Make expm work for more data types

[kcrisman]

This originally came up at this ask.sagemath question.

import scipy 
a=scipy.matrix([[1,0],[0,1]],dtype='complex256')
scipy.linalg.expm(a)
<snip>
ValueError: invalid type: complex256

And indeed the code only allows float64, float32, complex64, or complex128.

This is an enhancement request, since presumably the numerical analysis needed to correctly deal with this is nontrivial!

[argriffing]

Ignoring the numerical analysis I would have guessed that this might work for the newer expm, but I guess not.
TypeError: unsupported operand type(s) for *: 'numpy.float128' and 'long'
I'm not sure what that error means. Can we not multiply numpy float128 and python integers together?

[argriffing]

I want to say that this is a numpy problem.

>>> np.__version__
'1.7.1'
>>> np.array(42, dtype='float128') * long(42)
1764.0
>>> np.array(42, dtype='float128') * 1000111000111000111000111
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for *: 'numpy.float128' and 'long'

I think the float128 is still exotic and not completely supported by numpy in the same way that the smaller data types are supported. For example see this 2011 mailing list post. I'm not sure how many if any of those problems have been fixed in the numpy development version.

[rgommers]

Same for float96. So close this issue and open a numpy one?