{"id":"fdd009d4-7495-4e6e-bdba-1d2041c1fcd9","arxiv_id":"1908.01385","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Brownian motion conditioned to a thin tubular neighbourhood of a submanifold of a Riemannian manifold converges in finite-dimensional distributions to Brownian motion on the submanifold.","lead":"This paper proves that Brownian motion forced to stay inside a shrinking tube around a surface embedded in a curved space converges, at each finite collection of times, to Brownian motion living on that surface. The proof works for general Riemannian manifolds and shows the associated heat semigroups converge in every Sobolev norm.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Corollary 1 proves only one-dimensional marginals; the claimed finite-dimensional convergence rests on a false Markovian implication and is not established.","rationale":"Section 1.2.d contains an explicit assertion that convergence of one-dimensional marginals implies finite-dimensional distribution convergence for Markov processes. This is not a standard theorem and is false in general: one can construct Markov chains with the same one-dimensional marginals from a fixed initial state but different two-dimensional joint laws. The proof of Corollary 1 only establishes the one-dimensional marginal limit for initial points in L, and it does not address joint laws at multiple times, where the process after the first transition is no longer in L. The advertised conclusion in the abstract is therefore not supported by the provided argument. The gap is probably repairable: formula (7), together with the strong semigroup convergence in Theorem 1, should yield joint marginals by an induction over products of semigroups, provided denominators are uniformly controlled and intermediate vectors are strongly continuous. But this proof is absent. The reader's Lopatinskij concern in Section 4.1 is also real and would affect the Sobolev-space convergence, but the finite-dimensional distribution gap directly concerns the paper's stated process-level conclusion. The verdict remains CONDITIONAL: the main semigroup theorem may be sound, but both the Lopatinskij verification and the finite-dimensional marginal derivation need to be supplied.","tokens_in":27990,"tokens_out":14169,"duration_ms":163244,"concrete_test":"Write out the two-dimensional joint marginal using equation (7) with times s<t<T: E_x[f(x^ε_s)g(x^ε_t)] equals a ratio of products of semigroups, transformed by Σ_ε. Attempt to identify the limit via Theorem 1 as the corresponding Brownian joint expectation. Specifically verify that the intermediate vector w_ε = σ_ε^{-1}(g) · e^{-(T-t)H^0_ε/2} σ_ε^{-1}(√ρ) is a strongly continuous family in L^2(L(1), μ_Sa), and that the denominator e^{-T H^0_ε/2} σ_ε^{-1}(√ρ) is uniformly bounded away from zero near L for small ε. If the induction succeeds, replace the false Markovian assertion with this proof; if it cannot be carried out, Corollary 1 supports only one-dimensional marginal convergence, not finite-dimensional convergence.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In Section 1.2.d, the paper asserts that for Markov processes, convergence of the one-dimensional marginals implies convergence in finite-dimensional distributions. This assertion is false: a Markov process is not determined by its one-dimensional marginals from a fixed starting point, and different transition kernels can produce identical one-dimensional marginals. Corollary 1 proves exactly the one-dimensional marginal statement, lim_{ε→0} E_x[f(x^ε_t)] = E_x[f|L(x^0_t)] for x∈L, and the proof stops there. It does not control joint laws such as E_x[f(x^ε_s)g(x^ε_t)] for 0<s<t. After the first transition, x^ε_s is a point in the tube, not in L, so the convergence in Corollary 1, which is only stated for starting points in L, cannot be iterated without additional transition-kernel convergence and uniform positivity of the denominator in equation (7). The abstract's conclusion that the conditional Brownian motion 'converges in finite dimensional distributions' is therefore not supported by the text. The gap is likely fixable by applying Theorem 1 inductively to the product-of-semigroups expression obtained from (7), but that argument is not supplied.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies Brownian motion on a complete Riemannian manifold M conditioned to stay inside a small tubular neighbourhood L(ε) of a closed connected submanifold L. The authors identify the relevant Dirichlet generators H_ε on L(ε), rescale them unitarily to the fixed tube L(1), and subtract the ground-state energy λ_0 ε^{-2} of the vertical Laplacian. Their main result, Theorem 1, states that the renormalized semigroups e^{-t H_ε^0/2} converge strongly in H^{2n}(L(1)) for every n to E_0 e^{-t Δ_L/2} E_0, where E_0 projects onto the ground state of the vertical Dirichlet Laplacian on the fibres and Δ_L is the Laplace-Beltrami operator on L. The proof combines a perturbation expansion of the induced metric in terms of the Sasaki metric (Proposition 6), a Kato-type inequality for the perturbed quadratic forms (Corollary 2), Γ-convergence of the associated functionals (Proposition 3), resolvent convergence (Corollary 7), and elliptic a priori estimates for a family of ε-dependent boundary value problems (Section 4). On the probabilistic side, Corollary 1 interprets the L^2 convergence as convergence of one-dimensional marginals of conditional Brownian motion starting on L, and the abstract claims convergence in finite-dimensional distributions.","tokens_in":28223,"tokens_out":16951,"duration_ms":159353,"significance":"If the main semigroup convergence holds, it is a valuable contribution: it provides a direct spectral/geometric mechanism for the collapse of Dirichlet heat kernels on tubes to the heat kernel on the submanifold, with an explicit projection onto the vertical ground state. The analytic machinery—perturbation expansion, Kato-type inequality, Γ-convergence, and the uniform Sobolev estimates—is coherent and appears to support the L^2 and Sobolev convergence statements. The paper also has the merit of making the geometric structure (Sasaki metric, vertical and horizontal Laplacians) and the exact renormalization explicit, rather than relying on fitted parameters. The advertised probabilistic conclusion, however, is not established: the proof of Corollary 1 covers only one-dimensional marginals from starting points on L, and the claimed finite-dimensional convergence rests on a false general implication for Markov processes. This is a significant issue for the paper's stated scope, though the analytic core remains of interest.","major_comments":[{"comment":"The assertion that for Markov processes convergence of the one-dimensional marginals implies convergence in finite-dimensional distributions is false. Corollary 1 proves exactly the one-dimensional marginal statement lim_{ε→0} E_x[f(x^ε_t)] = E_x[f|_L(x^0_t)] for x∈L, and the proof stops there. It does not control joint laws such as E_x[f(x^ε_s)g(x^ε_t)] for 0<s<t. After the first transition, x^ε_s is a point in the tube L(ε), not in L, so Corollary 1—which is only stated for starting points in L—cannot be iterated. Thus the abstract's conclusion that the conditional Brownian motion 'converges in finite dimensional distributions' is not supported by the text. Please either supply an inductive argument based on equation (7) with a proof of transition-kernel convergence for arbitrary starting points, or restate the conclusion as convergence of one-dimensional marginals only.","section":"1.2.d, Corollary 1"},{"comment":"The boundary problem (25) is asserted to satisfy the Shapiro-Lopatinskij conditions uniformly in small ε, but no verification is provided. The uniform elliptic a priori estimate (26) in Proposition 9 is the key step leading to Proposition 10, Corollary 10, and the H^{2n} convergence in Theorem 1. Please give the principal-symbol computation for the boundary operators Δ_{Sa}^j u|∂ − ε^3 S_j(ε)u|∂ and a uniformity argument (for example, by perturbation from the standard Dirichlet boundary value problem for Δ_{Sa}^n) showing that the constant in (26) can be chosen independent of ε.","section":"4.1, after (25)"},{"comment":"In the proof of Proposition 12, the estimate of Proposition 10 is applied to v := Δ_{Sa}^{n-1}u − ε^3 S_{n-1}(ε)u, which is only shown to lie in H^1_0 ∩ H^2(L(1), μ_{Sa}). Proposition 10 is stated for u ∈ C^∞(H^0_ε), and it is not shown that v belongs to this class, nor that the inequality of Proposition 10 extends to all functions in H^2 ∩ H^1_0. Please state the precise domain on which Proposition 10 is valid and justify the application, or modify the argument (for example, by proving the n=1 estimate directly for the Dirichlet Laplacian on H^2 ∩ H^1_0).","section":"4.2.2, Proposition 12"}],"minor_comments":[{"comment":"The cut-off function φ is introduced with φ|L(1)=1 and φ|M\\L(r)=0, but the Feynman-Kac formula in 1.2.c uses the potential U without the cut-off. Please clarify that, on the events Ω_{s,t}^ε with ε≤1, the cut-off equals 1 on L(ε), so the two formulations agree.","section":"1.2.a"},{"comment":"In the displayed inequality in part c, the expression [1 − ε k_l/2]φ_{Sa,ε_n,α,w}(f) − (ε k_l/2)[α‖f‖² − ⟨w,f⟩ + ‖f‖²_{H^1_0}] does not follow algebraically from the preceding line; a term involving −(1/2)q^0_{Sa,ε_n}(f) appears to be missing. The subsequent lim sup bound is unaffected, but the line should be corrected.","section":"2.4, proof of Proposition 3, part c"},{"comment":"The cancellation of the powers of ε in the change of variables from Σ_ε to σ_ε^{*−1} is not shown explicitly. Writing out this cancellation (including the factor (ε^{m−l})^{±1/2}) would make the proof of Corollary 1 easier to follow.","section":"Corollary 1"},{"comment":"The abstract contains a typo ('neig hbourhood'), and there are several other small typos throughout the text. A careful proofreading pass is recommended.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The analytic core of the paper—the semigroup convergence in L^2 and in Sobolev spaces—appears sound and is a genuine contribution. The advertised probabilistic consequence (convergence in finite-dimensional distributions) is, however, not proven: the argument in Section 1.2.d rests on a false implication about Markov processes, and the proof of Corollary 1 only yields one-dimensional marginals. The regularity section also has a gap in the application of Proposition 10. These issues are local and fixable, but they affect the central claims of the paper as stated. If the authors can supply a correct FDD argument (or explicitly limit the claim to marginal convergence) and close the regularity gap, the paper would be suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe main theorem is the real thing: the semigroup convergence in L2 and in Sobolev spaces of arbitrary order for the renormalized Dirichlet operators on tubular neighbourhoods of a submanifold of a Riemannian manifold. The proof, via the Sasaki metric, canonical variation, Gamma-convergence of the forms, and elliptic a priori estimates, is coherent and plausible. The perturbation expansion (Prop. 6), the Kato-type estimate (Cor. 2), the epi-convergence (Prop. 3), and the resolvent convergence (Cor. 7) line up correctly. The result genuinely extends the Euclidean surface limit of Sidorova, Smolyanov, v. Weizsaecker and Wittich to general embeddings and upgrades the convergence to Sobolev spaces.\n\nBut the paper overreaches in its secondary claim. Section 1.2.d asserts that for Markov processes, convergence of the one-dimensional marginals implies convergence in finite-dimensional distributions. That is false. Corollary 1 proves exactly the one-dimensional marginal statement, and the proof stops there; it does not control joint laws like E_x[f(x^eps_s) g(x^eps_t)]. The abstract's 'converges in finite dimensional distributions' is therefore not supported. This is a genuine gap, though likely fixable by applying Theorem 1 inductively to the product-of-semigroups expression from (7), with uniform positivity of the denominator. But that argument is not supplied.\n\nThe second soft spot is the asserted Shapiro-Lopatinskij condition for the boundary value problem (25). Proposition 9 and the higher-order Sobolev convergence depend on uniform elliptic estimates; the verification that the boundary operators S_i(eps) satisfy the condition uniformly in small eps is not provided. This is a technical gap that a referee should inspect, but nothing in the text suggests it is fatal.\n\nThe citation pattern is fine: [10], [11], [13] are prior and planned work by the same group, but the cited results are real. The paper is honest that tightness is deferred to a later paper; that is not the problem.\n\nVerdict: the semigroup theorem deserves a serious referee. The paper should be published once the finite-dimensional claim is either proved or removed from the abstract, and the Lopatinskij point is addressed. I would send it to review.","headline":"The semigroup convergence theorem is a genuine extension and looks correct, but the abstract's finite-dimensional distribution claim is unsupported and the Lopatinskij verification is missing.","tokens_in":28757,"tokens_out":4390,"would_cite":true,"duration_ms":42204,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B10","47D07","60J65","28C20"],"pacs":[],"model":"deepseek-v4-flash","headline":"Thin-tube Brownian motion converges to Brownian motion on the submanifold","keywords":["Brownian motion","submanifold","conditional process","Dirichlet semigroups","tubular neighbourhoods","Sasaki metric","finite-dimensional distributions","Sobolev convergence"],"falsifier":"Compute the principal symbol of the boundary operators $S_i(\\varepsilon)$ in (25) for a concrete tube, for instance around a great circle in the unit sphere, and check the Shapiro–Lopatinskij condition uniformly in $\\varepsilon$; a degenerate symbol at any small $\\varepsilon$ would break Proposition 9, and a numerical simulation of the conditioned process that fails to approach the submanifold heat kernel would break Corollary 1.","tokens_in":27759,"feed_emoji":"🎲","tokens_out":7198,"duration_ms":76372,"temperature":0.7,"pith_summary":"This paper establishes that Brownian motion confined to a thin tube around a closed submanifold of a Riemannian manifold approaches Brownian motion on the submanifold as the tube radius goes to zero. The proof is analytic: it identifies the generator of the conditioned process as a Dirichlet Laplacian on the tube, rescales the tube to unit size, and subtracts the lowest eigenvalue of the fibre Laplacian. The main theorem shows that the renormalized semigroups converge in every Sobolev space to the heat semigroup of the submanifold. A sympathetic reader would care because this gives the surface limit of conditional Brownian motion in full regularity, without relying on stochastic differential equation techniques.","feed_headline":"Thin-tube Brownian motion converges to the submanifold","feed_subtitle":"The proof controls all Sobolev norms, so the conditioned process converges in finite-dimensional distributions.","key_machinery":"The central object is the Sasaki metric on the unit disc bundle $L(1)$ and its canonical variation: the quadratic form splits as $q_{Sa,\\varepsilon}=\\varepsilon^{-2}q_V+q_H$, so the vertical Dirichlet Laplacian has fixed eigenvalues $0<\\lambda_0<\\lambda_1<\\dots$ independent of $\\varepsilon$. Renormalizing by subtracting $\\varepsilon^{-2}\\lambda_0$ and projecting onto $E_0$ kills the fast vertical modes. The perturbation of the induced metric relative to the Sasaki metric is expanded via Jacobi fields into a leading curvature form $\\Omega$ plus $\\varepsilon r_\\varepsilon$, with $\\Omega$ vanishing on $E_0$; this is what makes the limit independent of ambient curvature at leading order. The final Sobolev convergence is carried by elliptic a priori estimates for the boundary value problem (25) for powers $\\Delta_{Sa}^n$, with boundary terms of order $\\varepsilon^3$.","core_discovery":"The paper's central claim is Theorem 1: for any strongly continuous family $u(\\varepsilon)$, $$\\lim_{\\varepsilon\\to 0} $e^{{-t/2 H^0_\\varepsilon}}$ u(\\varepsilon) = E_0 $e^{{-t/2 \\Delta_L}}$ E_0 u(0)$$ uniformly on compact subintervals of $(0,\\infty)$ in the Sobolev space $H^{2n}(L(1),\\mu_{Sa})$, for every $n\\ge 1$. Here $H^0_\\varepsilon$ is the Dirichlet Laplacian on the tube of radius $\\varepsilon$, pulled back to the fixed tube $L(1)$ and renormalized by subtracting $\\lambda_0/\\varepsilon^2$, where $\\lambda_0$ is the lowest eigenvalue of the vertical Dirichlet Laplacian on the fibre, and $E_0$ is the projection onto the corresponding eigenspace. Corollary 1 translates the semigroup statement into convergence of the conditional process: for $x\\in L$ and every smooth $f$, $E_x[f(x^\\varepsilon_t)]\\to E_x[f|_L(x^0_t)]$, meaning the conditioned Brownian motion in the tube converges in finite-dimensional distributions to Brownian motion on the submanifold.","pith_inferences":["Inference: the unverified uniform Shapiro–Lopatinskij condition for the boundary problem (25) is the sharpest spot to test the proof; computing the principal symbols of the boundary operators $S_i(\\varepsilon)$ in a concrete geometry would settle whether the Sobolev convergence argument goes through.","Inference: the same renormalization scheme—subtracting the lowest vertical eigenvalue and projecting onto its eigenspace—looks like a general averaging mechanism that could apply to other fast-diffusion limits, such as homogenization on fibre bundles, with the curvature term $\\Omega$ playing the role of a vanishing corrector.","Inference: one could test numerically whether the next-order correction to the semigroup is of order $\\varepsilon^2$ and whether it is governed by the ambient curvature tensor, since the leading perturbation term $\\Omega$ is curvature-dependent but vanishes on $E_0$; the paper does not compute this rate."],"forward_implications":["The conditioned Brownian motion in the tube converges in finite-dimensional distributions to Brownian motion on the submanifold (Corollary 1).","The convergence holds in $H^{2n}(L(1),\\mu_{Sa})$ for every $n\\ge 1$, uniformly on compact time intervals, so it is much stronger than $L^2$ convergence.","The limit process is the standard Laplace–Beltrami Brownian motion on $L$; no curvature-dependent drift or extra potential survives the renormalized limit.","The argument reduces the problem to a uniformly elliptic boundary-value problem for powers of the Laplacian, so the same Sobolev-regularity strategy applies to any operator with an analogous spectral decomposition.","The authors state that tightness of the path measures, which would upgrade finite-dimensional convergence to weak convergence, is treated in a subsequent paper."],"supporting_citations":[{"why":"Supplies the canonical variation of the Sasaki metric and the commuting vertical and horizontal Laplacians, giving the spectral decomposition behind the renormalization.","marker":"[2]"},{"why":"Provides the elliptic a priori estimates and Shapiro–Lopatinskij framework used in Proposition 9 to control Sobolev norms on smooth vectors.","marker":"[1]"},{"why":"Gives the epi-convergence machinery used to identify the limit of resolvent minimizers in Proposition 3 and Corollary 7.","marker":"[4]"},{"why":"Earlier SDE-based proof of the surface limit for embeddings into Euclidean space, which the present analytic approach generalizes.","marker":"[10]"},{"why":"Earlier result on Brownian motion in tubular neighbourhoods of embedded Riemannian manifolds, setting the target process convergence.","marker":"[11]"},{"why":"Direct integral decomposition and spectral theorem for the vertical Laplacian, used to define the eigenprojections $E_k$ and the gap $\\lambda_0$.","marker":"[9]"},{"why":"Supplies compactness and norm-equivalence results used in Lemma 2 and the Sobolev setting on the tube.","marker":"[12]"},{"why":"Jacobi-field expansions that yield the asymptotic metric comparison in Propositions 5 and 6, the source of the perturbation terms.","marker":"[3]"}],"fun_headline_variants":["Tube-radius limit: conditional Brownian motion converges to submanifold","Slim tubes confine Brownian motion to submanifold limits","Dirichlet semigroups on thin tubes converge to submanifold Brownian motion","Conditioned Brownian motion in tubes converges to submanifold process"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the boundary value problem (25) satisfies the Shapiro–Lopatinskij ellipticity conditions uniformly for small tube radii—the paper states this without giving the verification—because without it the uniform a priori estimate (26) and hence the Sobolev convergence do not follow.","fun_headline_variants_meta":{"raw":{"variants":["Tube-radius limit: conditional Brownian motion converges to submanifold","Slim tubes confine Brownian motion to submanifold limits","Dirichlet semigroups on thin tubes converge to submanifold Brownian motion","Conditioned Brownian motion in tubes converges to submanifold process"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000667,"raw_usage":{"total_tokens":3043,"prompt_tokens":942,"completion_tokens":2101,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":558,"completion_tokens_details":{"reasoning_tokens":2020}},"tokens_in":558,"tokens_out":2101,"duration_ms":13910,"temperature":1.0,"reasoning_tokens":2020,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:15:46.613390+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the principal symbol of the boundary operators $S_i(\\varepsilon)$ in (25) for a concrete tube, for instance around a great circle in the unit sphere, and check the Shapiro–Lopatinskij condition uniformly in $\\varepsilon$; a degenerate symbol at any small $\\varepsilon$ would break Proposition 9, and a numerical simulation of the conditioned process that fails to approach the submanifold heat kernel would break Corollary 1.","supporting_citations":[{"cited_title":"Berard-Bergery and J","cited_arxiv_id":null,"evidence_quote":"Supplies the canonical variation of the Sasaki metric and the commuting vertical and horizontal Laplacians, giving the spectral decomposition behind the renormalization."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the elliptic a priori estimates and Shapiro–Lopatinskij framework used in Proposition 9 to control Sobolev norms on smooth vectors."},{"cited_title":"Dal Maso","cited_arxiv_id":null,"evidence_quote":"Gives the epi-convergence machinery used to identify the limit of resolvent minimizers in Proposition 3 and Corollary 7."},{"cited_title":"Sidorova, O","cited_arxiv_id":null,"evidence_quote":"Earlier SDE-based proof of the surface limit for embeddings into Euclidean space, which the present analytic approach generalizes."},{"cited_title":"Sidorova, O","cited_arxiv_id":null,"evidence_quote":"Earlier result on Brownian motion in tubular neighbourhoods of embedded Riemannian manifolds, setting the target process convergence."},{"cited_title":"Reed and B","cited_arxiv_id":null,"evidence_quote":"Direct integral decomposition and spectral theorem for the vertical Laplacian, used to define the eigenprojections $E_k$ and the gap $\\lambda_0$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies compactness and norm-equivalence results used in Lemma 2 and the Sobolev setting on the tube."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Jacobi-field expansions that yield the asymptotic metric comparison in Propositions 5 and 6, the source of the perturbation terms."}],"review_version":1}