CODEWITHSUNDEEP
pythonfortran
Anna-Kat
Anna-Kat

Reputation: 203

Matmul in Fortran

I obtain different results from matmul in Fortran and Python. How to get the right one in Fortran. B is probably not understood as a 3x3 matrix. How to solve it, please?

k = 1 m = 3

In Fortran, I have:

 A(k:m,k):  0.88807383397711526       0.32505758367186816       0.32505758367186816     
 B(k:m,k:m):   1.0000000000000000        1.0000000000000000        1.0000000000000000       -1.0000000000000000        0.0000000000000000        1.0000000000000000        1.0000000000000000        0.0000000000000000        1.0000000000000000

matmul gives

1.5381890013208515      -0.56301625030524716        1.2131314176489834

In Python, I have:

A[k,k:m] [ 0.888074  0.325058  0.325058]
B[k:m,k:m] [[ 1.000000  1.000000  1.000000]
 [-1.000000  0.000000  1.000000]
 [ 1.000000  0.000000  1.000000]]

matmul gives

 [ 0.888074  0.888074  1.538189]

Upvotes: 1

Views: 290

Answers (1)

Breno
Breno

Reputation: 812

It depends on the order of the multiplication. Numpy considers 1D arrays without treating it as a row or column vector.

import numpy as np
A = np.array([0.88807383397711526, 0.32505758367186816, 0.32505758367186816])
B = np.array([[1, 1, 1], [-1, 0, 1], [1, 0, 1]], dtype=float)

If you do A@B, which is the same as np.matmul(A,B), assuming A is actually a "1D row array", you get

>>> A@B
array([0.88807383, 0.88807383, 1.538189  ])

But if you do the opposite, i.e. B@A which is the same as np.matmul(B,A), assuming A is actually a "1D column array" you get

>>> B@A
array([ 1.538189  , -0.56301625,  1.21313142])

Upvotes: 3

Related Questions