Missing complex conjugation in scipy.sparse.linalg.LinearOperator

[ktgeier]

I noticed that there is a complex conjugation missing in the implementation of scipy.sparse.linalg.interface._ScaledLinearOperator._adjoint:

    def _adjoint(self):
        A, alpha = self.args
        return A.H * alpha

The correct version should be:

    def _adjoint(self):
        A, alpha = self.args
        return A.H * np.conj(alpha)

Considering also the other issues posted for the LinearOperator class, I have the impression that the implementation is not very mature yet. However, in my opinion, it is a very useful interface for linear operators, especially for matrix-free implementations, and therefore I consider it worth to further improve on it.

Reproducing code example:

>>> import numpy as np
>>> from scipy.sparse.linalg import aslinearoperator

>>> mat = np.arange(9.0).reshape((3, 3))
>>> vec = np.arange(3.0)

>>> result_array = (1j * mat).conj().T.dot(vec)
>>> result_linop = (1j * aslinearoperator(mat)).adjoint().dot(vec)

>>> print(result_array)
[0.-15.j 0.-18.j 0.-21.j]

>>> print(result_linop)
[0.+15.j 0.+18.j 0.+21.j]

Scipy/Numpy/Python version information:

1.1.0 1.14.3 sys.version_info(major=3, minor=6, micro=5, releaselevel='final', serial=0)

[rgommers]

Thanks for the report @laplacesdemon64. Fixed in gh-8962