{"id":"1acf85e5-1aa6-43c2-b349-19ac229d0ede","arxiv_id":"2606.10098","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Adaptive CVaR sampling, two-stage optimization, and a heavy-hex-native deep-chain ansatz layout improve VQA results on dynamic portfolio optimization, with the new layout best on IBM hardware, though no quantum advantage over classical solvers is found.","lead":"This paper studies how sampling strategies, optimizer scheduling, and hardware-aware quantum circuit designs affect variational quantum algorithm performance on a 150-qubit dynamic portfolio optimization problem. A smart generalist might read it for concrete tactics to tune VQA workflows on current quantum hardware for financial optimization tasks.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"QPU layout comparison reports single-run objective values without variance or trial count, so 'best' ranking may reflect noise rather than design","rationale":"Reader correctly flags generalization risk, but the strongest claim is an empirical ranking on one device; its internal validity (statistical robustness of the QPU comparison) is the more immediate load-bearing point. If the concrete test shows the ranking is noise-sensitive, the claim requires qualification even on the reported instance.","tokens_in":1827,"tokens_out":331,"duration_ms":14625,"concrete_test":"In the results section, locate the QPU layout table or figure and count the number of independent circuit executions or shots per layout; if ≤3 runs or no std reported, re-execute the three layouts on a calibrated noise model of ibm_quebec (same 150-qubit instance) for 20 independent seeds and test whether the deep-chain mean objective remains strictly lowest at p<0.05.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim rests on ibm_quebec runs showing the heavy-hex-native deep-chain layout superior in final objective and CVaR tail. Because the comparison occurs on noisy hardware and the abstract (and likely results) give no indication of repeated executions, error bars, or statistical tests, any observed gap could arise from shot noise, calibration drift, or transpilation stochasticity rather than the data-guided or deep-chain modifications. Simulator studies are used only for CVaR/optimizer/depth selection; the decisive layout evidence is hardware-only and therefore vulnerable to this confound.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.3","summary":"The manuscript studies variational quantum algorithms for a 150-qubit dynamic portfolio optimization problem. It proposes an adaptive CVaR schedule that tightens the sampled tail, a two-stage optimizer (global PSO exploration followed by local NFT refinement), and two hardware-aware ansatz modifications (data-guided colored layout and heavy-hex-native deep-chain layout). Simulator experiments select CVaR, optimizer, and depth settings; the decisive ansatz-layout comparison is performed on the ibm_quebec QPU. The central empirical claim is that sampling strategy, optimizer scheduling, and hardware-aware layout materially affect performance, with the deep-chain layout achieving the best final objective value and CVaR-tail performance among tested layouts, although no quantum advantage is observed versus a state-of-the-art classical solver.","tokens_in":1956,"tokens_out":527,"duration_ms":21968,"significance":"If the reported performance ordering is confirmed with statistical controls, the work supplies concrete, actionable guidance on CVaR scheduling, optimizer staging, and post-transpilation ansatz layout for VQAs on heavy-hex hardware. The explicit statement that no quantum advantage is observed is a positive feature of the presentation.","major_comments":[{"comment":"§5 (QPU layout comparison): the reported ranking of ansatz layouts rests on single-run objective values and CVaR-tail metrics with no error bars, no statement of the number of independent executions or shots per run, and no statistical test. Because the decisive evidence for the heavy-hex-native deep-chain layout is hardware-only, the absence of variance estimates makes it impossible to determine whether the observed gaps reflect the proposed design or shot noise, calibration drift, or transpilation stochasticity.","section":"§5"},{"comment":"Abstract and §4–5: the claim that the proposed components 'materially affect performance' is load-bearing for the paper's contribution, yet the hardware results that support the layout component lack the repeated trials and statistical support required to substantiate a material effect.","section":"Abstract, §4–5"}],"minor_comments":[{"comment":"Table of optimizer hyperparameters and CVaR schedule parameters should be added to the methods section to support reproducibility.","section":"Methods"},{"comment":"Simulator figures should explicitly state the number of shots, random seeds, and number of independent runs used for each curve.","section":"Figures 3–4"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the careful review and for highlighting the need for statistical support in the hardware experiments. We address each major comment below and propose targeted revisions to improve transparency.","responses":[{"response":"We agree that the QPU layout comparison in §5 is based on single executions without error bars or statistical tests. This stems from limited access to the ibm_quebec device. In revision we will (i) state the exact shot count used per iteration, (ii) add an explicit limitations paragraph noting single-run hardware results and possible contributions from shot noise, calibration drift and transpilation variability, and (iii) qualify the ranking language to indicate that the deep-chain layout performed best in the reported single trial. Simulator experiments in §4, which used multiple independent runs, already show consistent ordering trends that informed the QPU test; we will cross-reference these to provide supporting context while acknowledging the hardware evidence remains preliminary.","revision_made":"partial","referee_comment":"[§5] §5 (QPU layout comparison): the reported ranking of ansatz layouts rests on single-run objective values and CVaR-tail metrics with no error bars, no statement of the number of independent executions or shots per run, and no statistical test. Because the decisive evidence for the heavy-hex-native deep-chain layout is hardware-only, the absence of variance estimates makes it impossible to determine whether the observed gaps reflect the proposed design or shot noise, calibration drift, or transpilation stochasticity."},{"response":"The overall claim rests on both simulator and hardware evidence. Simulator studies (§4) for CVaR scheduling and optimizer staging include repeated trials and statistical support. For the layout component we will revise the abstract and §5 to distinguish the two: the CVaR and optimizer effects are backed by multi-run simulator data, while the layout comparison is presented as an observed single-run ordering on hardware. We will replace the unqualified phrase “materially affect performance” with more precise wording that reflects the differing levels of statistical support, thereby preserving the contribution while addressing the referee’s concern about overstatement.","revision_made":"partial","referee_comment":"[Abstract, §4–5] Abstract and §4–5: the claim that the proposed components 'materially affect performance' is load-bearing for the paper's contribution, yet the hardware results that support the layout component lack the repeated trials and statistical support required to substantiate a material effect."}],"tokens_in":1542,"tokens_out":525,"duration_ms":16991,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper tests a few concrete changes to VQA workflow for dynamic portfolio optimization on a 150-qubit instance. It introduces an adaptive CVaR schedule that gradually tightens the sampled tail, a two-stage optimizer that uses PSO for global search then switches to NFT for local refinement, and two hardware-aware ansatz modifications: a data-guided colored layout and a heavy-hex-native deep-chain layout. Simulator runs are used to select CVaR, optimizer, and depth settings before the layout comparison moves to ibm_quebec hardware, where the deep-chain layout is reported to give the best final objective and CVaR-tail values.\n\nThe simulator portion looks like a reasonable way to narrow design choices before hardware time. The paper is also straightforward that no quantum advantage appears over a classical solver.\n\nThe main soft spot is the hardware layout comparison. The abstract and stress-test note give no sign of repeated executions, error bars, or statistical tests on the QPU runs. On noisy hardware, any single-run gap could easily trace to shot noise, calibration changes, or transpilation rather than the proposed layout changes. That weakens the central claim about which layout performed best.\n\nThe work is aimed at researchers already running VQAs on financial optimization problems with near-term devices. It supplies some practical pointers on those specific choices but stays narrow in scope.\n\nThe paper shows clear, step-by-step thinking about the workflow and honest reporting of the lack of advantage, so it deserves a serious referee. A referee could ask for more robust hardware statistics and check whether the simulator-to-hardware transfer holds up in the full methods.\n\nI would send it for peer review.","headline":"The paper gives targeted empirical tests of an adaptive CVaR schedule, two-stage optimizer, and two hardware-aware ansatz layouts on a 150-qubit portfolio problem, but the decisive QPU layout ranking rests on single runs without variance or repeats.","tokens_in":2469,"tokens_out":434,"would_cite":false,"duration_ms":15227,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A heavy-hex-native deep-chain ansatz layout achieves the best objective value and CVaR performance in VQA for a 150-qubit dynamic portfolio problem on ibm_quebec.","keywords":["variational quantum algorithms","dynamic portfolio optimization","CVaR sampling","ansatz layout","quantum hardware","optimizer scheduling","portfolio constraints"],"falsifier":"Running the same VQA workflow on a different quantum device or a portfolio instance of different size and checking whether the heavy-hex-native deep-chain layout still records the highest objective value and CVaR-tail score.","tokens_in":2712,"feed_emoji":"","tokens_out":501,"duration_ms":14408,"temperature":0.7,"pith_summary":"This paper examines how sampling objectives, classical optimizer choices, and quantum circuit layouts influence variational quantum algorithms applied to dynamic portfolio optimization. The work introduces an adaptive CVaR schedule that tightens the risk tail over iterations, a two-stage optimizer that pairs particle swarm exploration with Nakanishi-Fujii-Todo refinement, and two hardware-aware ansatz modifications. Simulator runs select sampling and optimizer settings, while the layout comparison runs directly on the ibm_quebec processor for a 150-qubit instance. The results establish that these workflow decisions change the quality of the obtained portfolios.","feed_headline":"Deep-chain ansatz layout wins on 150-qubit portfolio optimization","feed_subtitle":"The heavy-hex-native design records the strongest objective value and CVaR-tail score on ibm_quebec among tested layouts.","key_machinery":"The heavy-hex-native deep-chain ansatz layout, which increases native two-qubit interaction depth without extra routing overhead after transpilation.","core_discovery":"The paper establishes that sampling strategy, optimizer scheduling, and hardware-aware ansatz layout design materially affect VQA performance on dynamic portfolio optimization. In particular, on a 150-qubit instance executed on the ibm_quebec QPU, the proposed heavy-hex-native deep-chain layout achieves the best final objective value and CVaR-tail performance among the tested layouts.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Heavy-hex deep-chain ansatz achieves best objective in 150-qubit portfolio VQA","Hardware-aware ansatz design improves VQA for dynamic portfolio optimization","Optimizer scheduling and CVaR sampling boost 150-qubit portfolio VQA results","Deep-chain heavy-hex layout achieves best CVaR-tail score on ibm_quebec QPU"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The performance gains observed for the adaptive CVaR schedule, two-stage optimizer, and hardware-aware ansatz layouts on the specific 150-qubit instance and ibm_quebec device will translate to other problem sizes, constraint sets, or quantum hardware platforms.","fun_headline_variants_meta":{"raw":{"variants":["Heavy-hex deep-chain ansatz achieves best objective in 150-qubit portfolio VQA","Hardware-aware ansatz design improves VQA for dynamic portfolio optimization","Optimizer scheduling and CVaR sampling boost 150-qubit portfolio VQA results","Deep-chain heavy-hex layout achieves best CVaR-tail score on ibm_quebec QPU"]},"model":"grok-4.3","cost_usd":0.01468,"raw_usage":{"total_tokens":6361,"prompt_tokens":763,"num_sources_used":0,"completion_tokens":86,"cost_in_usd_ticks":146799500,"prompt_tokens_details":{"text_tokens":763,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":5512,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":763,"tokens_out":86,"duration_ms":29913,"temperature":1.0,"reasoning_tokens":5512,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T14:21:28.967128+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Running the same VQA workflow on a different quantum device or a portfolio instance of different size and checking whether the heavy-hex-native deep-chain layout still records the highest objective value and CVaR-tail score.","supporting_citations":[],"review_version":1}