{"id":"1e8f514a-a2ac-4706-92fa-91068ce86b51","arxiv_id":"2608.10081","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Sliding-window decoding of topological quantum codes is effectively described by a parity-conserving reaction-diffusion process in which the decoding rate 1/W acts as a relevant perturbation, yielding exponential memory-time scaling in the window size.","lead":"This paper proposes that sliding-window quantum error correction behaves at large scales like a simple reaction-diffusion process of Z2 charges, with memory time growing exponentially in the window size. It gives a theoretical benchmark for the tradeoff between decoding speed and accuracy in real-time quantum error correction.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The main risk is Eq. (3): the DPRM step-size assumption is plausible but only indirectly tested, and it is less critical for the memory-time crossover. A direct single-window displacement measurement would settle it.","rationale":"The reader identified the DPRM mapping as the weakest assumption; we agree with that identification but only partially agree with the stated consequence, since the W/L crossover collapse for the memory time is controlled by the activated nucleation rate τ (Eq. 2), not by the diffusion constant D. The paper's evidence for D(W) is real but indirect: the long-time scaling collapse in Fig. 2(a) is consistent with D ∝ W^(1/3), yet it does not directly measure the single-window step statistics, and footnote [39] concedes that slow charges are not directly observable in the numerics. The 2D toric-code diffusion scaling is explicitly untested. A direct single-window displacement measurement would validate or falsify Eq. (3) without ambiguity. Because the central 1/W-relevant perturbation claim and the exponential memory-time scaling are supported by independent numerical collapses (Fig. 1c and End Matter) and do not hinge on the DPRM step, the ACCEPT verdict remains appropriate; if the direct test failed, the parameter scaling in Eq. (3) would need revision while the main crossover result would likely survive.","tokens_in":14158,"tokens_out":24572,"duration_ms":265928,"concrete_test":"Numerically measure the single-round displacement distribution of a coarse-grained domain wall: initialize one domain wall at x=L/2 in the 1D repetition code (open boundaries, same noise model as Fig. 2), run SWD for exactly one commit window, read out the new wall position from the cumulative spin configuration, and record δr = x_after − x_before for many realizations and several W. Check whether Var(δr)^(1/2) ∝ W^(2/3) and whether successive-round displacements are uncorrelated (Cov(δr_n, δr_{n+1}) ≈ 0). If Var(δr)^(1/2) ∝ W^(1/2) or W, Eq. (3) fails; if displacements are correlated, the random-walk reduction of the kinetic model needs revision. The same procedure on the toric code would require tracking an anyon through the decoder's projected correction chains, but the 1D test is the minimal decisive check.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (3) sets D(W) ∝ W^{zζ−1} by assuming that in each commit window a slow charge's endpoint displacement is a DPRM step, |δr| ∝ W^ζ with ζ=2/3, and that successive steps are independent. Two things make this the weakest part of the central claim. First, the only published test (Fig. 2a) is an indirect long-time diffusion collapse; footnote [39] explicitly admits that slow charges are “always clouded by a background of fast charges, and are not easily distinguishable”, so no direct measurement of the single-window displacement distribution is presented. Second, the decoder's MWPM matching is a global optimization, and it is an assumption—not a derivation—that the resulting disorder seen by a coarse-grained charge is in the quenched DPRM universality class rather than, say, annealed or correlated from round to round. If ζ were 1/2 or steps were correlated, Eq. (3) would fail. Note, however, that the memory-time scaling Eq. (5) and the relevance of 1/W are governed by the activated rate τ (Eq. 2) and do not require the DPRM assumption, so the headline conclusion is less exposed. The 2D toric-code D(W) is explicitly untested in the End Matter.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an effective kinetic description of sliding-window decoding (SWD) for topological codes with deconfined Z2 point-like defects. The authors argue that, at length and time scales large compared to the window size W, SWD is equivalent to a parity-conserving reaction-diffusion process of coarse-grained charges, Eq. (1). They derive scaling relations for the pair-nucleation rate, tau(W) proportional to exp(-kappa sigma W), for the diffusion constant, D(W) proportional to W^{z zeta - 1}, and for the memory time, ln t_mem proportional to L Phi(W/L), with Phi linear for W << L and saturating for W >> L. These predictions are tested numerically for the 1D repetition code, the 2D toric code, modular SWD, and union-find/clustering decoders. The End Matter and Supplemental Material also provide tests using quasi-local decoder variants and a SWAP-noise model exhibiting an absorbing-state transition. The central claim is that the decoding rate 1/W is a relevant perturbation to the decodable phase, so that logical failure is exponentially suppressed in W, but not in L for any finite W.","tokens_in":14461,"tokens_out":5392,"duration_ms":53920,"significance":"If correct, this is a significant step: it provides the first concrete theoretical framework for real-time quantum error correction beyond the static equilibrium mapping, and it identifies W/L as the natural figure of merit for the speed-accuracy tradeoff. The paper's strengths are its extensive numerical evidence, its falsifiable scaling predictions, and the deliberate exploration of many microscopic variants (modular SWD, UF/clustering decoders, SWAP noise) that are all consistent with the same effective description. The authors are also commendably explicit about the limitations of their evidence, for example in footnote [39] and in the End Matter statement that the 2D toric-code diffusion scaling has not been directly tested. I see no internal inconsistency in the central derivation; the main risk is a physics assumption that is plausible but only indirectly tested.","major_comments":[{"comment":"The scaling D(W) proportional to W^{z zeta - 1} rests on the assumption that, in each commit window, a slow charge's endpoint displacement is an independent DPRM step with |delta r| proportional to W^zeta and zeta = 2/3, and that successive steps are independent. This assumption is plausible, but it is not derived from the MWPM matching dynamics, and footnote [39] explicitly states that slow charges are 'not easily distinguishable' from the fast background, so Fig. 2(a) is only an indirect long-time diffusion collapse. If the decoder-generated disorder were annealed or correlated from round to round, or if zeta were 1/2 instead of 2/3, Eq. (3) and the Fig. 2(a) interpretation would fail. I ask the authors to provide a direct measurement of the single-window displacement distribution for an isolated slow charge, or to clearly label Eq. (3) as a conjecture and soften the claims that depend on it. The memory-time crossover Eq. (5) is not affected by this concern because it depends on tau rather than on D.","section":"Effective model for SWD, Eq. (3) and footnote [39]"},{"comment":"The abstract claims broad applicability of the effective description, but for the 2D toric code the paper only tests the memory-time crossover in Fig. 5(b), and the End Matter explicitly states that the diffusion scaling has not been tested because no local observable tracks a coarse-grained anyon in a way analogous to the 1D spin-density Delta-rho. Since the diffusion exponent enters the claimed universal D(W) scaling in 2D, and the DPRM exponent is dimension-dependent (zeta for (2+1)-dimensional directed polymers), the 2D universality of Eq. (3) is currently unsupported. I recommend either supplying a targeted numerical test that isolates slow anyon motion (for example through a suitable two-point correlation function or a nonlocal filter), or explicitly labeling the 2D diffusion scaling as an open prediction rather than a tested consequence of the theory.","section":"End Matter, 'Crossover scaling for the 2D toric code'"}],"minor_comments":[{"comment":"The affiliation line for the University of Washington reads 'W A' and should be 'WA'.","section":"Title page affiliations"},{"comment":"The quantity Gamma_0 appears in Eq. (6) before it is introduced; consider defining it in the text immediately before the equation.","section":"Numerical results, Eq. (6)"},{"comment":"The caption of Fig. 2 could state explicitly that <Delta-rho> is the ensemble-averaged absolute spin-density difference, since the notation is otherwise only defined in the main text.","section":"Fig. 2"},{"comment":"The manuscript does not include a data or code availability statement; given the extensive Monte Carlo results, a note on availability of simulation code would be helpful for reproducibility.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"I largely agree with the reader's positive assessment of the core physics and with the skepticism about Eq. (3). The paper is strong and the central memory-time result is robust, but the DPRM step-size assumption is a load-bearing point for the diffusion scaling, and the 2D toric-code diffusion prediction is explicitly untested. A direct single-window displacement measurement would substantially strengthen the paper; absent that, the authors should present Eq. (3) as a conjecture and adjust the corresponding claims. This is fixable within the manuscript's scope, so I recommend major revision rather than acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me be direct: this is the first paper I know that gives an actual theoretical handle on sliding-window decoding. The central claim - that SWD at scales larger than W behaves as parity-conserving reaction-diffusion, with 1/W a relevant perturbation - is well argued, and the memory-time crossover in Eq. (5) is convincing. The numerical support is extensive: repetition code, toric code, modular SWD, union-find and clustering decoders, all collapse onto the expected curve. The paper also earns credit for flagging its own open ends, including the critical regime and the untested 2D diffusion. The real soft spot is Eq. (3), the DPRM step-size assumption. The stress-test note has this right: the only test is the indirect long-time diffusion collapse, and footnote [39] admits slow charges are not cleanly separable from fast background. So we don't have direct evidence on the single-window displacement distribution, and the MWPM global matching could in principle produce disorder that is not quenched DPRM. If zeta were different or steps correlated, Eq. (3) fails. For the headline results, though, this matters less: Eq. (5) and the activated memory time ride on tau (Eq. 2), not on D. That gives me confidence the main conclusion is not hanging by this thread. A couple of smaller notes. No code or data is shipped; that is a minor annoyance in a paper this numerical. The 2D toric-code diffusion is explicitly untested, which is fine but worth remembering when people cite the W^{1/3} law. The near-threshold regime is left open, and the paper says so. The citation pattern is clean, with prior SWD work properly distinguished from the new contribution. Who should read this? Anyone working on real-time decoding or on non-equilibrium descriptions of QEC. I would also point a student at it as a model of how to combine an effective theory with honest numerical checks. It deserves a serious referee, and I'd send it out.","headline":"A strong theory paper that finally gives sliding-window decoding a real effective description; the memory-time crossover is convincing, while the DPRM-based diffusion scaling is the part I'd push on.","tokens_in":705,"tokens_out":869,"would_cite":true,"duration_ms":28723,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Sliding-window quantum error correction is governed by a parity-conserving reaction-diffusion process.","keywords":["sliding-window decoding","real-time quantum error correction","reaction-diffusion process","parity-conserving dynamics","memory time","directed polymers in random media","topological codes","Z2 charges"],"falsifier":"Initialize a single domain wall in the 1D repetition code, run SWD with open boundaries at fixed $p<p_c$ and $W \\ll L^{2/3}$, and measure how the domain wall spreads. The paper predicts effective diffusion $D(W) \\propto W^{1/3}$, so a log-log plot of $D$ versus $W$ should have slope $1/3$; any other slope falsifies the directed-polymer assumption.","tokens_in":13991,"feed_emoji":"⚛️","tokens_out":12865,"duration_ms":104833,"temperature":0.7,"pith_summary":"Sliding-window decoding (SWD) is a practical real-time error-correction scheme in which a decoder repeatedly sees only a finite temporal window $W$ of syndrome data and irreversibly commits to corrections. This paper argues that despite the decoder's global, nonlocal matching, the long-time dynamics of the code is captured by a simple stochastic process: $\\mathbb{Z}_2$-charged point particles that diffuse, annihilate, and nucleate in pairs. The memory time of the code—how long logical information survives—then obeys $\\ln t_{\\mathrm{mem}} \\propto L\\,\\Phi(W/L)$, interpolating between exponential growth in $W$ for small windows and the static limit $L$ for huge windows. If correct, this gives a theoretical benchmark for real-time decoders and identifies the decoding rate $1/W$ as a relevant perturbation that prevents true memory at any finite window size.","feed_headline":"Sliding-window decoding obeys reaction-diffusion kinetics","feed_subtitle":"A single scaling curve links real-time and static decoding: the window size W sets how long logical memory survives.","key_machinery":"The central object is the worldline of a $\\mathbb{Z}_2$ charge in the combined error-and-correction chain $E\\oplus E'$; the window $W$ screens decoder corrections to length $O(W)$, making charge motion quasi-local. The key scaling input is the treatment of each window's worldline as a directed polymer in a random medium: a segment of temporal extent $W$ wanders spatially by $|\\delta r| \\propto W^{\\zeta}$, so a charge takes steps of that size every $W$ time units, giving $D(W) \\propto W^{z\\zeta-1}$ with $z=2$ and $\\zeta=2/3$ in (1+1) dimensions. A standard energy-counting argument for a pair to separate a distance $O(W)$ inside one window yields $\\tau \\propto e^{-\\kappa\\sigma W}$. These two relations feed the reaction-diffusion process, whose relaxation time sets the memory time and the crossover function $\\Phi$.","core_discovery":"The paper's central claim is that SWD, at time and length scales larger than the window size $W$, admits an effective description as a parity-conserving reaction-diffusion process $A+A \\rightleftharpoons \\varnothing$ with diffusion. The effective parameters inherit nontrivial $W$-dependence: pair nucleation is exponentially suppressed, $\\tau(W) \\propto e^{-\\kappa\\sigma W}$; the effective diffusion constant grows as $D(W) \\propto W^{z\\zeta-1}$; and annihilation is essentially immediate upon contact. The relaxation time of this process is identified with the memory time, giving $\\ln t_{\\mathrm{mem}} \\propto L\\,\\Phi(W/L)$ with $\\Phi$ linear for $W\\ll L$ and saturating for $W\\gg L$. The authors support this with numerical data collapse for the 1D repetition code and the 2D toric code, and with tests using alternative decoders.","pith_inferences":["A testable extension: if different decoders bias the disorder seen by worldlines differently, the wandering exponent $\\zeta$ would change; measuring the $W$-exponent of the effective diffusion constant across decoders would test whether the directed-polymer universality assumption is truly universal.","The relevant-perturbation picture suggests a design rule: persistent logical memory in real-time decoding requires making the decoding rate irrelevant, which is exactly what single-shot decoders aim to do; SWD shows the generic cost of not doing so.","The same machinery should apply to more general excitations: non-abelian or kinetically constrained charges would likely fall in different non-equilibrium universality classes, and the scaling-collapse method used here could detect them."],"forward_implications":["For any finite window $W$, the memory time no longer diverges with system size: it grows as $e^{\\alpha\\kappa\\sigma W}$ in the $W\\ll L$ regime and saturates to the static $e^{\\sigma L}$ form as $W\\gg L$, so finite-rate decoding always caps logical memory.","The single-variable scaling form $\\ln t_{\\mathrm{mem}} \\propto L\\,\\Phi(W/L)$ makes $W/L$ the natural figure of merit for the speed-accuracy tradeoff, usable for error budgeting under throughput constraints.","The effective description is unchanged across microscopic details: the same scaling collapse holds for modular sliding-window decoding, union-find and clustering decoders, and for the 2D toric code with point-like anyons.","Because $1/W$ is a relevant perturbation to the decodable phase, SWD is not a single-shot decoder for codes with point-like excitations: no matter how large $W$, the logical failure rate per round remains nonzero.","In one spatial dimension the relaxation exponent $\\alpha=1$ makes the memory time grow as $e^{\\alpha\\kappa\\sigma W}$, so each additional unit of window size exponentially suppresses logical failure."],"supporting_citations":[{"why":"Supplies the MWPM decoder and the equilibrium statistical-mechanics mapping whose line tension $\\sigma(p)$ enters the energy-counting nucleation estimate.","marker":"[12]"},{"why":"Supplies the parity-conserving reaction-diffusion universality class $A+A\\rightleftharpoons\\varnothing$ used as the effective model.","marker":"[28]"},{"why":"Bases the wandering exponent $\\zeta=2/3$ in (1+1)D on directed polymers in random media, yielding $D\\propto W^{z\\zeta-1}$.","marker":"[31]"},{"why":"Gives the relaxation-time exponent $\\alpha=1$ in one spatial dimension, connecting the nucleation rate $\\tau$ to the memory time.","marker":"[38]"},{"why":"Union-find decoder used to show the crossover scaling survives a change of decoder.","marker":"[55]"},{"why":"Clustering decoder used as an additional test that the effective reaction-diffusion description is decoder-independent.","marker":"[56]"},{"why":"Supplies the DP2 universality-class exponents used in the SWAP-noise absorbing-state phase transition test.","marker":"[60]"}],"fun_headline_variants":["Sliding-window decoding as reaction-diffusion process","Real-time QEC kinetics: window size matters","Decoding rate tunes quantum memory lifetime","Universal kinetics for real-time error correction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole scaling picture rests on the assumption that the way a charge wanders across one decoding window has the same statistics as a directed polymer in a random medium, with a wandering exponent $\\zeta=2/3$ in one spatial dimension; if the decoder-generated disorder does not belong to that universality class, the predicted diffusion constant $D \\propto W^{1/3}$ and the $W/L$ data collapse would fail.","fun_headline_variants_meta":{"raw":{"variants":["Sliding-window decoding as reaction-diffusion process","Real-time QEC kinetics: window size matters","Decoding rate tunes quantum memory lifetime","Universal kinetics for real-time error correction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000535,"raw_usage":{"total_tokens":2569,"prompt_tokens":938,"completion_tokens":1631,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":1574}},"tokens_in":554,"tokens_out":1631,"duration_ms":14875,"temperature":1.0,"reasoning_tokens":1574,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:14:48.082592+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Initialize a single domain wall in the 1D repetition code, run SWD with open boundaries at fixed $p<p_c$ and $W \\ll L^{2/3}$, and measure how the domain wall spreads. The paper predicts effective diffusion $D(W) \\propto W^{1/3}$, so a log-log plot of $D$ versus $W$ should have slope $1/3$; any other slope falsifies the directed-polymer assumption.","supporting_citations":[{"cited_title":"Field Theory of Branching and Annihilating Random Walks","cited_arxiv_id":"cond-mat/9704160","evidence_quote":"Supplies the parity-conserving reaction-diffusion universality class $A+A\\rightleftharpoons\\varnothing$ used as the effective model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Bases the wandering exponent $\\zeta=2/3$ in (1+1)D on directed polymers in random media, yielding $D\\propto W^{z\\zeta-1}$."},{"cited_title":"Rácz, Phys","cited_arxiv_id":null,"evidence_quote":"Gives the relaxation-time exponent $\\alpha=1$ in one spatial dimension, connecting the nucleation rate $\\tau$ to the memory time."},{"cited_title":"Fast Decoders for Topological Quantum Codes","cited_arxiv_id":"0911.0581","evidence_quote":"Clustering decoder used as an additional test that the effective reaction-diffusion description is decoder-independent."},{"cited_title":"range of interaction","cited_arxiv_id":null,"evidence_quote":"Supplies the DP2 universality-class exponents used in the SWAP-noise absorbing-state phase transition test."}],"review_version":1}