{"id":"ff086464-1475-40c8-a569-a239371befb9","arxiv_id":"2608.05007","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A quasimode expansion in powers of 1/|ln epsilon| yields rigorous asymptotics for the mean exit time and per-window exit probabilities of reflected Brownian motion in a 2D domain with small boundary holes.","lead":"This paper derives precise formulas for how long a Brownian motion stays in a 2D domain before escaping through tiny openings on the boundary, and which opening it uses, when it starts from the natural equilibrium inside the domain. The results give the first rigorous high-order asymptotics for the exit probabilities and improve on classical first-order estimates.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the quasimode expansions are internally consistent; Prop. 2.4 is the external regularity premise most worth verifying before accepting the arbitrary-order claims.","rationale":"The paper's central claim is a rigorous arbitrary-order asymptotic expansion for λ_0^ε and for the per-window exit probabilities Y^ε_{k0}. I read the full construction of the quasimode, the boundary-cancellation recursion, and the linear-system derivation in Section 4. The main algebraic steps check out: the substitution leading to Lemma 2.12 is coherent, the recursion in Lemma 2.13 fixes the b-coefficients uniquely, the determinant computation in Lemma 4.4 is consistent with the rank-one structure, and Lemma 4.5 gives the needed lower bound on p_M. The reader's weakest-assumption identification is correct: the uniform regularity import of Proposition 2.4 is the least self-contained structural premise. Without a uniform W^{1,3} bound, the C^{1/3} control of R_k^ε collapses, and with it Corollary 2.10 and the boundary cancellation. However, this is not an observed failure: the auxiliary data N_k is uniformly bounded in L^∞ by Lemma 2.2, so the standard Neumann regularity result should apply. I also examined the signs around Lemma 2.3 and found an arithmetic slip in the displayed computation of I(t): the equations give I(t) = +π, and the lemma's stated a_k = π/|Ω| is the value consistent with that correct sign. This is a local proof defect, not a flaw in the final expansion. The paper is explicit that higher-order terms in Theorem 4.6 are a recipe rather than closed formulas; that is a stated limitation, not an error. The reliance on [6] for the QSD interpretation is partially mitigated by Appendix A, which establishes the key measure-theoretic fact directly. Overall, the central claim survives my scrutiny, and the reader's ACCEPT verdict should remain unchanged.","tokens_in":36468,"tokens_out":30347,"duration_ms":356804,"concrete_test":"Verify [3, Prop. 2.3] directly for the data in (2.23): prove, or compute for a non-circular C^∞ model domain, that ||Rtilde_k(·,t)||_{W^{1,p}} ≤ C_p ||N_k(·,t)||_{L^∞(∂Ω)} with C_p independent of t, uniformly as t approaches the window endpoints T^±_{k,ε}. Then re-derive Corollary 2.10 from that bound. If the constant blows up near the endpoints, recompute the quasimode boundary values and the O(|K^ε|^M) cancellation; if it stays bounded, the central claim stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I found no concrete failure in the quasimode construction or in the algebraic expansions of Theorems 3.3 and 4.6. The single most load-bearing step is Proposition 2.4: the uniform W^{1,p} bound for Rtilde_k(·,t) is what produces R_k^ε ∈ C^{1/3} with a uniform constant (Corollary 2.5), and hence the uniform boundary values in Corollary 2.10 that make φ^ε = O(|K^ε|^M) on Γ_D. If the imported [3, Prop. 2.3] bound degenerated as t approaches the window endpoints, the O(|ε|^{1/6}) remainder would not be absorbed by the powers K^M and the quasimode would not cancel on Γ_D. I see no evidence of such degeneration: Lemma 2.2 gives N_k uniformly bounded in L^∞(∂Ω) in t, so standard Neumann theory should supply the needed uniform bound. This premise should be checked rather than assumed. Separately, the proof of Lemma 2.3 contains an arithmetic sign slip: from I2 = -π and (2.24), I3 → 0 and I1 = 0, one obtains I(t) = +π, not -π. The stated conclusion a_k = π/|Ω| is nevertheless the correct value, so this is a typographical defect in the proof, not a load-bearing objection.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a spectral quasimode approach to the narrow escape problem for a Brownian motion in a smooth two-dimensional domain with reflecting boundary, except for N small disjoint Dirichlet arcs on the boundary. The main objects are the principal eigenvalue lambda_0^epsilon of the mixed Laplacian and the boundary normal derivative of the principal eigenfunction, which give respectively the exponential rate of the exit time and the exit-point distribution under the quasi-stationary distribution. The authors construct a quasimode with explicitly determined coefficients and prove, to any prescribed order M, expansions of lambda_0^epsilon in powers of K_k^epsilon = -1/ln(epsilon_k/2), and of the per-window exit probabilities Y_k^epsilon as rational functions of a determinant p_M(K^epsilon). The leading term of the exit-point expansion is K_{k0}^epsilon / K^epsilon, recovering the expected geometric leading behavior.","tokens_in":36682,"tokens_out":18430,"duration_ms":206723,"significance":"If the results are correct, this is a substantial contribution: it gives the first rigorous arbitrary-order asymptotic expansion of the exit-point distribution in a general smooth two-dimensional domain, in addition to the mean exit time, and it extends the authors' previous disk result to the natural setting of exits through boundary arcs. The construction is genuinely quasimodal rather than fitted: the coefficients a_gamma, b_{gamma,k}, and K_k^epsilon are determined by compatibility, recursion, and boundary cancellation, and the expansions are then derived. The paper also provides explicit recursive formulas, a determinant lower bound, and a self-contained identification of the exit measure as a probability measure. The main estimates are spot-checkable and internally consistent. The only reservations are local presentation issues and the need to make the imported uniform regularity premise fully transparent.","major_comments":[],"minor_comments":[{"comment":"In the proof of Lemma 2.3, after (2.28) one has I2(t) = -pi + O(sqrt(delta)), so with I1(t)=0 and I3(t)=O(delta), equation (2.24) gives I(t) = 0 - (-pi) + O(delta) = +pi + O(sqrt(delta)), not -pi. The stated conclusion a_k = pi/|Omega| is nevertheless correct because the compatibility condition yields a_k|Omega| = I(t); the displayed value 'I(t) = -pi' and the sentence following (2.29) should be corrected.","section":"Lemma 2.3"},{"comment":"In the proof of Lemma 2.6, the computation of ||tilde w_k^epsilon||_1 silently replaces the interval [T^-_{k,epsilon}, T^+_{k,epsilon}] by [-T^+_{k,epsilon}, T^+_{k,epsilon}]. Using only (2.1)-(2.2), the missing piece is O(epsilon_k^{3/2}), not O(|epsilon|^2), because tilde w_k^epsilon is integrably singular at the endpoint. The later statement ||w_k^epsilon||_1 = pi epsilon_k (1+O(|epsilon|^{1/2})) is unaffected, but the displayed O(|epsilon|^2) estimate should be corrected or justified.","section":"Lemma 2.6"},{"comment":"The uniform W^{1,p} bound for tilde R_k(·,t) is imported from [3, Proposition 2.3] and is load-bearing: it feeds Corollary 2.10 and hence the boundary cancellation of the quasimode in Lemmas 2.12 and 2.13. Since the bound is asserted uniformly in t and epsilon, please include the precise statement of the imported result or a short proof that the constant depends only on Omega, p, and the uniform L^infty bound of N_k supplied by Lemma 2.2. I see no indication that the bound degenerates, but the uniformity should be made transparent.","section":"Proposition 2.4"},{"comment":"The symbol O(|K^epsilon|^infty) is used repeatedly (for example in (4.7) and in the proof of Theorem 4.6) but is never defined. A short definition, e.g., bounded by C_m |K^epsilon|^m for every m, would improve readability.","section":"Section 2.1"}],"recommendation":"minor_revision","confidential_remarks":"I am not aware of any novelty or attribution concerns, and the paper is clearly within the scope of the journal. The central derivation appears sound; the requested changes are local and should be straightforward to implement. The sign slip in Lemma 2.3 is a typographical defect in the proof rather than a mathematical error, since the final compatibility value is correct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a strong, careful paper that delivers the first rigorous arbitrary-order expansions for the narrow escape problem in general two-dimensional domains, including the per-window exit probabilities. It deserves a serious referee and, assuming the one flagged premise holds up, acceptance with minor revisions.\n\nWhat's new: prior work [23] covered the disk with interior holes; here the windows are actual boundary arcs on a smooth domain, the quasimode is deliberately not in the domain of the operator, and the main theorems give expansions to arbitrary order for both the principal eigenvalue (mean exit time) and the exit point distribution. The construction is recursive and messy—there's no closed form for the higher coefficients, and the authors are honest that it's a recipe—but the algebra is internally consistent. I spot-checked Lemma 2.3's flux, the exponents in Lemmas 2.6-2.7, and the determinant structure in Lemma 4.4: all coherent. The leading-order formulas match the physics (exit rate ~ sum 1/|ln ε_k|, exit probabilities ~ 1/|ln ε_k|), and no fitted parameters appear anywhere; the coefficients are fixed by compatibility and cancellation conditions.\n\nThe real soft spot is Proposition 2.4. The entire uniformity of the quasimode boundary cancellation rests on the imported W^{1,p} bound from [3] being uniform in t and ε. I can't see a degeneration: Lemma 2.2 gives N_k uniformly in L∞, so standard Neumann theory should supply it, but this is the one premise I'd want a referee to verify carefully rather than take on faith. The other issue is a sign slip in Lemma 2.3: the flux I(t) actually computes to +π, not -π, as a quick check of (2.24)-(2.28) shows. The conclusion a_k = π/|Ω| is still correct because the compatibility condition involves -I(t), so this is a typographical defect, not a load-bearing error, but it should be fixed. The self-citation [6] for QSD facts is unverifiable from here, but Appendix A gives the PDE proof and the statements are standard.\n\nWho this is for: anyone in PDE/spectral theory or applied probability working on metastability, narrow escape, or small-hole asymptotics. It's not light reading, but it's honest, self-contained, and the main claims are concrete. I'd bring it to a reading group and cite it. Send it to review.","headline":"Strong, careful paper delivering the first rigorous arbitrary-order expansions for the 2D narrow escape problem, including exit point laws; the only real flag is an imported regularity premise worth double-checking.","tokens_in":37324,"tokens_out":3172,"would_cite":true,"duration_ms":33258,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35P05","35J25","35C20","60J65"],"pacs":[],"model":"deepseek-v4-flash","headline":"For Brownian motion in a smooth 2D domain with N tiny exit windows, the paper proves that the exit rate and each window's exit probability have asymptotic expansions, to any order, in inverse powers of the logarithms of the window radii.","keywords":["narrow escape problem","quasi-stationary distribution","Brownian motion","Dirichlet-Neumann Laplacian","asymptotic expansion","quasimode","exit point distribution","logarithmic weights"],"falsifier":"Take a disk with one exit window of radius ε and compute λ_0^ε numerically (for instance by finite elements or a spectral method) for several values of ε. If the difference between λ_0^ε and the paper's M-th order expansion does not decay like |K^ε|^{M+1}, equivalently like |ln ε|^{-(M+1)}, then the remainder claim is false. Similarly, simulate a two-window domain and compare the empirical exit frequency through window 1 with the predicted leading term K_1^ε/K^ε and its first corrections.","tokens_in":36182,"feed_emoji":"🧮","tokens_out":5378,"duration_ms":64972,"temperature":0.7,"pith_summary":"The paper studies a Brownian particle in a smooth two-dimensional container whose boundary reflects the particle except for N small arcs (windows) through which it can escape. Assuming the particle starts in the quasi-stationary distribution, the exit time is exponential with rate given by the principal eigenvalue of the Laplacian with Dirichlet conditions on the windows and Neumann conditions elsewhere. The paper constructs a quasimode, an explicit approximate eigenfunction, in powers of K_k^ε = 1/|ln(ε_k/2)|, and proves that both the eigenvalue and the probabilities of leaving through each window admit asymptotic expansions to any prescribed order M. This gives, for general domains, a parameter-free spectral version of the narrow-escape asymptotics and, the authors state, the first mathematical expansion of the exit-point distribution.","feed_headline":"Brownian escape through tiny windows: expansions to any order","feed_subtitle":"Mean exit time and per-window exit odds follow inverse-log expansions on any smooth 2D domain, to any order.","key_machinery":"The quasimode is φ^ε(x) = 1 + Σ_{1≤|γ|≤M} (K^ε)^γ f_γ^ε(x), where each first-order f_k^ε is a sum of a singular logarithmic layer S_k^ε with arcsine weight w_k^ε(t) = χ_k^ε(t)(1−(t/T_{k,ε}^+)^2)^{-1/2} and a regular part R_k^ε solving a Neumann Poisson problem with compatibility constant a_k = π/|Ω|. Higher-order f_γ^ε are built recursively from v_γ and linear combinations of the f_k^ε, with coefficients b_{γ,j} chosen so that φ^ε is O(|K^ε|^M) uniformly on the Dirichlet windows. Green's formula then transfers the quasimode estimates to λ_0^ε, while, for the exit-point probabilities, auxiliary quasimodes with one window removed lead to a linear system whose leading matrix has an explicitly computed determinant p_M(K^ε); the rational-form expansion of the solution is what produces Theorem 4.6.","core_discovery":"For a C∞ bounded planar domain with N disjoint small exit arcs of radii ε_k, starting from the quasi-stationary distribution, the principal eigenvalue satisfies λ_0^ε = Σ_{1≤|γ|≤M} a_γ (K^ε)^γ + O(|K^ε|^{M+1}), with coefficients a_γ constructed recursively from solutions of linear elliptic problems and fixed by the requirement that the quasimode vanish approximately on the Dirichlet windows. The probability of exiting through window k_0 satisfies Y_{k_0}^ε = Σ_{q=0}^{M-1} Σ_{m=0}^q $Q^{{(k_0)}}$_{M,m,q}(K^ε)/p_M(K^ε)^{m+1} + O(|K^ε|^M), where p_M is an explicitly computed determinant, and the leading term is K_{k_0}^ε/K^ε. Because the quasimode is not in the domain of the operator but carries exact Neumann conditions and approximate Dirichlet conditions with controlled errors, the spectral computations are uniform in ε, yielding both the mean exit time (the inverse of the eigenvalue) and the full exit-point law in the small-window limit.","pith_inferences":["A direct Monte Carlo simulation of reflected Brownian motion in a disk or rectangle with two windows could test the leading prediction P_k ≈ K_k^ε/K^ε, and the paper's expansion predicts the next-order corrections that such a simulation could look for.","The construction suggests that the narrow-escape problem inherits an Eyring-Kramers-like structure for entropic metastability: the exponential rate is controlled by geometrically computable coefficients (starting with π/|Ω|) rather than by an energy barrier, and the higher-order coefficients encode boundary curvature and inter-window distances.","The determinant p_M(K^ε) organizing the exit-probability expansion points to an algebraic structure in inverse-logarithm variables that may also appear in related problems such as narrow capture or escape in tubes, where similar spectral quasimodes are used."],"forward_implications":["Under the quasi-stationary distribution, the exit time is exponentially distributed with rate given by the M-th order expansion of λ_0^ε, with error O(|K^ε|^{M+1}).","The probability of leaving through window k has leading term K_k^ε/K^ε, so among windows with comparable logarithmic sizes the exit odds are proportional to the inverse-log weights 1/|ln(ε_k/2)|.","The exit-point expansion is obtained for general smooth two-dimensional domains, not only disks, and to arbitrary order, extending the previously known first-order mean-exit-time asymptotics.","The coefficient recursion, driven by the local geometry at the window centers encoded in the functions f_j^0(x^{(k)}) and v_γ, provides a concrete algorithm for computing higher-order corrections to both the rate and the exit probabilities."],"supporting_citations":[{"why":"Supplies the uniform W^{1,p} regularity result for the auxiliary Neumann problem (2.23) that Proposition 2.4 imports and on which the uniformity of all remainder estimates rests.","marker":"[3]"},{"why":"Provides the layer-potential and singular-integral techniques underlying the construction of the singular part S_k^ε.","marker":"[2]"},{"why":"The preceding spectral approach for the disk that this paper extends to general 2D domains and to higher-order exit-point asymptotics.","marker":"[23]"},{"why":"Establishes the quasi-stationary distribution and the identities identifying λ_0^ε and ∂_n u_0^ε/λ_0^ε with the exit time and exit point laws.","marker":"[6]"},{"why":"Motivates the quasi-stationary distribution as the natural initial condition for metastable escape and parametrization of jump Markov models.","marker":"[11]"},{"why":"Provides the earlier layer-potential-based asymptotic expansions of the mean first exit time that the present quasimode construction partially draws on.","marker":"[1]"}],"fun_headline_variants":["To any order: Brownian escape from 2D domains","Inverse-log expansions for narrow escape","Any-order expansions for exit time and exit point","Spectral method yields full exit law for small windows"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on a uniform regularity estimate for a family of auxiliary Neumann problems: the solutions must stay uniformly bounded in $W^{{1,p}}$ as the windows shrink, with constants independent of t and ε, and if that uniformity degenerates the remainder bounds collapse.","fun_headline_variants_meta":{"raw":{"variants":["To any order: Brownian escape from 2D domains","Inverse-log expansions for narrow escape","Any-order expansions for exit time and exit point","Spectral method yields full exit law for small windows"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000233,"raw_usage":{"total_tokens":1445,"prompt_tokens":848,"completion_tokens":597,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":464,"completion_tokens_details":{"reasoning_tokens":536}},"tokens_in":464,"tokens_out":597,"duration_ms":7595,"temperature":1.0,"reasoning_tokens":536,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T11:40:43.202255+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a disk with one exit window of radius ε and compute λ_0^ε numerically (for instance by finite elements or a spectral method) for several values of ε. If the difference between λ_0^ε and the paper's M-th order expansion does not decay like |K^ε|^{M+1}, equivalently like |ln ε|^{-(M+1)}, then the remainder claim is false. Similarly, simulate a two-window domain and compare the empirical exit frequency through window 1 with the predicted leading term K_1^ε/K^ε and its first corrections.","supporting_citations":[{"cited_title":"Carillo, T","cited_arxiv_id":null,"evidence_quote":"Establishes the quasi-stationary distribution and the identities identifying λ_0^ε and ∂_n u_0^ε/λ_0^ε with the exit time and exit point laws."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the uniform W^{1,p} regularity result for the auxiliary Neumann problem (2.23) that Proposition 2.4 imports and on which the uniformity of all remainder estimates rests."},{"cited_title":"Ammari, H","cited_arxiv_id":null,"evidence_quote":"Provides the layer-potential and singular-integral techniques underlying the construction of the singular part S_k^ε."},{"cited_title":"Di Ges` u, T","cited_arxiv_id":null,"evidence_quote":"Motivates the quasi-stationary distribution as the natural initial condition for metastable escape and parametrization of jump Markov models."},{"cited_title":"Ammari, K","cited_arxiv_id":null,"evidence_quote":"Provides the earlier layer-potential-based asymptotic expansions of the mean first exit time that the present quasimode construction partially draws on."}],"review_version":1}