{"id":"1d97e964-7f8c-4c74-8266-d6289a8299ef","arxiv_id":"1908.08080","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A new embedding into locally compact spaces proves existence of solutions to martingale problems for probability measure valued jump-diffusions in generalized Wasserstein spaces, covering drift, diffusion, and infinite-activity jumps.","lead":"This paper proves a general existence theorem for probability measure valued jump-diffusions on Wasserstein-type spaces, embedding the state space into a locally compact space so classical martingale problem theory applies. The method covers drift, diffusion, and infinite-activity jumps, and gives direct existence proofs for McKean-Vlasov equations with common noise.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main theorem is sound, but condition (iii) is a genuinely ad hoc barrier assumption verified case-by-case, and the McKean-Vlasov application rests on unproved filtration compatibility.","rationale":"The reader's weakest_assumption identifies condition (iii) as the central technical bottleneck, and my read confirms this. The proof of Theorem 3.4 is structurally sound: the embedding T is a homeomorphism, D is dense in C0(X), the positive maximum principle is transferred correctly via Lemma 3.6(ii), and the Gronwall argument is valid because the positive parts of the approximating generator outputs are uniformly bounded, which justifies the limsup exchange. The bar for the theorem is therefore exactly whether condition (iii) can be verified for a given operator; the paper gives a workable sufficient condition in Lemma 6.3 and demonstrates it for the diffusion examples in Section 7. However, the abstract claims general dynamics including possibly infinite-activity jumps, and condition (iii) must be re-verified for each such jump structure; the sketches in Section 8 for jump-type common noise are less complete. Additionally, the McKean-Vlasov application in Theorem 8.2 is conditional on filtration compatibility assumptions that are not established by the martingale problem construction, and the remark that one can always construct a suitable ~W is unproved. This does not undermine the abstract existence theorem, but it does justify the reader's CONDITIONAL verdict rather than unconditional acceptance.","tokens_in":26469,"tokens_out":41479,"duration_ms":382609,"concrete_test":"Verify condition (iii) for a pure-jump Lévy-type operator with a jump kernel that moves mass to points with large w (e.g., N(μ,dν)=c 1_{⟨w,ν⟩>2⟨w,μ⟩}e^{-⟨w,ν⟩}dν): compute the barrier functions from Lemma 6.3 and check whether ~g_m^+ ≤ c'~f holds on X_c. If it fails, condition (iii) excludes jump operators with unbounded jump sizes, limiting the abstract's 'infinite-activity jumps' claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Condition (iii) of Theorem 3.4 is the load-bearing assumption: it constructs, for every c≥1, barrier functions ~f vanishing exactly on T(P_w)∩X_c, with control on the positive part of ~L~f_m. Without it, the auxiliary solution Y on the locally compact space X could charge the cemetery point Δ, and the martingale problem on P_w would not be solved. This condition is not implied by (i), (ii), or (iv); it must be checked for each operator. The paper offers Lemma 6.3 as a verification tool, and Theorem 7.1 checks it for diffusion-type coefficients, but the check itself requires the coefficient functions to have well-defined limits at the point at infinity (conditions 7.3 and 7.4). A further soft spot is in Section 8: Theorem 8.2 identifies X_t with the conditional law of Z_t only under the ad hoc compatibility conditions (W independent of G, and F_s, G_t conditionally independent given G_s), and the remark asserting that a suitable ~W can always be constructed is stated without proof. The proof of Theorem 3.4 itself is internally coherent; in particular, the Gronwall step is justified because the ~g_m^+ are uniformly bounded.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":26726,"tokens_out":13953,"duration_ms":132961,"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":[{"comment":"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.","section":"§8, Lemma 8.5"},{"comment":"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.","section":"§8, Theorem 8.2 and following remark"},{"comment":"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.","section":"§3, Theorem 3.4(iii)"}],"minor_comments":[{"comment":"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)=−∞.'","section":"§3, proof of Lemma 3.6(ii)"},{"comment":"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.","section":"§3 and §6"},{"comment":"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.","section":"§8, Theorem 8.2 and Lemma 8.6"},{"comment":"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.","section":"§1 and §7"},{"comment":"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.","section":"§7, Theorem 7.1"}],"recommendation":"major_revision","confidential_remarks":"The main abstract theorem appears sound, and the proof of Theorem 3.4 is careful. The obstacles to acceptance are concentrated in Section 8: the omitted quadratic-variation verification in Lemma 8.5 and the unproved construction of ~W after Theorem 8.2 are both load-bearing for the advertised McKean–Vlasov application. These gaps seem fixable within the scope of the manuscript, so I recommend major revision rather than rejection. The relation to the authors' prior work is acknowledged appropriately, and I found no circularity in the use of the optimality conditions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Nothing to hide: this is a strong paper. The embedding T(μ)=wμ into M_+(E^Δ) is a simple device that actually works, and it turns the non-locally-compact Wasserstein space P_w into a locally compact set where the classical Ethier–Kurtz theory applies. That is the main new thing, and the proof of Theorem 3.4 is coherent. Lemma 3.6(ii) is the key technical step and it works; the Gronwall step is justified because the ~g_m^+ are uniformly bounded. The C0-type test function algebra (3.3) is the right choice for making the pushed-forward test functions vanish at the cemetery state, and the overall approach genuinely bypasses particle-system approximations.\n\nThe positive maximum principle tools in Section 5 are also a real extension of Cuchiero–Larsson–Svaluto-Ferro to Lévy type operators, including infinite-activity jumps, and the verification toolkit in Sections 6–7 is practically useful.\n\nNow the soft spots, in proportion. Condition (iii) of Theorem 3.4 is ad hoc. It is not implied by the other conditions; it is a barrier assumption that must be checked for each operator, and it is exactly what keeps the auxiliary solution from charging the point at infinity Δ. The paper does provide tools (Lemma 6.3, Theorem 7.1), but the check itself requires coefficient limits at infinity. So the main theorem is honest but the user must do real work. I do not consider this a flaw in the core argument, but it is a limit of the method that readers should know about.\n\nThe McKean–Vlasov application in Section 8 is the weakest part. Lemma 8.5 states that the quadratic variation computation is omitted, and Theorem 8.2 needs explicit compatibility conditions on filtrations to identify the conditional law, with the remark that a suitable ~W can always be constructed stated without proof. These are not cosmetic. Anyone citing this for McKean–Vlasov equations will need those details or will need to work around them. The rest of the paper stands independently.\n\nCitation pattern is fine: the dependence on Cuchiero et al. (2019) is for independently established optimality conditions, and there is no circularity. No machine-checked proofs, but that is not expected for this kind of paper.\n\nBottom line: this deserves a serious referee and, after the Section 8 gaps are either filled or the claims reduced to what is proved, publication. I would cite it for the embedding and the abstract existence theorem; I would be careful citing the McKean–Vlasov part as is.","headline":"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.","tokens_in":27260,"tokens_out":1815,"would_cite":true,"duration_ms":16809,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J60","60J75","60G57"],"pacs":[],"model":"deepseek-v4-flash","headline":"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.","keywords":["probability measure valued processes","martingale problem","Wasserstein spaces","positive maximum principle","McKean–Vlasov equations","common noise","jump-diffusions","Lévy type operators"],"falsifier":"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.","tokens_in":26290,"feed_emoji":"📊","tokens_out":9562,"duration_ms":123091,"temperature":0.7,"pith_summary":"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.","feed_headline":"Measure-valued jump-diffusions exist on Wasserstein-type spaces","feed_subtitle":"A weighted embedding into a locally compact space unlocks classical existence theorems for drift, diffusion, and jumps.","key_machinery":"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.","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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)."],"supporting_citations":[{"why":"Supplies the classical locally compact existence theorem (Theorem 2.4) and the Gronwall inequality used to show the auxiliary solution avoids Δ.","marker":"Ethier and Kurtz (2005)"},{"why":"Provides the optimality conditions for test functions that the paper extends to prove the positive maximum principle for Lévy-type operators (Theorem 5.1).","marker":"Cuchiero et al. (2019)"},{"why":"Supplies the Fubini-type result (Lemma A.1) that identifies the conditional-law process with the solution of the linear equation in Theorem 8.2.","marker":"Kailath et al. (1978)"},{"why":"Classical result converting martingale problem solutions to weak solutions of SDEs, used in Lemma 8.5.","marker":"Stroock and Varadhan (1972)"},{"why":"Invariance result used to verify the diffusion-matrix condition in Theorem 5.1(iii) and to justify flows staying in E.","marker":"Da Prato and Frankowska (2004)"},{"why":"Supplies the flow argument used in the proof of Theorem 5.1(iii).","marker":"Abi Jaber et al. (2019)"}],"fun_headline_variants":["Measure-valued jump-diffusions via compact embedding","Jump-diffusions on Wasserstein spaces exist","Measure-valued SDEs solved via compactification","Existence for measure-valued jump-diffusions","Compact embedding yields measure-valued solutions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Measure-valued jump-diffusions via compact embedding","Jump-diffusions on Wasserstein spaces exist","Measure-valued SDEs solved via compactification","Existence for measure-valued jump-diffusions","Compact embedding yields measure-valued solutions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000188,"raw_usage":{"total_tokens":1295,"prompt_tokens":869,"completion_tokens":426,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":485,"completion_tokens_details":{"reasoning_tokens":358}},"tokens_in":485,"tokens_out":426,"duration_ms":14735,"temperature":1.0,"reasoning_tokens":358,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:50:38.231565+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[],"review_version":1}