Matrix right division fails due to decorator about sympy

[oscarbenjamin]

There is no matrix division operation in Python. The operation you show above is:

In [100]: b.T.solve(a.T).T
Out[100]: 
⎡1  2⎤
⎢    ⎥
⎣3  4⎦

The transposes are needed because you're expecting the product a*inv(b) For inv(b)*a it's just a.solve(b).

[chrisjohgorman]

Further investigation into the error after the decorator, line 2729, can be rewritten as return self * other.inv(), which is works, but is also probably not desired. I believe the original reading should be something like, in matrix terms, A * B**(-1). I will wait on playing with this further until someone more experienced than me can comment.

[cbm755]

It might be that SymPy is not intending to support system solves via this operator (I mean, maybe it was just a coincidence that it worked before). But I'm not sure the current state.

Do the docs give any indication of what should happen and what what classes of objects?

Assuming this is a regression, then Gaussian elimination would be better than explicitly inv computation for example.

[chrisjohgorman]

I took a look through the documentation and didn't see anything for division. So, I suppose that sympy is not supporting division as it's not a common matrix operation. Given this appears to be a regression, are you suggesting doing something like the following to get back our [1, 2; 3, 4] matrix?

a = Matrix([[9, 19], [19, 41]])
b = Matrix([[1, 3], [4, 8]])
c = b.row_join(eye(2))
d = c.rref()
e = d[0]
e.col_del(0)
e.col_del(0)
a*e

Or is there a better solution?

[cbm755]

Ok, we'll do this downstream then, thanks @oscarbenjamin for the hints on a.solve(b) and similar.