Pith. sign in

REVIEW 3 major objections 5 minor 26 references

A neural pre-decoder that adapts to changing hardware noise can cut logical errors and residual matching work without replacing the global decoder.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-31 07:51 UTC pith:2WEMQJ4V

load-bearing objection Solid hybrid pre-decoding systems paper with real Willow zero-shot numbers; the 110/110 OOD story over-credits adaptation because the baseline never saw the OOD axes. the 3 major comments →

arxiv 2607.28422 v1 pith:2WEMQJ4V submitted 2026-07-30 cs.LG quant-ph

QAdapt: A Noise-Adaptive Neural Pre-Decoding Framework for Quantum Error Correction

Ran Miao , Rui Luo , Xiaohan Shan , Xiaoming Sun This is my paper
classification cs.LG quant-ph
keywords quantum error correctionsurface codeneural decoderpre-decodingcontinual learningelastic weight consolidationout-of-distribution generalization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Fault-tolerant quantum computers need classical decoders that stay accurate as hardware noise drifts and as syndrome streams get denser. Fixed neural decoders often fail under that shift, and full retraining is costly. QAdapt is a modular pre-decoder that learns local space-time patterns in syndrome data, adapts sequentially across noise regimes while protecting earlier knowledge, and hands a cleaner residual syndrome to a standard matching decoder. On 110 synthetic out-of-distribution noise settings it always lowers logical error rate versus a fixed neural pre-decoder baseline, and on public Willow hardware data—with no target fine-tuning—it cuts logical error rate by up to about 5.8% and residual backend latency by up to about 9.3%. The claim is practical: adaptive classical pre-processing can widen the operating range of existing decoding stacks under nonstationary noise.

Core claim

QAdapt shows that a noise-adaptive neural pre-decoder—trained sequentially on physically motivated noise tasks with forgetting mitigation, then paired with an unchanged global matcher—consistently improves logical accuracy and residual decoding load under both synthetic distribution shift and zero-shot transfer to real Willow surface-code data.

What carries the argument

QAdapt’s local–global pipeline: HTNet extracts spatial, temporal, and joint syndrome correlations with adaptive fusion and axis–channel gating, then Q-EWC sequential training preserves important weights across noise tasks; thresholded local corrections produce a residual syndrome that PyMatching decodes globally.

Load-bearing premise

Training on a short, hand-chosen sequence of synthetic noise boosts, with fixed forgetting regularization, is enough for the local corrections to stay helpful under real hardware drift and residual graphs that differ from the training family.

What would settle it

Run the same residual-matching pipeline on paired experimental shots from a drifting device (or the anonymized cloud platform) with and without QAdapt: if logical error rate or residual matching cost does not improve relative to the fixed pre-decoder baseline—or worsens after further real drift—the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Existing matching-based QEC stacks can gain accuracy and residual throughput by inserting an adaptive pre-decoder without changing the code or global algorithm.
  • Sequential noise-task training with importance regularization can reduce the need for full target-domain retraining when hardware noise evolves.
  • Lower residual syndrome density and ambiguity can ease classical latency pressure as code distance and round count grow.
  • Zero-shot gains on external hardware data imply that learned local corrections can transfer beyond a single mapped noise model.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If online Fisher updates and drift detection work, pre-decoders could track calibration cycles continuously rather than only predefined task sequences.
  • The same local–global split may transfer to other stabilizer codes and backends if residual topology, not just density, is what the matcher needs simplified.
  • End-to-end control-stack timing (neural inference plus residual construction plus matching) will decide whether backend-only latency wins become system-level wins.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. QAdapt is a hybrid neural pre-decoding framework for rotated surface-code QEC. A heterogeneous spatiotemporal network (HTNet) predicts local corrections from detector tensors; residual syndromes are decoded by PyMatching. The model is trained sequentially on five circuit-level Pauli noise tasks (T0 base through T4 Z-bias) with diagonal Fisher EWC (Q-EWC) to limit catastrophic forgetting. Claims are: (i) under mapped T0, HTNet beats the Ising-fast pre-decoder on LER and residual matching latency; (ii) after T0–T4 adaptation, QAdapt reduces LER versus Ising-fast on all 110 synthetic OOD configurations built from the same noise axes; (iii) zero-shot on Google Willow d=5/7 data it cuts LER by up to 5.79% and residual PyMatching latency by up to 9.32%, without target fine-tuning. The paper positions QAdapt as a modular, decoder-compatible interface for nonstationary hardware noise.

Significance. If the adaptation story holds under fair controls, the work is a useful systems contribution: modular pre-decoding that improves both logical accuracy and backend matching load under distribution shift, without replacing MWPM. The evaluation scale (110 OOD configs; external Willow zero-shot; explicit residual-only timing scope) and the clear local–global pipeline are strengths. The architecture (separate spatial/temporal/joint branches, axis–channel gating, raw-evidence skip) is a concrete design for dense detector streams. The result would matter for FTQC control stacks where hardware drift and decoder latency are first-class constraints. Significance is currently limited by whether gains are truly from continual noise adaptation versus multi-task exposure and architecture capacity alone.

major comments (3)
  1. [§5.2, §6.2–6.4, Tables 3–4] §5.2, §6.2–6.4, Tables 3–4, App. C: The load-bearing OOD and Willow comparisons pit QAdapt (HTNet after sequential T0–T4 + Q-EWC) against Ising-fast trained only under fixed T0. The 110-point OOD grid multiplies exactly the axes that define T1–T4 (measurement, CNOT, idle, Z-bias). Every OOD point is therefore closer to QAdapt’s training distribution than to Ising-fast’s. The matched-T0 architecture win (Table 2: 15–18% LER) does not isolate whether further OOD/Willow deltas come from continual adaptation, from having seen related noise, or from HTNet capacity. A multi-task or jointly trained Ising-fast/HTNet baseline (and/or T0-only HTNet on the same OOD/Willow sets) is required before attributing the 110/110 and zero-shot gains to “noise-adaptive” pre-decoding.
  2. [§4.4, Eq. (9), §7.4] §4.4, Eq. (9), §7.4: The paper’s adaptation narrative rests on Q-EWC mitigating catastrophic forgetting, but there is no controlled comparison to unregularized sequential fine-tuning or to joint mixed-noise training. Without that, one cannot tell whether EWC (λ=100, diagonal Fisher from 65,536 samples) is doing necessary work or whether any multi-task exposure would suffice. Section 7.4 acknowledges this gap; it should be closed for the central claim, not left as future work.
  3. [§6, Tables 2–4] §6 and Tables 2–4: Logical error rates are reported as point estimates with no binomial confidence intervals, shot counts per cell (beyond Willow’s aggregate 400k/100k), or seed variation. For the smaller Willow deltas (e.g., d=7 LER 2.51%; d=5 latency 1.43%) and the mean OOD LER reductions (~3.2%), uncertainty quantification is needed to judge whether improvements are statistically reliable and stable across training seeds.
minor comments (5)
  1. [§7.1, Figure 3] §7.1 correctly notes the absence of module-by-module HTNet ablations (branch fusion, axis–channel gate, raw-evidence skip). These are not strictly required for the adaptation claim once baselines are fixed, but even a short ablation under T0 would strengthen the architecture section.
  2. [§5.5, Abstract] §5.5 / §7.2: Timing scope (residual PyMatching only) is stated clearly; still, a single end-to-end latency breakdown (neural inference + residual construction + matching) in an appendix would prevent over-reading of the “backend efficiency” language in the abstract.
  3. [Figure 5c] Figure 5c: “110/110 wins” is visually strong; consider also reporting mean and worst-case relative ΔLER and whether any wins reverse under CI overlap once uncertainty is added.
  4. [Appendix B, Table 6] App. B Table 6: Task epoch column appears cumulative (20/40/60/80/100) while the text says “20 epochs each”; clarify to avoid confusion about training budget.
  5. [§2] Related work cites concurrent pre-decoder arXiv notes (e.g., Chamberland et al. 2604.12841, QuantiSpect 2607.18204); ensure citation status and differentiation remain accurate at publication.

Circularity Check

0 steps flagged

No significant circularity: empirical pre-decoder comparisons against external baselines and Willow, not definitional or fit-forced predictions.

full rationale

QAdapt is an empirical systems/ML paper. Its load-bearing claims are measured logical error rates and residual PyMatching latencies versus Ising-fast on a synthetic OOD grid and on Google’s external Willow data (zero-shot). Those metrics are not algebraic restatements of the T0–T4 task schedule, the EWC λ, the Fisher sample count, or the OOD multipliers. HTNet’s architecture and Q-EWC are design choices evaluated by experiment, not uniqueness theorems or self-cited forced forms. Citations supporting the pipeline (EWC, PyMatching, Stim, Willow, prior pre-decoding) are external or standard methods, not author-only uniqueness results that close the argument. The high-load subset threshold (cloud-platform density 0.13106) only selects which OOD points to highlight; LER deltas on those points remain independent measurements. Experimental confounds (e.g., Ising-fast trained only on T0 while QAdapt sees T0–T4) affect causal attribution of gains, not circularity of the reported quantities. No self-definitional loop, fitted-input-as-prediction, or renaming of a known identity was found.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on standard QEC/decoding machinery plus several author-chosen modeling and training knobs (noise-task sequence, EWC strength, architecture widths, density anchor). No new physical entities are postulated; the load-bearing leaps are that the synthetic task sequence and residual-matching interface transfer to real nonstationary hardware.

free parameters (6)
  • EWC lambda = 100
    Regularization strength in the Q-EWC loss; set by hand to 100 and used for all sequential tasks.
  • Fisher sample count per task = 65536
    Number of samples used to estimate each diagonal Fisher state after T0–T3.
  • HTNet width and depth hyperparameters = C=112, expand=168, 3 blocks
    Stem channels 112, expanded width 168, three blocks, six joint-convolution groups, receptive field of nine; chosen design knobs that define model capacity.
  • OOD noise multipliers and axis set = 11 axes × 5 multipliers × {d=7,9}
    Multipliers {1.2,1.5,2.0,2.5,3.0} and 11 axis combinations define the entire synthetic evaluation grid.
  • T1–T4 noise scale factor = 1.5× selected T0 parameters
    Task sequence scales selected T0 parameters by 1.5 to create continual-learning tasks.
  • High-load density anchor = 0.13106
    Cloud-platform mean density 0.13106 used to define the high-load subset before method comparison.
axioms (5)
  • domain assumption Rotated surface-code memory under circuit-level Pauli noise is an adequate proxy for the decoding problem of interest.
    All synthetic training and OOD evaluation use Stim-sampled rotated-surface-code memory circuits (Section 5.1).
  • domain assumption Local neural corrections composed with a fixed global MWPM decoder preserve logical consistency while reducing backend work.
    Pipeline definition s_res = s ⊕ H ê and PyMatching on the residual (Eq. 2, Sections 3.1 and 4.5); follows prior local–global pre-decoding.
  • ad hoc to paper Diagonal Fisher EWC with sequential tasks T0–T4 mitigates catastrophic forgetting enough for useful noise adaptation.
    Q-EWC loss (Eq. 9) and task schedule (Table 6) are design choices without a controlled comparison to unregularized fine-tuning in the reported results.
  • domain assumption Syndrome density is a useful common workload indicator even though it is not a complete noise descriptor.
    Stated in Sections 3.1–3.3 and used to motivate the grid and high-load subset.
  • domain assumption Willow open data can be treated as an external zero-shot distribution without target calibration.
    Section 5.4 protocol; improvements are interpreted as transfer beyond the mapped T0 family.
invented entities (2)
  • HTNet no independent evidence
    purpose: Heterogeneous spatiotemporal network that separates spatial, temporal, and joint syndrome correlations with adaptive fusion and axis-channel gating for local corrections.
    Core learned module of QAdapt (Section 4.2–4.3, Figure 3); architectural invention, not a physical entity.
  • Q-EWC task sequence (T0–T4) no independent evidence
    purpose: Hand-specified continual-learning curriculum over base, measurement, CNOT, idle, and Z-bias enhanced noise.
    Defines the adaptation protocol (Section 4.4, Table 6); success is measured only inside this curriculum plus OOD multipliers of the same axes.

pith-pipeline@v1.2.0-daily-grok45 · 17487 in / 3566 out tokens · 57054 ms · 2026-07-31T07:51:40.458855+00:00 · methodology

0 comments
read the original abstract

Fault-tolerant quantum computing (FTQC) relies on quantum error correction to suppress physical errors and preserve logical information at scale. In practice, however, performance is constrained not only by physical noise but also by the latency of classical decoders processing rapidly generated syndrome data. This challenge is exacerbated by hardware noise that is strong, heterogeneous, and nonstationary, as well as by the simulation-to-hardware distribution shift that can substantially degrade fixed neural decoders. We present QAdapt, a noise-adaptive neural pre-decoding framework for surface-code quantum error correction. QAdapt captures local spatiotemporal correlations in syndrome data, sequentially adapts to evolving noise conditions while mitigating catastrophic forgetting, and forwards the residual syndrome to a conventional global decoder. Across 110 synthetic out-of-distribution noise configurations for rotated surface-code memory circuits, QAdapt consistently reduces the logical error rate relative to the neural pre-decoding baseline. On Google's Willow benchmark data, without target-domain fine-tuning, it achieves reductions of up to 5.79 percent in logical error rate and 9.32 percent in backend decoding latency on the residual syndrome. These results demonstrate that QAdapt provides a practical and decoder-compatible approach to improving the robustness and backend decoding efficiency of quantum error correction under evolving hardware noise.

Figures

Figures reproduced from arXiv: 2607.28422 by Ran Miao, Rui Luo, Xiaohan Shan, Xiaoming Sun.

Figure 1
Figure 1. Figure 1: Input syndrome load across hardware references and the selected synthetic OOD grid. (a) Willow and the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: QAdapt workflow. The design connects hardware-informed noise modeling, heterogeneous spatiotemporal [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: HTNet architecture. (a) The detector tensor passes through a 112-channel 3-D stem, three HTNet [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mapped-noise comparison between HTNet and Ising-fast. (a) LER at [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Synthetic OOD evaluation over the five retained noise multipliers. (a) Mean LER over 55 configurations [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Zero-shot transfer to Willow at ten rounds. (a) LER at [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

26 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    V. V. Sivak, A. Eickbusch, B. Royer, S. Singh, I. Tsioutsios, S. Ganjam, A. Miano, B. L. Brock, A. Z. Ding, L. Frunzio, S. M. Girvin, R. J. Schoelkopf, and M. H. Devoret. Real-time quantum error correction beyond break-even.Nature, 616: 50–55, 2023. doi: 10.1038/s41586-023-05782-6

    doi:10.1038/s41586-023-05782-6 2023

  2. [2]

    Yoder, Youngseok Kim, Muyuan Li, Edward H

    Neereja Sundaresan, Theodore J. Yoder, Youngseok Kim, Muyuan Li, Edward H. Chen, Grace Harper, Ted Thor- beck, Andrew W. Cross, Antonio D. Córcoles, and Maika Takita. Demonstrating multi-round subsystem quantum error correction using matching and maximum likelihood decoders.Nature Communications, 14:2852, 2023. doi: 10.1038/s41467-023-38247-5

    doi:10.1038/s41467-023-38247-5 2023

  3. [3]

    Browne, Kenton M

    Luka Skoric, Dan E. Browne, Kenton M. Barnes, Neil I. Gillespie, and Earl T. Campbell. Parallel window decod- ing enables scalable fault tolerant quantum computation. Nature Communications, 14:7040, 2023. doi: 10.1038/ s41467-023-42482-1

    2023

  4. [4]

    Blunt, Archibald Ruban, Jimmy McDaniel, Joseph A

    Laura Caune, Luka Skoric, Nick S. Blunt, Archibald Ruban, Jimmy McDaniel, Joseph A. Valery, Andrew D. Patter- son, Alexander V. Gramolin, Joonas Majaniemi, Kenton M. Barnes, Tomasz Bialas, Okan Buğdaycı, Ophelia Craw- ford, György P. Gehér, Hari Krovi, Elisha Matekole, Can- berk Topal, Stefano Poletto, Michael Bryant, Kalan Snyder, Neil I. Gillespie, Glen...

    Pith/arXiv arXiv doi:10.48550/arxiv.2410.05202 2024

  5. [5]

    Hockings, Andrew C

    Evan T. Hockings, Andrew C. Doherty, and Robin Harper. Improving error suppression with noise-aware decoding. arXiv preprint arXiv:2502.21044, 2025. doi: 10.48550/arXiv. 2502.21044

    Pith/arXiv arXiv doi:10.48550/arxiv.2502.21044 2025

  6. [6]

    Devansh Bhardwaj, Evangelia Takou, Yingjia Lin, and Ken- neth R. Brown. Adaptive estimation of drifting noise in quantum error correction.arXiv preprint arXiv:2511.09491,

    arXiv

  7. [7]

    Topological quantum memory.Journal of Math- ematical Physics, 43(9):4452–4505, 2002

    Eric Dennis, Alexei Kitaev, Andrew Landahl, and John Preskill. Topological quantum memory.Journal of Math- ematical Physics, 43(9):4452–4505, 2002. doi: 10.1063/1. 1499754

    doi:10.1063/1 2002

  8. [8]

    Fowler, Matteo Mariantoni, John M

    Austin G. Fowler, Matteo Mariantoni, John M. Martinis, and Andrew N. Cleland. Surface codes: Towards practical large-scale quantum computation.Physical Review A, 86(3): 032324, Sep 2012. doi: 10.1103/PhysRevA.86.032324

    doi:10.1103/physreva.86.032324 2012

  9. [9]

    Barbara M. Terhal. Quantum error correction for quantum memories.Reviews of Modern Physics, 87(2):307–346, Apr

  10. [10]

    Suppressing quantum errors by scaling a surface code logical qubit.Nature, 614:676–681, 2023

    Google Quantum AI. Suppressing quantum errors by scaling a surface code logical qubit.Nature, 614:676–681, 2023. doi: 10.1038/s41586-022-05434-1

    doi:10.1038/s41586-022-05434-1 2023

  11. [11]

    Quantum error correction below the surface code threshold.Nature, 638: 920–926, 2025

    Google Quantum AI and Collaborators. Quantum error correction below the surface code threshold.Nature, 638: 920–926, 2025. doi: 10.1038/s41586-024-08449-y

    doi:10.1038/s41586-024-08449-y 2025

  12. [12]

    PyMatching: A Python package for decod- ing quantum codes with minimum-weight perfect matching

    Oscar Higgott. PyMatching: A Python package for decod- ing quantum codes with minimum-weight perfect matching. arXiv preprint arXiv:2105.13082, 2021. doi: 10.48550/arXiv. 2105.13082

    Pith/arXiv arXiv doi:10.48550/arxiv.2105.13082 2021

  13. [13]

    Crespo, and Josu Etxezarreta Martinez

    Antonio deMarti iOlius, Patricio Fuentes, Román Orús, Pe- dro M. Crespo, and Josu Etxezarreta Martinez. Decoding algorithms for surface codes.Quantum, 8:1498, 2024. doi: 10.22331/q-2024-10-10-1498

    doi:10.22331/q-2024-10-10-1498 2024

  14. [14]

    Nicolas Delfosse, Andres Paz, Alexander Vaschillo, and Krysta M. Svore. How to choose a decoder for a fault- tolerant quantum computer? the speed vs accuracy trade-off. arXiv preprint arXiv:2310.15313, 2023. doi: 10.48550/arXiv. 2310.15313

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.15313 2023

  15. [15]

    Techniques for combining fast local decoders with global decoders un- der circuit-level noise.Quantum Science and Technology, 8: 045011, 2023

    Christopher Chamberland, Luis Goncalves, Prasahnt Sivara- jah, Eric Peterson, and Sebastian Grimberg. Techniques for combining fast local decoders with global decoders un- der circuit-level noise.Quantum Science and Technology, 8: 045011, 2023. doi: 10.1088/2058-9565/ace64d

    work page doi:10.1088/2058-9565/ace64d 2023

  16. [16]

    Fast and accurate AI-based pre- decoders for surface codes.arXiv preprint arXiv:2604.12841,

    Christopher Chamberland, Jan Olle, Muyuan Li, Scott Thornton, and Igor Baratta. Fast and accurate AI-based pre- decoders for surface codes.arXiv preprint arXiv:2604.12841,

    Pith/arXiv arXiv

  17. [17]

    Sparse blossom: Cor- recting a million errors per core second with minimum- weight matching.Quantum, 9:1600, 2025

    Oscar Higgott and Craig Gidney. Sparse blossom: Cor- recting a million errors per core second with minimum- weight matching.Quantum, 9:1600, 2025. doi: 10.22331/ q-2025-01-20-1600

    2025

  18. [18]

    Varbanov, Marc Serra-Peralta, David Byfield, and Barbara M

    Boris M. Varbanov, Marc Serra-Peralta, David Byfield, and Barbara M. Terhal. Neural network decoder for near-term surface-code experiments.arXiv preprint arXiv:2307.03280,

    Pith/arXiv arXiv

  19. [19]

    Senior, Francisco J

    Johannes Bausch, Andrew W. Senior, Francisco J. H. Heras, Thomas Edlich, Alex Davies, Michael Newman, Cody Jones, Kevin Satzinger, Murphy Yuezhen Niu, Sam Blackwell, George Holland, Dvir Kafri, Juan Atalaya, Craig Gid- ney, Demis Hassabis, Sergio Boixo, Hartmut Neven, and Pushmeet Kohli. Learning high-accuracy error decoding for quantum processors.Nature,...

    doi:10.1038/s41586-024-08148-8 2024

  20. [20]

    QuantiSpect: A Structure-Aware Lightweight 3D CNN Pre-Decoder for Scalable Surface Code Quantum Error Correction

    Pan Gao, Xu-Sheng Xu, Ji-Ze Han, Jing-Wei Wen, Ling Qian, Xu-Dong Lv, Run-Qing Zhang, Xiao-Xiao Hu, and Gui-Lu Long. QuantiSpect: A structure-aware lightweight 3d CNN pre-decoder for scalable surface code quantum error correction.arXiv preprint arXiv:2607.18204, 2026. doi: 10.48550/arXiv.2607.18204

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2607.18204 2026

  21. [21]

    Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, Demis Hassabis, Claudia Clopath, Dharshan Kumaran, and Raia Hadsell

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A. Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, Demis Hassabis, Claudia Clopath, Dharshan Kumaran, and Raia Hadsell. Overcoming catastrophic forgetting in neural networks.Proceedings of the National Academy of Sciences, 114(13):3521–...

    doi:10.1073/pnas.1611835114 2017

  22. [22]

    Stim: A fast stabilizer circuit simulator

    Craig Gidney. Stim: A fast stabilizer circuit simulator. Quantum, 5:497, 2021. doi: 10.22331/q-2021-07-06-497. 11

    doi:10.22331/q-2021-07-06-497 2021

  23. [2015]

    doi: 10.1103/RevModPhys.87.307

    doi:10.1103/revmodphys.87.307

  24. [2023]

    doi: 10.48550/arXiv.2307.03280

    Pith/arXiv arXiv doi:10.48550/arxiv.2307.03280

  25. [2025]

    doi: 10.48550/arXiv.2511.09491

    doi:10.48550/arxiv.2511.09491

  26. [2026]

    doi: 10.48550/arXiv.2604.12841

    Pith/arXiv arXiv doi:10.48550/arxiv.2604.12841