Pith. sign in

REVIEW 3 major objections 6 minor 28 references

This paper proves global well-posedness, uniform support confinement, and explicit stationary states for a nonlocal attraction-repulsion transport equation with power kernels.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 08:38 UTC pith:YIRVR7CM

load-bearing objection The well-posedness and support-confinement half is real and carefully done; the long-time convergence half rests on a dissipation identity whose proof is only sketched, so this is a solid conditional contribution rather than a complete theory. the 3 major comments →

arxiv 2607.04424 v2 pith:YIRVR7CM submitted 2026-07-05 math.AP

The nonlocal attraction-repulsion transport equation with power kernels

Massimo Fornasier , Hui Huang , Lukang Sun This is my paper
classification math.AP MSC 35Q7035A0135B4049Q22
keywords nonlocal transport equationattraction-repulsionpower-law kernelsLagrangian well-posednesssupport confinementzero-flux stationary statesfractional Laplacianmaximum mean discrepancy gradient flow
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper studies a continuity equation on R^d in which a probability density is pulled toward a fixed background measure ω by the attraction kernel |x|^{1+a} and pushed apart by self-repulsion |x|^{1+r}, with 0≤a,r<1. It claims that this singular, non-convex Wasserstein gradient flow is globally well-posed in the Lagrangian class, with explicit L∞ and moment bounds, and that in the attraction-dominant regime a>r (or a=r with ω(R^d)>1) compactly supported initial data stay inside a fixed ball forever. For the attractive-dominant nonquadratic range it reduces the zero-flux stationarity condition to a fractional exterior Dirichlet problem, yielding explicit stationary measures in dimensions 1, 2, and 3. It also proves subsequential convergence of every bounded-energy trajectory to a zero-flux stationary state. The results matter because this equation is the Wasserstein gradient flow of an MMD energy with negative-distance kernel, where the usual displacement-convexity tools fail.

Core claim

The central claim is that the zero-flux stationary condition ϕ(K_a*ω − K_r*ϕ)=0 is not just a PDE identity but a free-boundary problem with a fractional Laplacian. Writing F = ψ_a*ω − ψ_r*ϕ and L_r = c_{d,r}(−Δ)^{(d+1+r)/2}, the paper shows the inverse relation ϕ = Q_{a,r} − L_r F, where Q_{a,r} = L_r(ψ_a*ω); stationarity is then equivalent to L_r F = Q_{a,r} on the complement of the support plus ∇F=0 on the support. With the additional homogeneous exterior Dirichlet ansatz H=0 outside the support, this becomes a fractional Dirichlet problem whose Green representation gives explicit stationary profiles: intervals in d=1 via a finite Riesz equation, disks in d=2 via the Boggio–Kelvin kernel,

What carries the argument

The squared-radius regularization ψ_{s,ε}(x)=(|x|^2+ε)^{(1+s)/2} is the engine of the well-posedness proof: it preserves convexity and the sign of the Laplacian, yielding uniform L∞ and moment estimates, and a two-scale decomposition of the kernel K_s into a small ball around the singularity and a smooth far part passes the nonlinear force to the limit. For the stationarity half, the Fourier multiplier operator L_r with symbol c|ξ|^{d+1+r} inverts the repulsion kernel ψ_r and converts the zero-flux condition into the fractional exterior Dirichlet problem; the Boggio–Kelvin Green kernel then supplies the explicit stationary states.

Load-bearing premise

The load-bearing premise is the energy-dissipation identity d/dt E(ϕ_t)=−∫|K_a*ω−K_r*ϕ_t|^2 dϕ_t for Lagrangian solutions, whose proof is deferred in Remark 8.2; a second premise is the homogeneous exterior Dirichlet ansatz H=0 on Ω^c, which is stated not to follow from the free-boundary equation but underlies the explicit stationary states.

What would settle it

Compute, for a concrete Lagrangian solution (for instance d=1, a=0.8, r=0.2, ω(x)=0.9·1_{[-1,1]}(x)), the quantity d/dt E(ϕ_t)+∫|K_a*ω−K_r*ϕ_t|^2 dϕ_t over several times; if it is not identically zero, Proposition 8.1 is false and the convergence argument collapses. Separately, simulate the two-particle symmetric configuration with a=r=s>1 and ω(R^d)>1: the paper predicts finite-time escape, so any simulation showing permanent confinement would refute the claimed sharpness of the support theorem.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • For every T>0 and every admissible initial density, there exists a unique Lagrangian distributional solution with finite-time L∞ and moment bounds, so the model is predictive on R^d.
  • In the attractive-dominant regime with compact support, solutions remain confined in a fixed ball for all time, and the paper shows this fails when a=r>1.
  • Stationary states of the attractive-dominant equation are characterized by a fractional exterior Dirichlet problem, giving explicit formulas in d=1,2,3 and a numerical check in 1D and 2D.
  • Every global solution with bounded energy and uniform moments has a subsequence converging narrowly to a zero-flux stationary state; full convergence follows if the omega-limit set contains only one such state.
  • In the MMD case a=r with ω(R^d)=1, well-posedness, the dissipation identity, and subsequential convergence hold, but uniform support confinement and full convergence generally do not.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The free-boundary reduction suggests that equilibria can be computed by solving one fractional elliptic problem rather than by running the particle dynamics; the paper's own direct reconstruction in d=2 is a first instance of this.
  • If the stationary characterization is combined with the background measure ω alone, it yields the equilibrium population mismatch quantified by the cumulative distribution function introduced in the introduction, making the 'unequal access' interpretation directly computable.
  • The convergence result currently rests on a dissipation identity whose proof is deferred; a rigorous verification or a counterexample for that identity would decide whether convergence is subsequential only or full.
  • A natural testable extension is the borderline case a=r with ω(R^d)=1 and compactly supported ω in d≥2, where the paper's confinement argument does not apply and the expected large-time profile is ω.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper studies the nonlocal continuity equation ∂tφ = div(φ(∇V − K_r * φ)) on R^d, with V = ψ_a * ω for a prescribed background density ω and kernels ψ_s(x)=|x|^{1+s}, K_s=∇ψ_s, a,r∈[0,1). The main results are: (i) global Lagrangian well-posedness via a squared-radius regularization, including L∞ and moment bounds, W^{n,∞} propagation, and uniqueness in the Lagrangian class (Theorem 4.1, Proposition 4.2); (ii) uniform confinement of the support in the attractive-dominant case a>r, or a=r with ω(R^d)>1, with a counterexample for the superlinear equal-exponent case (Theorem 5.6, Example 10.3); (iii) a free-boundary characterization of zero-flux stationary states in the range 0≤r≤a<1, based on an inverse operator L_r and a fractional exterior Dirichlet problem (Section 6, Proposition 6.5); (iv) explicit stationary-state constructions in dimensions 1, 2, and 3, with numerical simulations in d=1,2; and (v) a subsequential convergence result to zero-flux stationary states, conditional on an energy-dissipation identity (Section 8). The proof of well-posedness is built on averaged Lipschitz estimates and a careful passage from the regularized to the singular problem; the stationary-state examples are verified by direct force/mass identities in the three-dimensional cases.

Significance. If the convergence program is completed, the paper would provide a comprehensive theory for a nonlocal transport equation with singular, non-λ-convex interaction kernels, substantially extending earlier one-dimensional results by Di Francesco–Fornasier–Hütter–Matthes and the MMD-gradient-flow literature to arbitrary dimension and to the discordant regime a≠r. The well-posedness part is detailed and appears technically sound: Lemma 3.3 supplies a uniform averaged Lipschitz estimate for the singular force, and the compactness argument of Theorem 4.1 is presented in enough detail to be checked. The explicit zero-flux stationary states in d=3 are genuinely verified by closed-form computation, and the numerical experiments in d=1 are consistent with the theoretical characterization. The main reservation concerns Section 8: the energy-dissipation identity (Proposition 8.1), on which the convergence theorem and its corollary rest, is only proved formally and its rigorous justification is deferred to Remark 8.2. Until that identity is proved for the Lagrangian solution class, Theorem 8.5 and Corollary 8.10 remain conditional. The paper also honestly flags that the exterior Dirichlet ansatz

major comments (3)
  1. [§8, Proposition 8.1 and Remark 8.2] The energy-dissipation identity d/dt E(φ_t) = −∫|K_a*ω − K_r*φ_t|² dφ_t is load-bearing for (8.2), Theorem 8.5, and Corollary 8.10, but the proof is purely formal: it differentiates E and integrates by parts without confronting the regularity of φ_t. The solution built in Theorem 4.1 is only an L^1∩L^∞ density with finite p-moments (p>1 in Assumption 2.1); this does not even guarantee that E(φ_t) is finite when 1+r>p, and B_t = K_a*ω − K_r*φ_t is discontinuous for r=0 and singular near the diagonal for r>0. Remark 8.2 defers the result to a regularization 'similar to Theorem 4.1', but the compactness obtained there gives B_ε→B only in L^1_t L∞_loc(Q); passing the quadratic dissipation term ∫|B_ε|² dφ_ε to the limit requires strong L²(dφ_t dt)-convergence or a compensated-compactness argument, which is not supplied. Please either provide a complete proof of Proposition 8.1 under stated as
  2. [§6, Proposition 6.5 and Theorem 6.2] The abstract and introduction state that the stationarity condition is reduced to a fractional exterior Dirichlet problem. However, Proposition 6.5 explicitly acknowledges that the exterior homogeneous condition H=0 in Ω^c 'does not follow from (6.10)' and is an additional ansatz. All the explicit stationary-state constructions in Section 7 use this ansatz. Theorem 6.2 itself is a reformulation of the zero-flux condition via the inverse identity φ = Q_{a,r} − L_r F; it does not by itself determine Ω and F—the 'free-boundary conditions' (6.12)–(6.13) and the convolution gauge condition must still be verified. Please qualify the claim of a full characterization and state clearly that the fractional Dirichlet problem provides candidate states, which are stationary only after the extra conditions are verified. In its current form, the title of Section 6 overstates the reduction.
  3. [§7.2, Proposition 7.3 and the numerical reconstruction] For the two-dimensional example the 'explicit' stationary measure is not fully determined by the theory: the support radius is fixed at R=0.33 and the remaining global multiplicative constant is fixed by the mass-one constraint and by minimizing the residual of the force identity. This means the d=2 construction is partly calibrated rather than predictive. Since the abstract advertises explicit examples in d∈{1,2,3}, either give an analytic determination of these constants (including the normalization constants of the fractional Laplacian and Fourier transform in (7.16)–(7.25)), or clearly classify the d=2 result as a numerical demonstration of the reconstruction formula rather than an explicit closed-form stationary state.
minor comments (6)
  1. [Throughout] The equation numbering has apparent errors: after (3.27) the text labels formulas (3.30) and (3.31), and later (4.30) appears twice (one occurrence after (4.27)). Please renumber consistently.
  2. [§1, Declaration] In the Declaration of generative AI use, 'ChaGPT' should be 'ChatGPT'.
  3. [§2, Definition 2.3] The zero-flux stationary condition (2.14) requires the integral to be well-defined; for r=0 the kernel K_0 is discontinuous at the origin. State explicitly the convention K_0(0)=0 (it is used later, e.g., in Theorem 8.5).
  4. [§4, Theorem 4.1] The theorem states a,r∈[0,1], while the abstract and most of the paper restrict to a,r∈[0,1). The endpoint a=r=1 is said to be simple in the Introduction, but it is not excluded in Theorem 4.1. Please clarify whether the theorem covers [0,1] or [0,1) in the main statements.
  5. [§5, Theorem 5.6 and Remark 5.7] The counterexample in Example 10.3 is for a=r=s>1 and shows particle escape; this is cited in Remark 5.7 as 'a≥r>1'. The phrasing 'for a≥r>1' is slightly broader than the example, since the paper only treats a=r. Please state precisely which parameter range the counterexample covers.
  6. [§7.2, Eq. (7.30)] The definition of A_q(ξ,η) divides by 2π but the angular integration variable is not explicitly bounded; please specify θ∈[0,2π] and note the convention for ξ=η, r=0 where the kernel may be singular.

Circularity Check

0 steps flagged

No significant circularity: the well-posedness and support theorems are self-contained, and the stationary states are constructed under an explicitly admitted ansatz and then verified directly. The main caveats — a deferred proof of the dissipation identity and a fitted normalization in a 2D numerical example — are rigor/numerical issues, not circular reductions.

full rationale

The central derivation chain is not circular. Theorem 4.1 obtains global Lagrangian well-posedness from a squared-radius regularization with uniform L∞ and moment bounds, compactness, and a uniqueness argument; no fitted parameter or target conclusion is presupposed. Theorem 5.6's support confinement rests on energy-sublevel and diameter estimates, with the endpoint case treated separately, and does not reduce to its conclusion. In Section 6, the operator L_r is introduced only after an injectivity proof for ψ_r-convolutions, so the identity ϕ = Q_{a,r} − L_rF is an honest inversion rather than a definition tailor-made to produce the stated stationary form. The stationary-state constructions in Section 7 are explicitly based on an exterior Dirichlet ansatz (H=0 on Ω^c), which the paper itself states 'does not follow from (6.10)', and the candidates are then checked against the zero-flux and mass conditions by direct computation (Propositions 7.5 and 7.6). This is an acknowledged assumption plus verification, not a smuggled conclusion. The convergence program of Section 8 is conditional on Proposition 8.1, whose proof is only sketched: Remark 8.2 says 'one can adapt a similar argument from Theorem 4.1... and then pass to the limit'. That is a genuine gap in delivered rigor, but it is not circular: the dissipation identity is not derived from the convergence statement it supports. The 2D numerical example fixes R=0.33 and a global normalization constant by mass and residual minimization, so that particular numerical comparison is partly fitted; however, it is an illustrative experiment, not the basis of the paper's main theorems. Minor self-citations such as [15] and [20] are contextual and not load-bearing for the core well-posedness or support results. Overall, I find no step in which a prediction is equivalent by construction to its inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 7 axioms · 0 invented entities

The central claims rest on standard Fourier/finite-part calculus for homogeneous kernels, standard Green-function representations, and the standing L^1 cap L^infinity moment hypotheses. The paper-specific burden is concentrated in three items: the energy-dissipation identity whose proof is deferred (Prop. 8.1/Remark 8.2), the H=0 exterior ansatz that converts the free-boundary problem into a solvable Dirichlet problem (Prop. 6.5), and a fitted normalization constant in the 2D numerical curve (Section 7.2). Two quantities are numerically fitted (R=0.33 and the normalization constant); the 1D support half-width L is selected by the mass free-boundary condition and is not a free fit. No new physical entities are postulated.

free parameters (2)
  • 2D stationary support radius R = 0.33
    In the two-dimensional example (Section 7.2) the support radius is fixed at R=0.33 after (7.32), without deriving it from a free-boundary condition as in the 1D experiment; all comparison curves use this hand-fixed value.
  • Global normalization constant of the 2D direct reconstruction = not quoted; set by mass-one constraint and residual minimization of (7.31)
    The 'independent' direct-quadrature curve (7.24) is rescaled by minimizing the residual of the radial zero-flux system it is meant to confirm, so its normalization is fitted rather than derived from the Fourier conventions.
axioms (7)
  • domain assumption Assumption 2.1: omega, phi_0 in L^1 cap L^infinity with finite <x>^p moments
    Standing hypotheses for all well-posedness results; the L^infinity bound (4.2), moment bound (4.3), and compactness arguments all build directly on them.
  • standard math Conditional negative definiteness of |x|^q for 0<q<=2 (Lemma 10.1)
    Used in Proposition 5.1, case a=r with omega(R^d)>1, to show the energy is bounded below; the lemma is proven in the appendix via the Riesz cosine representation.
  • standard math Fourier transform of the homogeneous kernel: psi_hat_s = gamma_{d,s} fp|xi|^{-d-1-s}
    Invoked in Section 6 to define the inversion operator L_r. The paper proves injectivity of the representing measure on its own; the transform formula itself is taken from Gelfand-Shilov/Grafakos.
  • standard math Boggio-Kelvin Green kernel for the exterior-ball fractional Dirichlet problem (Example 6.7, eq. (7.21))
    Classical representation used to write the 2D candidate (7.24). The candidate is only declared stationary after independently verifying the force and mass conditions.
  • ad hoc to paper Homogeneous exterior Dirichlet ansatz: H=0 in Omega^c (Prop. 6.5)
    The paper states it 'does not follow from (6.10); it is an additional homogeneous Dirichlet ansatz'. Every explicit stationary-state reconstruction in Section 7 rests on it, so the characterization is not closed — it selects one admissible branch of stationary states.
  • domain assumption Energy-dissipation identity (8.1) for Lagrangian solutions (Prop. 8.1)
    Rigorous proof deferred in Remark 8.2; the time-averaged dissipation (8.2), the approximating sequence (8.3), and Theorem 8.5 all depend on it.
  • domain assumption Moment compactness Assumption 8.3: sup_t int |x|^q dphi_t < inf for some q > max{a,r}
    Required for tightness in Theorem 8.5; Remark 8.4 shows it follows from energy-sublevel estimates in the attractive regimes and from [20]'s moment bounds when a=r<1, but for other regimes it is a genuinely separate hypothesis.

pith-pipeline@v1.3.0-alltime-deepseek · 41862 in / 22067 out tokens · 221114 ms · 2026-08-02T08:38:41.589040+00:00 · methodology

0 comments
read the original abstract

We study a nonlocal continuity equation on $\mathbb{R}^d$ in which a probability density is driven by the competition between attraction toward a prescribed background measure $\omega$ and self-repulsion among particles, governed respectively by the power-law kernels $\psi_a(x) = |x|^{1+a}$ and $\psi_r(x) = |x|^{1+r}$ with exponents $a, r \in [0,1)$. We establish global Lagrangian well-posedness via a squared-radius regularization, obtaining uniform $L^\infty$ and moment bounds, $W^{n,\infty}$ regularity, and uniqueness in the Lagrangian class. When the initial data is compactly supported and attraction dominates ($a > r$, or $a = r$ with $\omega(\mathbb{R}^d) > 1$), we prove that the support remains uniformly bounded at all time; a counterexample shows this fails for $a = r > 1$. For the attractive-dominant nonquadratic range $0 \leq r \leq a < 1$, we characterize zero-flux stationary states via a free-boundary problem involving a fractional Laplacian operator, reducing the stationarity condition to a fractional exterior Dirichlet problem. This characterization allow us to exhibit explicit examples of stationary measures in dimensions $d \in \{1,2,3\}$. Numerical particle simulations confirm agreement with the theoretical stationary profiles. Finally, we prove that every global solution with bounded energy and uniform moment bounds converges to a zero-flux stationary state.

Figures

Figures reproduced from arXiv: 2607.04424 by Hui Huang, Lukang Sun, Massimo Fornasier.

Figure 1
Figure 1. Figure 1: One-dimensional verification in the attractive-dominant regime [PITH_FULL_IMAGE:figures/full_fig_p039_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two-dimensional numerical verification for [PITH_FULL_IMAGE:figures/full_fig_p042_2.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 2 linked inside Pith

  1. [1]

    Ambrosio, N

    L. Ambrosio, N. Gigli, and G. Savar´ e.Gradient flows in metric spaces and in the space of probability measures. Lectures in Mathematics ETH Z¨ urich. Birkh¨ auser Verlag, Basel, second edition, 2008

    2008

  2. [2]

    Arbel, A

    M. Arbel, A. Gretton, J. Niles-Weed, and P. Rigollet. Maximum mean discrepancy gradient flow.Advances in Neural Information Processing Systems, 32, 2019. 55 Result Kind Holds? Thm. 4.1: well-posedness, L∞/moment bounds,W n,∞ regularity, uniqueness in La- grangian class Global well-posedness✓ Thm. 5.6 Uniform support confine- ment × Prop. 6.5, Thm. 6.2 Zer...

    2019

  3. [3]

    C. Berg, J. P. R. Christensen, and P. Ressel.Harmonic Analysis on Semigroups: Theory of Positive Definite and Related Functions, volume 100 ofGraduate Texts in Mathematics. Springer-Verlag, New York, 1984

    1984

  4. [4]

    Bogdan and T

    K. Bogdan and T. Byczkowski. Potential theory for theα-stable schr¨ odinger operator on bounded lipschitz domains.Studia Mathematica, 133(1):53–92, 1999

    1999

  5. [5]

    T. Boggio. Sulle funzioni di Green d’ordinem.Rendiconti del Circolo Matematico di Palermo, 20:97–135, 1905

    1905

  6. [6]

    J. A. Ca˜ nizo, J. A. Carrillo, and F. S. Patacchini. Existence of compactly supported global min- imisers for the interaction energy.Archive for Rational Mechanics and Analysis, 217(3):1197– 1217, 2015

    2015

  7. [7]

    J. A. Carrillo, K. Craig, and Y. Yao. Aggregation-diffusion equations: Dynamics, asymptotics, and singular limits. In N. Bellomo, P. Degond, and E. Tadmor, editors,Active Particles, Volume 2: Advances in Theory, Models, and Applications, Modeling and Simulation in Science, Engineering and Technology, pages 65–108. Birkh¨ auser, Cham, 2019

    2019

  8. [8]

    J. A. Carrillo, M. G. Delgadino, and A. Mellet. Regularity of local minimizers of the interaction energy via obstacle problems.Communications in Mathematical Physics, 343(3):747–781, 2016

    2016

  9. [9]

    J. A. Carrillo, M. Di Francesco, A. Figalli, T. Laurent, and D. Slepˇ cev. Global-in-time weak measure solutions and finite-time aggregation for nonlocal interaction equations.Duke Math- ematical Journal, 156(2):229–271, 2011. 56

    2011

  10. [10]

    J. A. Carrillo, M. Di Francesco, A. Figalli, T. Laurent, and D. Slepˇ cev. Confinement in nonlocal interaction equations.Nonlinear Analysis: Theory, Methods & Applications, 75(2):550–558, 2012

    2012

  11. [11]

    J. A. Carrillo, R. J. McCann, and C. Villani. Kinetic equilibration rates for granular media and related equations: Entropy dissipation and mass transportation estimates.Revista Matem´ atica Iberoamericana, 19(3):971–1018, 2003

    2003

  12. [12]

    J. A. Carrillo, R. J. McCann, and C. Villani. Contractions in the 2-Wasserstein length space and thermalization of granular media.Archive for Rational Mechanics and Analysis, 179(2):217–263, 2006

    2006

  13. [13]

    Chizat, M

    L. Chizat, M. Colombo, R. Colombo, and X. Fern´ andez-Real. Quantitative convergence of Wasserstein gradient flows of kernel mean discrepancies, 2026. Preprint, arXiv:2603.01977

    arXiv 2026

  14. [14]

    Colasanto, M

    F. Colasanto, M. Focardi, M. Fornasier, and F. Mattesini. Sharp rates of MMD empirical estimation with power kernels.arXiv preprint arXiv:2605.18497, 2026. Submitted 18 May 2026, available athttps://arxiv.org/abs/2605.18497

    Pith/arXiv arXiv 2026

  15. [15]

    Di Francesco, M

    M. Di Francesco, M. Fornasier, J.-C. H¨ utter, and D. Matthes. Asymptotic behavior of gradient flows driven by nonlocal power repulsion and attraction potentials in one dimension.SIAM Journal on Mathematical Analysis, 48(5):3232–3270, 2016

    2016

  16. [16]

    Dipierro and H.-C

    S. Dipierro and H.-C. Grunau. Boggio’s formula for fractional polyharmonic Dirichlet problems. Annali di Matematica Pura ed Applicata (1923-), 196(4):1327–1344, 2017

    1923

  17. [17]

    Duong, V

    R. Duong, V. Stein, R. Beinert, J. Hertrich, and G. Steidl. Wasserstein gradient flows of MMD functionals with distance kernel and Cauchy problems on quantile functions.ESAIM: Control, Optimisation and Calculus of Variations, 32:10, 2026. Published online 11 February 2026

    2026

  18. [18]

    B. Dyda. Fractional calculus for power functions and eigenvalues of the fractional laplacian. Fractional Calculus and Applied Analysis, 15(4):536–555, 2012

    2012

  19. [19]

    Fornasier, J

    M. Fornasier, J. Haˇ skovec, and G. Steidl. Consistency of variational continuous-domain quan- tization via kinetic theory.Applicable Analysis, 92(6):1283–1298, 2013

    2013

  20. [20]

    Fornasier and J.-C

    M. Fornasier and J.-C. H¨ utter. Consistency of probability measure quantization by means of power repulsion–attraction potentials.Journal of Fourier Analysis and Applications, 22(3):694–749, 2016

    2016

  21. [21]

    Fournier and A

    N. Fournier and A. Guillin. On the rate of convergence in Wasserstein distance of the empirical measure.Probab. Theory Related Fields, 162(3-4):707–738, 2015

    2015

  22. [22]

    I. M. Gelfand and G. E. Shilov.Generalized Functions, Volume 1: Properties and Operations. Academic Press, New York, 1964

    1964

  23. [23]

    Grafakos.Classical Fourier Analysis, volume 249 ofGraduate Texts in Mathematics

    L. Grafakos.Classical Fourier Analysis, volume 249 ofGraduate Texts in Mathematics. Springer, New York, 3 edition, 2014

    2014

  24. [24]

    Gretton, K

    A. Gretton, K. M. Borgwardt, M. J. Rasch, B. Sch¨ olkopf, and A. Smola. A kernel two-sample test.Journal of Machine Learning Research, 13:723–773, 2012

    2012

  25. [25]

    Hertrich, R

    J. Hertrich, R. Gruber, K. Banachewicz, P. Wald, and M. Genzel. Generative sliced MMD flows with Riesz kernels.arXiv preprint arXiv:2305.11462, 2023. 57

    Pith/arXiv arXiv 2023

  26. [26]

    Rosenzweig, D

    M. Rosenzweig, D. Slepˇ cev, and L. Wang. Wasserstein gradient flow of maximum mean dis- crepancy with negative distance kernel. Work in progress, known to the authors by private communication, 2025

    2025

  27. [27]

    E. M. Stein and G. Weiss.Introduction to Fourier Analysis on Euclidean Spaces. Princeton University Press, Princeton, 1971

    1971

  28. [28]

    Steinwart and A

    I. Steinwart and A. Christmann.Support Vector Machines. Springer, New York, 2008. Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held ...

    2008