sparse.linalg.expm_multiply causes ValueError when fed with (dense, sparse, *args)

[daniel-edler]

The following works

• expm_multiply(A, B) without optional parameter (which is equivalent to _expm_multiply_simple(A, B))
• expm_multiply(A, B, start, stop, num) where
  • type(A) == numpy.ndarray and type(B) == scipy.sparse.csc.csc_matrix
  • and of course type(A) == numpy.ndarray and type(B) == numpy.ndarray

The following fails

...and throws a ValueError: setting an array element with a sequence exception:

• expm_multiply(A, B, start, stop, num) where
  • type(A) == scipy.sparse.csc.csc_matrix and type(B) == numpy.ndarray
  • type(A) == scipy.sparse.csc.csc_matrix and type(B) == scipy.sparse.csc.csc_matrix

Example

import numpy as np
import scipy.sparse.linalg as linalg
from scipy.sparse import csc_matrix

A = np.array([[1, 0], [0, 1]])
B = csc_matrix([[np.exp(-1.), np.exp(-2.)],[0, 1]])

linalg.expm_multiply(A, B, start=1, stop=2, num=2)

Attempt to understand the problem

The traceback points to line 597 in scipy/scipy/sparse/linalg/_expm_multiply.py

 X[0] = _expm_multiply_simple_core(A, B, t_0, mu, m_star, s)

where _expm_multiply_simple_core(A, B, t_0, mu, m_star, s) returns a sparse matrix but X is a numpy matrix of type np.result_type(A.dtype, B.dtype, float) (In my case a float64) . So either the np.result_type function should have returned the type object to store the sparse matrices or in such cases _expm_multiply_simple_core(A, B, t_0, mu, m_star, s) should be transformed to a dense matrix.

Scipy/Numpy/Python version information:

1.1.0 1.15.1 sys.version_info(major=3, minor=6, micro=7, releaselevel='final', serial=0)