Performance bottleneck when computing expectation values with default.clifford

[JeremieGince]

Hi,

I am the creator of MatchCake, a free-fermions simulator built on top of PennyLane. As part of my simulations, I rely on a Clifford-based subroutine to compute expectation values using PennyLane’s default.clifford device.

This subroutine is implemented here:
https://github.com/MatchCake/MatchCake/blob/864063be8bece84d0238fbfc648e1fd2e1ecacb9/src/matchcake/devices/expval_strategies/clifford_expval/clifford_expval_strategy.py#L26.

While profiling MatchCake within another project, I noticed that this subroutine is a significant performance bottleneck. In particular, computing expectation values via DefaultClifford._measure_expectation appears to take substantially more time than expected.

I wanted to ask whether you have any suggestions for optimizing this workflow. Maybe there is a way to batch the expectation value observables for instance?

This may, in a way, be related to the following issue: #8946.

Thank you very much for your time and insights.

Best regards,
Jérémie Gince

[CatalinaAlbornoz]

Hi @JeremieGince , thanks for opening this issue! Our team will review and triage this issue. We'll let you know when we have any insights.