or algorithm to eliminate
unknowns and to back substitute.

• As in the case of the solution of two equations, the

technique for n equations consists of two phases:
1. Forward elimination of unknowns.
2. Back substitution.

23
 Naive Gauss Elimination Method
• Consider the following system of n equations.

a11x1 + a12x2 + ... + a1nxn = b1 (1)

a21x1 + a22x2 + ... + a2nxn = b2 (2)

...
an1x1 + an2x2 + ... + annxn = bn (n)

Form the augmented matrix of [A|B].

Step 1 : Forward Elimination: Reduce the system to an upper triangular

system.

1.1- First eliminate x1 from 2nd to nth equations.

- Multiply the 1st eqn. by a21/a11 & subtract it from the 2nd equation.
This is the new 2nd eqn.
- Multiply the 1st eqn. by a31/a11 & subtract it from the 3rd equation.
This is the new 3rd eqn.
...
- Multiply the 1st eqn. by an1/a11 & subtract it from the nth equation.
This is the new nth eqn. 24
Naive Gauss Elimination Method (cont’d)
Note:
- In these steps the 1st eqn is the pivot equation and a11 is
the pivot element.
- Note that a division by zero may occur if the pivot
element is zero. Naive-Gauss Elimination does not check
for this.
a11 a12 a13  a1n   x 1  b1 
0 a22 a23  a2n   x 2  b2 
    
The modified system is  0 a32 a33  a3n   x 3   b3 
 
          
   
indicates that the  0 an2 an3  ann   x n  bn 
system is modified once.

25
Naive Gauss Elimination Method (cont’d)

1.2- Now eliminate x2 from 3rd to nth equations.

 a 11 a 12 a 13  a 1n   x 1  b1 
 0 a 22 a 23  a 2 n   x  b  
The modified system is    2   2 
 0 0 a 33
  a 3n   x 3   b 3 
     
     
   
 0 0 a n3   
a nn  x n  b n 

Repeat steps (1.1) and (1.2) upto (1.n-1).

�a11 a12 a13 L a1n ��x 1 � � b1 �
we will get this upper �0 a � a � L a 2� ��x � � b ��
� 22 23 n �� 2 � � 2 �
triangular system  �0 �
0 a33� L a3� � �
� � �
x  b �
�
�
n �3� � 3 �
� �
�M M M O M � �M� � M �
� � � ( n 1) �
� 0
���� 0 0 0 ( n 1) �
a nn �x n � �bn �
26
 Naive Gauss Elimination Method (cont’d)
Step 2 : Back substitution
Find the unknowns starting from the last equation.
1. Last equation involves only xn. Solve for it. ( n 1)
b
xn  n
( n 1)
a nn
2. Use this xn in the (n-1)th equation and solve for xn-1.
...
3. Use all previously calculated x values in the 1st eqn
and solve for x1.
n
bi
( i 1)
 a
j  i 1
i 1
ij xj
xi  ( i 1)
for i  n-1, n-2, ..., 1
a ii 27
Summary of Naive Gauss Elimination Method

28
 Pseudo-code of Naive Gauss Elimination Method

(a) Forward Elimination (b) Back substitution

k,j

29
Naive Gauss Elimination Method
Example 1

Solve the following system using Naive Gauss Elimination.

6x1 – 2x2 + 2x3 + 4x4 = 16

12x1 – 8x2 + 6x3 + 10x4 = 26
3x1 – 13x2 + 9x3 + 3x4 = -19
-6x1 + 4x2 + x3 - 18x4 = -34

Step 0: Form the augmented matrix

6 –2 2 4 | 16
12 –8 6 10 | 26 R2-2R1
3 –13 9 3 | -19 R3-0.5R1
-6 4 1 -18 | -34 R4-(-R1) 30
 Naive Gauss Elimination Method
Example 1 (cont’d)
Step 1: Forward elimination

1. Eliminate x1 6–2 2 4 | 16 (Does not change. Pivot is 6)

0 –4 2 2 | -6
0 –12 8 1 | -27 R3-6R2
0 2 3 -14 | -18 R4-(-0.5R2)

2. Eliminate x2 6–2 2 4 | 16 (Does not change.)

0 –4 2 2 | -6 (Does not change. Pivot is-4)
0 0 2 -5 | -9
0 0 4 -13 | -21 R4-2R3

3. Eliminate x3 6 –2 2 4 | 16 (Does not change.)

0 –4 2 2 | -6 (Does not change.)
0 0 2 -5 | -9 (Does not change. Pivot is 2)
0 0 0 -3 | -3

31
 Naive Gauss Elimination Method
Example 1 (cont’d)
Step 2: Back substitution

Find x4 x4 =(-3)/(-3) = 1

Find x3 x3 =(-9+5*1)/2 = -2

Find x2 x2 =(-6-2*(-2)-2*1)/(-4) = 1

Find x1 x1 =(16+2*1-2*(-2)-4*1)/6= 3

32
 Naive Gauss Elimination Method Example 2
(Using 6 Significant Figures)
3.0 x1 - 0.1 x2 - 0.2 x3 = 7.85
0.1 x1 + 7.0 x2 - 0.3 x3 = -19.3 R2-(0.1/3)R1
0.3 x1 - 0.2 x2 + 10.0 x3 = 71.4 R3-(0.3/3)R1

Step 1: Forward elimination

3.00000 x1- 0.100000 x2 - 0.200000 x3 = 7.85000

7.00333 x2 - 0.293333 x3 = -19.5617
- 0.190000 x2 + 10.0200 x3 = 70.6150

3.00000 x1- 0.100000 x2 - 0.20000 x3 = 7.85000

7.00333 x2 - 0.293333 x3 = -19.5617

10.0120 x3 = 70.0843

33
 Naive Gauss Elimination Method Example 2
(cont’d)
Step 2: Back substitution

x3 = 7.00003
x2 = -2.50000
x1 = 3.00000

Exact solution:
x3 = 7.0
x2 = -2.5
x1 = 3.0

34
 Pitfalls of Gauss Elimination Methods

1. Division by zero
2 x2 + 3 x 3 = 8
a11 = 0
4 x1 + 6 x2 + 7 x3 = -3 (the pivot element)
2 x1 + x 2 + 6 x3 = 5

It is possible that during both elimination and back-

substitution phases a division by zero can occur.
2. Round-off errors
In the previous example where up to 6 digits were kept
during the calculations and still we end up with close to the
real solution.
x3 = 7.00003, instead of x3 = 7.0

35
 Pitfalls of Gauss Elimination (cont’d)
3. Ill-conditioned systems
x1 + 2x2 = 10
1.1x1 + 2x2 = 10.4  x1 = 4.0 & x2 = 3.0

x1 + 2x2 = 10
1.05x1 + 2x2 = 10.4  x1 = 8.0 & x2 = 1.0

Ill conditioned systems are those where small changes in

coefficients result in large change in solution. Alternatively, it
happens when two or more equations are nearly identical,
resulting a wide ranges of answers to approximately satisfy
the equations. Since round off errors can induce small
changes in the coefficients, these changes can lead to large
solution errors.
36
 Pitfalls of Gauss Elimination (cont’d)
4. Singular systems.
• When two equations are identical, we would loose one
degree of freedom and be dealing with case of n-1
equations for n unknowns.

To check for singularity:

• After getting the forward elimination process and
getting the triangle system, then the determinant for
such a system is the product of all the diagonal
elements. If a zero diagonal element is created, the
determinant is Zero then we have a singular system.
• The determinant of a singular system is zero.

37
 Techniques for Improving Solutions
1. Use of more significant figures to solve for the
round-off error.
2. Pivoting. If a pivot element is zero, elimination step
leads to division by zero. The same problem may arise,
when the pivot element is close to zero. This Problem
can be avoided by:
 Partial pivoting. Switching the rows so that the
largest element is the pivot element.
 Complete pivoting. Searching for the largest element
in all rows and columns then switching.
3. Scaling
 Solve problem of ill-conditioned system.
 Minimize round-off error

38
 Partial Pivoting

Before each row is normalized, find the largest

available coefficient in the column below the
pivot element. The rows can then be switched
so that the largest element is the pivot
element so that the largest coefficient is used
as a pivot.

39
 Use of more significant figures to solve for the round-off error
:Example.

Use Gauss Elimination to solve these 2 equations:

(keeping only 4 sig. figures)

0.0003 x1 + 3.0000 x2 = 2.0001

1.0000 x1 + 1.0000 x2 = 1.000

0.0003 x1 + 3.0000 x2 = 2.0001

- 9999.0 x2 = -6666.0

Solve: x2 = 0.6667 & x1 = 0.0

The exact solution is x2 = 2/3 & x1 = 1/3 40

Use of more significant figures to solve for the round-
off error :Example (cont’d).

2 2.0001  3(2 / 3)
x2  x1 
3 0.0003
Significant
Figures x2 x1

3 0.667 -3.33
4 0.6667 0.000
5 0.66667 0.3000
6 0.666667 0.33000
7 0.6666667 0.333000

41
 Pivoting: Example
Now, solving the pervious example using the partial
pivoting technique:
1.0000 x1+ 1.0000 x2 = 1.000
0.0003 x1+ 3.0000 x2 = 2.0001
The pivot is 1.0
1.0000 x1+ 1.0000 x2 = 1.000
2.9997 x2 = 1.9998
x2 = 0.6667 & x1=0.3333
Checking the effect of the # of significant digits:
# of dig x2 x1 Ea% in x1

4 0.6667 0.3333 0.01

5 0.66667 0.33333 0.001
42
 Scaling: Example
• Solve the following equations using naïve gauss elimination:
(keeping only 3 sig. figures)
2 x1+ 100,000 x2 = 100,000
x1 + x2 = 2.0
• Forward elimination:
2 x1+ 100.000 x2 = 100.000
- 50,0000 x2 = -50,0000
Solve x2 = 1.00 & x1 = 0.00

• The exact solution is x1 = 1.00002 & x2 = 0.99998

43
 Scaling: Example (cont’d)
B) Using the scaling algorithm to solve:
2 x1+ 100,000 x2 = 100,000
x1 + x2 = 2.0
Scaling the first equation by dividing by 100,000:
0.00002 x1+ x2 = 1.0
x1 + x2 = 2.0
Rows are pivoted:
x1 + x2 = 2.0
0.00002 x1+ x2 = 1.0
Forward elimination yield:
x1 + x2 = 2.0
x2 = 1.00 44
 Scaling: Example (cont’d)
C) The scaled coefficient indicate that pivoting is necessary.
We therefore pivot but retain the original coefficient to give:
x1 + x2 = 2.0
2 x1+ 100,000 x2 = 100,000

Forward elimination yields:

x1 + x2 = 2.0
100,000 x2 = 100,000
Solve: x2 = 1.00 & x1 = 1.00

Thus, scaling was useful in determining whether pivoting was

necessary, but the equation themselves did not require
scaling to arrive at a correct result.
45
 Determinate Evaluation

The determinate can be simply evaluated at the end of

the forward elimination step, when the program
employs partial pivoting:

( 1)
( n 1) p
D a a a L a
' ''
11 22 33 nn

where:
p represents the number of times that rows are pivoted

46
Example: Gauss Elimination
2x 2 x 4 0
2x 1 2x 2 3x 3 2x 4  2
4x 1 3x 2 x 4  7
6x 1 x 2 6 x 3 5x 4 6

a) Forward Elimination

�
0 2 0 1 0� 6 1 6 5
� 6�
� � � �
2 2 3 2
� 2 � 2 2 3 2
� 2 �
��
�R 4
R 1 ��
�
�
4 3 0 1 7 � �
4 3 0 1 7 �
� � � �
6 1 6 5
� 6� 0 2 0 1
� 0�
47
 Example: Gauss Elimination (cont’d)
6 1 6 5
� 6�
� �
2 2 3 2
� 2 �R 2  0.33333 �
R1
�
4 3 0 1 7 �R 3  0.66667 �
R1
� �
0 2 0 1
� 0�
�
6 1 6 5 6 �
� �
0 1.6667
� 5 3.6667 4 �
��
�R 3
R 2 ��
�
�
0 3.6667 4 4.3333 11�
� �
0
� 2 0 1 0 �
�
6 1 6 5 6 �
� �
0 3.6667 4 4.3333
� 11�
�
0 1.6667 5 3.6667 4 �
� �
0
� 2 0 1 0 �
 Example: Gauss Elimination (cont’d)
�
6 1 6 5 6 �
� �
0 3.6667 4 4.3333
� 11�
�
0 1.6667 5 3.6667 4 �R 3  0.45455 �
R2
� �
0
� 2 0 1 0 �R 4  0.54545 �
R2

�
6 1 6 5 6 �
� �
0 3.6667
� 4 4.3333 11 �
�
0 0 6.8182 5.6364 9.0001�
� �
0
� 0 2.1818 3.3636 5.9999 �R 4  0.32000 �
R3

�
6 1 6 5 6 �
� �
0 3.6667
� 4 4.3333 11 �
�
0 0 6.8182 5.6364 9.0001�
� �
0
� 0 0 1.5600 3.1199 �
49
 Example: Gauss Elimination (cont’d)
�
6 1 6 5 6 �
� �
0 3.6667
� 4 4.3333 11 �
�
0 0 6.8182 5.6364 9.0001�
� �
0
� 0 0 1.5600 3.1199 �

b) Back Substitution
3.1199
x4   1.9999
1.5600
9.0001  5.6364 ( 1.9999 )
x3   0.33325
6.8182
11  4.3333 ( 1.9999 )  4 ( 0.33325 )
x2   1.0000
3.6667
6  5 ( 1.9999 )  6 ( 0.33325 )  1( 1.0000 )
x1   0.50000 50
6
 Gauss-Jordan Elimination

• It is a variation of Gauss elimination. The

major differences are:
– When an unknown is eliminated, it is eliminated
from all other equations rather than just the
subsequent ones.
– All rows are normalized by dividing them by their
pivot elements.
– Elimination step results in an identity matrix.
– It is not necessary to employ back substitution to
obtain solution.

51
 Gauss-Jordan Elimination- Example

�
0 2 0 1 0� 1 0.16667 1 0.83335
� 1�
� � �� �R 4 � �
2 2 3 2
� 2 �R 1��� 2
� 2 3 2 2 �
�
4 3 0 1 7 � R 4 / 6.0 �
4 3 0 1 7 �
� � � �
6 1 6 5
� 6� 0
� 2 0 1 0�

1 0.16667 1 0.83335
� 1�
� �
2
� 2 3 2 2 �R 2  2 �
R1
�
4 3 0 1 7 �R 3  4 �
R1
� �
0
� 2 0 1 0�

52
 1 0.16667 1 0.83335
� 1�
� �
0 1.6667
� 5 3.6667 2 �
�
0 3.6667 4 4.3334 7 �
� �
0
� 2 0 1 0�

Dividing the 2nd row by 1.6667 and reducing the second column.
(operating above the diagonal as well as below) gives:

�
1 0 1.5 1.2000 1.4000 �
� �
0
� 1 2.9999 2.2000 2.4000�
�
0 0 15.000 12.400 19.800�
� �
0
� 0 5.9998 3.4000 4.8000 �
Divide the 3rd row by 15.000 and make the elements in the 3rd
Column zero.
53
 �
1 0 0 0.04000 0.58000 �
� �
0
� 1 0 0.27993 1.5599 �
�
0 0 1 0.82667 1.3200 �
� �
0
� 0 0 1.5599 3.1197 �
Divide the 4th row by 1.5599 and create zero above the diagonal in the fourth
column.

�
1 0 0 0 0.49999�
� �
0
� 1 0 0 1.0001 �
�
0 0 1 0 0.33326�
� �
0
� 0 0 1 1.9999 �

Note: Gauss-Jordan method requires almost 50 % more operations than

Gauss elimination; therefore it is not recommended to use it.
54