{"id":"8b0bb0cf-2576-4696-a538-1bb7c9e76fb6","arxiv_id":"2505.16904","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Claims existence, uniqueness, and positivity for a stochastic predator-prey model, but the positivity proof does not rule out hitting zero.","lead":"This paper adds demographic noise to the Rosenzweig-MacArthur predator-prey model and claims a unique positive solution that never goes extinct. Generalist readers may care because the paper tries to connect stochastic calculus with ecological persistence, but the main proof has a boundary flaw.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.5's domain invariance is not established and is in fact false: the Lyapunov function V=(1+N^2+P^2)^α does not blow up at N=0 or P=0, and the square-root diffusion makes the zero boundary attainable, so P((N(t),P(t))∈D)=1 cannot hold in general.","rationale":"The reader's weakest assumption is exactly the gap I find: the Lyapunov function controls only the far boundary, not the axes. My independent Lamperti/Bessel calculation shows the boundary is not merely unexcluded but actually attainable, so the paper's central claim fails, not just its proof. I also note that the truncating sets K_n fail Theorem 4.3's precompactness hypothesis and that the coefficients are not Lipschitz on K_n near zero, but the boundary-attainability point is the most decisive single concern. Because this directly contradicts Theorem 4.5 and Remark 4.6, the reader's REJECT is appropriate; no adjustment is needed.","tokens_in":17217,"tokens_out":12227,"duration_ms":92093,"concrete_test":"Carry out Feller's boundary test for the Y=√N diffusion obtained by Itô's formula from (1.4), holding P fixed at a positive value. Near Y=0 the SDE is dY≈((1-mP)Y/2-(1+mP)/(8Y))dt+(1/2)√(1+mP)dB^1. Its scale density is p(Y)≈Y and its speed measure density is m(Y)≈4/[(1+mP)Y]. Since ∫_0^ε p(Y)dY<∞ while ∫_0^ε m(Y)dY=∞, the boundary 0 is an exit (attainable) boundary. Run this computation explicitly for m=3, c=1, k=3 with P=0.6; if the boundary is attainable, Theorem 4.5's domain-invariance claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step in the proof of Theorem 4.5 is the estimate controlling τ_n = inf{t≥0: (N_t,P_t)∉K_n}, with K_n=(0,n]^2. Since D∖K_n = {(N,P): N>n or P>n}, the quantity inf_{D∖K_n} V = (1+n^2)^α is governed entirely by the far boundary; V=(1+N^2+P^2)^α remains bounded on every strip near the axes. Therefore the Itô/Chebyshev estimate only proves that the solution does not escape to infinity. It gives no control over the event that N_t or P_t hits 0, and if the process exits U through the zero boundary, X_{τ_n}∉U, so the application of Theorem 4.3 and the Itô formula on U is no longer justified. The missing event is real: near N=0 with P>0, equation (1.4) has the form dN≈(1-mP)N dt+√(N(1+mP))dB^1. Setting Y=√N gives dY≈[(1-mP)Y/2-(1+mP)/(8Y)]dt+(1/2)√(1+mP)dB^1, whose 1/Y drift is the same as that of a 0-dimensional Bessel process; Feller's boundary test classifies 0 as an attainable boundary. Hence P(N_t,P_t>0 for all t)=1 is false in general, and extinction is possible. The same gap invalidates Remark 4.6 and the persistence interpretation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a stochastic version of the Rosenzweig-MacArthur predator-prey model driven by internal demographic noise, with diffusion coefficients proportional to population densities. It presents a general existence-and-uniqueness criterion for autonomous SDEs on an open submanifold of R^n (Theorem 4.3), then applies it to the stochastic Rosenzweig-MacArthur model to claim global well-posedness and invariance of the positive quadrant D (Theorem 4.5). The paper further derives L^p moment estimates and an almost-sure exponential growth bound, and supports the analysis with Euler-Maruyama simulations showing deviations from the deterministic model.","tokens_in":17584,"tokens_out":10066,"duration_ms":83022,"significance":"If the central results were correct, the paper would offer a useful framework for SDEs constrained to open submanifolds and a rigorous persistence statement for an intrinsically stochastic predator-prey model. The Lyapunov computations in Theorem 4.5 and the L^p estimates in Proposition 4.8 are explicit and checkable, and the general idea of using Lyapunov functions that do not blow up at every boundary is worth exploring. However, the main domain-invariance claim is false: the chosen Lyapunov function does not blow up at the zero boundary, and the square-root noise structure makes that boundary attainable. The proof of the general criterion also contains a load-bearing gap. Since the paper's biological conclusion of persistence and non-extinction rests on the invalid Theorem 4.5, the current manuscript is not suitable for publication.","major_comments":[{"comment":"The sets K_n = (0,n]^2 are closed in D but not precompact: their closure in D is still (0,n]^2, which is not compact because a sequence such as (1/m,1) has no convergent subsequence in D. Hence no open precompact subset ~K_n of D can contain K_n, and the hypothesis of Theorem 4.3 requiring the existence of such ~K_n fails. Therefore Theorem 4.3 cannot be applied to the model, and the proof of Theorem 4.5 collapses at this point.","section":"Section 4, Theorem 4.5"},{"comment":"The assertion that 'since for each n, supp ~mu_n, supp ~sigma_n ⊆ ~K_n and ~K_n is precompact, it is easy to see each mu_n and sigma_n are globally Lipschitz continuous' is false: a continuous function with compact support need not be Lipschitz (e.g., sqrt(x) on [0,1]). In the application the coefficients are C^1 on D, so a smooth extension with compact support could be used instead, but the argument as written does not establish the Lipschitz condition required for the classical existence theorem.","section":"Section 4, Theorem 4.3 proof"},{"comment":"The Lyapunov function V = (1+N^2+P^2)^alpha is bounded on the strips 0<N<=n, 0<P<=n near the axes. Consequently the estimate P(tau_n <= t) <= ... / inf_{D\\K_n} V only controls the event that the process escapes to large values of N or P; it gives no control over the event that N or P hits zero. If the process exits D through the zero boundary, then X_{tau_n} is not in D, V is not defined at that point, and the lower bound V(X_{tau_n}) >= inf_{D\\K_n} V used in the proof fails. Thus the proof does not establish P((N(t),P(t)) in D for all t) = 1.","section":"Section 4, Theorem 4.5 proof and Remark 4.6"},{"comment":"The claimed invariance is not merely unproved but is false. Near N=0 with P>0, the prey equation in (1.4) behaves like dN ≈ (1-mP)N dt + sqrt(N(1+mP)) dB^1. Writing Y = sqrt(N) gives dY ≈ [(1-mP)Y/2 - (1+mP)/(8Y)] dt + (1/2) sqrt(1+mP) dB^1, whose -1/Y drift is that of a Bessel-type process with dimension less than 2. Feller's boundary classification then shows that 0 is an attainable boundary, so with positive probability N can hit zero in finite time. Hence P((N(t),P(t)) in D for all t>=0) = 1 cannot hold for general initial data, contradicting Theorem 4.5 and the persistence interpretation in Remark 4.6.","section":"Section 4, Theorem 4.5 statement"}],"minor_comments":[{"comment":"In the statement of Lemma 4.1, the equation is labelled (3.2), but it should be (4.2) to match the surrounding text.","section":"Section 4, Lemma 4.1"},{"comment":"In the definition of the infinitesimal generator, the domain of f should be R+×U, not R+×R+, since V is only defined on U.","section":"Section 4, Theorem 4.3"},{"comment":"The figure caption is internally inconsistent: it refers to panels (a), (b), (e) for the source case and (c), (d), (f) for the sink case, but then describes the trajectory as panel (c) and (f), while the text also mentions panel (e). The labels should be corrected.","section":"Section 5, Figure 2"},{"comment":"There are several typographical errors, including 'wiith' in Lemma 4.2 and 'mainfolds' in Section 3; a careful proofreading pass is needed.","section":"Throughout"},{"comment":"The stability and limit-cycle results for the deterministic Rosenzweig-MacArthur model are cited to an unpublished note [Gun12]; the authors should cite the original literature (e.g., Rosenzweig and MacArthur 1963, and Cheng 1981) for these classical results.","section":"References"}],"recommendation":"reject","confidential_remarks":"The central result of the paper, Theorem 4.5, contradicts classical Feller boundary theory for square-root diffusions. The issue is not a matter of presentation: the model can hit the zero boundary, so the claimed persistence is false. A substantial revision would be needed, likely changing the main claim to well-posedness on the closed domain [0,∞)^2 with absorbing boundary, rather than invariance of the open domain. The paper in its current form is not publishable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Rough take: the model is a legitimate new object of study, and the paper does a solid job of laying out the deterministic background and the explicit Lyapunov computations. But the central existence/invariance theorem is not supported: the proof of Theorem 4.3 asserts that a continuous compactly supported extension is globally Lipschitz, which is false (sqrt(N) on (0,n] is not Lipschitz), and the truncating sets K_n=(0,n]^2 are not precompact in D. More importantly, the Lyapunov function V=(1+N^2+P^2)^α only controls the far boundary; it is bounded on the strips near N=0 or P=0, so the estimate cannot rule out hitting the zero boundary. The square-root noise makes 0 an attainable boundary for the prey or predator density, so the claimed a.s. domain invariance is likely false. The Feller boundary test argument in the stress-test note is convincing: near N=0 with P fixed, the diffusion has the form of a Bessel-type process with accessible boundary. That is a load-bearing flaw, not a minor gap.\n\nWhat is genuinely useful: the paper frames a stochastic Rosenzweig-MacArthur model driven by internal demographic noise, computes the generator on V explicitly, and lists plausible Lp moment estimates. The deterministic part is a competent review. The general criterion in Theorem 4.3 is a modest variant of Lemma 4.1 from Abu91/Kha12, so novelty is limited but not zero. The numerical experiments are illustrative but do not test the zero-boundary issue.\n\nBottom line: the paper needs substantial revision before it can be taken seriously. A corrected version would need different truncating sets bounded away from zero, a Lyapunov function that blows up at the boundary, or a reformulation that allows extinction. As it stands, the main claim that D is invariant is not established and, on the Bessel-process heuristics, false. I would not cite the theorem, but the model and the deterministic summary might be worth a mention in related work. For peer review: yes, a serious editor should send it to a referee rather than desk reject; the flaw is subtle enough that a careful referee report could help the authors rewrite it properly.","headline":"The stochastic Rosenzweig-MacArthur model is a natural new object and the deterministic review is clean, but the central domain-invariance proof collapses on a false Lipschitz-extension step and a Lyapunov function blind to the zero boundary; the main claim is not established and likely false.","tokens_in":18084,"tokens_out":6557,"would_cite":false,"duration_ms":51820,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34D20","37H10","37N25"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that a stochastic Rosenzweig-MacArthur predator-prey model with internal demographic noise has a unique strong solution, and that the positive quadrant is invariant, so predator and prey densities remain positive for all…","keywords":["Rosenzweig-MacArthur model","internal demographic noise","stochastic differential equations","strong solution","domain invariance","Lyapunov function","predator-prey persistence","Lp estimates"],"falsifier":"Compute the quantity $\\inf_{(N,P)\\in D\\setminus K_n} V(N,P)$ for $V=(1+N^2+P^2)^\\alpha$ and $K_n=(0,n]^2$; since $D\\setminus K_n$ contains points such as $(\\varepsilon,n)$ with $\\varepsilon\\downarrow0$, the infimum stays near $1+n^2$ rather than going to infinity, so condition (ii) of Theorem 4.3 is not satisfied by the stated $V$. A direct check of whether the proof can be repaired, together with a pathwise simulation that stops at the first hit of zero, would settle whether the claimed invariance of $D$ actually holds.","tokens_in":17030,"feed_emoji":"🐾","tokens_out":7692,"duration_ms":41952,"temperature":0.7,"pith_summary":"This paper introduces a stochastic version of the Rosenzweig-MacArthur predator-prey model in which randomness comes only from internal demographic fluctuations, with noise intensity proportional to population densities. Its central claim is that for every positive initial condition the system has a unique strong solution and that the positive quadrant is invariant, so with probability one both densities remain positive for all finite times. To get there, the authors prove a general existence-and-uniqueness criterion for autonomous SDEs on an open submanifold of Euclidean space, using a Lyapunov function of the form $V=(1+N^2+P^2)^\\alpha$. They also prove $p$-th moment bounds and an almost-sure asymptotic estimate showing at most exponential growth. Taken together, the results are used to argue that internal noise does not have the same destabilizing effect as external environmental noise and does not lead to extinction.","feed_headline":"Unique solution keeps stochastic predator-prey densities positive","feed_subtitle":"Internal demographic noise alone is claimed not to drive either species extinct, with growth at most exponential.","key_machinery":"The load-bearing object is the Lyapunov function $V(N,P)=(1+N^2+P^2)^\\alpha$ with $\\alpha>2$, together with the localization criterion of Theorem 4.3. The criterion says that if the drift $\\mu$ and diffusion $\\sigma$ are Lipschitz on each truncating set $K_n$, and if there is a nonnegative $V$ with $LV\\le \\alpha V$ and $\\inf_{D\\setminus K_n}V\\to\\infty$, then a unique strong solution exists and the open domain $D$ is invariant. The noise coefficients $\\sigma_{11}=\\sqrt{N(1+N/k)+mNP/(1+N)}$ and $\\sigma_{22}=\\sqrt{cP+mNP/(1+N)}$ encode the internal-demographic-noise assumption that variance scales with population density, and it is this structure that makes the generator estimate $LV\\le C_1V$ possible. The Euler-Maruyama scheme is then used in simulations to compare stochastic paths with the deterministic limit cycle or stable equilibrium.","core_discovery":"The paper's central discovery is Theorem 4.5: for initial data in $D=\\{(N,P):N>0,P>0\\}$, the stochastic Rosenzweig-MacArthur system (1.4) admits a unique strong solution, and $D$ is invariant, meaning $P((N(t),P(t))\\in D)=1$ for every $t\\ge 0$. The proof rests on a new criterion (Theorem 4.3) for existence and uniqueness of autonomous SDEs on an open submanifold of $\\mathbb{R}^d$, which relaxes earlier localization requirements and permits Lyapunov functions such as $V=(1+N^2+P^2)^\\alpha$ with $\\alpha>2$. The diffusion coefficients are chosen so that the infinitesimal generator satisfies $LV\\le C_1V$, and the criterion then yields both global existence and invariance of the open domain. The paper also derives $p$-th moment estimates and an almost-sure Lyapunov-exponent bound $\\limsup_{t\\to\\infty}\\frac{1}{t}\\log\\|X_t\\|\\le C_1=\\frac{5+6m+c}{4}+\\frac{1}{2k}$, which it interprets as at-most-exponential population growth. The invariance of $D$ is emphasized as the rigorous content of biological nonnegativity and is used to conclude that internal stochasticity alone cannot cause extinction.","pith_inferences":["The theorem proves positivity for each fixed finite $t$; it does not by itself prevent $\\liminf_{t\\to\\infty}N(t)$ or $\\liminf_{t\\to\\infty}P(t)$ from being zero, so the paper's language about persistence goes a step beyond what the invariance theorem alone establishes.","A natural test of the claimed invariance is to classify the boundary behavior of the square-root diffusion coefficients: since the noise vanishes as $N$ or $P$ approaches zero, Feller or boundary-classification criteria could decide whether hitting the boundary in finite time is truly impossible or merely has probability zero under the constructed solution.","The same Lyapunov-localization template could be used to prove well-posedness for other predator-prey or epidemiological SDEs with state-dependent internal noise and a natural positive-invariant domain."],"forward_implications":["Starting from any positive initial densities, the solution stays in the positive quadrant for every finite time, so the model predicts that internal demographic fluctuations alone do not produce finite-time extinction.","The generalized existence-uniqueness criterion applies to other Lotka-Volterra-type systems posed on open submanifolds and permits Lyapunov functions that earlier localization criteria excluded.","The $p$-th moment and Lyapunov exponent estimates bound population growth at most exponentially, with explicit constants in terms of the model parameters.","For parameter regimes where the deterministic system has a stable limit cycle or a stable positive equilibrium, the stochastic paths fluctuate around those attractors while remaining positive."],"supporting_citations":[{"why":"Supplies the localization-type criterion for existence and uniqueness of strong solutions on an open domain with Lyapunov estimates and domain invariance, which Theorem 4.3 adapts.","marker":"[Abu91, Kha12]"},{"why":"Used for the extension lemma for continuous functions on closed subsets of a manifold, the gluing step in the localization proof.","marker":"[LL12]"},{"why":"Provides the standard localization argument that solutions of truncated SDEs agree up to exit times.","marker":"[Fri75]"},{"why":"Supplies the Lp moment estimates and the almost-sure Lyapunov exponent bound used in Propositions 4.8 and Theorem 4.9.","marker":"[Mao07]"},{"why":"Gives the deterministic boundedness and local stability or limit-cycle results for the Rosenzweig-MacArthur system that the stochastic analysis extends and compares against.","marker":"[Gun12]"}],"fun_headline_variants":["Stochastic predator-prey model keeps both species alive","Internal noise alone cannot drive predator-prey extinct","Unique strong solution preserves positive densities in stochastic ecology","New criterion proves well-posedness for stochastic Rosenzweig-MacArthur","Invariant positive domain: stochastic predator-prey model has unique solution"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The localization proof assumes that the Lyapunov function becomes large outside each truncating set, but the chosen function $V=(1+N^2+P^2)^\\alpha$ grows only when densities become large, not when a density approaches zero, so the proof does not actually rule out hitting the zero boundary.","fun_headline_variants_meta":{"raw":{"variants":["Stochastic predator-prey model keeps both species alive","Internal noise alone cannot drive predator-prey extinct","Unique strong solution preserves positive densities in stochastic ecology","New criterion proves well-posedness for stochastic Rosenzweig-MacArthur","Invariant positive domain: stochastic predator-prey model has unique solution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000268,"raw_usage":{"total_tokens":1695,"prompt_tokens":1101,"completion_tokens":594,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":717,"completion_tokens_details":{"reasoning_tokens":512}},"tokens_in":717,"tokens_out":594,"duration_ms":5233,"temperature":1.0,"reasoning_tokens":512,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:54:20.951828+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the quantity $\\inf_{(N,P)\\in D\\setminus K_n} V(N,P)$ for $V=(1+N^2+P^2)^\\alpha$ and $K_n=(0,n]^2$; since $D\\setminus K_n$ contains points such as $(\\varepsilon,n)$ with $\\varepsilon\\downarrow0$, the infimum stays near $1+n^2$ rather than going to infinity, so condition (ii) of Theorem 4.3 is not satisfied by the stated $V$. A direct check of whether the proof can be repaired, together with a pathwise simulation that stops at the first hit of zero, would settle whether the claimed invariance of $D$ actually holds.","supporting_citations":[],"review_version":1}