MPS simulator can handle any topology (still required to have 1- or 2-qubit gates)

[tonybruguier]

No description provided.

[tonybruguier]

Hi,
I tried to be careful, but the size of some tensors is quite big (up to 8-dimensions!) and so it's possible that I made a mistake somewhere. I tried to have good test coverage, but I'm more than happy to add more tests, more comments, more assert statements, etc...

Thanks a lot in advance and this is not particularly urgent.

[mpharrigan]

density simulator -> mps simulator?

For this, I think you should really look into quimb (and the cirq.contrib.quimb simulator) to get a handle on the indices. The nicety afforded by quimb (or tensor network libraries in general) is you get to provide names for the indices instead of relying on their relative ordering or np.einsum letters or tensor[:, :, 1, :, :]

[tonybruguier]

I followed your advice and used Quimb, and the code is much nicer indeed.

The unit tests do pass on my machine, but not on Github because it cannot import quimb. I looked around and it seems other files in Cirq can import it, so I am not sure what is wrong. Do you have a suggestion?

Thanks so much for teaching me about Quimb! I'm a bit concerned that Cirq is just a subset of Quimb :)

PTAL

[mpharrigan]

Looking really good. Main outstanding thing is docstrings and answering some of my questions. I explicitly marked some of my other comments as "optional", since this PR is functional and readable at present and don't want to block merging it on those ideas

[tonybruguier]

Thanks. PTAL.

[mpharrigan]

sweet

[tonybruguier]

Thanks for all the help. I think I've addressed all the comments, but happy to do more if required.