Difference between revisions of "Software tutorial/Matrix operations"

From Process Model Formulation and Solution: 3E4
Jump to navigation Jump to search
m
Line 1: Line 1:
{{Navigation|Book=Software tutorial|previous=Vectors and arrays|current=Tutorial index|next=}}
{{Navigation|Book=Software tutorial|previous=Vectors and arrays|current=Tutorial index|next=}}
== Matrix operations ==
== LU decomposition ==
 
In this section we will show how to calculate the LU decomposition, matrix inverse, norms and condition number of a matrix in MATLAB and Python.


Let \(A = \left(\begin{array}{cccc}    4 & 0  & 1 \\ 0  & 6  & 2  \\ 3.2  & 7  &  4 \end{array} \right)\) and let \(b = \left(\begin{array}{cccc} 9.1 \\ 2  \\ 0 \end{array} \right) \)
Let \(A = \left(\begin{array}{cccc}    4 & 0  & 1 \\ 0  & 6  & 2  \\ 3.2  & 7  &  4 \end{array} \right)\) and let \(b = \left(\begin{array}{cccc} 9.1 \\ 2  \\ 0 \end{array} \right) \)
Line 102: Line 100:
</syntaxhighlight>
</syntaxhighlight>
|}
|}
== Matrix determinant ==
One way to test if a matrix is invertable is to calculate the determinant.  If it is non-zero, then \(A\) is invertable.
Let \(A = \left(\begin{array}{cccc}    4 & 0  & 1 \\ 0  & 6  & 2  \\ 3.2  & 7  &  4 \end{array} \right)\)
{| class="wikitable"
|-
! MATLAB
! Python
|-
| width="50%" valign="top" |
<syntaxhighlight lang="matlab">
A = [4, 0, 1; 0, 6, 2; 3.2, 7, 4];
>> det(A)
ans =
  20.8000
</syntaxhighlight>
| width="50%" valign="top" |
<syntaxhighlight lang="python">
import numpy as np
from scipy.linalg import *
A = np.array([[4, 0, 1],[0, 6, 2],[3.2, 7, 4]])
>>> det(A)
20.800000000000011
</syntaxhighlight>
|}
== Matrix inverse ==
The matrix inverse is defined so that \(A A^{-1} = I\).  Let \(A = \left(\begin{array}{cccc}    4 & 0  & 1 \\ 0  & 6  & 2  \\ 3.2  & 7  &  4 \end{array} \right)\)
{| class="wikitable"
|-
! MATLAB
! Python
|-
| width="50%" valign="top" |
<syntaxhighlight lang="matlab">
A = [4, 0, 1; 0, 6, 2; 3.2, 7, 4];
>> inv(A)
ans =
    0.4808    0.3365  -0.2885
    0.3077    0.6154  -0.3846
  -0.9231  -1.3462    1.1538
>> A * inv(A)
ans =
    1.0000        0        0
  -0.0000    1.0000    0.0000
        0        0    1.0000
</syntaxhighlight>
| width="50%" valign="top" |
<syntaxhighlight lang="python">
import numpy as np
from scipy.linalg import *
A = np.array([[4, 0, 1],[0, 6, 2],[3.2, 7, 4]])
>>> inv(A)
array([[ 0.48076923,  0.33653846, -0.28846154],
      [ 0.30769231,  0.61538462, -0.38461538],
      [-0.92307692, -1.34615385,  1.15384615]])
>>> np.dot(A, inv(A))
array([[  1.00000000e+00,  0.00000000e+00,  0.00000000e+00],
      [ -2.22044605e-16,  1.00000000e+00,  4.44089210e-16],
      [  0.00000000e+00,  0.00000000e+00,  1.00000000e+00]])
</syntaxhighlight>
Notice the small, essentially zero elements in the off-diagonal elements
|}
<!-- matrix inverse, determinant norms and condition number of a matrix in MATLAB and Python. -->

Revision as of 00:20, 5 October 2010

← Vectors and arrays (previous step) Tutorial index

LU decomposition

Let \(A = \left(\begin{array}{cccc} 4 & 0 & 1 \\ 0 & 6 & 2 \\ 3.2 & 7 & 4 \end{array} \right)\) and let \(b = \left(\begin{array}{cccc} 9.1 \\ 2 \\ 0 \end{array} \right) \)

MATLAB Python
A = [4, 0, 1; 0, 6, 2; 3.2, 7, 4];

% LU decomposition of A
[L,U,P] = lu(A)
L =
    1.0000         0         0
    0.8000    1.0000         0
         0    0.8571    1.0000
U =
    4.0000         0    1.0000
         0    7.0000    3.2000
         0         0   -0.7429
P =
         1         0         0
         0         0         1
         0         1         0

The P matrix shows that rows 2 and rows 3 were interchanged during the Gauss elimination pivoting steps.

import numpy as np
from scipy.linalg import *

A = np.array([[4, 0, 1],[0, 6, 2],[3.2, 7, 4]])
LU, P = lu_factor(A)

>>> LU
array([[ 4.        ,  0.        ,  1.        ],
       [ 0.8       ,  7.        ,  3.2       ],
       [ 0.        ,  0.85714286, -0.74285714]])

>>> P
array([0, 2, 2])

The LU matrices are overlayed in the output. Notice that the results are the same as MATLAB's, except we do not need to store the diagonal set of ones from the L matrix.

The P vector is also a bit cryptic - it should be read from left to right, in order. The way to read it is that row \(i\) of matrix \(A\) was interchanged with row P[i] in \(A\), where i is the index into vector P.

  • row 0 in \(A\) was interchanged with row 0: i.e. no pivoting for the first row
  • row 1 in \(A\) was interchanged with row 2: i.e. rows 2 and 3 in \(A\) were swapped
  • row 2 in \(A\) was interchanged with row 2: i.e. no further pivoting.

Once we have \(L, U\) and \(P\) we can solve the system of equations for a given vector, \(b\). It takes a bit more work to do this in MATLAB than with Python. Recall from class that \(A = LU\), but with a permutation or pivot matrix, it becomes \(PA = LU\), or alternatively: \(A = P^{-1}LU\)

\[ \begin{align} Ax &= b \\ P^{-1}LUx &= b \\ \text{Let}\qquad Ux &= y \\ \text{Then solve}\qquad \left(P^{-1}L\right)y &= b \qquad \text{for}\,\, y \\ \text{Finally, solve}\qquad Ux &= y \qquad \text{for}\,\, x \end{align} \]

MATLAB Python
b = [9.1, 2, 0]';
[L,U,P] = lu(A);
y = mldivide(inv(P)*L, b);
x = mldivide(U, y);
>> x
x =
    5.0481
    4.0308
  -11.0923

% Verify the solution:
>> A*x - b
ans =
   1.0e-14 *

    0.1776
         0
         0

The small values of the solution error indicate that our LU solution is accurate. You should compare the LU solution with solution from mldivide(A, b).

LU, P = lu_factor(A)
b = np.array([9.1, 2, 0])
x = lu_solve((LU,P), b)

>>> x
array([  5.04807692,   4.03076923, -11.09230769])

# Verify the solution:
>>> np.dot(A, x) - b
array([  1.77635684e-15,   0.00000000e+00,   0.00000000e+00])

Matrix determinant

One way to test if a matrix is invertable is to calculate the determinant. If it is non-zero, then \(A\) is invertable.

Let \(A = \left(\begin{array}{cccc} 4 & 0 & 1 \\ 0 & 6 & 2 \\ 3.2 & 7 & 4 \end{array} \right)\)

MATLAB Python
A = [4, 0, 1; 0, 6, 2; 3.2, 7, 4];
>> det(A)
ans =
   20.8000
import numpy as np
from scipy.linalg import *

A = np.array([[4, 0, 1],[0, 6, 2],[3.2, 7, 4]])
>>> det(A)
20.800000000000011

Matrix inverse

The matrix inverse is defined so that \(A A^{-1} = I\). Let \(A = \left(\begin{array}{cccc} 4 & 0 & 1 \\ 0 & 6 & 2 \\ 3.2 & 7 & 4 \end{array} \right)\)

MATLAB Python
A = [4, 0, 1; 0, 6, 2; 3.2, 7, 4];
>> inv(A)
ans =
    0.4808    0.3365   -0.2885
    0.3077    0.6154   -0.3846
   -0.9231   -1.3462    1.1538

>> A * inv(A)
ans =
    1.0000         0         0
   -0.0000    1.0000    0.0000
         0         0    1.0000
import numpy as np
from scipy.linalg import *

A = np.array([[4, 0, 1],[0, 6, 2],[3.2, 7, 4]])
>>> inv(A)
array([[ 0.48076923,  0.33653846, -0.28846154],
       [ 0.30769231,  0.61538462, -0.38461538],
       [-0.92307692, -1.34615385,  1.15384615]])

>>> np.dot(A, inv(A))
array([[  1.00000000e+00,   0.00000000e+00,   0.00000000e+00],
       [ -2.22044605e-16,   1.00000000e+00,   4.44089210e-16],
       [  0.00000000e+00,   0.00000000e+00,   1.00000000e+00]])

Notice the small, essentially zero elements in the off-diagonal elements