Source code for mitiq.cdr.clifford_training_data

# Copyright (C) 2021 Unitary Fund
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <https://www.gnu.org/licenses/>.

"""Functions for mapping circuits to (near) Clifford circuits."""
from typing import List, Optional, Sequence, Union, Any, cast

import numpy as np

import cirq
from cirq.circuits import Circuit

from mitiq.interface import (
    accept_any_qprogram_as_input,
    atomic_one_to_many_converter,
)

# Z gates with these angles/exponents are Clifford gates.
_CLIFFORD_EXPONENTS = np.array([0.0, 0.5, 1.0, 1.5])
_CLIFFORD_ANGLES = [exponent * np.pi for exponent in _CLIFFORD_EXPONENTS]


[docs]@atomic_one_to_many_converter def generate_training_circuits( circuit: Circuit, num_training_circuits: int, fraction_non_clifford: float, method_select: str = "uniform", method_replace: str = "closest", random_state: Optional[Union[int, np.random.RandomState]] = None, **kwargs: Any, ) -> List[Circuit]: r"""Returns a list of (near) Clifford circuits obtained by replacing (some) non-Clifford gates in the input circuit by Clifford gates. The way in which non-Clifford gates are selected to be replaced is determined by ``method_select`` and ``method_replace``. In the Clifford Data Regression (CDR) method [Czarnik2020]_, data generated from these circuits is used as a training set to learn the effect of noise. Args: circuit: A circuit of interest assumed to be compiled into the gate set {Rz, sqrt(X), CNOT}, or such that all the non-Clifford gates are contained in the Rz rotations. num_training_circuits: Number of circuits in the returned training set. fraction_non_clifford: The (approximate) fraction of non-Clifford gates in each returned circuit. method_select: Method by which non-Clifford gates are selected to be replaced by Clifford gates. Options are 'uniform' or 'gaussian'. method_replace: Method by which selected non-Clifford gates are replaced by Clifford gates. Options are 'uniform', 'gaussian' or 'closest'. random_state: Seed for sampling. kwargs: Available keyword arguments are: - sigma_select (float): Width of the Gaussian distribution used for ``method_select='gaussian'``. - sigma_replace (float): Width of the Gaussian distribution used for ``method_replace='gaussian'``. .. [Czarnik2020] : Piotr Czarnik, Andrew Arramsmith, Patrick Coles, Lukasz Cincio, "Error mitigation with Clifford quantum circuit data," (https://arxiv.org/abs/2005.10189). """ if random_state is None or isinstance(random_state, int): random_state = np.random.RandomState(random_state) # Find the non-Clifford operations in the circuit. operations = np.array(list(circuit.all_operations())) non_clifford_indices_and_ops = np.array( [ [i, op] for i, op in enumerate(operations) if not cirq.has_stabilizer_effect(op) ] ) if len(non_clifford_indices_and_ops) == 0: raise ValueError("Circuit is already Clifford.") non_clifford_indices = np.int32(non_clifford_indices_and_ops[:, 0]) non_clifford_ops = non_clifford_indices_and_ops[:, 1] # Replace (some of) the non-Clifford operations. near_clifford_circuits = [] for _ in range(num_training_circuits): new_ops = _map_to_near_clifford( non_clifford_ops, fraction_non_clifford, method_select, method_replace, random_state, **kwargs, ) operations[non_clifford_indices] = new_ops near_clifford_circuits.append(Circuit(operations)) return near_clifford_circuits
[docs]@accept_any_qprogram_as_input def is_clifford(circuit: Circuit) -> bool: """Returns True if the input argument is Clifford, else False. Args: circuit: A single operation, list of operations, or circuit. """ return all( cirq.has_stabilizer_effect(op) for op in circuit.all_operations() )
[docs]@accept_any_qprogram_as_input def count_non_cliffords(circuit: Circuit) -> int: """Returns the number of non-Clifford operations in the circuit. Assumes the circuit consists of only Rz, Rx, and CNOT operations. Args: circuit: Circuit to count the number of non-Clifford operations in. """ return sum( not cirq.has_stabilizer_effect(op) for op in circuit.all_operations() )
def _map_to_near_clifford( non_clifford_ops: Sequence[cirq.ops.Operation], fraction_non_clifford: float, method_select: str = "uniform", method_replace: str = "closest", random_state: Optional[np.random.RandomState] = None, **kwargs: Any, ) -> Sequence[cirq.ops.Operation]: """Returns the list of non-Clifford operations with some of these replaced by Clifford operations. Args: non_clifford_ops: A sequence of non-Clifford operations. fraction_non_clifford: The (approximate) fraction of non-Clifford operations in the returned list. method_select: The way in which the non-Clifford gates are selected to be replaced by Clifford gates. Options are 'uniform' or 'gaussian'. method_replace: The way in which selected non-Clifford gates are replaced by Clifford gates. Options are 'uniform', 'gaussian' or 'closest'. random_state: Seed for sampling. kwargs: Additional options for selection / replacement methods. sigma_select (float): Width of the Gaussian distribution used for ``method_select='gaussian'``. sigma_replace (float): Width of the Gaussian distribution used for ``method_replace='gaussian'``. """ sigma_select: float = kwargs.get("sigma_select", 0.5) sigma_replace: float = kwargs.get("sigma_replace", 0.5) # Select (indices of) operations to replace. indices_of_selected_ops = _select( non_clifford_ops, fraction_non_clifford, method_select, sigma_select, random_state, ) # Replace selected operations. clifford_ops: Sequence[cirq.ops.Operation] = _replace( [non_clifford_ops[i] for i in indices_of_selected_ops], method_replace, sigma_replace, random_state, ) # Return sequence of (near) Clifford operations. return [ cast(List[cirq.ops.Operation], clifford_ops).pop(0) if i in indices_of_selected_ops else op for (i, op) in enumerate(non_clifford_ops) ] def _select( non_clifford_ops: Sequence[cirq.ops.Operation], fraction_non_clifford: float, method: str = "uniform", sigma: Optional[float] = 1.0, random_state: Optional[np.random.RandomState] = None, ) -> List[int]: """Returns indices of non-Clifford operations selected (to be replaced) according to some method. Args: non_clifford_ops: Sequence of non-Clifford operations. fraction_non_clifford: fraction of non-Clifford gates to change. method: {'uniform', 'gaussian'} method to use to select Clifford gates to replace. sigma: width of probability distribution used in selection of non-Clifford gates to replace, only has effect if method_select = 'gaussian'. random_state: Random state for sampling. """ if random_state is None: random_state = np.random num_non_cliff = len(non_clifford_ops) num_to_replace = int(round(fraction_non_clifford * num_non_cliff)) # Get the distribution for how to select operations. if method == "uniform": distribution = 1.0 / num_non_cliff * np.ones(shape=(num_non_cliff,)) elif method == "gaussian": non_clifford_angles = np.array( [ op.gate.exponent * np.pi # type: ignore for op in non_clifford_ops ] ) probabilities = _angle_to_proximity(non_clifford_angles, sigma) distribution = [k / sum(probabilities) for k in probabilities] else: raise ValueError( f"Arg `method_select` must be 'uniform' or 'gaussian' but was " f"{method}." ) # Select (indices of) non-Clifford operations to replace. selected_indices = cast(np.random.RandomState, random_state).choice( range(num_non_cliff), num_non_cliff - num_to_replace, replace=False, p=distribution, ) return [int(i) for i in sorted(selected_indices)] def _replace( non_clifford_ops: Sequence[cirq.ops.Operation], method: str = "uniform", sigma: float = 1.0, random_state: Optional[np.random.RandomState] = None, ) -> Sequence[cirq.ops.Operation]: """Function that takes the non-Clifford angles and replacement and selection specifications, returning the projected angles according to a specific method. Args: non_clifford_ops: array of non-Clifford angles. method: {'uniform', 'gaussian', 'closest'} method to use to replace selected non-Clifford gates. sigma: width of probability distribution used in replacement of selected non-Clifford gates, only has effect if method_replace = 'gaussian'. random_state: Seed for sampling. Returns: rz_non_clifford_replaced: the selected non-Clifford gates replaced by a Clifford according to some method. Raises: Exception: If argument 'method_replace' is not either 'closest', 'uniform' or 'gaussian'. """ if random_state is None: random_state = np.random # TODO: Update these functions to act on operations instead of angles. non_clifford_angles = np.array( [op.gate.exponent * np.pi for op in non_clifford_ops] # type: ignore ) if method == "closest": clifford_angles = _closest_clifford(non_clifford_angles) elif method == "uniform": clifford_angles = _random_clifford( len(non_clifford_angles), random_state ) elif method == "gaussian": clifford_angles = _probabilistic_angle_to_clifford( non_clifford_angles, sigma, random_state ) else: raise ValueError( f"Arg `method_replace` must be 'closest', 'uniform', or 'gaussian'" f" but was {method}." ) # TODO: Write function to replace the angles in a list of operations? return [ cirq.ops.rz(a).on(*q) for (a, q) in zip( clifford_angles, [op.qubits for op in non_clifford_ops], ) ] def _random_clifford( num_angles: int, random_state: np.random.RandomState ) -> np.ndarray: """Returns an array of Clifford angles chosen uniformly at random. Args: num_angles: Number of Clifford angles to return in array. random_state: Random state for sampling. """ return np.array( [random_state.choice(_CLIFFORD_ANGLES) for _ in range(num_angles)] ) @np.vectorize def _closest_clifford(angles: np.ndarray) -> float: """Returns the nearest Clifford angles to the input angles. Args: non_Clifford_ops: Non-Clifford opperations. """ ang_scaled = angles / (np.pi / 2) # if just one min value, return the corresponding nearest cliff. if ( abs((ang_scaled / 0.5) - 1) > 10 ** (-6) and abs((ang_scaled / 0.5) - 3) > 10 ** (-6) and (abs((ang_scaled / 0.5) - 5) > 10 ** (-6)) ): index = int(np.round(ang_scaled)) % 4 return _CLIFFORD_ANGLES[index] # If equidistant between two Clifford angles, randomly choose one. else: index_list = [ang_scaled - 0.5, ang_scaled + 0.5] index = int(np.random.choice(index_list)) return _CLIFFORD_ANGLES[index] @np.vectorize def _is_clifford_angle(angles: np.ndarray, tol: float = 10 ** -5,) -> bool: """Function to check if a given angle is Clifford. Args: angles: rotation angle in the Rz gate. """ angles = angles % (2 * np.pi) closest_clifford_angle = _closest_clifford(angles) if abs(closest_clifford_angle - angles) < tol: return True return False def _angle_to_proximities(angle: np.ndarray, sigma: float) -> List[float]: """Returns probability distribution based on distance from angles to Clifford gates. Args: angle: angle to form probability distribution. Returns: discrete value of probability distribution calculated from exp(-(diff/sigma)^2) where diff is the distance from each angle and the Clifford gates. """ s_matrix = cirq.unitary(cirq.S) rz_matrix = cirq.unitary(cirq.rz(angle % (2 * np.pi))) # TODO: Update loop / if. dists = [] for exponent in range(4): if exponent == 0: exponent = 4 diff = np.linalg.norm(rz_matrix - s_matrix ** exponent) dists.append(np.exp(-((diff / sigma) ** 2))) return dists @np.vectorize def _angle_to_proximity(angle: np.ndarray, sigma: float) -> float: """Returns probability distribution based on distance from angles to Clifford gates. Args: angle: angle to form probability distribution. Returns: discrete value of probability distribution calculated from exp(-(dist/sigma)^2) where dist = sum(dists) is the sum of distances from each Clifford gate. """ dists = _angle_to_proximities(angle, sigma) return np.max(dists) @np.vectorize def _probabilistic_angle_to_clifford( angles: np.ndarray, sigma: float, random_state: np.random.RandomState, ) -> float: """Returns a Clifford angle sampled from the distribution prob = exp(-(dist/sigma)^2) where dist is the Frobenius norm from the 4 clifford angles and the gate of interest. Args: angles: Non-Clifford angles. sigma: Width of probability distribution. """ dists = _angle_to_proximities(angles, sigma) cliff_ang = random_state.choice( _CLIFFORD_ANGLES, 1, replace=False, p=np.array(dists) / np.sum(dists) ) return cliff_ang