{"id":"99d1ccb2-2163-418c-89dc-12085f5e38e6","arxiv_id":"2505.02282","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"FQAOA-SCLFM, a QAOA variant with a Hartree-Fock driver that encodes the procurement target as a self-consistent local field, beats XY-QAOA and the prior FQAOA in expected cost on all eight small noiseless instances tested.","lead":"An electricity aggregator chooses which households to call on for demand response; this paper adds a self-consistent mean-field local field to a fermionic quantum optimizer so that the chosen portfolio stays close to a target total while minimizing fluctuation. Noiseless simulations on 20 households show lower expected cost than two earlier QAOA variants on all eight tested time windows.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The performance claim is measured against a penalized proxy cost (Eq. 4) that the paper admits can violate the feasibility constraint (Eq. 3); the reported advantage may not transfer to the actual procurement objective.","rationale":"The reader identified the HF linearization in Eq. (15) as the weakest assumption. I do not dispute that concern, but I regard the proxy-objective issue as more load-bearing because it affects the interpretability of even the reported results, not only their generalization. The paper explicitly acknowledges that Eq. (4) can violate Eq. (3), so the central claim of 'stable procurement' is assessed on a metric that is admitted to permit infeasible portfolios. The HF approximation, by contrast, is a heuristic for constructing a driver and initial state, and its usefulness on the tested instances is supported by the numerical results; its failure would only weaken robustness beyond the tested scope. The proposed brute-force re-evaluation on the original constrained objective would settle whether the Table I ranking corresponds to actual procurement quality. Since the reader's verdict is already CONDITIONAL and the concern sharpens the condition, I keep the same verdict rather than moving it.","tokens_in":11025,"tokens_out":10057,"duration_ms":132136,"concrete_test":"For each time period T, enumerate all C(20,5)=15,504 cardinality-M portfolios (or solve the QUBO exactly) and compute: (i) the true constrained optimum of Eqs. (1)-(3), (ii) the fraction of FQAOA-SCLFM, FQAOA, and XY-QAOA samples at p=10 that violate Eq. (3), and (iii) the expected variance E[σ^2_{t,tot}] restricted to feasible samples. If FQAOA-SCLFM's lower ΔE_T/W_T in Table I is accompanied by an infeasibility rate no larger than the baselines and a smaller gap to the true constrained optimum, the concern is resolved; otherwise the performance claim is an artifact of the proxy cost. This check uses only data already implied by the paper's model parameters and requires no new hardware.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is about stable procurement of a target negawatt, but the quantity actually minimized and reported in Table I is the penalized mean-squared-error proxy E_{T,x} of Eq. (4), with the penalty coefficient fixed to 1. Section II states that this cost function 'guarantees efficiency and feasibility ... while violating the inequality constraint in Eq. (3)'. Thus the paper's own text concedes that the optimized portfolios need not satisfy P_{t,proc} ≤ Σ_l E[p_{t,l}] x_l ≤ P_{t,proc} + δ. The relative ranking of FQAOA-SCLFM, FQAOA, and XY-QAOA is therefore a ranking on a surrogate objective, not on the original variance-minimization problem of Eqs. (1)-(3). No argument is given that the minimizer of Eq. (4) at unit penalty weight is close to the true constrained optimum, or that the lower ΔE_T values in Table I correspond to more feasible or lower-variance feasible portfolios. Figure 3 shows final states roughly near P'_{t,proc}, but 'roughly' is not quantified as an infeasibility rate, and the cost expectation in Eq. (26) is measured relative to E_{T,min}, the minimum of the proxy, not of the constrained problem. If the proxy is a poor Lagrangian relaxation, the headline advantage could be an artifact of optimizing an infeasible objective.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes FQAOA-SCLFM, a fermionic QAOA variant whose driver Hamiltonian is augmented by a self-consistent local field obtained from a Hartree-Fock approximation of the soft target-power constraint. The method is applied to an electricity demand portfolio problem in which a binary vector x selects M DR participants and the objective is to procure a target negawatt with small variance. The paper introduces a penalized quadratic cost function, maps it to a fermionic Hamiltonian, linearizes the penalty term in the driver through Eq. (15), and reports noiseless simulations on eight time windows using a 20-participant dataset. Table I reports that FQAOA-SCLFM achieves lower expected penalized costs than XY-QAOA and the previous FQAOA at QAOA levels p=0, 1, and 10 across all eight windows.","tokens_in":11393,"tokens_out":8244,"duration_ms":107616,"significance":"The algebraic derivation of Eq. (15) as a linearized, self-consistent version of the driver in Eq. (14) is straightforward and appears correct, and the paper provides a complete grid of simulation results over all time windows and three QAOA levels. If the reported advantage survives evaluation against the original inequality-constrained problem, the SCLFM construction is a useful and inexpensive way to encode a soft constraint into the QAOA driver and initial state, adding only pL extra single-qubit rotations. The main limitations are that the benchmark is a surrogate objective with no feasibility statistics, the simulations lack repeated-run statistics, and the instance set is narrow; these limitations currently make the significance conditional.","major_comments":[{"comment":"The problem stated in Eqs. (1)-(3) is variance minimization under the inequality constraint P_{t,proc} <= sum_l E[p_{t,l}] x_l <= P_{t,proc} + delta, but the quantity minimized and reported is the unit-weight penalized proxy E_{t,x} of Eq. (4). Section II itself concedes that this proxy 'violat[es] the inequality constraint in Eq. (3)', yet Table I reports only Delta E_T relative to E_{T,min}, the minimum of the proxy. A lower Delta E_T therefore does not by itself establish better performance on the actual procurement objective. Please report the fraction of sampled portfolios that violate Eq. (3), the distribution of sum_l E[p_{t,l}] x_l for each algorithm, and the expectation of the original variance under the constraint (or a bound quantifying the suboptimality introduced by the unit-weight penalty). Without such data, the claim that FQAOA-SCLFM procures negawatt 'more stably' is not supported.","section":"Section II, Eq. (4) and Table I"},{"comment":"Table I reports a single value per algorithm, instance, and QAOA level, with no description of the number of BFGS restarts, random seeds, or confidence intervals. Several of the claimed advantages are small enough to be optimizer noise, e.g. T=6, p=1 (XY-QAOA 0.060 vs FQAOA 0.056), T=9, p=10 (FQAOA 0.037 vs FQAOA-SCLFM 0.036), and T=0, p=1 (XY-QAOA 0.073 vs FQAOA 0.066). Since the central claim is uniform outperformance over all instances, the paper should provide repeated optimization runs, report best/median statistics, and give error bars or a statistical test.","section":"Section IV.B.2, Table I"},{"comment":"The sentence 'For any given T and finite p, FQAOA outperforms XY-QAOA, and FQAOA-SCLFM further outperforms FQAOA' is contradicted by Table I: at p=0, T=3, FQAOA has Delta E_T/W_T = 0.123 versus 0.117 for XY-QAOA, and at p=0, T=12 the two values are equal at 0.177. Please correct this statement or restrict it to the actual central claim, namely that FQAOA-SCLFM outperforms both baselines in the table.","section":"Section IV.B.2, text after Table I"},{"comment":"The HF driver is obtained by replacing (P_t - P^HF_{t,tot})^2 with zero, but the quality of this approximation is not checked. Since the p=0 advantage of FQAOA-SCLFM appears to come from the HF initial state being close to the proxy optimum, the paper should quantify this closeness, for instance by comparing the HF state energy to the exact E_{T,min} on a few windows or by reporting overlap or energy-gap diagnostics. This is particularly important because E_{T,min} can be computed exactly for L=20, M=5 by brute force, yet the paper does not state how E_{T,min} was obtained.","section":"Section III.B, Eq. (15)"}],"minor_comments":[{"comment":"Equation (A.2) sets n^HF_{l,0} = 1/M_T, which does not describe a uniform distribution with M_T particles among L sites; the sum over l would be L/M_T, not M_T. This should presumably read n^HF_{l,0} = M_T/L, which is the occupation probability of the uniform superposition under the constraint.","section":"Appendix A, Eq. (A.2)"},{"comment":"The stated parameters are inconsistent: P'_{t,proc} = P_{t,proc} + delta/2 with P_{t,proc}=1 and P'_{t,proc}=1.5 implies delta = 1, not delta = 11 as written.","section":"Section IV.A"},{"comment":"Equation (19) denotes the local-field unitary as U_n(beta), while Eq. (22) defines U_I(beta); please make the notation consistent.","section":"Section III.D, Eqs. (19) and (22)"},{"comment":"The phrase 'guarantees efficiency and feasibility ... while violating the inequality constraint in Eq. (3)' is self-contradictory; please rephrase to state clearly that Eq. (4) is an unconstrained penalized surrogate whose minimizer may be infeasible for Eq. (3).","section":"Section II, text after Eq. (4)"},{"comment":"The text contains minor language errors, e.g. 'selecton' should be 'selection' and 'is the difficult to solve' should be 'is difficult to solve'.","section":"Section IV.B.2"}],"recommendation":"major_revision","confidential_remarks":"The core idea is reasonable and the algebraic derivation is sound, but the central empirical claim is currently evaluated against a surrogate objective with no feasibility reporting and no repeated-run statistics. I recommend major revision rather than rejection because these gaps are addressable within the scope of the manuscript; if additional feasibility analysis were to overturn the ranking, rejection would be appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The SCLFM driver is a genuine new mechanism: replacing the quadratic soft constraint in the driver with a self-consistent linear field is not a one-line tweak, and the HF fixed-point loop plus the extra U_I rotation make it a distinct contribution to the FQAOA literature. The derivation of Eq. (15) from Eq. (14) is algebraically fine, and the numerics in Table I are internally consistent: on the penalized cost E_T,x, FQAOA-SCLFM beats both XY-QAOA and the earlier FQAOA on all eight time windows at p=0,1,10.\n\nThat said, the paper's own text concedes the central weakness: Eq. (4) \"guarantees efficiency and feasibility ... while violating the inequality constraint in Eq. (3).\" All reported costs, including the E_T,min baseline in Eq. (26), are measured against this penalized proxy, not against the original variance-minimization-with-inequality problem. So we do not actually know whether the lower ΔE_T values correspond to portfolios that are feasible or closer to the true constrained optimum. The stress-test note is on target here, and the paper does not offer an infeasibility rate or any argument that the unit penalty weight makes the proxy a good Lagrangian relaxation.\n\nOther soft spots, in proportion: the demonstration is tiny (20 variables, noiseless simulation only), there are no error bars or multiple optimization runs, and no code or data is shipped (though the input data is from a cited public database). There is also no serious classical baseline—random sampling under the constraint appears only as an inset in Fig. 4, and a simple local search or greedy would likely do well on these instances. For an algorithmic paper these gaps are survivable, but they undercut any practical claim about stable negawatt procurement.\n\nWhat is worth keeping: the idea of using the soft-constraint structure to build a problem-adapted driver, and solving it self-consistently at the mean-field level, is clever and could transfer to other soft-constrained QUBO-like problems. The p=0 advantage is exactly what you'd expect from a mean-field initialization that already respects the bias toward P'_t,proc, and the paper's explicit acknowledgment of the Eq. (3) violation is to its credit.\n\nMy take: this deserves a serious referee, not a desk reject—the mechanism is novel and the math is sound. But the referee should push for feasibility tracking, a classical baseline, and a statement about what happens to the advantage when Eq. (3) is actually enforced. As it stands, I'd read it as a contribution to QAOA driver design, not as evidence about power portfolio optimization.","headline":"A real new QAOA driver mechanism, but the reported advantage is measured against a penalized proxy that the paper itself concedes can violate the actual feasibility constraint.","tokens_in":11861,"tokens_out":3538,"would_cite":false,"duration_ms":43102,"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":"A fermionic QAOA variant embeds the power-procurement constraint in its driver Hamiltonian and, in noiseless simulation, achieves lower expected cost than two standard QAOA baselines on all eight demand periods tested.","keywords":["electric power demand portfolio optimization","negawatt trading","Fermionic QAOA","self-consistent local field modulation","Hartree-Fock driver Hamiltonian","demand response","constrained quantum optimization","portfolio optimization"],"falsifier":"Run the same noiseless comparison at a procurement target deliberately far from the range of achievable totals (for example, set $P'_{t,\\mathrm{proc}}$ to the maximum achievable total with $M_T=5$, instead of $1.5\\,\\mathrm{kWh}$): if FQAOA-SCLFM's expected cost at $p=1$ no longer beats the previous FQAOA, the SCLFM advantage is conditional on the Hartree-Fock approximation's accuracy rather than intrinsic to the driver construction.","tokens_in":10863,"feed_emoji":"⚡","tokens_out":9329,"duration_ms":107829,"temperature":0.7,"pith_summary":"The paper proposes FQAOA-SCLFM, a version of the Fermionic Quantum Approximate Optimization Algorithm that builds the constraint on the target amount of procured power directly into the driver Hamiltonian, rather than leaving it only to the cost function. It does this by replacing the quadratic penalty on the total-negawatt deviation with a linear local field whose strength is fixed self-consistently by a Hartree-Fock calculation, so the driver's ground state is a single Slater determinant and can be prepared efficiently. Applied to an electricity demand portfolio problem—choosing which demand-response participants to call so that total negawatt matches the desired procurement with low variance—the method yields lower expected cost than XY-QAOA and the previous FQAOA in every tested demand period at QAOA levels $p=0$, $1$, and $10$ in noiseless simulation. If this holds outside the simulated instances, aggregators could procure negawatt portfolios that meet the target more reliably with the same or fewer quantum operations.","feed_headline":"QAOA driver encodes power target, beats both baselines in every case","feed_subtitle":"Fermionic QAOA with self-consistent local field modulation lowers expected cost on all eight demand windows modeled.","key_machinery":"Self-consistent local field modulation (SCLFM): the construction of a free-fermion driver Hamiltonian $$\\hat{H}^{\\mathrm{HF}}_{d,T} = -t_{\\mathrm{hop}}\\sum_l (\\hat{c}^{\\dagger}_{l+1}\\hat{c}_l + \\mathrm{h.c.}) + \\frac{2}{N_T}\\sum_t ($P^{{\\mathrm{HF}}$}_{t,\\mathrm{tot}} - P'_{t,\\mathrm{proc}})\\sum_l \\mathbb{E}[p_{t,l}]\\hat{n}_l - \\frac{1}{N_T}\\sum_t\\left[($P^{{\\mathrm{HF}}$}_{t,\\mathrm{tot}})^2 - (P'_{t,\\mathrm{proc}})^2\\right],$$ with $P^{\\mathrm{HF}}_{t,\\mathrm{tot}}=\\sum_l \\mathbb{E}[p_{t,l}]\\langle \\phi_{\\mathrm{HF}}|\\hat{n}_l|\\phi_{\\mathrm{HF}}\\rangle$ computed self-consistently from the driver's own ground state. This replaces the quadratic soft constraint by a linear local field while keeping the Hamiltonian quadratic in fermionic operators, so the ground state remains a single Slater determinant satisfying the particle-number constraint. The same Hartree-Fock iteration yields the initial state, and the mixing unitary factorizes into number, boundary, odd, and even parts whose implementation adds only $pL$ single-qubit Pauli-$Z$ gates to the previous FQAOA circuit.","core_discovery":"The central discovery, on the paper's own terms, is that the self-consistent local field modulation removes a known defect of the previous FQAOA driver: the soft constraint $(1/N_T)\\sum_t (\\hat{P}_t - P'_{t,\\mathrm{proc}})^2$ is quadratic in the number operator, so its ground state is not a single Slater determinant and the driver violates the design condition that the initial state be efficiently preparable. The proposed driver keeps only the linear part of this penalty, evaluated at a Hartree-Fock fixed point, and solves the resulting free-fermion Hamiltonian self-consistently. In noiseless simulations on a 20-participant, five-request problem fixed to $P'_{t,\\mathrm{proc}}=1.5\\,\\mathrm{kWh}$ across eight three-hour periods, the resulting algorithm reaches expected cost (Eq. 26) below both XY-QAOA and previous FQAOA at every QAOA level tested; at $p=10$ the total negawatt $P_{t,\\mathrm{tot}} \\pm \\sigma_{t,\\mathrm{tot}}$ roughly satisfies the original balance inequality, and at $p=0$ it already sits near the target in the difficult evening period. The additional circuit cost is only $pL$ single-qubit $Z$ rotations.","pith_inferences":["One consequence the authors leave implicit is that most of the advantage is likely coming from the initial state: already at $p=0$ the Hartree-Fock state puts the total negawatt near the target, so a cleaner test would be to keep the SCLFM initial state but swap in the standard FQAOA driver during mixing, isolating the contribution of the driver itself.","The results cover one operating point ($P'_{t,\\mathrm{proc}}=1.5\\,\\mathrm{kWh}$, $M_T=5$, $L=20$) and one residential dataset; a testable extension is to vary the procurement target, particularly to values far from the Hartree-Fock solution, where the linearization of the quadratic penalty should lose accuracy.","On portfolios with many near-degenerate low-cost solutions, the Hartree-Fock fixed point may be a poor proxy for the true optimum; an adaptive scheme that re-estimates the local field from the QAOA output distribution rather than only from the Hartree-Fock state is a natural next step, but it is not what this paper proposes."],"forward_implications":["FQAOA-SCLFM reports lower expected cost than both XY-QAOA and the previous FQAOA in all eight demand periods at QAOA levels $p=0$, $1$, and $10$, so on this problem it is the better performing of the three algorithms at the depths tested.","At $p=10$, the optimized state yields total negawatt $P_{t,\\mathrm{tot}}\\pm\\sigma_{t,\\mathrm{tot}}$ that roughly satisfies the balance condition, meaning procurement risk is reduced enough for near-target delivery.","Because the SCLFM driver is derived from the soft constraint, the same ansatz construction applies to other constrained combinatorial optimizations whose cost has one hard cardinality constraint and one quadratic soft penalty.","The added resource cost is only $pL$ single-qubit Pauli-$Z$ rotations, so the improvement does not require deeper circuits than the previous FQAOA.","The $T=18$ and $T=21$ instances, which have large positive and negative negawatt covariances, show the largest relative improvement, suggesting the benefit is most pronounced on strongly fluctuating demand periods."],"supporting_citations":[{"why":"Supplies the electric power demand portfolio formulation and the model parameters estimated from residential data that define the problem instances.","marker":"[8]"},{"why":"Provides the negawatt procurement method and the supply-demand balance condition that the cost function penalizes.","marker":"[9]"},{"why":"Introduces the original QAOA framework that FQAOA extends.","marker":"[14]"},{"why":"Defines the previous FQAOA ansatz, the driver-design conditions I-III, and the FQAOA baseline results.","marker":"[18]"},{"why":"Gives the circuit implementation of the mixing unitary and the gate-count table used to compute FQAOA-SCLFM overhead.","marker":"[19]"},{"why":"Defines the XY-mixer variant of QAOA used as the XY-QAOA baseline.","marker":"[20]"},{"why":"Provides a constrained optimization demonstration of XY-QAOA that supports using it as a practical baseline on constrained problems.","marker":"[21]"},{"why":"Is the fast quantum circuit simulator used for all noiseless simulation results in Table I and Figs. 3-4.","marker":"[22]"},{"why":"Is the residential electricity-usage database from which the 20 participants' negawatt expectations and covariances are estimated.","marker":"[23]"}],"fun_headline_variants":["FQAOA-SCLFM beats XY-QAOA and prior FQAOA on all test cases","FQAOA driver with local field modulation beats baselines in every test","Fermionic QAOA with SCLFM wins all comparisons in study","Driver encodes power target, FQAOA beats all baselines in every case","New FQAOA variant beats both QAOA baselines on every demand window"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Hartree-Fock estimate of the total procured power is close enough to the true optimum that the quadratic procurement penalty can be replaced by a linear local field, and that the penalized cost function is an acceptable stand-in for the original balance inequality even though it can violate it.","fun_headline_variants_meta":{"raw":{"variants":["FQAOA-SCLFM beats XY-QAOA and prior FQAOA on all test cases","FQAOA driver with local field modulation beats baselines in every test","Fermionic QAOA with SCLFM wins all comparisons in study","Driver encodes power target, FQAOA beats all baselines in every case","New FQAOA variant beats both QAOA baselines on every demand window"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001518,"raw_usage":{"total_tokens":6074,"prompt_tokens":929,"completion_tokens":5145,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":5041}},"tokens_in":545,"tokens_out":5145,"duration_ms":38504,"temperature":1.0,"reasoning_tokens":5041,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:56:37.097620+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same noiseless comparison at a procurement target deliberately far from the range of achievable totals (for example, set $P'_{t,\\mathrm{proc}}$ to the maximum achievable total with $M_T=5$, instead of $1.5\\,\\mathrm{kWh}$): if FQAOA-SCLFM's expected cost at $p=1$ no longer beats the previous FQAOA, the SCLFM advantage is conditional on the Hartree-Fock approximation's accuracy rather than intrinsic to the driver construction.","supporting_citations":[{"cited_title":"A design method for electric power demand portfolio and its basic investigation -Application to design of aggregator portfolio-","cited_arxiv_id":null,"evidence_quote":"Supplies the electric power demand portfolio formulation and the model parameters estimated from residential data that define the problem instances."},{"cited_title":"A design method for electric power demand portfolio and its basic investigation -Proposition of a method for a resource aggregator to procure negawatt effectively-","cited_arxiv_id":null,"evidence_quote":"Provides the negawatt procurement method and the supply-demand balance condition that the cost function penalizes."},{"cited_title":"Fermionic quantum approximate optimization algorithm","cited_arxiv_id":null,"evidence_quote":"Defines the previous FQAOA ansatz, the driver-design conditions I-III, and the FQAOA baseline results."},{"cited_title":"Experimental Demon- stration of Fermionic QAOA with One-Dimensional Cyclic Driver Hamiltonian","cited_arxiv_id":null,"evidence_quote":"Gives the circuit implementation of the mixing unitary and the gate-count table used to compute FQAOA-SCLFM overhead."},{"cited_title":"XY mixers: analytical and numerical results for the quantum alternating operator ansatz","cited_arxiv_id":null,"evidence_quote":"Defines the XY-mixer variant of QAOA used as the XY-QAOA baseline."},{"cited_title":"Constrained Quantum Optimization for Extractive Summarization on a Trapped-ion Quantum Computer","cited_arxiv_id":"2206.06290","evidence_quote":"Provides a constrained optimization demonstration of XY-QAOA that supports using it as a practical baseline on constrained problems."},{"cited_title":"Qulacs: a fast and versatile quantum circuit simulator for research purpose","cited_arxiv_id":null,"evidence_quote":"Is the fast quantum circuit simulator used for all noiseless simulation results in Table I and Figs. 3-4."},{"cited_title":"In this paper, we follow the study [8] and use the electricity usage of 20 residences in September 1-15, 2003 as the amount of negawatt","cited_arxiv_id":null,"evidence_quote":"Is the residential electricity-usage database from which the 20 participants' negawatt expectations and covariances are estimated."}],"review_version":1}