{"id":"8e408e43-2eef-4609-a8c9-8aa8b9c69ca5","arxiv_id":"1908.08846","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A reduced-basis method for parameterized stationary Maxwell optimal control with Gauss's law is proven to converge uniformly as snapshot samples densify, and is equipped with residual-based a posteriori error bounds.","lead":"This paper develops and analyzes a reduced-basis method for optimal control problems governed by parameterized stationary Maxwell's equations with Gauss's law, proving uniform convergence and residual-based a posteriori error estimates. It matters because many-query electromagnetic design problems require cheap surrogate models that come with rigorous error control.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 6.2 applies discrete coercivity (3.9) to E_N−Ê_h, which is not in the discrete divergence-free space; the residual R_E vanishes on gradients, so the a posteriori estimator of Theorem 6.5 misses a divergence residual and is not established.","rationale":"The reader's conditional verdict focused on snapshot consistency and reduced coercivity; those are legitimate gaps but likely fixable with standard arguments. The stress test found a more direct flaw: Lemma 6.2 uses the full-space discrete coercivity (3.9) on a function that is not discrete divergence-free, and the defined residual does not control the divergence part of the error. This invalidates the proof of Theorem 6.3 and the absolute estimator of Theorem 6.5 as stated. Because the a posteriori estimator is one of the paper's two central results, the appropriate verdict is REJECT pending a corrected estimator that includes a divergence residual or a proof that the missing term is controlled. This is not a stylistic or completeness concern; it is a correctness defect in a displayed lemma and its applications.","tokens_in":20424,"tokens_out":18084,"duration_ms":184652,"concrete_test":"Run a scalar or 3D edge-element experiment on a coarse mesh with V_N a proper subspace of V_h. Pick ρ with zero V_N-projection but nonzero V_h-projection, set u_N=0 and E_N=0, and solve (6.1) for Ê_h. If ‖Ê_h−E_N‖_{H(curl)}>0 while ‖R_E‖_*=0, Lemma 6.2 and the derived estimator are refuted. Alternatively, augment the estimator with the divergence residual term and check whether the corrected bound restores effectivity.","verdict_should_be":"REJECT","load_bearing_attack":"Section 6's a posteriori theory hinges on Lemma 6.2, which asserts ‖E_N−Ê_h‖_{H(curl)} ≤ C_Ω^σ ‖R_E‖_* and the analogous bound for F. The proof invokes the discrete coercivity (3.9), but (3.9) is stated only for v_h∈D_h^{(ǫ)} = {E_h∈E_h : (ǫE_h,∇φ_h)=0 ∀φ_h∈V_h}. While Ê_h from (6.1) is fully divergence-free, the reduced state E_N only satisfies the second equation of (4.4) against V_N⊂V_h. Thus v=E_N−Ê_h is generally not in D_h^{(ǫ)}, and the inequality ‖v‖^2_{H(curl)} ≤ C_Ω^σ(σ^{-1}∇×v,∇×v) does not follow. Moreover, because u_N satisfies ∇·(ǫu_N)=0, R_E(∇φ_h)=(ǫu_N,∇φ_h)=0 for all φ_h∈V_h; the residual is blind to gradient components. If v is a nonzero gradient (nonzero divergence residual, zero curl residual), then R_E=0 while ‖v‖>0, so Lemma 6.2 is false in that configuration. A concrete instance: choose ρ orthogonal to V_N but not to V_h, set u_N=0 and E_N=0; then (4.4) holds, Ê_h is a nonzero curl-free solution of (6.1), and R_E=0. Since Theorems 6.3 and 6.5 use Lemma 6.2 to bound true errors by residual norms, the advertised absolute a posteriori estimator is not established; an additional term involving the discrete divergence residual D(φ_h)=(ǫE_N,∇φ_h)+(ρ,φ_h) on the full space V_h is required.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a reduced basis method for control-constrained optimal control problems governed by a parameterized stationary Maxwell system with Gauss's law. The state is discretized with Nédélec finite elements, the control is treated by variational discretization, and the reduced spaces are built from state, adjoint, and Helmholtz gradient snapshots. The main theoretical results are a uniform convergence theorem for the reduced optimal control as the snapshot set becomes dense (Theorem 5.5) and residual-based absolute and relative a posteriori error estimators for the reduced state, adjoint, control, and cost functional (Theorems 6.3, 6.4, and 6.5). The convergence result is conditional on a snapshot consistency assumption, and the a posteriori results are derived from estimates involving the state and adjoint residuals.","tokens_in":20891,"tokens_out":6486,"duration_ms":69049,"significance":"If the advertised results were fully established, the paper would be a useful contribution to certified reduced-order methods for Maxwell-type optimal control: it identifies a parameter-separability mechanism, gives an explicit convergence rate in terms of the snapshot density and Hölder exponents, and proposes residual-based estimators that avoid solving the high-dimensional problem. The main theorems are stated with detailed proofs, and the convergence rate is concrete. However, the central a posteriori error estimates rest on a lemma whose proof is invalid, and the convergence argument relies on an unproved consistency property. The scope of the claimed contribution is therefore currently not met, though the framework appears repairable by adding a divergence residual term and by supplying the missing consistency proof.","major_comments":[{"comment":"The proof of the upper bound in Lemma 6.2 applies the discrete coercivity inequality (3.9) to v = E_N^*(μ) - Ê_h(μ). However, (3.9) is stated only for v_h in D_h^{(ε)}, the discrete ε-divergence-free subspace of E_h, and E_N^* satisfies the divergence condition (4.5b) only against test functions in V_N, not against the full V_h. Consequently E_N^* - Ê_h is not generally in D_h^{(ε)}, and the first inequality in Lemma 6.2 is not justified. The issue is load-bearing: since R_E(∇φ_h) = (ε u_N^*, ∇φ_h) = 0 for all φ_h ∈ V_h, the residual R_E is blind to gradient components. In the configuration u_N^* = 0, E_N^* = 0, with ρ orthogonal to V_N but not to V_h, the reduced equations (4.4) hold, Ê_h is a nonzero curl-free solution of (6.1), and R_E = 0; the asserted upper bound would then imply ‖Ê_h‖_{H(curl)} = 0, which is false. Theorems 6.3, 6.4, and 6.5 all inherit this gap. The estimator must include a discrete divergence residual term such as D(φ_h) = (ε E_N^*, ∇φ_h) + (ρ, φ_h) on the full space V_h, and Lemma 6.2 must be reproved with that term.","section":"Section 6, Lemma 6.2"},{"comment":"The uniform convergence theorem is conditional on the snapshot consistency assumption u_h^*(μ) = u_N^*(μ) for all μ ∈ P_N, stated immediately before Theorem 5.5 and justified only by a reference to [1, p. A283]. This property is nontrivial in the present setting because the reduced state and adjoint satisfy the divergence equations in (4.5) only against V_N, not against the full V_h, and the greedy spaces include gradient components precisely to recover the full discrete Gauss law. If this consistency fails, the dense-snapshot argument in the proof of Theorem 5.5 collapses. The authors should either prove the consistency property for their construction of E_N and V_N or state and verify the precise result from [1] that implies it.","section":"Section 5, Theorem 5.5"},{"comment":"Lemma 5.4 supplies the parameter-Hölder continuity of the discrete and reduced optimal controls that is used in the proof of Theorem 5.5, but its proof is omitted with the remark that it follows exactly as in the continuous case. Since this lemma is load-bearing for the convergence theorem, the manuscript should include the proof or a detailed indication of how the continuous arguments in Lemma 5.3 transfer to the discrete and reduced settings, including the role of the discrete coercivity assumption (3.9) on (E_N, V_N).","section":"Section 5, Lemma 5.4"},{"comment":"The existence of a unique solution to the reduced problem (P_N) is conditioned on the assumption that the coercivity inequality (3.9) holds on (E_N, V_N). No criterion or verification procedure is given for this assumption, and the greedy construction in Algorithm 4.1 does not include an inf-sup or coercivity certification step. Because the reduced spaces contain gradients of snapshots, this condition is not automatic, and the convergence and estimator results depend on it.","section":"Section 4, reduced problem setup"}],"minor_comments":[{"comment":"The constants in Lemma 5.3 and Theorem 6.4 involve quantities such as ‖u‖_{R^3}, although u is a function; the authors should define this as a uniform bound on the pointwise box constraints or replace it with max(|u|, |u|).","section":"Section 2 and 3, notation"},{"comment":"In the estimate for J4 + J5 the expression contains a typo: the term should be ‖E_e^*(μ2)‖_{L^2(D)}, not E_e^*(μ2)‖_{L^2(D)} without a norm.","section":"Section 5, Lemma 5.3 proof"},{"comment":"In the second displayed inequality of Lemma 5.4 the norm notation appears as ‖·‖^2_{L(Ω)}; it should be ‖·‖_{L^2(Ω)}.","section":"Section 5, Lemma 5.4 statement"},{"comment":"The relative estimator (6.12) divides by ‖u_N^*(μ)‖_{L^2(Ω)}; the case ‖u_N^*(μ)‖ = 0 should be treated separately or excluded explicitly.","section":"Section 6, Theorem 6.5"},{"comment":"The proof of Theorem 3.1 is omitted with the comment that it is standard; since the theorem is used to justify the high-dimensional discretization, a short proof sketch or a precise reference for the stated curl and divergence error estimates would improve the paper.","section":"Section 3.3, Theorem 3.1"},{"comment":"There are several typographical issues, including 'exasperated' for 'exacerbated' and a missing footnote marker in the cost functional definition; these should be corrected in revision.","section":"Introduction"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's a posteriori section contains a genuine mathematical gap in Lemma 6.2 that propagates to the main estimator theorems. The fix is identifiable: include a discrete divergence residual term on the full V_h and reprove the bounds. The convergence section also needs either a proof of the snapshot consistency assumption or an explicit verification of the cited result in [1]. These are substantial but appear addressable within the manuscript's intended scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First: the advertised absolute a posteriori estimator is not established. Lemma 6.2 applies the discrete coercivity (3.9) to v = E_N^* - Ê_h, but (3.9) only holds on the discrete divergence-free space D_h^(ε). E_N^* solves the reduced second equation only against V_N ⊂ V_h, so v generally fails the full divergence constraint. The residual R_E annihilates gradients because u_N^* is divergence-free, so a nonzero gradient component of v yields R_E = 0 while ||v||_{H(curl)} > 0. Theorem 6.3's upper bound and Theorem 6.5 collapse. The fix is standard: add a discrete divergence residual term over V_h to the estimator. This is not a cosmetic gap; it is the paper's main new result.\n\nWhat is genuinely there: the convergence theorem 5.5 is a clean dense-snapshot argument, assuming the consistency property u_h^* = u_N^* on snapshots and Hölder continuity of affine coefficients. The first-order optimality systems and the greedy algorithm are carefully laid out. The algebra in the proofs of Theorems 6.3–6.4 is internally consistent except for the missing coercivity.\n\nSoft spots in proportion: Theorem 3.1 is stated without proof ('standard arguments'), Lemma 5.4 is dismissed as 'follows exactly as in the continuous case', and the consistency assumption is not proved here—it is cited to [1, pp. A283]. Those are acceptable in a theory paper only if the cited result covers exactly this setting; I did not check that in detail, but the burden is real. No numerical experiments, so effectivity of even a corrected estimator is unknown.\n\nWho it is for: researchers in certified reduced basis methods for saddle-point PDE-constrained optimization, especially Maxwell. It is an incremental extension of an established template to a new system. With the divergence residual fix and a proof or clean citation for consistency, the paper could be a solid contribution; in its current form I would not rely on the a posteriori bounds.\n\nRecommendation: send to peer review, but expect major revision. The convergence part is likely salvageable; the a posteriori part needs the divergence term.","headline":"First RB treatment of constrained Maxwell optimal control; the convergence part is plausible, but the advertised a posteriori estimator rests on a false application of discrete coercivity and needs a divergence residual term.","tokens_in":21336,"tokens_out":3379,"would_cite":false,"duration_ms":30987,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q61","35Q93","65M60","65M12","65K10","49M25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs a reduced-basis surrogate for control-constrained optimal control of stationary Maxwell's equations with Gauss's law, proves uniform convergence of the reduced optimal controls to the finite-element ones as snapshot…","keywords":["Maxwell's system","Parameterized partial differential equation","Optimal control","Reduced basis method","Model order reduction","Convergence analysis","A posteriori error estimates"],"falsifier":"Run the proposed greedy scheme on a family of Maxwell optimal control problems with Hölder-continuous affine coefficients, and compare ||u*_h(µ) - u*_N(µ)|| to κ_N^γ over a fine parameter grid; if the ratio grows without bound as N increases, or if one exhibits a problem where the reduced solution does not reproduce the finite-element solution at a snapshot, the central convergence claim would be refuted. For the a posteriori bound, compute the residual norms and the true error on random parameters; an unbounded ratio between true error and the estimator would contradict Theorem 6.3.","tokens_in":20234,"feed_emoji":"⚡","tokens_out":6395,"duration_ms":57583,"temperature":0.7,"pith_summary":"This paper builds a reduced-basis surrogate for optimal control problems governed by a parameter-dependent stationary Maxwell system: a control drives the electric field through curl and divergence equations with Gauss's law, while dielectric, magnetic permeability, and charge density depend on parameters from a compact set. The authors prove that if the snapshot parameter set is dense enough and the affine coefficients are Hölder continuous, the reduced optimal controls converge uniformly to the finite-element optimal controls at a rate controlled by the snapshot density. They further prove that the combined error in control, state, and adjoint is bounded above and below by computable norms of the state and adjoint residuals, yielding an absolute a posteriori estimator and, under a mild condition, a relative one. A sympathetic reader would care because fast repeated queries of such optimization problems are otherwise expensive, and the paper supplies the missing convergence and certification theory for a reduced-basis approach in this setting.","feed_headline":"Surrogate Maxwell control converges as snapshots fill parameter space","feed_subtitle":"Dense snapshots and Hölder coefficients give uniform convergence; residual norms certify the error.","key_machinery":"The load-bearing objects are the reduced basis spaces E_N and V_N built by a greedy algorithm from snapshots of the optimal state E*_h(µ_i), the adjoint F*_h(µ_i), and their potential parts, together with parameter separability expansions $σ^{{-1}}$ = Σ Θ_q^σ $σ_q^{{-1}}$, ǫ = Σ Θ_q^ǫ ǫ_q, and analogous expansions for the desired state and control. The argument runs on three mechanisms: the discrete Helmholtz decomposition and the coercivity inequality on the reduced spaces, which makes the constrained saddle-point system well posed; Hölder continuity of the coefficient functions Θ, which turns parameter distance into control distance; and the residual functionals R_E and R_F whose dual norms bound the state and adjoint gaps above and below through the Riesz representation theorem. The residual-based bounds are what make the greedy sampling error estimator computable without solving the high-dimensional problem.","core_discovery":"The central claim is that the reduced-basis solution of the parameterized Maxwell optimal control problem is a faithful surrogate in a uniform sense: for every parameter µ in the compact parameter set, the L² error between the reduced optimal control u*_N(µ) and the finite-element target u*_h(µ) is bounded by C κ_N^γ, where κ_N is the largest distance from a parameter to the snapshot sample and γ is half the smallest Hölder exponent of the affine coefficients. The proof adds and subtracts the solution at the nearest snapshot, using Hölder continuity to convert parameter proximity into control proximity and using the snapshot consistency assumption to cancel the error at sample parameters. For the estimator part, the paper defines feasible state and adjoint fields from the reduced control, forms the residuals of the reduced state and adjoint equations, and proves that the true errors are trapped between two positive multiples of the residual norms; this gives an absolute a posteriori estimator for the control and an analogous bound for the cost functional.","pith_inferences":["Inference: the same dense-snapshot plus Hölder argument should transfer to other constrained saddle-point optimal control problems, such as Stokes or mixed elasticity, whenever a discrete coercivity condition and snapshot consistency hold.","Inference: the residual-based estimator could be used online as a genuinely a posteriori stopping criterion, but the paper does not include numerical tests; a natural experiment is to monitor the ratio of true error to estimator as the greedy algorithm enriches the basis.","Inference: if the snapshot consistency assumption fails, the convergence rate would need an extra interpolation term; testing with a discretization known not to reproduce reduced solutions exactly would isolate that dependence."],"forward_implications":["If the snapshot parameters are dense, the reduced optimal controls converge uniformly in the parameter set at a rate fixed by the smoothness of the parameter-to-coefficient maps.","The state and adjoint residuals provide both upper and lower bounds for the control, state, and adjoint errors, so an online computation can certify how far the reduced solution is from the full finite-element solution.","The cost-functional error is also controlled by the same residual norms with explicit constants, enabling certified evaluation of the objective without solving the full problem.","A relative error estimator holds whenever the absolute estimator is no larger than half the norm of the reduced control, giving a practical stopping criterion for the greedy algorithm.","The theory covers parameter-dependent dielectric, permeability, charge density, desired state, and desired control, so a single reduced basis can be reused across many parameter queries."],"supporting_citations":[{"why":"Supplies the snapshot consistency assumption that the reduced solution reproduces the finite-element optimal control at sample parameters.","marker":"[1]"},{"why":"The empirical interpolation method that yields the affine parameter separability used in the Hölder-continuity argument.","marker":"[5]"},{"why":"States the consistency property of reduced basis schemes that the paper invokes for the state equation.","marker":"[17]"},{"why":"The variational discretization concept used to keep the control infinite-dimensional in the finite-element and reduced problems.","marker":"[23]"},{"why":"Provides the discrete Poincaré-Friedrichs inequality used to establish coercivity of the discrete and reduced saddle-point systems.","marker":"[25]"},{"why":"Supplies the H(curl) and H(div) functional framework and the discrete Helmholtz decomposition used throughout.","marker":"[31]"},{"why":"The Nédélec edge element spaces used to discretize the electric and adjoint fields.","marker":"[33]"},{"why":"Edge element analysis of stationary Maxwell optimal control with Gauss's law, used for well-posedness and the compact embedding result.","marker":"[38]"}],"fun_headline_variants":["Maxwell control surrogates converge with denser snapshots","Snapshot density drives uniform Maxwell control error","Residual-based error certifies reduced Maxwell control","Reduced Maxwell control: uniform bounds from snapshot coverage","A posteriori Maxwell control: snapshot gap sets error"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes that the reduced basis solution agrees exactly with the finite-element solution at every snapshot parameter, a consistency condition cited to another paper rather than proved here; if it fails, the uniform convergence bound no longer follows.","fun_headline_variants_meta":{"raw":{"variants":["Maxwell control surrogates converge with denser snapshots","Snapshot density drives uniform Maxwell control error","Residual-based error certifies reduced Maxwell control","Reduced Maxwell control: uniform bounds from snapshot coverage","A posteriori Maxwell control: snapshot gap sets error"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000348,"raw_usage":{"total_tokens":1868,"prompt_tokens":871,"completion_tokens":997,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":487,"completion_tokens_details":{"reasoning_tokens":923}},"tokens_in":487,"tokens_out":997,"duration_ms":9835,"temperature":1.0,"reasoning_tokens":923,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:28:54.132009+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the proposed greedy scheme on a family of Maxwell optimal control problems with Hölder-continuous affine coefficients, and compare ||u*_h(µ) - u*_N(µ)|| to κ_N^γ over a fine parameter grid; if the ratio grows without bound as N increases, or if one exhibits a problem where the reduced solution does not reproduce the finite-element solution at a snapshot, the central convergence claim would be refuted. For the a posteriori bound, compute the residual norms and the true error on random parameters; an unbounded ratio between true error and the estimator would contradict Theorem 6.3.","supporting_citations":[{"cited_title":"Ali and M","cited_arxiv_id":null,"evidence_quote":"Supplies the snapshot consistency assumption that the reduced solution reproduces the finite-element optimal control at sample parameters."},{"cited_title":"Barrault, Y","cited_arxiv_id":null,"evidence_quote":"The empirical interpolation method that yields the affine parameter separability used in the Hölder-continuity argument."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"States the consistency property of reduced basis schemes that the paper invokes for the state equation."},{"cited_title":"Hinze , A variational discretization concept in control constrain ed optimization: the linear- quadratic case, Comput","cited_arxiv_id":null,"evidence_quote":"The variational discretization concept used to keep the control infinite-dimensional in the finite-element and reduced problems."},{"cited_title":"Hiptmair , Finite elements in computational electromagnetism , Acta Numer","cited_arxiv_id":null,"evidence_quote":"Provides the discrete Poincaré-Friedrichs inequality used to establish coercivity of the discrete and reduced saddle-point systems."},{"cited_title":"Monk , Finite Element Methods for Maxwell’s Equations , New York: Oxford University Press, 2003","cited_arxiv_id":null,"evidence_quote":"Supplies the H(curl) and H(div) functional framework and the discrete Helmholtz decomposition used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Nédélec edge element spaces used to discretize the electric and adjoint fields."},{"cited_title":"Yousept and J","cited_arxiv_id":null,"evidence_quote":"Edge element analysis of stationary Maxwell optimal control with Gauss's law, used for well-posedness and the compact embedding result."}],"review_version":1}