{"id":"d7daf9f3-521a-448c-9e2b-293bcff33935","arxiv_id":"1908.07168","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A new type of extended backward stochastic Volterra integral equation is shown to provide a probabilistic representation for a non-local quasilinear parabolic PDE, generalizing the Pardoux-Peng Feynman-Kac formula.","lead":"This paper introduces a new class of backward stochastic Volterra integral equations and proves that the solutions generate solutions to a family of non-local parabolic partial differential equations. It extends the nonlinear Feynman-Kac formula from classical backward stochastic differential equations to this broader, time-inconsistent setting.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 5.2's proof invokes Theorem 4.7, which requires (B.5), not the stated (B.4); the final Riemann-sum/stochastic-integral limit is also asserted without proof.","rationale":"The reader's verdict of CONDITIONAL is appropriate, but the most load-bearing defect is sharper than the one singled out. Theorem 5.2's statement assumes (F.1)-(B.4), yet its proof uses Theorem 4.7, which is explicitly proved under (F.1)+(B.5). This is not a cosmetic issue: (B.5) supplies the uniform t-modulus used by Theorem 3.2 to get the continuity in the diagonal and the C^{0,0,2} regularity of Theta_hat. Without that regularity, the estimates (5.4)-(5.5) and the Taylor/Ito expansion leading to (5.6) have no justification. I therefore partially agree with the reader's focus on Section 5, but the precise missing hypothesis is more specific than the unproved convergence of the stochastic integrals. That latter step is also a genuine gap, because the paper does not show that the two stochastic-integral sums in (5.6) vanish as the mesh goes to zero; it only asserts a dominated convergence argument after postulating (5.7)-(5.10). Both gaps are repairable: either strengthen the theorem's assumptions to (B.5) and prove the limit, or replace the invoked regularity by a direct proof under (B.4). For this reason the paper should remain CONDITIONAL rather than ACCEPT or REJECT: the central idea is plausible and the likely fix is routine, but the main theorem as written is not fully proved.","tokens_in":29614,"tokens_out":23363,"duration_ms":250886,"concrete_test":"Analytical check of the assumption mismatch: in 1D, take b=0, sigma=1, psi=0, g(t,x)=cos(t x). This satisfies (B.4) but not (B.5), since sup_x |cos(t1 x)-cos(t2 x)|=2 for suitable x. Compute the explicit EBSVIE solution and see whether Theta_hat(t,s,x)=Y^{s,x}(t,s) is C^{0,0,2} and satisfies (5.1); if the diagonal regularity fails, Theorem 5.2 needs (B.5). Independently, supply the missing proof that the stochastic-integral sums in (5.6) converge to 0, e.g. via Ito isometry and a Holder estimate |Z^{s,x}(t,r)-Z^{r,x}(t,r)|<=K|s-r|^alpha; without such an estimate the last limit is unproved.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is Theorem 5.2, and its proof begins by invoking Theorem 4.7 to obtain C^{0,0,2}-regularity of {Y^{s,x}(t,s)} in (t,s,x). But Theorem 4.7 is stated and proved under (F.1)+(B.5), not merely (B.4). (B.5) adds a uniform modulus of continuity in t for g and psi that is not implied by (B.4): for example, g(t,x)=cos(tx) satisfies the C^3_b spatial bounds in (B.4) yet has no uniform modulus in t over x in R^d. All subsequent steps---(5.4)-(5.5), the Ito expansions leading to (5.6), and the final differentiation of the integral identity---rely on that regularity, so the theorem as stated is not established. Separately, the passage from (5.6) to the PDE is compressed into 'by dominated convergence theorem' after asserting (5.7)-(5.10) via Kolmogorov's theorem; the convergence to zero of the two stochastic-integral sums is not demonstrated. Even with (B.5) added, that last limit requires a written proof.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces extended backward stochastic Volterra integral equations (EBSVIEs), in which the generator depends on both Y(t,r) and the diagonal value Y(r,r). It establishes well-posedness in H^p under Lipschitz conditions (Theorem 3.1), a continuity and a priori estimate for the diagonal process Y(t,t) (Theorem 3.2), Malliavin differentiability of the solution (Theorem 4.1), Markovian regularity results (Corollary 4.2, Theorem 4.7), and finally a two-way Feynman--Kac correspondence between the Markovian EBSVIE (1.17) and the nonlocal quasilinear parabolic system (5.1). The main advertised result is Theorem 5.2: under (F.1)--(B.4), the deterministic function \\hat{\\Theta}(t,s,x)=Y^{s,x}(t,s) is the unique classical solution of (5.1), generalizing the Pardoux--Peng nonlinear Feynman--Kac formula.","tokens_in":29764,"tokens_out":24184,"duration_ms":250098,"significance":"If the main theorem is established, the paper gives a probabilistic representation for a class of nonlocal parabolic PDEs that arise in time-inconsistent control, and it does so without requiring the uniform ellipticity condition (1.10) that appears in Wang--Yong [30]. The structure of the paper is sensible: Theorem 5.1 is a direct Itô-argument, the well-posedness framework for EBSVIEs is useful in itself, and the explicit treatment of the diagonal coupling Y(r,r) is a genuine extension of earlier BSVIE results. However, the advertised converse direction is not proved as written: the proof of Theorem 5.2 relies on regularity results stated under a stronger assumption, and several load-bearing convergence and identification steps are asserted rather than shown. These gaps appear fixable, but they require nontrivial additional work.","major_comments":[{"comment":"Theorem 5.2 is stated under (F.1)--(B.4), but its proof begins by invoking Theorem 4.7, which is stated and proved under (F.1)--(B.5), not (B.4). The extra condition in (B.5) is a uniform modulus of continuity in t for both g and ψ, and it is not implied by (B.4): for example g(t,x)=cos(tx) satisfies the C^3_b spatial bounds required by (B.4) for each t, yet it has no uniform modulus in t over x∈R^d. This is load-bearing because \\hat{\\Theta}(t,s,x)=Y^{s,x}(t,s) depends on t through the first argument of Y, and the proof of Theorem 4.7 uses the uniform t-modulus through Theorem 3.2. The theorem as stated is therefore not established; either add (B.5) to the statement of Theorem 5.2 and Section 5, or prove the needed regularity under (B.4) alone.","section":"Section 5, Theorem 5.2 versus Theorem 4.7"},{"comment":"The passage from the Riemann-sum identity (5.6) to the PDE is the core of Theorem 5.2, but it is not proved. After asserting the Hölder bounds (5.7)--(5.10) via Kolmogorov's continuity theorem, the text says 'by dominated convergence theorem' and lets the mesh go to zero. This does not cover the two stochastic-integral sums in (5.6); no argument is given that they converge to zero in L^2 or in probability. One needs to prove, for example via Itô isometry and the bounds (5.9)--(5.10), that the sum of the stochastic integrals vanishes in the limit, and also to justify the corresponding convergence of the deterministic sums. Since the derivation of the PDE (5.1) depends entirely on this limit passage, a complete proof is required.","section":"Section 5, equations (5.6)--(5.10)"},{"comment":"The identification of Z^{t,x}(s,r) with the Malliavin derivative D_r Y^{t,x}(s,r) is asserted without proof. Corollary 4.2 ends with the 'Moreover' sentence claiming this identification, but the proof stops after establishing the equation (4.12); the identification is not derived. Similarly, equation (4.2) in Theorem 4.1 is stated as 'In addition' without a proof in the Picard-iteration argument. Lemma 4.4 and the subsequent uniform estimates for Z, in particular (5.9)--(5.10), depend on this identification. This is a nontrivial Malliavin-calculus step and should be proved explicitly.","section":"Corollary 4.2 and Lemma 4.4"},{"comment":"The proof of Theorem 4.7 does not establish the full C^{0,0,0,2} regularity that Theorem 5.2 uses. The argument shows moment bounds for first-order finite differences and states that this implies twice differentiability in x, but it does not prove existence and continuity of the second derivatives, and it does not derive the uniform bound |\\hat{\\Theta}_{xx}|≤C_p that is asserted in (5.5). To justify (5.5) and the dominated-convergence step in Theorem 5.2, one needs second-difference estimates and a Kolmogorov-type argument yielding a continuous version of the second derivative. As written, the regularity conclusion is under-supported.","section":"Theorem 4.7, equations (4.27)--(4.33)"}],"minor_comments":[{"comment":"The deduction of the pathwise bound |Y(t,s)-Y(t',s)|≤Cρ(|t-t'|) from the conditional estimate (3.21) uses that Y(t,s)-Y(t',s) is F_s-measurable; this measurability should be stated explicitly. Without that remark the step looks unjustified, although it is in fact correct.","section":"Theorem 3.2, after (3.21)"},{"comment":"The regularity statement after (5.3) says the solution belongs to C^{0,0,2}(∆[0,T]×R^d; R^d); the codomain should be R^m, matching the values of Y. This is a typo but should be corrected.","section":"Theorem 5.2, statement"},{"comment":"The condition (B.5) contains a typo: 'x,y,,y′' should read 'x,y,y′'. Please also state explicitly that the modulus of continuity is uniform with respect to the spatial variables, as the example in the major comment shows this uniformity is essential.","section":"Assumption (B.5)"},{"comment":"The expression Y^{s,x}(t,s) uses the extension of Y^{t,x}(s,r) to first arguments s that are smaller than the initial time; this extension is constructed in Proposition 4.6 and equation (4.18), but Theorem 5.2 should recall that convention to avoid ambiguity.","section":"Definition of \\hat{\\Theta}, equation (5.3)"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a natural and worthwhile generalization, and the gaps identified above appear to be fillable by adding the missing regularity assumption and supplying the omitted analytic and Malliavin-calculus arguments. I do not see a fatal flaw in the main idea, but the central theorem is not proved as stated. I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a genuine new object, the EBSVIE, and the intended result — a probabilistic representation for the non-local PDE class (5.1) that includes equilibrium HJB equations — is important within stochastic control. If the proof of Theorem 5.2 is made rigorous, it is a solid step beyond Wang–Yong and Pardoux–Peng.\n\nWhat is good: Theorem 3.1 (well-posedness) is a clean adaptation of Yong's contraction method, and Theorem 5.1, the Itô direction from PDE to EBSVIE, is correct and short. The Malliavin regularity program is ambitious and gives the right kind of statement. The writing is clear and the citations are fair; the new content is not in the cited papers.\n\nThe soft spots are in the converse, Theorem 5.2, and they are concrete.\n\nFirst, assumption mismatch. The theorem is stated under (F.1)-(B.4), but the proof begins by invoking Theorem 4.7, which is proved under (B.5). (B.5) adds a uniform-in-x modulus of continuity in t for g and ψ. (B.4) only gives continuity and C^3_b in the spatial variables; it does not imply (B.5) — take g(t,x)=cos(tx). So the theorem as stated is not established. This is the most serious issue.\n\nSecond, the limit passage from the Riemann sum (5.6) to the PDE. The text says 'by dominated convergence theorem' after asserting (5.7)-(5.10). The estimates are plausible, but the convergence of the two stochastic integral sums to zero is not demonstrated; dominated convergence does not handle Itô integrals directly. I expect it can be fixed with BDG and an L^2 estimate, but the proof is not there.\n\nThird, two smaller things: in Theorem 3.2 the deduction of a.s. joint continuity from L^p bounds is too quick, and Corollary 4.2 asserts the Malliavin-derivative-to-Z identification without proof. Both are likely repairable, but they contribute to a conditional verdict.\n\nThe author's claim that Theorem 5.2 gives a nonlinear Feynman-Kac formula for possibly degenerate non-local PDEs is a good and falsifiable claim. The tools are imported, but the target is new.\n\nWho should read: stochastic analysts working with BSVIEs, Malliavin calculus, or time-inconsistent control. They should read it with a red pen on Section 5. It deserves a serious referee — the result is worth the referee's time even though the current version should come back for revision.\n\nMy recommendation: send to peer review, but the referee should demand a corrected Theorem 5.2 (either add (B.5) to the assumptions or prove the regularity under (B.4)) and a written proof of the limiting step.","headline":"Genuinely new extension of the Feynman-Kac formula to non-local PDEs via EBSVIEs, but Theorem 5.2 is under-assumed and its key limiting argument is not written out.","tokens_in":30388,"tokens_out":11427,"would_cite":false,"duration_ms":106871,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H20","45D05","35K40","35K59"],"pacs":[],"model":"deepseek-v4-flash","headline":"The solution of an EBSVIE is the unique classical solution of a non-local quasilinear parabolic PDE system.","keywords":["extended backward stochastic Volterra integral equation","backward stochastic differential equations","non-local quasilinear parabolic equations","nonlinear Feynman-Kac formula","Malliavin calculus","probabilistic representation","time-inconsistent control"],"falsifier":"For a concrete smooth case with $m=d=1$, $b=0$, $\\sigma=1$, $g(t,s,x,y,y',z)=y'$, and $\\psi(t,x)=x^2$, simulate the EBSVIE and compute the residual of the PDE (5.1) at fixed $(t,s,x)$; the residual must tend to zero as the time partition is refined, otherwise the limit passage in Theorem 5.2 fails.","tokens_in":29324,"feed_emoji":"🎲","tokens_out":11413,"duration_ms":107924,"temperature":0.7,"pith_summary":"This paper introduces extended backward stochastic Volterra integral equations (EBSVIEs), in which the generator depends on the value of the unknown on the diagonal, $Y(r,r)$, as well as on the current value and the noise term. It proves that, under Lipschitz conditions, an EBSVIE has a unique adapted solution, and it uses Malliavin calculus to show higher-order regularity of that solution in the Markovian setting. The main result is a converse Feynman-Kac statement: the function $\\hat{\\Theta}(t,s,x)=Y^{s,x}(t,s)$ built from the EBSVIE solution is the unique classical solution of a non-local quasilinear parabolic PDE system. This matters because it gives a probabilistic representation for PDEs that couple different time layers, the kind that appears in time-inconsistent optimal control.","feed_headline":"Stochastic Volterra solutions solve non-local quasilinear PDEs","feed_subtitle":"A Feynman-Kac bridge now reaches PDEs whose generator couples two time layers.","key_machinery":"The engine is the Markovian EBSVIE (1.17), a family of backward stochastic differential equations indexed by $s\\in[t,T)$, whose generator at time $\\tau$ depends on the solution triple $(Y(s,\\tau), Y(\\tau,\\tau), Z(s,\\tau))$; the diagonal process $Y(\\tau,\\tau)$ is what makes the resulting PDE non-local. The argument proceeds through three linked pieces: a contraction-and-induction proof that the EBSVIE has a unique adapted solution in the space $H^p[0,T]$; Malliavin calculus showing that $Z^{s,x}(s,r)=\\nabla Y^{s,x}(s,r)(\\nabla X^{s,x}(r))^{-1}\\sigma(r,X^{s,x}(r))$, which gives regularity of the solution in the initial data; and, in Theorem 5.2, a Riemann-sum identity (5.6) that is passed to the limit by the Holder estimates and dominated convergence to obtain the PDE (5.1).","core_discovery":"The paper's central claim is a two-way Feynman-Kac correspondence. For the Markovian EBSVIE (1.17) under assumptions (F.1)--(B.4), the deterministic function $\\hat{\\Theta}(t,s,x)\\triangleq Y^{s,x}(t,s)$ is the unique classical solution of the non-local quasilinear parabolic system (5.1). Conversely, if $\\Theta$ is a classical solution of (5.1), then $Y^{s,x}(s,r)=\\Theta(s,r,X^{s,x}(r))$ and $Z^{s,x}(s,r)=\\Theta_x(s,r,X^{s,x}(r))\\sigma(r,X^{s,x}(r))$ form the adapted solution of the EBSVIE. This extends the nonlinear Feynman-Kac formula to equations whose nonlinear term involves both $\\Theta(t,s,x)$ and the diagonal value $\\Theta(s,s,x)$, and it allows the diffusion coefficient to be degenerate.","pith_inferences":["A natural next step would be to extend the same diagonal-coupling mechanism to path-dependent or mean-field EBSVIEs, since the non-locality enters only through $Y(r,r)$ and the Malliavin step does not use the special field structure.","The identity $Z^{s,x}(s,r)=\\nabla Y^{s,x}(s,r)\\sigma(r,X^{s,x}(r))$ suggests a Monte Carlo scheme that regresses finite-difference estimates of the gradient of $Y$ to approximate $Z$ and then evaluates the PDE residual; the paper does not describe such a scheme.","Because the non-local PDE (1.16) is the equilibrium Hamilton-Jacobi-Bellman equation of time-inconsistent control, the representation gives a probabilistic route to equilibrium value functions and strategies, an application the introduction leaves implicit."],"forward_implications":["Under (F.1)--(B.4), the function $\\hat{\\Theta}$ defined by (5.3) is the unique classical solution of the non-local quasilinear parabolic system (5.1).","Any classical solution of (5.1) yields the EBSVIE solution through (5.2), so the PDE and the stochastic equation determine each other.","The representation applies to possibly degenerate diffusion coefficients, since no uniform ellipticity condition such as (1.10) is imposed.","The regularity proven for the EBSVIE solution implies the PDE system has a classical solution, not merely a weak or viscosity solution."],"supporting_citations":[{"why":"It supplies the Markovian forward-backward SDE regularity and the nonlinear Feynman-Kac theorem that this paper extends.","marker":"[21]"},{"why":"It supplies the BSVIE representation theorem that Theorem 5.1 mirrors.","marker":"[30]"},{"why":"It supplies the well-posedness framework for BSVIEs, including the contraction argument that is adapted to EBSVIEs.","marker":"[36]"},{"why":"It supplies the Malliavin calculus used to differentiate the EBSVIE solution and to identify Z with a derivative of Y.","marker":"[18]"},{"why":"It provides the standard BSDE estimates and SDE regularity facts used throughout the estimates.","marker":"[38]"},{"why":"It supplies the Kolmogorov continuity criterion behind the Holder estimates in the limit passage.","marker":"[9]"}],"fun_headline_variants":["EBSVIEs connect non-local quasilinear PDE systems","Feynman-Kac bridge links Volterra to parabolic PDEs","Two-time-layer Feynman-Kac for degenerate diffusion","Volterra solutions satisfy non-local parabolic PDEs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The transition from the Riemann-sum identity (5.6) to the PDE requires uniform Holder estimates (5.7)--(5.10) and the convergence to zero of the stochastic integral terms, and these are asserted in the text rather than proved.","fun_headline_variants_meta":{"raw":{"variants":["EBSVIEs connect non-local quasilinear PDE systems","Feynman-Kac bridge links Volterra to parabolic PDEs","Two-time-layer Feynman-Kac for degenerate diffusion","Volterra solutions satisfy non-local parabolic PDEs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000812,"raw_usage":{"total_tokens":3525,"prompt_tokens":875,"completion_tokens":2650,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":491,"completion_tokens_details":{"reasoning_tokens":2583}},"tokens_in":491,"tokens_out":2650,"duration_ms":21114,"temperature":1.0,"reasoning_tokens":2583,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:24:35.279836+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a concrete smooth case with $m=d=1$, $b=0$, $\\sigma=1$, $g(t,s,x,y,y',z)=y'$, and $\\psi(t,x)=x^2$, simulate the EBSVIE and compute the residual of the PDE (5.1) at fixed $(t,s,x)$; the residual must tend to zero as the time partition is refined, otherwise the limit passage in Theorem 5.2 fails.","supporting_citations":[{"cited_title":"Pardouc and S","cited_arxiv_id":null,"evidence_quote":"It supplies the Markovian forward-backward SDE regularity and the nonlinear Feynman-Kac theorem that this paper extends."},{"cited_title":"Wang and J","cited_arxiv_id":null,"evidence_quote":"It supplies the BSVIE representation theorem that Theorem 5.1 mirrors."},{"cited_title":"Yong, Well-posedness and regularity of backward stochast ic Volterra integral equations, Probab","cited_arxiv_id":null,"evidence_quote":"It supplies the well-posedness framework for BSVIEs, including the contraction argument that is adapted to EBSVIEs."},{"cited_title":"Nualart, The Malliavin calculus and related topics, Springer, Heidelberg, 1995","cited_arxiv_id":null,"evidence_quote":"It supplies the Malliavin calculus used to differentiate the EBSVIE solution and to identify Z with a derivative of Y."},{"cited_title":"Yong and X","cited_arxiv_id":null,"evidence_quote":"It provides the standard BSDE estimates and SDE regularity facts used throughout the estimates."},{"cited_title":"Friz and M","cited_arxiv_id":null,"evidence_quote":"It supplies the Kolmogorov continuity criterion behind the Holder estimates in the limit passage."}],"review_version":1}