Noisy Simulation

Model real hardware imperfections by attaching a noise model to a kernel and running it on a density-matrix backend.

Selecting a noisy backend

import qalgora

qalgora.set_target("density-matrix-cpu")   # tracks the full density matrix

Building a noise model

Compose channels (bit-flip, depolarization, amplitude damping) and bind them to specific gates and qubits.

noise = qalgora.NoiseModel()

# 1% depolarizing error after every Hadamard on qubit 0
noise.add_channel("h", [0], qalgora.DepolarizationChannel(0.01))

# amplitude damping + depolarizing on every `x` gate (a single-qubit channel)
# to model 2-qubit noise on CNOT / controlled-X, bind to that gate name or a
# 2-qubit channel instead (per the reference build's supported API)
noise.add_all_qubit_channel("x", qalgora.AmplitudeDampingChannel(0.02))
noise.add_all_qubit_channel("x", qalgora.DepolarizationChannel(0.02))

# apply one channel to a gate on EVERY qubit, without listing them
noise.add_all_qubit_channel("h", qalgora.DepolarizationChannel(0.005))

Running with noise

@qalgora.kernel
def bell():
    q = qalgora.qvector(2)
    h(q[0])
    x.ctrl(q[0], q[1])
    mz(q)

ideal = qalgora.sample(bell, shots_count=2000)
noisy = qalgora.sample(bell, shots_count=2000, noise_model=noise)

print("ideal:", ideal)   # ~{ 00, 11 }
print("noisy:", noisy)   # leaks into 01 / 10

Built-in channels

Every channel is ultimately a list of KrausOperator matrices; the named channels above are convenience constructors that build the right Kraus set for you.

Custom Kraus channel

import numpy as np
# a custom channel from its Kraus operators (must satisfy Σ Kᵢ†Kᵢ = I)
k0 = np.array([[1, 0], [0, np.sqrt(0.9)]])
k1 = np.array([[0, np.sqrt(0.1)], [0, 0]])
noise.add_channel("z", [1], qalgora.KrausChannel([k0, k1]))

Inline noise & trajectory simulation

Apply a channel at a specific point with apply_noise inside a kernel — this runs on the reference build. For many qubits, a planned GPU trajectory backend would sample noise stochastically as a statistical approximation, instead of tracking the full density matrix.

@qalgora.kernel
def with_inline_noise():
    q = qalgora.qvector(2)
    h(q[0])
    qalgora.apply_noise(qalgora.DepolarizationChannel, 0.01, q[0])
    mz(q)

# Planned GPU trajectory backend — not in the open reference build yet:
# qalgora.set_target("gpu")   # statistical noise sampling, scales past density matrix