def test_random_pauli(self):
"""Test the Random Pauli circuit."""
circuit = PauliTwoDesign(4, seed=12, reps=1)
qr = QuantumRegister(4, "q")
params = circuit.ordered_parameters
# expected circuit for the random seed 12
expected = QuantumCircuit(qr)
# initial RYs
expected.ry(np.pi / 4, qr)
# first random Pauli layer
expected.ry(params[0], 0)
expected.rx(params[1], 1)
expected.rz(params[2], 2)
expected.rz(params[3], 3)
# entanglement
expected.cz(0, 1)
expected.cz(2, 3)
expected.cz(1, 2)
# second random Pauli layer
expected.rx(params[4], 0)
expected.rx(params[5], 1)
expected.rx(params[6], 2)
expected.rx(params[7], 3)
self.assertEqual(circuit.decompose(), expected)
def test_assign_keeps_one_initial_layer(self):
"""Test assigning parameters does not add an additional initial layer."""
circuit = PauliTwoDesign(2)
values = list(range(circuit.num_parameters))
bound0 = circuit.assign_parameters(values)
bound1 = circuit.assign_parameters(values)
bound2 = circuit.assign_parameters(values)
self.assertEqual(bound0, bound1)
self.assertEqual(bound0, bound2)
def test_resize(self):
"""Test resizing the Random Pauli circuit preserves the gates."""
circuit = PauliTwoDesign(1)
top_gates = [op.name for op, _, _ in circuit.data]
circuit.num_qubits = 3
with self.subTest('assert existing gates remain'):
new_top_gates = []
for op, qargs, _ in circuit:
if qargs == [circuit.qubits[0]]: # if top qubit
new_top_gates.append(op.name)
self.assertEqual(top_gates, new_top_gates)
def test_pauli_two_design(self):
"""Test standard gradient descent on the Pauli two-design example."""
circuit = PauliTwoDesign(3, reps=3, seed=2)
parameters = list(circuit.parameters)
obs = Z ^ Z ^ I
expr = ~StateFn(obs) @ StateFn(circuit)
initial_point = np.array([
0.1822308,
-0.27254251,
0.83684425,
0.86153976,
-0.7111668,
0.82766631,
0.97867993,
0.46136964,
2.27079901,
0.13382699,
0.29589915,
0.64883193,
])
def objective(x):
return expr.bind_parameters(dict(zip(parameters, x))).eval().real
optimizer = GradientDescent(maxiter=100,
learning_rate=0.1,
perturbation=0.1)
result = optimizer.optimize(circuit.num_parameters,
objective,
initial_point=initial_point)
self.assertLess(result[1], -0.95) # final loss
self.assertEqual(result[2], 100) # function evaluations
def test_resize(self):
"""Test resizing the Random Pauli circuit preserves the gates."""
circuit = PauliTwoDesign(1)
top_gates = [
instruction.operation.name
for instruction in circuit.decompose().data
]
circuit.num_qubits = 3
decomposed = circuit.decompose()
with self.subTest("assert existing gates remain"):
new_top_gates = []
for instruction in decomposed:
if instruction.qubits == (
decomposed.qubits[0], ): # if top qubit
new_top_gates.append(instruction.operation.name)
self.assertEqual(top_gates, new_top_gates)
def test_random_pauli(self):
"""Test the Random Pauli circuit."""
circuit = PauliTwoDesign(4, seed=12, reps=1)
qr = QuantumRegister(4, "q")
params = circuit.ordered_parameters
# expected circuit for the random seed 12
#
# ┌─────────┐┌──────────┐ ┌──────────┐
# q_0: ┤ Ry(π/4) ├┤ Ry(θ[0]) ├─■─┤ Rx(θ[4]) ├────────────
# ├─────────┤├──────────┤ │ └──────────┘┌──────────┐
# q_1: ┤ Ry(π/4) ├┤ Rx(θ[1]) ├─■──────■──────┤ Rx(θ[5]) ├
# ├─────────┤├──────────┤ │ ├──────────┤
# q_2: ┤ Ry(π/4) ├┤ Rz(θ[2]) ├─■──────■──────┤ Rx(θ[6]) ├
# ├─────────┤├──────────┤ │ ┌──────────┐└──────────┘
# q_3: ┤ Ry(π/4) ├┤ Rz(θ[3]) ├─■─┤ Rx(θ[7]) ├────────────
# └─────────┘└──────────┘ └──────────┘
expected = QuantumCircuit(qr)
# initial RYs
expected.ry(np.pi / 4, qr)
# first random Pauli layer
expected.ry(params[0], 0)
expected.rx(params[1], 1)
expected.rz(params[2], 2)
expected.rz(params[3], 3)
# entanglement
expected.cz(0, 1)
expected.cz(2, 3)
expected.cz(1, 2)
# second random Pauli layer
expected.rx(params[4], 0)
expected.rx(params[5], 1)
expected.rx(params[6], 2)
expected.rx(params[7], 3)
self.assertEqual(circuit.decompose(), expected)
def test_pauli_two_design(self, method):
"""Test SPSA on the Pauli two-design example."""
circuit = PauliTwoDesign(3, reps=1, seed=1)
parameters = list(circuit.parameters)
obs = Z ^ Z ^ I
expr = ~StateFn(obs) @ StateFn(circuit)
initial_point = np.array([
0.82311034, 0.02611798, 0.21077064, 0.61842177, 0.09828447,
0.62013131
])
def objective(x):
return expr.bind_parameters(dict(zip(parameters, x))).eval().real
settings = {"maxiter": 100, "blocking": True, "allowed_increase": 0}
if method == "2spsa":
settings["second_order"] = True
settings["regularization"] = 0.01
expected_nfev = settings["maxiter"] * 5 + 1
elif method == "qnspsa":
settings["fidelity"] = QNSPSA.get_fidelity(circuit)
settings["regularization"] = 0.001
settings["learning_rate"] = 0.05
settings["perturbation"] = 0.05
expected_nfev = settings["maxiter"] * 7 + 1
else:
expected_nfev = settings["maxiter"] * 3 + 1
if method == "qnspsa":
spsa = QNSPSA(**settings)
else:
spsa = SPSA(**settings)
with self.assertWarns(DeprecationWarning):
result = spsa.optimize(circuit.num_parameters,
objective,
initial_point=initial_point)
with self.subTest("check final accuracy"):
self.assertLess(result[1], -0.95) # final loss
with self.subTest("check number of function calls"):
self.assertEqual(result[2], expected_nfev) # function evaluations