def __init__(self, molecule, rz_index):
"""See class fields for parameter definitions.
"""
super()
self.molecule = molecule
self.rz_index = rz_index
# Get circuit, cqp and slice_index.
pickle_file_name = "{}_circuits.pickle".format(molecule.lower())
pickle_file_path = os.path.join(HPO_DATA_PATH, pickle_file_name)
with open(pickle_file_path, "rb") as f:
rz_indices, circuit_list = pickle.load(f)
slice_index = rz_indices[self.rz_index]
circuit, connected_qubit_pairs = circuit_list[slice_index]
self.slice_index = slice_index
self.circuit = circuit
self.connected_qubit_pairs = connected_qubit_pairs
self.unitary = get_unitary(circuit)
self.file_name = "s{}".format(slice_index)
datadir_name = "uccsd_{}".format(molecule.lower())
self.data_path = os.path.join(TIME_DATA_PATH, datadir_name)
# TODO: We assume TIME_DATA_PATH exists.
if not os.path.exists(self.data_path):
os.mkdir(self.data_path)
# Get grape config.
# Set grape parameters.
num_qubits = self.circuit.width()
H0 = np.zeros((NUM_STATES ** num_qubits, NUM_STATES ** num_qubits))
Hops, Hnames = get_Hops_and_Hnames(num_qubits, NUM_STATES, self.connected_qubit_pairs)
states_concerned_list = get_full_states_concerned_list(num_qubits, NUM_STATES)
maxA = get_maxA(num_qubits, NUM_STATES, self.connected_qubit_pairs)
reg_coeffs = {}
self.grape_config = {
"H0": H0,
"Hops": Hops,
"Hnames": Hnames,
"states_concerned_list": states_concerned_list,
"reg_coeffs": reg_coeffs,
"maxA": maxA,
}
self.grape_config.update(GRAPE_TASK_CONFIG)
# Get hyperparameters by choosing the configuration with the lowest loss.
best_data = {"loss": 1}
hpo_file_name = "{}.json".format(self.file_name)
hpo_file_path = os.path.join(HPO_DATA_PATH, datadir_name, hpo_file_name)
with open(hpo_file_path) as f:
line = f.readline()
while line:
data = json.loads(line)
if data["loss"] < best_data["loss"]:
best_data = data
line = f.readline()
#ENDWITH
self.lr = best_data["lr"]
self.decay = best_data["decay"]
def __init__(self, molecule, slice_index, circuit, connected_qubit_pairs):
"""See corresponding class field declarations above for other arguments.
"""
super()
self.molecule = molecule
self.slice_index = slice_index
self.circuit = circuit
self.connected_qubit_pairs = connected_qubit_pairs
self.unitary = get_unitary(self.circuit)
if self.circuit.depth() > REDUCED_CIRCUIT_DEPTH_CUTOFF:
self.pulse_time = get_max_pulse_time(
self.circuit) * PULSE_TIME_MULTIPLIER
else:
self.pulse_time = get_max_pulse_time(
self.circuit) * REDUCED_PULSE_TIME_MULTIPLIER
self.file_name = "s{}".format(self.slice_index)
self.data_path = os.path.join(BASE_DATA_PATH,
"uccsd_{}".format(molecule.lower()))
# TODO: We assume BASE_DATA_PATH exists.
if not os.path.exists(self.data_path):
os.mkdir(self.data_path)
# Set grape parameters.
num_qubits = self.circuit.width()
H0 = np.zeros((NUM_STATES**num_qubits, NUM_STATES**num_qubits))
Hops, Hnames = get_Hops_and_Hnames(num_qubits, NUM_STATES,
self.connected_qubit_pairs)
states_concerned_list = get_full_states_concerned_list(
num_qubits, NUM_STATES)
maxA = get_maxA(num_qubits, NUM_STATES, self.connected_qubit_pairs)
reg_coeffs = {}
self.grape_config = {
"H0": H0,
"Hops": Hops,
"Hnames": Hnames,
"states_concerned_list": states_concerned_list,
"reg_coeffs": reg_coeffs,
"maxA": maxA,
}
self.grape_config.update(GRAPE_TASK_CONFIG)
NPS = 1 / SPN
### GATE DATA ###
# fqc/experiments/Gate_Times.ipynb
GATE_TIMES = {'h': 1.4, 'cx': 3.8, 'rz': 0.4, 'rx': 2.5, 'x': 2.5, 'swap': 7.4}
_RZ = GATE_TIMES['rz']
### GRAPE CONSTANTS ###
REG_COEFFS = {}
QUBIT2_CQP = get_nearest_neighbor_coupling_list(1, 2, directed=False)
QUBIT2_H0 = np.zeros((NUM_STATES**2, NUM_STATES**2))
QUBIT2_HOPS, QUBIT2_HNAMES = get_Hops_and_Hnames(2, NUM_STATES, QUBIT2_CQP)
QUBIT2_SCL = get_full_states_concerned_list(2, NUM_STATES)
QUBIT2_MPA = get_maxA(2, NUM_STATES, QUBIT2_CQP)
GRAPE_QUBIT2_CONFIG = {
"H0": QUBIT2_H0,
"Hops": QUBIT2_HOPS,
"Hnames": QUBIT2_HNAMES,
"states_concerned_list": QUBIT2_SCL,
"reg_coeffs": REG_COEFFS,
"maxA": QUBIT2_MPA,
}
QUBIT4_CQP = get_nearest_neighbor_coupling_list(2, 2, directed=False)
QUBIT4_H0 = np.zeros((NUM_STATES**4, NUM_STATES**4))
QUBIT4_HOPS, QUBIT4_HNAMES = get_Hops_and_Hnames(4, NUM_STATES, QUBIT4_CQP)
QUBIT4_SCL = get_full_states_concerned_list(4, NUM_STATES)
QUBIT4_MPA = get_maxA(4, NUM_STATES, QUBIT4_CQP)
from quantum_optimal_control.core import hamiltonian
d = 2 # this is the number of energy levels to consider (i.e. d-level qudits)
max_iterations = 6000
decay = max_iterations / 2
convergence = {'rate':0.01, 'max_iterations': max_iterations,
'conv_target':1e-3, 'learning_rate_decay':decay, 'min_grad': 1e-12, 'update_step': 20}
reg_coeffs = {}
N = 4
connected_qubit_pairs = util.get_nearest_neighbor_coupling_list(2, 2, directed=False)
H0 = np.zeros((d ** N, d ** N))
Hops, Hnames = hamiltonian.get_Hops_and_Hnames(N, d, connected_qubit_pairs)
states_concerned_list = hamiltonian.get_full_states_concerned_list(N, d)
maxA = hamiltonian.get_maxA(N, d, connected_qubit_pairs)
circuit = uccsd.get_uccsd_circuit('LiH')
slices = uccsd.get_uccsd_slices(circuit, granularity=1)
slices = [slice for slice in slices if not slice.parameterized]
def binary_search_for_shortest_pulse_time(min_time, max_time, tolerance=1):
"""Search between [min_time, max_time] up to 1ns tolerance. Assumes 20 steps per ns."""
min_steps, max_steps = min_time * 20, max_time * 20
while min_steps + 20 * tolerance < max_steps: # just estimate to +- 1ns
# Time backoff for binary search.
BACKOFF = 1.2
# Binary search nanosecond granularity.
BNS_GRANULARITY = 10
# Grape args.
DATA_PATH = ""
# Define hardware specific parameters.
num_qubits = 4
num_states = 2
connected_qubit_pairs = get_nearest_neighbor_coupling_list(2, 2, directed=False)
H0 = np.zeros((num_states ** num_qubits, num_states ** num_qubits))
Hops, Hnames = get_Hops_and_Hnames(num_qubits, num_states,
connected_qubit_pairs)
states_concerned_list = get_full_states_concerned_list(num_qubits, num_states)
maxA = get_maxA(num_qubits, num_states, connected_qubit_pairs)
# Define convergence parameters and penalties.
convergence = {'rate': 2e-2, 'conv_target': 1e-3,
'max_iterations': 1e3, 'learning_rate_decay': 1e3}
reg_coeffs = {}
use_gpu = False
sparse_H = False
show_plots = False
method = 'ADAM'
# Steps per nanosecond and nanoseconds per step
spn = 20.0
nps = 1 / spn
# Get slices to perform qoc on. The initial angle of each RZ
ANGLE_CONSTANT = 180
TIME_MULTIPLIERS = [0.25, 0.5, 0.75]
TIME_MULTIPLIER_CONSTANT = 0.5
JOBS_PER_SLICE = 5
BROADWELL_CORE_COUNT = 14
# Grape constants.
NUM_QUBITS = 4
NUM_STATES = 2
CONNECTED_QUBIT_PAIRS = get_nearest_neighbor_coupling_list(2,
2,
directed=False)
H0 = np.zeros((NUM_STATES**NUM_QUBITS, NUM_STATES**NUM_QUBITS))
Hops, Hnames = get_Hops_and_Hnames(NUM_QUBITS, NUM_STATES,
CONNECTED_QUBIT_PAIRS)
STATES_CONCERNED_LIST = get_full_states_concerned_list(NUM_QUBITS, NUM_STATES)
MAX_AMPLITUDE = get_maxA(NUM_QUBITS, NUM_STATES, CONNECTED_QUBIT_PAIRS)
METHOD = 'ADAM'
MAX_GRAPE_ITERATIONS = 1e3
DECAY = 1e3
REG_COEFFS = {}
USE_GPU = False
SPARSE_H = False
SHOW_PLOTS = False
# Wave steps per nanosecond of pulse time.
SPN = 20
# Get slices and information.
SLICE_GRANULARITY = 2
UCCSD_LIH_FULL_CIRCUIT = optimize_circuit(
get_uccsd_circuit('LiH', UCCSD_LIH_THETA), CONNECTED_QUBIT_PAIRS)