Different pauli_error result from previous releases for two-qubit gate.

[JamesB-1qbit]

Informations

• Qiskit Aer version: 0.10.1
• Python version: 3.9.9
• Operating system: MacOS

What is the current behavior?

The tensor of two pauli_error gates after a two qubit gate is different from previous versions. Applying the gate

import qiskit.providers.aer.noise as noise
error_1 = noise.pauli_error([('X', 1/3), ('Y', 1/3), ('Z', 1/3), ('I', 0)])
error_2 = error_1.tensor(error_1)

to a circuit with only a cx gate. (essentially identity circuit) results in the dictionary

{'00': 557, '01' 467}

Steps to reproduce the problem

from qiskit import QuantumCircuit, execute, Aer
import qiskit.providers.aer.noise as noise
error_1 = noise.pauli_error([('X', 1/3), ('Y', 1/3), ('Z', 1/3), ('I', 0)])
err = error_1.tensor(error_1)

noise_model = noise.NoiseModel()
noise_model.add_all_qubit_quantum_error(err, ['cx'])

circ = QuantumCircuit(2, 2)
circ.cx(0, 1)
circ.measure([0, 1], [0, 1])

result = execute(circ, Aer.get_backend('qasm_simulator'), noise_model=noise_model).result()
counts = result.get_counts()
print(counts)

What is the expected behavior?

In previous versions, it results in something close to what I would expect given my understanding of the pauli_error which would be.

counts = {'01': 2 / 9 * shots, '11': 4 / 9 * shots, '10': 2 / 9 * shots, '00': 1 / 9 * shots}

Suggested solutions

Revert to the old behavior or explain in the documentation what a tensored pauli_error is supposed to do in practice.

[chriseclectic]

It looks like a bug in the QuantumError.to_dict method, which is flatting all instructions to qubits [0, ..., n-1]. I will put in a fix for this