Wrong result when applying unitary matrix using cuquantum about qibojit

[scarrazza]

I confirm this issue, and if I do:

matrix = expm(1j * (matrix + matrix.T.conj())).flatten().reshape(2,2)

the issue disappears. Most likely expm is delivering an array with order != C, while reshape reprojects in C.
@stavros11 please try and let me know, an workaround is to cast matrix in C.

[stavros11]

I confirm this issue, and if I do:

matrix = expm(1j * (matrix + matrix.T.conj())).flatten().reshape(2,2)

the issue disappears. Most likely expm is delivering an array with order != C, while reshape reprojects in C. @stavros11 please try and let me know, an workaround is to cast matrix in C.

Thank you for finding the fix, it works for me too. Indeed doing np.array(x, order='C') works too. Yesterday I was trying with just np.array and it wasn't working, I am assuming it stays in the existing order if we don't specify a new one. I guess we could fix this by changing the platform casts to have an order argument. I will open a PR implementing this here.

We could just fix the qibo tests that use expm but I believe it is better to implement it here as some user may use it as well.

[scarrazza]

@stavros11 thanks, and I agree with the solution.