Question

I have the following problem using numpy 1.3.0 and MATLAB 7.9.0:
the python code

import numpy as np    
Lu = [[1.01250000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[-0.00250000000000000,1.01250000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,-0.00250000000000000,1.01250000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0,0,0,0,0],[0,0,-0.00250000000000000,1.01250000000000,0,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0,0,0,0],[-0.00250000000000000,0,0,0,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0,0,0],[0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0,0],[0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0,0],[0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,0,0,0,-0.00250000000000000,0,0,0,0,0,0,0,0],[0,0,0,0,-0.00250000000000000,0,0,0,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0,0],[0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0,0],[0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0,0,0],[0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,0,0,0,-0.00250000000000000,0,0,0,0],[0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,0,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0,0],[0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0,0],[0,0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,-0.00250000000000000,0,0,-0.00250000000000000,0],[0,0,0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01000000000000,0,0,0,-0.00250000000000000],[0,0,0,0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,0,1.01250000000000,-0.00250000000000000,0,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01250000000000,-0.00250000000000000,0],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01250000000000,-0.00250000000000000],[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,-0.00250000000000000,0,0,-0.00250000000000000,1.01250000000000]]
rhs = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0, 0,0, 0,  0, 0, 0.0050, 0.0050, 0.0050, 0.0050]

Lu = np.array(Lu)
rhs = np.array(rhs)
ans = np.linalg.solve(Lu,rhs)
print ans

produces the output

[  1.87241716e-13   1.89545264e-13   1.89545264e-13   1.87241716e-13
   7.56433496e-11   7.63890449e-11   7.63890449e-11   7.56433496e-11
   3.04833369e-08   3.07089522e-08   3.07089522e-08   3.04833369e-08
   1.22844835e-05   1.23451480e-05   1.23451480e-05   1.22844835e-05
   4.95055571e-03   4.96277946e-03   4.96277946e-03   4.95055571e-03]

whereas using backslash in MATLAB produces the output

 0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0000    0.0050    0.0050    0.0050    0.0050

I have not found any other system of linear algebraic equation where numpy and matlab produce different solutions. I am currently out of town and therefore unable to check if a different version of numpy (on a different computer) would give the correct result. Is np.linalg.solve not the correct function to use solve this system (system matrix Lu is sparse)? Is this a bug in my version of numpy? Is there some problem in my code?

Thanks!

Answer 1

Actually, those are likely the same solution. MATLAB is rounding to 0.0000 (when printing at least), while python is giving you much more detailed numbers (some problems might be from floating point rounding errors). The only numbers that are showing up as 0.0050 are the ones that are e-03. All of the other numbers are smaller than 0.0005, so it might get rounded to 0.0000.