Pith. sign in

REVIEW 3 major objections 5 minor 28 references

Existence of probability measure valued jump-diffusions in generalized Wasserstein spaces

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper proves that probability-measure-valued jump-diffusions on generalized Wasserstein spaces exist, by embedding those spaces into locally compact ones where classical martingale-problem theory applies.

desk verdict A real existence theorem for measure-valued jump-diffusions in Wasserstein spaces, with an ad hoc but checkable barrier condition and some unfinished business in the McKean–Vlasov application. read the letter →

arxiv 1908.08080 v2 pith:4TQTTGMX submitted 2019-08-21 math.PR

Martin Larsson , Sara Svaluto-Ferro This is my paper · ORCID
classification math.PR MSC 60J6060J7560G57
keywords probabilitymeasurevaluedprocessesmartingaleproblemWassersteinspacespositivemaximumprincipleMcKean–Vlasovequationscommonnoisejump-diffusionsLévytypeoperators
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 proves that a wide class of probability-measure-valued jump-diffusions—processes whose state is itself a probability measure, evolving through drift, diffusion, and jumps that may have infinite activity—exist as solutions to martingale problems, without first going through interacting particle systems and a large-population limit. The key step is to take a Wasserstein-type space $P_w$ of measures with finite $w$-moment and map it into the locally compact space $X=\{\nu\in M_+(E^\Delta):\langle w^{-1},\nu\rangle=1\}$ by $T(\mu)(dx)=w(x)\mu(dx)$, where $E^\Delta$ is the one-point compactification of the underlying space. On $X$, classical existence theory for martingale problems applies, and a barrier-function condition ensures the resulting process never charges the added point $\Delta$, so it lives in $P_w$. The paper also gives verifiable tools—optimality conditions for the positive maximum principle and continuity and growth checks for operators of Lévy type—and applies the result to mean-field particle systems with common noise and to McKean–Vlasov equations.

What carries the argument

The load-bearing object is the embedding $T(\mu)(dx)=w(x)\mu(dx)$ from $P_w$ into $X=\{\nu\in M_+(E^\Delta):\langle w^{-1},\nu\rangle=1\}$, with $E^\Delta$ the one-point compactification of $E$; it is a topological embedding because convergence in $P_w$ is weak convergence plus convergence of $\langle w,\mu\rangle$. Its effect is to turn the non-locally-compact Wasserstein space into a closed subset of a locally compact Polish space, so the classical equivalence between the positive maximum principle and existence of possibly killed martingale-problem solutions (Theorem 2.4, from Ethier and Kurtz) becomes available. The auxiliary process on $X$ is then kept away from $\Delta$ by the barrier condition (iii), which is verified in applications through approximations by the generator's graph and through the optimality conditions of Theorem 5.1 for Lévy-type operators.

What would settle it

The cleanest way to refute the claim would be to exhibit a Lévy-type operator satisfying the analytic conditions (i), (ii), and (iv) of Theorem 3.4 together with the growth bounds of Theorem 7.1 at $\gamma=0$, and to show—analytically or by an exact simulation of the auxiliary process on $X$—that the solution charges the cemetery point $\Delta$ with positive probability. Since the theorem identifies $\Delta$-charging with killing, any such example would contradict both Theorem 3.4 and Theorem 7.1's 'no killing' conclusion.

Watch

Extended reading notes

Core claim

The central claim, Theorem 3.4, is that a linear operator $L$ on the test-function algebra $D_w$ generated by $\langle\phi,\mu\rangle e^{-\langle w,\mu\rangle}$ produces a possibly killed martingale-problem solution on $P_w$ for every initial measure, provided $L$ satisfies the positive maximum principle (i), maps $D_w$ to functions of $C_0$ type (ii), and admits the barrier functions of condition (iii) for every mass level; condition (iv) then upgrades 'possibly killed' to 'stays in $P_w$'. The proof pushes the martingale problem through the embedding $T$, solves the auxiliary problem on the locally compact space $X$, and uses Gronwall's inequality together with (iii) to conclude that the auxiliary solution never places mass at the cemetery point $\Delta$, hence corresponds to a true $P_w$-valued solution. Concrete sufficient conditions for (i)–(iv) are developed for Lévy-type operators with drift, diffusion, and possibly non-summable jumps, and the method yields existence for McKean–Vlasov equations with common Brownian or jump noise.

Load-bearing premise

Everything rests on condition (iii) of Theorem 3.4: for every bounded mass level one must be able to approximate, in the bounded-pointwise sense from the generator's graph, a barrier function that is zero exactly when the embedded measure has no mass at the added point $\Delta$, with its positive part controlled by the barrier; this ad hoc condition is what prevents the auxiliary solution from escaping to $\Delta$ and must be checked separately in each application.

Editorial extensions

If this is right

  • Existence of measure-valued processes with drift, diffusion, and infinite-activity jumps is obtained directly on the limiting space, not as a limit of particle systems.
  • For mean-field particle systems with common Brownian or Poisson noise, the limiting empirical distribution is realized as a solution of the martingale problem, with the conditional-law identification holding under the stated compatibility conditions.
  • Under the linear-growth version of the growth conditions, all moments $\mathbb{E}[\langle w,X_t\rangle^k]$ are finite and the solution remains at all times in $P_w$ (Proposition 7.6).
  • The framework covers distribution-dependent Fleming–Viot-type models where the sampling-replacement rate depends on the whole type distribution, not just on the pair of types.
  • If the linear equation (8.3) has a uniqueness property, the constructed solution is a genuine weak solution of the McKean–Vlasov equation, and a uniqueness criterion is given for polynomial coefficients (Remark 8.3).

Reading between the lines

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

  • Because the proof only uses that $w\geq 1$, $w(x)\to\infty$, and $w^{-1}$ extends continuously to $\Delta$, the same embedding should work for weight functions growing faster than any polynomial, reaching Wasserstein spaces with super-polynomial moment conditions.
  • The compatibility conditions in Theorem 8.2 look removable: the paper notes without proof that a change of Brownian motion can make it independent of the relevant filtration, so the existence statement for McKean–Vlasov equations with common noise may hold without the conditional-independence assumption.
  • Condition (iii) is the only assumption that is not automatically implied by the positive maximum principle; a natural extension would be to find systematically verifiable sufficient conditions for it, perhaps in terms of the behaviour of the Lévy kernel near infinity.
  • The same compactification idea could be applied to other non-locally-compact state spaces of measures, such as spaces of sub-probability measures with tempered moments, or spaces of measures on Riemannian manifolds with a chosen weight.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper develops an abstract existence theory for probability-measure-valued jump-diffusions on generalized Wasserstein spaces P_w. The central device, introduced in Section 3, embeds P_w into the locally compact space X = {ν ∈ M_+(E^Δ) : ⟨w^{-1},ν⟩ = 1} via T(μ) = wμ, so that classical Ethier–Kurtz martingale-problem existence theory can be applied to an auxiliary problem on X. The main abstract theorem, Theorem 3.4, asserts existence of possibly killed solutions under the positive maximum principle, a C0-type condition, and a barrier condition (iii), with a further condition (iv) guaranteeing non-killing. Sections 4–6 provide tools for verifying these conditions for Lévy-type operators, including extensions of optimality conditions from prior work. Sections 7–8 apply the results to McKean–Vlasov diffusions with common noise, including a discussion of uniqueness for the conditional-law equation. The proof of Theorem 3.4 is internally coherent: Lemma 3.6(ii) provides the required point-preserving approximation of measures in X by measures in T(P_w), and the subsequent Gronwall argument is justified by the uniform boundedness of the approximating functions.

Significance. If the main theorem is accepted, the paper gives a genuinely useful existence route for measure-valued processes with drift, diffusion, and possibly infinite-activity jumps in Wasserstein-type spaces, bypassing particle-system approximations. The embedding of P_w into a locally compact cone is simple, explicit, and likely to be reusable. The paper also supplies practically checkable criteria for the positive maximum principle and for the required C0-type conditions, including extensions of earlier optimality results; these contributions are substantial. I found no circularity: the reliance on Cuchiero et al. (2019) and on classical Ethier–Kurtz theory is legitimate, and the main abstract proof is not circular. However, the value of the Section 8 application is currently reduced by two explicitly omitted proof details and by compatibility conditions that are stated but not established; these issues are fixable but are load-bearing for the advertised McKean–Vlasov existence result.

major comments (3)
  1. [§8, Lemma 8.5] The proof of Lemma 8.5 leaves the central verification undone: the process ⟨φ,X_t⟩−∫_0^t⟨B_{X_s}φ,X_s⟩ds−∫_0^t⟨τ_{X_s}∇φ,X_s⟩^⊤ dW^0_s is asserted to be constant after 'verifying that its quadratic variation is zero; we omit the details,' and the proof of (8.1) is dismissed as 'similar.' These identities are exactly what Theorem 8.2 needs to identify X as the conditional law, so this is not a routine omission. Moreover, the argument that h(μ,z,x,x_0)=ψ(x,x_0) lies in D(H) requires justification, since D_w does not contain the constant function 1 and the domain D(H) was defined using products with f ∈ D_w. Please supply the quadratic-variation computation, or a precise reference that covers it, and provide the needed density or localization argument for the coordinate functions.
  2. [§8, Theorem 8.2 and following remark] The compatibility conditions in Theorem 8.2—independence of W from G and conditional independence of F_s and G_t given G_s—are stated as hypotheses under which (8.3) holds for Y_t = P(Z_t ∈ · | G_t), but the paper never proves that a solution produced by Corollary 8.4 satisfies them. The remark asserting that a suitable ~W can always be constructed is introduced with 'let us also mention (without proof).' As a consequence, the advertised McKean–Vlasov existence result is conditional on unproved structural assumptions. Please either prove the construction of ~W, restrict the theorem to an explicitly verifiable subclass of initial data and coefficients where the compatibility conditions are automatic, or state clearly that the conditional-law identification is an additional assumption rather than a consequence of the martingale-problem construction.
  3. [§3, Theorem 3.4(iii)] Condition (iii) is genuinely ad hoc: it is not implied by conditions (i), (ii), or (iv), and its verification is delegated to Lemma 6.3 and then to Theorem 7.1, where it requires additional asymptotic conditions on the coefficients at infinity, such as (7.3)–(7.4). Because this condition is load-bearing for the step showing that the auxiliary solution does not charge the cemetery point Δ, I recommend that the introduction and the discussion after Theorem 3.4 state more explicitly that the method requires a case-by-case barrier verification, rather than merely presenting condition (iii) as one of several technical assumptions. Adding a simple example where condition (iii) fails would help calibrate the scope of the abstract result.
minor comments (5)
  1. [§3, proof of Lemma 3.6(ii)] The displayed limits for α(t) state 'lim_{t→∞} α(t)=∞ and lim_{t→−∞} α(t)=−∞,' but t is restricted to (1,∞); the second limit should be 'lim_{t↓1} α(t)=−∞.'
  2. [§3 and §6] The paper uses 'bp-closure' without defining it at first use in Theorem 3.4 and Definition 6.2. Please define the bounded-pointwise closure explicitly, or give a reference, so that readers do not have to infer it from the surrounding text.
  3. [§8, Theorem 8.2 and Lemma 8.6] The compatibility conditions refer to a filtration F that is not defined in the theorem or lemma. Please identify F explicitly; presumably it is the filtration generated by (X,Z,W,W^0), but this should be stated.
  4. [§1 and §7] The introduction's claim that the method 'allows for general dynamics' should be tempered by a sentence acknowledging that condition (iii) of Theorem 3.4 is an additional barrier condition verified case by case. Theorem 7.1 provides one verification, but the abstract result itself does not imply it.
  5. [§7, Theorem 7.1] The proof of Theorem 7.1 is written only for d=1, while the statement and conditions (7.1)–(7.4) are for general d. Please indicate how the multidimensional case follows, especially regarding the interpretation of the limits at infinity in (7.3) and the verification of Lemma 6.3 in higher dimensions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the embedding-plus-martingale argument is self-contained, with only independent self-citations for auxiliary optimality conditions.

full rationale

The derivation chain is not circular. Theorem 3.4 is a conditional existence result: given an operator L whose generator satisfies (i)-(iv), the proof transports the problem to the locally compact space X via the embedding T, proves that the pushed-forward operator tilde-L satisfies the positive maximum principle, and invokes the classical Ethier-Kurtz martingale existence theorem. Condition (iii) is a genuine Lyapunov/barrier hypothesis: it postulates functions in the bp-closure of the restricted graph whose zero set is exactly T(P_w) intersect X_c, and the proof uses these functions only to show that the auxiliary solution cannot charge the cemetery point Delta. That is an input sufficient condition, not a disguised version of the conclusion. Condition (iv) is similarly a standard non-explosion test. The paper's same-author citations (notably Cuchiero, Larsson, and Svaluto-Ferro 2019) supply optimality conditions used to verify the positive maximum principle; those results are proved independently in the cited work, do not assume the present existence theorem or condition (iii), and are not fitted to any data. No parameter is estimated and no 'prediction' is defined in terms of the quantity it is supposed to predict. The unproved remark in Section 8 about constructing the process tilde-W, and the implicit nature of the filtration compatibility conditions in Theorem 8.2, are rigor or applicability gaps rather than circularity: they identify hypotheses under which the conditional law can be recognized as a solution of the linear equation, but they do not reduce the main theorem to its own conclusion. Therefore no circular step can be exhibited from the text.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No data fitting is involved. The central claim rests on applying classical martingale problem theory to a constructed locally compact embedding; the auxiliary conditions (i)-(iv) are abstract hypotheses verified in applications via the provided lemmas. No new physical or mathematical entities are postulated.

assumptions (5)
  • standard math Classical existence theory for martingale problems on locally compact spaces (Ethier and Kurtz 2005, Theorem 4.5.4), quoted as Theorem 2.4.
    Used in the proof of Theorem 3.4 to obtain a possibly killed solution to the auxiliary martingale problem on the locally compact space X.
  • standard math Optimality conditions for differentiable functions of probability measures at extrema (Cuchiero, Larsson, Svaluto-Ferro 2019, Theorem 3.1), used in Theorem 5.1(i)-(ii).
    These published results are the basis for verifying the positive maximum principle for operators of Lévy type.
  • standard math Stochastic invariance condition for E-valued diffusions (Da Prato and Frankowska 2004; Abi Jaber, Bouchard, and Illand 2019), used in Theorem 5.1(iii).
    The condition (5.1) is the standard way to preserve the closed subset E under diffusion dynamics.
  • standard math Fubini type theorem for stochastic integrals under conditional independence (Kailath, Segall, and Zakai 1978), stated as Lemma A.1.
    Used in Lemma 8.6 to derive the equation for the conditional law process Y_t.
  • domain assumption The operator L is assumed to be of Lévy type (4.1) in the applications, with coefficient continuity and growth conditions (7.1)-(7.2).
    Theorems 7.1 and 7.4 rely on these structural coefficient assumptions to verify the abstract conditions of Theorem 3.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Existence of probability measure valued jump-diffusions in generalized Wasserstein spaces." pith.science (2026). https://pith.science/paper/4TQTTGMX

@misc{pith2026190808080,
  author       = {Pith},
  title        = {Pith review of: Existence of probability measure valued jump-diffusions in generalized Wasserstein spaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4TQTTGMX}},
  note         = {Machine review of arXiv:1908.08080}
}
read the original abstract

We study existence of probability measure valued jump-diffusions described by martingale problems. We develop a simple device that allows us to embed Wasserstein spaces and other similar spaces of probability measures into locally compact spaces where classical existence theory for martingale problems can be applied. The method allows for general dynamics including drift, diffusion, and possibly infinite-activity jumps. We also develop tools for verifying the required conditions on the generator, including the positive maximum principle and certain continuity and growth conditions. To illustrate the abstract results, we consider large particle systems with mean-field interaction and common noise.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 27 canonical work pages

  1. [1]

    Abi Jaber, B

    E. Abi Jaber, B. Bouchard, and C. Illand. Stochastic invariance of closed sets with non-lipschitz coefficients. Stochastic Processes and their Applications, 129 0 (5): 0 1726 -- 1748, 2019

    work page 2019

  2. [2]

    Carmona and F

    R. Carmona and F. Delarue. Probabilistic Theory of Mean Field Games with Applications I-II . Springer, 2017

    work page 2017

  3. [3]

    Cuchiero

    C. Cuchiero. Polynomial processes in stochastic portfolio theory. Stochastic processes and their applications, 129 0 (5): 0 1829--1872, 2019

    work page 2019

  4. [4]

    Cuchiero, M

    C. Cuchiero, M. Larsson, and S. Svaluto-Ferro. Probability measure-valued polynomial diffusions. Electronic Journal of Probability, 24, 2019

    work page 2019

  5. [5]

    D. Dawson. The critical measure diffusion process. Zeitschrift f \"u r Wahrscheinlichkeitstheorie und Verwandte Gebiete , 40 0 (2): 0 125--145, 1977

    work page 1977

  6. [6]

    D. Dawson. Geostochastic calculus. The Canadian Journal of Statistics / La Revue Canadienne de Statistique, 6 0 (2): 0 143--168, 1978

    work page 1978

  7. [7]

    D. Dawson. Measure-valued M arkov processes. In \' E cole d'\' E t\'e de P robabilit\'es de S aint- F lour XXI --1991 , Lecture Notes in Mathematics, pages 1--260. Springer, 1993

    work page 1991

  8. [8]

    Dawson and J

    D. Dawson and J. Vaillancourt. Stochastic McKean-Vlasov equations . Nonlinear Differential Equations and Applications NoDEA, 2 0 (2): 0 199--229, 1995

    work page 1995

Show all 28 references
  1. [9]

    Etheridge

    A. Etheridge. Some Mathematical Models from Population Genetics: \'E cole D' \'E t \'e de Probabilit \'e s de Saint--Flour XXXIX-2009 . Lecture Notes in Mathematics. Springer, 2011

    2009

  2. [10]

    S. N. Ethier and T. G. Kurtz. The Infinitely-Many-Alleles Model with Selection as a Measure-Valued Diffusion, pages 72--86. Springer, Berlin, 1987

    1987

  3. [11]

    S. N. Ethier and T. G. Kurtz. Fleming-- V iot processes in population genetics. SIAM Journal on Control and Optimization, 31 0 (2): 0 345--386, 1993

    1993

  4. [12]

    S. N. Ethier and T. G. Kurtz. Markov Processes: Characterization and Convergence. Wiley Series in Probability and Statistics. Wiley, 2 edition, 2005

    2005

  5. [13]

    Fernholz

    R. Fernholz. Stochastic Portfolio Theory. Applications of Mathematics. Springer-Verlag, New York, 2002

    2002

  6. [14]

    Fernholz and I

    R. Fernholz and I. Karatzas. Stochastic portfolio theory: an overview. Handbook of numerical analysis, 15: 0 89--167, 2009

    2009

  7. [15]

    Filipovi \'c and M

    D. Filipovi \'c and M. Larsson. Polynomial Jump-Diffusion Models . ArXiv e-prints, 2017. URL https://arxiv.org/abs/1711.08043

    2017 arXiv

  8. [16]

    W. H. Fleming and M. Viot. Some measure-valued M arkov processes in population genetics theory. Indiana Univ. Math. J., 28 0 (5): 0 817--843, 1979

    1979

  9. [17]

    Florchinger and F

    P. Florchinger and F. Le Gland. Particle approximation for first order stochastic partial differential equations. In Applied stochastic analysis, pages 121--133. Springer, 1992

    1992

  10. [18]

    F \"o llmer and A

    H. F \"o llmer and A. Schied. Stochastic Finance: An Introduction in Discrete Time . De Gruyter studies in mathematics. Walter de Gruyter, 2 edition, 2004

    2004

  11. [19]

    K. Huang. Statistical mechanics. Wiley, 1987

    1987

  12. [20]

    Kailath, A

    T. Kailath, A. Segall, and M. Zakai. Fubini-Type Theorems for Stochastic Integrals . Sankhy \=a : The Indian Journal of Statistics, Series A (1961-2002) , 40 0 (2): 0 138--143, 1978

    1961

  13. [21]

    A. Klenke. Probability Theory: A Comprehensive Course . Universitext. Springer London, 2 edition, 2013

    2013

  14. [22]

    Kurtz and J

    T. Kurtz and J. Xiong. Particle representations for a class of nonlinear SPDEs . Stochastic Processes and their Applications, 83, 1999

    1999

  15. [23]

    E. Perkins. Dawson-Watanabe Superprocesses and Measure-valued Diffusions . In \`E cole d' \`E t \`e de Probabilit \`e s de Saint-Flour XXIX--1999 , Lecture Notes in Mathematics, pages 125--329. Springer, 2002

    1999

  16. [24]

    Da Prato and H

    G. Da Prato and H. Frankowska. Invariance of stochastic control systems with deterministic arguments. Journal of Differential Equations, 200 0 (1): 0 18 -- 52, 2004

    2004

  17. [25]

    D. W. Stroock and S. R. S. Varadhan. On the support of diffusion processes with applications to the strong maximum principle . In Proceedings of the Sixth Berkeley Symposium on Mathematical Statistics and Probability (Univ. California, Berkeley, Calif., 1970/1971), volume 3, p...

    1970

  18. [26]

    A. S. Sznitman. Topics in Propagation of Chaos . In \'E cole d' \'E t \'e de Probabilit \'e s de Saint-Flour XIX--1989 , Lecture Notes in Mathematics, pages 165--251. Springer, 1991

    1989

  19. [27]

    C. Villani. Optimal transport: old and new, volume 338. Springer Science & Business Media, 2008

    2008

  20. [28]

    Watanabe

    S. Watanabe. A limit theorem of branching processes and continuous state branching processes. J. Math. Kyoto Univ., 8 0 (1): 0 141--167, 1968

    1968

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.