PRISM-Q (Python)

Python bindings for PRISM-Q, a performance-first quantum circuit simulator written in Rust.

Install

pip install prism-q

The Linux and Windows wheels include the CUDA backends. They need an NVIDIA driver and NVRTC, which the cuda12 extra installs:

pip install "prism-q[cuda12]"

prism_q.gpu_info() reports the device, or why it cannot be used.

Quick start

import prism_q

# Build a Bell state and run it.
circuit = prism_q.CircuitBuilder(2).h(0).cx(0, 1).build()
outcome = prism_q.simulate(circuit).seed(42).run()
print(outcome.probabilities)        # array([0.5, 0., 0., 0.5])

# Parse OpenQASM and sample shots.
qasm = """
OPENQASM 3.0;
qubit[2] q;
bit[2] c;
h q[0];
cx q[0], q[1];
c = measure q;
"""
result = prism_q.simulate(prism_q.parse_qasm(qasm)).seed(7).shots(1000)
print(result.counts())              # {'00': ~500, '11': ~500}

# Exact statevector as a NumPy array.
sv = prism_q.simulate(prism_q.circuits.ghz(3)).seed(1).state_vector()
print(sv.dtype, sv.shape)           # complex128 (8,)

Bit ordering

Count keys and measurement bit indices are LSB-first: in a bitstring key, the leftmost character is classical bit 0, and q[0] is the least-significant qubit. This is reversed relative to Qiskit. For example, the state where only q[0] is set reads as "10...", not "...01".

Features

• Fluent CircuitBuilder and OpenQASM 3.0 parsing.
• Classical control on measured bits (ClassicalCondition, guarded with an else branch) and measurement in a Pauli basis or of a Pauli product.
• OpenQASM 3.0 export (Circuit.to_qasm) and text, SVG, and Jupyter circuit drawing.
• Reusable circuit generators (prism_q.circuits): QFT, GHZ, QAOA, hardware-efficient ansatz, quantum volume, and more.
• Backend selection via BackendKind (statevector, stabilizer, sparse, MPS, Pauli propagation, ...).
• Parameter sweeps without a rebuild (Parameters, PreparedCircuit) plus adjoint gradients.
• Noise models (NoiseModel, NoiseChannel, two-qubit Kraus channels included) for shot sampling.
• Compiled Clifford samplers (CompiledSampler) with streamed marginals, parities, and correlators, plus analytic noisy marginals (noisy_marginals_analytical).
• Native QEC programs (QecProgram) with detector and observable sampling and feed-forward corrections.
• NumPy output for probabilities, statevectors, and QEC bit matrices.
• Pickling for circuits, parameters, noise models, and QEC programs, so they cross multiprocessing, Ray, and Dask boundaries.

License

MIT OR Apache-2.0