AerSimulator from NoiseModel does not work with dynamic circuits

[anedumla]

Informations

• Qiskit Aer version: 0.13.0
• Python version: 3.10.8
• Operating system:

What is the current behavior?

The custom noise is not applied when the circuit contains dynamic instructions. The output of the following example is {'0 0': 1024}, i.e. the result without noise.

from qiskit import QuantumCircuit

from qiskit_aer.noise import NoiseModel, pauli_error
from qiskit_aer import AerSimulator

# Create the noise model
p_error = 0.3

readout_error = pauli_error([('X', p_error), ('I', 1 - p_error)])
noise_bit_flip = NoiseModel()
noise_bit_flip.add_all_qubit_quantum_error(readout_error, ['id'])

backend = AerSimulator(noise_model=noise_bit_flip)

test = QuantumCircuit(1,1)
test.id(0)
test.measure(0,0)
with test.if_test((0, 1)):
        test.x(0)
test.measure_all()
backend.run(test).result().get_counts()

However, when commenting out the lines containing dynamic instructions:

with test.if_test((0, 1)):
        test.x(0)

the output is {'0 0': 730, '1 1': 294}, as expected.

[doichanj]

I think this script always returns 0 at mid-measure and always skips x gate.

Following circuit works well.

test = QuantumCircuit(1,1)
test.h(0)
test.measure(0,0)
with test.if_test((0, 1)):
        test.x(0)
test.measure_all()

[anedumla]

In my original script, since the only noise is attached to the id gate, and from the results of the example without the dynamic instruction, the expected result would be {'0 0': 730, '0 1': 294}. I.e., because the id gate is noisy and flips the qubit with prob 0.3 to 1, you would expect that with probability 0.3, the dynamic X gate acts and the final result is with prob. 0.3 0 1, because measurements are noiseless in this example.

[doichanj]

I found id gate is removed by qiskit.compiler.transpile called from here

qiskit-aer/qiskit_aer/backends/aer_compiler.py

Lines 101 to 103 in 880df28

	compiled_circ = transpile(
	self._inline_circuit(circuit, None, None), basis_gates=basis_gates
	)

I think we can avoid by adding optimization_level=0, I will make PR for this fix