New virtual error cancellation method strips residual errors from logical quantum circuits, in simulation

Google Quantum AI

Virtual error cancellation removes leftover logical errors from fault-tolerant quantum circuits using only classical post-processing of recorded error-check data. It needs no extra calibration circuits and no change to how the processor runs. In numerical simulations it outperformed common error-reduction approaches, but it has not been tested on hardware.

• It uses two decoders: a fast one that runs in real time on the hardware and a slow, highly accurate one run offline.
• Instead of discarding shots where the decoders disagree, it gives them negative weights, so the errors cancel on average.
• In simulation of a distance-7 surface code at physical error rate 0.005, the residual logical error falls below 7.95 × 10⁻¹⁰, against about 1.32 × 10⁻⁶ unmitigated.
• Sampling overhead is about 1.3 at distance 7, versus about 4.0 for standard logical error cancellation, and approaches 1 as code distance grows.
• The benchmark covers one code at one error rate, and the article reports no qubit counts, gate counts or hardware runs.