Matrix Operations with SCILAB

By
Gilberto E. Urroz, Ph.D., P.E.

Distributed by

i nfoClearinghouse.com

2001 Gilberto E. Urroz

All Rights Reserved

A "zip" file containing all of the programs in this document (and other
SCILAB documents at InfoClearinghouse.com) can be downloaded at the
following site:
http://www.engineering.usu.edu/cee/faculty/gurro/Software_Calculators/Scil
ab_Docs/ScilabBookFunctions.zip

The author's SCILAB web page can be accessed at:

http://www.engineering.usu.edu/cee/faculty/gurro/Scilab.html

Please report any errors in this document to: [email protected]

Gaussian and Gauss-Jordan elimination

Gaussian elimination using a system of equations
Forward elimination
Backward substitution
Gaussian elimination using matrices
Gaussian elimination algorithm
Augmenting a matrix in SCILAB
Algorithm for first step in forward elimination
SCILAB example for first step in forward elimination
Algorithm for second step in forward elimination
SCILAB example for second step in forward elimination
Algorithm for third and subsequent steps in the forward elimination
SCILAB example implementing the full forward elimination algorithm
Algorithm for backward substitution
SCILAB example for calculating the unknown xs
SCILAB function for Gaussian elimination
Pivoting
Gaussian elimination with partial pivoting
Solving multiple set of equations with the same coefficient matrix
Gaussian elimination for multiple sets of linear equations
Calculating an inverse matrix using Gaussian elimination
Calculating the determinant with Gaussian elimination
Gauss-Jordan elimination
Calculating the inverse through Gauss-Jordan elimination

2
2
2
2
3
4
4
5
5
6
6
7
8
8
9
9
11
13
15
16
17
17
20
21

Eigenvalues and eigenvectors

Calculating the first eigenvector
Calculating eigenvectors with a user-defined function
Generating the characteristic equation for a matrix
Generalized eigenvalue problem

22
24
25
27
29

Sparse matrices
Creating sparse matrices
Getting information about a sparse matrix
Sparse matrix with unit entries
Sparse identity matrices
Sparse matrix with random entries
Sparse matrices with zero entries
Visualizing sparse matrices
Factorization of sparse matrices
Solution to system of linear equations involving sparse matrices
Solution to system of linear equations using the inverse of a sparse matrix

31
31
33
33
34
34
35
35
36
39
40

Solution to a system of linear equations with a tri-diagonal matrix of coefficients

Solution to tri-diagonal systems of linear equations using sparse matrices

41
43

Iterative solutions to systems of linear equations

Jacobi iterative method

45
46

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

Gaussian and Gauss-Jordan elimination

Gaussian elimination is a procedure by which the square matrix of coefficients belonging to a
system of n linear equations in n unknowns is reduced to an upper-triangular matrix (echelon
form) through a series of row operations. This procedure is known as forward elimination.
The reduction of the coefficient matrix to an upper-triangular form allows for the solution of
all n unknowns, utilizing only one equation at a time, in a procedure known as backward
substitution.

Gaussian elimination using a system of equations

To illustrate the Gaussian elimination procedure we will use the following system of 3
equations in 3 unknowns:
2X +4Y+6Z = 14,
3X -2Y+ Z = -3,
4X +2Y -Z = -4.
Forward elimination
First, we replace the second equation by (equation 2 (3/2)* equation 1), and the third by
(equation 1 (4/2)*equation 1), to get
2X +4Y+6Z = 14,
-8Y-8Z = -24,
-6Y-13Z = -32.
Next, replace the third equation with (equation 3 - (6/8)*equation 2), to get
2X +4Y+6Z = 14,
-8Y-8Z = -24,
-7Z = -14.
Backward substitution
The system of equations is now in an upper-triangular form, allowing us to solve for Z first, as
Z = -14/-7 = 2.
We then replace this value into equation 2 to solve for Y, as
Y = (-24 +8Z)/(-8) = (-24 +82)/(-8) = 1.
Finally, we replace the values of Y and Z into equation 1 to solve for X as
X = (14-4Y-6Z)/2 = (14-4(1)-6(2))/2 =-1.

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

The solution is, therefore,

X = -1, Y = 1, Z = 2.

Gaussian elimination using matrices

The system of equations used in the example above, i.e.,
2X +4Y+6Z = 14,
3X -2Y+ Z = -3,
4X +2Y -Z = -4.
Can be written as a matrix equation Ax = b, if we use:

6
2 4
X
14

A = 3 2 1 , x = Y , b = 3.
4 2 1
Z
4

To obtain a solution to the system matrix equation using Gaussian elimination, we first create
what is known as the augmented matrix corresponding to A, i.e.,

A aug

2 4
6 14

= 3 2 1 3 .
4 2 1 4

The matrix Aaug is nothing more than the original matrix A with a new row, corresponding to the
elements of the vector b, added (i.e., augmented) to the right of the rightmost column of A.
Once the augmented matrix is put together, we can proceed to perform row operations on it
that will reduce the original A matrix into an upper-triangular matrix similar to what we did
with the system of equations shown earlier. If you perform the forward elimination by hand,
you would write the following:

A aug

2 4
6 14 2 4
6
14 2 4
6 14

= 3 2 1 3 0 8 8 24 . 0 8 8 24
4 2 1 4 0 6 13 32 0 0 7 14

The symbol ( is congruent to) indicates that what follows is equivalent to the previous
matrix with some row (or column) operations involved.
After the augmented matrix is reduced as shown above, we can proceed to perform the
backward substitution by converting the augmented matrix into a system of equations and
solving for x3, x2, and x1 in that order, as performed earlier.

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

Gaussian elimination algorithm

In this section we present the algorithm for performing Gaussian elimination on an augmented
matrix. The augmented matrix results from the system of linear equations represented by the
matricial equation Ax = b. The matrix A and vectors x and b are given by

a11
a
A = 21
M

a n1

a12
a 22
M
an 2

L a1m
x1
b1

L a2m
x
, x = 2 , b = 2
M
M
O M

L a nm
xm
bn

The augmented matrix is:

Aaug

a11

a
= 21
M

a n1

L a1m b1

L a 2 m b2
.
O M
M

L a nm bn

a12
a 22
M
an2

Thus, the augmented matrix can be referred to as (Aaug)nx(n+1) = [aij], with ai,n+1 = bi, for i = 1,2,
, n.
__________________________________________________________________________________
Augmenting a matrix in SCILAB
In SCILAB augmenting a matrix A by a column vector b is straightforward. For example, for the
linear system introduced earlier, namely,

6
2 4
14

A = 3 2 1 , b = 3,
4 2 1
4

an augmented matrix, Aaug, is obtained as follows:

-->A = [2,4,6;3,-2,1;4,2,-1], b = [1;-3;-4]
A =
!
!
!

2.
3.
4.

4.
- 2.
2.

6. !
1. !
- 1. !

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

!
1. !
! - 3. !
! - 4. !
-->A_aug = [A b]
A_aug =
!
!
!

2.
3.
4.

4.
- 2.
2.

6.
1.
- 1.

1. !
- 3. !
- 4. !

__________________________________________________________________________________
Algorithm for first step in forward elimination
After the augmented matrix has been created the first elimination pass will fill with zeros
those values in the first column below a11, i.e.,

a11

0
M

a12
a * 22
M
a *n 2

a1,n +1
L a1m

L a * 2 m a *2,n +1
.
M
O
M

L a *nm a *n ,n +1

The procedure is indicated by the expressions:

a*21 = a*31 = = a*n1 = 0; j = 1 (first column)
a*2,j = a2j - a1j a21/a11; j = 2,3, . . . , n+1
a*3j = a3,j - a1j a31/a11; j = 2,3, . . . , n+1
.
.
.
.
.
.
.
.
.
a*n,j = an,j - a1j an1/a11; j = 2,3, . . . , n+1
or, more concisely, as
a*ij = aij - a1j ai1/a11; for i = 2,3, n; and, j = 2,3,. . .,n+1
a*ij = 0; for i = 2,3, , n; and, j = 1
__________________________________________________________________________________
SCILAB example for first step in forward elimination
To illustrate this first step using SCILAB, we first copy the augmented matrix we created into
an array a by using:
-->a = A_aug
a =

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

!
!
!

2.
3.
4.

4.
- 2.
2.

6.
1.
- 1.

1. !
- 3. !
- 4. !

Next, we define n as 3 and implement the first step in the forward elimination through the use
of for..end constructs:
--> n = 3;
-->for i=2:n, for j=2:n+1, a(i,j)=a(i,j)-a(1,j)*a(i,1)/a(1,1); end; a(i,1) = 0;
end;

To see the modified augmented matrix we use:

-->a
a =
!
!
!

2.
0.
0.

4.
- 8.
- 6.

6.
- 8.
- 13.

14. !
- 24. !
- 32. !

__________________________________________________________________________________
Algorithm for second step in forward elimination
After the second elimination pass, the generic augmented matrix is:

a11

0
0

M
0

a12
a *22
0
M
0

a13
a *23
a * *33
M
a * *n 3

a1, n+1
L
a1n

L a *2 n a *2,n +1
L a * *3n a * *3,n +1

M
O
M

L a * *nn a * *n ,n +1

with rows i = 3, 4, n, recalculated as:

a**ij = a*ij - a*2j a*i2/a*22; for i = 3,4, n; and, j = 3,4,. . .,n+1
a**ij = 0; for i = 3,4,,n; and, j = 2
__________________________________________________________________________________
SCILAB example for second step in forward elimination
For the SCILAB example under consideration, this second step will be the final step in the
forward elimination. This step is implemented as follows:
--> for i=3:n, for j=3:n+1, a(i,j)=a(i,j)-a(2,j)*a(i,2)/a(2,2); end; a(i,2) =
0; end;

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

The result is:

a
!
!
!

=
2.
0.
0.

4.
- 8.
0.

6.
- 8.
- 7.

14. !
- 24. !
- 14. !

__________________________________________________________________________________

Algorithm for third and subsequent steps in the forward elimination

For a general nx(n+1) augmented matrix, we can predict that in the next (third) elimination
pass the new coefficients in the augmented matrix can be written as:
a***ij = a**ij - a**3j a**i3/a**33; for i = 4,5, n; and, j = 4,5,. . .,n+1
a***ij = 0; for i = 4,5,,n; and, j = 3
Notice that the number of asterisks in the new coefficients calculated for each elimination pass
corresponds to the number of the elimination pass, i.e., we calculated a*ij in the first pass, a**ij
in the second pass, and so on. We may venture to summarize all the equations we have
written for the new coefficients as:

a(k)ij = a(k-1)ij - a(k-1)kj a(k-1)ik/a(k-1)kk; for i = k+1,k+2, n; and, j = k+1,k+2,. . .,n+1,

a(k)ij = 0; for i = k+1,k+2,,n; and, j = 1,2,,k
where k = 1,2,. . .,n-1.
Notice that k represents the order of the elimination pass, i represents the row index, and j
represents the column index.
After the forward elimination passes have been completed, the modified augmented matrix
should look like this:

a11

0
0

M
0

a12
a (1) 22
0
M
0
0

a13
a (1) 23
a ( 2) 33
M
0
0

L
a1,n 1
L
a (1) 2,n 1
L
a ( 2) 3,n 1
O
M
L a ( n 2) n 1,n 1
L
0

a1n
a (1) 2 n
a ( 2 ) 3n
M

a ( n 2 ) n 1, n
a ( n 1) n ,n

( n2)

a
n 1, n +1

a ( n 1) n ,n +1
a1,n +1
a (1) 2,n +1
a ( 2) 3,n +1
M

If we use the array aij to store the modified coefficients, we can simply write:
aij = aij - akjaik/akk; i = k+1,k+2,...,n; j = k+1,k+2,...,n+1.
aij = 0; for i = k+1,k+2,,n; and, j = k

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

for k = 1,2,...,n-1.
Note: By using the array aij to store the newly calculated coefficients you loose the information
stored in the original array aij. Therefore, when implementing the algorithm in SCILAB, if you
need to keep the original array available for any reason, you may want to copy it into a
different array as we did in the SCILAB example presented earlier.
__________________________________________________________________________________

SCILAB example implementing the full forward elimination algorithm

For the 3x4 augmented matrix presented earlier, the following SCILAB commands implement
the algorithm for forward elimination:
-->a = A_aug
a =
!
!
!

2.
3.
4.

4.
- 2.
2.

6.
1.
- 1.

1. !
- 3. !
- 4. !

-->n=3;
-->for k=1:n-1, for i=k+1:n, for j=k+1:n+1, a(i,j)=a(i,j)-a(k,j)*a(i,k)/a(k,k);
end; for j = 1:k, a(i,j) = 0; end; end; end;

-->a
a =
!
!
!

2.
0.
0.

4.
- 8.
0.

6.
- 8.
- 7.

14. !
- 24. !
- 14. !

__________________________________________________________________________________
Algorithm for backward substitution
The next step in the algorithm is to calculate the solution x1, x2,..., xn, by back substitution,
starting with

xn = an,n+1/ann.
and continuing with

xn-1 = (an-1,n+1-an-1,nxn)/an-1,n-1

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

xn-2 = (an-2,n+1-an-2,n-1xn-1-an-2,nxn)/an-2,n-2
.
.

x1 =

a1,n +1 a1k x k
k =2

a11

In summary, the solution is obtained by first calculating:

xn = an,n+1/ann.
and, then, calculating

xi =

ai , n +1

k = i +1

aii

ik

xk
,

for i = n-1, n-2, ..., 2,1 (i.e., counting backwards from (n-1) to 1).
__________________________________________________________________________________
SCILAB example for calculating the unknown xs
Using the current value of the matrix a that resulted from the forward elimination, we can
calculate the unknowns in the linear system as follows. First, the last unknown is:
-->x(n) = a(n,n+1)/a(n,n);

Then, we calculate the remaining unknowns using:

-->for i = n-1:-1:1, sumk=0;
x(i)=(a(i,n+1)-sumk)/a(i,i); end;

for

k=i+1:n,

sumk=sumk+a(i,k)*x(k);

end;

The solution is:

-->x
x =
! - 1. !
!
1. !
!
2. !

_________________________________________________________________________________
SCILAB function for Gaussian elimination

Download at InfoClearinghouse.com

2001 - Gilberto E. Urroz

The following SCILAB function implements the solution of the system of linear equations,
Ax=b. The functions arguments are a nxn matrix A and a nx1 vector b. The function returns
the solution x.
function [x] = gausselim(A,b)
//This function obtains the solution to the system of
//linear equations A*x = b, given the matrix of coefficients A
//and the right-hand side vector, b
[nA,mA] = size(A)
[nb,mb] = size(b)
if nA<>mA then
error('gausselim - Matrix A must be square');
abort;
elseif mA<>nb then
error('gausselim - incompatible dimensions between A and b');
abort;
end;
a = [A b];

//Matrix augmentation

//Forward elimination
n = nA;
for k=1:n-1
for i=k+1:n
for j=k+1:n+1
a(i,j)=a(i,j)-a(k,j)*a(i,k)/a(k,k);
end;
end;
end;
//Backward substitution
x(n) = a(n,n+1)/a(n,n);
for i = n-1:-1:1
sumk=0
for k=i+1:n
sumk=sumk+a(i,k)*x(k);
end;
x(i)=(a(i,n+1)-sumk)/a(i,i);
end;
//End function

Note: In this function we did not include the statements that produce the zero values in the
lower triangular part of the augmented matrix. These terms are not involved in the solution
at all, and were used earlier only to illustrate the effects of the Gaussian elimination.
Application of the function gausselim to the problem under consideration produces:
->A = [2,4,6;3,-2,1;4,2,-1]; b = [14;-3;-4];
-->getf('gausselim')

//Loading the function

Download at InfoClearinghouse.com

10

2001 - Gilberto E. Urroz

-->gausselim(A,b)
ans =

//Function call without assignment

! - 1. !
!
1. !
!
2. !
-->x = gausselim(A,b)
x =

//Function call with assignment

! - 1. !
!
1. !
!
2. !

This simple implementation of Gaussian elimination is commonly referred to as nave Gaussian

elimination, because it does not account for the possibility of having zero terms in the
diagonal. Such terms will produce a division by zero in the Gaussian elimination algorithm
effectively terminating the process in SCILAB. A technique known as partial pivoting can be
implemented to account for the case of zero diagonal terms. Details on pivoting are presented
next.
__________________________________________________________________________________

Pivoting
If you look carefully at the row operations in the examples shown above, you will notice that
many of those operations divide a row by its corresponding element in the main diagonal. This
element is called a pivot element, or simply, a pivot. In many situations it is possible that the
pivot element become zero, in which case a division by zero occurs. Also, to improve the
numerical solution of a system of equations using Gaussian or Gauss-Jordan elimination, it is
recommended that the pivot be the element with the largest absolute value in a given column.
This operation is called partial pivoting. To follow this recommendation is it often necessary
to exchange rows in the augmented matrix while performing a Gaussian or Gauss-Jordan
elimination.
While performing pivoting in a matrix elimination procedure, you can improve the numerical
solution even more by selecting as the pivot the element with the largest absolute value in the
column and row of interest. This operation may require exchanging not only rows, but also
columns, in some pivoting operations. When row and column exchanges are allowed in
pivoting, the procedure is known as full pivoting.
When exchanging rows and columns in partial or full pivoting, it is necessary to keep track of
the exchanges because the order of the unknowns in the solution is altered by those exchanges.
One way to keep track of column exchanges in partial or full pivoting mode, is to create a
permutation matrix P = Inn, at the beginning of the procedure. Any row or column exchange
required in the augmented matrix Aaug is also registered as a row or column exchange,
respectively, in the permutation matrix. When the solution is achieved, then, we multiply the
permutation matrix by the unknown vector x to obtain the order of the unknowns in the
solution. In other words, the final solution is given by Px = b, where b is the last column of
the augmented matrix after the solution has been found.

Download at InfoClearinghouse.com

11

2001 - Gilberto E. Urroz

A function such as lu in SCILAB automatically take care of using partial or full pivoting in the
solution of linear systems. No special provision is necessary in the call to this function to
activate pivoting. The function lu take care of reincorporating the permutation matrix into
the final result.
__________________________________________________________________________________
Consider, as an example, the linear system shown below:
2Y + 3Z = 7,
2X +
3Z = 13,
8X +16Y - Z = -3.
The augmented matrix is:

A aug

7
0 2 3

= 2 0 3 13 .
8 16 1 3

The zero in element (1,1) will not allow the simple Gaussian elimination to proceed since a
division by zero will be required. Trying a solution with function gausselim, described earlier,
produces an error:
-->A = [0,2,3;2,0,3;8,16,-1], b = [7;13;-3]
A =
!
!
!
b

0.
2.
8.
=

2.
0.
16.

3. !
3. !
- 1. !

!
7. !
!
13. !
! - 3. !
-->getf('gausselim')
-->x = gausselim(A,b)
!--error
27
division by zero...
at line
26 of function gausselim
x = gausselim(A,b)

called by :

As expected, gausselim produces a division-by-zero error. However, lu, which already

incorporates pivoting, will produce the following solution through LU decomposition:
-->[L,U,P] = lu(A)
P =
!
!
!
U

0.
0.
1.
=

0.
1.
0.

1. !
0. !
0. !

Download at InfoClearinghouse.com

12

2001 - Gilberto E. Urroz

!
!
!
L
!
!
!

8.
0.
0.
=

16.
- 4.
0.

1.
.25
0.

0.
1.
- .5

- 1.
!
3.25 !
4.625 !

0. !
0. !
1. !

The right-hand side vector is modified by using the permutation matrix:

-->c = P*b
c =
! - 3. !
!
13. !
!
7. !

The solution to the system of linear equations through LU decomposition proceeds in two parts:
-->y = L\c
y =
! - 3.
!
!
13.75 !
!
13.875 !
-->x = U\y
x =
!
2. !
! - 1. !
!
3. !

Gaussian elimination with partial pivoting

In this section we modify the function gausselim to incorporate partial pivoting. The resulting
function is called gausselimPP. The algorithm for partial pivoting requires us to compare the
current pivot at the k-th pass of the forward elimination part, namely akk, with all the
elements of the same column, aij, for i>k. When the largest absolute value of the column is
found, the corresponding row and row k are exchanged before proceeding with the forward
elimination algorithm. The backward substitution part of the algorithm proceeds as in the
original gausselim function.
Here is the listing of function gausselimPP:
function [x] = gausselimPP(A,b)
//This function obtains the solution to the system of
//linear equations A*x = b, given the matrix of coefficients A
//and the right-hand side vector, b. Gaussian elimination with
//partial pivoting.

Download at InfoClearinghouse.com

13

2001 - Gilberto E. Urroz

[nA,mA] = size(A)
[nb,mb] = size(b)
if nA<>mA then
error('gausselim - Matrix A must be square');
abort;
elseif mA<>nb then
error('gausselim - incompatible dimensions between A and b');
abort;
end;
a = [A b];
n = nA
;

// Augmented matrix
// Matrix size

//Forward elimination with partial pivoting

for k=1:n-1
kpivot = k; amax = abs(a(k,k));
//Pivoting
for i=k+1:n
if abs(a(i,k))>amax then
kpivot = i; amax = a(k,i);
end;
end;
temp = a(kpivot,:); a(kpivot,:) = a(k,:); a(k,:) = temp;
for i=k+1:n
//Forward elimination
for j=k+1:n+1
a(i,j)=a(i,j)-a(k,j)*a(i,k)/a(k,k);
end;
end;
end;
//Backward substitution
x(n) = a(n,n+1)/a(n,n);
for i = n-1:-1:1
sumk=0
for k=i+1:n
sumk=sumk+a(i,k)*x(k);
end;
x(i)=(a(i,n+1)-sumk)/a(i,i);
end;
//End function

Application of the function gausselimPP for the case under consideration is shown next:
-->A = [0,2,3;2,0,3;8,16,-1]; b = [7;13;-3];
-->getf('gausselimPP')
-->gausselimPP(A,b)
ans =
!
2. !
! - 1. !
!
3. !

Download at InfoClearinghouse.com

14

2001 - Gilberto E. Urroz

Solving multiple set of equations with the same coefficient matrix

Suppose that you want to solve the following three sets of equations:
x1 +2x2+3x3 = 14,
3x1 -2 x2+ x3 = 2,
4x1 +2 x2 - x3 = 5,

2x1 +4x2+6x3 = 9,
3x1 -2x2+ x3 = -5,
4x1 +2x2 - x3 = 19,

2x1 +4x2+6 x3 = -2,

3x1 -2x2+ x3 = 2,
4x1 +2x2 - x3 = 12.

We can write the three systems of equations as a single matrix equation: AX = B, where

3
1 2
x11

A = 3 2 1 , X = x 21
4 2 1
x31

x12
x 22
x32

x13
14 9 2

x 23 , B = 2 5 2 .
5 19 12
x33

In the unknown matrix X, the first sub-index identifies the original sub-index, while the second
one identifies to which system of linear equations a particular variable belongs.
The solution to the system, AX = B, can be found in SCILAB using the backward slash operator,
i.e., X = A\B, or the inverse matrix, X = A-1B. For the matrices defined above we can write:
-->A = [1,2,3;3,-2,1;4,2,-1], B=[14,9,-2;2,-5,2;5,19,12]
A =
!
1.
2.
3. !
!
3. - 2.
1. !
!
4.
2. - 1. !
B =
!
14.
9.
- 2. !
!
2.
- 5.
2. !
!
5.
19.
12. !

Solution using the backward slash operator:

-->X1 = A\B
X1 =
!
!
!

1.
2.
3.

2.
5.
- 1.

2. !
1. !
- 2. !

Solution using the inverse matrix of A:

-->X2 = inv(A)*B
X2 =
!
!
!

1.
2.
3.

2.
5.
- 1.

2. !
1. !
- 2. !

In both cases, the result is:

Download at InfoClearinghouse.com

2
1 2

x := 2 5
1 .
3 1 2
15

2001 - Gilberto E. Urroz

Gaussian elimination for multiple sets of linear equations

In this section, function gausselimPP is modified to account for a multiple set of linear
equations. The augmented matrix for this case is Aaug = [A B], where A is a nxn matrix, and B
is a nxm matrix. The solution is now a nxn matrix X.
function [x] = gausselimm(A,B)
//This function obtains the solution to the system of
//linear equations A*X = B, given the nxn matrix of coefficients A
//and the nxm right-hand side matrix, B. Matrix X is nxm.
[nA,mA] = size(A)
[nB,mB] = size(B)
if nA<>mA then
error('gausselim - Matrix A must be square');
abort;
elseif mA<>nB then
error('gausselim - incompatible dimensions between A and b');
abort;
end;
a = [A B];
n = nA
;
m = mB
;

// Augmented matrix
// Number of rows and columns in A, rows in B
// Number of columns in B

//Forward elimination with partial pivoting

for k=1:n-1
kpivot = k; amax = abs(a(k,k));
//Pivoting
for i=k+1:n
if abs(a(i,k))>amax then
kpivot = i; amax = a(k,i);
end;
end;
temp = a(kpivot,:); a(kpivot,:) = a(k,:); a(k,:) = temp;
for i=k+1:n
//Forward elimination
for j=k+1:n+m
a(i,j)=a(i,j)-a(k,j)*a(i,k)/a(k,k);
end;
end;
end;
//Backward substitution
for j = 1:m
x(n,j) = a(n,n+j)/a(n,n);
for i = n-1:-1:1
sumk=0

Download at InfoClearinghouse.com

16

2001 - Gilberto E. Urroz

for k=i+1:n
sumk=sumk+a(i,k)*x(k,j);
end;
x(i,j)=(a(i,n+j)-sumk)/a(i,i);
end;
end;

//End function
Next, function gausselimm is applied to the matrices A and B defined earlier:
-->A = [1,2,3;3,-2,1;4,2,-1]; B=[14,9,-2;2,-5,2;5,19,12];
-->getf('gausselimm')
-->gausselimm(A,B)
ans =
!
!
!

1.
2.
3.

2.
5.
- 1.

2. !
1. !
- 2. !

Calculating an inverse matrix using Gaussian elimination

If the right-hand side matrix B of a set of multiple systems of linear equations is the nxn
identity matrix, i.e., B = Inxn, the solution X from the matrix system AX = B, is the inverse of
matrix A, i.e., X = A-1.
For example, to find the inverse of the matrix A presented earlier through the use of function
gausselimm, you can use the following SCILAB command:
-->Ainv = gausselimm(A,eye(3,3))
Ainv =
!
!
!

0.
.125
.25

.1428571
.2321429
.1071429

.1428571 !
.1428571 !
.1428571 !

You can check that this result is the same as that obtained from SCILABs function inv:
-->inv(A)
ans =
!
!
!

0.
.125
.25

.1428571
.2321429
.1071429

.1428571 !
.1428571 !
.1428571 !

Calculating the determinant with Gaussian elimination

You can prove that the determinant of a matrix can be obtained by multiplying the elements in
the main diagonal after the forward elimination part of a Gaussian elimination procedure is

Download at InfoClearinghouse.com

17

2001 - Gilberto E. Urroz

completed. In other words, after completing the forward elimination of a nave Gaussian
elimination, the determinant of a nxn matrix A = [aij] is calculated as
n

det( A) = a kk .
k =1

If partial pivoting is included, however, the sign of the determinant changes according to the
number of row switches included in the pivoting process. In such case, if N represents the
number of row exchanges, the determinant is given by
n

det( A) = (1) N a kk .
k =1

The calculation of the determinant is included in the Gaussian elimination function called
gausselimd. The call to this function includes return variables for the solution X to the system
AX = B as well as for the determinant of A. A listing of the function is shown below:
function [x,detA] = gausselimd(A,B)
//This function obtains the solution to the system of
//linear equations A*X = B, given the nxn matrix of coefficients A
//and the nxm right-hand side matrix, B. Matrix X is nxm.
[nA,mA] = size(A);
[nB,mB] = size(B);
if nA<>mA then
error('gausselim - Matrix A must be square');
abort;
elseif mA<>nB then
error('gausselim - incompatible dimensions between A and b');
abort;
end;
a = [A B];
// Augmented matrix
n = nA
;
// Number of rows and columns in A, rows in B
m = mB
;
// Number of columns in B
//Forward elimination with partial pivoting
nswitch = 0;
for k=1:n-1
kpivot = k; amax = abs(a(k,k));
//Pivoting
for i=k+1:n
if abs(a(i,k))>amax then
kpivot = i; amax = a(k,i);
nswitch = nswitch+1
end;
end;
temp = a(kpivot,:); a(kpivot,:) = a(k,:); a(k,:) = temp;
for i=k+1:n
//Forward elimination
for j=k+1:n+m
a(i,j)=a(i,j)-a(k,j)*a(i,k)/a(k,k);
end;
end;
end;

Download at InfoClearinghouse.com

18

2001 - Gilberto E. Urroz

//Calculating the determinant

detA = 1;
for k = 1:n
detA = detA*a(k,k);
end;
detB = (-1)^nswitch*detA;
//Checking for singular matrix
epsilon = 1.0e-10;
if abs(detB)<epsilon then
error('gausselimd - singular matrix');
abort;
end;
//Backward substitution
for j = 1:m
x(n,j) = a(n,n+j)/a(n,n);
for i = n-1:-1:1
sumk=0
for k=i+1:n
sumk=sumk+a(i,k)*x(k,j);
end;
x(i,j)=(a(i,n+j)-sumk)/a(i,i);
end;
end;
//End function

An application of function gausselimd for a non-singular matrix A is shown next:

-->A = [3,5,-1;2,2,3;1,1,2]
A =
!
!
!

3.
2.
1.

5.
2.
1.

- 1. !
3. !
2. !

-->b = [-4;17;11]
b =
! - 4. !
!
17. !
!
11. !

-->getf('gausselimd')
-->[x,detA] = gausselimd(A,b)
detA =
- 2.
=

!
2. !
! - 1. !
!
5. !

Download at InfoClearinghouse.com

19

2001 - Gilberto E. Urroz

For a singular matrix A, function gausselimd provides an error message:

-->A = [2,3,1;4,6,2;1,1,2]
A =
!
!
!

2.
4.
1.

3.
6.
1.

1. !
2. !
2. !

-->[x,detA] = gausselimd(A,b)
!--error 9999
gausselimd - singular matrix
at line
53 of function gausselimd
[x,detA] = gausselimd(A,b)

called by :

Gauss-Jordan elimination
Gauss-Jordan elimination consists in continuing the row operations in the augmented matrix
that results from the forward elimination from a Gaussian elimination process, until an identity
matrix is obtained in place of the original A matrix. For example, for the following augmented
matrix, the forward elimination procedure results in:

A aug

2 4
6 14 2 4
6
14 2 4
6 14

= 3 2 1 3 0 8 8 24 . 0 8 8 24
4 2 1 4 0 6 13 32 0 0 7 14

We can continue performing row operations until the augmented matrix gets reduced to:

1 2 3 7 1 2 3 7 1 2 0 1 1 0 0 1

0 1 1 3 0 1 1 1 0 1 0 1 0 1 0 1 .
0 0 1 2 0 0 1 2 0 0 1 2 0 0 1 2

The first matrix is the same as that one obtained before at the end of the forward elimination
process, with the exception that all rows have been divided by the corresponding diagonal
term, i.e., row 1 was divided by 2, row 2 was divided by -8, and row 3 was divided by 7. The
final matrix is equivalent to the equations: X = -1, Y = 1, Z= 2, which is the solution to the
original system of equations.

Download at InfoClearinghouse.com

20

2001 - Gilberto E. Urroz

SCILAB provides the function rref (row-reduced echelon form) that can be used to obtain a
solution to a system of linear equations using Gauss-Jordan elimination. The function requires
as argument an augmented matrix, for example:
-->A = [2,4,6;3,-2,1;4,2,-1]; b = [14;-3;-4];
-->A_aug = [A b]
A_aug =
!
!
!

2.
3.
4.

4.
- 2.
2.

6.
1.
- 1.

14. !
- 3. !
- 4. !

-->rref(A_aug)
ans =
!
!
!

1.
0.
0.

0.
1.
0.

0.
0.
1.

- 1. !
1. !
2. !

The last result indicates that the solution to the system is x 1 = -1, x2 = 1, and x3 = 2.

Calculating the inverse through Gauss-Jordan elimination

The calculation of an inverse matrix is equivalent to calculating the solution to the augmented
system [A | I ]. For example, for the matrix A

3
1 2

A = 3 2 1 ,
4 2 1
we would write this augmented matrix as

A aug

1 2
3 1 0 0

= 3 2 1 0 1 0.
4 2 1 0 0 1

The SCILAB commands needed to obtain the inverse using Gauss-Jordan elimination are:
-->A = [1,2,3;3,-2,1;4,2,-1]
A =
!
1.
2.
3. !
!
3. - 2.
1. !
!
4.
2. - 1. !
-->A_aug = [A eye(3,3)]
A_aug =
!
1.
2.
3.
1.
!
3. - 2.
1.
0.
!
4.
2. - 1.
0.

Download at InfoClearinghouse.com

//Original matrix

//Augmented matrix
0.
1.
0.

0. !
0. !
1. !

21

2001 - Gilberto E. Urroz

The next command produces the Gauss-Jordan elimination:

-->A_aug_inv = rref(A_aug)
A_aug_inv =
!
1.
0.
0.
0.
!
0.
1.
0.
.125
!
0.
0.
1.
.25

.1428571
.2321429
.1071429

.1428571 !
.1428571 !
.1428571 !

The inverse is obtained by extracting rows 4, 5, and 6, of the previous result:

-->A_inv_1 = [A_aug_inv(:,4), A_aug_inv(:,5), A_aug_inv(:,6)]
A_inv_1 =
!
!
!

0.
.125
.25

.1428571
.2321429
.1071429

.1428571 !
.1428571 !
.1428571 !

To check that the matrix A_inv_1 is indeed the inverse of A use:

-->A*A_inv_1
ans =
!
!
!

1.
0.
0.

0.
1.
0.

0. !
0. !
1. !

This exercise is presented to illustrate the calculation of inverse matrices through Gauss-Jordan
elimination. In practice, you should use the function inv in SCILAB to obtain inverse matrices.
For the case under consideration, for example, the inverse is obtained from:
-->A_inv_2 = inv(A)
A_inv_2 =
!
!
!

0.
.125
.25

.1428571
.2321429
.1071429

.1428571 !
.1428571 !
.1428571 !

The result is exactly the same as found above using Gauss-Jordan elimination.

Eigenvalues and eigenvectors

Given a square matrix A, we can write the eigenvalue equation
Ax = x,
where the values of that satisfy the equation are known as the eigenvalues of matrix A. For
each value of , we can find, from the same equation, values of x that satisfy the eigenvalue
equation. These values of x are known as the eigenvectors of matrix A.
The eigenvalues equation can be written also as

Download at InfoClearinghouse.com

22

2001 - Gilberto E. Urroz

(A I)x = 0.
This equation will have a non-trivial solution only if the characteristic matrix, (A I), is
singular, i.e., if
det(A I) = 0.
The last equation generates an algebraic equation involving a polynomial of order n for a
square matrix Ann. The resulting equation is known as the characteristic polynomial of matrix
A. Solving the characteristic polynomial produces the eigenvalues of the matrix.
Using SCILAB we can obtain the characteristic matrix, characteristic polynomial, and
eigenvalues of a matrix as shown below for a symmetric matrix

3 2 5
A = 2 3 6.
5
6 4
First, we define the matrix A:
-->A = [3,-2,5;-2,3,6;5,6,4]
A =
!
3.
! - 2.
!
5.

- 2.
3.
6.

5. !
6. !
4. !

Next, the characteristic matrix B = (A I) corresponding to a square matrix A is calculated as

follows. Notice that we define the symbolic variable lam (for lambda, ) before forming the
characteristic matrix:
-->lam = poly(0,'lam')
lam =
lam
-->charMat = A-lam*eye(3,3)

charMat

!
3 - lam
!
! - 2
!
!
5

- 2

3 - lam

4 - lam

!
!
!
!
!

The characteristic equation can be determined using function poly with matrix A and the
variable name lam as arguments, i.e.,
-->charPoly = poly(A,'lam')
charPoly =

Download at InfoClearinghouse.com

23

2001 - Gilberto E. Urroz

2
3
283 - 32lam -10lam + lam

The function spec produces the eigenvalues for matrix A:

-->lam = spec(A)
lam =
! - 5.4409348 !
!
4.9650189 !
!
10.475916 !

Calculating the first eigenvector

We can form the characteristic matrix for a particular eigenvalue, say 1 = lam(1), by using:
-->B1 = A - lam(1)*eye(3,3)
B1 =
! 8.4409348 - 2.
5.
!
! - 2.
8.4409348 6.
!
! 5.
6.
9.4409348 !
This, of course, is a singular matrix, which you can check by calculating the determinant of B1:
--> det(B1)
ans
=
5.973E-14

which, although not exactly zero, is close enough to zero to ensure singularity.
Because the characteristic matrix is singular, there is no unique solution to the problem (A1I)x = 0. The equivalent system of linear equations is:
2x2 +
5x3 = 0
8.4409348x1 6x3 = 0
- 2x1 + 8.4409348x2 +
6 x2 + 9.4409348x3 = 0
5x1 +
The three equations are linearly dependent. This means that we can select an arbitrary value
of one of the solutions, say, x3 = 1, and solve two of the equations for the other two solutions,
x1 and x2. We could try, for example, to solve the first two equations (with x3 = 1), re-written
as:
8.4409348x1 2x2 = -5
- 2x1 + 8.4409348x2 = -6
Using SCILAB we can get the solution to this system as follows:
-->C1 = B1(1:2,1:2), b1 = -B1(3,1:2)
C1 =
!

8.4409348

- 2.

Download at InfoClearinghouse.com

24

2001 - Gilberto E. Urroz

! - 2.

8.4409348 !

-->b1 = -B1(1:2,3)
b1 =
! - 5. !
! - 6. !

The solution for 1 = [x1,x2]T is:

-->xi1 = C1\b1
xi1 =
! ! -

.8060249 !
.9018017 !

The eigenvector x1 is, therefore:

-->x1 = [xi1;1]
x1 =
! - .8060249 !
! - .9018017 !
!
1.
!

If we want to find the corresponding unit eigenvector, ex1, we can use:

-->e_x1 = x1/norm(x1)
e_x1 =
! - .5135977 !
! - .5746266 !
! .6371983 !

Calculating eigenvectors with a user-defined function

Following the procedure outlined above for calculating the first eigenvector, we can write a
SCILAB user-defined function to calculate all the eigenvectors of a matrix. The following is a
listing of this function:
function [x,lam] = eigenvectors(A)
//Calculates unit eigenvectors of matrix A
//returning a matrix x whose columns are
//the eigenvectors. The function also
//returns the eigenvalues of the matrix.
[n,m] = size(A);
if m<>n then
error('eigenvectors - matrix A is not square');
abort;
end;
lam = spec(A)';

Download at InfoClearinghouse.com

//Eigenvalues of matrix A

25

2001 - Gilberto E. Urroz

x = [];
for k = 1:n
B = A - lam(k)*eye(n,n);
C = B(1:n-1,1:n-1);
b = -B(1:n-1,n);
y = C\b;
y = [y;1];
y = y/norm(y);
x = [x y];
end;

//Characteristic matrix
//Coeff. matrix for reduced system
//RHS vector for reduced system
//Solution for reduced system
//Complete eigenvector
//Make unit eigenvector
//Add eigenvector to matrix

//End of function

Applying function eigenvectors to the matrix A used earlier produces the following eigenvalues
and eigenvectors:
-->getf('eigenvectors')
-->[x,lam] = eigenvectors(A)
lam =
! - 5.4409348
x =
! ! !

.5135977
.5746266
.6371983

4.9650189

10.475916 !

.7711676
.6347298
.0491799

.3761887 !
.5166454 !
.7691291 !

We can separate the eigenvectors x1, x2, x3 from matrix x by using:

-->x1 = x(1:3,1), x2 = x(1:3,2), x3 = x(1:3,3)
x1 =
! ! !
x2

.5135977 !
.5746266 !
.6371983 !
=

!
! !
x3

.7711676 !
.6347298 !
.0491799 !
=

!
!
!

.3761887 !
.5166454 !
.7691291 !

The eigenvectors corresponding to a symmetric matrix are orthogonal to each other, i.e., xixj
= 0, for ij. Checking these results for the eigenvectors found above:
-->x1'*x2, x1'*x3, x2'*x3
ans =
- 3.678E-16
ans =

Download at InfoClearinghouse.com

26

2001 - Gilberto E. Urroz

- 5.551E-17
ans =
9.576E-16

For a non-symmetric matrix, the eigenvalues may include complex numbers:

Example 1 - non-symmetric matrix, real eigenvalues
-->A = [3, 2, -1; 1, 1, 2; 5, 5, -2]
!
!
!

3.
1.
5.

2.
1.
5.

- 1. !
2. !
- 2. !

-->[x,lam] = eigenvectors(A)
lam

.9234927
x

4.1829571

- 3.1064499 !

! !
!

=
.5974148
.7551776
.2698191

.3437456
.5746662
.7426962

.2916085 !
.4752998 !
.8300931 !

Example 2 - non-symmetric matrix, complex eigenvalues

-->A = [2, 2, -3; 3, 3, -2; 1, 1, 1]
A =
!
!
!

2.
3.
1.

2.
3.
1.

- 3. !
- 2. !
1. !

-->[x,lam] = eigenvectors(A)
warning
matrix is close to singular or badly scaled.
results may be inaccurate. rcond =
7.6862E-17
lam
!

=
4.646E-16
=

3. - i

! - .7071068
!
.7071068
!
2.826E-16

.25 .75 .5

3. + i

.25i
.25i

.25 +
.75 +
.5

.25i !
.25i !
!

Generating the characteristic equation for a matrix

Download at InfoClearinghouse.com

27

2001 - Gilberto E. Urroz

This section represents an additional exercise on SCILAB function programming, since we

already know that SCILAB can generate the characteristic equation for a square matrix through
function poly. The method coded in the function uses the equations:
pn+1 = -1.0, Bn+1 = A
Bj = A(Bj+1 - pj+1I), for j = n,n-1, , 3, 2,1
Pj = (-1)jpj, j = 1, 2, ,n+1
The matrix I shown in the formula is the nxn identity matrix.
polynomial will be contained in the 1x(n+1) vector

The coefficients of the

p = [p1 p2 pn+1].
The characteristic equation is given by
p1+p2+p3 2+ +pn n-1+pn+1 n = 0.
Here is a listing of the function:
function [p]=chreq(A)
//This function generates the coefficients of the characteristic
//equation for a square matrix A
[m n]=size(A);
if(m <> n)then
error('matrix is not square.')
abort
end;
I = eye(n,n);
//Identity matrix
p = zeros(1,n);
//Matrix (1xn) filled with zeroes
p(1,n+1) = -1.0;
B = A;
p(1,n)=trace(B);
for j = n-1:-1:1,
B = A*(B-p(1,j+1)*I),
p(1,j) = trace(B)/(n-j+1),
end;
p = (-1)^n*p;
p = poly(p,"lmbd","coeff");
p
//end function chreq

In this function we used the function poly to generate the final result in the function. In the
call to poly shown above, we use the vector of coefficients p as the first argument, "lmbd" is
the independent variable , and "coeff" is required to indicate that the vector p represent the
coefficients of the polynomial. The following commands show you how to load the function and
run it for a particular matrix:
-->getf('chreq')
-->A = [1 3 1;2 5 -1;2 7 -1]
A =
!

1.

3.

1. !

Download at InfoClearinghouse.com

28

2001 - Gilberto E. Urroz

!
!

2.
2.

5.
7.

- 1. !
- 1. !

-->CheqA = chreq(A)
CheqA =
2
3
- 6 - 2lmbd - 5lmbd + lmbd

Generalized eigenvalue problem

The generalized eigenvalue problem involves square matrices A and B of the same size, and
consists in finding the values of the scalar and the vectors x that satisfy the matricial
equation:
Ax = Bx.
The following function, geingenvectors (generalized eigenvectors), calculates the eigenvalues
and eigenvectors from the generalized eigenvalue problem Ax = Bx, or (A-B)x = 0.
function [x,lam] = geigenvectors(A,B)
//Calculates unit eigenvectors of matrix A
//returning a matrix x whose columns are
//the eigenvectors. The function also
//returns the eigenvalues of the matrix.
[nA,mA] = size(A);
[nB,mB] = size(B);
if (mA<>nA | mB<>nB) then
error('geigenvectors - matrix A or B not square');
abort;
end;
if nA<>nB then
error('geigenvectors - matrix A and B have different dimensions');
abort;
end;
lam = poly(0,'lam');
chPoly = det(A-B*lam);
lam = roots(chPoly)';

//Define variable "lam"

//Characteristic polynomial
//Eigenvalues of matrix A

x = []; n = nA;
for k = 1:n
BB = A - lam(k)*B;
CC = BB(1:n-1,1:n-1);
bb = -BB(1:n-1,n);
y = CC\bb;
y = [y;1];
y = y/norm(y);
x = [x y];
end;

Download at InfoClearinghouse.com

//Characteristic matrix
//Coeff. matrix for reduced system
//RHS vector for reduced system
//Solution for reduced system
//Complete eigenvector
//Make unit eigenvector
//Add eigenvector to matrix

29

2001 - Gilberto E. Urroz

//End of function

An application of this function is presented next:

-->A = [4,1,6;8,5,8;1,5,5]
A =
!
!
!

4.
8.
1.

1.
5.
5.

6. !
8. !
5. !

-->B = [3,9,7;3,3,2;9,3,4]
B =
!
!
!

3.
3.
9.

9.
3.
3.

7. !
2. !
4. !

-->getf('geigenvectors')
-->[x,lam] = geigenvectors(A,B)
lam =
! - .2055713 - 1.1759636i
x =

.2055713 + 1.1759636i

- 1.5333019 !

! ! !

.2828249 +
.5392352 .7450009

.2828249 .5392352 +
.7450009

.0422024i
.2691247i

Download at InfoClearinghouse.com

30

.0422024i
.2691247i

.0202307 !
.7437197 !
.6681854 !

2001 - Gilberto E. Urroz

Sparse matrices
Sparse matrices are those that have a large percentage of zero elements. When a matrix is
defined as sparse in SCILAB, only those non-zero elements are stored. The regular definition
of a matrix, also referred to as a full matrix, implies that SCILAB stores all elements of the
matrix, zero or otherwise.

Creating sparse matrices

SCILAB provides the function sparse to define sparse matrices. The simplest call to the
function is
A_sparse = sparse(A)
Where A is a sparse or full matrix. Obviously, if A is already a sparse matrix, no action is taken
by the call to function sparse. If A is a full matrix, the function sparse squeezes out those zero
elements from A. For example,
-->A = [2, 0, -1; 0, 1, 0; 0, 0 , 2]
A =
!
!
!

2.
0.
0.

0.
1.
0.

- 1. !
0. !
2. !

-->As = sparse(A)
As =
(

3,

3) sparse matrix

(
(
(
(

1,
1,
2,
3,

1)
3)
2)
3)

2.
- 1.
1.
2.

Notice that SCILAB reports the size of the matrix (3,3), and those non-zero elements only.
These are the only elements stored in memory. Thus, sparse matrices are useful in minimizing
memory storage particularly when large-size matrices, with relatively small density of non-zero
elements, are involved.
Alternatively, a call to sparse of the form
A_sparse = sparse(index,values)
Where index is a nx2 matrix and values is a nx1 vector such that values(i,1) represents the
element in row index(i,1) and column index(i,2) of the sparse matrix. For example,
-->row = [2, 2, 3, 3, 6, 6, 10]
row =
!

2.

2.

3.

3.

6.

-->col = [1, 2, 2, 3, 1, 4, 2]
col =

Download at InfoClearinghouse.com

//row position for non-zero elements

6.

10. !

//column position for non-zero elements

31

2001 - Gilberto E. Urroz

1.

2.

2.

3.

1.

4.

2. !

-->val = [-0.5, 0.3, 0.2, 1.5, 4.2, -1.1, 2.0] //values of non-zero elements
val =
! -

.5

.3

.2

1.5

-->index = [row' col']

index =
!
!
!
!
!
!
!

2.
2.
3.
3.
6.
6.
10.

1.
2.
2.
3.
1.
4.
2.

4.2

- 1.1

2. !

//indices of non-zero elements (i,j)

!
!
!
!
!
!
!

-->As = sparse(index,val)
As =
(

10,

4) sparse matrix

(
(
(
(
(
(
(

2,
2,
3,
3,
6,
6,
10,

1)
2)
2)
3)
1)
4)
2)

//creating the sparse matrix

.5
.3
.2
1.5
4.2
- 1.1
2.

The function full converts a sparse matrix into a full matrix. For example,
-->A =full(As)
A =
!
0.
! - .5
!
0.
!
0.
!
0.
!
4.2
!
0.
!
0.
!
0.
!
0.

0.
.3
.2
0.
0.
0.
0.
0.
0.
2.

0.
0.
1.5
0.
0.
0.
0.
0.
0.
0.

0.
0.
0.
0.
0.
- 1.1
0.
0.
0.
0.

!
!
!
!
!
!
!
!
!
!

The following call to sparse includes defining the row and column dimensions of the sparse
matrix besides providing indices and values of the non-zero values:
A_sparse = sparse(index,values,dim)
Here, dim is a 1x2 vector with the row and column dimensions of the sparse matrix. As an
example, we can try:
-->As = sparse(index,val,[10,12])

Download at InfoClearinghouse.com

32

2001 - Gilberto E. Urroz

As

10,

(
(
(
(
(
(
(

2,
2,
3,
3,
6,
6,
10,

12) sparse matrix

1)
2)
2)
3)
1)
4)
2)

.5
.3
.2
1.5
4.2
- 1.1
2.

Getting information about a sparse matrix

The function spget, through a call of the form
[index, val, dim] = spget(As)
returns the location of the values contained in val described by the row and column indices
(listed in the first and second columns of index), as well as the row and column dimensions of
the matrix contained in vector dim. For example, for the matrix As defined above, function
spget produces:
-->[index, values, dim] = spget(As)
dim =
!

10.
values

12. !
=

! - .5 !
!
.3 !
!
.2 !
!
1.5 !
!
4.2 !
! - 1.1 !
!
2. !
index =
!
!
!
!
!
!
!

2.
2.
3.
3.
6.
6.
10.

1.
2.
2.
3.
1.
4.
2.

!
!
!
!
!
!
!

Sparse matrix with unit entries

To create a matrix including values of 1.0 with the same structure of an existing sparse matrix
use the function spones, for example:

Download at InfoClearinghouse.com

33

2001 - Gilberto E. Urroz

-->A1 = spones(As)
A1 =
(

10,

(
(
(
(
(
(
(

2,
2,
3,
3,
6,
6,
10,

12) sparse matrix

1)
2)
2)
3)
1)
4)
2)

1.
1.
1.
1.
1.
1.
1.

Sparse identity matrices

To create a sparse identity matrix use the function speye. For example, giving the dimensions
of the desired identity matrix we can use:
-->speye(6,6)
ans =
(

6,

6) sparse matrix

(
(
(
(
(
(

1,
2,
3,
4,
5,
6,

1)
2)
3)
4)
5)
6)

1.
1.
1.
1.
1.
1.

To produce a sparse identity matrix with the same dimensions of an existing matrix As use:
-->speye(As)
ans =
(

10,

12) sparse matrix

(
(
(
(
(
(
(
(
(
(

1,
2,
3,
4,
5,
6,
7,
8,
9,
10,

1)
2)
3)
4)
5)
6)
7)
8)
9)
10)

1.
1.
1.
1.
1.
1.
1.
1.
1.
1.

Sparse matrix with random entries

The function sprand generates a sparse matrix with random non-zero entries. The function
requires the dimensions of the matrix as the two first arguments, and a density of non-zero

Download at InfoClearinghouse.com

34

2001 - Gilberto E. Urroz

elements as the third argument. The density of non-zero values is given as a number between
zero and one, for example:
-->As = sprand(6,5,0.2)
As =
(

6,

5) sparse matrix

(
(
(
(
(
(

1,
2,
2,
3,
3,
4,

3)
1)
4)
2)
5)
2)

.0500420
.9931210
.7485507
.6488563
.4104059
.9923191

To see the full matrix use:

-->full(As)
ans =
!
!
!
!
!
!

0.
.9931210
0.
0.
0.
0.

0.
0.
.6488563
.9923191
0.
0.

.0500420
0.
0.
0.
0.
0.

0.
.7485507
0.
0.
0.
0.

0.
0.
.4104059
0.
0.
0.

!
!
!
!
!
!

Sparse matrices with zero entries

The function spzeros is used to define a sparse matrix given the dimensions of the matrix.
Since all the elements of the matrix are zero, what the function does is to reserve memory
space for the matrix ensuring its sparse character. A couple of examples of application of
spzeros are shown next:
->spzeros(As)
ans =
(
6,
5) zero sparse matrix
-->spzeros(6,4)
ans =
(
6,
4) zero sparse matrix

The function spzeros can be used, for example, in programming SCILAB functions that require
sparse matrices to reserve a matrix name for future use.

Visualizing sparse matrices

The following function, spplot (sparse matrix plot), can be used to visualize the distribution of
non-zero elements for relatively large sparse matrices. A listing of the function follows:
function spplot(As)
//Plot a schematic of non-zero elements
//for a sparse matrix As

Download at InfoClearinghouse.com

35

2001 - Gilberto E. Urroz

wnumber = input('Enter graphics window number:');

A = sparse(As);
[index,vals,dim] = spget(A);
xx=index(:,1);yy=index(:,2);
xset('window',wnumber);
xset('mark',-2,2);
plot2d(xx,yy,-2);
//end function

The function requires as input the name of the sparse matrix and it requests the number of the
graphics window where the plot is to be displayed. The following is an example that uses this
function to visualize a sparse matrix of dimensions 40x40 with a density of non-zero numbers of
0.2:
-->As = sprand(40,40,0.2);
-->getf('spplot')
-->spplot(As)
Enter graphics window number:
--> 2

The result, shown in SCILAB graphics window number 2 is reproduced next:

Factorization of sparse matrices

SCILAB provides functions lufact, luget, and ludel, to produce the LU factorization of a sparse
matrix A. Function lufact returns the rank of the matrix as well as a pointer or handle to the
LU factors, which are then retrieved with function luget. Function ludel is required after
performing a LU factorization to clear up the pointer or handle generated with lufact.
The call to function lufact has the general form:

Download at InfoClearinghouse.com

36

2001 - Gilberto E. Urroz

[hand,rk] = lufact(As)
where hand is the handle or pointer, rk is the rank of sparse matrix As. Be aware that hand is
a pointer to locate the LU factors in memory. SCILAB produces no display for hand.
The call to function luget has the general form:
[P,L,U,Q] = luget(hand)
where P and Q are permutation matrices and L and U are the LU factors of the sparse matrix
that generated hand through a call to function lufact. Matrices P, Q, L, and U are related to
matrix As by P*Q*L*U = As.
After a LU factorization is completed, it is necessary to call function ludel to clear the pointer
generated in the call to function lufact, i.e., use
ludel(hand)
The following example shows the LU factorization of a 6x6 randomly-generated sparse matrix
with a density of non-zero numbers of 0.5:
-->As = sprand(5,5,0.5);

To see the full matrix just generated try:

-->full(As)
ans =
!
!
!
!
!

.7019967
.5018194
0.
.7860680
0.

.7354560
0.
.4732295
0.
.9456872

0.
.6522234
.9388094
0.
.2401141

0.
.2921187
.6533222
.9933566
.4494063

.5098139
.3776263
0.
0.
0.

!
!
!
!
!

The first step in the LU factorization is:

-->[hand,rk] = lufact(As)
rk =
5.
hand =

The rank of the matrix is 5. Notice that SCILAB shows nothing for the handle or pointer hand.
The second step in the LU factorization is to get the matrices P, L, U, and Q, such that P*L*U*Q
= As, i.e.,
-->[P,L,U,Q] = luget(hand);

These matrices are shown in full as follows:

-->full(P)
ans =
!
!

0.
0.

0.
0.

1.
0.

0.
1.

Download at InfoClearinghouse.com

0. !
0. !

37

2001 - Gilberto E. Urroz

!
!
!

0.
1.
0.

1.
0.
0.

0.
0.
0.

0.
0.
0.

0. !
0. !
1. !

-->full(L)
ans =
!
!
!
!
!

.9933566
.6533222
0.
.2921187
.4494063

0.
.9388094
0.
.6522234
.2401141

0.
0.
.7019967
.6298297
- .2233988

0.
0.
0.
- .9886182
1.0586985

0.
.5040741
1.047663
1.
0.

0.
0.
.726234
.0806958
1.

0.
0.
0.
0.
.0768072

!
!
!
!
!

-->full(U)
ans =
!
!
!
!
!

1.
0.
0.
0.
0.

0.
1.
0.
0.
0.

.7913251
.5506871
1.
0.
0.

!
!
!
!
!

-->full(Q)
ans =
!
!
!
!
!

0.
0.
1.
0.
0.

0.
0.
0.
1.
0.

0.
1.
0.
0.
0.

1.
0.
0.
0.
0.

0.
0.
0.
0.
1.

!
!
!
!
!

To check if the product P*L*U*Q is indeed equal to the original sparse matrix As, use:
-->full(P*L*U*Q-As)
ans =
!
!
!
!
!

0.
0.
0.
0.
0.

1.110E-16
0.
0.
0.
1.110E-16

0.
0.
0.
0.
0.

0.
0.
0.
0.
0.

0.
0.
0.
0.
0.

!
!
!
!
!

The resulting matrix has a couple of small non-zero elements. These can be cleared by using
the function clean as follows:
-->clean(full(P*L*U*Q-As))
ans =
!
!
!
!
!

0.
0.
0.
0.
0.

0.
0.
0.
0.
0.

0.
0.
0.
0.
0.

0.
0.
0.
0.
0.

0.
0.
0.
0.
0.

!
!
!
!
!

At this point we can use function ludel to clear up the handle or pointer used in the
factorization:
--> ludel(hand)

Download at InfoClearinghouse.com

38

2001 - Gilberto E. Urroz

Solution to system of linear equations involving sparse matrices

The solution to a system of linear equations of the form Ax = b, where A is a sparse matrix and
b is a full column vector, can be accomplished by using functions lufact and lusolve. Function
lufact provides the handle or pointer hand, from [hand,rk] = lufact(A), which then used by
lusolve through the function call
x = lusolve(hand,b)
The following example illustrates the uses of lufact and lusolve to solve a system of linear
equations whose matrix of coefficients is sparse:
-->A=sprand(6,6,0.6);

//Generate a random, sparse matrix of coefficients

-->b=rand(6,1);

//Generate a random, full right-hand side vector

First, we get the rank and handle for the LU factorization of matrix A:
-->[hand,rk] = lufact(A)
rk =
5.
hand =

Next, we use function lusolve with the handle hand and right-hand side vector b to obtain the
solution to the system of linear equations:
-->x = lusolve(hand,b)
x =
! - .8137973
! - .5726203
!
.8288245
!
.9152823
!
.4984395
!
0.

!
!
!
!
!
!

Function lusolve also allows for the direct solution of the system of linear equations without
having to use lufact first. For the case under consideration the SCILAB command is:
-->x = lusolve(A,b)
!--error
singular matrix

19

However, matrix A for this case is singular (earlier we found that its rank was 5, i.e., smaller
than the number of rows or columns, thus indicating a singular matrix), and no solution is
obtained. The use of lufact combined with lusolve, as shown earlier, forces a solution by
making one of the values equal to zero and solving for the remaining five values.
After completing the solution we need to clear the handle for the LU factorization by using:
-->ludel(hand)

Download at InfoClearinghouse.com

39

2001 - Gilberto E. Urroz

A second example of solving a system of linear equations is shown next:

-->A = sprand(5,5,0.5);
-->b = full(sprand(5,1,0.6));
Notice that to generate b, which must be a full vector, we first generated a sparse random
vector with a non-zero value density of 0.6. Next, we used the function full to convert that
sparse vector into a full one.
A call to function lufact will provide the rank of matrix A:
-->[hand,rk] = lufact(A)
rk =
5.
hand =

Because the matrix has full rank (i.e., its rank is the same as the number of rows or columns),
it is possible to find the solution by using:
-->x = lusolve(A,b)
x =
! - .2813594 !
!
.5590971 !
!
1.0143263 !
! - .1810458 !
! - 1.1233045 !

Do not forget to clear the handle generated with lufact by using: -->ludel(hand)

Solution to system of linear equations using the inverse of a sparse

matrix
If sparse matrix A in the system of linear equations Ax = b is non-singular (i.e., a full-rank
matrix), it is possible to find the inverse of A, A-1, through function inv. The solution x can then
be found with x = A-1b. For example, using matrix A from the previous example we can write:
-->Ainv = inv(A)
Ainv =

//Calculate inverse of a sparse matrix

5,

5) sparse matrix

(
(
(
(
(
(
(
(
(
(
(
(

1,
1,
1,
1,
1,
2,
2,
2,
2,
2,
3,
4,

1)
2)
3)
4)
5)
1)
2)
3)
4)
5)
1)
1)

.7235355
.7404212
.3915968
2.0631067
.1957261
1.4075482
1.2851264
- .6179589
.0607043
- .5012662
2.9506345
- 2.1683689

Download at InfoClearinghouse.com

40

2001 - Gilberto E. Urroz

(
(
(
(
(
(

4,
5,
5,
5,
5,
5,

3)
1)
2)
3)
4)
5)

1.8737551
- 2.4138658
- .0298745
- .9542841
- 2.1488971
2.5469151

-->x = Ainv*b
x =
! - .2813594 !
!
.5590971 !
!
1.0143263 !
! - .1810458 !
! - 1.1233045 !

Alternatively, you can also use left-division to obtain the solution:

-->x=A\b
x =
! - .2813594 !
!
.5590971 !
!
1.0143263 !
! - .1810458 !
! - 1.1233045 !

Solution to a system of linear equations with a tridiagonal matrix of coefficients

A tri-diagonal matrix is a square matrix such that the only non-zero elements are those in the
main diagonal and those in the two diagonals immediately off the main one. A tri-diagonal
matrix system will look as follows:

a11
a
21
0

M
0

0
0

a12
a 22
a32
M
0
0
0

0
a 23
a33
M
0
0
0

L
0
L
0
L
0
O
M
L a n 2,n 2
L a n 1, n 2
0
L

0
0
0
M
a n 2,n 1
a n 1,n 1
a n 1, n

x1 b1

x 2 b2
x3 b3

M = M
x n 2 bn 2

a n 1,n x n 1 bn 1
a nn x n bn
0
0
0
M
0

This system can also be written as Ax = b, where the nxn matrix A, and the nx1 vectors x
and b are easily identified from the previous expression.
Since each column in the matrix of coefficients only uses three elements, we can enter the
data as the elements of a nx3 matrix,

Download at InfoClearinghouse.com

41

2001 - Gilberto E. Urroz

 a11
a
21
a 31
A=
M
a n1,1

a12
a 22
a32
M
a n 1, 2
a n ,n 1

0
a 23
a33

M
a n 1,3

a nn

Thomas algorithm for the solution of the tri-diagonal system of linear equations is an
adaptation of the Gaussian elimination procedure. It consists of a forward elimination
accomplished through the recurrence formulas:
ai2 = ai2 - ai1 ai-1,3/ai-1,2, bi = bi - ai1bi-1/ai-1,2, for i = 2, 3, , n
The backward substitution step is performed through the following equations:
xn = bn/an2,

and

xi = (bi - ai3xi+1/ai2), for i = n-1, n-2, , 3, 2, 1 (counting backwards).

The following function implements the Thomas algorithm for the solution of a system of linear
equations whose matrix of coefficients is a tri-diagonal matrix:
function [x] = Tridiag(A,b)
//Uses Thomas algorithm for tridiagonal matrices
//The matrix A is entered as a matrix of three columns and n rows
//containing the main diagonal and its two closest diagonals
[n m] = size(A); // determine matrix size
//Check if number of columns = 3.
if m <> 3 then
error('Matrix needs to have three columns only.')
abort;
end
[nv mv] = size(b); // determine vector size
if ((mv <> 1) | (nv <> n)) then
error('Vector needs to be a column vector or it does not have same
number of rows as matrix')
end
//Recalculate matrix A and vector b
AA = A; bb = b;
for i = 2:n
AA(i,2) = AA(i,2)-AA(i,1)*AA(i-1,3)/AA(i-1,2);
bb(i)
= bb(i) - AA(i,1)*bb(i-1)/AA(i-1,2);
end;
//Back calculation of solution
x(n) = bb(n)/AA(n,2);
for i = n-1:-1:1
x(i) = (bb(i)-AA(i,3)*x(i+1))/AA(i,2)
end;
//show solution
x
//end function

Download at InfoClearinghouse.com

42

2001 - Gilberto E. Urroz

As an exercise, enter the following SCILAB commands:

-->A = [0,4,-1;-1,4,-1;-1,4,-1;-1,4,0]; b=[150;20;150;100];
-->getf('Tridiag')
-->x = Tridiag(A,b)
x =
!
!
!
!

44.976077
29.904306
54.641148
38.660287

!
!
!
!

Solution to tri-diagonal systems of linear equations using sparse

matrices
Because a tri-diagonal matrix is basically a sparse matrix, we can use functions lufact and
lusolve to obtain the solution to a tri-diagonal system of linear equations. The following
function takes the nx3 matrix that represents a tri-diagonal matrix and produces the index and
values of the corresponding sparse matrix:
function [index,val] = tritosparse(A)
//Converts the nx3 matrix A that represents
//a tri-diagonal matrix, into the corresponding
//nxn sparse matrix
krow = 0; kcol = 0; kval = 0;
krow
kcol
kval
krow
kcol
kval

=
=
=
=
=
=

krow+1;
kcol+1;
kval+1;
krow+1;
kcol+1;
kval+1;

for i = 2:n-1
krow =
kcol =
kval =
krow =
kcol =
kval =
krow =
kcol =
kval =
end;
krow
kcol
kval
krow
kcol
kval

=
=
=
=
=
=

row(krow)
col(kcol)
val(kval)
row(krow)
col(kcol)
val(kval)

krow+1;
kcol+1;
kval+1;
krow+1;
kcol+1;
kval+1;
krow+1;
kcol+1;
kval+1;

krow+1;
kcol+1;
kval+1;
krow+1;
kcol+1;
kval+1;

=
=
=
=
=
=

1;
1;
A(1,2)
1;
2;
A(1,3)

row(krow)
col(kcol)
val(kval)
row(krow)
col(kcol)
val(kval)
row(krow)
col(kcol)
val(kval)

row(krow)
col(kcol)
val(kval)
row(krow)
col(kcol)
val(kval)

Download at InfoClearinghouse.com

=
=
=
=
=
=

=
=
=
=
=
=
=
=
=

i;
i-1;
A(i,1)
i;
i;
A(i,2)
i;
i+1;
A(i,3)

n;
n-1;
A(n,1)
n;
n;
A(n,2)

43

2001 - Gilberto E. Urroz

index = [row col];

//end function

An application, using the compact tri-dimensional matrix A presented earlier is shown below:
First, we load the function:
-->getf('tritosparse')

A call to the function with argument A produces the values and index for a sparse matrix:
-->[index,val] = tritosparse(A)
val =
!
4. !
! - 1. !
! - 1. !
!
4. !
! - 1. !
! - 1. !
!
4. !
! - 1. !
! - 1. !
!
4. !
index
1.
1.
2.
2.
2.
3.
3.
3.
4.
4.

!
!
!
!
!
!
!
!
!
!

=
1.
2.
1.
2.
3.
2.
3.
4.
3.
4.

!
!
!
!
!
!
!
!
!
!

Next, we put together the sparse matrix using the matrix index and the vector val:
-->As = sparse(index,val)
As =
(

4,

4) sparse matrix

(
(
(
(
(
(
(
(
(
(

1,
1,
2,
2,
2,
3,
3,
3,
4,
4,

1)
2)
1)
2)
3)
2)
3)
4)
3)
4)

4.
- 1.
- 1.
4.
- 1.
- 1.
4.
- 1.
- 1.
4.

To see the full form of the sparse matrix we use:

-->full(As)

Download at InfoClearinghouse.com

44

2001 - Gilberto E. Urroz

ans

!
4.
! - 1.
!
0.
!
0.

- 1.
4.
- 1.
0.

0.
- 1.
4.
- 1.

0.
0.
- 1.
4.

!
!
!
!

For the right-hand side vector previously defined we can use function lusolve to obtain the
solution of the tri-diagonal system of linear equations:
-->x = lusolve(As,b)
x =
!
!
!
!

44.976077
29.904306
54.641148
38.660287

!
!
!
!

Iterative solutions to systems of linear equations

With a powerful numerical environment like SCILAB, which provides a number of preprogrammed functions to obtain the solution to systems of linear equations, there is really no
need to use iterative methods. They are presented here only as programming exercises.
Iterative methods result from re-arranging the system of linear equations
a11x1 + a12x2 + a13x3 + + a1,n-1x n-1 + a1,nx n = b1,
a21x1 + a22x2 + a23x3 + + a2,n-1x n-1 + a2,nx n = b2,
.
.
.

.
.
.
.
.
.

.
.
.
an1x1 + an2x2 + an3x3 + + an,n-1x n-1 + an,nx n = bn.
as
x1 = (b1 - a12x2 - a13x3- - a1nxn)/a11
x2 = (b2 - a22x2 - a23x3- - a2nxn)/a22
.

xn = (bn - an2x2 - an3x3- - annxn)/ann

The solution is obtained by first assuming a first guess for the solution x(0) = [x10, x20, , xn0],
which is then replaced in the right-hand side of the recursive equations shown above to
produce a new solution x(1). With this new value in the right-hand side, a new approximation,
x(2) is obtained, and so on until the solution converges to a fixed value.
If you replace each value of xi in the recursive equations as it gets calculated the approach is
called the Gauss-Seidel method. An alternative, presented below as a SCILAB function, is the
Jacobi iterative method.

Download at InfoClearinghouse.com

45

2001 - Gilberto E. Urroz

Jacobi iterative method

In the Jacobi iterative method the new values of xis are replaced only once in each iteration.
This method calculates the residual, R, from the equation Ax = b, as R = Ax - b, and adjusts
the value of R recalculating x until the average of the residuals is smaller than a set value of
tolerance, .
To program the Jacobi iterative method we create three files, as follows:
function [x]= Jacobi(A,b)
//Uses Jacobi iterative process to calculate x
getf('xbar.txt'); // this function calculates average absolute value
getf('Nextx.txt');// this function calculates a new value of x
[n m] = size(A); // determine size of matrix A
// check if matrix is square
if n <> m then
error('Matrix must be square');
abort;
end;
//initialize variables
x = zeros(n,1);
//x = [0 0 ... 0]
Itmax = 100.; eps = 0.001;
//max iterations and tolerance
Iter = 1; R = b-A*x; Rave = xbar(R);
//first value of residuals
disp(Iter, 'iteration'); disp(Rave,' Average residual:')
while ((Rave > eps) & (Iter < Itmax))
//Loop while no convergence
x = Nextx(x,R,A)
R = b-A*x;
Rave = xbar(R);
Iter = Iter + 1
disp(Iter, 'iteration'); disp(Rave,' Average residual:')
end;
x
//end function Jacobi

function [xb] = xbar(x)

//Calculates average of absolute values of vector x
[n m] = size(x);
suma = 0.0;
for i = 1:n
suma = suma + abs(x(i,1));
end;
xb = suma/n
//end function xbar

function [x]=Nextx(x,R,A)
//calculates residual average for the system A*x = b
[n m] = size(A);
for i = 1:n
x(i,1) = x(i,1) + R(i,1)/A(i,i)
end;

Download at InfoClearinghouse.com

46

2001 - Gilberto E. Urroz

x
//end function Nextx

As an exercise, within SCILAB enter the following:

-->A = [4 -1 0 0; -1 4 -1 0; 0 -1 4 -1; 0 0 -1 4]; b = [150; 200; 150; 100];
-->getf('Jacobi.txt')
-->Jacobi(A,b)
A few things to notice in these functions:
1.
Function Jacobi loads the functions xbar.txt and Nextx.txt.
2.
The function size, when used with a matrix or vector, requires two
components. For that reason, whenever we use it in any of the functions Jacobi, xbar or
Nextx, we always use [n m] = size(). Therefore, n = number of rows, m = number of
columns.
3.
To initialize the vector x we use the statement: x = zeros(n,1). The general
form of the function zeros is zeros(m,n). See also, help ones.
4.
The vector x having one column is treated as a matrix. See also the use of
vector x in functions xbar.txt and Nextx.txt.
5.
The function disp (display) prints the values in the list of variables, from the
last to the first. Thus, the command disp(iter, 'iteration: ') prints a line with
'iteration:' and the next line with the value of iter.
6.
A while loop is used in function Jacobi.txt, which uses a compound logical
expression, namely, ((Rave > eps) & (Iter < Itmax)). So, as long as the average residual
is larger than the tolerance (eps) and the number of iterations is less than the maximum
number of iterations allowed (Itmax), the loop statements are repeated.
7.
In function xbar, the word suma is used to hold a sum. The word sum is
reserved by SCILAB (try: help sum) and cannot be redefined as a variable. By the way, suma is
Spanish for sum.

Download at InfoClearinghouse.com

47

2001 - Gilberto E. Urroz

REFERENCES (for all SCILAB documents at InfoClearinghouse.com)

Abramowitz, M. and I.A. Stegun (editors), 1965,"Handbook of Mathematical Functions with Formulas, Graphs, and
Mathematical Tables," Dover Publications, Inc., New York.
Arora, J.S., 1985, "Introduction to Optimum Design," Class notes, The University of Iowa, Iowa City, Iowa.
Asian Institute of Technology, 1969, "Hydraulic Laboratory Manual," AIT - Bangkok, Thailand.
Berge, P., Y. Pomeau, and C. Vidal, 1984,"Order within chaos - Towards a deterministic approach to turbulence," John
Wiley & Sons, New York.
Bras, R.L. and I. Rodriguez-Iturbe, 1985,"Random Functions and Hydrology," Addison-Wesley Publishing Company,
Reading, Massachussetts.
Brogan, W.L., 1974,"Modern Control Theory," QPI series, Quantum Publisher Incorporated, New York.
Browne, M., 1999, "Schaum's Outline of Theory and Problems of Physics for Engineering and Science," Schaum's
outlines, McGraw-Hill, New York.
Farlow, Stanley J., 1982, "Partial Differential Equations for Scientists and Engineers," Dover Publications Inc., New
York.
Friedman, B., 1956 (reissued 1990), "Principles and Techniques of Applied Mathematics," Dover Publications Inc., New
York.
Gomez, C. (editor), 1999, Engineering and Scientific Computing with Scilab, Birkhuser, Boston.
Gullberg, J., 1997, "Mathematics - From the Birth of Numbers," W. W. Norton & Company, New York.
Harman, T.L., J. Dabney, and N. Richert, 2000, "Advanced Engineering Mathematics with MATLAB - Second edition,"
Brooks/Cole - Thompson Learning, Australia.
Harris, J.W., and H. Stocker, 1998, "Handbook of Mathematics and Computational Science," Springer, New York.
Hsu, H.P., 1984, "Applied Fourier Analysis," Harcourt Brace Jovanovich College Outline Series, Harcourt Brace
Jovanovich, Publishers, San Diego.
Journel, A.G., 1989, "Fundamentals of Geostatistics in Five Lessons," Short Course Presented at the 28th International
Geological Congress, Washington, D.C., American Geophysical Union, Washington, D.C.
Julien, P.Y., 1998,Erosion and Sedimentation, Cambridge University Press, Cambridge CB2 2RU, U.K.
Keener, J.P., 1988, "Principles of Applied Mathematics - Transformation and Approximation," Addison-Wesley
Publishing Company, Redwood City, California.
Kitanidis, P.K., 1997,Introduction to Geostatistics - Applications in Hydogeology, Cambridge University Press,
Cambridge CB2 2RU, U.K.
Koch, G.S., Jr., and R. F. Link, 1971, "Statistical Analysis of Geological Data - Volumes I and II," Dover Publications,
Inc., New York.
Korn, G.A. and T.M. Korn, 1968, "Mathematical Handbook for Scientists and Engineers," Dover Publications, Inc., New
York.
Kottegoda, N. T., and R. Rosso, 1997, "Probability, Statistics, and Reliability for Civil and Environmental Engineers,"
The Mc-Graw Hill Companies, Inc., New York.
Kreysig, E., 1983, "Advanced Engineering Mathematics - Fifth Edition," John Wiley & Sons, New York.
Lindfield, G. and J. Penny, 2000, "Numerical Methods Using Matlab," Prentice Hall, Upper Saddle River, New Jersey.
Magrab, E.B., S. Azarm, B. Balachandran, J. Duncan, K. Herold, and G. Walsh, 2000, "An Engineer's Guide to
MATLAB", Prentice Hall, Upper Saddle River, N.J., U.S.A.

Download at InfoClearinghouse.com

48

2001 - Gilberto E. Urroz

McCuen, R.H., 1989,Hydrologic Analysis and Design - second edition, Prentice Hall, Upper Saddle River, New Jersey.
Middleton, G.V., 2000, "Data Analysis in the Earth Sciences Using Matlab," Prentice Hall, Upper Saddle River, New
Jersey.
Montgomery, D.C., G.C. Runger, and N.F. Hubele, 1998, "Engineering Statistics," John Wiley & Sons, Inc.
Newland, D.E., 1993, "An Introduction to Random Vibrations, Spectral & Wavelet Analysis - Third Edition," Longman
Scientific and Technical, New York.
Nicols, G., 1995, Introduction to Nonlinear Science, Cambridge University Press, Cambridge CB2 2RU, U.K.
Parker, T.S. and L.O. Chua, , "Practical Numerical Algorithms for Chaotic Systems, 1989, Springer-Verlag, New York.
Peitgen, H-O. and D. Saupe (editors), 1988, "The Science of Fractal Images," Springer-Verlag, New York.
Peitgen, H-O., H. Jrgens, and D. Saupe, 1992, "Chaos and Fractals - New Frontiers of Science," Springer-Verlag, New
York.
Press, W.H., B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling, 1989, Numerical Recipes - The Art of Scientific
Computing (FORTRAN version), Cambridge University Press, Cambridge CB2 2RU, U.K.
Raghunath, H.M., 1985, "Hydrology - Principles, Analysis and Design," Wiley Eastern Limited, New Delhi, India.
Recktenwald, G., 2000, "Numerical Methods with Matlab - Implementation and Application," Prentice Hall, Upper
Saddle River, N.J., U.S.A.
Rothenberg, R.I., 1991, "Probability and Statistics," Harcourt Brace Jovanovich College Outline Series, Harcourt Brace
Jovanovich, Publishers, San Diego, CA.
Sagan, H., 1961,"Boundary and Eigenvalue Problems in Mathematical Physics," Dover Publications, Inc., New York.
Spanos, A., 1999,"Probability Theory and Statistical Inference - Econometric Modeling with Observational Data,"
Cambridge University Press, Cambridge CB2 2RU, U.K.
Spiegel, M. R., 1971 (second printing, 1999), "Schaum's Outline of Theory and Problems of Advanced Mathematics for
Engineers and Scientists," Schaum's Outline Series, McGraw-Hill, New York.
Tanis, E.A., 1987, "Statistics II - Estimation and Tests of Hypotheses," Harcourt Brace Jovanovich College Outline
Series, Harcourt Brace Jovanovich, Publishers, Fort Worth, TX.
Tinker, M. and R. Lambourne, 2000, "Further Mathematics for the Physical Sciences," John Wiley & Sons, LTD.,
Chichester, U.K.
Tolstov, G.P., 1962, "Fourier Series," (Translated from the Russian by R. A. Silverman), Dover Publications, New York.
Tveito, A. and R. Winther, 1998, "Introduction to Partial Differential Equations - A Computational Approach," Texts in
Applied Mathematics 29, Springer, New York.
Urroz, G., 2000, "Science and Engineering Mathematics with the HP 49 G - Volumes I & II", www.greatunpublished.com,
Charleston, S.C.
Urroz, G., 2001, "Applied Engineering Mathematics with Maple", www.greatunpublished.com, Charleston, S.C.
Winnick, J., , "Chemical Engineering Thermodynamics - An Introduction to Thermodynamics for Undergraduate
Engineering Students," John Wiley & Sons, Inc., New York.

Download at InfoClearinghouse.com

49

2001 - Gilberto E. Urroz