{"id":"b8ec270e-2627-44b2-a3d9-48bee4a01768","arxiv_id":"2502.02754","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For spider diffusions with a local-time dependent spinning measure, the paper proves Itô calculus, no atom at the junction, Feynman-Kac representations, local-time approximations, the strong Markov property, and vertex scattering limits.","lead":"This paper proves several trajectory properties for a Walsh spider diffusion whose choice of outgoing ray at the central vertex depends on how much local time the process has spent there. It establishes an Itô formula, non-atomicity at the junction, a Feynman-Kac representation, local-time approximations, the strong Markov property, and the instantaneous scattering distribution.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.1's proof mishandles the initial excursion when x⋆>0: the n=0 term in (10) is not controlled by the non-stickiness bound, so the ε→0 limit in (15) is unjustified as written.","rationale":"The paper aims to establish a full Itô calculus for a Walsh spider process whose spinning measure depends on its own local time. The central claim, Theorem 3.1, is plausible and the surrounding applications are coherent, but the proof of that theorem contains a concrete internal gap: the excursion decomposition used to compute quadratic variation does not control the initial stretch before the first hitting of level ε when the process starts away from 0. The n = 0 term in decomposition (10) has quadratic variation supported on [0, θ^ε_0), where x > ε, and the non-stickiness estimate controls only the occupation time near 0. Thus the ε → 0 limit in (15) is not justified as written. This is an internal consistency issue in the present manuscript rather than a disagreement with the field's consensus, and it is independent of whether the companion paper's non-stickiness estimate is correct. The reader's weakest-assumption identification of the non-stickiness estimate is reasonable, but my main concern is the additional proof gap that remains even assuming that estimate. Because the gap is local and repairable, I do not see grounds for rejecting the paper; the appropriate disposition is conditional acceptance with a request to repair the proof of Theorem 3.1 and to clarify the use of [9] in Section 4.2. The paper provides detailed computations and a coherent set of consequences, so the central claim is not obviously false, only incompletely proved at a load-bearing step.","tokens_in":28007,"tokens_out":15728,"duration_ms":156429,"concrete_test":"Specialize the Step 1 computation of Theorem 3.1 to a Walsh Brownian motion on two edges, with σ ≡ 1, b ≡ 0, constant α, and starting point x⋆ > 0. For this process representation (2) holds with W a standard Brownian motion and l the local time at 0. Compute explicitly ⟨M^{1,0,ε}(f)⟩_t = ∫_0^{t∧θ^ε_0} (f')^2(x(s)) ds and compare it with the claimed bound (14). Since x(s) > ε on [0, θ^ε_0), the integral ∫_0^t 1_{x(u)≤ε} du is zero on that interval, so (14) fails for this term. Then verify that adding the missing initial-segment contribution to (15) yields exactly the standard Itô formula, confirming the theorem remains true but the written proof requires a localization argument.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Step 1 of the proof of Theorem 3.1, the excursion decomposition (10) sets τ^ε_0 = 0 and θ^ε_0 = inf{s ≥ 0 : x(s) = ε}. If the starting point satisfies x⋆ > ε, then on [0, θ^ε_0) the process stays strictly above ε and the local time is constant. The term M^{1,0,ε}(f) therefore has quadratic variation ∫_0^{t∧θ^ε_0} σ_{i(s)}^2(s,x(s),l(s)) (∂_x f_{i(s)}(x(s)))^2 ds, which is nonzero and does not vanish as ε → 0. The displayed estimate (14), bounding the whole sum ∑_n M^{1,n,ε}(f) by C ∫_0^t 1_{x(u)≤ε} du, is consequently false for n = 0 whenever x⋆ > ε. The subsequent comparison, which subtracts the M^{2,n,ε} intervals and bounds the remainder by 2C ∫_0^t 1_{x(u)≤ε} du, is therefore not justified; the initial segment contributes a non-negligible term that is not killed by the non-stickiness estimate. This is an internal gap in the proof of the paper's central Itô formula, independent of whether the imported non-stickiness estimate from [17, Proposition 5.2] is valid. The gap is local and appears repairable, for instance by first stopping at the first hitting time of 0 and applying the excursion argument only after that time, but as written the passage from (14) to (15) is incomplete for starting points away from 0.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Walsh spider diffusion constructed in the companion paper [17], whose spinning measure depends on time and on the local time at the junction vertex. The main results are: an Itô formula driven by the same Brownian motion as the underlying diffusion (Theorem 3.1); non-atomicity of the law of x(t) at zero and existence of a density (Section 4); a Feynman-Kac representation for linear parabolic systems with a local-time Kirchhoff boundary condition (Theorem 5.2); two local-time approximations, a downcrossing representation and a mean-value approximation (Section 6); the strong Markov property (Section 7); and a characterization of the instantaneous scattering distribution at the vertex as the spinning coefficients α_i(t,ℓ) (Proposition 8.1). The proofs systematically rely on the non-stickiness estimate imported from [17, Proposition 5.2] and on the well-posedness of the spider martingale problem from [17].","tokens_in":28302,"tokens_out":10925,"duration_ms":110691,"significance":"If the results are correct, this is a valuable extension of Walsh spider calculus to the case where the spinning measure is selected by the process's own local time. The paper is well organized as a sequence of independent problems, and the statements are precise about which results are imported from [17] and [18]. The Itô formula, if established, constitutes the foundation for the subsequent trajectory, approximation, Markov, and scattering results; the paper also makes original use of Fournier-Printems density techniques and extends Lévy's classical local-time approximations to a genuinely non-Markovian setting. The proofs are detailed but not machine-checked, and the central derivation contains a repairable gap in the excursion decomposition; the absolute-continuity argument also contains an erroneous estimate.","major_comments":[{"comment":"The term M^{1,0,ε}_t is not controlled by the displayed estimate. Since τ^ε_0 = 0 and θ^ε_0 = inf{s≥0 : x(s)=ε}, if x⋆ > ε, the initial interval [0, θ^ε_0) lies entirely above level ε, so its quadratic variation ∫_0^{t∧θ^ε_0} σ^2_{i(s)}(∂_x f_{i(s)})^2 ds does not vanish as ε→0. Therefore inequality (14), bounding ⟨Σ_n M^{1,n,ε}(f)⟩_t by C∫_0^t 1_{x(u)≤ε} du, is false for the n=0 term, and the passage to (15) is not justified for starting points away from 0. This is an internal gap in the proof of the paper's central Itô formula, independent of the validity of the non-stickiness estimate. The gap appears repairable, for example by first stopping at the first hitting time of 0 and applying the excursion decomposition only after that time; as written, the proof is incomplete.","section":""},{"comment":"The bound E[(Y(1)-Z_ε)^2] ≤ Cε^2 is incorrect. With Z_ε = y_0 - W(1-ε), the Brownian increment W(1)-W(1-ε) contributes ε, not ε^2, so the first term on the right-hand side should be of order ε and the final bound is of order ε, not ε^2. Moreover, the remainder ∫_{1-ε}^1 h_{i(u)}(u,x(u),l(u)) du is not independent of Z_ε, and the integrand contains the chaotic edge process i(u); hence the hypotheses of Fournier-Printems [9, Theorem 3.1] are not verified as stated. The same issue affects Section 4.3, where the density claim for V(t) is asserted by appealing to the method of Fournier-Printems without checking the relevant conditions. Consequently, the absolute-continuity conclusions of Section 4 are not established by the arguments given.","section":""},{"comment":"The reasoning that the exponential martingale property of E_t(f) implies ∫_0^· ⟨Θ_s(f), d(m_s, M_s(f))⟩ ≡ 0 is not justified as written. What actually yields the conclusion is that the continuous local martingale N_t = ∫_0^t ⟨Θ_s(f), d(m_s, M_s(f))⟩ has zero quadratic variation, hence is constant; the exponential-martingale property alone does not imply the vanishing of N. This step is repairable by invoking the standard zero-quadratic-variation fact, but the proof needs to be rewritten.","section":""}],"minor_comments":[{"comment":"There are several typos and notational inconsistencies: 'Feynmann-Kac' should be 'Feynman-Kac'; 'Skohokhod' should be 'Skorokhod'; 'where introduced' should be 'were introduced'; and the notation C^{1,2,1}_b(J_T × [0+∞)]) contains an extra bracket.","section":""},{"comment":"In the display after defining m_t, the equality ⟨m⟩_t = ∫_0^t σ^2_{i(s)}(s,x(s),l(s)) dW(s) = ⟨x⟩_t should have ds in place of dW(s).","section":""},{"comment":"The sum '∑_{j=0} α_j(s,l(s))∂_x u_i(s,0,l(s))' should run from j=1 to I, not from j=0.","section":""},{"comment":"The approximation (23) uses F(θ^ε_{n+1}) - F(τ^ε_n), while the stopping times are defined with θ^ε_n; the proof clarifies the intended intervals, but the statement should align the notation to avoid ambiguity.","section":""},{"comment":"The phrase 'this ladder case is excluded' should read 'this latter case is excluded'.","section":""},{"comment":"The symbol ℓ(t) is introduced for the integrated boundary term ∫_0^t Σ_j α_j(u,l(u))/σ_j(u,0+,l(u)) dl(u), which is easy to confuse with the local time l(t); a different notation would improve readability.","section":""}],"recommendation":"major_revision","confidential_remarks":"The paper is not self-contained: it imports existence, uniqueness, non-stickiness, and the extended martingale property from the companion paper [17] and PDE well-posedness from [18]. This is acceptable because those results are accepted or published and are distinct from the target results, but the reader should have [17] available. The main theorem's proof contains a repairable but real gap in the initial excursion term, and Section 4.2 contains an erroneous estimate; these must be fixed before the paper can be accepted. Once repaired, the paper would make a solid contribution to the local-time-dependent spider literature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is the natural follow-up to the authors' construction of Walsh spider diffusions with a spinning measure that depends on the process's own local time. The results are new: constant spinning measure cases were known (Freidlin–Sheu, Barlow–Pitman–Yor), but the local-time-dependent case is not covered by those. The paper gives an Itô formula, absolute continuity of the one-dimensional marginals, a Feynman–Kac representation, local-time approximations, the strong Markov property, and a scattering limit. That is a coherent and useful package for this class of processes, and the paper is honest about what is imported from the companion paper [17] and the PDE paper [18]. The scattering proof and the downcrossing representation are clean and the non-stickiness estimate is correctly identified as the load-bearing imported input.\n\nThe stress-test note is right and lands on the central proof. In Theorem 3.1, Step 1, the excursion decomposition sets τ^ε_0 = 0 and θ^ε_0 as the first hitting time of ε. If the starting point x* > ε, the n = 0 segment [0, θ^ε_0) stays above ε, so the bound (14) — the whole sum of M^{1,n,ε} controlled by C∫_0^t 1_{x(u)≤ε} du — is simply false for that term. The initial segment contributes a nonzero quadratic variation that does not vanish as ε→0. The proof as written therefore does not establish the Itô formula for starting points away from zero. This is a local gap and it looks repairable, for example by applying the excursion argument only after the first hitting time of 0, or by treating the first excursion separately. But it is a genuine flaw in the paper's main theorem, not a cosmetic issue.\n\nTwo smaller soft spots. The strong Markov section restarts the process at \"time t\" after a stopping time τ; it should restart at τ, and the time-inhomogeneous mechanism is left ambiguous enough that the proof of Lemma 7.1 does not quite convince. The application of Fournier–Printems in Section 4.2 is asserted rather than verified: the drift contains the chaotic index i(u), and the paper says the good integrability properties suffice without checking the exact hypotheses. The reader's note that there is an order-ε versus ε² error in the absolute continuity estimate did not survive reading — the displayed computation gives Cε² — but the unverified application of [9] remains a legitimate concern. The Feynman–Kac section leans on the authors' own PDE well-posedness result from [18]; that is acceptable since [18] is published, but the representation is conditional on that regularity framework.\n\nWho is this for? Specialists in Walsh diffusions, Markov processes on graphs, and stochastic control with boundary conditions. The paper deserves a serious referee: the objects are new, the theorems are plausible, and the gaps are identifiable and likely repairable. I would recommend engaging with it, but the referee should insist on a repaired Itô formula proof and a clarified strong Markov statement before acceptance.","headline":"A plausible and genuinely new companion paper on spider diffusions with local-time-dependent spinning; the central Itô formula proof has a repairable but real gap in the initial excursion term.","tokens_in":28889,"tokens_out":4657,"would_cite":true,"duration_ms":44200,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J60","60J55","60H05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves Itô's formula for a Walsh spider diffusion whose spinning measure is a function of its own local time at the junction.","keywords":["Walsh spider diffusion","local time","spinning measure","Itô formula","strong Markov property","Feynman-Kac representation","star-shaped network","non-stickiness"],"falsifier":"Simulate the companion-paper construction on two rays with α_1(l)=1/(1+l), α_2(l)=l/(1+l), record the branch chosen at first exit from a radius δ around the vertex, and compare the empirical frequency to α_1(t,ℓ) at the local-time level of that exit; Proposition 8.1 predicts agreement as δ→0. Alternatively, evaluate both sides of the Itô formula (7) for a smooth test function in the simulation and check that any systematic discrepancy vanishes as the excursion cutoff ε→0.","tokens_in":27745,"feed_emoji":"🕸️","tokens_out":6585,"duration_ms":60923,"temperature":0.7,"pith_summary":"This paper establishes Itô's formula for the Walsh spider diffusion constructed in the companion paper, where the probabilities of choosing a ray at the junction are not fixed but depend on the process's own local time at the junction. The main result, Theorem 3.1, writes a smooth test function of the process and its local time as a stochastic integral against the same Brownian motion that drives the radial coordinate, plus drift terms and boundary local-time terms. From this formula the authors derive a density for the marginal law with no atom at the junction, two L1 approximations of the local time, the strong Markov property, and a characterization of the instantaneous scattering distribution along each ray. The paper also gives a Feynman-Kac representation for linear parabolic systems with a local-time Kirchhoff boundary condition. If the results are correct, the process of the companion paper is not merely a solution of a martingale problem but a full-fledged Markov diffusion with a complete calculus.","feed_headline":"Spider diffusions that pick rays by local time get an Itô formula","feed_subtitle":"The ray chosen at the junction depends on time spent there, and the paper proves the stochastic calculus for it.","key_machinery":"The paper's central object is the martingale problem (Spi-Mar) for the spider process (x,i,l) on the star-shaped network J, where the spinning coefficients α_i(t,l) and the diffusion coefficients depend on the local time l at the vertex. The load-bearing mechanism is the excursion decomposition of the radial coordinate x around the vertex, using the hitting times θε_n of the level ε and τε_n of the vertex 0, together with the companion paper's non-stickiness bound E∫_0^T 1_{x(u)≤ε}du ≤ Cε. That bound makes the accumulated quadratic variation of excursion endpoints vanish as ε→0, so that the Itô formula's boundary term comes only from local time and the α-weighted spatial derivatives at 0; the same ε-removal turns occupation-time formulas into the two L1 approximations of the local time.","core_discovery":"For every bounded test function f in $C^{{1,2,1}}$_b(J_T × R_+), the Itô formula (7) holds almost surely, writing f_{i(t)}(t,x(t),l(t)) - f_{i*}(0,x_*,0) as the sum of a stochastic integral against the Brownian motion W appearing in the martingale problem, the usual time and spatial drift terms, and the boundary contribution ∂_l f(u,0,l(u)) dl(u) plus the weighted sum of spatial derivatives at the vertex. The proof decomposes the path into excursions away from the vertex, computes the quadratic variation piecewise on each excursion, and then sends the excursion radius ε to zero using the non-stickiness estimate of the companion paper. A subsequent exponential-martingale argument identifies the stochastic integral, and the formula is extended from product-form test functions to the full class. A central corollary is that for each fixed t>0 the law of x(t) is absolutely continuous with respect to Lebesgue measure and has no atom at the vertex.","pith_inferences":["Editorial inference: the Itô formula together with the strong Markov property suggests that the joint process (x,i,l) is a genuine Markov process on the augmented state space J × R_+, although the paper does not exhibit its infinitesimal generator on the whole space.","Editorial inference: one could test the scattering law empirically by counting first-exit rays from a small ball in a simulation of the companion-paper construction; the limiting frequencies should equal α_i evaluated at the current local-time level, and any systematic disagreement would localize a failure of the non-stickiness estimate.","Editorial inference: the Feynman-Kac representation is proved for linear parabolic systems; the same probabilistic representation should transfer to nonlinear Hamilton-Jacobi-Bellman settings with local-time Kirchhoff boundary transmission, using the comparison principle already available for such systems, although this paper does not make that step.","Editorial inference: the absolute-continuity proof only addresses the marginal law of the radial coordinate x(t); a parallel argument using the squared process suggests the pair (x(t),l(t)) may also have a density away from the vertex, but that is not claimed here."],"forward_implications":["Every test function in C^{1,2,1}_b admits the same Itô representation, so the process of the companion paper can be used in stochastic calculus arguments that need both a driving Brownian motion and a local-time boundary term.","For each fixed t>0 the law of x(t) is absolutely continuous with respect to Lebesgue measure, so the process has no atom at the junction at a fixed time.","The local time at the vertex is approximated in L1 both by ε times the number of downcrossings from ε to 0 and by the normalized occupation time near 0, extending the classical one-dimensional formulas to the local-time-dependent spider.","Solutions of the backward linear parabolic system with local-time Kirchhoff boundary condition admit the explicit probabilistic representation given in Theorem 5.2.","Upon reaching the vertex with local time ℓ at time t, the probability of scattering to ray i tends to α_i(t,ℓ) as the observation radius shrinks to zero."],"supporting_citations":[{"why":"Provides existence and uniqueness for the spider martingale problem and the non-stickiness estimate E∫1_{x≤ε}du ≤ Cε on which every ε-limit in this paper relies.","marker":"[17]"},{"why":"Supplies the classical Itô formula and the scattering law for constant spinning measure, the template this paper extends.","marker":"[7]"},{"why":"Establishes existence of the spider process with constant spinning measure, used as the starting building block for the construction in [17].","marker":"[8]"},{"why":"Proves well-posedness of the parabolic systems with local-time Kirchhoff condition whose solution the Feynman-Kac formula represents.","marker":"[18]"},{"why":"Gives the absolute-continuity argument adapted in Section 4 to rule out an atom at zero.","marker":"[9]"},{"why":"Provides the classical downcrossing representation of local time that Proposition 6.1 extends.","marker":"[13]"},{"why":"Supplies the conditional-distribution and martingale-problem machinery used for the strong Markov property.","marker":"[20]"},{"why":"Supplies the standard extension argument from product-form test functions to the full class in the final step of the Itô formula proof.","marker":"[21]"}],"fun_headline_variants":["Spider walks pick rays by local time, Ito formula proven","Local-time-chosen spider rays get Ito calculus","When a spider's ray is its local time: Ito calculus","Ito formula for spider diffusions with local-time-selected spin"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the companion paper's non-stickiness estimate E∫_0^T 1_{x(u)≤ε}du ≤ Cε, which lets every ε-limit in this paper be sent to zero; without it, excursions near the junction need not be negligible and the Itô formula, density, approximations, and scattering limit would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Spider walks pick rays by local time, Ito formula proven","Local-time-chosen spider rays get Ito calculus","When a spider's ray is its local time: Ito calculus","Ito formula for spider diffusions with local-time-selected spin"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000892,"raw_usage":{"total_tokens":3825,"prompt_tokens":900,"completion_tokens":2925,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":516,"completion_tokens_details":{"reasoning_tokens":2856}},"tokens_in":516,"tokens_out":2925,"duration_ms":18800,"temperature":1.0,"reasoning_tokens":2856,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T11:15:07.525399+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the companion-paper construction on two rays with α_1(l)=1/(1+l), α_2(l)=l/(1+l), record the branch chosen at first exit from a radius δ around the vertex, and compare the empirical frequency to α_1(t,ℓ) at the local-time level of that exit; Proposition 8.1 predicts agreement as δ→0. Alternatively, evaluate both sides of the Itô formula (7) for a smooth test function in the simulation and check that any systematic discrepancy vanishes as the excursion cutoff ε→0.","supporting_citations":[{"cited_title":"Martingale problem for a Walsh spider process with spinning measure se- lected from its own local-time","cited_arxiv_id":null,"evidence_quote":"Provides existence and uniqueness for the spider martingale problem and the non-stickiness estimate E∫1_{x≤ε}du ≤ Cε on which every ε-limit in this paper relies."},{"cited_title":"Diffusion processes on graphs: stochastic differential equations, large devia- tion principle","cited_arxiv_id":null,"evidence_quote":"Supplies the classical Itô formula and the scattering law for constant spinning measure, the template this paper extends."},{"cited_title":"Diffusion processes on gra phs and the averaging principle, Ann","cited_arxiv_id":null,"evidence_quote":"Establishes existence of the spider process with constant spinning measure, used as the starting building block for the construction in [17]."},{"cited_title":"Well posedness of linear Parab olic partial differential equation posed on a star-shaped network with local-time Kirchhoff’s boundary c ondition at the vertex","cited_arxiv_id":null,"evidence_quote":"Proves well-posedness of the parabolic systems with local-time Kirchhoff condition whose solution the Feynman-Kac formula represents."},{"cited_title":"Absolute continuity for som e one-dimensional processes","cited_arxiv_id":null,"evidence_quote":"Gives the absolute-continuity argument adapted in Section 4 to rule out an atom at zero."},{"cited_title":"Brownian Motion and Stocha stic Calculus, volume 113 of Graduate Texts in Mathematics","cited_arxiv_id":null,"evidence_quote":"Provides the classical downcrossing representation of local time that Proposition 6.1 extends."},{"cited_title":"Multidimensional Diffus ion Processes, 2004","cited_arxiv_id":null,"evidence_quote":"Supplies the conditional-distribution and martingale-problem machinery used for the strong Markov property."},{"cited_title":"Continuous martingales and Brownia n motion","cited_arxiv_id":null,"evidence_quote":"Supplies the standard extension argument from product-form test functions to the full class in the final step of the Itô formula proof."}],"review_version":1}