{"id":"9ff417f4-1c62-43bb-94a6-313522acc131","arxiv_id":"2507.09325","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In discrete-time ecological models, environmental stochasticity can increase, decrease, or leave unchanged the expected stationary population size, depending on the model, the noisy parameter, and the deterministic baseline used for comparison.","lead":"Random environmental variation changes the long-run average population size in several classic ecology models. The paper shows the effect is not universal: noise can shrink, inflate, or leave the expected population unchanged, depending on the model, which parameter fluctuates, and the chosen no-noise baseline.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Small-noise predator-prey expansion uses an unproved global-stability conjecture; Section 4.4 is conditional on Streipert et al.'s conjecture for E*.","rationale":"The reader's weakest-assumption analysis identifies the same load-bearing concern: the small-noise predator-prey formulas rest on global asymptotic stability of E*, which the cited literature only conjectures. This is genuinely the least secure input to the paper's Section 4.4 claims, and the manuscript itself flags the gap. I have verified that the large-noise results—Beverton-Holt, Ricker, Hassell, and the two-species Ricker—are Jensen-based and internally sound, so the central claim that environmental stochasticity can decrease, increase, or leave unchanged the expected population size stands on its own. The Hassell large-noise result provides a robust example of an increase, and the Ricker random-r result provides a robust example of no change. The mechanical typos the reader notes (O(ρ^3) versus O(ρ^2), E ln b_i versus E b_i, and the missing factor 2 in Corollary 4.2's cross term) are real but do not alter the qualitative conclusions in the applications. I therefore see no reason to move the verdict away from CONDITIONAL: the paper's main message is supported, but the small-noise predator-prey section needs either a proof of global stability, a citation to one, or an explicit restriction to the local basin of E*.","tokens_in":22267,"tokens_out":33857,"duration_ms":366291,"concrete_test":"Run a deterministic basin-of-attraction test for map (32) over the parameter region d < γK < 1+2d, including the Section 4.4 slice d = r, Kγ = 2r, α/γ = p. Iterate the map for 10^5 steps from a fine grid of positive initial conditions, e.g. X0 ∈ [0.01K, 10K], Y0 ∈ [0.01 r/α, 10 r/α], and record the omega-limit. If any trajectory fails to converge to E* = (d/γ, r(γK−d)/(αγK)), then Theorem 4.1's global-attraction hypothesis is violated and the Section 4.4 expansion is only local. A fully affirmative numerical result would support the conjecture but would not prove it, so an independent Lyapunov-function proof would be the definitive analytic check.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.4 applies Theorem 4.1 to the interior predator-prey fixed point E* = (d/γ, r(γK−d)/(αγK)). Theorem 4.1 explicitly assumes the deterministic system has a globally attracting fixed point in R^n_{++}, and Appendix A (Assumption B3) likewise requires every deterministic trajectory to converge to an equilibrium. However, Section 3.3 states that Streipert et al. (2022) proved only local asymptotic stability of E* under d < γK < 1+2d, and that global stability is a conjecture. The paper neither proves this conjecture nor cites a proof. Consequently, the small-noise formulas around E* and the biological reading that \"noise increases the prey\" are justified only for the local basin of E*; they do not support convergence from every positive initial condition, which is what Theorem 4.1 promises. If the deterministic system has additional attractors, the stationary distribution reached under small noise may depend on the initial basin, and the displayed expansion may describe only one local invariant measure. A secondary but concrete issue: Corollary 4.2's expression for \\tilde A is missing the factor 2 on the cross term relative to Theorem 4.1's vec(A) equation; the model applications in Sections 4.1-4.2 appear to use the correct expression, but the stated corollary is internally inconsistent. These issues do not undermine the central abstract claim, which is independently supported by the large-noise Hassell result EN∞ > N and the Ricker random-r result EX∞ = K, but they do justify keeping the paper's small-noise predator-prey claims conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies expected population sizes at stationarity for stochastic discrete-time ecological models. Two tools are used: (i) the stationarity condition r_i(mu)=0 for per-capita growth rates under an invariant measure, combined with Jensen's inequality, to obtain inequalities for large environmental noise; and (ii) small-noise Taylor expansions from Cuello (2019) around deterministic fixed points to obtain O(rho^2) corrections to the mean. The large-noise results give: Beverton-Holt with random carrying capacity satisfies E X_infty < K together with a lower bound; Ricker with random K satisfies E X_infty = 1/E(1/K_1) < K; Ricker with random r satisfies E X_infty = K; Hassell with random K satisfies E N_infty > N-bar; two-species Ricker preserves the deterministic means; and a Streipert-type predator-prey model gives prey mean > d/gamma. The small-noise results treat Beverton-Holt and Hassell with two random parameters, showing that noise correlation can reverse the sign of the effect, and give an expansion for the predator-prey model. Simulations illustrate the main trends.","tokens_in":22613,"tokens_out":14292,"duration_ms":152601,"significance":"If the results are correct, the paper makes a useful contribution by showing rigorously that environmental fluctuations can decrease, increase, or leave unchanged the expected population size, depending on the model, on which parameters are randomized, and on the deterministic baseline chosen. The large-noise inequalities are elegant, essentially parameter-free consequences of stationarity and Jensen's inequality, and they identify the Hassell model as a counterpoint to the Cushing-Henson intuition. The paper also gives a careful discussion, following Chesson, of how the choice of no-noise baseline affects conclusions. The small-noise expansions provide explicit, falsifiable formulas for correlation effects. The main weaknesses are the unverified global-stability premise in the predator-prey small-noise application and an internal factor-of-two error in Corollary 4.2.","major_comments":[{"comment":"The small-noise predator-prey expansion is applied to the interior fixed point E* = (d/gamma, r(gamma K - d)/(alpha gamma K)), but Theorem 4.1 assumes that the deterministic system has a globally attracting fixed point in R^n_{++}. In Section 3.3 the authors state that Streipert et al. (2022) proved only local asymptotic stability under d < gamma K < 1 + 2d and conjectured global stability; Appendix A, Assumption B3, similarly requires every deterministic trajectory to converge to an equilibrium. The paper neither proves this conjecture nor rules out other attractors in the interior. Therefore the formulas in Section 4.4 are justified, at best, for initial conditions in the local basin of E* (as in Cuello's local Theorem A.1), not for every positive initial condition as Theorem 4.1 promises. The authors should either prove the global-stability conjecture, explicitly restrict the application to the local basin, or present the Section 4.4 formulas as conditional on Streipert et al.'s conjecture.","section":"Section 4.4 and Theorem 4.1"},{"comment":"The displayed expression for \\tilde A in Corollary 4.2 contains the cross term (partial F/partial xi_1)(partial F/partial xi_2) E(xi_1 xi_2) without a factor of 2. In Theorem 4.1, vec(A) = (I - D_x F \\otimes D_x F)^{-1}(D_e F \\otimes D_e F) vec(Cov(e)) sums over all ordered pairs (j1,j2), so the off-diagonal covariance contributes 2(partial F/partial xi_1)(partial F/partial xi_2) Cov(xi_1, xi_2). The direct second-derivative term in the same corollary has the correct factor 2, making the corollary internally inconsistent. The applications in Sections 4.1 and 4.2 appear to use the correct expression, but the stated corollary should be corrected.","section":"Corollary 4.2"}],"minor_comments":[{"comment":"The sentence 'the change in mean is O(rho^3)' should read O(rho^2), since Eq. (27) has an explicit rho^2 term; the following phrase also contains a typo ('te mean').","section":"Section 4, after Eq. (27)"},{"comment":"The formula 'r_i(delta_0) = E ln b_i(1)' should be 'E[b_i(1)]', since the log of the Ricker fitness is b_i(1), not the log of the coefficient.","section":"Section 3.2.1, before Eq. (18)"},{"comment":"The sentence 'Here we see how the correlations play an important role' is duplicated immediately before Section 4.1.","section":"Section 4, introduction"},{"comment":"The displayed expansion is for E(X(infinity)) + E(Y(infinity)), but the biological interpretation refers to the expected prey population. Since the expansion is for the sum, it does not by itself show that the prey marginal mean increases; the prey-increase claim is supported by Eq. (22) and by simulations, but the presentation should be clarified, and the missing parenthesis in 'E(Y (infinity)' should be fixed.","section":"Section 4.4"},{"comment":"The sentence 'If r > 1 we get that as t tends to infinity one has X_t to X_infinity in distribution' is too terse: the deterministic Ricker map is not globally stable for all r > 1, so the relevant range of r (or the precise stochastic persistence theorem) should be stated.","section":"Section 3.1.2"},{"comment":"The noise strength is called t and is also used as the time index elsewhere; this is confusing and should be renamed, for example to epsilon or sigma.","section":"Section 5.1"}],"recommendation":"major_revision","confidential_remarks":"The paper's central abstract claim is independently supported by the large-noise sections, in particular the Hassell, Ricker, and Beverton-Holt results, so the Section 4.4 global-stability issue, while serious, is fixable by reframing or by explicitly marking the result as conditional. The Corollary 4.2 factor-of-two error should be corrected before publication. The manuscript is suitable in scope for a mathematical biology or q-bio journal. No concerns about citation or novelty."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: the large-noise part of this paper is the real contribution, and it holds up. The small-noise section is conditional in places and has a few mechanical slips, but nothing that sinks the central message.\n\nWhat's new: using the zero-growth condition r_i(µ)=0 plus Jensen, they prove directional statements for expected stationary population size that were not in the literature. Specifically, stochastic Hassell with random K gives EN∞ > N̄ (noise helps), Ricker with random r gives EX∞ = K exactly (noise neutral), while Beverton-Holt with random K gives EX∞ < K (noise harmful). The Ricker random-r result and the Hassell increase are new as far as I can tell. The Hassell example is nicely framed as an artifact of baseline choice: compare with EK or E(1/K) and you reverse the sign. That's a clean and honest point, and they credit Chesson for the baseline ambiguity.\n\nThe two-species Ricker calculation is a neat application: linearity of log-fitness gives exact expectations and shows noise does not change the deterministic fixed point. The predator-prey large-noise bound X∞ > d/γ is simple and uses only Jensen. These results are transparent and reproducible—there's no hidden fitting.\n\nWhere it's soft: the small-noise predator-prey section (4.4) rests on Theorem 4.1's global-attracting fixed point assumption. But the deterministic model from Streipert et al. has only a proved locally stable interior fixed point under d<γK<1+2d; global stability is explicitly a conjecture in that paper. The authors cite it as a conjecture but then apply the global result as if it were proved. That makes the small-noise formulas around E* conditional on an unproved premise. They could fix this by stating the result as local (convergence from the basin of E*) or by citing a global proof if one exists. There are also smaller mechanical issues: in the two-species Ricker section, r_i(δ0) is written as E ln b_i(1) when it should be E[b_i(1)]; and Corollary 4.2's expression for à is missing a factor of 2 on the cross term relative to Theorem 4.1. These are typos, but the paper's small-noise section is largely asserted formulas without derivations or code, so the typos are harder to catch.\n\nThe simulations are illustrative but not fully reproducible: no code, no data, no error bars. That's not fatal for the theory but lowers confidence in the unproven simulation claims about large-noise predator-prey behavior.\n\nBottom line: the large-noise theorems are solid and worth citing. The small-noise section is not load-bearing for the abstract's main claim—that claim is independently supported by the Hassell and Ricker results. For a theoretical ecology audience, this deserves referee time. I'd advise a conditional accept path: require the authors to either prove or clearly localize the global-stability assumption in Section 4.4 and fix the algebraic slips.","headline":"Large-noise results are solid and novel; small-noise section has a load-bearing unproved assumption and some typos, but the paper deserves peer review.","tokens_in":23173,"tokens_out":3901,"would_cite":true,"duration_ms":38091,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92D25","92D40","60J05","37N25"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes that environmental stochasticity can decrease, increase, or leave unchanged the expected population size, depending on which parameters are random and on which deterministic system is taken as the no-noise baseline.","keywords":["stochastic population dynamics","expected population size","invariant probability measure","Beverton-Holt model","Ricker model","Hassell model","predator-prey model","small-noise expansion"],"falsifier":"Simulate the stochastic predator-prey model with initial prey density far above the carrying capacity and parameters where only local stability of the interior equilibrium is guaranteed; if the long-run sample mean of the prey does not approach the predicted $d/\\gamma$ plus the second-order correction, the global-attraction premise behind the small-noise result would be refuted.","tokens_in":22001,"feed_emoji":"🌦️","tokens_out":11932,"duration_ms":123452,"temperature":0.7,"pith_summary":"Environmental randomness does not have a single built-in effect on population abundance. This paper studies discrete-time population models at stationarity and shows that long-run expected population size can fall, rise, or stay equal to the deterministic equilibrium, depending on the model, on which parameters fluctuate, and on which deterministic system is chosen as the no-noise baseline. The proofs exploit the fact that per-capita growth rates are zero at stationarity, which turns the problem of comparing population sizes into inequalities like Jensen's inequality, and small-noise expansions are used where exact formulas are unavailable. The concrete payoffs include a strict decrease for Beverton-Holt and Ricker models with random carrying capacity, no change for Ricker with random growth rate, an increase for the Hassell model with random competition, and an increase in expected prey abundance for a predator-prey model.","feed_headline":"Random environments can raise, lower, or hold population size steady","feed_subtitle":"In classic models, the outcome depends on which parameters fluctuate and on your no-noise baseline.","key_machinery":"The load-bearing mechanism is the stationarity condition $r_i(\\mu)=0$ for the realized per-capita growth rate (invasion rate) of any species supported by an invariant probability measure $\\mu$. Since $r_i$ is the expectation of $\\ln f_i$ under the stationary distribution, setting it to zero yields equations that connect the stationary population mean to the moments of the random parameters; Jensen's inequality then converts these equations into inequalities between the stochastic mean and the deterministic equilibrium. For models where the stationary distribution cannot be solved, the paper applies a small-noise Taylor expansion around a globally attracting fixed point, which expresses the change in expected abundance as an explicit second-order correction involving the Jacobian, Hessian, and noise covariance of the dynamics.","core_discovery":"The paper's central claim is that environmental stochasticity can push the stationary expected population size in any direction, and the direction is controlled by the model structure, by the identity of the fluctuating parameters, and by the comparison baseline. For the Beverton-Holt model with i.i.d. random carrying capacity $K_t$, the stationary mean satisfies $E X_\\infty < K = E K_1$, so noise is always detrimental; for the Ricker model the same is true when the carrying capacity is random, but when only the growth rate $r_t$ fluctuates one gets $E X_\\infty = K$, so noise is neutral. For the Hassell model with random $K_t$, the stationary mean satisfies $E N_\\infty > \\bar N$, so noise is beneficial. The same reasoning applied to a two-species discrete Lotka-Volterra system shows the stationary means equal the deterministic stable fixed point, while for a predator-prey model the expected prey abundance satisfies $E X_\\infty > d/\\gamma$, exceeding its deterministic value $d/\\gamma$.","pith_inferences":["Beyond the paper, the zero-growth-at-stationarity trick suggests a general criterion for when noise helps or hurts in single-species maps, without solving for the invariant measure.","The small-noise results imply that the sign of the correlation between fluctuations in growth rate and carrying capacity is a measurable predictor of whether noise raises or lowers abundance; field studies could test this by estimating that covariance.","The baseline-dependence result suggests that empirical comparisons of stochastic and deterministic predictions should always report whether the deterministic baseline uses the arithmetic or harmonic mean of the fluctuating parameter.","For the predator-prey model, if global convergence to the interior fixed point ever fails, the predicted prey increase may only hold near equilibrium; the large-noise simulations in the paper already hint that the predator's response can reverse as noise grows."],"forward_implications":["In single-species models with a concave log-growth factor, random carrying capacity lowers expected abundance below the deterministic equilibrium, so management targets based on deterministic models may overestimate abundance.","In the Ricker model, the effect of noise disappears if only the growth rate fluctuates: expected abundance stays equal to the deterministic carrying capacity.","In the Hassell model, random competition raises expected abundance above the deterministic equilibrium, so noise can be beneficial for population size.","For two-species discrete Lotka-Volterra dynamics, the stationary expected abundances coincide with the deterministic globally stable fixed point, so noise has no effect on mean abundance.","In the predator-prey model, stochasticity increases expected prey abundance above $d/\\gamma$; whether the predator mean rises or falls depends on parameter values."],"supporting_citations":[{"why":"Conjectures that a periodic environment lowers the average population in the Beverton-Holt model, the deterministic analogue the paper revisits in a stochastic setting.","marker":"Cushing & Henson (2002)"},{"why":"Proves the stochastic Cushing-Henson result for Beverton-Holt, supplying the large-noise comparison that expected abundance falls below the mean carrying capacity.","marker":"Haskell & Sacker (2005)"},{"why":"Supplies the small-noise Taylor expansion around a globally attracting fixed point that powers Theorem 4.1 and all the second-order correction formulas.","marker":"Cuello (2019)"},{"why":"Provides the predator-prey model and the local stability condition for its interior fixed point, which serves as the deterministic baseline for the prey-abundance result.","marker":"Streipert et al. (2022)"},{"why":"Identifies the ambiguity in choosing the deterministic baseline, which the paper uses to reconcile opposite noise effects across different models.","marker":"Chesson (1991)"},{"why":"Supplies the ergodic small-noise convergence results underlying the expansion used for invariant-measure approximations.","marker":"Stenflo (1998)"},{"why":"Gives the stochastic persistence theorem that guarantees a unique invariant probability measure under invasion-rate conditions, used throughout for the large-noise results.","marker":"Hening et al. (2021)"}],"fun_headline_variants":["Noise can push population size up, down, or not at all","Environmental randomness flips expected population size","Population mean under noise: up, down, or unchanged","Model structure decides how noise shifts population size","Stochastic ecological dynamics: noise's mixed effect on size"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The small-noise predator-prey expansions require the deterministic model to converge to its interior equilibrium from every starting population, whereas the underlying model only establishes local convergence near that equilibrium; if far-away trajectories escape to another attractor, the predicted mean shifts need not hold.","fun_headline_variants_meta":{"raw":{"variants":["Noise can push population size up, down, or not at all","Environmental randomness flips expected population size","Population mean under noise: up, down, or unchanged","Model structure decides how noise shifts population size","Stochastic ecological dynamics: noise's mixed effect on size"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000445,"raw_usage":{"total_tokens":2242,"prompt_tokens":928,"completion_tokens":1314,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":1237}},"tokens_in":544,"tokens_out":1314,"duration_ms":12845,"temperature":1.0,"reasoning_tokens":1237,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:02:10.287161+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the stochastic predator-prey model with initial prey density far above the carrying capacity and parameters where only local stability of the interior equilibrium is guaranteed; if the long-run sample mean of the prey does not approach the predicted $d/\\gamma$ plus the second-order correction, the global-attraction premise behind the small-noise result would be refuted.","supporting_citations":[],"review_version":1}