Files
final-qibotn/src/qibotn/backends/quimb.py
Mattia Robbiano 9853e86deb fixing grad nan
2025-09-20 15:15:40 +02:00

465 lines
17 KiB
Python

import re
import warnings
from collections import Counter, defaultdict
import numpy as np
import jax
import jax.numpy as jnp
import quimb as qu
import quimb.tensor as qtn
from qibo.backends import NumpyBackend
from qibo.config import raise_error
from qibo.result import QuantumState
from qibotn.backends.abstract import QibotnBackend
from qibotn.result import TensorNetworkResult
from qibo.gates.abstract import ParametrizedGate
GATE_MAP = {
"h": "H",
"x": "X",
"y": "Y",
"z": "Z",
"s": "S",
"sdg": "SDG",
"t": "T",
"tdg": "TDG",
"sx": "SX",
"sxdg": "SXDG",
"rx": "RX",
"ry": "RY",
"rz": "RZ",
"u1": "U1",
"u2": "U2",
"u3": "U3",
"cx": "CX",
"cnot": "CNOT",
"cy": "CY",
"cz": "CZ",
"iswap": "ISWAP",
"swap": "SWAP",
"ccx": "CCX",
"toffoli": "CCX",
"ccz": "CCZ",
"cswap": "CSWAP",
"fredkin": "CSWAP",
"crx": "CRX",
"cry": "CRY",
"crz": "CRZ",
"fsim": "FSIM",
"rxx": "RXX",
"ryy": "RYY",
"rzz": "RZZ",
"m": None, # measurement, skip
}
class QuimbBackend(QibotnBackend, NumpyBackend):
def __init__(self):
super().__init__()
self.name = "qibotn"
self.platform = "quimb"
self.configure_tn_simulation()
self.setup_backend_specifics()
def configure_tn_simulation(
self,
ansatz: str = "any",
max_bond_dimension: int = 10,
n_most_frequent_states: int = 100,
):
"""
Configure tensor network simulation.
Args:
ansatz : str, optional
The tensor network ansatz to use. Currently, only "MPS" or "any" is supported. In the second case
the generic Circuit Quimb class is used.
max_bond_dimension : int, optional
The maximum bond dimension for the MPS ansatz. Default is 10.
Notes:
- The ansatz determines the tensor network structure used for simulation. Currently, only "MPS" is supported.
- The `max_bond_dimension` parameter controls the maximum allowed bond dimension for the MPS ansatz.
"""
self.ansatz = ansatz
self.max_bond_dimension = max_bond_dimension
self.n_most_frequent_states = n_most_frequent_states
def setup_backend_specifics(self, qimb_backend="numpy", optimizer="auto-hq"):
"""Setup backend specifics.
Args:
qimb_backend: str
The backend to use for the quimb tensor network simulation.
optimizer: str, optional
The optimizer to use for the quimb tensor network simulation.
"""
self.backend = qimb_backend
self.optimizer = optimizer
def execute_circuit(
self,
circuit,
initial_state=None,
nshots=None,
return_array=False,
):
"""
Execute a quantum circuit using the specified tensor network ansatz and initial state.
Args:
circuit : QuantumCircuit
The quantum circuit to be executed.
initial_state : array-like, optional
The initial state of the quantum system. Only supported for Matrix Product States (MPS) ansatz.
nshots : int, optional
The number of shots for sampling the circuit. If None, no sampling is performed, and the full statevector is used.
return_array : bool, optional
If True, returns the statevector as a dense array. Default is False.
Returns:
TensorNetworkResult
An object containing the results of the circuit execution, including:
- nqubits: Number of qubits in the circuit.
- backend: The backend used for execution.
- measures: The measurement frequencies if nshots is specified, otherwise None.
- measured_probabilities: A dictionary of computational basis states and their probabilities.
- prob_type: The type of probability computation used (currently "default").
- statevector: The final statevector as a dense array if return_array is True, otherwise None.
Raises:
ValueError
If an initial state is provided but the ansatz is not "MPS".
Notes:
- The ansatz determines the tensor network structure used for simulation. Currently, only "MPS" is supported.
- If `initial_state` is provided, it must be compatible with the MPS ansatz.
- The `nshots` parameter enables sampling from the circuit's output distribution. If not specified, the full statevector is computed.
"""
if initial_state is not None and self.ansatz == "MPS":
initial_state = qtn.tensor_1d.MatrixProductState.from_dense(
initial_state, 2
) # 2 is the physical dimension
elif initial_state is not None:
raise_error(
ValueError, "Initial state not None supported only for MPS ansatz."
)
circ_ansatz = (
qtn.circuit.CircuitMPS if self.ansatz == "MPS" else qtn.circuit.Circuit
)
circ_quimb = circ_ansatz.from_openqasm2_str(
circuit.to_qasm(), psi0=initial_state
)
if nshots:
frequencies = Counter(circ_quimb.sample(nshots))
main_frequencies = {
state: count
for state, count in frequencies.most_common(self.n_most_frequent_states)
}
computational_states = list(main_frequencies.keys())
amplitudes = {
state: circ_quimb.amplitude(state) for state in computational_states
}
measured_probabilities = {
state: abs(amplitude) ** 2 for state, amplitude in amplitudes.items()
}
else:
frequencies = None
measured_probabilities = None
statevector = (
circ_quimb.to_dense(backend=self.backend, optimize=self.optimizer)
if return_array
else None
)
return TensorNetworkResult(
nqubits=circuit.nqubits,
backend=self,
measures=frequencies,
measured_probabilities=measured_probabilities,
prob_type="default",
statevector=statevector,
)
def expectation(self, circuit, observable):
"""
Compute the expectation value of a Qibo-friendly ``observable`` on the Tensor Network constructed from a Qibo ``circuit``.
This method takes a Qibo-style symbolic Hamiltonian (e.g., `X(0)*Z(1) + 2.0*Y(2)*Z(0)`)
as the observable, converts it into a Quimb observable and computes its expectation
value using the provided circuit.
Args:
circuit: A Qibo quantum circuit object on which the expectation value
is computed.
observable: The observable whose expectation value we want to compute.
This must be provided in the symbolic Hamiltonian form supported by Qibo
(e.g., `X(0)*Y(1)` or `Z(0)*Z(1) + 1.5*Y(2)`).
Returns:
float: The expectation value (real part).
"""
'''Convert Qibo observables to Quimb'''
operators_list, sites_list, coeffs_list = self._qiboobs_to_quimbobs(observable)
'''Convert Qibo circuit to Quimb circuit'''
parameters = circuit.get_parameters()
quimb_circuit = self._qibo_circuit_to_quimb(
circuit, quimb_circuit_type=qtn.Circuit, to_backend=jnp.array, convert_eager=True
)
quimb_parameters = {
key: jnp.asarray(parameters[i]) for i, key in enumerate(quimb_circuit.get_params().keys())
}
quimb_circuit.set_params(quimb_parameters)
'''Compute expectation value'''
expectation_value = 0.0
for ops, sites, coeffs in zip(operators_list, sites_list, coeffs_list):
exp_values = quimb_circuit.local_expectation(
ops,
where=sites,
backend=self.backend,
optimize=self.optimizer
)
expectation_value = expectation_value + coeffs * exp_values
return jnp.real(expectation_value)
def expectation_old(self, circuit, observable):
"""Compute the expectation value of a Qibo-friendly ``observable`` on the Tensor Network constructed from a Qibo ``circuit``.
This method takes a Qibo-style symbolic Hamiltonian (e.g., `X(0)*Z(1) + 2.0*Y(2)*Z(0)`)
as the observable, converts it into a Quimb observable and computes its expectation
value using the provided circuit. In case of multiple terms on the same group of qubits, they can be computed in a single contraction.
A grouping procedure is applied to optimize the number of contractions performed.
Args:
circuit: A Qibo quantum circuit object on which the expectation value
is computed.
observable: The observable whose expectation value we want to compute.
This must be provided in the symbolic Hamiltonian form supported by Qibo
(e.g., `X(0)*Y(1)` or `Z(0)*Z(1) + 1.5*Y(2)`).
Returns:
float: The expectation value (real part).
"""
# Map the Qibo observable to Quimb operators and group local operators on the same sites
# for computing them in a single contraction. This does not work with CircuitMPS for some now
# for Quimb 1.11.1
operators_list, sites_list, coeffs_list = self._qiboobs_to_quimbobs(observable)
sites_list_grouped, operators_list_grouped, coeffs_list_grouped = (
self._group_by_tuples(sites_list, operators_list, coeffs_list)
)
if self.ansatz == "MPS":
if len(sites_list) - len(sites_list_grouped) > 10:
warnings.warn(
"More than 10 local operators on the same sites are not being grouped as this is not compatible with CircuitMPS. Expected value computation can be more efficient without an MPS ansatz."
)
circ_ansatz = qtn.circuit.CircuitMPS
circ = circ_ansatz.from_openqasm2_str(circuit.to_qasm())
expectation_value = 0.0
for ops, sites, coeffs in zip(operators_list, sites_list, coeffs_list):
exp_values = circ.local_expectation(
ops, where=sites, backend=self.backend, optimize=self.optimizer
)
expectation_value += np.dot(coeffs, exp_values)
return np.real(expectation_value)
else:
circ_ansatz = qtn.circuit.Circuit
circ = circ_ansatz.from_openqasm2_str(circuit.to_qasm())
expectation_value = 0.0
for ops, sites, coeffs in zip(
operators_list_grouped, sites_list_grouped, coeffs_list_grouped
):
exp_values = circ.local_expectation(
ops, where=sites, backend=self.backend, optimize=self.optimizer
)
expectation_value += np.dot(coeffs, exp_values)
return np.real(expectation_value)
def _qiboobs_to_quimbobs(self, hamiltonian):
"""
Convert a Qibo SymbolicHamiltonian into a Quimb-compatible decomposition.
Returns three lists:
- operators_list: Quimb operators (tensor products of Pauli matrices).
- sites_list: tuples of qubit indices the operators act on.
- coeffs_list: coefficients for each term.
"""
factor_pattern = re.compile(r"([^\d]+)(\d+)")
operators_list = []
sites_list = []
coeffs_list = []
for term in hamiltonian.terms:
coeff = term.coefficient
term_ops = []
term_sites = []
for factor in term.factors:
match = factor_pattern.match(str(factor))
if not match:
raise ValueError(
f"Factor '{str(factor)}' does not match the expected format."
)
operator_name = match.group(1)
qubit_index = int(match.group(2))
# Build the single-qubit operator
if operator_name not in {"X", "Y", "Z", "I"}:
raise ValueError(f"Unsupported operator {operator_name}")
op = qu.pauli(operator_name)
term_ops.append(op)
term_sites.append(qubit_index)
# Build the tensor product if more than one factor
if term_ops:
full_op = term_ops[0]
for op in term_ops[1:]:
full_op = full_op & op
else:
# Identity term (just coefficient)
full_op = qu.eye(2)
operators_list.append(full_op)
sites_list.append(tuple(term_sites))
coeffs_list.append(coeff)
return operators_list, sites_list, coeffs_list
def _group_by_tuples(self, A, B, C):
"""
Groups the elements of B and C by the unique tuples in A.
Parameters:
A (list of tuples): key tuples (can contain duplicates)
B (list): values aligned with A
C (list): values aligned with A
Returns:
(A_new, B_new, C_new):
A_new: list of unique tuples
B_new: list of lists of grouped values from B
C_new: list of lists of grouped values from C
"""
grouped_B = defaultdict(list)
grouped_C = defaultdict(list)
for a, b, c in zip(A, B, C):
grouped_B[a].append(b)
grouped_C[a].append(c)
A_new = list(grouped_B.keys())
B_new = list(grouped_B.values())
C_new = list(grouped_C.values())
return A_new, B_new, C_new
# def _qibo_circuit_to_quimb(self, qibo_circ, quimb_circuit_type=qtn.Circuit, **circuit_kwargs):
# """
# Convert a Qibo Circuit to a Quimb Circuit.
# Parameters
# ----------
# qibo_circ : qibo.models.circuit.Circuit
# The circuit to convert.
# quimb_circuit_type : type
# The Quimb circuit class to use (Circuit, CircuitMPS, etc).
# circuit_kwargs : dict
# Extra arguments to pass to the Quimb circuit constructor.
# Returns
# -------
# circ : quimb.tensor.circuit.Circuit
# The converted circuit.
# """
# nqubits = qibo_circ.nqubits
# quimb_gates = []
# circ = quimb_circuit_type(nqubits, **circuit_kwargs)
# for gate in qibo_circ.queue:
# gname = getattr(gate, "name", None)
# qname = GATE_MAP.get(gname, None)
# if qname is None:
# continue # skip measurements and unknown gates
# # Handle parametrized gates (Qibo: .parameters, Quimb: expects flat tuple)
# params = getattr(gate, "parameters", ())
# qubits = getattr(gate, "qubits", ())
# # Quimb expects (*params, *qubits)
# gate_spec = (qname,) + tuple(params) + tuple(qubits)
# quimb_gates.append(gate_spec)
# circ.apply_gates(quimb_gates)
# return circ
def _qibo_circuit_to_quimb(self, qibo_circ, quimb_circuit_type=qtn.Circuit, **circuit_kwargs):
"""
Convert a Qibo Circuit to a Quimb Circuit.
Parameters
----------
qibo_circ : qibo.models.circuit.Circuit
The circuit to convert.
quimb_circuit_type : type
The Quimb circuit class to use (Circuit, CircuitMPS, etc).
circuit_kwargs : dict
Extra arguments to pass to the Quimb circuit constructor.
Returns
-------
circ : quimb.tensor.circuit.Circuit
The converted circuit.
"""
nqubits = qibo_circ.nqubits
circ = quimb_circuit_type(nqubits, **circuit_kwargs)
for gate in qibo_circ.queue:
gname = getattr(gate, "name", None)
qname = GATE_MAP.get(gname, None)
if qname is None:
continue # skip measurements and unknown gates
params = getattr(gate, "parameters", ())
qubits = getattr(gate, "qubits", ())
# Check if the gate is parametrized
is_parametrized = isinstance(gate, ParametrizedGate)
if is_parametrized:
circ.apply_gate(
qname,
*params,
*qubits,
parametrized= is_parametrized
)
else:
circ.apply_gate(
qname,
*params,
*qubits,
)
return circ