Sampler with shots=None does not calculate resets exactly

[garrison]

Informations

• Qiskit Aer version:
• Python version:
• Operating system:

What is the current behavior?

A circuit with reset operation(s) is not guaranteed to be deterministic when shots=None, a setting which is supposed to cause probabilities to be calculated exactly.

Steps to reproduce the problem

from qiskit_aer.primitives import Sampler as AerSampler
qc = QuantumCircuit(2, 2)
qc.h(0)
qc.cx(0, 1)
qc.reset(0)
qc.measure(0, 0)
qc.measure(1, 1)
AerSampler().run(qc, shots=None).result().quasi_dists

is not deterministic. The last line evaluates to (on successive runs):

• [{0: 0.5166015625, 2: 0.4833984375}]
• [{0: 0.4833984375, 2: 0.5166015625}]
• [{0: 0.505859375, 2: 0.494140625}]

What is the expected behavior?

Should evaluate to [{0: 0.5, 2: 0.5}]

Suggested solutions

[none]

Similar to #1810, I have created my own Sampler implementation based on Qiskit Terra's Statevector that supports resets exactly: https://github.com/Qiskit-Extensions/circuit-knitting-toolbox/blob/main/circuit_knitting_toolbox/utils/simulation.py

[hhorii]

I roughly read simulation.py. Here is my understanding:

• If measure or reset come, a Statevector is coped to two Statevector to simulate two cases when 0 and 1 are measured,
• c_if is not supported

This is very similar to #1766 (and #1766 supports c_if and other control-flow operations). We would like to re-think this issue after merging #1766.