def _make_random_change(self, compiled_sequence: UnitarySequence) -> None:
'''
Chooses and implements a random change to the specified
unitary sequence.
'''
count_append = np.count_nonzero([
CompilerAction.is_append(action) for action in list(CompilerAction)
])
count_non_append = len(CompilerAction) - count_append
p_append = self.append_probability / count_append
p_non_append = (1 - self.append_probability) / count_non_append
action = np.random.choice(
list(CompilerAction),
1,
p=[
p_append if CompilerAction.is_append(action) else p_non_append
for action in list(CompilerAction)
])
new_sequence_entry = None
if CompilerAction.is_append(action):
new_sequence_entry = Compiler.create_random_sequence_entry(
self.dimension, self.unitary_primitives,
self.unitary_primitive_probabilities)
if action == CompilerAction.AppendFirst:
compiled_sequence.append_first(new_sequence_entry)
elif action == CompilerAction.AppendLast:
compiled_sequence.append_last(new_sequence_entry)
elif action == CompilerAction.RemoveFirst:
compiled_sequence.remove_first()
elif action == CompilerAction.RemoveLast:
compiled_sequence.remove_last()
def test_append_and_remove(self) -> None:
dimension = 2
identity = Unitary.identity(dimension)
sequence = UnitarySequence(dimension)
assert sequence.get_length() == 0
assert sequence.product().close_to(identity)
sequence.append_first(
UnitarySequenceEntry(UnitaryDefinitions.sigmax(), [0]))
assert sequence.get_length() == 1
assert sequence.product().close_to(UnitaryDefinitions.sigmax())
sequence.append_last(
UnitarySequenceEntry(UnitaryDefinitions.sigmay(), [0]))
assert sequence.get_length() == 2
assert sequence.product().close_to(UnitaryDefinitions.sigmaz())
sequence.append_first(
UnitarySequenceEntry(UnitaryDefinitions.sigmaz(), [0]))
assert sequence.get_length() == 3
assert sequence.product().close_to(identity)
sequence.remove_last()
assert sequence.get_length() == 2
assert sequence.product().close_to(UnitaryDefinitions.sigmay())
sequence.remove_first()
assert sequence.get_length() == 1
assert sequence.product().close_to(UnitaryDefinitions.sigmax())
sequence.remove_first()
assert sequence.get_length() == 0
assert sequence.product().close_to(identity)
def compile_rav_sequence(
self,
time: float,
max_t_step: float,
threshold: float,
allow_simultaneous_terms: bool = False
) -> CompilerResult:
'''
Returns a randomized analog verification (RAV) sequence as
per Shaffer et al., arXiv:2003.04500 (2020). The sequence of
unitaries is built from terms of this Hamiltonian by first
generating a random sequence and then using STOQ to compile
the inverse such that the full sequence approximately
implements the identity operation.
:param time: The total time to evolve the system in the
initial randomly-generated sequence.
:type time: float
:param max_t_step: The maximum time to use for a single Hamiltonian
term at each step of the sequence.
:type max_t_step: float
:param threshold: The overlap with the target unitary at which to
stop the STOQ compilation, defaults to None. A value of 1.0
implies an exact compilation.
:type threshold: float
:param allow_simultaneous_terms: Whether to allow multiple
Hamiltonian terms to be executed simultaneously in the resulting
sequence, defaults to False.
:type allow_simultaneous_terms: bool, optional
:return: A sequence of unitaries implementing RAV.
:rtype: CompilerResult
'''
# Generate a random sequence, mostly forward in time
forward_probability = 0.8
unitary_primitives = self._get_unitary_primitives(
max_t_step, allow_simultaneous_terms)
apply_to = list(range(self.get_qubit_count()))
random_sequence = UnitarySequence(self.get_dimension())
total_time = 0.0
while total_time < time:
t_step = (max_t_step * np.random.random_sample()) * (
1 if np.random.random_sample() < forward_probability else -1)
u_step = np.random.choice(
unitary_primitives).get_unitary().as_unitary([t_step])
random_sequence.append_last(UnitarySequenceEntry(u_step, apply_to))
total_time += np.abs(t_step)
# Calculate the product of this sequence and invert it
target_unitary = random_sequence.product().inverse()
# Call _compile_stoq_sequence_from_unitary to compile a new sequence
# implementing the inverse
result = self._compile_stoq_sequence_for_target_unitary(
target_unitary, max_t_step, threshold, allow_simultaneous_terms)
# Return the CompilerResult with the combined sequence
result.compiled_sequence = UnitarySequence.combine(
random_sequence, result.compiled_sequence)
return result
def _make_random_change_layered(
self, compiled_sequence: UnitarySequence,
unitary_primitive_counts: Dict[UnitaryPrimitive, int]) -> None:
'''
Chooses and implements a random change to the specified
layered unitary sequence.
'''
count_append = np.count_nonzero([
CompilerAction.is_append(action) for action in list(CompilerAction)
])
count_non_append = len(CompilerAction) - count_append
p_append = self.append_probability / count_append
p_non_append = (1 - self.append_probability) / count_non_append
action = np.random.choice(
list(CompilerAction),
1,
p=[
p_append if CompilerAction.is_append(action) else p_non_append
for action in list(CompilerAction)
])
layer_length = sum(unitary_primitive_counts.values())
new_layer = None
if CompilerAction.is_append(action):
new_layer = Compiler.create_random_layer(self.dimension,
unitary_primitive_counts)
# only save the undo state for the first modification to the sequence
# this is so that a future undo() call will reverse the entire action
if action == CompilerAction.AppendFirst:
for i, sequence_entry in enumerate(new_layer):
compiled_sequence.append_first(sequence_entry,
save_undo=(i == 0))
elif action == CompilerAction.AppendLast:
for i, sequence_entry in enumerate(new_layer):
compiled_sequence.append_last(sequence_entry,
save_undo=(i == 0))
elif action == CompilerAction.RemoveFirst:
for i in range(layer_length):
compiled_sequence.remove_first(save_undo=(i == 0))
elif action == CompilerAction.RemoveLast:
for i in range(layer_length):
compiled_sequence.remove_last(save_undo=(i == 0))
def generate_layered_rav_sequence(
dimension: int,
unitary_primitive_counts: Dict[UnitaryPrimitive, int],
layer_count: int,
threshold: float,
stoq_append_probability: float = 0.5,
max_step_count: int = 10000
) -> CompilerResult:
'''
Implements layered randomized analog verification (RAV).
:param dimension: [description]
:type dimension: int
:param unitary_primitive_counts: Specifies the fixed set of unitary
primitives to be contained in each layer of the compilation.
Each key is the unitary primitive to be included, and each
value is the count of that unitary primitive per layer.
:type unitary_primitive_counts: Dict[UnitaryPrimitive, int]
:param layer_count: The number of layers to create in the initial
randomly-generated sequence.
:type layer_count: int
:param threshold: The overlap with the target unitary at which to
stop compilation, defaults to None. A value of 1.0 implies an exact
compilation. If None, a threshold of 1.0 is used.
:type threshold: float
:param stoq_append_probability: Probability of appending a new gate
at each step in the compilation, defaults to 0.5.
:type stoq_append_probability: float, optional
:param max_step_count: Maximum number of steps to perform while
attempting to perform the approximate compilation,
defaults to 10000. Compilation of the inversion sequence will
terminate after this number of steps regardless of whether the
threshold has been reached.
:type max_step_count: int, optional
:return: The result of the compilation, including the layered
RAV sequence.
:rtype: CompilerResult
'''
assert isinstance(unitary_primitive_counts, dict)
assert np.all([
isinstance(primitive, UnitaryPrimitive)
for primitive in unitary_primitive_counts.keys()])
assert np.all([
primitive.get_unitary().get_dimension() <= dimension
for primitive in unitary_primitive_counts.keys()])
assert np.all([
isinstance(count, int)
for count in unitary_primitive_counts.values()])
assert layer_count >= 0
assert threshold >= 0.0 and threshold <= 1.0
# Generate a random sequence of the desired number of layers
# Total sequence length will therefore be
# sum(unitary_primitive_counts.values()) * layer_count
random_sequence = UnitarySequence(dimension)
for _ in range(layer_count):
layer = Compiler.create_random_layer(
dimension, unitary_primitive_counts)
for sequence_entry in layer:
random_sequence.append_last(sequence_entry)
# Skip inverse compilation if threshold or max_step_count is zero
if threshold == 0.0 or max_step_count == 0:
return CompilerResult(random_sequence, [], 0.0)
# Calculate the product of this sequence and invert it
target_unitary = random_sequence.product().inverse()
# Use Compiler to compile a new sequence implementing the inverse
compiler = Compiler(
dimension,
list(unitary_primitive_counts.keys()),
append_probability=stoq_append_probability)
result = compiler.compile_layered(
target_unitary, unitary_primitive_counts,
threshold, max_step_count)
# Return the CompilerResult with the combined sequence
result.compiled_sequence = UnitarySequence.combine(
random_sequence, result.compiled_sequence)
return result
def generate_rav_sequence(
dimension: int,
unitary_primitives: List[UnitaryPrimitive],
sequence_length: int,
threshold: float,
stoq_append_probability: float = 0.5,
unitary_primitive_probabilities: Optional[List[float]] = None,
max_step_count: int = 10000
) -> CompilerResult:
'''
Implements randomized analog verification (RAV) as per
Shaffer et al., arXiv:2003.04500 (2020).
:param dimension: The dimension of the state space. For an n-qubit
system, dimension should be set to 2**n.
:type dimension: int
:param unitary_primitives: The unitary primitives to be used for
the compilation.
:type unitary_primitives: List[UnitaryPrimitive]
:param sequence_length: The length of the initial randomly-generated
sequence.
:type sequence_length: int
:param threshold: The overlap with the target unitary at which to
stop compilation, defaults to None. A value of 1.0 implies an exact
compilation. If None, a threshold of 1.0 is used.
:type threshold: float
:param stoq_append_probability: Probability of appending a new gate
at each step in the compilation, defaults to 0.5.
:type stoq_append_probability: float, optional
:param unitary_primitive_probabilities: The probability for STOQ to
choose each of the primitives specified in unitary_primitives when
proposing new gates at each step of the compilation process,
defaults to None. If not specified, each unitary primitive is
chosen with uniform probability.
:type unitary_primitive_probabilities: Optional[List[float]], optional
:param max_step_count: Maximum number of steps to perform while
attempting to perform the approximate compilation,
defaults to 10000. Compilation of the inversion sequence will
terminate after this number of steps regardless of whether the
threshold has been reached.
:type max_step_count: int, optional
:return: The result of the compilation, including the RAV sequence.
:rtype: CompilerResult
'''
assert (isinstance(unitary_primitives, list)
or isinstance(unitary_primitives, np.ndarray))
assert np.all([
isinstance(primitive, UnitaryPrimitive)
for primitive in unitary_primitives])
assert np.all([
primitive.get_unitary().get_dimension() <= dimension
for primitive in unitary_primitives])
assert sequence_length >= 0
assert threshold >= 0.0 and threshold <= 1.0
# Generate a random sequence of the desired length
random_sequence = UnitarySequence(dimension)
for _ in range(sequence_length):
new_sequence_entry = Compiler.create_random_sequence_entry(
dimension, unitary_primitives, unitary_primitive_probabilities)
random_sequence.append_last(new_sequence_entry)
# Skip inverse compilation if threshold or max_step_count is zero
if threshold == 0.0 or max_step_count == 0:
return CompilerResult(random_sequence, [], 0.0)
# Calculate the product of this sequence and invert it
target_unitary = random_sequence.product().inverse()
# Use Compiler to compile a new sequence implementing the inverse
compiler = Compiler(
dimension,
unitary_primitives,
unitary_primitive_probabilities,
append_probability=stoq_append_probability)
result = compiler.compile(
target_unitary, threshold, max_step_count)
# Return the CompilerResult with the combined sequence
result.compiled_sequence = UnitarySequence.combine(
random_sequence, result.compiled_sequence)
return result