def __init__(self, layers, domain, ansatz, function):
self.function = function
super().__init__(layers, domain, ansatz)
self.target = self.function(self.domain)
self.target = 2 * (self.target - np.min(self.target)) / (
np.max(self.target) - np.min(self.target)) - 1
self.H = Hamiltonian(1, matrices._Z)
self.classical = globals()[f"classical_real_{self.ansatz}"]
def __init__(self, layers, domain, function, ansatz):
self.function = function
super().__init__(layers, domain, ansatz)
self.target = np.array(list(self.function(self.domain)))
self.target = 2 * (self.target - np.min(self.target)) / (
np.max(self.target) - np.min(self.target)) - 1
self.H = Hamiltonian(1, matrices._Z)
self.classical = classical_real_Weighted_2D
def test_trotter_hamiltonian_initialization_errors():
"""Test errors in initialization of ``TrotterHamiltonian``."""
# Wrong type of terms
with pytest.raises(TypeError):
ham = TrotterHamiltonian({(0, 1): "abc"})
# Wrong type of parts
with pytest.raises(TypeError):
ham = TrotterHamiltonian([(0, 1)])
# Wrong number of target qubits
with pytest.raises(ValueError):
ham = TrotterHamiltonian({(0, 1): TFIM(nqubits=3, numpy=True)})
# Same targets multiple times
h = TFIM(nqubits=2, numpy=True)
with pytest.raises(ValueError):
ham = TrotterHamiltonian({(0, 1): h}, {(0, 1): h})
# Different term matrix types
h2 = Hamiltonian(2, np.eye(4, dtype=np.float32), numpy=True)
with pytest.raises(TypeError):
ham = TrotterHamiltonian({(0, 1): h, (1, 2): h2})
# ``from_twoqubit_term`` initialization with nqubits < 0
with pytest.raises(ValueError):
ham = TrotterHamiltonian.from_twoqubit_term(-2, h)
# ``from_twoqubit_term`` initialization with more than 2 targets
h = TFIM(nqubits=3, numpy=True)
with pytest.raises(ValueError):
ham = TrotterHamiltonian.from_twoqubit_term(4, h)
def qpdf_hamiltonian(nqubits, z_qubit=0):
"""Precomputes Hamiltonian.
Args:
nqubits (int): number of qubits.
z_qubit (int): qubit where the Z measurement is applied, must be z_qubit < nqubits
Returns:
An Hamiltonian object.
"""
eye = matrices.I
if z_qubit == 0:
h = matrices.Z
for _ in range(nqubits - 1):
h = np.kron(eye, h)
elif z_qubit == nqubits - 1:
h = eye
for _ in range(nqubits - 2):
h = np.kron(eye, h)
h = np.kron(matrices.Z, h)
else:
h = eye
for _ in range(nqubits - 1):
if _ + 1 == z_qubit:
h = np.kron(matrices.Z, h)
else:
h = np.kron(eye, h)
return Hamiltonian(nqubits, h)
def from_symbolic(cls, symbolic_hamiltonian, symbol_map, ground_state=None):
"""Creates a ``TrotterHamiltonian`` from a symbolic Hamiltonian.
We refer to the :ref:`How to define custom Hamiltonians using symbols? <symbolicham-example>`
example for more details.
Args:
symbolic_hamiltonian (sympy.Expr): The full Hamiltonian written
with symbols.
symbol_map (dict): Dictionary that maps each symbol that appears in
the Hamiltonian to a pair of (target, matrix).
ground_state (Callable): Optional callable with no arguments that
returns the ground state of this ``TrotterHamiltonian``.
See :class:`qibo.base.hamiltonians.TrotterHamiltonian` for more
details.
Returns:
A :class:`qibo.base.hamiltonians.TrotterHamiltonian` object that
implements the given symbolic Hamiltonian.
"""
from qibo.hamiltonians import Hamiltonian
terms, constant = _SymbolicHamiltonian(
symbolic_hamiltonian, symbol_map).trotter_terms()
terms = {k: Hamiltonian(len(k), v, numpy=True)
for k, v in terms.items()}
return cls.from_dictionary(terms, ground_state=ground_state) + constant
def construct_terms(terms):
"""Helper method for `from_symbolic`.
Constructs the term dictionary by using the same
:class:`qibo.abstractions.hamiltonians.Hamiltonian` object for terms that
have equal matrix representation. This is done for efficiency during
the exponentiation of terms.
Args:
terms (dict): Dictionary that maps tuples of targets to the matrix
that acts on these on targets.
Returns:
terms (dict): Dictionary that maps tuples of targets to the
Hamiltonian term that acts on these on targets.
"""
from qibo.hamiltonians import Hamiltonian
unique_matrices = []
hterms = {}
for targets, matrix in terms.items():
flag = True
for m, h in unique_matrices:
if K.np.array_equal(matrix, m):
ham = h
flag = False
break
if flag:
ham = Hamiltonian(len(targets), matrix, numpy=True)
unique_matrices.append((matrix, ham))
hterms[targets] = ham
return hterms
def test_from_symbolic_application_hamiltonian():
"""Check ``from_symbolic`` for a specific four-qubit Hamiltonian."""
import sympy
z1, z2, z3, z4 = sympy.symbols("z1 z2 z3 z4")
symmap = {z: (i, matrices.Z) for i, z in enumerate([z1, z2, z3, z4])}
symham = (z1 * z2 - 0.5 * z1 * z3 + 2 * z2 * z3 + 0.35 * z2 +
0.25 * z3 * z4 + 0.5 * z3 + z4 - z1)
# Check that Trotter dense matrix agrees will full Hamiltonian matrix
fham = Hamiltonian.from_symbolic(symham, symmap)
tham = TrotterHamiltonian.from_symbolic(symham, symmap)
np.testing.assert_allclose(tham.dense.matrix, fham.matrix)
# Check that no one-qubit terms exist in the Trotter Hamiltonian
# (this means that merging was successful)
first_targets = set()
for part in tham.parts:
for targets, term in part.items():
first_targets.add(targets[0])
assert len(targets) == 2
assert term.nqubits == 2
assert first_targets == set(range(4))
# Check making an ``X`` Hamiltonian compatible with ``tham``
xham = X(nqubits=4, trotter=True)
cxham = tham.make_compatible(xham)
assert not tham.is_compatible(xham)
assert tham.is_compatible(cxham)
np.testing.assert_allclose(xham.dense.matrix, cxham.dense.matrix)
def __init__(self, layers, domain, ansatz, real, imag):
self.f_real = real
self.f_imag = imag
# self.function.name = lambda:self.real.name + '_' + self.imag.name
super().__init__(layers, domain, ansatz)
self.real = self.f_real(self.domain)
self.real = 2 * (self.real - np.min(self.real)) / (
np.max(self.real) - np.min(self.real)) - 1
self.imag = self.f_imag(self.domain)
self.imag = 2 * (self.imag - np.min(self.imag)) / (
np.max(self.imag) - np.min(self.imag)) - 1
self.target = self.real + 1j * self.imag
self.target /= np.max(np.abs(self.target))
self.H = [Hamiltonian(1, matrices._X), Hamiltonian(1, matrices._Y)]
self.classical = globals()[f"classical_complex_{self.ansatz}"]
def test_trotter_hamiltonian_make_compatible_onequbit_terms():
"""Check ``make_compatible`` when the two-qubit Hamiltonian has one-qubit terms."""
term1 = Hamiltonian(1, matrices.Z, numpy=True)
term2 = Hamiltonian(2, np.kron(matrices.Z, matrices.Z), numpy=True)
terms = {
(0, 1): term2,
(0, 2): -0.5 * term2,
(1, 2): 2 * term2,
(1, ): 0.35 * term1,
(2, 3): 0.25 * term2,
(2, ): 0.5 * term1,
(3, ): term1
}
tham = TrotterHamiltonian.from_dictionary(terms) + 1.5
xham = X(nqubits=4, trotter=True)
cxham = tham.make_compatible(xham)
assert not tham.is_compatible(xham)
assert tham.is_compatible(cxham)
np.testing.assert_allclose(xham.dense.matrix, cxham.dense.matrix)
def test_trotter_hamiltonian_three_qubit_term(backend):
"""Test creating ``TrotterHamiltonian`` with three qubit term."""
import qibo
from scipy.linalg import expm
original_backend = qibo.get_backend()
qibo.set_backend(backend)
m1 = utils.random_numpy_hermitian(3)
m2 = utils.random_numpy_hermitian(2)
m3 = utils.random_numpy_hermitian(1)
term1 = Hamiltonian(3, m1, numpy=True)
term2 = Hamiltonian(2, m2, numpy=True)
term3 = Hamiltonian(1, m3, numpy=True)
parts = [{(0, 1, 2): term1}, {(2, 3): term2, (1, ): term3}]
trotter_h = TrotterHamiltonian(*parts)
# Test that the `TrotterHamiltonian` dense matrix is correct
eye = np.eye(2, dtype=m1.dtype)
mm1 = np.kron(m1, eye)
mm2 = np.kron(np.kron(eye, eye), m2)
mm3 = np.kron(np.kron(eye, m3), np.kron(eye, eye))
target_h = Hamiltonian(4, mm1 + mm2 + mm3)
np.testing.assert_allclose(trotter_h.dense.matrix, target_h.matrix)
dt = 1e-2
initial_state = utils.random_numpy_state(4)
if backend == "custom":
with pytest.raises(NotImplementedError):
circuit = trotter_h.circuit(dt=dt)
else:
circuit = trotter_h.circuit(dt=dt)
final_state = circuit(np.copy(initial_state))
u = [expm(-0.5j * dt * m) for m in [mm1, mm2, mm3]]
target_state = u[2].dot(u[1].dot(u[0])).dot(initial_state)
target_state = u[0].dot(u[1].dot(u[2])).dot(target_state)
np.testing.assert_allclose(final_state, target_state)
qibo.set_backend(original_backend)
def test_hamiltonian_initialization():
"""Testing hamiltonian initialization errors."""
import tensorflow as tf
dtype = K.dtypes('DTYPECPX')
with pytest.raises(TypeError):
H = Hamiltonian(2, "test")
H1 = Hamiltonian(2, np.eye(4))
H1 = Hamiltonian(2, np.eye(4), numpy=True)
H1 = Hamiltonian(2, tf.eye(4, dtype=dtype))
H1 = Hamiltonian(2, tf.eye(4, dtype=dtype), numpy=True)
with pytest.raises(ValueError):
H1 = Hamiltonian(-2, np.eye(4))
with pytest.raises(RuntimeError):
H2 = Hamiltonian(np.eye(2), np.eye(4))
with pytest.raises(ValueError):
H3 = Hamiltonian(4, np.eye(10))
def test_x_hamiltonian_from_symbols(nqubits, trotter):
"""Check creating sum(X) Hamiltonian using sympy."""
import sympy
x_symbols = sympy.symbols(" ".join((f"X{i}" for i in range(nqubits))))
symham = -sum(x_symbols)
symmap = {x: (i, matrices.X) for i, x in enumerate(x_symbols)}
target_matrix = X(nqubits).matrix
if trotter:
trotter_ham = TrotterHamiltonian.from_symbolic(symham, symmap)
final_matrix = trotter_ham.dense.matrix
else:
full_ham = Hamiltonian.from_symbolic(symham, symmap)
final_matrix = full_ham.matrix
np.testing.assert_allclose(final_matrix, target_matrix)
def test_tfim_hamiltonian_from_symbols(nqubits, trotter):
"""Check creating TFIM Hamiltonian using sympy."""
import sympy
h = 0.5
z_symbols = sympy.symbols(" ".join((f"Z{i}" for i in range(nqubits))))
x_symbols = sympy.symbols(" ".join((f"X{i}" for i in range(nqubits))))
symham = sum(z_symbols[i] * z_symbols[i + 1] for i in range(nqubits - 1))
symham += z_symbols[0] * z_symbols[-1]
symham += h * sum(x_symbols)
symmap = {z: (i, matrices.Z) for i, z in enumerate(z_symbols)}
symmap.update({x: (i, matrices.X) for i, x in enumerate(x_symbols)})
target_matrix = TFIM(nqubits, h=h).matrix
if trotter:
trotter_ham = TrotterHamiltonian.from_symbolic(-symham, symmap)
final_matrix = trotter_ham.dense.matrix
else:
full_ham = Hamiltonian.from_symbolic(-symham, symmap)
final_matrix = full_ham.matrix
np.testing.assert_allclose(final_matrix, target_matrix)
def test_three_qubit_term_hamiltonian_from_symbols(trotter):
"""Check creating Hamiltonian with three-qubit interaction using sympy."""
import sympy
from qibo import matrices
x_symbols = sympy.symbols(" ".join((f"X{i}" for i in range(4))))
y_symbols = sympy.symbols(" ".join((f"Y{i}" for i in range(4))))
z_symbols = sympy.symbols(" ".join((f"Z{i}" for i in range(4))))
symmap = {x: (i, matrices.X) for i, x in enumerate(x_symbols)}
symmap.update({x: (i, matrices.Y) for i, x in enumerate(y_symbols)})
symmap.update({x: (i, matrices.Z) for i, x in enumerate(z_symbols)})
symham = x_symbols[0] * y_symbols[1] * z_symbols[2]
symham += 0.5 * y_symbols[0] * z_symbols[1] * x_symbols[3]
symham += z_symbols[0] * x_symbols[2]
symham += -3 * x_symbols[1] * y_symbols[3]
symham += y_symbols[2]
symham += 1.5 * z_symbols[1]
symham -= 2
target_matrix = np.kron(np.kron(matrices.X, matrices.Y),
np.kron(matrices.Z, matrices.I))
target_matrix += 0.5 * np.kron(np.kron(matrices.Y, matrices.Z),
np.kron(matrices.I, matrices.X))
target_matrix += np.kron(np.kron(matrices.Z, matrices.I),
np.kron(matrices.X, matrices.I))
target_matrix += -3 * np.kron(np.kron(matrices.I, matrices.X),
np.kron(matrices.I, matrices.Y))
target_matrix += np.kron(np.kron(matrices.I, matrices.I),
np.kron(matrices.Y, matrices.I))
target_matrix += 1.5 * np.kron(np.kron(matrices.I, matrices.Z),
np.kron(matrices.I, matrices.I))
target_matrix -= 2 * np.eye(2**4, dtype=target_matrix.dtype)
if trotter:
trotter_ham = TrotterHamiltonian.from_symbolic(symham, symmap)
final_matrix = trotter_ham.dense.matrix
else:
full_ham = Hamiltonian.from_symbolic(symham, symmap)
final_matrix = full_ham.matrix
np.testing.assert_allclose(final_matrix, target_matrix)
def test_from_symbolic_with_power(trotter):
"""Check ``from_symbolic`` when the expression contains powers."""
import sympy
z = sympy.symbols(" ".join((f"Z{i}" for i in range(3))))
symham = z[0]**2 - z[1]**2 + 3 * z[1] - 2 * z[0] * z[2] + +1
matrix = utils.random_numpy_hermitian(1)
symmap = {x: (i, matrix) for i, x in enumerate(z)}
if trotter:
ham = TrotterHamiltonian.from_symbolic(symham, symmap)
final_matrix = ham.dense.matrix
else:
ham = Hamiltonian.from_symbolic(symham, symmap)
final_matrix = ham.matrix
matrix2 = matrix.dot(matrix)
eye = np.eye(2, dtype=matrix.dtype)
target_matrix = np.kron(np.kron(matrix2, eye), eye)
target_matrix -= np.kron(np.kron(eye, matrix2), eye)
target_matrix += 3 * np.kron(np.kron(eye, matrix), eye)
target_matrix -= 2 * np.kron(np.kron(matrix, eye), matrix)
target_matrix += np.eye(8, dtype=matrix.dtype)
np.testing.assert_allclose(final_matrix, target_matrix)
def expectation(self, state, normalize=False):
return Hamiltonian.expectation(self, state, normalize)
class Approximant_real_2D(Approximant):
def __init__(self, layers, domain, function, ansatz):
self.function = function
super().__init__(layers, domain, ansatz)
self.target = np.array(list(self.function(self.domain)))
self.target = 2 * (self.target - np.min(self.target)) / (
np.max(self.target) - np.min(self.target)) - 1
self.H = Hamiltonian(1, matrices._Z)
self.classical = classical_real_Weighted_2D
def name_folder(self, quantum=True):
folder = self.ansatz + '/' + self.function.name + '/%s_layers' % (
self.layers)
if quantum:
folder = 'quantum/' + folder
else:
folder = 'classical/' + folder
folder = 'results/' + folder
import os
try:
l = os.listdir(folder)
l = [int(_) for _ in l]
l.sort()
trial = int(l[-1]) + 1
except:
trial = 0
os.makedirs(folder)
fold_name = folder + '/%s' % (trial)
os.makedirs(fold_name)
return fold_name, trial
def cf_one_point(self, x, f):
state = self.get_state(x)
o = self.H.expectation(state)
cf = (o - f)**2
return cf
def paint_representation_2D(self, name):
fig = plt.figure()
axs = fig.gca(projection='3d')
axs.plot_trisurf(self.domain[:, 0],
self.domain[:, 1],
self.target,
color='black',
label='Target Function',
alpha=0.5)
outcomes = np.zeros_like(self.target)
for j, x in enumerate(self.domain):
state = self.get_state(x)
outcomes[j] = self.H.expectation(state)
axs.scatter(self.domain[:, 0],
self.domain[:, 1],
outcomes,
color='C1',
label='Quantum ' + self.ansatz + ' model')
fig.savefig(name)
plt.close(fig)
def paint_representation_2D_classical(self, prediction, name):
fig = plt.figure()
axs = fig.gca(projection='3d')
prediction = np.array(list(prediction))
axs.plot_trisurf(self.domain[:, 0],
self.domain[:, 1],
self.target,
color='black',
label='Target Function',
alpha=0.5)
axs.scatter(self.domain[:, 0],
self.domain[:, 1],
prediction,
color='C0',
label='Classical ' + self.ansatz + ' model')
fig.savefig(name)
#plt.show()
plt.close(fig)
class Approximant_real(Approximant):
def __init__(self, layers, domain, ansatz, function):
self.function = function
super().__init__(layers, domain, ansatz)
self.target = self.function(self.domain)
self.target = 2 * (self.target - np.min(self.target)) / (
np.max(self.target) - np.min(self.target)) - 1
self.H = Hamiltonian(1, matrices._Z)
self.classical = globals()[f"classical_real_{self.ansatz}"]
def name_folder(self, quantum=True):
folder = self.ansatz + '/' + self.function.name + '/%s_layers' % (
self.layers)
if quantum:
folder = 'quantum/' + folder
else:
folder = 'classical/' + folder
folder = 'results/' + folder
import os
try:
l = os.listdir(folder)
l = [int(_) for _ in l]
l.sort()
trial = int(l[-1]) + 1
except:
trial = 0
os.makedirs(folder)
fold_name = folder + '/%s' % (trial)
os.makedirs(fold_name)
return fold_name, trial
def cf_one_point(self, x, f):
state = self.get_state(x)
o = self.H.expectation(state)
cf = (o - f)**2
return cf
def derivative_cf_one_point(self, x, f):
self.theta_with_x(x)
derivatives = np.zeros_like(self.params)
index = 0
ch_index = 0
for l in range(self.layers - 1):
cir_params_ = self.cir_params.copy()
delta = np.zeros_like(cir_params_)
delta[ch_index] = np.pi / 2
self.C.set_parameters(cir_params_ + delta)
state = self.C()
z1 = self.H.expectation(state)
self.C.set_parameters(cir_params_ - delta)
state = self.C()
z2 = self.H.expectation(state)
derivatives[index + 1] = 0.5 * ((z1**2 - z2**2) - 2 * f *
(z1 - z2))
derivatives[index] = x * derivatives[index + 1]
index += 2
ch_index += 1
cir_params_ = self.cir_params.copy()
delta = np.zeros_like(cir_params_)
delta[ch_index] = np.pi / 2
self.C.set_parameters(cir_params_ + delta)
state = self.C()
z1 = self.H.expectation(state)
self.C.set_parameters(cir_params_ - delta)
state = self.C()
z2 = self.H.expectation(state)
derivatives[index] = 0.5 * ((z1**2 - z2**2) - 2 * f * (z1 - z2))
index += 1
ch_index += 1
cir_params_ = self.cir_params.copy()
delta = np.zeros_like(cir_params_)
delta[ch_index] = np.pi / 2
self.C.set_parameters(cir_params_ + delta)
state = self.C()
z1 = self.H.expectation(state)
self.C.set_parameters(cir_params_ - delta)
state = self.C()
z2 = self.H.expectation(state)
derivatives[index + 1] = 0.5 * ((z1**2 - z2**2) - 2 * f * (z1 - z2))
derivatives[index] = x * derivatives[index + 1]
index += 2
ch_index += 1
return derivatives
def derivative_cf(self, params):
try:
params = params.flatten()
except:
pass
self.set_parameters(params)
derivatives = np.zeros_like(params)
for x, t in zip(self.domain, self.target):
derivatives += self.derivative_cf_one_point(x, t)
derivatives /= len(self.domain)
return derivatives
def paint_representation_1D(self, name):
fig, axs = plt.subplots()
axs.plot(self.domain,
self.target,
color='black',
label='Target Function')
outcomes = np.zeros_like(self.domain)
for j, x in enumerate(self.domain):
state = self.get_state(x)
outcomes[j] = self.H.expectation(state)
axs.plot(self.domain,
outcomes,
color='C1',
label='Quantum ' + self.ansatz + ' model')
axs.legend()
fig.savefig(name)
plt.close(fig)
def paint_representation_1D_classical(self, prediction, name):
fig, axs = plt.subplots()
axs.plot(self.domain,
self.target,
color='black',
label='Target Function')
axs.plot(self.domain,
prediction,
color='C0',
label='Classical ' + self.ansatz + ' model')
axs.legend()
fig.savefig(name)
plt.close(fig)
def paint_representation_2D(self):
from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
if self.num_functions == 1:
ax = fig.gca(projection='3d')
print('shape', self.target[0].shape)
ax.plot_trisurf(self.domain[:, 0],
self.domain[:, 1],
self.target[:, 0],
label='Target Function',
linewidth=0.2,
antialiased=True)
outcomes = np.zeros_like(self.domain)
for j, x in enumerate(self.domain):
C = self.circuit(x)
state = C.execute()
outcomes[j] = self.hamiltonian[0].expectation(state)
ax.plot_trisurf(self.domain[:, 0],
self.domain[:, 1],
outcomes[:, 0] + 0.1,
label='Approximation',
linewidth=0.2,
antialiased=True)
#ax.legend()
else:
for i in range(self.num_functions):
ax = fig.add_subplot(1, 1, i + 1, projection='3d')
ax.plot(self.domain,
self.functions[i](self.domain).flatten(),
color='black')
outcomes = np.zeros_like(self.domain)
for j, x in enumerate(self.domain):
C = self.circuit(x)
state = C.execute()
outcomes[j] = self.hamiltonian[i].expectation(state)
ax.scatter(self.domain[:, 0],
self.domain[:, 1],
outcomes,
color='C0',
label=self.measurements[i])
ax.legend()
fig.savefig(name)
plt.close(fig)
def paint_historical(self, name):
import matplotlib.pyplot as plt
fig, axs = plt.subplots(nrows=2)
axs[0].plot(np.arange(len(self.hist_chi)), self.hist_chi)
axs[0].set(yscale='log', ylabel=r'$\chi^2$')
hist_params = np.array(self.hist_params)
for i in range(len(self.params)):
axs[1].plot(np.arange(len(self.hist_chi)), hist_params[:, i],
'C%s' % i)
axs[1].set(ylabel='Parameter', xlabel='Function evaluation')
fig.suptitle('Historical behaviour', fontsize=16)
fig.savefig(name)