{"id":"0fd8b24e-8c2f-426e-87e0-c06484c5dea8","arxiv_id":"1908.05100","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For catalytic branching random walks on Z^d, the projection of the front onto each normal direction satisfies a limit theorem with limit distribution phi, extending one-dimensional single-catalyst results to multiple catalysts and higher dimensions.","lead":"This paper proves that in a supercritical catalytic branching random walk on a multidimensional lattice, the particle cloud's front fluctuates around its deterministic limiting shape according to a non-trivial limiting distribution. The result extends earlier one-dimensional single-catalyst theorems to any finite number of catalysts and any dimension, using a projection method.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The N>1 extension rests on an unverified application of Crump's matrix renewal theorem; the asymptotic for sum_k G^{*k} * I with positive constants K_i^{(N)} is asserted, not proved, and its hypotheses (direct Riemann integrability, Perron eigenvalue, positivity) are not checked.","rationale":"The reader's conditional verdict identifies the same gap. My reading of Section 3 confirms that the full proof is only written for d=1, N=1; the N>1 paragraph (p. 18) and the d>1 paragraph (p. 19) contain the two 'omitted details' passages. The matrix renewal asymptotic is not a cosmetic omission: it is the mechanism that converts the renewal-theoretic input into the c_* constants that pin down the solution class C_theta. Without it, the limit in Theorem 1 is not derived for the advertised generality. This is a completeness/correctness-risk concern, not an observed contradiction; it does not require rejection. The concrete test above would settle whether Crump's corollary applies. I agree with the reader's weakest-assumption and keep the conditional verdict.","tokens_in":22309,"tokens_out":10186,"duration_ms":113576,"concrete_test":"Pull Crump (1970), Corollary 3.1(i), and check its hypotheses against G and I^{(N)} for a concrete two-catalyst model on Z (e.g., nearest-neighbor symmetric random walk, two catalysts at 0 and a>0, binary branching at each catalyst). Compute (a) the Perron root of integral e^{-nu t} dG(t) and verify it equals 1 with positive eigenvectors; (b) direct Riemann integrability of e^{-nu t} I^{(N)}(t; mu t + y) for each coordinate; (c) the resulting K_1^{(N)}, K_2^{(N)} and compare with a Monte Carlo estimate of lim_{t->infinity} e^{ry - r{mu t + y}} P_{w_i}(M_t > mu t + y). If the constants are finite, positive, and match simulation, the omitted step can be filled; if not, the proof fails for N>1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The advertised extension from the fully proved d=1 single-catalyst case to N>1 (and then d>1) turns on the asymptotic sum_{k>=0} G^{*k} * I(t; mu t + y) ~ e^{-ry + r{mu t + y}} (K_1^{(N)}, ..., K_N^{(N)})^T, stated in the proof of Corollary 1 immediately after eq. (43). The text says 'Completely similarly to Lemma 3, using Corollary 3.1, item (i), of paper [20]' and then omits the constants as 'cumbersome and superfluous.' This is a genuine load-bearing step: without it, Lemma 5, the counterpart of Lemma 7, and hence the c* identification for each catalyst do not go through. To make the step valid one would have to verify at least: (a) the matrix measure G(t) with entries delta_{ij} alpha_i m_i G_i(t) + (1 - alpha_i) G_i * {}^{W_j}F_{w_i,w_j} satisfies the hypotheses of Crump's corollary, including the Perron-Frobenius condition at nu and the correct arithmetic/non-lattice renewal classification; (b) the forcing vector e^{-nu t} I^{(N)}(t; mu t + y) is directly Riemann integrable, with coordinatewise bounds analogous to Lemma 4; (c) the limit constants K_i^{(N)} are finite and strictly positive, since the counterpart of Lemma 7 needs theta_i > 0 for each i. None of these points is checked in the manuscript. The same applies to the d>1 case, where the paper replaces S by <S,r> and says 'the further details can be omitted'; the projected walk can be lattice or non-lattice depending on r, and the renewal theorem must be applied to the projected random walk with the appropriate arithmetic span. Thus the full-strength Theorem 1, as stated for arbitrary finite catalysts and d>1, is not supported by the written proof.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a supercritical catalytic branching random walk (CBRW) on Z^d with a finite catalyst set W. For r in the level set R = {H(r) = nu}, it defines M_t(r) = max_z <X_z(t), r> and claims that P_x(M_t(r) - nu t <= y) - phi(e^{-y + chi(t;y)}; x) -> 0 as t -> infinity, where phi is the solution of the integral equations (8)-(9). The proof is carried out in detail for d = 1 with a single catalyst, using a nonlinear renewal equation for the tail probability and large-deviation estimates. The extension to N > 1 and d > 1 is sketched via a matrix renewal theorem of Crump. As a corollary, the one-dimensional result extends the Carmona-Hu theorem to an arbitrary finite number of catalysts.","tokens_in":22767,"tokens_out":6506,"duration_ms":61765,"significance":"If correct, Theorem 1 gives a complete description of the fluctuations of the population front around the limiting shape on the non-degeneracy set, for an arbitrary finite number of catalysts and any dimension. The d = 1, N = 1 proof is detailed and internally consistent, and it produces explicit constants, notably c* in Lemma 7. The paper also highlights an interesting contrast with standard branching random walks, where a logarithmic correction is present, whereas here the fluctuations are stochastically bounded. However, the advertised multidimensional and multi-catalyst generalization rests on an unverified matrix renewal asymptotic, so the full significance of the paper as stated is conditional on filling that gap.","major_comments":[{"comment":"The asymptotic sum_{k>=0} G^{*k} * I(t; mu t + y) ~ e^{-ry + r{mu t + y}} (K_1^{(N)}, ..., K_N^{(N)})^T is asserted with 'Completely similarly to Lemma 3, using Corollary 3.1, item (i), of paper [20]', and the constants are omitted as 'cumbersome and superfluous'. This step is load-bearing: it is the only justification for the N > 1 counterparts of Lemmas 3, 5, 7, and 8, and hence for the identification of the limit in Theorem 1 for N > 1. The hypotheses of Crump's matrix renewal theorem are not verified: the matrix measure G(t) with entries delta_{ij} alpha_i m_i G_i(t) + (1 - alpha_i) G_i * {}^{W_j}F_{w_i,w_j}(t) must be shown to satisfy the Perron-Frobenius condition at nu, the forcing vector e^{-nu t} I^{(N)}(t; mu t + y) must be directly Riemann integrable with bounds analogous to Lemma 4, and the constants K_i^{(N)} must be finite and strictly positive. None of these points is checked in the manuscript. Without these verifications, the proof of Corollary 1 for N > 1 is incomplete.","section":"Proof of Corollary 1, paragraph after eq. (43)"},{"comment":"The reduction to the projected walk <S(t), r> and the remark 'the further details can be omitted' leave unaddressed the lattice/non-lattice classification of the renewal theorem for the projected walk. The paper notes in Section 2 that <S(t), r> can be lattice or non-lattice depending on r, but the proof sketch does not verify which renewal case applies in the matrix renewal argument, nor does it provide the d > 1 analogue of the constant c*. Consequently, Theorem 1 for d > 1 is not established; only the d = 1, N = 1 case is proved in full.","section":"Proof of Theorem 1, d > 1 case"},{"comment":"The uniqueness of phi in the class C_theta is invoked with theta_i defined as lim_{y -> +infty} lim_{t -> infty} e^{y - chi(t;y)} P_{w_i}(M_t(r) > nu t + y). For N > 1, this requires the positivity and finiteness of the constants K_i^{(N)} mentioned in the proof of Corollary 1. Since those constants are not proved to exist, the definition of the class C_theta and the uniqueness statement for the system (8) are not justified in the multi-catalyst case.","section":"Theorem 1 and eqs. (8)-(9), definition of C_theta"}],"minor_comments":[{"comment":"The definition of limsup_{t -> infinity} {A_t} using binary-rational times is written as a union/intersection over m, k, n; it would be helpful to state explicitly that this defines the event of infinitely many visits at dyadic times and that it has the same probability as the usual continuous-time limsup event.","section":"Section 2, definition of I"},{"comment":"The relation between the correction function chi(t;y) in Theorem 1 and the fractional part {mu t + y} in Corollary 1 could be stated explicitly; in the one-dimensional lattice case, chi(t;y) should reduce to {mu t + y} up to the lattice span.","section":"Theorem 1 and Corollary 1, chi(t;y)"},{"comment":"The constant c* is displayed as e^{-r}(1 - F*_{0,0}(nu))(1 - alpha_1 G*_1(nu)) / sqrt(2(1 - e^{-r}) H'(r)) times an integral; the derivation of the factor (1 - e^{-r}) is not transparent. A short explanation would improve readability.","section":"Proof of Lemma 3, relation (25)"},{"comment":"The proof relies heavily on the author's previous works [10]-[13] and [15]-[17]; the paper would be easier to verify if the specific results used from these papers were stated as lemmas where they enter the argument.","section":"References and use of prior work"},{"comment":"There are numerous typographical and spacing errors in the title and text (for example, 'W alk' and 'T heorem'); the manuscript needs a careful proofreading before resubmission.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is heavily self-referential: Theorem 1 and its proof depend on results from [10]-[13] without full statements. The central gap in the matrix renewal step is not a matter of disagreement with current consensus but a missing verification of hypotheses; the authors should either prove the renewal lemma for the matrix case or explicitly restrict the main theorem to the fully proved d = 1, N = 1 case. Given the importance of the claimed generalization, major revision is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Ekaterina's paper is worth a serious referee, but the referee should be told upfront that the fully proved part is narrower than the abstract suggests. The d=1, single-catalyst case is done carefully: the integral equation for E(t;u), the renewal argument in Lemma 3, the constant c* computed from first principles, and the iteration/limiting argument in Lemma 8 all hang together. I checked the steps that looked delicate -- the Laplace integral, the fractional-part correction, the passage from binary-rational times to real t in Lemma 2 -- and they are plausible and mostly self-contained. The projection method for d>1, replacing X_z(t) by <X_z(t),r>, is a clean way to make sense of front fluctuations on a curved limiting surface. That is the genuine new idea.\n\nThe soft spot is exactly where the reader's report puts it: the move from N=1 to N>1 in Corollary 1 and then to d>1 in Theorem 1. The proof says 'Completely similarly to Lemma 3, using Corollary 3.1, item (i), of [20]' and then omits the constants K_i^{(N)} as cumbersome. That is load-bearing. To justify the asymptotic for the renewal series of the matrix G(t) one has to check the hypotheses of Crump's theorem: Perron-Frobenius structure at the Malthusian exponent, arithmetic/non-lattice classification for the entries, direct Riemann integrability of the forcing vector, and positivity of K_i^{(N)}. None of those checks is in the manuscript. The d>1 case adds another gap: the projected walk <S(t),r> can be lattice or non-lattice depending on r, and the paper just says the details can be omitted. The statement of Theorem 1 as written, for arbitrary finite N and d>1, is therefore not supported by the written proof. I do not see a direct error in what is written; the extension is plausible and likely true, but it is a sketch, not a proof.\n\nThe self-citation density is high, but in this area the author genuinely did prove the earlier classification and front-shape results, so I don't treat it as a red flag. The limit function phi is defined as a solution of a decomposition equation, not assumed to be the limiting object, so the circularity concern is unfounded.\n\nBottom line: the detailed d=1 proof and the projection method justify sending this to a competent referee. The referee should be asked to either supply the missing renewal-theorem verification and the d>1 lattice/non-lattice handling, or to restate Theorem 1 with the N>1/d>1 parts as conditional. I would not desk-reject; I also would not accept on the current text.","headline":"The d=1 single-catalyst proof is solid and the projection idea is genuinely useful, but the advertised N>1 and d>1 generalizations ride on an unverified matrix renewal step.","tokens_in":23238,"tokens_out":2321,"would_cite":true,"duration_ms":22688,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J80","60F05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The front of a supercritical catalytic branching walk has a complete fluctuation law, given by the integral-equation function $\\phi$.","keywords":["catalytic branching random walk","supercritical regime","spread of population","propagation front","front fluctuations","multidimensional lattice","integral equations","renewal theory"],"falsifier":"Simulate a two-catalyst supercritical catalytic branching walk on $\\mathbb{Z}$ with binary branching and exponential waiting times, solve (8)--(9) numerically, and compare the empirical distribution of $M_t-\\mu t$ with $\\phi(e^{-ry+\\{\\mu t+y\\}};x)$ for large $t$. The theorem predicts agreement to $o(1)$; a persistent discrepancy would refute it.","tokens_in":22106,"feed_emoji":"🧬","tokens_out":14187,"duration_ms":126475,"temperature":0.7,"pith_summary":"The paper aims to prove that the random position of the population front in a supercritical catalytic branching random walk on $\\mathbb{Z}^d$ has non-trivial, fully describable fluctuations around its deterministic linear speed. Earlier work on the integer line proved that the rightmost particle $M_t$ satisfies $M_t/t\\to\\mu$ almost surely and, for a single catalyst, found a limit law for $M_t-\\mu t$. This paper extends that limit law to any finite number of catalysts and to all dimensions by studying, for each direction $r$ on the limiting surface, the maximal projection $M_t(r)$ and its difference from $\\nu t$. The complete description is a single function $\\phi$ solving a system of integral equations, and it also encodes the probability that the population dies out locally. If correct, the result closes the fluctuation problem for light-tailed catalytic branching walks on the set where the population spreads.","feed_headline":"Front fluctuations in catalytic branching walk now have a limit law","feed_subtitle":"One integral-equation function gives the full fluctuation law for every point of the front's limiting shape.","key_machinery":"The central object is the function $\\phi(\\lambda;x)$, $\\lambda\\ge0$, $x\\in\\mathbb{Z}^d$, the unique solution of the nonlinear integral system (8)--(9). At each catalyst $w_j$, equation (8) balances the branching contribution, through the offspring generating function $f_j$ and the exponential clock $G_j$, against the chance that the particle leaves the catalyst and later returns, while equation (9) expresses $\\phi$ away from catalysts as a renewal-type average over first hitting times of the catalyst set under taboo, that is, avoiding the other catalysts before the hit. The proof machinery consists of writing the exceedance probability $\\mathbb{P}_x(M_t(r)>u)$ as a renewal equation (Lemma 1 for one catalyst, system (42) in general), iterating the renewal kernel $G$, estimating the boundary term $I(t;u)$ by large-deviation bounds for the underlying random walk, and applying Laplace asymptotics to the resulting renewal sums. The multidimensional extension projects all particle positions onto the normal direction $r$ at a point of the limiting surface $\\mathcal{P}$ and applies the same argument to the scalar walk $\\langle S(t),r\\rangle$.","core_discovery":"Theorem 1 states that, for every starting point $x$, every $r$ with $H(r)=\\nu$, and every real $y$, $$\\lim_{t\\to\\infty}\\bigl(\\mathbb{P}_x(M_t(r)-\\nu t\\le y)-\\$\\varphi$($e^{{-y+\\chi(t;y)}}$;x)\\bigr)=0.$$ Here $M_t(r)$ is the largest value of $\\langle X_z(t),r\\rangle$ among particles alive at time $t$, $\\nu$ is the Malthusian parameter of the supercritical regime, and $\\chi(t;y)$ is a lattice-spacing correction that is $r^*\\{\\nu t/r^*+y/r^*\\}$ when the projected walk is lattice and $0$ otherwise. The function $\\phi(\\lambda;x)$ is defined as the unique solution of the integral system (8)--(9), built from the offspring laws, the exponential waiting times at catalysts, and taboo hitting-time distributions of the underlying random walk. As $\\lambda\\to\\infty$, $\\phi(\\lambda;x)$ tends to $1-\\mathbb{P}_x(I)$, the probability of local extinction, so Theorem 1 gives a complete fluctuation description on the event $I$ of infinitely many catalyst visits. For $d=1$, it reduces to $\\mathbb{P}_x(M_t-\\mu t\\le y)-\\phi(e^{-ry+\\{\\mu t+y\\}};x)\\to0$ with any finite number of catalysts, generalizing the earlier single-catalyst result.","pith_inferences":["The proof suggests a two-sided reading not made explicit in the paper: the front fluctuation law and the population-size limit law are governed by the same $\\phi$, so progress on population-size asymptotics should translate directly into sharper front-fluctuation statements.","A concrete testable extension would be a numerical simulation of a two-catalyst walk on $\\mathbb{Z}$ with exponential clocks, comparing the empirical distribution of $M_t-\\mu t$ with the theorem's $\\phi(e^{-ry+\\{\\mu t+y\\}};x)$; this would also indirectly probe the omitted renewal constants.","If the matrix-renewal asymptotic holds for a given catalyst configuration, the same argument should extend to asymmetric jump kernels and to catalysts with distinct offspring laws; if the asymptotic fails, the multidimensional several-catalyst statement would need a genuinely different proof."],"forward_implications":["On $\\mathbb{Z}$, the maximum $M_t$ satisfies the strong law $M_t/t\\to\\mu$ and $M_t-\\mu t$ converges in distribution after the lattice correction, now with any finite number of catalysts rather than only one.","At each point of the limiting surface in $\\mathbb{Z}^d$, the particle-cloud front has bounded fluctuations in the normal direction, with limit law given by $\\phi$; fluctuations in other directions are asymptotically invisible at the $\\nu t$ scale.","The function $\\phi$ simultaneously gives the fluctuation distribution and the local-extinction probability $1-\\mathbb{P}_x(I)$, so the theorem covers the whole non-degeneracy event $I$.","Unlike homogeneous branching random walks, catalytic branching walks show no logarithmic correction: after subtracting $\\nu t$, only stochastically bounded fluctuations remain."],"supporting_citations":[{"why":"Establishes the one-dimensional strong law $M_t/t\\to\\mu$ and the single-catalyst fluctuation limit that this paper generalizes.","marker":"[18]"},{"why":"Provides the multidimensional limiting-shape surface $\\mathcal{P}$ and the a.s. spread estimates motivating the projection method.","marker":"[12]"},{"why":"Supplies the nonlinear integral equation for $\\mathbb{P}_0(M_t>u)$ and the uniqueness of the solution class for $\\phi$.","marker":"[13]"},{"why":"Defines the supercritical regime and the Malthusian parameter $\\nu$ used throughout.","marker":"[10]"},{"why":"Identifies $\\phi$ as a Laplace transform of limit population sizes and gives the local-extinction probability $1-\\mathbb{P}_x(I)$.","marker":"[11]"},{"why":"Matrix renewal theorem used to obtain the exponential renewal asymptotic in the several-catalyst case.","marker":"[20]"},{"why":"Large-deviation asymptotics for the underlying random walk used to control the boundary term $I(t;u)$.","marker":"[7]"}],"fun_headline_variants":["Exact front fluctuation law for catalytic branching walk","Multidimensional limit law for catalytic branching front","Full fluctuation theorem for catalytic branching front","Front fluctuations solved in catalytic branching walk"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, for several catalysts and in dimensions $d>1$, repeated visits to the catalysts settle into the regular exponential pattern predicted by a standard renewal theorem, with positive constants whose explicit form the paper leaves out; if that settling fails, the limit theorem for those cases does not follow.","fun_headline_variants_meta":{"raw":{"variants":["Exact front fluctuation law for catalytic branching walk","Multidimensional limit law for catalytic branching front","Full fluctuation theorem for catalytic branching front","Front fluctuations solved in catalytic branching walk"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000354,"raw_usage":{"total_tokens":2004,"prompt_tokens":1100,"completion_tokens":904,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":716,"completion_tokens_details":{"reasoning_tokens":850}},"tokens_in":716,"tokens_out":904,"duration_ms":9505,"temperature":1.0,"reasoning_tokens":850,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:23:24.496588+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a two-catalyst supercritical catalytic branching walk on $\\mathbb{Z}$ with binary branching and exponential waiting times, solve (8)--(9) numerically, and compare the empirical distribution of $M_t-\\mu t$ with $\\phi(e^{-ry+\\{\\mu t+y\\}};x)$ for large $t$. The theorem predicts agreement to $o(1)$; a persistent discrepancy would refute it.","supporting_citations":[{"cited_title":"The spread of a catalytic branching random w alk","cited_arxiv_id":null,"evidence_quote":"Establishes the one-dimensional strong law $M_t/t\\to\\mu$ and the single-catalyst fluctuation limit that this paper generalizes."},{"cited_title":"Spread of a catalytic branching random walk on a m ultidimensional lattice","cited_arxiv_id":null,"evidence_quote":"Provides the multidimensional limiting-shape surface $\\mathcal{P}$ and the a.s. spread estimates motivating the projection method."},{"cited_title":"Maximum of Catalytic Branching Random Walk with Regularly Varying Tails","cited_arxiv_id":"1808.01465","evidence_quote":"Supplies the nonlinear integral equation for $\\mathbb{P}_0(M_t>u)$ and the uniqueness of the solution class for $\\phi$."},{"cited_title":"Complete classification of catalytic branching pro cesses","cited_arxiv_id":null,"evidence_quote":"Defines the supercritical regime and the Malthusian parameter $\\nu$ used throughout."},{"cited_title":"Strong and weak convergence of the population size in a supercritical catalytic branching process","cited_arxiv_id":null,"evidence_quote":"Identifies $\\phi$ as a Laplace transform of limit population sizes and gives the local-extinction probability $1-\\mathbb{P}_x(I)$."},{"cited_title":"On systems of renewal equations","cited_arxiv_id":null,"evidence_quote":"Matrix renewal theorem used to obtain the exponential renewal asymptotic in the several-catalyst case."},{"cited_title":"Asymptotic Analysis of Random Walks","cited_arxiv_id":null,"evidence_quote":"Large-deviation asymptotics for the underlying random walk used to control the boundary term $I(t;u)$."}],"review_version":1}