Noise Model

A noise model is a mathematical representation of the errors that occur during quantum computation on real hardware. It captures the effects of decoherence, gate imperfections, measurement errors, and other imperfections, allowing classical simulators to approximate the behavior of a noisy quantum processor. Noise models are essential for algorithm development, error mitigation strategy design, and benchmarking.

Types of noise

Noise in quantum computers comes from several sources, each modeled differently:

Gate errors: Each gate application introduces a small error. Common models include:

• Depolarizing noise: with probability $p$, the qubit state is replaced by the maximally mixed state $I/2$. The quantum channel is $\mathcal{E}(\rho) = (1-p)\rho + \frac{p}{3}(X\rho X + Y\rho Y + Z\rho Z)$
• Pauli noise: with specified probabilities, an $X$, $Y$, or $Z$ error is applied after the gate
• Coherent errors: systematic over-rotation or under-rotation of gates, modeled as a unitary close to but not equal to the ideal gate

Decoherence: Idle qubits lose quantum information over time. Modeled using the $T_1$ (amplitude damping) and $T_2$ (phase damping) relaxation times:

• Amplitude damping: $|1\rangle$ decays to $|0\rangle$ with probability $1 - e^{-t/T_1}$
• Phase damping: off-diagonal elements of the density matrix decay as $e^{-t/T_2}$

Readout errors: The measurement result may not match the actual qubit state. Modeled as a classical confusion matrix:

$P(\text{measure } i | \text{state } j) = M_{ij}$

For example, a qubit in $|0\rangle$ might be read as 1 with probability 2%, and a qubit in $|1\rangle$ might be read as 0 with probability 5%.

Crosstalk: Operations on one qubit affect neighboring qubits. This is harder to model and is often omitted from simple noise models.

Building noise models

Noise models can be constructed in several ways:

• From calibration data: Hardware providers publish error rates, $T_1$/$T_2$ times, and readout fidelities for each qubit. These can be assembled into a device-specific noise model.
• From tomographic characterization: Process tomography or randomized benchmarking can characterize the actual noise channels.
• Simplified models: For theoretical analysis, uniform depolarizing noise (same error rate on every gate) is a common simplification.

In Qiskit, a noise model matching a real backend can be created from its calibration data:

from qiskit_aer.noise import NoiseModel
from qiskit_ibm_runtime import QiskitRuntimeService

service = QiskitRuntimeService()
backend = service.backend("ibm_sherbrooke")
noise_model = NoiseModel.from_backend(backend)

Why it matters for learners

Before running circuits on quantum hardware (which costs time and money), testing them against a noise model on a classical simulator is standard practice. Noise models also help you understand which parts of your circuit are most affected by errors, guiding optimization efforts. The gap between a noise model’s predictions and actual hardware results reveals the limitations of the noise model itself, which is an active area of research.

See also

• Decoherence
• Depolarizing Channel
• Gate Fidelity
• Quantum Error Mitigation
• Randomized Benchmarking