In transpile options, set routing = "none"

[kevinsung]

I think the optimization level can be left to the default. Routing should not be performed, so we set it to "none".

[nkanazawa1989]

I think disabling the gate optimization is also important. If we optimize the gate depending on angles, it could induce nonuniform error contribution and diverge the curve. In principle, GEM experiment measures the trend of metrics against the drift from Clifford limit, and we should assume underlying physical error is constant. @ehchen please confirm.

[kevinsung]

I think disabling the gate optimization is also important. If we optimize the gate depending on angles, it could induce nonuniform error contribution and diverge the curve. In principle, GEM experiment measures the trend of metrics against the drift from Clifford limit, and we should assume underlying physical error is constant. @ehchen please confirm.

In that case, we should actually set optimization_level = 0, right?

[kevinsung]

Ok after further discussion, we should keep optimization_level = 1 so that single-qubit gates are optimized.

[kevinsung]

@nkanazawa1989 does the PR as is look good to you?

[nkanazawa1989]

Looks good. Usually routing is not called because the circuit is coupling aware and the PM has a checker pass before calling routing. Perhaps there could be some edge case that the experiment makes edge between non-coupled qubits though. With this change PM can raise an error instead of inserting SWAPs.