germsel.find_germs() does not support operations.ComposedOp() and is limited in large models.

[Pablovinas]

germsel.find_germs() Fails with Composed Gates in Two-Qubit Systems

When performing GST on a two-qubit system , the function germsel.find_germs() fails to find germs if any of the gates in the model are defined using operations.ComposedOp(). It is usually the case that some layers in two-qubit systems are built using the composition of one-qubit gates via tensor product. If one considers crosstalk, each of those layers should be regarded as an independent gate to characterise. However, germsel.find_germs() currently only supports gates defined directly in the two-qubit Hilbert space and not those created using operations.ComposedOp(). If that is the case, an error is encountered with the message: TypeError: unsupported operand type(s) for *: 'float' and 'ComposedOp'. This is because the function randomize_with_unitary()contains the code line: mdl_randomized.operations[opLabel] = _op.FullArbitraryOp(_np.dot(randOp, gate)). To solve this issue, I think it would be necessary to substitute gate -> gate.to_dense() in the mentioned code line, so that the .dot() product is compatible with the format of the matrix.

Gate Label Limitation in Circuit.to_pythonstr()

Additionally, I have observed that germsel.find_germs() utilises the function Circuit.to_pythonstr(). This encodes circuits as python strings, using a single letter for each gate in the model, and therefore imposes a limit of 26 distinct gates on it. While 26 gates might seem sufficient in most cases, the number of layers to characterise can scale rapidly beyond that limit when crosstalk is considered. I would suggest using a more flexible labelling system.

[coreyostrove]

Thanks for the report, @Pablovinas. Would you be a able to attach a minimal python script which reproduces the behavior you observed with germ selection for models containing ComposedOp?

[Pablovinas]

Hi @coreyostrove, thanks for your answer! I am not able to attach a script here, but you can find this minimal example .ipynb file I just uploaded on my GitHub profile. It provides some further detail.