scipy > 1.9.3 breaks QuarticFermionicSimulationGate decompose method.

[fdmalone]

Tracking this here. A bug was introduced into scipy for linalg.sqrtm for complex valued matrices which appears starting v1.10.0 (see #18250). This causes our tests to fail for QuarticFermionicSimulationGate due to loss of precision. I'll pin the scipy dependency to < 1.10.0 for the moment.

[yurivict]

@fdmalone

Do you know if there is a fix for this?

[fdmalone]

It hasn't been fixed in scipy yet, but there does look to be an open PR. If there's no more activity there we may need to implement our own OR relax our test thresholds. The square root function is only used once in openfermion.

[mpharrigan]

This is becoming increasingly more of an issue, xref #851 . It's not clear from the linked issue whether there will be a fix or not.

Can we just bump our input precision to complex256?

[fdmalone]

I'll look into it. Worst case we could probably just write a custom square root function.

[fdmalone]

Fixed with #864