stabilizer method fails with SamplerV2

[kevinsung]

Informations

• Qiskit Aer version: 0.14.2
• Python version: 3.12.3
• Operating system: Arch Linux

What is the current behavior?

stabilizer method fails with SamplerV2 even though it works with V1 Sampler.

Steps to reproduce the problem

import numpy as np
from qiskit.circuit.library import EfficientSU2
from qiskit_aer.primitives import Sampler, SamplerV2

n_qubits = 8
circuit = EfficientSU2(n_qubits)
circuit.measure_all()
rng = np.random.default_rng(1234)
params = rng.choice(
    [0, np.pi / 2, np.pi, 3 * np.pi / 2],
    size=circuit.num_parameters,
)

# works
sampler = Sampler(backend_options=dict(method="stabilizer"))
job = sampler.run(circuit.decompose(), params, shots=10)
quasis = job.result().quasi_dists[0]

# throws error
samplerv2 = SamplerV2(options=dict(backend_options=dict(method="stabilizer")))
pub = (circuit.decompose(), params, 10)
job = samplerv2.run([pub])
result = job.result()

QiskitError: 'ERROR:  [Experiment 0] Circuit EfficientSU2 contains invalid instructions {"gates": {ry}} for "stabilizer" method.Circuit EfficientSU2 contains invalid parameters  for "stabilizer" method. ,  ERROR: Circuit EfficientSU2 contains invalid instructions {"gates": {ry}} for "stabilizer" method.Circuit EfficientSU2 contains invalid parameters  for "stabilizer" method.'

What is the expected behavior?

Should work

Suggested solutions

[doichanj]

I think V2 primitives only accepts ISA circuits, so the circuits should be transpiled before passing to run() function (V1 primitive runs because it calls transpiler inside)

[kevinsung]

How do I obtain the appropriate transpiler target from the SamplerV2 instance?

[doichanj]

circuit = transpile(circuit, AerSimulator(method='stabilizer')) Do you think it is acceptable for users, or not?

[kevinsung]

It's less than ideal that the user has to initialize the AerSimulator object separately from the SamplerV2. This is not easy to discover.

[doichanj]

I do not know how users use qiskit.primitives with transpiler but I think we have to provide the similar way to use qiskit_aer.primitives

[kevinsung]

With qiskit.primitives, it seems that no transpilation is required. It can even handle the EfficientSU2 gate directly (for qiskit_aer.primitives I had to do circuit.decompose() to decompose that gate.

For example, this works:

import numpy as np
from qiskit.circuit.library import EfficientSU2
from qiskit.primitives import StatevectorSampler

n_qubits = 8
circuit = EfficientSU2(n_qubits)
circuit.measure_all()
rng = np.random.default_rng(1234)
params = rng.choice(
    [0, np.pi / 2, np.pi, 3 * np.pi / 2],
    size=circuit.num_parameters,
)

samplerv2 = StatevectorSampler()
pub = (circuit, params, 10)
job = samplerv2.run([pub])
result = job.result()

[kevinsung]

To match this behavior, the primitives in Qiskit Aer should not require any transpilation. It should handle any gate that can be simulated.

[kevinsung]

Closing in favor of #2165