Pith. sign in

REVIEW 6 minor 34 references

No-gap second-order conditions for optimization problems involving transport distances

T0 review · 0 major / 6 minor · reviewed 2026-07-31 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read No-gap second-order conditions for measure optimization with transport regularization are equivalent to quadratic growth in the bounded-Lipschitz norm.

desk verdict Solid transfer of the no-gap weak-★ SOC machine to Kantorovich transport: NDC via dual descent, explicit D'', equivalence to BL quadratic growth, under clearly flagged potential-regularity assumptions. read the letter →

arxiv 2607.28264 v1 pith:A6TLSDUD submitted 2026-07-30 math.OC

Nicolas Borchard , Christian Meyer , Gerd Wachsmuth This is my paper · ORCID
classification math.OC MSC 49K40
keywords no-gapsecond-orderconditionstransportdistanceweak-starsecondsubderivativeKantorovichpotentialquadraticgrowthoptimalcontrolinmeasurespacesnon-degeneracycondition
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper studies minimization of a smooth functional plus a multiple of the transport distance from a fixed prior measure, posed in the space of measures. Classical second-order theory fails because the transport distance is nonsmooth and the natural measure norm does not support quadratic growth. By lifting the problem to the dual of continuously differentiable functions and using weak-star second subderivatives, the authors prove that a simple coercivity condition on that subderivative is equivalent to quadratic growth of the objective near a local minimizer, once a regularity assumption on the associated Kantorovich potential holds. Under stronger smoothness they also compute the subderivative explicitly as an infimum of a weighted quadratic form over densities that represent the direction. The result supplies usable second-order tests for sparse and transport-regularized optimal control problems in measure space.

What carries the argument

The weak-star second subderivative of the transport distance, obtained via duality from the twice Hadamard-differentiable pre-conjugate functional built from the c-bar conjugate; its positivity, together with a dual descent inequality that yields the non-degeneracy condition, carries the no-gap equivalence.

What would settle it

Construct a compact domain, continuous cost, and smooth objective for which a Kantorovich potential exists but violates the quadratic-growth condition, then check whether quadratic growth of the objective still holds in the bounded-Lipschitz norm; if it does, the claimed necessity fails.

Watch

Extended reading notes

Core claim

Under mild smoothness of the smooth part of the objective and a quadratic-growth regularity condition on a Kantorovich potential, the second-order condition that the second derivative of the smooth part plus a positive multiple of the weak-star second subderivative of the transport distance is strictly positive for every nonzero direction is necessary and sufficient for quadratic growth of the objective in the bounded-Lipschitz norm.

Load-bearing premise

A Kantorovich potential for the candidate minimizer must grow at least quadratically away from the optimal transport map, uniformly or with a controllable exceptional set; without that growth the key dual descent inequality already fails.

Editorial extensions

If this is right

  • Quadratic growth in the bounded-Lipschitz norm is fully characterized by a checkable second-order inequality involving only the smooth part and the transport subderivative.
  • Under C-squared costs and potentials the subderivative reduces to an explicit infimum of a matrix-weighted L2 quadratic form over representing densities.
  • The same second-order condition implies strong metric subregularity of the first-order optimality map for power costs with exponent at least two.
  • The theory applies directly to linear-quadratic elliptic control problems with separated observation and control domains, giving no-gap second-order tests in measure space.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same duality-plus-lift strategy may extend to other weakly continuous regularizers that admit an explicit pre-conjugate with a verifiable descent property.
  • If non-degeneracy can be recovered in a slightly smoother dual space, the method could cover semilinear state equations without artificial smoothing of the control operator.
  • Explicit second-subderivative formulas open the door to second-order algorithms and local convergence rates for transport-regularized sparse control.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

0 major / 6 minor

Summary. The paper derives no-gap second-order optimality conditions for problems of the form min F(u) + (α/2)D(u) on measures, where D is the Kantorovich transport cost to a fixed prior. The problem is lifted to C¹(ω₁)⋆ so that the non-degeneracy condition (NDC) can hold; under a quadratic-growth assumption on a Kantorovich potential (Ass. 3.8), dual descent for the pre-conjugate J yields NDC, and the abstract theory of weak-⋆ second subderivatives then equates the SOC F''(ū)h² + (α/2)D''(ū,φ̄;h) > 0 with quadratic growth in the bounded-Lipschitz norm (Thm. 5.2). Under the stronger Ass. 4.2 the weak-⋆ second subderivative of D is computed explicitly via Hadamard second differentiability of J, conjugacy, and a diagonal recovery-sequence argument (Thm. 4.13), and the results are illustrated on a linear-quadratic elliptic control problem with separated observation domain.

Significance. No-gap second-order conditions for transport-regularized problems in measure space have been missing; the only prior second-variation formula ([6]) requires smooth densities and domains and does not yield SOC. The paper supplies a complete chain—space choice justified by counter-examples (Ex. 2.7, 3.13), dual descent under a verifiable potential-growth hypothesis, explicit formula for D'' via disintegration, and an application to sparse optimal control—building cleanly on the authors’ abstract weak-⋆ framework and the Radon-norm analogue. The explicit matrix field H and the equivalence with BL-quadratic growth are concrete, usable contributions; the limitations for semilinear equations are honestly flagged. If the stated regularity on the Kantorovich potential is accepted, the result is a solid advance for second-order analysis of transport-regularized measure optimization.

minor comments (6)
  1. [§2.1, §2.4] In Ass. 2.1(v) and (2.12) the same symbol ū is used both for the measure and for its image under Π; a brief remark that the identification is used throughout would help the reader keep track of the ambient space.
  2. [§3] Lemma 3.5 gives a convenient C² criterion for Ass. 3.3, but the subsequent weaker Ass. 3.8 is the one actually used for NDC. A one-sentence pointer after Lemma 3.5 that Ass. 3.8 is the working hypothesis for §3 would clarify the logical hierarchy.
  3. [Lemma 3.9] The constant C̃ appearing in Lemma 3.9 is never made explicit; even a rough expression in terms of C_L, γ, C, u₀(ω₀) would make the descent inequality more transparent for later quantitative work.
  4. [Theorem 4.13] In the proof of Thm. 4.13 the diagonal sequence is constructed via a dense countable set (f_i); it would be helpful to recall that bounded sets in C¹⋆ are weak-⋆ metrizable (already used in Lemma 4.12) so that the argument is fully sequential.
  5. [Example 5.5] Example 5.5 assumes the observation domain has positive distance to ω₁; a short remark that the same argument works for any compactly supported right-hand side whose support is separated from Λ would slightly enlarge the scope without extra work.
  6. [Title page, throughout] Typographical: “OPTIMIZA TION”, “TRANSPOR T”, “INVOL VING”, “DIST ANCES” in the title; “F akultät” in the affiliations; occasional missing spaces after commas in displayed formulae (e.g. after t_k in Def. 2.3).

Circularity Check

1 steps flagged · score 1.0 of 10

No significant circularity: transport-specific dual estimates and D'' formula are independently derived; self-citations supply only the abstract SOC machine.

  1. self citation load bearing [Section 2.2, Theorem 2.5; cf. also lifting strategy §2.3–2.4 citing [33]]
    "Now, we are in position to state the second-order optimality result. Theorem 2.5 ([3, Thm 2.20]). ... Here, we will pursue the same strategy and treat (P) as a problem in C¹(ω₁)⋆."

    The abstract equivalence (NDC + coercivity of F''+G'' ⇔ quadratic growth) and the C¹⋆ quotient-space setup are taken from overlapping-author papers [3,33] rather than re-proved. This is ordinary reuse of a framework, not a definitional loop: the paper still has to establish NDC and compute D'' for the transport cost, which it does with independent dual arguments. Not load-bearing for the transport claims; flagged only as minor self-citation.

full rationale

The paper applies the authors' prior abstract no-gap framework (Theorem 2.5 from [3], weak-★ second subderivatives from [31,12]) and the C¹⋆-lifting strategy of the Radon-norm analogue [33], then derives the transport-specific ingredients in full: the dual descent inequality for J under Assumption 3.8 (Lemmas 3.7–3.9), NDC (Theorem 3.11), Hadamard second differentiability of J (Theorem 4.6), the conjugate formula for D'' with weak-★ epi-differentiability (Theorem 4.13), and the equivalence to BL-quadratic growth (Theorem 5.2). None of these steps redefine the target as an input, fit a parameter and relabel it a prediction, or import a uniqueness theorem that forces the main claim. Counterexamples 2.7 and 3.13 are original and show why the space/potential hypotheses cannot be dropped. Self-citation is present but not load-bearing for the transport content; score 1 reflects only that minor dependence on the authors' abstract machine.

Assumptions & free parameters 0 free parameters · 6 assumptions · 2 invented entities

The central equivalence rests on the abstract no-gap theorem from prior work, on geometric and regularity assumptions that make the dual of C¹ well-behaved and the Kantorovich potential quadratically growing, and on twice weak-⋆ differentiability of the smooth part F. No numerical free parameters appear; the 'constants' γ, ε, C in the potential assumptions are structural hypotheses, not fits.

assumptions (6)
  • standard math Abstract no-gap SOC theorem: under sequential weak-⋆ continuity of h↦F''(x)h², NDC plus positivity of F''+G'' is equivalent to quadratic growth (Thm 2.5 / [3, Thm 2.20]).
    Imported as the black-box engine; the paper's contribution is verifying its hypotheses for G = (α/2)D.
  • domain assumption Standing data: compact ω₀,ω₁; int ω₁ uniformly locally quasiconvex and closure of its interior; continuous cost c; u₀≥0 with full support; ū≥0 with equal mass (Ass. 2.1).
    Standard compact OT setting; quasiconvexity needed for Lip control and BL ~ C¹⋆ equivalence on measures.
  • domain assumption Kantorovich potential regularity: quadratic growth of c(x,·)−φ̄ about T̄ (Ass. 3.3) or the weaker threshold form (Ass. 3.8).
    Essential for dual descent of J and hence NDC; not automatic for arbitrary continuous c.
  • domain assumption For explicit D'': c ∈ C², supp(ū)⊂int(ω₁), φ̄ ∈ C² near supp(ū), and Ass. 3.3 (Ass. 4.2).
    Needed for calmness of argmin maps, Taylor expansion, and the Hessian formula for J''.
  • domain assumption F twice differentiable in the weak-⋆ sense (2.4)/(5.1) with h↦F''(ū)h² sequentially weak-⋆ continuous.
    Required by the abstract theorem; verified for linear elliptic control with separated observation, not for general semilinear control in C¹⋆.
  • standard math Classical Kantorovich duality and existence of measurable transport maps for continuous costs on compact sets.
    Used throughout via Santambrogio and measurable-selection theorems.
invented entities (2)
  • Generalized transport distance D on C¹(ω₁)⋆ (∞ off ran(Π)) independent evidence
    purpose: Lift the original measure problem so that NDC and quadratic growth can hold.
    Standard extension-by-infinity of a functional defined on measures; not a new physical entity.
  • Matrix field H(η) built by disintegrating inverted partial Hessians of c−φ̄ along T̄-fibers independent evidence
    purpose: Express J''(φ̄;ψ) and the conjugate formula for D''(ū,φ̄;·).
    Derived object, not postulated; reduces to 2Id−∇²φ̄ for quadratic cost.

how reviews work

0 comments
Cite this review

Pith. "Pith review of No-gap second-order conditions for optimization problems involving transport distances." pith.science (2026). https://pith.science/paper/A6TLSDUD

@misc{pith2026260728264,
  author       = {Pith},
  title        = {Pith review of: No-gap second-order conditions for optimization problems involving transport distances},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A6TLSDUD}},
  note         = {Machine review of arXiv:2607.28264}
}
abstract

We consider optimization problems in the space of measures. As a regularization term, the problem includes the transport distance to a given prior measure. For the derivation of second-order optimality conditions of no-gap type, the theory of weak-$\star$ second subderivatives is used which will lead to an equivalence with quadratic growth under additional assumptions on the smooth part of the objective and on the Kantorovich potential, i.e., the solution of the dual transport problem. Further, the weak-$\star$ second subderivative is calculated and weak-$\star$ epidifferentiability is proven. Finally, the results are applied to optimal control problems in measure space.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 14 canonical work pages

  1. [6]

    Delgadino, and Jun Kitagawa.Stability of optimal transport maps and second variation of the 2-Monge-Kantorovich distance

    Francisco-Unai Caja-Lopez, Matias G. Delgadino, and Jun Kitagawa.Stability of optimal transport maps and second variation of the 2-Monge-Kantorovich distance. 2026. arXiv:2605.24232

    arXiv 2026

  2. [1]

    Basel: Birkh¨ auser, 2005.doi: 10.1007/b137080

    Luigi Ambrosio, Nicola Gigli, and Giuseppe Savar´ e.Gradient flows in metric spaces and in the space of probability measures. Basel: Birkh¨ auser, 2005.doi: 10.1007/b137080

    doi:10.1007/b137080 2005

  3. [2]

    Birkh¨ auser Boston, MA, 2008.doi:10.1007/978-0-8176-4848-0

    Jean-Pierre Aubin and H´ el` ene Frankowska.Set-Valued Analysis. Birkh¨ auser Boston, MA, 2008.doi:10.1007/978-0-8176-4848-0

    doi:10.1007/978-0-8176-4848-0 2008

  4. [3]

    Second-order conditions for spatio- temporally sparse optimal control via second subderivatives

    Nicolas Borchard and Gerd Wachsmuth. “Second-order conditions for spatio- temporally sparse optimal control via second subderivatives”. In:Journal of Nonsmooth Analysis and Optimization5 (2024).doi: 10.46298/jnsao-2024- 12604

    work page doi:10.46298/jnsao-2024- 2024

  5. [4]

    Inverse problems in spaces of measures

    Kristian Bredies and Hanna Katriina Pikkarainen. “Inverse problems in spaces of measures”. In:ESAIM Control Optim. Calc. Var.19.1 (2013), pp. 190–218. doi:10.1051/cocv/2011205

    arXiv 2013

  6. [5]

    Universitext

    Haim Brezis.Functional analysis, Sobolev spaces and partial differential equa- tions. Universitext. Springer, New York, 2011.doi: 10.1007/978- 0- 387- 70914-7

    doi:10.1007/978- 2011

  7. [7]

    Second Order Analysis for Bang-Bang Control Problems of PDEs

    Eduardo Casas. “Second Order Analysis for Bang-Bang Control Problems of PDEs”. In:SIAM Journal on Control and Optimization50.4 (2012), pp. 2355– 2372.doi:10.1137/120862892

    work page doi:10.1137/120862892 2012

  8. [8]

    Optimal control of semilinear elliptic equations in measure spaces

    Eduardo Casas and Karl Kunisch. “Optimal control of semilinear elliptic equations in measure spaces”. In:SIAM J. Control Optim.52.1 (2014), pp. 339– 364.doi:10.1137/13092188X

    work page doi:10.1137/13092188x 2014

Show all 34 references
  1. [9]

    Second Order Analysis for Optimal Control Problems: Improving Results Expected From Abstract Theory

    Eduardo Casas and Fredi Tr¨ oltzsch. “Second Order Analysis for Optimal Control Problems: Improving Results Expected From Abstract Theory”. In: SIAM Journal on Optimization22.1 (2012), pp. 261–279.doi: 10 . 1137 / 110840406. 38 REFERENCES

    2012

  2. [10]

    Sufficient Second- Order Conditions for Bang-Bang Control Problems

    Eduardo Casas, Daniel Wachsmuth, and Gerd Wachsmuth. “Sufficient Second- Order Conditions for Bang-Bang Control Problems”. In:SIAM Journal on Con- trol and Optimization55.5 (2017), pp. 3066–3090.doi: 10.1137/16M1099674

    2017 doi

  3. [11]

    Differential Sensitivity Analysis of Variational Inequalities with Locally Lipschitz Continuous Solution Operators

    Constantin Christof and Gerd Wachsmuth. “Differential Sensitivity Analysis of Variational Inequalities with Locally Lipschitz Continuous Solution Operators”. In:Applied Mathematics & Optimization81.1 (2019), pp. 23–62.doi: 10.1007/ s00245-018-09553-y

    2019

  4. [12]

    No-Gap Second-Order Conditions via a Directional Curvature Functional

    Constantin Christof and Gerd Wachsmuth. “No-Gap Second-Order Conditions via a Directional Curvature Functional”. In:SIAM Journal on Optimization 28.3 (2018), pp. 2097–2130.doi:10.1137/17M1140418

    2018 doi

  5. [13]

    A duality-based approach to elliptic control problems in non-reflexive Banach spaces

    Christian Clason and Karl Kunisch. “A duality-based approach to elliptic control problems in non-reflexive Banach spaces”. In:ESAIM Control Optim. Calc. Var.17.1 (2011), pp. 243–266.doi:10.1051/cocv/2010003

    2011

  6. [14]

    A measure space approach to optimal source placement

    Christian Clason and Karl Kunisch. “A measure space approach to optimal source placement”. In:Comput. Optim. Appl.53.1 (2012), pp. 155–171.doi: 10.1007/s10589-011-9444-9

    2012 doi

  7. [15]

    A note on the approximation of elliptic control problems with bang-bang controls

    Klaus Deckelnick and Michael Hinze. “A note on the approximation of elliptic control problems with bang-bang controls”. In:Comput. Optim. Appl.51.2 (2012), pp. 931–939.doi:10.1007/s10589-010-9365-z

    2012 doi

  8. [16]

    Generalized second-order derivatives of convex functions in reflexive Banach spaces

    Chi Ngoc Do. “Generalized second-order derivatives of convex functions in reflexive Banach spaces”. In:Transactions of the American Mathematical Society334.1 (1992), pp. 281–301.doi:10.2307/2153983

    1992 doi

  9. [17]

    Exact support recovery for sparse spikes deconvolution

    Vincent Duval and Gabriel Peyr´ e. “Exact support recovery for sparse spikes deconvolution”. In:Found. Comput. Math.15.5 (2015), pp. 1315–1355.doi: 10.1007/s10208-014-9228-6

    2015 doi

  10. [18]

    On Nemytskij operators in Lp-spaces of abstract functions

    Hyman Goldberg, Winfried Kampowsky, and Fredi Tr¨ oltzsch. “On Nemytskij operators in Lp-spaces of abstract functions”. In:Mathematische Nachrichten 155 (1992), pp. 127–140.doi:10.1002/mana.19921550110

    1992 doi

  11. [19]

    The inhomogeneous Dirichlet problem in Lipschitz domains

    David Jerison and Carlos E. Kenig. “The inhomogeneous Dirichlet problem in Lipschitz domains”. In:J. Funct. Anal.130.1 (1995), pp. 161–219.doi: 10.1006/jfan.1995.1067

    1995

  12. [20]

    Second-order epi-derivatives of integral functionals

    A. B. Levy. “Second-order epi-derivatives of integral functionals”. In:Set- Valued Analysis1.4 (1993), pp. 379–392.doi:10.1007/bf01027827

    1993 doi

  13. [21]

    Christian Meyer and Gerd Wachsmuth.Optimal control of the Poisson equa- tion with transport regularization: Properties of optimal transport plans and transport map. 2025. arXiv:2506.02808

    2025 arXiv

  14. [22]

    A sparse control approach to optimal sensor placement in PDE-constrained parameter estimation problems

    Ira Neitzel, Konstantin Pieper, Boris Vexler, and Daniel Walter. “A sparse control approach to optimal sensor placement in PDE-constrained parameter estimation problems”. In:Numerische Mathematik143.4 (2019), pp. 943–984. doi:10.1007/s00211-019-01073-3

    2019 doi

  15. [23]

    A Priori Error Analysis for Discretization of Sparse Elliptic Optimal Control Problems in Measure Space

    Konstantin Pieper and Boris Vexler. “A Priori Error Analysis for Discretization of Sparse Elliptic Optimal Control Problems in Measure Space”. In:SIAM Journal on Control and Optimization51.4 (2013), pp. 2788–2808.doi: 10. 1137/120889137

    2013

  16. [24]

    Ponce.Elliptic PDEs, Measures and Capacities

    Augusto C. Ponce.Elliptic PDEs, Measures and Capacities. Vol. 23. EMS Tracts Math. Z¨ urich: European Mathematical Society (EMS), 2016.doi: 10.4171/140. REFERENCES 39

    2016 doi

  17. [25]

    Integrals which are convex functionals. II

    R. T. Rockafellar. “Integrals which are convex functionals. II.” In:Pacific Jour- nal of Mathematics39.2 (1971), pp. 439–469.url: https://projecteuclid. org:443/euclid.pjm/1102969571

    1971

  18. [26]

    Ralph Tyrrell Rockafellar and Roger J. B. Wets.Variational Analysis. Springer Berlin Heidelberg, 1998.doi:10.1007/978-3-642-02431-3

    1998 doi

  19. [27]

    Springer International Publishing, 2015.doi:10.1007/978-3-319-20828-2

    Filippo Santambrogio.Optimal Transport for Applied Mathematicians: Cal- culus of Variations, PDEs, and Modeling. Springer International Publishing, 2015.doi:10.1007/978-3-319-20828-2

    2015 doi

  20. [28]

    H. H. Schaefer and M. P. Wolff.Topological Vector Spaces. Springer New York, 1999.doi:10.1007/978-1-4612-1468-7

    1999 doi

  21. [29]

    Sparsity Regularization for Radon Mea- sures

    Otmar Scherzer and Birgit Walch. “Sparsity Regularization for Radon Mea- sures”. In:Scale Space and Variational Methods in Computer Vision. Springer Berlin Heidelberg, 2009, pp. 452–463.doi: 10.1007/978-3-642-02256-2_38

    2009 doi

  22. [30]

    Daniel Wachsmuth and Gerd Wachsmuth.Continuous differentiability of the signum function and Newton ’s method for bang-bang control. 2025. arXiv: 2509.24829

    2025

  23. [31]

    Second-order conditions for non- uniformly convex integrands: quadratic growth in L1

    Daniel Wachsmuth and Gerd Wachsmuth. “Second-order conditions for non- uniformly convex integrands: quadratic growth in L1”. In:Journal of Non- smooth Analysis and Optimization3 (2022).doi: 10.46298/jnsao- 2022- 8733

    2022 doi

  24. [32]

    Convergence and Regularization Results for Optimal Control Problems with Sparsity Functional

    Gerd Wachsmuth and Daniel Wachsmuth. “Convergence and Regularization Results for Optimal Control Problems with Sparsity Functional”. In:ESAIM: Control, Optimisation and Calculus of Variations17.3 (2011), pp. 858–886. doi:10.1051/cocv/2010027

    2011

  25. [33]

    Gerd Wachsmuth and Daniel Walter.No-gap second-order conditions for minimization problems in spaces of measures. 2024. arXiv:2403.12001

    2024 arXiv

  26. [34]

    Gerd Wachsmuth and Daniel Walter.Proximal gradient methods in Banach spaces. 2025. arXiv:2509.24685. BTU Cottbus-Senftenberg, F akult¨at 1, F achgebiet Optimale Steuerung, Platz der Deutschen Einheit 1, 03046 Cottbus, Germany Email address:nicolas.borchard@b-tu.de Technische U...

    2025

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.