{"id":"0d5dd9b5-0203-40a8-99d2-7be3368e6034","arxiv_id":"2506.02808","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For Poisson optimal control with transport regularization, optimal controls are shown to lack interior atoms under smooth costs, to be induced by a continuous transport map under curvature conditions, and to be absolutely continuous in many metric-cost cases.","lead":"This paper studies optimal control of the Poisson equation where the control is a measure and the cost penalizes the transport distance to a given prior. It proves that optimal controls have regular structure: no interior atoms for smooth transport costs, and absolute continuity for Wasserstein-1 costs.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the optimality system is sound, and the structural theorems are valid under their explicitly stated adjoint-regularity hypotheses, whose limited verifiability is already noted by the reader.","rationale":"The paper's central claim, Theorem 4.2, is a standard adjoint/subdifferential derivation and is supported by complete proofs. The structural theorems are honestly conditional, with the extra adjoint regularity stated as assumptions and not disguised as consequences of Theorem 4.2. The reader's weakest-assumption analysis identifies exactly the right scope limitation, but a limitation of applicability is not a correctness defect. Sections 5, 6, and 7 were checked for sign errors, exponent mismatches, and hidden assumptions; none were found. The stress-test therefore does not change the reader's ACCEPT verdict.","tokens_in":85,"tokens_out":46283,"duration_ms":1145868,"concrete_test":"Independently re-derive Proposition 7.5 from Theorem 5.9 and standard elliptic regularity: starting from y in L^{∞}(Ω), verify that the adjoint p solving -Δp = y - y_d is in W^{2,r}_{loc}(Ω) for all r < ∞ on a bounded Lipschitz domain; if this bootstrap fails on some admissible domain, then the regularity hypothesis of Theorem 7.2 is not guaranteed in the intended tracking-type application.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No load-bearing flaw was found in the central argument. Theorem 4.2 follows from the convexity of the transport-regularization functional and the Fenchel-conjugate characterization in Lemma A.1/Proposition A.3; the Kantorovich duality step and the support inclusion in Lemma 4.4 are consistent. The main limitation, already identified by the reader, is that several structural results (Theorems 6.3, 7.2, 8.1, 8.3) assume regularity of the adjoint state beyond what Theorem 4.2 guarantees, such as p in W^{2,r}_{loc} with r > d or continuous differentiability. These hypotheses are stated explicitly and are verified in special settings (Corollary 6.7, Proposition 7.5), so the paper does not overclaim: it presents conditional structural theorems rather than unconditional ones. I also checked the more technical estimates in Sections 5 and 8 and found no internal inconsistency that would invalidate the conclusions as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies an optimal control problem for the Poisson equation in which the control is a Borel measure on a compact set ω1 and the Tikhonov regularization is the generalized transportation distance D^c_{u0}(u) to a fixed prior. Existence of optimal controls is proved by the direct method, and a first-order optimality system is derived via Fenchel conjugation and Kantorovich duality: Theorem 4.2 gives an adjoint state p satisfying the adjoint Poisson equation and the inclusion 0 ∈ p + α∂D^c_{u0}(ū), with the subdifferential characterized by a dual Kantorovich problem. The associated support inclusion for optimal transport plans is then used to derive structural properties of optimal controls. The main results are: non-atomicity of optimal controls in the interior of ω1 for smooth tracking-type costs (Section 5); existence and regularity of an optimal transport map under strong convexity of the cost and a curvature condition on the adjoint state (Section 6); absolute continuity of the optimal control for power-type costs when the prior is absolutely continuous and the adjoint has additional W^{2,r} regularity (Section 7); and, for metric costs, non-atomicity or absolute continuity with respect to H^{d-1} under different regularity assumptions on the adjoint state (Section 8). Two worked examples are provided to show sharpness.","tokens_in":35292,"tokens_out":8210,"duration_ms":92231,"significance":"If the results are correct, the paper gives a systematic and largely self-contained PDE-constrained framework for extracting structural information—non-atomicity, existence of transport maps, absolute continuity—from optimality conditions when the control space is a space of measures with transport regularization. The proofs are detailed, the constants in the estimates are explicit, and the conditional structural theorems are phrased with clearly stated hypotheses on the adjoint state. A particular strength is that the paper does not overclaim: the extra adjoint regularity needed in Theorems 6.3, 7.2, 8.1, and 8.3 is stated explicitly, and the paper itself verifies those hypotheses in several special settings (Corollary 6.7, Proposition 7.5). The companion numerical paper [5] is used only as confirmation, not as an ingredient in the proofs. The examples in Section 8 demonstrate that the results are close to sharp.","major_comments":[],"minor_comments":[{"comment":"The paper repeatedly writes 'ω1 = Ω' (for instance in Theorem 5.9, Theorem 7.4, and Proposition 7.5), although Assumption 2.1 fixes ω1 to be compact and Ω to be an open bounded Lipschitz domain. Please state explicitly that these statements are to be read with ω1 = \\overline{Ω}, or introduce a closure convention at the beginning of the paper.","section":"Sections 5 and 7"},{"comment":"The proof uses [33, Lemma 5.8] and asserts that the smoothness assumption on the boundary is only needed to check unique solvability of the Poisson equation, so that the lemma extends to Lipschitz domains via Lemma 3.1. Since this step is load-bearing for the H^1_0-regularity and capacity conclusions, please expand the argument into a short proof or provide a precise reference for the Lipschitz-domain version.","section":"Theorem 5.9"},{"comment":"The sentence 'the optimal objective value of the dual Kantorovich problem a does not change' contains a stray 'a'; please correct this typo.","section":"Remark 4.3"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is a careful and substantive contribution that fits the journal well. The remaining issues are local and presentational; the notation ω1 = Ω should be clarified before publication. I saw no indication of citation-pattern or disclosure problems."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper to know: Meyer and Wachsmuth give the first systematic structural analysis of optimal control of the Poisson equation with a transport distance (Wasserstein-type) as Tikhonov regularization in the control space of measures. They prove existence, derive first-order optimality conditions via Fenchel conjugation and Kantorovich duality, and then extract structural properties of optimal controls: non-atomicity in the interior for smooth costs, existence of a continuous transport map under a strong convexity plus curvature condition, and absolute continuity results for power-type and metric costs. The proofs are detailed and appear correct; the constants are explicit and there are no fitted parameters. The companion paper [5] is cited only for numerical confirmation, not as an input, so no circularity.\n\nThe strongest part is the honest treatment of scope. Theorems 6.3, 7.2, 8.1, and 8.3 are conditional on extra regularity of the adjoint state (W^{2,r}_{loc} with r > d, or C^1 with Lipschitz gradient), which the optimality system in Theorem 4.2 does not guarantee. The authors verify these hypotheses only in special settings, e.g., tracking objectives with separated observation domain (Corollary 6.7) or the full-domain W2 case (Proposition 7.5). They flag these limitations explicitly in remarks. That is not a flaw in the paper; it is a clear boundary condition on the results. A reader should not expect unconditional structural theorems.\n\nThere are minor technical weak points. In Theorem 5.9 the proof invokes [33, Lemma 5.8] which is stated for smooth boundaries, and the authors argue the smoothness is only used for unique solvability, replaceable by their Lemma 3.1. That is probably fine, but a referee may want the replacement spelled out. The boundary regularity results (Theorems 7.4 and 8.3) rely on a coarea formula for Lipschitz domains; the arguments are plausible but somewhat compressed.\n\nOverall, this is a serious, carefully written paper. It will be useful to researchers in measure-valued PDE-constrained optimization, especially those interested in sparse or structured controls. It deserves a serious referee and, assuming the minor points are addressed, publication.","headline":"Solid, honest structural analysis of measure-valued optimal control with transport regularization; the main theorems are conditional on adjoint-regularity hypotheses the authors flag clearly.","tokens_in":35705,"tokens_out":2481,"would_cite":true,"duration_ms":24377,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49K20","49N60","49J20","35J08"],"pacs":[],"model":"deepseek-v4-flash","headline":"Transport-regularized PDE control: adjoint state carries the geometry","keywords":["optimal control of PDEs","measure-valued control","optimal transport regularization","Wasserstein distance","Kantorovich duality","first-order necessary optimality conditions","transport map","sparsity of optimal controls"],"falsifier":"Compute the optimal control for a radial, spherically symmetric instance with $d=3$, smooth quadratic transport cost, a tracking-type objective with smooth desired state, and a prior with an $L^\\infty$ density whose support avoids the boundary. Corollary 5.6 predicts no atom at the center; a computed optimal control with a positive Dirac there would refute it, while an atom-free control supports the claim. The same experiment with metric costs and an absolutely continuous prior should instead produce a control with nonzero $H^2$-singular part, as Example 8.4 illustrates.","tokens_in":34833,"feed_emoji":"🎯","tokens_out":7450,"duration_ms":79044,"temperature":0.7,"pith_summary":"The paper studies an optimal control problem in which the control is a Borel measure and the Tikhonov penalty is not the usual total variation but the transportation distance to a given prior measure. It establishes existence of optimal controls and a first-order optimality system in which the adjoint state of the Poisson equation acts as a Kantorovich potential for the optimal transport plan. From that system the authors derive structural facts: under smooth quadratic costs the optimal control has no interior atoms; for strongly convex costs transport is realized by a Hölder or Lipschitz map; for power-type costs an absolutely continuous prior with a regular adjoint state forces an absolutely continuous optimal control; for metric (Wasserstein-1) costs the control is absolutely continuous with respect to the $(d-1)$-dimensional Hausdorff measure, with an example showing Lebesgue absolute continuity can fail. The upshot is that transport regularization imposes geometric restrictions on optimal controls that the Radon-norm regularizer does not.","feed_headline":"Transport-regularized PDE control: adjoint state carries the geometry","feed_subtitle":"A first-order system turns the Poisson adjoint into a Kantorovich potential, fixing where optimal controls can concentrate.","key_machinery":"The central object is the generalized transportation distance $D^c_{u_0}(u)$, defined by the Kantorovich problem with marginals $u_0$ and $u$, used as the Tikhonov term. The load-bearing identity is the subdifferential characterization from Appendix A: a continuous function $\\psi$ lies in $\\partial D^c_{u_0}(u)$ exactly when $(\\psi^c, \\psi)$ solves the dual Kantorovich problem; inserting $\\psi = -p/\\alpha$ turns the adjoint equation into a Kantorovich potential condition. The supporting mechanism is Lemma 4.4's support inclusion, which confines $\\operatorname{supp}(\\bar\\pi)$ to the set where $c(x,\\xi) + p(\\xi)/\\alpha$ is minimized in $\\xi$; all subsequent results on non-atomicity, transport maps, and absolute continuity are consequences of analyzing these minimizers with elliptic regularity of $p$.","core_discovery":"On the paper's own terms, the central discovery is that the adjoint state, which classical adjoint calculus would treat as a Lagrange multiplier, doubles as the c-conjugate potential in Kantorovich duality. Consequently the optimality condition $0 \\in p + \\alpha\\, \\partial D^c_{u_0}(u)$ is equivalent to $(-p/\\alpha|_{\\omega_1})^c$ and $-p/\\alpha|_{\\omega_1}$ solving the dual Kantorovich problem, and every optimal transport plan has support confined to pairs $(x,\\xi)$ where $\\xi$ minimizes $c(x,\\cdot) + p(\\cdot)/\\alpha$. This single mechanism carries the structural results: elliptic regularity of $p$, obtained through Green functions, maximum principles, and interior estimates, is converted into restrictions on where mass can travel and therefore on where the optimal control can live. The paper proves existence, gives sufficiency under convexity, and instantiates the mechanism for tracking-type objectives, strongly convex costs, power-type costs, and metric costs, including a sharp example where the optimal control is not Lebesgue absolutely continuous although the prior is.","pith_inferences":["The adjoint-as-potential mechanism suggests a transfer principle: any PDE whose Green function has a known singularity should yield analogous optimal-transport regularity statements, with the singularity of the Green function setting the threshold for atom suppression.","The smoothness dichotomy (C^2 costs suppress interior atoms, metric costs allow them) points to a cost-regularity phase transition; a testable conjecture is that the critical threshold is the degree of differentiability of $c$ in its second argument.","In inverse-problem practice, choosing a Wasserstein-1 regularizer should concentrate optimal controls on lower-dimensional sets even from absolutely continuous priors, whereas a Wasserstein-2 regularizer should spread mass with a density; this difference is directly testable on the same Poisson inverse problem.","The condition (6.2) and the bound (6.10) are checkable a priori, so one could design benchmarks below the threshold and test whether the computed optimal plan ceases to be induced by a continuous map."],"forward_implications":["If the optimality system of Theorem 4.2 is valid, every locally optimal control comes with an adjoint state solving the dual Kantorovich problem, so numerical first-order methods can be checked against Kantorovich duality gaps.","In the tracking-type case with $C^2$ costs, optimal controls have no atoms in the interior of the control domain; Dirac masses can only sit on the boundary, and with metric costs an interior atom can appear only if the prior has an atom at the same point.","Under strong convexity plus the curvature condition (6.2), the optimal plan is unique and induced by a Hölder-$1/2$ map $T$ with $T_\\# u_0 = u$, so the support and Hausdorff dimension of the control are bounded by those of the prior.","With power-type costs and an essentially bounded prior density, a $W^{2,r}_{\\mathrm{loc}}$ adjoint with $r>d$ gives absolute continuity of the optimal control in the interior with local density in $L^{r/d}$; under additional boundary regularity the density is bounded.","For metric costs, the optimal control is absolutely continuous with respect to $H^{d-1}$ in the interior, and Example 8.4 shows this cannot be improved to Lebesgue absolute continuity."],"supporting_citations":[{"why":"Supplies the Kantorovich duality, c-conjugate machinery, transport maps, and transport rays used in Theorem 4.2, Lemma 4.4, and Sections 6-8.","marker":"[35]"},{"why":"Gives the Wasserstein-distance interpretation of $D^c_{u_0}$ and the stability of transport plans used in Remark 3.3 and the continuity discussion in the introduction.","marker":"[39]"},{"why":"Provides the exponent $p_\\Omega$ and solvability of the Poisson equation on Lipschitz domains, underpinning well-posedness of the state and adjoint equations.","marker":"[22]"},{"why":"Supplies the Green function, greatest harmonic minorant, and superharmonicity results used for the Green-potential representation in Section 5.","marker":"[1]"},{"why":"Gives the maximum principles and interior elliptic regularity estimates used in Lemma 5.5, Theorem 5.10, and Corollary 8.6.","marker":"[19]"},{"why":"Provides the coarea inequality used in Theorem 8.3 to prove absolute continuity with respect to $H^{d-1}$.","marker":"[17]"},{"why":"Supplies the area formula for $W^{1,n}$ mappings used for the change-of-variables estimate in Theorem 7.2.","marker":"[28]"},{"why":"The numerical companion paper whose computed controls exhibit the absolute continuity behavior claimed in Section 7 and referenced in Remark 7.6.","marker":"[5]"}],"fun_headline_variants":["Adjoint state morphs into Kantorovich potential for Poisson control","Poisson control: transport plan support pinned by adjoint as potential","Optimal transport meets PDE control: adjoint reveals support geometry","Where can optimal controls live? Adjoint's dual role answers"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the adjoint state coming out of the first-order system is regular enough (continuously differentiable, or in $W^{2,r}_{\\mathrm{loc}}$ with $r>d$) for the Kantorovich potential to be analyzed pointwise; the optimality system itself does not guarantee this regularity, and the paper verifies it only in special settings.","fun_headline_variants_meta":{"raw":{"variants":["Adjoint state morphs into Kantorovich potential for Poisson control","Poisson control: transport plan support pinned by adjoint as potential","Optimal transport meets PDE control: adjoint reveals support geometry","Where can optimal controls live? Adjoint's dual role answers"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000779,"raw_usage":{"total_tokens":3394,"prompt_tokens":849,"completion_tokens":2545,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":465,"completion_tokens_details":{"reasoning_tokens":2472}},"tokens_in":465,"tokens_out":2545,"duration_ms":20605,"temperature":1.0,"reasoning_tokens":2472,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:15:48.423103+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the optimal control for a radial, spherically symmetric instance with $d=3$, smooth quadratic transport cost, a tracking-type objective with smooth desired state, and a prior with an $L^\\infty$ density whose support avoids the boundary. Corollary 5.6 predicts no atom at the center; a computed optimal control with a positive Dirac there would refute it, while an atom-free control supports the claim. The same experiment with metric costs and an absolutely continuous prior should instead produce a control with nonzero $H^2$-singular part, as Example 8.4 illustrates.","supporting_citations":[{"cited_title":"Optimal Transport for Applied Mathematicians: Cal- culus of Variations, PDEs, and Modeling","cited_arxiv_id":null,"evidence_quote":"Supplies the Kantorovich duality, c-conjugate machinery, transport maps, and transport rays used in Theorem 4.2, Lemma 4.4, and Sections 6-8."},{"cited_title":"Optimal Transport","cited_arxiv_id":null,"evidence_quote":"Gives the Wasserstein-distance interpretation of $D^c_{u_0}$ and the stability of transport plans used in Remark 3.3 and the continuity discussion in the introduction."},{"cited_title":"The coarea inequality","cited_arxiv_id":null,"evidence_quote":"Provides the coarea inequality used in Theorem 8.3 to prove absolute continuity with respect to $H^{d-1}$."},{"cited_title":"The area formula forW 1,n-mappings","cited_arxiv_id":null,"evidence_quote":"Supplies the area formula for $W^{1,n}$ mappings used for the change-of-variables estimate in Theorem 7.2."},{"cited_title":"Numerical solution of optimal control problems using quadratic transport regularization","cited_arxiv_id":null,"evidence_quote":"The numerical companion paper whose computed controls exhibit the absolute continuity behavior claimed in Section 7 and referenced in Remark 7.6."}],"review_version":1}