{"id":"8361acc4-20ea-4e6c-9608-df3df131d8dd","arxiv_id":"2607.24468","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"Under Gompertz growth with L1–L∞ treatment constraints, minimizers are uniform when growth is constant, maximizers are bang-bang, and in 1D large diffusion they concentrate at a domain endpoint.","lead":"For a stationary tumor model with Gompertz growth, uniform treatment uniquely minimizes total population when growth is constant, while maximizers are always bang-bang and, in 1D at large diffusion, concentrate at a boundary endpoint. The work extends logistic optimal-control theory and reports a diffusion-monotone population effect not seen in the logistic case.","discovery_kind":"extension","skeptic_critique":null,"referee_report":{"model":"moonshotai/kimi-k3","summary":"The paper studies optimization of the total steady-state population J_d(m) = ∫_Ω u for a reaction–diffusion tumor model with Gompertz growth, where the control m (a density-dependent removal/treatment rate) is subject to 0 ≤ m ≤ m̄(x) and ∫m = M. The authors prove existence and uniqueness of a positive steady state with uniform bounds ε̄ ≤ u < K (Prop. 2.1); that for constant growth rate s the constant weight M/|Ω| is the unique minimizer (Thm 1.1, via the w = ln u transformation and Jensen); that every maximizer is of generalized bang-bang type, m* = m̄ χ_{ω*} with ω* a sublevel set of the switching function φ* = u*p* (Thm 1.2, via first- and second-order conditions in the spirit of Mazari–Nadin–Privat and Ferreri–Mazari-Fouquer–Prunier); and that in one dimension with constant s and m̄ ≡ 1, for all sufficiently large d the optimal set is an interval at one endpoint of the domain (Thm 1.3, via a first-order expansion in 1/d recast as a min–min rearrangement problem). Numerical simulations explore heterogeneous s, localized admissible regions, the 2D case, and report a monotone dependence of the optimal value on d, in contrast with the logistic setting.","tokens_in":43429,"tokens_out":7832,"duration_ms":700342,"significance":"If correct, the paper extends the well-developed Mazari–Nadin–Privat theory of total-population optimization from logistic to Gompertz growth, a nonlinearity with singular derivative at zero, and does so under L¹–L∞ constraints with a spatially varying upper bound m̄(x) that may vanish on a set of positive measure — a localized-treatment regime that is open even for the logistic model. The 1D large-diffusion endpoint characterization is a sharp, falsifiable structural result, and the numerically observed monotone dependence of the optimized population on d (in contrast to the fragmentation phenomenon in the logistic case) identifies a genuinely new qualitative feature worth further study. The analysis is supported by clean uniform two-sided bounds on the state (ε̄ ≤ u < K), a complete first/second-order sensitivity and adjoint apparatus, and transparent, reproducible-looking numerics. The contribution is incremental relative to the logistic program but nontrivial and clearly executed.","major_comments":[{"comment":"§4.1, proof of Theorem 1.2 (around Eq. (4.15)): the two load-bearing estimates — J''(m*)[g,g] ≥ A1||g||²_{W^{-1,2}} − B2||g||²_{W^{-2,2}} and its small-support improvement (4.15) — are imported from [16, Propositions A.2, 2.1.6, 2.1.8]. Yet Remark 3.2 states explicitly that the Gompertz nonlinearity 'does not fit the hypotheses on Q' of [16]. If those propositions are purely functional-analytic and independent of the reaction term, the authors should say so and verify the hypotheses in their setting (or reproduce the short proofs); otherwise the second-variation argument contains an unverified step. This is compounded by [16] being an unpublished preprint. This is the only point in the proof of the central Theorem 1.2 that I could not check from the manuscript alone.","section":"§4.1 (proof of Theorem 1.2)"},{"comment":"§7, proof of Theorem 1.3: Theorem 7.2 and Corollary 6.9 yield convergence of m*_d and z_d only up to subsequences, and the limit problem J1 has two maximizers, χ_{(0,M)} and χ_{(1−M,1)}. The proof derives the endpoint-interval conclusion along one convergent subsequence, but the theorem is stated for every d > d̂. The bridging step is missing: one needs the standard contradiction/compactness argument (if the conclusion failed for a sequence d_k → ∞, extract a subsequence converging to one of the two limits and apply the C^1 convergence of z_d to reach a contradiction). Please add this explicitly, and clarify that d̂ is uniform with respect to the choice of maximizer m*_d (the problem may have several).","section":"§7 (proof of Theorem 1.3)"},{"comment":"Proof of Corollary 6.4: the chain '∫η = ∫sη/s > (1/s)∫sη ≥ 0' is not justified as written. Since η is not known to be signed, ∫s(x)η(x)/s(x) dx ≥ (1/s̄)∫(sη)⁺ + (1/s̲)∫(sη)⁻, which does not follow from ∫sη > 0. For constant s the claim ∫η > 0 does follow from (7.10), ∫|∇η|² = U∞ s ∫η, so one fix is to restrict the corollary to constant s; otherwise a genuine argument for ∫η > 0 (e.g. via a Neumann Green-function representation and the normalization (6.6)) is needed. The corollary is not used in the main theorems, but the gap should be closed or the statement weakened.","section":"§6, Corollary 6.4"}],"minor_comments":[{"comment":"Typos/grammar: 'contradicton' and 'impossibile' (proof of Prop. 2.1); 'minumum' and 'Ley us briefly' (Prop. 4.1); 'analize' and 'definiton' (§6, Lemma 6.8); 'possibile' (several places); 'extremals' (Thm 7.2); 'with u is the solution' after (1.9); brace typo in (4.6), '0 < m*(x) < m(x}'.","section":"Throughout"},{"comment":"Cross-references: 'Proposition 3.6' in the proof of Lemma 6.6 should be Lemma 3.6; 'as p → ∞' in the same proof should be 'as d → ∞'. The quantity J∞(m) in Corollary 6.4 is used without definition (presumably ∫U∞ = lim_{d→∞} J_d(m)).","section":"§3, §6"},{"comment":"Hypotheses (1.7), (5.3), (5.6): please state that the pointwise inequalities on m̄ are meant a.e. in Ω. In Proposition 3.1, 'inf_Ω t(x)' should be ess inf. In Remark 2.2(2), the indicated test functions for the Brezis–Oswald argument look garbled ('testing the equation of u1 with u2²/u1 and the equation solved by u2 with u2'); please check.","section":"§1, §2, §3, §5"},{"comment":"Figure 1 (right): the text says the grid is logarithmic in [1e-8, 1] but the axis is drawn linearly on [0,1]; likewise check the axis scaling in Figure 4 (top right), where the logarithmic scale is mentioned only in the caption. A consistent log-axis would make the small-d behavior (the interesting regime) visible.","section":"§8, Figures 1 and 4"},{"comment":"The numerical optimizer is a local method (projected L-BFGS with best-of-several restarts); the manuscript is mostly careful to phrase conclusions as numerical evidence, and the honest discussion of the delicate small-d branch in Figure 6 is appreciated. I suggest one sentence in §8.1 stating tolerances/grid size and that global optimality of the computed candidates is not certified.","section":"§8.1"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a competent math.AP optimal-control paper that takes the Lou / Mazari–Nadin–Privat / Nagahara–Yanagida total-population program and redoes it for stationary Gompertz kinetics. The three load-bearing theorems are real and carefully proved: unique constant minimizer when s is constant (Thm 1.1, via the log change of variables plus Jensen), bang-bang maximizers under a spatially varying upper bound m (Thm 1.2, second-variation argument adapted to the log singularity and the possible vanishing set of m), and 1D large-d endpoint support (Thm 1.3, first-order expansion + rearrangement). Existence/uniqueness of the positive steady state for every d > 0 is cleaner than the logistic case and is handled properly.\n\nWhat is genuinely new is the Gompertz-specific analysis (infinite derivative at zero, always-existing positive state independent of the sign of M) and the fact that the upper bound m is allowed to vanish on a positive-measure set. The numerics are useful: they confirm bang-bang structure, explore heterogeneous s and localized admissible regions, and show a monotone dependence of the optimized total population on d that does not appear in the logistic literature. That last claim is only computational; the paper is honest that small-d theory and heterogeneous minimizers remain open.\n\nSoft spots are proportional and mostly already flagged by the authors. Theorem 1.3 needs constant s, m ≡ 1, and d large enough that the 1/d expansion controls the true maximizer; outside that package the endpoint conclusion need not hold. No code is shipped, so the monotone-J*(d) observation is not independently checkable. Citation pattern is appropriate: the logistic precursors are credited and the self-citations are background, not load-bearing.\n\nThis is for people already working on spectral/optimal-control problems for reaction–diffusion population models or mathematical oncology. A serious referee should see it. I would accept it for peer review and would cite the bang-bang and constant-minimizer results if I were writing in the area.","headline":"Solid, incremental extension of the logistic total-population program to Gompertz: unique constant minimizer, bang-bang maximizers, and 1D large-d endpoint support are proved cleanly; the “new vs logistic” monotonicity claim is only numerical.","tokens_in":42386,"tokens_out":543,"would_cite":true,"duration_ms":10621,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35J25","35B40","35Q92","49J20"],"pacs":[],"model":"grok-4.5","headline":"Under Gompertz growth, the worst treatment for total tumor load is bang-bang; with constant growth and large diffusion it sits at one endpoint of the domain.","keywords":["Gompertz growth","reaction-diffusion equations","optimal control","bang-bang controls","tumor population","switching function","L1-L∞ constraints"],"falsifier":"In one dimension with constant growth and intensity bound equal to 1, compute a maximizer for a sequence of large diffusion values; if the positivity set fails to converge to an interval of length M glued to 0 or to 1, the large-diffusion characterization is false.","tokens_in":42529,"feed_emoji":"ღ","tokens_out":857,"duration_ms":14550,"temperature":0.7,"pith_summary":"This paper studies how to place a limited treatment budget so as to minimize or maximize the total steady-state tumor population when cells follow Gompertz growth and diffuse in space. The treatment is a density-dependent removal rate constrained in both intensity and total mass. When the intrinsic growth rate is spatially constant, the unique minimizer is the uniform distribution of treatment. For the opposite problem of maximizing total population, every optimal control is bang-bang: it saturates the local intensity bound on a measurable set of the right total mass and vanishes elsewhere. In one space dimension, when diffusion is large enough and the intensity bound is constant, that active set is simply an interval glued to one endpoint of the domain. Numerical experiments confirm the bang-bang structure, show how heterogeneity and restricted treatment regions reshape the optimizers, and reveal that the optimized total population varies monotonically with the diffusion coefficient—unlike the logistic models previously studied.","feed_headline":"Worst tumor treatment is bang-bang; large diffusion pins it to an end","feed_subtitle":"Gompertz models force maximizers onto endpoint intervals and reverse the logistic diffusion trend","key_machinery":"The switching function φ = u p, where u is the unique positive steady state and p solves the adjoint linearization; first- and second-order Gateaux derivatives of the total-mass map with respect to admissible control perturbations force any maximizer to be extreme and to coincide with a sublevel set of φ.","core_discovery":"Every maximizer of total steady-state tumor mass under L1–L∞ treatment constraints is of bang-bang type, equal to the pointwise intensity bound on a measurable set that is a sublevel set of the switching function formed by the product of state and adjoint; when growth is constant, the unique minimizer is the spatially uniform treatment of the same total mass.","pith_inferences":["The same large-diffusion rearrangement argument should place maximizers near corners of a square, matching the two-dimensional numerics already shown.","If the small-diffusion limit of the optimized mass can be shown to equal the large-diffusion limit only for constant controls, fragmentation is excluded for every diffusion rate.","The free-boundary regularity of the bang-bang interface remains open and would follow from adapting existing free-boundary techniques once the switching function is known to be non-degenerate."],"forward_implications":["Uniform treatment is the unique best strategy for minimizing total load when growth is homogeneous and the intensity bound permits it.","The worst treatment always concentrates maximal intensity on a proper subset whose geometry is read off the switching function.","For large motility in one dimension the worst placement is an endpoint interval, giving an explicit geometric rule.","Optimized total population decreases monotonically with diffusion, ruling out the small-diffusion fragmentation seen in logistic models.","Heterogeneous growth or restricted admissible regions force the active set to align with high-growth zones or with the admissible support, still in bang-bang form."],"fun_headline_variants":["Maximizers of tumor mass are bang-bang under Gompertz growth","Uniform treatment uniquely minimizes steady tumor when growth is constant","Large diffusion pins bang-bang maximizers to a domain endpoint","Optimal tumor-mass maximizers equal intensity bound on a sublevel set","Diffusion raises optimized total population unlike the logistic case"],"cache_read_input_tokens":32896,"weakest_assumption_plain":"The sharp one-dimensional claim that the active set sticks to an endpoint requires constant growth rate, constant intensity bound, and diffusion large enough that the first-order expansion in 1/d already controls the true maximizer.","fun_headline_variants_meta":{"raw":{"variants":["Maximizers of tumor mass are bang-bang under Gompertz growth","Uniform treatment uniquely minimizes steady tumor when growth is constant","Large diffusion pins bang-bang maximizers to a domain endpoint","Optimal tumor-mass maximizers equal intensity bound on a sublevel set","Diffusion raises optimized total population unlike the logistic case"]},"model":"grok-4.5","effort":"low","cost_usd":0.003295,"raw_usage":{"total_tokens":1129,"prompt_tokens":761,"num_sources_used":0,"completion_tokens":87,"cost_in_usd_ticks":32948000,"prompt_tokens_details":{"text_tokens":761,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":281,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":761,"tokens_out":87,"duration_ms":5034,"temperature":1.0,"reasoning_tokens":281,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T14:01:29.811740+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"In one dimension with constant growth and intensity bound equal to 1, compute a maximizer for a sequence of large diffusion values; if the positivity set fails to converge to an interval of length M glued to 0 or to 1, the large-diffusion characterization is false.","supporting_citations":[],"review_version":1}