{"id":"242873a2-6a94-4268-8fac-aac5bdbd3b23","arxiv_id":"2412.14095","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The paper proposes and tests a CVQLS-augmented interior point method for optimal power flow, with heuristic convergence improvements, reporting reliable solutions on small systems and an unsupported proxy for larger ones.","lead":"This paper integrates a coherent variational quantum linear solver into the interior point method for solving optimal power flow. It shows promising results on small test systems, but the claimed scalability to larger grids relies on a noise-injection proxy rather than actual quantum simulation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 118/300-bus scalability results rest entirely on a noise-injection proxy that perturbs the classical Newton matrices without running CVQLS; the paper provides no evidence that this proxy reproduces CVQLS behavior, so the scalability claim is unsupported.","rationale":"The reader's verdict was CONDITIONAL, with the weakest assumption being that the 118/300-bus results rely on a Qiskit noise-injection proxy rather than actual CVQLS execution. My independent read converges on the same concern, and I find it genuinely load-bearing: the abstract and conclusion explicitly tout scalability testing on 118- and 300-bus systems, and Section V.A admits that CVQLS could not be simulated for these cases. The paper's own language in Section V.F—'introduced noise into our linear system's matrix H_k and vector r_k'—shows that the proxy is classical data perturbation, not a quantum circuit simulation. This cannot validate the central claim that CVQLS, with its tailored ansatz and sequential initialization, is the right quantum linear solver for large OPF instances. I do not see a basis to move the verdict to REJECT: the small-system experiments (2-bus on hardware, 3/5-bus on PennyLane simulator) are genuine, and the proposed techniques may still be useful in a small-scale demonstration. However, the scalability claim, a key part of the paper's contribution, is unsupported by the current evidence. The conditionality of the reader's verdict already captures this; the required condition is that the noise-injection proxy must be validated, or the 118/300-bus results must be re-derived with a true CVQLS simulation or a properly justified error model. Other potential weaknesses, such as the lack of statistical rigor in the VQLS/CVQLS comparison and the hand-tuned mu-correction thresholds, are secondary to this proxy because even if those were fixed, the large-system results would still not demonstrate quantum scalability. The concrete test I propose would settle the matter directly: on a case where both the genuine CVQLS simulator and the proxy are feasible, compare their behaviors; disagreement would demonstrate the proxy's invalidity, while agreement would give some confidence in the large-system numbers. Until then, the scalability claim should be regarded as unverified, not as an established result.","tokens_in":15871,"tokens_out":4725,"duration_ms":45674,"concrete_test":"Reproduce the 3-bus DCOPF case, for which genuine PennyLane CVQLS results exist (Fig. 13), using the same noise-injection scheme that was applied to the 118/300-bus systems: add Qiskit-noise-style perturbations to H_k and r_k, solve the Newton system classically, and run the same QIPM with Algorithm II. Compare the resulting objective trajectory and final generator output to the CVQLS simulator results. If the final objective differs by more than the 5% accuracy observed on the 2-bus hardware run, or if the trajectory qualitatively differs, the noise-injection proxy is invalid for extrapolating scalability. Separately, the authors should release the exact noise model parameters (error rates, gate set, how noise was added to H_k and r_k) so the proxy can be audited.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central scalability claim depends on Section V.F, where for the 118-bus and 300-bus cases CVQLS is not executed. Section V.A explicitly states that encoding the large Hessians caused out-of-memory errors (Table VI), and that 'for the larger systems, we modeled quantum errors and noise using the Qiskit noise simulator and incorporated these into the computations.' Section V.F then says the simulator 'introduced noise into our linear system's matrix H_k and vector r_k.' This is not a faithful simulation of CVQLS on noisy hardware: it replaces the variational circuit, its ansatz, the classical-quantum optimization loop, and measurement shot noise with an unspecified additive perturbation to the classical linear system data. Such a proxy cannot capture the very phenomena the paper claims to manage, such as barren plateaus, ansatz expressibility, or CVQLS's supposed stability with ill-conditioned matrices. Consequently, the 118/300-bus QIPM results demonstrate only that a classical IPM tolerates some perturbation of H_k and r_k; they say nothing about whether the CVQLS-augmented QIPM scales. Because the abstract advertises 'We use a quantum noise simulator to test scalability' as evidence for the approach, this is the most load-bearing weakness of the paper.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a quantum-augmented interior point method (QIPM) for optimal power flow (OPF), replacing the classical Newton linear solve with a coherent variational quantum linear solver (CVQLS). It argues that CVQLS is preferable to HHL and VQLS for OPF because of its stability with ill-conditioned matrices, and it introduces three enhancements: an OPF-tailored ansatz, sequential parameter initialization across IPM iterations, and a μ-correction rule that stabilizes the central path under inexact quantum solves. The method is tested on a 2-bus system on real IBMQ hardware, on 3- and 5-bus systems with the PennyLane simulator, and on 118- and 300-bus systems using Qiskit noise simulation. The small-system trajectories match classical IPM, and an ablation over 1,000 random load cases suggests that removing the μ-correction causes frequent convergence failures. The paper concludes that CVQLS-augmented QIPM provides reliable OPF solutions while acknowledging current hardware limitations.","tokens_in":16195,"tokens_out":5488,"duration_ms":49832,"significance":"If the scalability evidence were valid, this would be a useful practical integration of a variational quantum linear solver into power-system optimization, and the real-hardware 2-bus demonstration (192 circuits) is a genuine contribution. The paper also credits its own limitations honestly and provides an ablation of the μ-correction, which is a constructive step. However, the central scalability claim currently rests on a noise-injection proxy that does not execute CVQLS on the large systems, so the contribution at present is best viewed as a small-system feasibility study plus a classical sensitivity analysis. The comparison of VQLS and CVQLS on a few IPM matrices is suggestive but not a systematic validation of the ill-conditioning stability claim.","major_comments":[{"comment":"The 118- and 300-bus results in Figs. 15 and 16 are not CVQLS simulations. Section V.A states that encoding large Hessians caused out-of-memory errors and that 'we modeled quantum errors and noise using the Qiskit noise simulator and incorporated these into the computations,' and Section V.F says this simulator 'introduced noise into our linear system's matrix H_k and vector r_k.' This replaces the variational circuit, the ansatz, the classical-quantum optimization loop, and measurement shot noise with an unspecified additive perturbation of the classical Newton data. Such a proxy cannot validate the paper's claims about barren-plateau avoidance, ansatz expressibility, or stability with ill-conditioned matrices. The abstract's statement 'We use a quantum noise simulator to test scalability' is therefore misleading: the large-system experiments demonstrate only that a classical IPM tolerates some perturbation of H_k and r_k. To support the scalability claim, the authors should either run actual CVQLS circuits for intermediate systems (e.g., 14-, 30-, or 57-bus) or explicitly reframe the large-system experiments as a classical sensitivity analysis, with the quantum experiments limited to the small systems.","section":"V.A and V.F"},{"comment":"The complexity expression in Eq. (21) is presented as the time complexity of the CVQLS-augmented IPM, but CVQLS is a variational algorithm whose runtime is not established by a theorem in this paper. The cited bounds in [13] hold under assumptions (e.g., block-encodings, effective condition number, state-preparation guarantees) that the authors do not verify for OPF KKT matrices. Moreover, the numerical results show QIPM requiring more IPM iterations than classical IPM, which is inconsistent with assuming O(log(1/ε)) outer iterations with exact Newton steps. The authors should either derive Eq. (21) from explicitly stated assumptions or replace it with a carefully qualified statement that separates the quantum linear-solve step from the outer IPM loop and notes the absence of a proven variational convergence guarantee.","section":"IV.F, Eq. (21)"},{"comment":"The central motivation for selecting CVQLS is its 'stability with ill-conditioned matrices.' The supporting evidence in Section V.C consists of a few runs where VQLS fails with Adam or COBYLA (Figs. 5, 7) and CVQLS succeeds (Figs. 8, 9). No experiment varies the condition number of the KKT matrix, nor is there a systematic comparison of VQLS and CVQLS on the same sequence of IPM iterations with reported condition numbers and solution errors. Since the entire approach is justified by this advantage, please add a controlled comparison across a range of condition numbers and IPM iterations, or explicitly soften the claim to 'empirically observed on the tested cases.'","section":"II.C and V.C"}],"minor_comments":[{"comment":"The μ-correction ablation reports 'more than 60% of the simulations failed to converge' in every test system, with exact percentages said to be in Table VIII, but Table VIII is not populated in the manuscript text provided. Please ensure the table (or equivalent numbers) appears, since this is the only quantitative support for the necessity of the μ-correction.","section":"V.G / Table VIII"},{"comment":"The heading 'Varaitional Quantum Optimization Landscape' contains a typo; it should read 'Variational.' Also, the plural 'ansatzs' appears twice in Section IV.B and should be 'ansatze' or 'ansatz circuits.'","section":"IV.B"},{"comment":"The text contains 'roper initialization' (missing 'P') and a duplicate section numbering: both Section V and the Conclusion are labeled 'V.' Please correct these formatting issues.","section":"IV.C and V"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the large-system scalability claim, which is built on a noise-injection proxy rather than on actual CVQLS execution. If the authors can supply genuine CVQLS runs on at least one intermediate system (or substantially reframe the large-system results as a classical sensitivity analysis), the paper could become publishable as a small-system demonstration with an honest discussion of scalability. The real-hardware 2-bus experiment and the μ-correction ablation are useful and should be preserved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper's small-system results are the real contribution. The 3- and 5-bus PennyLane runs actually execute CVQLS inside the IPM loop, the 2-bus run on an IBMQ device is a genuine hardware demonstration, and the central-path mu-correction heuristic is a sensible answer to a real problem: inexact Newton steps need a gentler barrier-parameter schedule. Warm-starting CVQLS parameters across IPM iterations also clearly helps and is supported by the cost-function plots. That part deserves a read.\n\nThe soft spot is Section V.F. For 118- and 300-bus systems, encoding the Hessian blows up memory, so instead of running CVQLS they inject Qiskit noise into the classical H_k and r_k. That is not a faithful simulation of CVQLS on noisy hardware. It cannot capture the variational optimization, the ansatz expressibility, measurement shot noise, or the barren-plateau behavior that Section IV is built around. The resulting convergence curves show a classical IPM can tolerate some perturbation of its Newton data; they do not show CVQLS scaling. The abstract's 'We use a quantum noise simulator to test scalability' overstates what was done. To the authors' credit, the complexity analysis in IV.F is explicitly conditional on fault-tolerant hardware and calls itself a theoretical upper-bound, but the numerical evidence for the 118/300-bus cases does not support the scalability claim.\n\nThe other issues are minor. The VQLS-vs-CVQLS comparison is anecdotal with single runs and no error bars. The mu-correction threshold and averaging window tau are hand-tuned. The ansatz choice is empirical. None of this is fatal for a methods paper, but it should be labeled as such.\n\nI'd send this to review, but with the expectation of a major revision on Section V.F. The fix is either a genuine CVQLS simulation at whatever size is feasible, or a relabeling of the large-system results as a sensitivity study of classical IPM to noisy Newton data. The small-system story is solid enough to be worth refereeing; the scalability claim needs to be reined in.","headline":"Small-system CVQLS-IPM integration is genuine and the mu-correction heuristic is sensible, but the 118/300-bus scalability evidence is a classical noise-injection proxy, not a CVQLS execution, so the scaling claims do not follow.","tokens_in":16677,"tokens_out":2772,"would_cite":false,"duration_ms":23893,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","90C51"],"pacs":["03.67.Ac","03.67.Lx"],"model":"deepseek-v4-flash","headline":"A coherent variational quantum linear solver can replace the hard linear solve inside interior-point optimal power flow.","keywords":["optimal power flow","quantum linear solver","coherent variational quantum linear solver","interior point method","variational quantum algorithm","ill-conditioned matrices","quantum noise simulation","power systems optimization"],"falsifier":"Run the same CVQLS-augmented IPM on a real noisy quantum processor, or on a hardware-validated noise model calibrated with measured gate errors, crosstalk, and readout errors, for a system of at least ten buses. If the objective value or the KKT residuals deviate sharply from the paper's noise-injected simulations while the classical IPM converges, the central scalability claim is falsified.","tokens_in":15660,"feed_emoji":"⚡","tokens_out":6564,"duration_ms":58932,"temperature":0.7,"pith_summary":"This paper tries to establish that optimal power flow—the optimization problem that sets generator outputs and voltages in an electricity grid—can be solved by an interior point method whose inner linear equations are solved by a coherent variational quantum linear solver (CVQLS) rather than a classical direct solver. The authors argue that CVQLS is the practical quantum solver for this task because the Hessian matrices appearing in IPM iterations are ill-conditioned, and CVQLS remains stable where other variational solvers struggle. They add three engineering fixes: a shallow ansatz tailored to the OPF matrices, warm-start parameter initialization from the previous IPM iteration, and a modified barrier-parameter update that keeps the IPM near the central path. With these fixes, they report solutions close to the classical optimum on test systems from 2 to 300 buses, including a genuine run on commercial quantum hardware for the 2-bus case.","feed_headline":"Quantum solver tames ill-conditioned power-flow matrices","feed_subtitle":"With warm-started parameters and a corrected central path, QIPM tracks the classical optimum across test systems.","key_machinery":"The workhorse is CVQLS used as a drop-in replacement for the direct linear solve inside IPM: a variational circuit $V(\\omega)$ prepares the state $|\\Delta x_k\\rangle$ such that $H_k|\\Delta x_k\\rangle$ is proportional to $|-r_k\\rangle$, with the cost function evaluated through local Hadamard tests. The supporting machinery is the Pauli decomposition of the Hermitian Hessian into controlled unitaries, a shallow $R_y$–$CZ$ ansatz chosen empirically, sequential initialization of the variational parameters from the previous IPM iteration, and a central-path correction that freezes the barrier parameter $\\mu$ when the averaged objective shows large fluctuations.","core_discovery":"The central claim is that the Newton-step linear system $H_k \\Delta x_k = -r_k$ arising at each IPM iteration can be solved by CVQLS, with the Hermitian matrix $H_k$ encoded through Pauli decomposition and the solution prepared by a shallow variational circuit. On this basis, the paper claims that CVQLS outperforms HHL and VQLS for OPF because of its stability with ill-conditioned matrices, that a shallow ansatz with sequential parameter initialization converges reliably while deeper ansatzes overfit or hit barren plateaus, that the Adam optimizer works best for CVQLS, and that a modified $\\mu$-update is necessary to keep inexact quantum solves on the central path. The paper further claims that without this $\\mu$-correction, more than 60% of one thousand random demand scenarios fail to converge, and that with it, the quantum-augmented IPM tracks the classical IPM objective on systems up to 300 buses under simulated device noise.","pith_inferences":["Beyond the paper's own claims: if the noise-injection proxy used for the 118- and 300-bus cases does not faithfully reproduce real hardware noise, those large-system results do not demonstrate scalability; the genuine quantum evidence is the 2-bus hardware run and the small simulator runs.","Beyond the paper: the sequential warm-starting and central-path correction are general recipes that could benefit other variational quantum algorithms applied to ill-conditioned optimization problems.","Beyond the paper: the complexity analysis assumes fault-tolerant hardware, so the paper does not establish a near-term speedup; real queue times, gate errors, and measurement overhead could erase any theoretical gain.","Beyond the paper: a direct comparison of CVQLS against classical sparse direct solvers on the same KKT systems, measuring wall-clock time and accuracy, would clarify when the quantum route actually wins."],"forward_implications":["Every expensive linear solve inside IPM-based OPF can be replaced by a quantum subroutine, leaving the classical parts of IPM at roughly $O(n)$ per iteration.","CVQLS becomes the preferred quantum linear solver for ill-conditioned power-system matrices, so future quantum OPF work can focus on it rather than on HHL or VQLS.","Warm-starting variational parameters across IPM iterations turns a sequence of slowly changing linear systems into a tractable variational task.","The $\\mu$-correction is a necessary ingredient for inexact or noisy linear solvers; without it, most random demand scenarios fail optimality or feasibility tests.","Under ideal fault-tolerant hardware assumptions, the per-iteration quantum solve complexity grows polynomially in condition number and accuracy but only logarithmically in system size."],"supporting_citations":[{"why":"Introduces the HHL quantum linear solver, the baseline whose fault-tolerance requirements motivate the variational alternative.","marker":"[12]"},{"why":"Supplies the variational quantum linear solver and its local Hadamard-test cost function, which CVQLS builds upon.","marker":"[13]"},{"why":"Provides the coherent variational quantum linear solver implementation whose coherence-preserving scheme the paper adopts.","marker":"[24]"},{"why":"Supplies the standard OPF formulations and test systems used for reference solutions and comparison.","marker":"[26]"},{"why":"Gives the reduced KKT system and Hessian expressions used to build the Newton system solved by CVQLS.","marker":"[28]"},{"why":"Provides the interior point method background and the $O(\\log(1/\\varepsilon))$ iteration complexity used in the analysis.","marker":"[11]"},{"why":"Establishes the classical IPM complexity estimates that the quantum speedup claim is measured against.","marker":"[9]"},{"why":"Details the Pauli spin matrix decomposition used to encode the Hermitian Hessian into quantum circuits.","marker":"[30]"},{"why":"Analyzes cost-function-dependent barren plateaus, motivating the ansatz and initialization choices.","marker":"[23]"},{"why":"Discusses variational quantum algorithms and ansatz dependence, framing the trainability challenges addressed here.","marker":"[17]"}],"fun_headline_variants":["CVQLS-augmented IPM for reliable optimal power flow","Quantum linear solver stabilizes power flow optimization","CVQLS outperforms HHL and VQLS in power flow","Central path correction keeps quantum IPM on track"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The large-system scalability results rest on the assumption that injecting simulated device noise into the classical Hessian and right-hand side reproduces what CVQLS would actually do on noisy quantum hardware; if that proxy is wrong, the 118- and 300-bus results do not demonstrate the method scales.","fun_headline_variants_meta":{"raw":{"variants":["CVQLS-augmented IPM for reliable optimal power flow","Quantum linear solver stabilizes power flow optimization","CVQLS outperforms HHL and VQLS in power flow","Central path correction keeps quantum IPM on track"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000445,"raw_usage":{"total_tokens":2286,"prompt_tokens":1017,"completion_tokens":1269,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":1202}},"tokens_in":633,"tokens_out":1269,"duration_ms":10871,"temperature":1.0,"reasoning_tokens":1202,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T12:29:14.998070+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same CVQLS-augmented IPM on a real noisy quantum processor, or on a hardware-validated noise model calibrated with measured gate errors, crosstalk, and readout errors, for a system of at least ten buses. If the objective value or the KKT residuals deviate sharply from the paper's noise-injected simulations while the classical IPM converges, the central scalability claim is falsified.","supporting_citations":[{"cited_title":"Variational quantum linear solver,","cited_arxiv_id":null,"evidence_quote":"Supplies the variational quantum linear solver and its local Hadamard-test cost function, which CVQLS builds upon."},{"cited_title":"[Online]","cited_arxiv_id":null,"evidence_quote":"Provides the coherent variational quantum linear solver implementation whose coherence-preserving scheme the paper adopts."},{"cited_title":"Matpower 6.0 user’s manual,","cited_arxiv_id":null,"evidence_quote":"Supplies the standard OPF formulations and test systems used for reference solutions and comparison."},{"cited_title":"An interior point method based optimal power flow,","cited_arxiv_id":null,"evidence_quote":"Gives the reduced KKT system and Hessian expressions used to build the Newton system solved by CVQLS."},{"cited_title":"Interior point methods 25 years later,","cited_arxiv_id":null,"evidence_quote":"Provides the interior point method background and the $O(\\log(1/\\varepsilon))$ iteration complexity used in the analysis."},{"cited_title":"A parallel quadratic programming method for dynamic optimization problems,","cited_arxiv_id":null,"evidence_quote":"Establishes the classical IPM complexity estimates that the quantum speedup claim is measured against."},{"cited_title":"Cost function dependent barren plateaus in shallow parametrized quantum circuits,","cited_arxiv_id":null,"evidence_quote":"Analyzes cost-function-dependent barren plateaus, motivating the ansatz and initialization choices."},{"cited_title":"Variational quantum algorithms,","cited_arxiv_id":null,"evidence_quote":"Discusses variational quantum algorithms and ansatz dependence, framing the trainability challenges addressed here."}],"review_version":1}