Incorrect results from statevector simulation when fusion is enabled

[hvenev-insait]

Informations

• Qiskit Aer version: 0.15.1
• Python version: 3.12.7
• Operating system: Fedora 40

What is the current behavior?

On some circuits AerSimulator(method='statevector', fusion_enable=...) produces very different results based on the value of fusion_enable. I think the fusion_enable=False result is the correct one.

Steps to reproduce the problem

import qiskit
import qiskit_aer
import numpy

q = qiskit.circuit.QuantumRegister(15)
circuit = qiskit.QuantumCircuit(q)
circuit.h(q[0])
circuit.cx(q[1], q[0])
circuit.x(q[0])
circuit.s(q[0])
circuit.x(q[0])
circuit.save_statevector()

sim1 = qiskit_aer.AerSimulator(method='statevector', precision='double', device='CPU', fusion_enable=True)
sim2 = qiskit_aer.AerSimulator(method='statevector', precision='double', device='CPU', fusion_enable=False)

res1 = sim1.run(circuit, shots=1).result()
res2 = sim2.run(circuit, shots=1).result()

sv1 = numpy.asarray(res1.data()['statevector'])[:4]
# [0.7071067811865476, 0.7071067811865475j, 0, 0]

sv2 = numpy.asarray(res2.data()['statevector'])[:4]
# [0.7071067811865476j, 0.7071067811865475, 0, 0]

# remove global phase
sv1 /= sv1[0]
sv2 /= sv2[0]

print(list(sv1))
# [1, 0.9999999999999998j, 0, 0]

print(list(sv2))
# [1, -0.9999999999999998j, 0, 0]

What is the expected behavior?

The difference between simulation results with and without fusion should be minimal.

Suggested solutions