Difference in eigenvalues in MatLab and numpy for hermitian matrices

1
votes

I'm doing some work with hermitian matrices. The initial work was done in matlab but I'm now doing some follow on work in python. I'm using numpy in python but I'm getting descrepencies between the eigenvalues reported in python and those in MatLab, can anyone tell me what's going wrong here?

I have more examples of this issue but a quick example is:

import numpy as np
mat = [[-1,1,-1],[1,-1,1],[-1,1,1]]
w,v = np.linalg.eig(mat)

gives 

w = array([ -2.56155281e+00,   2.01878445e-16,   1.56155281e+00])

In MatLab - 

mat = [-1,1,-1;1,-1,1;-1,1,1]
eig(mat)

ans =

   -2.5616
         0
    1.5616

Is this MatLab being imprecise or Python throwing errors? I have many others that agree on the values, the disagreement is an exception on my full data set. Any help in explaining this issue would be great.

Thanks

1
pythonmatlabnumpymatrixeigenvalue
Output from Matalb with format long: -2.561552812808829, 1.561552812808831 – Anton Protopopov
If your matrices are hermitian then use eigh instead of eig. – percusse
Possible duplicate of How to store more than 4 decimal places in an array in MATLAB – horchler

1 Answers

2
votes

Matlab is rounding the values for display. Try the following to get the same results as in python.

format long
mat = [-1,1,-1;1,-1,1;-1,1,1]
eig(mat)