Using mitiq ZNE with qiskit_runtime.Sampler.

[Z035]

We are running a SamplingVQE with qiskits new qiskit_ibm_runtime.Sampler . The SampleVQE creates a jobs for every iteration with a unique ansatz. In each job, the sampler run and measure the circuit multiple times, where the result is a distribution of class QuasiDistribution. (see code below)

My question is this: I want to incorperate mitiq ZNE with the Sampler. But since ZNE returns an expectation value instead of a distribution, how should I do it?

[andreamari]

Unless you apply strong customization of the source code, mitiq.zne acts on expectation values and is not able to act on intermediate quasi-distributions.

In practice, in order to use mitiq.zne you need an executor, i.e., a function mapping circuits (without symbolic parameters) to expectation values. See e.g. this Qiskit example.

[natestemen]

We have talked about adding the functionality to mitigate distributions before. I guess we could run the noise scaled circuits as normal, but instead of receiving expectation values to extrapolate, we get counts which we use to extrapolate by converting the counts to probabilities. @andreamari do you think this would be useful?

[andreamari]

It could be useful. But it would also require a quite radical change in the structure of Mitiq code, since now most techniques are based on post-processing of expectation values.