Curator's Take
AI Commentary
This article overturns the long‑standing belief that randomized compiling erases coherent‑error signatures by showing that the discarded twirl labels actually retain the full Fisher information about those errors. By correlating outcomes with the per‑shot random gate choices, researchers can now extract systematic error parameters at the quantum‑limited precision without any extra circuit overhead—a capability that could dramatically improve calibration and error mitigation on near‑term devices such as IBM’s 127‑qubit processor. The result bridges a gap between noise‑randomization techniques and precise hardware diagnostics, though it relies on preserving and processing the label data, which current software stacks may need to support explicitly.
— Mark Eatherly
Summary
Randomized compiling (RC) is the standard technique for converting coherent (systematic) gate errors into stochastic noise. The prevailing view is that the twirl destroys coherent-error information. An exact Fisher-information conservation law shows the opposite: the coherent-error information RC removes from the averaged output is preserved in full in the twirl labels -- the per-shot random gate choices that standard RC discards. Retaining the labels and forming a label-outcome correlation recovers the coherent error parameters at the quantum Fisher-information limit, unbiased under all standard incoherent channels, at zero additional circuit cost. Two theorems are proved, the conservation law is verified to machine precision across 12 circuit families, and recovery is confirmed on a 127-qubit IBM Quantum processor (ibm_marrakesh). The labeled estimator recovers injected coherent phases to within 0.0063 rad of the true value across all depths tested, while the standard marginal estimator returns near-zero signal at every depth.