Performance difference for generating the unitary of a circuit using AerSimulator vs quantum_info.Operator

[ACE07-Sev]

Informations

• Qiskit AER version: 0.14.2
• Operating system: WSL : Ubuntu

Greetings there,

Hope all are well. I had a quick question I was hoping to ask. I understand there are two ways we can generate the unitary of a circuit. One is using qiskit.quantum_info.Operator, and the other is using AerSimulator like so:

# Construct quantum circuit without measure
circ = QuantumCircuit(2)
circ.h(0)
circ.cx(0, 1)
circ.save_unitary()

# Transpile for simulator
simulator = AerSimulator(method = 'unitary')
circ = transpile(circ, simulator)

# Run and get unitary
result = simulator.run(circ).result()
unitary = result.get_unitary(circ)
print("Circuit unitary:\n", np.asarray(unitary).round(5))

I was wondering what the main difference between the two approaches is, and if someone wants to generate the unitary (which is admittedly an expensive operation as we have to contract all the gates together) if they can do so using GPU (and whether this is then only doable with AerSimulator and thus the difference).

[doichanj]

I think it is better to use qiskit.quantum_info for smaller number of qubits, but Aer is suitable to make large unitary matrix, because Aer has some overheads. I think if you want to make > 7qubits unitary matrix, Aer will be faster

[ACE07-Sev]

I see. Would I be able to get the unitary matrix using GPU as a device when using Aer?

[doichanj]

Yes, you can use GPU by setting option device=GPU