{"id":"0e447892-9848-4152-8f6c-107b61661ab5","arxiv_id":"2506.19538","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A constrained quantum-enhanced MCMC algorithm with an approximate Benincasa-Dowker Hamiltonian samples causal sets, claiming a cubic scaling advantage in spectral gap on small instances.","lead":"This paper builds a quantum-enhanced Markov chain Monte Carlo algorithm to sample causal sets, the discrete structures behind Causal Set Theory, adding a constraint Hamiltonian and a qubit Hamiltonian for the Benincasa-Dowker action. It reports a roughly cubic thermalization slowdown improvement in simulations of very small causal sets.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Uniform-sampling spectral gaps are combined across underspecified parameter samples; the cubic advantage may be an artifact of parameter selection on very small N.","rationale":"The reader's weakest_assumption points to the inconsistency between Eq. (7) and Eq. (G2) for the BD Hamiltonian, and that is a genuine defect for the weighted-sampling algorithm. However, the strongest claim in the abstract and conclusion is the cubic uniform-sampling advantage, and that claim depends on the uniform-sampling spectral-gap data, not on the BD Hamiltonian. The more direct threat to that central claim is the underspecified procedure by which the quantum proposals' parameters are sampled and combined. If the reported spectral gap is the best over a small parameter grid, the factor-of-three reduction in k is not a property of a single algorithm. If it is a mixture over parameters, the transition matrix must be defined explicitly and the comparison with fixed classical moves made fair. Both cases are testable by re-running the same simulations with per-parameter spectral gaps and by extending the cardinality range. The finite-size caveat is acknowledged by the authors, but the paper does not state how many N values enter the fits or how sensitive k_Q is to the smallest points. This concern does not by itself overturn the paper; it reinforces the conditional verdict, because the central scaling claim needs a clearer and more robust numerical basis before it can be accepted as stated.","tokens_in":13972,"tokens_out":5311,"duration_ms":60274,"concrete_test":"Recompute Figure 1 separately for each listed parameter combination (r_TC in {0.7,0.9}, t in {3,10}) at each N, without pooling; fit delta proportional to exp(-kN) for each combination and report best/median/worst k_Q. If the best-case k_Q is used in the paper and the median is >= 0.3, or if extending N by one point changes k_Q by more than the quoted error, the cubic advantage is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline claim is the cubic uniform-sampling advantage (k_Q = 0.17(5) vs k_C = 0.51(4), Fig. 1). The most load-bearing weakness is not the BD Hamiltonian but the way the quantum spectral gaps are produced. The text says that for the quantum algorithms 'we sample the parameters' (r_TC, r_BD, t) and then 'combining these samples, the transition matrix and thus spectral gap can be calculated,' with jackknife error. This is not a well-defined Markov chain unless the proposal distribution at every step is the same fixed mixture over parameter values; if instead the reported delta is the best over sampled parameters, the advantage is overstated, and if it is an average, the comparison with fixed classical moves needs a precise transition-matrix definition. Only 2x2 parameter choices are listed for uniform sampling (r_TC in {0.7,0.9}, t in {3,10}), so the fit delta proportional to exp(-kN) rests on very few N values and on a parameter-combining rule that is underspecified. The paper itself concedes simulations are limited to very small cardinality; an exponential fit to a handful of points with two free parameters can produce k = 0.17 with small quoted error while being non-robust. The Eq. (7)/(G2) inconsistency noted by the reader is real but affects the weighted-sampling section, whose advantage is not quantified; the central abstract/conclusion claim is the uniform-sampling exponent.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper adapts Quantum-enhanced Markov Chain Monte Carlo (QeMCMC) to sample causal sets of fixed cardinality, both uniformly and weighted by an approximate 4d Benincasa–Dowker (BD) action. It introduces a transitive-closure penalty Hamiltonian to enforce causal-set validity and derives a qubit Hamiltonian intended to represent the BD action. Spectral-gap simulations compare the quantum proposal against classical move sets from the literature. The headline result is an exponential thermalization-time exponent k_Q = 0.17(5) for the quantum uniform-sampling proposal versus k_C = 0.51(4) for the classical algorithm, described as a cubic scaling advantage. Weighted-sampling results are presented but not quantified as a scaling advantage.","tokens_in":14186,"tokens_out":7253,"duration_ms":73670,"significance":"If the uniform-sampling exponent comparison survives scrutiny, this is a notable empirical extension of QeMCMC from unconstrained Ising-type problems to a constrained combinatorial space of direct physical interest. The paper makes concrete, reproducible contributions: an open-source implementation, a constraint Hamiltonian for causal sets, and a first qubit Hamiltonian approximating the BD action. It also states its limitations honestly, including the quadratic qubit requirement and the very small system sizes. However, the central quantitative claim rests on a small number of low-cardinality simulations and on a parameter-combining rule that is not fully specified. In addition, the BD Hamiltonian used in the weighted-sampling section has an internal inconsistency between the main-text expression and the Pauli decomposition in the appendix, so the weighted-sampling results cannot currently be interpreted as sampling the claimed distribution.","major_comments":[{"comment":"The reported quantum spectral gaps are obtained by sampling the algorithmic parameters (r_TC and t) and then \"combining these samples\" to compute the transition matrix and spectral gap. This procedure is not defined precisely enough to support the central claim. If different proposals in the Markov chain are generated with different parameter values, the chain is time-inhomogeneous and its spectral gap is not well-defined; if instead the reported gap is an average over separately computed gaps, or a best value over the sampled parameters, the comparison with the fixed classical chains is biased. The manuscript must specify exactly how the parameter samples enter a single transition matrix (for example, as a fixed mixture over proposals) or report the full distribution of gaps over the parameter grid. Without this, the quoted k_Q = 0.17(5) and the cubic-advantage claim are not robust.","section":"Simulated Results; Appendix G"},{"comment":"The exponential fit delta proportional to exp(-kN) is performed on \"very small cardinality\" simulations, but the actual values of N, the number of independent parameter samples, and the number of fit points are not stated. With only two values of r_TC and two values of t listed in Appendix G, the jackknife error bars reflect sampling over these few points, not the uncertainty of the exponential fit. Please report the N values used, the fit residuals, and the sensitivity of k_Q to excluding each data point. As presented, the factor-of-three reduction in k could be an artifact of the fit range and the parameter-combining rule.","section":"Fig. 1; Simulated Results"},{"comment":"The qubit Hamiltonian used in the weighted simulations is not the Pauli representation of Eq. (7). Eq. (7) has a leading N term proportional to epsilon^{-1/2} and a linear-in-C coefficient proportional to epsilon^{1/2}, while Eq. (G2) has an N coefficient proportional to epsilon^{1/2} and a linear coefficient proportional to epsilon^{3/2}; the cubic coefficients differ by an additional power of epsilon and by a numerical factor. As written, the two expressions describe different Hamiltonians, so the weighted-sampling simulations may be sampling from a distribution that is not the intended BD partition function. The derivation in Appendix A should be reconciled with the implementation, and the weighted-sampling claims in the conclusion should be conditioned on the corrected Hamiltonian.","section":"Eq. (7) versus Eq. (G2); Appendix G"},{"comment":"The statement that \"both algorithms exhibit a super-quadratic scaling\" is stronger than the quantified results. For uniform sampling, the ratio k_C/k_Q is about 3, which is cubic. For weighted sampling, Fig. 2 gives k_Q = 2.09(8) versus k_C = 4.3(4), a ratio of about 2.1, and the text explicitly declines to quantify a weighted-sampling advantage because of temperature dependence. Please state precisely which claims are quantified and avoid attributing a super-quadratic advantage to the weighted algorithm unless the corrected Hamiltonian yields a stable ratio strictly below 1/2.","section":"Conclusion; Fig. 2"}],"minor_comments":[{"comment":"In the sentence defining the sampled variables, gamma_BD is assigned twice: the text gives gamma_BD = (1 - r_TC) r_BD and then gamma_BD = (1 - r_TC)(1 - r_BD); the second assignment should presumably be gamma_mix = (1 - r_TC)(1 - r_BD).","section":"Appendix G"},{"comment":"The captions of Figs. 3 and 4 appear to swap the dimension labels: Fig. 3 is described in the text as the 2d BD action case but the caption says \"for 4d BD action\" with d = 2, and Fig. 4 is described as the 4d case but the caption says \"for 2d BD action\" with d = 4.","section":"Appendix E, Figs. 3 and 4"},{"comment":"The term \"super-quadratic\" is used without a definition; the text should state the precise criterion, for example k_Q < k_C/2, at first use.","section":"Simulated Results"},{"comment":"The prefactor 4*sqrt(6)/sqrt(epsilon) in Eq. (7) diverges as epsilon goes to 0, while the text states that the epsilon much less than 1 regime is the relevant one; this suggests a typo in the placement of epsilon, which should be corrected together with Eq. (G2).","section":"Eq. (7)"}],"recommendation":"major_revision","confidential_remarks":"I have no conflict of interest. The paper is within the scope of the journal and is honest about its limitations, with code provided. The main risks are the underspecified parameter-combining rule in the quantum spectral-gap computation and the inconsistency between Eq. (7) and Eq. (G2). If the authors can clarify the transition-matrix construction, re-analyze the uniform-sampling fit, and repair the weighted-sampling Hamiltonian, the work could become publishable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"X, here's my read of 2506.19538. The genuinely new pieces are worth taking seriously: extending QeMCMC to constrained spaces with a transitive-closure penalty, and deriving a qubit Hamiltonian for the Benincasa-Dowker action. Those are real, reusable contributions, and the paper is honest that it has no analytical proof of advantage, non-optimal parameters, and quadratic qubit cost. The code is on GitHub, though without a commit hash.\n\nThe problem is that the headline claim—cubic quantum scaling advantage from spectral-gap fits—rests on very thin ice. The fits use tiny cardinalities, few parameter samples, and a parameter-combining rule that is underspecified. The text says they sample parameters and then combine samples to build the transition matrix and gap, with jackknife error. That is not a well-defined Markov chain unless the proposal is a fixed mixture; if it's the best over samples, the advantage is overstated. I don't see that resolved in the paper. With only a handful of N values, an exponential fit with two free parameters can easily produce stable-looking k with overlapping errors. The stress test is right: this is the load-bearing weakness.\n\nSecond soft spot: the weighted-sampling Hamiltonian. Eq. (7) and Eq. (G2) have inconsistent prefactors in ε (one implies ε^{1/2}/ε^{3/2}, the other ε^{3/2}/ε^{5/2}, plus a √6 discrepancy). The authors don't flag this. Since the weighted section is not quantified anyway, it doesn't destroy the paper, but it means the weighted simulations may be sampling a different distribution than claimed, and the appendix comparison of classical vs quantum at low T is suspect until re-derived.\n\nWhat holds up: the uniform sampling algorithm itself is sound in construction, detailed balance and ergodicity arguments are fine, and the constrained QeMCMC idea is a step beyond Layden. If the parameter-combining rule is clarified and the fits re-done with a fixed mixture or robust statistics, the uniform result might survive at a weaker level. The BD Hamiltonian derivation, once fixed, is a useful tool.\n\nWho is this for? People working on quantum algorithms for discrete gravity, and QeMCMC practitioners. It deserves a serious referee—the ideas are substantive—but it needs major revision before publication. I would not cite it as is. Bring to reading group? Maybe, as a case study in fragile scaling claims.","headline":"New constrained-QeMCMC ingredients are real, but the cubic speedup claim rests on underspecified spectral-gap fits at tiny N; worth refereeing after major revision.","tokens_in":14774,"tokens_out":1762,"would_cite":false,"duration_ms":17979,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that two quantum-enhanced Markov-chain algorithms sample causal-set spacetimes with a threefold smaller slowdown exponent than classical chains, one of them weighting samples by the Benincasa-Dowker action.","keywords":["causal set theory","quantum-enhanced Markov chain Monte Carlo","Benincasa-Dowker action","quantum gravity","spectral gap","transitive closure constraint","quantum sampling algorithms","thermalization scaling"],"falsifier":"Compute, on a fixed small ensemble of causal sets, both $H_{\\mathrm{BD}\\varepsilon}$ from Eq. (G2) and the action $S^{(4)}_\\varepsilon$ from Eq. (2), and check whether the Metropolis acceptance probabilities $\\exp(-\\beta\\Delta S)$ match those built from the Hamiltonian actually simulated; the mismatched prefactors mean the two will disagree at $\\varepsilon = 0.1$, which would show the weighted benchmark sampled a different action. Independently, re-run the spectral-gap fits with the coefficient-matched Hamiltonian and check whether $k_Q \\approx k_C/3$ survives; if the ratio changes, the cubic claim falls with it. A third check targets the uniform sampler: obtain proposal statistics at larger $N$ through coarse-graining or quantum-inspired classical simulation and test whether $\\delta \\propto e^{-0.17N}$ still fits.","tokens_in":13674,"feed_emoji":"⚛️","tokens_out":22356,"duration_ms":187944,"temperature":0.7,"pith_summary":"Quantum computing could speed up the study of causal sets, a discrete picture of spacetime built from partially ordered events. The paper adapts quantum-enhanced Markov chain Monte Carlo to this setting: it adds a penalty term that keeps quantum proposals inside the space of valid causal sets, and it derives a qubit Hamiltonian encoding the Benincasa-Dowker action, the causal-set counterpart of the Einstein-Hilbert action. The central empirical claim is a super-quadratic scaling advantage: for uniform sampling, the spectral gap of the quantum chain decays as $\\delta \\propto e^{-kN}$ with $k = 0.17(5)$, about a third of the classical $k = 0.51(4)$, so thermalization slows down roughly cubically more slowly as cardinality grows. If the claim holds, simulations could reach larger causal sets and, for the first time, sample the causal-set partition function without dimension restrictions.","feed_headline":"Quantum proposals shrink causal-set sampling slowdown by 3x","feed_subtitle":"Spectral-gap fits put the quantum chain's decay exponent at about 0.17, versus 0.51 classically.","key_machinery":"Three pieces carry the argument. (1) The transitive-closure constraint $$H_{\\mathrm{TC}} = P \\sum_{i<j<k} C_{ij}C_{jk}(1-C_{ik})$$ adds a penalty $P$ for every triple of events that would violate the defining transitivity of a partial order, making valid causal sets the degenerate ground space and suppressing proposals into the exponentially larger space of directed acyclic graphs. (2) The qubit Hamiltonian $$H_{\\mathrm{BD}\\varepsilon} = \\frac{4\\sqrt{6}}{\\sqrt{\\varepsilon}}\\left[N - \\varepsilon \\sum_{k<m} C_{km}\\left(1 - 10\\varepsilon \\sum_{k<l<m} C_{kl}C_{lm}\\right)\\right]$$ encodes the 4d Benincasa-Dowker action to first order in the small parameter $\\varepsilon$, using products of at most three of the $q = N(N-1)/2$ relation qubits. (3) The QeMCMC proposal itself: starting from the current causal set, evolve under $H = \\gamma_{\\mathrm{TC}}H_{\\mathrm{TC}} + \\gamma_{\\mathrm{BD}}H_{\\mathrm{BD}\\varepsilon} + \\gamma_{\\mathrm{mix}}\\sum_i X_i$, measure, and classically accept the result only if it passes a transitivity check and, for weighted sampling, the Metropolis-Hastings criterion. The figure of merit is the spectral gap $\\delta$ of the resulting Markov chain, since $\\delta^{-1}$ bounds the thermalization time on both sides.","core_discovery":"The paper claims that the QeMCMC proposal step can be adapted to constrained configuration spaces, with causal-set sampling as the test case. The uniform-sampling algorithm replaces classical moves with quantum real-time evolution under $H_{\\mathrm{uniform}} = (1-\\gamma)H_{\\mathrm{TC}} + \\gamma H_{\\mathrm{mix}}$, where $H_{\\mathrm{TC}}$ penalizes every triple with $i \\prec j$, $j \\prec k$, but not $i \\prec k$; valid causal sets become degenerate ground states. Spectral-gap fits give $k_Q = 0.17(5)$ versus $k_C = 0.51(4)$ in $\\delta \\propto e^{-kN}$, a factor-of-three, so-called cubic reduction of the slowdown exponent. The weighted-sampling algorithm adds a problem Hamiltonian $H_{\\mathrm{BD}\\varepsilon}$ that approximates the 4d Benincasa-Dowker action to first order in $\\varepsilon$ with at most cubic interactions, then accepts proposals via the Metropolis-Hastings rule using the action difference. At temperature $T = 0.004$ the quantum chain again decays far more slowly ($k_Q = 2.09(8)$ versus $k_C = 4.3(4)$), although the paper deliberately attaches no single advantage number to this case because the comparison depends strongly on temperature.","pith_inferences":["The cubic claim rests on exponential fits over very small cardinalities, since the quadratic qubit count confines direct simulation to roughly $N \\le 6$; the persistence of the factor-of-three ratio at larger $N$ is an extrapolation, and running the same fits with the coarse-graining the authors suggest would test it.","A classical proposal that samples directed acyclic graphs and then performs transitive reduction, a strategy the paper notes in passing, could narrow the measured gap; that comparison would sharpen rather than refute the reported advantage.","The BD Hamiltonian is derived for $\\varepsilon \\ll 1$, yet the simulations use $\\varepsilon = 0.1$, a value the paper itself notes is far above physical estimates near $10^{-20}$; re-benchmarking at smaller $\\varepsilon$ would show whether the weighted advantage survives in the physically relevant regime.","Because the paper identifies the inverse temperature with the squared ratio of the discreteness scale to the Planck length, its low-temperature advantage maps onto the regime of large discreteness, where classical samplers freeze and quantum-gravity effects are strongest."],"forward_implications":["The uniform sampler's thermalization time would grow as $e^{0.17N}$ rather than $e^{0.51N}$, extending the range of cardinalities at which causal-set ensembles can be explored without restricting dimension.","The constraint-term recipe generalizes: any problem whose feasible configurations are the ground states of a low-degree Hamiltonian can inherit the constrained QeMCMC proposal.","The BD-action Hamiltonian enables the first MCMC weighted by the causal-set action over the full space of causal sets, opening numerical study of the causal-set partition function in the Euclidean setting.","Because quantum noise only slows thermalization and does not bias the accepted samples, the algorithms are candidates for early fault-tolerant quantum hardware.","At low temperature in 4d, the quantum weighted sampler avoids the critical slowing down that defeats the classical link move, so the practical advantage grows as $\\beta$ increases."],"supporting_citations":[{"why":"Supplies the QeMCMC proposal method, real-time evolution under a problem-plus-mixing Hamiltonian, that both algorithms adapt to a constrained space.","marker":"[26]"},{"why":"Provides the classical uniform-sampling MCMC over causal sets whose spectral gap is the benchmark for the cubic-advantage claim.","marker":"[24]"},{"why":"Defines the Benincasa-Dowker action that the problem Hamiltonian is built to approximate.","marker":"[3]"},{"why":"Gives the dimension-dependent form of the BD action used to extend the Hamiltonian derivation to arbitrary dimension.","marker":"[13]"},{"why":"Supplies the asymptotic enumeration of causal sets, showing the raw bitstring space is exponentially larger and motivating the transitive-closure constraint.","marker":"[25]"},{"why":"Cited for the fact that no analytic proof of QeMCMC speedup exists, which is why the central claim rests on spectral-gap simulations.","marker":"[35]"},{"why":"Supplies the Metropolis-Hastings acceptance rule that turns uniform proposals into BD-action-weighted sampling.","marker":"[31]"}],"fun_headline_variants":["Quantum MCMC trims causal-set sampling slowdown 3x","Causal-set sampling gains super-quadratic quantum edge","QeMCMC adapts to causal sets with 3x slower decay","Quantum chain cuts causal-set slowdown exponent threefold","Quantum proposal speeds causal-set sampling by 3x"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the qubit Hamiltonian in Eq. (7) faithfully encodes the 4d Benincasa-Dowker action; the paper's own Pauli form in Eq. (G2) has coefficients scaling as $\\varepsilon^{3/2}$ and $\\varepsilon^{5/2}$ where Eq. (7) implies $\\varepsilon^{1/2}$ and $\\varepsilon^{3/2}$ (with an extra $\\sqrt{6}$), so the weighted chain may be sampling a different distribution from the claimed partition function.","fun_headline_variants_meta":{"raw":{"variants":["Quantum MCMC trims causal-set sampling slowdown 3x","Causal-set sampling gains super-quadratic quantum edge","QeMCMC adapts to causal sets with 3x slower decay","Quantum chain cuts causal-set slowdown exponent threefold","Quantum proposal speeds causal-set sampling by 3x"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000314,"raw_usage":{"total_tokens":1820,"prompt_tokens":1020,"completion_tokens":800,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":636,"completion_tokens_details":{"reasoning_tokens":717}},"tokens_in":636,"tokens_out":800,"duration_ms":7996,"temperature":1.0,"reasoning_tokens":717,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:32:10.608444+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, on a fixed small ensemble of causal sets, both $H_{\\mathrm{BD}\\varepsilon}$ from Eq. (G2) and the action $S^{(4)}_\\varepsilon$ from Eq. (2), and check whether the Metropolis acceptance probabilities $\\exp(-\\beta\\Delta S)$ match those built from the Hamiltonian actually simulated; the mismatched prefactors mean the two will disagree at $\\varepsilon = 0.1$, which would show the weighted benchmark sampled a different action. Independently, re-run the spectral-gap fits with the coefficient-matched Hamiltonian and check whether $k_Q \\approx k_C/3$ survives; if the ratio changes, the cubic claim falls with it. A third check targets the uniform sampler: obtain proposal statistics at larger $N$ through coarse-graining or quantum-inspired classical simulation and test whether $\\delta \\propto e^{-0.17N}$ still fits.","supporting_citations":[{"cited_title":"Scalar curvature of a causal set.Physical review letters, 104(18):181301, 2010","cited_arxiv_id":null,"evidence_quote":"Defines the Benincasa-Dowker action that the problem Hamiltonian is built to approximate."},{"cited_title":"Causal set d’alembertians for various dimensions.Classical and Quantum Gravity, 30(19):195016, 2013","cited_arxiv_id":null,"evidence_quote":"Gives the dimension-dependent form of the BD action used to extend the Hamiltonian derivation to arbitrary dimension."},{"cited_title":"Asymptotic enumeration of partial orders on a finite set.Transac- tions of the American Mathematical Society, 205:205– 220, 1975","cited_arxiv_id":null,"evidence_quote":"Supplies the asymptotic enumeration of causal sets, showing the raw bitstring space is exponentially larger and motivating the transitive-closure constraint."},{"cited_title":"Bounding the speedup of the quantum-enhanced markov-chain monte carlo algorithm","cited_arxiv_id":null,"evidence_quote":"Cited for the fact that no analytic proof of QeMCMC speedup exists, which is why the central claim rests on spectral-gap simulations."}],"review_version":2}