There is a problem with lsqr with certain matrices. A minimal example is given by the following:

scipy.sparse.linalg.lsqr(np.array([1,4]),np.array([1]))

which fails due to acond being 0 and therefore having a float division with 0 in test3 = 1/acond. Changing the matrix a bit to scipy.sparse.linalg.lsqr(np.array([1,5]),np.array([1])) works. Matlab lsqr can handle both.

Furthermore is the output print(msg[0]) in the case arnorm==0 annoying. It should only be printed to stdout if show is set.