Yael Ben-Haim

Probably can close this issue, @hhorii will decide.

@merav-aharoni @hhorii I remember that there were discussions about the measure sampling being implemented twice - in the statevector and the MPS simulator. I think there's a plan to unify...

@hhorii Please don't close this issue until the last point is resolved, thanks

The following pull request may be related: https://github.com/Qiskit/qiskit-terra/pull/5190 In Terra, probably the way to go is to allow initialization of `Statevector` from a dictionary. Then either to directly allow `initialize`...

Summary of a conversation: - Simulating the circuit until the expectation value only once - this is possible only in the case of noiseless simulation. Readout error is still possible....

The controller should not instruct the simulator to copy the data structure before changing the basis for each term. Because for some simulators it may be more efficient to switch...

Above, under "Need to support four types of users", we describe four cases, which we will now call "modes". Steps of work: - [ ] Assume that only modes 1...

Expectation value with shot noise is not really a snapshot, because it performs operations. It is not an instruction either, because it splits the circuit to new circuits, and there...

As for mode 4, the user can ask for the regular (accurate) expectation value snapshot, also when the noise model is not empty. Internally this means that an optimization similar...

Current steps: - [x] `sample_measure` optimization with readout error. - [x] Make `sample_measure` work on all the simulators together, if not done already. - [ ] Use the `sample_measure` mechanism...