{"id":"1c194b32-34d0-45cc-945a-31fb7dea34c9","arxiv_id":"2603.17506","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.5,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"Adaptive encoding of continuous variables for quantum annealing improves continuous-field accuracy by orders of magnitude under a fixed binary budget on a composite-rod design problem.","lead":"An adaptive range-update scheme lets quantum annealers encode continuous variables more accurately without adding binary bits. On a structural size-optimization benchmark it cuts field error by more than three orders of magnitude under a fixed qubit budget.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The reader’s weakest-assumption note correctly flags that the five heuristic range-update rules (Cases 1–5 scaled by 1/4 and ρ) may fail to concentrate resolution usefully on larger or more oscillatory problems. That risk, however, is not load-bearing for the strongest claim, which is confined to the concrete, fully documented improvement on the published composite-rod benchmark under fixed binary budget. The manuscript supplies matching best-approximation curves, robustness checks, data availability, and the same hardware as the reference, leaving no internal inconsistency or unsupported leap in the reported numbers. Consequently the ACCEPT verdict stands without adjustment.","tokens_in":20131,"tokens_out":450,"duration_ms":19163,"concrete_test":"Download the archived data from the paper’s DOI (https://doi.org/10.48436/vmpfx-80w27), recompute the final relative H1 error of F(x) for the adaptive-encoding run against the analytic solution, and verify that it equals 6.12e-6 while the recovered design matches the known optimum of the two-element rod.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that adaptive encoding improves relative H1 error of the continuous force field from 1.59e-2 to 6.12e-6 (more than three orders of magnitude) under an identical fixed binary budget of N=26 on the composite-rod benchmark, with QA attaining the encoding-optimal solution at each penalty iteration—is directly and quantitatively supported by Section 3.2, Table 3, and Figure 10. Coincidence of QA solutions with the best-approximation error isolates the gain to progressive range contraction (Eqs. 7–8) rather than solver artifacts. Robustness sweeps over ρ, initial ranges, and n_reads further confirm stability on this instance. The heuristic character of the five update rules and the modest problem size are real limitations for extrapolation, but they do not undermine the demonstrated precision–resource trade-off on the published benchmark itself.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper addresses continuous-variable encoding for quantum annealing (QA) in mixed discrete–continuous engineering optimization. It first shows empirically, on a 1D fluid–structure interaction testbed, that increasing fixed bit depth per continuous variable improves the encoding-optimal resolution but, beyond moderate N, fails to improve (and can stall) QA solution quality on current hardware, consistent with size-dependent integrated control errors. It then introduces an adaptive encoding that keeps the binary budget fixed while updating each continuous variable’s representable interval via three contraction rules (driven by successive iterates) and two saturation-triggered expansions (Eqs. 7–8, Algorithm 1). The scheme is embedded in a quadratic-penalty treatment of a fully coupled minimum-complementary-energy QUBO for structural size optimization, so that each outer iteration re-solves the joint design–field objective. On the published two-element composite-rod benchmark, with the same N=26 binary budget as the fixed-encoding baseline, the method reduces the relative H1 error of the force field from 1.59×10−2 to 6.12×10−6, with QA matching the encoding-optimal solution at each penalty step; parameter sweeps over ρ, initial ranges, and number of reads indicate robustness on this instance.","tokens_in":20357,"tokens_out":1243,"duration_ms":35110,"significance":"If the reported precision–resource trade-off holds, the work supplies concrete, hardware-aware guidance for encoding continuous fields in QA without inflating problem size or abandoning fully coupled global search—an issue that currently forces many applied-QA workflows into hybrid decoupling. Strengths that support credibility include: direct comparison to a published fixed-encoding baseline on the same D-Wave Advantage system; isolation of encoding gains via an explicit best-approximation reference; multi-run medians and interquartile ranges; open data (DOI) and reference to open-source tooling (EngiOptiQA). The update rules are heuristic and the structural benchmark is modest, so broader claims about general mixed-variable engineering problems remain provisional; within those bounds the empirical result is useful for the applied quantum-computing and computational-mechanics communities.","major_comments":[{"comment":"§2.6, Eq. (33) and Figs. 7 and 10: the “best approximation” is defined as the exact minimizer of H_QUBO over the encoded domain, and the design-optimization plots show QA coinciding with it. The manuscript never states how this reference is obtained in practice (exact classical enumeration, digital annealer, or projection of the analytic continuum solution onto the current grid). For the rod with N_total=26 this is feasible classically; for the FSI study with N up to 12 per coefficient it is not obviously so. Without that procedure, the claim that hardware (rather than encoding) limits fixed-depth performance, and that QA attains encoding-optimal solutions under adaptation, cannot be fully audited.","section":null},{"comment":"§2.3.2 (Cases 1–5) and §3.2 / Conclusion: the five range-update rules are free heuristics (including the fixed 1/4 scale and tunable ρ). The central claim on the composite-rod benchmark is well supported, but the paper’s broader assertion of “practical guidance” and improved precision on current hardware rests on the untested assumption that these rules continue to concentrate resolution usefully and do not trap iterates in an infeasible or suboptimal interval for larger, more oscillatory, or higher-dimensional fields. A short, explicit discussion of failure modes (or a second, slightly larger example) would make the load-bearing extrapolation proportionate to the evidence.","section":null}],"minor_comments":[{"comment":"Fig. 6a: the range-update box writes both bounds as a^{(k+1)}_{i,min}; the second should be a^{(k+1)}_{i,max}.","section":null},{"comment":"§2.3.2 / Algorithm 1: the text defines δ^{(k)}_{i,min} using y^{(k)}_{i,min} while the algorithm uses the previous bounds y^{(k−1)}_{i,min}; align the indexing to avoid confusion when implementing the rules.","section":null},{"comment":"Table 1 and §2.4: the mapping of complementary-energy design optimization onto the generic template (1) is clear, but a one-line note that static admissibility is enforced by a quadratic penalty (rather than hard constraints in the QUBO) would help readers who skip to the results.","section":null},{"comment":"Fig. 9: the two nearly identical row pairs appear redundant; a single pair with a clear inset for the a0 expansion event would improve readability.","section":null},{"comment":"Abstract and §4: phrases such as “the framework generalizes beyond structural design” slightly overreach relative to the single structural benchmark; soft wording already present elsewhere (“indicates,” “offers practical guidance”) should be used consistently.","section":null},{"comment":"Minor typographical issues: “as as gradient-based” (Introduction); “J¨ulich” encoding; ensure consistent use of H_1 vs H1 in figure labels.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The central empirical claim is solid and the comparison to the authors’ own prior fixed-encoding baseline is fair and transparent. I do not see a novelty or self-citation problem that would affect the decision. Fit for a computational-engineering / applied-quantum venue is good. My recommendation is minor revision mainly to force an explicit statement of how the best-approximation reference is computed and a more proportionate discussion of the heuristic rules’ untested failure modes; either could be handled in a short revision without new experiments if the authors prefer."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing is that they show fixed bit-depth encodings can stop helping—and even hurt—on current D-Wave hardware once problem size grows, then give a simple adaptive range-update scheme that recovers three-plus orders of magnitude in H1 error on the same composite-rod size-optimization benchmark from their earlier paper, still with N=26 binaries total.\n\nWhat is actually new is the five-case contraction/expansion logic (driven by successive iterates and saturation) embedded inside a quadratic-penalty loop that re-solves the full coupled complementary-energy QUBO each outer iteration. That keeps the joint design–stress search instead of the usual hybrid split. The empirical error analysis on the FSI piston is clean: best-approximation error keeps falling with N while QA plateaus after N≈5, and chain breaks are not the culprit. On the design problem the QA samples track the encoding-optimal reference at every penalty step, so the gain really is from progressive range tightening. Data DOI and EngiOptiQA code are supplied; parameter sweeps over ρ, initial ranges, and reads look careful.\n\nSoft spots are real but proportionate. The benchmark is a two-element rod; the update rules are pure heuristics with free factors (ρ, 1/4); everything is on one Advantage 4.1 system. Extrapolation to larger or more oscillatory problems is unproven. None of that undercuts the demonstrated precision–resource trade-off on the published instance.\n\nThis is for people already doing Ising/QUBO structural or mixed-variable engineering optimization who need better continuous encodings without blowing the qubit budget. Math and citations look solid; self-cites are to the necessary prior formulation. I would send it to referees and would cite it if I were encoding continuous fields for annealers. Worth a look in reading group if the group cares about applied QA.","headline":"Solid practical fix for continuous encodings on QA: multi-order accuracy gain on a published rod benchmark under fixed binary budget, with honest hardware diagnostics.","tokens_in":20938,"tokens_out":476,"would_cite":true,"duration_ms":8281,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Adaptive range updates let quantum annealing refine continuous fields under a fixed binary budget, beating fixed encodings by orders of magnitude on a structural design benchmark.","keywords":["quantum annealing","adaptive encoding","mixed discrete-continuous optimization","structural design optimization","QUBO","minimum complementary energy","composite rod"],"falsifier":"Re-run the same composite-rod (or a larger multi-element) size-optimization problem with the adaptive rules and measure whether the final relative H1 error of the force field still falls below 10^{-5} under the original binary budget; a persistent plateau near the fixed-encoding error of ~10^{-2} would falsify the claimed precision gain.","tokens_in":21047,"feed_emoji":"⚛️","tokens_out":635,"duration_ms":6178,"temperature":0.7,"pith_summary":"Engineering design often couples discrete choices (sizes, materials, topology) to continuous fields (stress, displacement). Quantum annealing can search such problems jointly, but it only handles binary variables, so continuous quantities must be encoded. Fixed bit-depth encodings either waste bits or, on present hardware, can make accuracy worse as the problem grows. This paper introduces an adaptive encoding that keeps the number of binary variables fixed while repeatedly shrinking or expanding each continuous variable's representable interval according to recent iterates. The strategy is embedded in a quadratic-penalty loop that re-solves the full coupled objective at every step, so the annealer still sees the joint problem. On a published composite-rod size-optimization benchmark the method improves continuous-field accuracy by more than three orders of magnitude under the same binary budget, showing that range adaptation can deliver higher effective precision without enlarging the logical problem.","feed_headline":"Adaptive ranges beat fixed bit depth on quantum annealing","feed_subtitle":"Same binary budget, three orders better continuous-field accuracy on a rod design benchmark","key_machinery":"Adaptive encoding update rules: five cases (three contraction rules driven by successive iterate signs or equality, two saturation-triggered expansions) that revise each continuous variable’s interval bounds while the number of binary variables N stays constant; the rules are applied inside a quadratic-penalty loop that re-encodes and re-solves the full coupled QUBO at every outer iteration.","core_discovery":"Simply increasing bit depth in a fixed continuous-variable encoding does not improve—and can degrade—solution quality on current quantum-annealing hardware. An adaptive strategy that contracts and expands representable ranges iteration by iteration, while holding the binary budget fixed, restores precision and yields orders-of-magnitude better continuous-field accuracy on a fully coupled structural design problem.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Adaptive ranges beat fixed bits on quantum annealers","QA continuous precision rises with adaptive encoding","Fixed bit depth degrades QA: adapt ranges instead","Adaptive encoding improves mixed-variable QA solutions","Same qubits, better continuous accuracy via adaptive ranges"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The five simple range-update heuristics will keep concentrating resolution usefully and will not trap the iterates inside an infeasible or suboptimal interval when the problem becomes larger or more oscillatory than the two-element rod.","fun_headline_variants_meta":{"raw":{"variants":["Adaptive ranges beat fixed bits on quantum annealers","QA continuous precision rises with adaptive encoding","Fixed bit depth degrades QA: adapt ranges instead","Adaptive encoding improves mixed-variable QA solutions","Same qubits, better continuous accuracy via adaptive ranges"]},"model":"grok-4.5","effort":"low","cost_usd":0.004124,"raw_usage":{"total_tokens":1266,"prompt_tokens":820,"num_sources_used":0,"completion_tokens":53,"cost_in_usd_ticks":41240000,"prompt_tokens_details":{"text_tokens":820,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":393,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":820,"tokens_out":53,"duration_ms":4167,"temperature":1.0,"reasoning_tokens":393,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T23:07:12.208273+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Re-run the same composite-rod (or a larger multi-element) size-optimization problem with the adaptive rules and measure whether the final relative H1 error of the force field still falls below 10^{-5} under the original binary budget; a persistent plateau near the fixed-encoding error of ~10^{-2} would falsify the claimed precision gain.","supporting_citations":[],"review_version":1}