Quantum Error Mitigation Overview

Techniques that reduce the effect of noise on quantum results without the full overhead of error correction.

Mitigation versus correction

Quantum error correction encodes logical qubits into many physical qubits and actively removes errors, but it demands large qubit overhead and high-fidelity gates beyond current hardware. Error mitigation instead accepts noisy runs and applies classical post-processing, often across many circuit executions, to estimate what a noiseless result would have been. It does not restore a clean quantum state; it corrects estimates of observables.

The core idea

Most mitigation methods target expectation values . They run circuits under controlled variations of noise or with modified circuits, collect the noisy expectation values, and combine them so that the leading noise contribution cancels. The result is a better estimate at the cost of extra measurements, often many more shots, but no extra qubits.

Major techniques

• Zero-noise extrapolation: amplify noise deliberately, then extrapolate to zero noise (see zero-noise extrapolation).
• Probabilistic error cancellation: invert the noise channel by sampling from a quasi-probability decomposition (see probabilistic error cancellation).
• Measurement error mitigation: correct readout errors using a calibrated confusion matrix.
• Clifford data regression and symmetry verification: use classically simulable references or conserved quantities to correct bias.

Costs and trade-offs

Mitigation trades qubits for samples. The required number of measurements often grows exponentially with circuit depth or error rate, because the methods estimate small corrections against sampling noise. This limits mitigation to modest circuits. It also generally targets expectation values, not full state recovery, and can amplify statistical variance even as it reduces bias.

Role in the roadmap

Error mitigation extends the useful reach of noisy intermediate-scale devices and is a bridge to full fault tolerance. Honest reporting states both the raw and mitigated results and the sampling overhead, since a mitigated number is a statistical estimate, not a certified value. In any simulation-based engineering study, mitigated quantum results should be validated against classical references where possible.