qevedo-simulator

Python bindings of qvd, a fast quantum circuit simulator written in Rust. It installs as qevedo-simulator and imports as qevedo.simulator.

pip install qevedo-simulator          # add [numpy] for NumPy state vectors

from qevedo.simulator import Circuit, expectation, run, statevector

bell = Circuit(2).h(0).cx(0, 1).measure_all()
result = run(bell, shots=1000, seed=1)
print(result.counts, result.backend)  # {'00': ..., '11': ...} stabilizer

# OpenQASM 2 or 3 text works anywhere a circuit does.
result = run("""
OPENQASM 3.0;
include "stdgates.inc";
qubit[3] q;
bit[3] c;
h q[0];
cx q[0], q[1];
ccx q[0], q[1], q[2];
c = measure q;
""", shots=1000)

state = statevector(Circuit(3).h(0).cx(0, 1).t(2))  # NumPy complex128 array

# Expectation values by Pauli propagation, far beyond a state vector:
# a Pauli string, a {string: coefficient} dict or a Qiskit SparsePauliOp.
e = expectation(Circuit(2).h(0).cx(0, 1), {"Z0 Z1": 1.0, "X0 X1": 0.5})
print(e.value, e.truncated_one_norm)  # 1.5, and a bound on its error (0)

Clifford circuits can carry Pauli noise channels (x_error, y_error, z_error, pauli_channel_1, depolarize1, depolarize2, Stim's channels), sampled independently in every shot by the stabilizer backend. sample returns the raw measurement records, measurement-major like Stim's FlipSimulator, and to_stim writes the same circuit for Stim:

from qevedo.simulator import Circuit, sample

c = Circuit(3)
for _ in range(10):                      # a small repetition code
    c.reset(1).x_error(1, 0.001)
    c.cx(0, 1).depolarize2(0, 1, 0.001).cx(2, 1).depolarize2(2, 1, 0.001)
    c.x_error(1, 0.001).measure(1, c.num_clbits)
bits = sample(c, 100_000, seed=1)        # NumPy bool array, (10, 100000)
print(c.to_stim().splitlines()[:3])      # ['R 1', 'X_ERROR(0.001) 1', 'CX 0 1']

Error-correction circuits written in Stim's format (text or a stim.Circuit, for example from stim.Circuit.generated) run directly. sample_detectors returns detection events and observable flips, shot-major as Stim's detector sampler returns them and as decoders take them; from_stim gives the Circuit with its detectors and observables as lists of measurement indices, and sample takes Stim text too:

import pymatching, stim
from qevedo.simulator import sample_detectors

circuit = stim.Circuit.generated("surface_code:rotated_memory_z", distance=5, rounds=5,
                                 after_clifford_depolarization=0.003)
detectors, observables = sample_detectors(circuit, 100_000, seed=1)  # (100000, 120), (100000, 1)
matching = pymatching.Matching.from_detector_error_model(circuit.detector_error_model(decompose_errors=True))
logical_error_rate = (matching.decode_batch(detectors) != observables).any(axis=1).mean()

The reader takes Stim's Clifford gates, M/MX/MY/MR/MRX/MRY with measurement-flip probabilities (M(0.01)) and inverted targets (M !3), R/RX/RY, MPAD, the Pauli channels, DETECTOR, OBSERVABLE_INCLUDE and nested REPEAT blocks; anything else (MPP, correlated errors, classical control, ...) is an error naming the line.

run chooses the simulation method automatically:

Pass backend= to force one. The result reports the method that ran, its timings, and the MPS fidelity estimate or the near-Clifford peak active qubits. Simulations release the GIL.

expectation propagates the observable backwards through the circuit (Cliffords for free, small coefficients dropped), for up to 1024 qubits: <Z62> of IBM's 127-qubit, 20-step kicked Ising circuit takes 0.25–9.6 s at the thresholds in the main README's benchmark.

x86-64 wheels are built for AVX2 + FMA processors (2013–2015 or newer); other x86-64 machines can build from source with a Rust toolchain.

License: MIT.