Decoder Model Compatibility Provides Information beyond the Logical Gap under Drifting and Correlated Quantum Noise
Aaron C. Hoyt, Chunshu Wu, Shuwen Kan, Sean Garner, Avimita Chatterjee, Drew Rebar, Norman M. Tubman, Katherine Klymko, Chenxu Liu, Ang Li, Samuel Stein
quant-phSubmitted Sep 24, 2026• Announced Sep 25, 2026• 0 citations
Quantum error-correcting decoders provide both corrections and confidence estimates for the logical-error risk of individual measurement records. Such confidence can guide postselection, but it depends on the accuracy of the decoder noise model. When the physical device drifts away from that model, confidence measures such as the logical gap can become miscalibrated. We show that this mismatch can itself provide useful information. On IBM quantum hardware, syndrome statistics drift across consecutive acquisition windows, and the resulting detector histories differ substantially from those predicted by the stationary detector error model used for decoding. We observe a similar hardware-model mismatch in independent Google surface-code memory data. We exploit this effect by assigning additional rejection weight to detector histories that are unlikely under the assumed model. For minimum-weight perfect matching (MWPM), the minimum weight $W$ provides an inexpensive proxy for syndrome surprisal. On held-out IBM repetition-code memories, likelihood-aware postselection reduces the retained logical-error probability (LEP) by $10.6\%$ at $d=7$ and $28.8\%$ at $d=9$, both at $15\%$ rejection. On Google ten-round $d=5$ XZZX surface-code memory data, it reduces retained LEP by $75.4\%$ at $90\%$ rejection. Applied to Google's magic-state cultivation experiment with the Tesseract decoder, it reduces tomography infidelity by $34.7\%$ with $30\%$ additional grafting-stage rejection, or increases accepted yield by $30.4\%$ at a target fidelity of $97.0\%$. Burst-noise simulations show that the advantage persists as code distance and syndrome-extraction volume increase. The method reuses quantities already produced during logical-gap decoding and requires only an additional fitting step.
AI Summary
The paper asks whether decoder-model compatibility adds useful logical-risk information beyond the logical gap when physical noise differs from a frozen decoding model. It distinguishes relative confidence between logical sectors from the absolute likelihood of a detector history. For MWPM, the minimum sector cost $W=\min(C_0,C_1)$ approximates syndrome surprisal at fixed gap $\Delta=|C_1-C_0|$. Postselection combines percentile-normalized $\Delta$ and $W$, with one coefficient chosen on earlier validation data.
A likelihood-response calculation shows that mismatch changes logical odds only when it reweights competing sectors differently. Exactly enumerated small circuits isolate this effect from MWPM approximation errors; whole-shot burst simulations test it through distance $15$ with $d$ extraction rounds. Hardware tests use chronological held-out IBM repetition-code records, public Google surface-code memories, and Google magic-state cultivation data decoded with Tesseract. At matched acceptance, retained logical-error probability falls relative to gap-only selection by $10.6\%$ at IBM $d=7$ and $28.8\%$ at $d=9$, both at $15\%$ rejection. Google ten-round $d=5$ memories show a $75.4\%$ reduction at $90\%$ rejection, with a $95\%$ interval of $[50.8\%,100\%]$. Three-cycle cultivation shows $34.7\%$ lower tomography infidelity at $30\%$ additional rejection, or $30.4\%$ higher accepted yield at $97\%$ target fidelity.
Compatibility alone does not imply logical risk, and $W$ is not established as optimal. Benefits vary with decoder prior, duration, and acceptance. Scaling tests use a specific whole-shot burst model; bootstrap intervals condition on fitted policies. Cultivation ordering is available only within each tomography basis, and yield gains are not demonstrated end-to-end computational speedups.
Some signs of AI-generated textMissing comparison
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
Quantum error-correcting decoders provide both corrections and confidence estimates for the logical-error risk of individual measurement records. Such confidence can guide postselection, but it depends on the accuracy of the decoder noise model. When the physical device drifts away from that model, confidence measures such as the logical gap can become miscalibrated. We show that this mismatch can itself provide useful information. On IBM quantum hardware, syndrome statistics drift across consecutive acquisition windows, and the resulting detector histories differ substantially from those predicted by the stationary detector error model used for decoding. We observe a similar hardware-model mismatch in independent Google surface-code memory data. We exploit this effect by assigning additional rejection weight to detector histories that are unlikely under the assumed model. For minimum-weight perfect matching (MWPM), the minimum weight W provides an inexpensive proxy for syndrome surprisal. On held-out IBM repetition-code memories, likelihood-aware postselection reduces the retained logical-error probability (LEP) by 10.6% at d=7 and 28.8% at d=9, both at 15% rejection. On Google ten-round d=5 XZZX surface-code memory data, it reduces retained LEP by 75.4% at 90% rejection. Applied to Google's magic-state cultivation experiment with the Tesseract decoder, it reduces tomography infidelity by 34.7% with 30% additional grafting-stage rejection, or increases accepted yield by 30.4% at a target fidelity of 97.0%. Burst-noise simulations show that the advantage persists as code distance and syndrome-extraction volume increase. The method reuses quantities already produced during logical-gap decoding and requires only an additional fitting step.
The paper asks whether decoder-model compatibility adds useful logical-risk information beyond the logical gap when physical noise differs from a frozen decoding model. It distinguishes relative confidence between logical sectors from the absolute likelihood of a detector history. For MWPM, the minimum sector cost W=min(C0,C1) approximates syndrome surprisal at fixed gap Δ=∣C1−C0∣. Postselection combines percentile-normalized Δ and W, with one coefficient chosen on earlier validation data.
A likelihood-response calculation shows that mismatch changes logical odds only when it reweights competing sectors differently. Exactly enumerated small circuits isolate this effect from MWPM approximation errors; whole-shot burst simulations test it through distance 15 with d extraction rounds. Hardware tests use chronological held-out IBM repetition-code records, public Google surface-code memories, and Google magic-state cultivation data decoded with Tesseract. At matched acceptance, retained logical-error probability falls relative to gap-only selection by 10.6% at IBM d=7 and 28.8% at d=9, both at 15% rejection. Google ten-round d=5 memories show a 75.4% reduction at 90% rejection, with a 95% interval of [50.8%,100%]. Three-cycle cultivation shows 34.7% lower tomography infidelity at 30% additional rejection, or 30.4% higher accepted yield at 97% target fidelity.
Compatibility alone does not imply logical risk, and W is not established as optimal. Benefits vary with decoder prior, duration, and acceptance. Scaling tests use a specific whole-shot burst model; bootstrap intervals condition on fitted policies. Cultivation ordering is available only within each tomography basis, and yield gains are not demonstrated end-to-end computational speedups.
Decoders for quantum error correction (QEC) experiments rely on detector error models (DEMs), which encode, for each error, its probability and the detectors and logical observables it flips. Here we show that estimating DEM event probabilities from experimental syndromes is feasible, avoids independent device benchmarking, and produces useful decoder priors for estimating and reducing decoded logical error probabilities. We evaluate our methods using open-source data from surface-code memory experiments performed on Google's Willow chip, and we carry out analogous surface-code experiments on IBM's \texttt{ibm\_miami} processor. Despite the different physical error scales of the Google and IBM devices, in both cases our estimated DEMs improve logical error probabilities relative to baseline device-informed DEMs, typically at the 5%−10% level and with larger gains in some IBM cases, without additional calibration circuits, decoder fine-tuning, or supervised fitting to logical outcomes.
Abstract Quantum error-correction decoders must convert noisy syndrome records into logical decisions while operating under strict latency and reliability constraints. Minimum-weight perfect matching (MWPM) remains a strong baseline for surface-code decoding because it is fast, interpretable, and exact for its graphlike optimization problem [1–8]. However, the graphlike independent-error approximation can become mismatched when syndrome records contain temporal drift, measurement corruption, crosstalk, bursts, or mixtures of mechanisms [9–18]. We introduce SAGE-QEC, a failure-aware selective decoder that retains nominal PyMatching as its default decision and learns when an observable syndrome structure warrants an alternative candidate. SAGE-QEC uses train-frozen structure thresholds, a candidate ranker, validation-only policy selection, and a paired blind evaluation. Across 45 simulator conditions, SAGE-QEC reduced the aggregate logical error rate from 27.1285% for PyMatching to 24.8320%, corresponding to an 8.47% relative reduction. The campaign used rotated surface-code memory circuits at distances 3, 5, and 7, nominal circuit parameters 0.002, 0.003, and 0.005, five seeds, and four test regimes withheld from training: Burst-3, Temporal Drift, Crosstalk, and Mixed Noise. The test set contained 81,000 paired shots. All twelve distance-by-regime aggregate comparisons favored SAGE-QEC in the executed campaign, with paired bootstrap intervals below zero for SAGE-QEC minus PyMatching. The result is evidence of a reproducible simulator-level improvement, not a claim of universal decoder superiority. SAGE-QEC currently uses logical-output proxy candidates rather than a complete physical correction search, and its intervention rate remains high under distribution shift. It has not yet been compared with a fully calibrated correlated-noise space-time PyMatching baseline or hardware calibration data. We therefore characterize SAGE-QEC as a strong qualification-stage prototype and release a complete audit protocol for the next industrial validation stage.
To utilize quantum error-correcting codes, a decoder must infer the logical sector from the measured syndrome. Beyond a hard logical decision, some decoders provide soft information estimating its reliability. For minimum-weight perfect matching (MWPM), a common confidence measure is the complementary, or logical, gap. Neural-network decoders trained with binary cross entropy naturally output a different score, the logit, whose relation to the matching gap remains unclear. Here, we test whether the logit of a graph neural network (GNN) decoder behaves as a learned proxy for the logical gap. Using a pretrained GNN for the rotated surface code under uniform circuit-level noise [PRR, 7(2):023181, 2025], we compare its logit with the MWPM gap shot by shot on identical syndromes. The native GNN decoder-logit pair yields a lower post-selected logical error rate than the native MWPM decoder-gap pair over the operationally relevant range of acceptance rates. Exchanging the scores between decoders isolates score discrimination from hard-decision accuracy and shows that both contribute to the advantage. At larger distances each score best ranks failures of its own decoder, whereas at small distance, where the bounded matching gap saturates, the learned confidence better discriminates MWPM failures than the gap itself. Finally, GNN confidence is substantially better described by the posterior logistic form implied by an ideal log-likelihood ratio, while the MWPM gap shows systematic departures. Thus, a neural-network decoder trained only on syndromes and logical labels learns a gap-like confidence proxy whose native scale more closely follows the expected posterior calibration form, supporting learned soft output when MWPM gap estimates are unavailable, costly, or insufficiently expressive.
Fault-tolerant quantum computers use decoders to monitor for errors and find a plausible correction. A decoder may provide a \emph{decoder confidence score} (DCS) to gauge its success. We adopt a \emph{swim distance} DCS, computed from the shortest path between syndrome clusters. By contracting tensor networks, we compare its performance under phenomenological noise to the well-known \emph{complementary gap} and find that both reliably estimate the logical error probability (LEP) in a decoding window. We explore ways to use this to mitigate the LEP in entire logical circuits. For shallow circuits, we just abort if any decoding window produces an exceptionally low DCS: for one window of a distance-13 surface code under circuit-level noise, rejecting a mere 0.1% of possible DCS values improves the window's mean LEP by more than 5 orders of magnitude. For larger algorithms comprising up to billions of windows, DCS-based rejection remains effective for enhancing observable estimation. Moreover, one can use the DCS to assign each circuit's output a unique LEP, and use it as a basis for maximum likelihood estimation. This can reduce the effects of noise by an order of magnitude at no quantum cost; we explore the combination of these methods.
Quantum error correction can be enhanced by post-selecting out runs that are likely to produce a logical failure, but the most accurate measures for that require costly decoder-level information. We introduce a practical, decoder-agnostic post-selection method that learns directly from syndrome data. The method trains a supervised classifier to distinguish between syndromes from low- and high-noise regimes, and then uses the classifier's output as an abort score for new runs, without requiring logical-error labels, correction operators, or code-specific likelihood calculations. We validate the approach in three complementary settings: circuit-level simulations of the Gross bivariate-bicycle code, code-capacity simulations of the surface code, and experimental logical magic-state distillation data from the QuEra neutral-atom processor. In the Gross and surface codes, learned syndrome post-selection reduces the conditional logical error rate at a fixed acceptance rate, with performance comparable to syndrome-weight filtering. For the surface code, the learned classifier reveals a post-selection transition distinct from the conventional decoding threshold. In the experimental data, the machine-learning score outperforms syndrome-weight post-selection and, when combined with logical-gap filtering, improves the output fidelity beyond using the logical gap alone. These results show that syndrome-only learning provides a scalable and hardware-compatible route to improving the reliability of quantum error correction.
Quantum error correction can suppress logical errors exponentially with code distance when physical noise is below threshold. Realizing this on hardware requires tight coordination among quantum gates, mid-circuit measurements, and qubit coherence. Device variation, noise asymmetry, and correlations across space and time complicate the translation of these capabilities into logical error suppression. Here, we study rotated surface-code memory on an IBM Nighthawk r2 processor. Performance varies across tested processor placements, which we attribute to differences in qubit quality. We optimize the code by combining short readout, fused measurement and reset, and an XX echo on data qubits during ancilla readout. Compared with standard readout without reset or echo, the optimized syndrome extraction circuit reduces the median per-round logical error rate by factors of 3.8 in the X basis and 1.4 in the Z basis. In memory experiments of up to 150 rounds, the distance advantage weakens, alongside long-lived correlations that may reflect a persistent error component. Under our optimization recipe, increasing the protected distance from 3 to 6 while holding the other distance fixed reduces the per-round logical error rate from 0.53% to 0.24% in Z memory and from 0.52% to 0.22% in X memory. The corresponding error-suppression factors under maximum-likelihood decoding are 1.68 and 1.85. These results connect circuit compilation and optimization to surface-code memory performance and limitations, and motivate QPU and QEC circuit codesign for fault-tolerant quantum computing primitives on current cloud hardware.