{"id":"8af529b8-55b0-41a8-b025-84a3bc2f54de","arxiv_id":"2607.14063","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Catalysts built from sign-propagated path patterns reshape the Hamming-distance structure of Ising landscapes and increase near-solution probability in quantum annealing simulations.","lead":"This paper designs extra two-spin interactions (\"diagonal catalysts\") for quantum annealing, built only from the problem's couplings, not its answer. On simulated sparse optimization problems, these catalysts shift the final probability distribution toward near-optimal configurations, improving the chance of finding good solutions in short runs.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"All quantitative gains rest on a fixed first-order Trotter step Δt=0.1 with no convergence check; a rerun at smaller step could materially change the headline improvements.","rationale":"I read the paper's argument as follows: Theorem 1 gives exact static shell moments, the path construction gives a J-only catalyst, and numerical simulations show dynamic gains. The static part is proven and verified. The dynamic part is the novel quantitative claim. The weakest link is the numerical integration. The reader identified the same. A secondary concern is that the path order m per family was selected after inspecting all orders on the same instances, which may inflate the quoted gains; however, even if that is true, the qualitative claim is likely robust. The Trotter issue, by contrast, could invalidate all dynamics numbers. Therefore I retain the CONDITIONAL verdict and propose a concrete convergence check. No internal inconsistency in Theorem 1; the proof appears correct.","tokens_in":15992,"tokens_out":6301,"duration_ms":61535,"concrete_test":"Rerun the 3-regular 200-instance ensemble at T=6 and T=10 with Δt=0.05, 0.02, and 0.01 using the provided QiliSDK scripts, and recompute the median paired gains for P0.95 and P(δ≤2). If the median gains (e.g., +351% at T=6) drift by more than 10% relative, or the fraction of improved instances drops below 95%, then the headline numbers are not converged and the supporting evidence for the central claim weakens. Also compute total-variation distance between final distributions at Δt=0.1 and Δt=0.01 for a handful of instances.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that diagonal catalysts multiply near-solution probability (e.g., +351% median top-band mass at T=6)—is supported entirely by state-vector simulations that use a fixed first-order Trotter step Δt=0.1. The Supplemental 'Numerical Dynamics and Normalization' states 'No adaptive ODE tolerance is used; the numerical accuracy parameter is the fixed product-formula step size Δt' and notes 'the same instances can be rerun at smaller Δt for step-size checks,' but no such convergence study is reported. For n=20 with peak-normalized couplings, H(s) contains the transverse driver plus O(n^2) ZZ terms with coefficients of order 1, scaled by s(1-s); the first-order product formula has global error O(Δt) with a prefactor depending on commutator norms that can be large. Diabatic annealing outcomes are sensitive to small perturbations: Trotterization introduces a time-dependent error Hamiltonian that can induce non-adiabatic transitions similar to the avoided-crossing physics the catalyst is intended to modify. Therefore the quantitative gains in Tables S1–S2, including the 99% improved fraction, may be artifacts of the discretization. The absence of a reported Δt→0 check leaves this unresolved.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a mathematical framework, built on Hamming shells, for understanding the energy–distance structure of field-free Ising QUBOs. Theorem 1 gives exact shell means and variances for any two-body Ising operator. Using this framework, the authors propose a diagonal ZZ-catalyst constructed only from the coupling matrix J via sign propagation along self-avoiding paths (Eq. (14)), with no information about the ground state. They test the catalyst in state-vector simulations on 200 instances each of 3-, 4-, and 5-regular graphs and on Sherrington–Kirkpatrick models at n=20, reporting large improvements in top-band energy mass and near-solution Hamming mass at sweep times T=2,6,10, relative both to the bare anneal and to an equal-peak-coupling baseline C=H_P.","tokens_in":16296,"tokens_out":5918,"duration_ms":60769,"significance":"If the numerical claims hold, this is a valuable result: the catalyst is two-local, diagonal, solution-free, and implementable with controls native to current annealers. Theorem 1 is exact and its proof in Appendix A is clean and parameter-free. The benchmark design is careful in several respects: paired per-instance comparisons, an equal-budget control, bootstrap confidence intervals, and a public data repository. The central dynamic claim, however, is currently supported only by fixed-step Trotter simulations with no convergence check. In addition, the per-family path order and the omission of path-edge contributions appear to be chosen after seeing results, which introduces a risk of overfitting in the reported gains. These issues are fixable but are load-bearing for the paper's central claim.","major_comments":[{"comment":"All quantitative results, including the headline '+351% median top-band mass' and '99% improved' at T=6, are produced with a fixed first-order Trotter step Δt=0.1. The Supplemental explicitly states that no adaptive ODE tolerance is used and that 'the same instances can be rerun at smaller Δt for step-size checks,' but no such check is reported. With n=20 and peak-normalized O(n^2) ZZ terms, the product-formula error is not negligible by construction, and diabatic annealing distributions can be sensitive to small perturbations. I request a convergence study (e.g., Δt=0.05, 0.02, 0.01) on at least a subset of instances for the main comparisons, reporting how the median paired gains and improved fractions change. Without this, the quantitative claims are not yet supported.","section":"Supplemental Material, 'Numerical Dynamics and Normalization'"},{"comment":"The path order m is fixed per family (m=4,3,2) because 'the order that maximizes the median paired gain ... decreases with degree.' This is a post hoc choice made after inspecting the simulation results. Moreover, Eq. (14) omits path-edge contributions because this 'empirically improves funneling.' These choices are not derived from Theorem 1 or from any prespecified rule. As reported, the gains may overstate the performance of a predetermined catalyst. The authors should either provide a predictive rule for choosing m from J alone, report results for all orders for each family, or separate order selection from evaluation using a held-out design.","section":"Main text, Results and Eq. (14)"},{"comment":"The construction's starting assumption is that sign propagation along paths gives patterns g^(p) that track the global ground state. Appendix B shows that a path contributes to the funnel only when (2φ_p−1)^2 > 1/(m+1), and that for independent random agreement the expected value equals the threshold. The actual fidelity distribution φ_p on the benchmark instances is never reported. Without evidence that the path patterns are sufficiently aligned, the empirical gains could in principle arise from the added density of a catalyst whose patterns are not tracking the solution. I ask for the distribution of |2φ_p−1| on the tested instances, and ideally a randomized-pattern control with the same weights and support, to confirm that the structure, not merely the added couplings, is responsible for the gains.","section":"Appendix B and Eq. (12)"}],"minor_comments":[{"comment":"The transverse-field scale Γ is introduced informally. Please state its relation to the driver coefficient in Eq. (2), e.g., Γ=1 there, or define it through the Hamiltonian norm.","section":"Introduction, Eq. (3)"},{"comment":"The caption mentions 'seed H = 4' but 'seed' is not defined. Please clarify or remove the term.","section":"Fig. 1"},{"comment":"The phrase 'ground-state patterns of small frustration-free subproblems' is slightly misleading: a path is a tree and hence is automatically frustration-free. Consider saying 'small tree subproblems' or 'sign patterns on tree subgraphs.'","section":"Abstract"},{"comment":"Instances with zero baseline probability are omitted from ratio summaries. This can bias the reported improved fraction and median gain when the baseline is occasionally zero. Please report how many instances are omitted for each protocol and threshold.","section":"Supplemental Material, 'Relative-Improvement Statistics'"},{"comment":"The sign convention in (B2) is easy to misread. Please spell out the expansion of −M_p^2 after discarding the constant, so that the reader can verify that a negative C_id(z*) corresponds to deepening the funnel.","section":"Appendix B, Eq. (B2)"}],"recommendation":"major_revision","confidential_remarks":"The paper has a sound exact result and a sensible heuristic, but the numerical verification is incomplete as submitted. The fixed Trotter step is the main obstacle: the authors themselves note that a step-size check is possible but do not perform it. If a convergence study confirms the gains, I would view the paper as publishable after the order-selection and fidelity issues are addressed transparently."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this paper has a real theoretical result and a plausible heuristic construction, but the banner numbers (+351% top-band mass, etc.) come from a single fixed Trotter step Δt=0.1 with no convergence check. That is a genuine omission, not something I can wave away.\n\nWhat is new and good: Theorem 1 gives exact Hamming-shell means and variances for any field-free two-body Ising operator, with a short parameter-free proof. That is a clean addition to the diagonal-catalyst literature. The path-based ZZ construction is solution-independent, built only from J, and the path-edge omission plus orthogonalization is a sensible design. The static diagnostics in Fig. 4 match the theory, and the benchmarks are reasonably thorough: 200 instances per family, an equal-peak-coupling baseline C=HP, bootstrap intervals, and data/scripts on GitHub. The authors also explicitly note the order–fidelity trade-off and report the dense-graph decay.\n\nSoft spots, in order of size. First, the Trotter step. The supplement says the accuracy parameter is the fixed product-formula step Δt, and notes the same instances can be rerun at smaller Δt, but no such check is reported. For n=20 with O(n^2) ZZ terms and peak-normalized couplings, the first-order product-formula error is not self-evidently negligible, and diabatic transitions are exactly the kind of thing that Trotter error can distort. This does not sink the paper, but it means the quantitative gains are provisional until a step-size study is shown. Second, the per-family path order m is selected after seeing the results. The paper is transparent that it fixes m per family, but the choice was outcome-driven; that is a mild selection effect, not a deal-breaker, since gains persist across orders 2–5. Third, the path-fidelity heuristic has no instance-level guarantee; the authors acknowledge this in Appendix B, and it mostly affects the size of the effect, not the sign.\n\nMy take: the central framework holds up. The static shell analysis is solid, and the construction is transparent enough that the missing Trotter check is fixable rather than fundamental. The paper deserves a serious referee; I would send it to review with a request for a convergence study and a cross-validated choice of m.\n\nWho gets value: anyone working on diagonal catalysts, QAOA phase separators, or landscape design for quantum annealing. I would cite the theorem.\n\nRecommendation: accept for peer review, with revision.","headline":"The shell-moment theorem and path-based catalyst construction are genuinely new and worth refereeing, but the headline gains rest on a single fixed Trotter step that the authors themselves leave unchecked.","tokens_in":16789,"tokens_out":3524,"would_cite":true,"duration_ms":29835,"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":"This paper claims that a two-body diagonal catalyst built solely from the coupling signs and magnitudes of an Ising optimization problem—with no knowledge of its solution—can reshape the annealing landscape so that finite-time quantum annea","keywords":["quantum annealing","diagonal catalysts","Hamming shells","Ising spin glass","ZZ couplings","path sign propagation","shell-moment theorem","approximate optimization"],"falsifier":"Compute, on the 200 3-regular n=20 instances, the median squared alignment |2φ_p − 1|^2 for the order-4 paths used in the catalyst; if it is at or below the random baseline 1/(m+1), the catalyst has no meaningful estimate of the ground state. Separately, rerun the T=6 protocol at Trotter steps 0.05 and 0.025; if the per-instance paired gains move by more than the quoted interquartile ranges, the reported dynamics are not converged.","tokens_in":15887,"feed_emoji":"⚛️","tokens_out":7417,"duration_ms":62785,"temperature":0.7,"pith_summary":"The paper tries to establish that a solution-blind, diagonal 'catalyst'—a set of extra ZZ couplings computed from the problem's own coupling graph—can make quantum annealing concentrate its final states near the true solution at short sweep times. The key move is a shell-moment theorem: for any two-body Ising cost, the average energy on the Hamming shell at distance d from the solution is a fixed parabola in d, and a diagonal catalyst can only rescale that parabola while independently shrinking its width. Using that lens, the authors build catalysts by propagating signs along self-avoiding paths in the coupling graph, a procedure that uses no information about the answer. On 200 random 3-regular 20-spin instances at sweep time T=6, the median probability within 95% of the ground-state energy rises from 0.067 to 0.324, and the probability within two spin flips of a ground state rises from 0.10 to 0.35; gains persist, weakened, on fully connected models. A sympathetic reader would care because it offers a hardware-native, purely diagonal way to improve approximate optimization on annealers without knowing the solution in advance.","feed_headline":"Solution-blind catalyst triples near-solution anneal hits","feed_subtitle":"Adding ZZ terms from coupling signs lifts final near-solution probability from 0.10 to 0.35.","key_machinery":"The load-bearing objects are Hamming shells S_{d,z*}—the sets of configurations at distance d from a reference solution z*—and the exact shell-moment identities (Theorem 1), which give the mean and variance of any field-free two-body Ising operator over a shell. These identities decompose the landscape into a fixed parabolic funnel (μ_2(d) H(z*)) plus within-shell fluctuations; the catalyst's job is to deepen the funnel and narrow the fluctuations. The construction uses open-path sign propagation: along each m-edge self-avoiding path, signs are propagated according to g_{v_{a+1}} = −sgn(J_{v_a v_{a+1}}) g_{v_a}, which always yields a locally satisfying pattern; weighting each vertex by its i","core_discovery":"The central claim is that the energy-versus-Hamming-distance landscape of an Ising problem can be reshaped by a two-body diagonal catalyst assembled from the problem couplings alone, and that this reshaping translates into a many-fold increase in the probability that a finite-time anneal lands at or near the ground state. Theorem 1 pins down what is possible: the shell-mean energy is exactly μ_2(d) times the ground-state energy, with μ_2(d) purely combinatorial, so a diagonal catalyst controls only the overall funnel steepness, while the shell variance is controlled separately. The constructive step propagates signs along self-avoiding paths—each path's propagated pattern is an exact ground","pith_inferences":["Editorial: the shell-moment theorem suggests a static screening tool the paper does not fully exploit—one could search over subgraph families beyond paths (stars, trees, small frustrated loops) for the family that minimizes the scale-invariant shell-overlap metric ov_d before running any dynamics, making catalyst selection a precomputed property of the coupling matrix.","Editorial: the near-solution Hamming concentration that persists on fully connected models at all sweep times points to a mechanism distinct from spectral-gap widening; a direct test would compare the instantaneous gap spectra of catalyzed and uncatalyzed protocols on the same instances, which the paper reports only indirectly through final distributions.","Editorial: as an explicit testable extension, the same catalyst could be inserted into QAOA as an additional diagonal phase separator—something the paper lists as a possible direction—and one could benchmark fixed-depth QAOA with and without the catalyst on 3-regular MaxCut instances; because the catalyst is built from J alone, this is a drop-in modification.","Editorial: the construction's reliance on a fixed Trotter step of 0.1 in the numerical dynamics is the kind of detail a convergence study could settle; if the gains persist at Δt=0.05 and 0.025, the claim becomes robust to discretization."],"forward_implications":["On sparse 3-regular instances, the near-solution probability mass (within two spin flips of the ground state) at T=6 rises from a median of 0.10 to 0.35, a +239% paired gain on 94% of instances, so annealers become useful as near-optimal solvers even when they miss the exact ground state.","The catalyst outperforms the equal peak-coupling control C=H_P by +77% in top-band mass and +102% in near-solution mass at T=6, indicating the distribution of coupling strength matters, not just its total scale.","Gains persist, though weaker, on fully connected Sherrington–Kirkpatrick instances (+22% median top-band gain at T=2, with every instance improved in near-solution Hamming mass), so the geometric funneling mechanism is not limited to sparse graphs.","The path order m is a tunable design axis: the optimal order decreases with graph degree (m=4,3,2 for 3-,4-,5-regular graphs at longer sweeps), letting practitioners trade reach against pattern fidelity on the fly.","Because the final distribution is closer to the solution in Hamming distance even when the ground state is not sampled, the method composes naturally with local post-processing or reverse-annealing warm starts."],"fun_headline_variants":["ZZ catalysts reshape anneal funnel, boost hits","Coupling-sign catalysts triple near-solution anneal hits","Diagonal catalysts steepen anneal energy landscape","ZZ terms tune anneal funnel, multiply near-solution odds","Catalyst from coupling signs lifts anneal success rate"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire construction leans on the hope that sign-propagated patterns on open paths track the true ground-state alignment better than random (the fidelity condition of Appendix B), and on the numerical simulations' fixed Trotter step Δt=0.1 with no reported step-size convergence check; if either gives way, the reported funneling gains are not established.","fun_headline_variants_meta":{"raw":{"variants":["ZZ catalysts reshape anneal funnel, boost hits","Coupling-sign catalysts triple near-solution anneal hits","Diagonal catalysts steepen anneal energy landscape","ZZ terms tune anneal funnel, multiply near-solution odds","Catalyst from coupling signs lifts anneal success rate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000241,"raw_usage":{"total_tokens":1282,"prompt_tokens":594,"completion_tokens":688,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":338,"completion_tokens_details":{"reasoning_tokens":612}},"tokens_in":338,"tokens_out":688,"duration_ms":6890,"temperature":1.0,"reasoning_tokens":612,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T02:53:03.003124+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, on the 200 3-regular n=20 instances, the median squared alignment |2φ_p − 1|^2 for the order-4 paths used in the catalyst; if it is at or below the random baseline 1/(m+1), the catalyst has no meaningful estimate of the ground state. Separately, rerun the T=6 protocol at Trotter steps 0.05 and 0.025; if the per-instance paired gains move by more than the quoted interquartile ranges, the reported dynamics are not converged.","supporting_citations":[],"review_version":1}