{"id":"05f0d494-da7a-4b35-85ce-a6aaf79638f9","arxiv_id":"1908.03827","paper_version":3,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A first-order weak-selection expansion of fixation probability is derived for arbitrary population structures and initial configurations, reducing the problem to solving a linear system of polynomial size.","lead":"This paper derives a general formula for the probability that a new trait takes over a population when the trait's effect is small. The formula works for arbitrary population structures and starting conditions, turning an exponentially hard calculation into a polynomial one.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; Theorem 2 is sound under its explicit smoothness assumption.","rationale":"The reader's verdict ACCEPT is appropriate. All load-bearing steps are explicitly justified: the amended-chain construction, the coalescent uniqueness argument, and the interchange of δ-derivative with the sojourn-time operator. The smoothness assumption is the most fragile premise but is an explicit, standard scope condition rather than an unsupported step. The paper also transparently flags the fixed-degree caveat for polynomial complexity. I therefore see no reason to alter the verdict.","tokens_in":31922,"tokens_out":32333,"duration_ms":360997,"concrete_test":"As a numerical consistency check, implement Theorem 2 for the Moran process with N=3 (or N=4) for each singleton initial state ξ={i}: compute c^k_{ij} from Eq. (11), solve (35) for η^ξ_I, evaluate (34), and compare the resulting d/ds ρ_A at s=0 with the exact derivative of (1-1/r)/(1-1/r^N) using r=1+s. A match confirms the index/sign conventions and the uniqueness argument for the restricted system.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the proof chain from Proposition 1 through Theorem 2. The sojourn-time/stationary-distribution identity (Eq. 18) is valid, the expansion in Theorem 1 follows from the exact identity ρ_A(ξ)=ξ̂+⟨Δ̂_sel⟩_ξ plus the vanishing of Δ̂_sel at δ=0, and the reduction to the linear system (35) is consistent: the coalescent chain restricted to subsets of size ≤D+1 has a simple eigenvalue 1, so normalization (35b) selects the unique solution. The weakest condition is indeed the smoothness of p_{R,α}(x) in δ (third assumption, Section 2.1). This is a genuine scope restriction: hard-threshold or sign-type selection rules are excluded because d/dδ e_{ij}(x) need not exist. However, the paper states this assumption explicitly and all applications use smooth fecundity maps (linear or exponential), so it is not a hidden flaw. The only caveat worth noting is that the 'polynomial complexity' claim requires the degree D to be fixed; the paper itself acknowledges this in Remark 1 and the Discussion.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a first-order weak-selection expansion for fixation probabilities in a broad class of finite-population evolutionary models with arbitrary spatial structure and frequency-dependent selection. The main result, Theorem 2 (Eq. 34), expresses the first derivative of the fixation probability ρ_A(ξ) at δ=0 in terms of reproductive values π_i, Fourier coefficients c_{ji}^I of the first-order effects of selection on replacement probabilities, and neutral sojourn times η^ξ_I that solve an explicit linear system (Eq. 35) on subsets of size at most D+1, where D is the degree of the process. This reduces the computation from exponential to polynomial complexity in N for fixed D. The paper also derives extensions to stochastic mutant appearance (Section 5), compares fixation probabilities of two types (Section 6), and applies the framework to constant-fecundity models on graphs with death-Birth and Birth-death updating (Section 7) and to the donation game on weighted and regular graphs (Section 8), recovering and generalizing several known results. The proof strategy uses a sojourn-time/stationary-distribution identity for an amended Markov chain (Proposition 1), a coalescent representation, and prior results from the authors' modeling framework.","tokens_in":32118,"tokens_out":14450,"duration_ms":146844,"significance":"If the results hold, this is a significant contribution to the mathematical theory of evolutionary dynamics. The central formula is explicit and computable: for any fixed degree D, the first-order weak-selection coefficient is obtained by solving a linear system of size O(N^{D+1}), rather than by analyzing the exponentially large state space. The applications are also valuable: they provide new weak-selection expansions for arbitrary initial configurations on weighted graphs and unify results of Chen et al. (2016), Allen et al. (2017), and others. The proofs of the central theorem are coherent and the paper is careful to state its assumptions, particularly the smoothness of the replacement rule in δ (Section 2.1). The main limitations—the smoothness requirement and the fixed-degree condition for polynomial complexity—are explicitly acknowledged. The paper does not supply machine-checked proofs or code, but the derivations are standard and reproducible from the stated equations. Overall, this is a solid and useful mathematical contribution that deserves publication after minor revisions.","major_comments":[],"minor_comments":[{"comment":"The definitions of the type-B mutant-appearance distribution use the same state notation as for type A, writing μ_B(1_i)=...; since μ_B is an initial distribution for B mutants in an all-A population, the paper's own subsequent identity E_{μ_B}[ξ_i]=1−μ_i shows that the intended support is the complementary state \\bar{1_i}. Please correct the notation consistently, as the literal reading contradicts the equations that follow.","section":"Section 5, Examples 1–3 (Eqs. 39–40, 54)"},{"comment":"The duality ρ_A(r;ξ)=1−ρ_A(r^{−1};ξ) and the averaging identity in Eq. (58) appear to omit the complement of the initial configuration on the right-hand side; restoring \\bar ξ makes Eq. (60) coherent, since ξ_I+\\bar ξ_I is then 1 when all individuals in I have the same type. The missing overbars should be added throughout this subsection.","section":"Section 7, Eqs. (57)–(60)"},{"comment":"The derivations of Propositions 2 and 3 are summarized only as 'a series of simplifications' of Eq. (59). Because these are advertised as new results for arbitrary initial states on weighted graphs, the intermediate algebra should be supplied, at least in an appendix or supplement, so that readers can verify the reduction from the general linear system (35).","section":"Section 7.1.1 and 7.1.2, Propositions 2 and 3"},{"comment":"The passage from Theorem 2's general η-system to the pair-level system (93) is compressed; in particular, the factors N/2 in Eq. (93a) and N in Eq. (93b) follow from the death-Birth coalescent recurrences but are not shown. A short derivation of (93a)–(93b) from Eq. (35) would make the application verifiable.","section":"Section 8, proof of Proposition 4"},{"comment":"The smoothness assumption on the replacement rule in δ is an explicit scope restriction: models with hard-threshold or sign-type selection rules, for which d/dδ e_{ij}(x) at δ=0 need not exist, are excluded. This is stated clearly and all applications use smooth fecundity maps, so I regard it as a limitation rather than a defect.","section":"Section 2.1, third modeling assumption"},{"comment":"The caption states that Γ2 is 'unequivocally better' for cooperation, but this comparison is made for a specific parameter choice (b=10, c=1) and depends on the magnitude of the first-order effect; the caption should state this parameter dependence explicitly to avoid overgeneralization.","section":"Section 8.3 and Figure 3"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript relies heavily on the authors' own prior framework (Allen and Tarnita 2014; Allen and McAvoy 2019), including the reproductive-value martingale property used in the proof of Theorem 2 and the deterministic fecundity mapping cited in Section 8. This is a normal research program and not a circularity concern, since the present paper's central derivation is otherwise self-contained. The only issues I found are notational and presentational. I would be comfortable with acceptance after the authors fix the complement notation and provide the requested derivations for the applied propositions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read the McAvoy–Allen paper on weak-selection fixation probabilities. The central result is Theorem 2: for a broad class of evolutionary models with arbitrary spatial structure, the fixation probability from any initial configuration expands to first order in selection intensity as ξ̂ + δ Σ π_i Σ c^I_ji (η^ξ_{i∪I} − η^ξ_{j∪I}) + O(δ²), where η solves a linear system of size O(N^{D+1}) for degree D. That reduces an exponentially hard problem to polynomial time for fixed D, and it generalizes several earlier results that only worked for regular graphs or uniform initialization. The donation-game result on arbitrary weighted graphs (Proposition 4) is new and subsumes Chen et al. 2016 and Allen et al. 2017.\n\nThe proof is solid. The sojourn-time/stationary-distribution identity (Proposition 1) is clean, and the uniqueness argument for η via the coalescent chain is convincing. I checked the step from Theorem 1 to Theorem 2 and it holds; the expansion of e_ij as a pseudo-Boolean function is standard but the bookkeeping is careful. The applications are concrete, and the regular-graph corollary reproduces Chen's known result, which is a good sanity check. The Section 8.3 caveat about comparing population structures via critical benefit-to-cost ratios is a genuinely useful caution.\n\nSoft spots are real but modest. The smoothness assumption on the replacement rule (third assumption, Section 2.1) is explicit, but it excludes hard-threshold or sign-type selection rules; that is a genuine scope restriction, not a hidden flaw. The 'series of simplifications' leading to Propositions 2–3 are compressed; I had to take some algebra on faith, but the results are consistent with known special cases. The polynomial complexity claim only holds for fixed degree D, which the paper acknowledges. And there are no explicit bounds on the O(δ²) remainder—standard in this literature, but worth flagging.\n\nThe paper leans on the authors' own framework (Allen–McAvoy 2019), which is legitimate since the new theorem is proven from stated assumptions, not by citing that framework's conclusions. Self-citation here is appropriate.\n\nBottom line: this deserves a serious referee and, in my view, acceptance. It is a genuine contribution that unifies and extends a large body of work in evolutionary graph theory. I'd cite it, and I'd bring it to our reading group.","headline":"A rigorous, general weak-selection expansion for fixation probabilities that unifies and extends prior results; deserves serious refereeing and likely acceptance.","tokens_in":32615,"tokens_out":2330,"would_cite":true,"duration_ms":23191,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92D15","60J20"],"pacs":[],"model":"deepseek-v4-flash","headline":"For any population structure, weak-selection fixation probability reduces to a linear system of polynomial size.","keywords":["fixation probability","weak selection","population structure","reproductive value","sojourn time","coalescent","evolutionary games on graphs","perturbation expansion"],"falsifier":"For a small smooth model (say a 4-cycle with death-Birth updating and the donation game), solve the exact $2^N$-state Markov chain for $\\rho_A(\\xi)$ at several small $\\delta$, compute the right side of Eq. (34) by solving the $\\eta$-system, and check that the difference is $O(\\delta^2)$; a difference linear in $\\delta$ would refute the expansion.","tokens_in":31725,"feed_emoji":"🧬","tokens_out":11850,"duration_ms":106741,"temperature":0.7,"pith_summary":"Evolutionary biologists ask whether a new mutant type is likely to take over a population, but the answer depends on spatial structure and on where the mutant first appears, and exact fixation probabilities are generally hard to compute. This paper proves that under weak selection—small mutant effect—the fixation probability from any initial configuration has a first-order expansion in the selection intensity $\\delta$, built from neutral reproductive values, neutral sojourn times, and first-order selection effects on replacement probabilities. The sojourn times solve a linear system of size polynomial in the population size $N$ whenever the process has bounded degree $D$, so the exponential cost of exact computation disappears. The paper uses the expansion to recover and unify earlier results for constant fecundity and for the donation game on graphs, and to compare population structures by the magnitude, not just the sign, of selection's first-order effect.","feed_headline":"Weak selection makes fixation probabilities polynomial to compute","feed_subtitle":"For any starting pattern and structure, one linear system of polynomial size gives selection's first-order effect.","key_machinery":"The argument is carried by three linked objects. An amended Markov chain that resets to the initial state $\\xi$ with probability $u$ after absorption lets Proposition 1 identify neutral sojourn times with the $u$-derivative of the amended stationary distribution at $u=0$. The sojourn-time quantities $\\eta^\\xi_I$ are the unique solution of the linear system (35), and they measure, under neutral drift from $\\xi$, the expected accumulation of $\\hat{x}-x_I$ before absorption. The selection coefficient $c_{ji}^I$ comes from the pseudo-Boolean (multi-linear polynomial) representation of the first derivative of the transmission probabilities, and the degree $D$ is the largest $|I|$ that appears; only $\\eta^\\xi_I$ for $|I|\\le D+1$ are needed. Uniqueness of $\\eta^\\xi$ follows from the set-valued coalescent chain, whose stationary distribution on singleton sets is exactly the reproductive-value vector $\\pi$.","core_discovery":"The central claim is Theorem 2: for any fixed initial configuration $\\xi$ in the paper's modeling framework, the fixation probability of type $A$ satisfies $$\\rho_A(\\xi) = \\hat{\\xi} + \\delta \\sum_{i=1}^N \\pi_i \\sum_{j=1}^N \\sum_{\\substack{I\\subseteq\\{1,\\dots,N\\}\\\\0\\le |I|\\le D_{ji}}} c_{ji}^I \\left(\\eta^\\xi_{i\\cup I} - \\eta^\\xi_{j\\cup I}\\right) + O(\\$delta^{2}$),$$ where $\\hat{\\xi}$ is the reproductive-value-weighted frequency of $A$, $\\pi_i$ are neutral reproductive values, $c_{ji}^I$ are the Fourier coefficients of the first derivative at $\\delta=0$ of the marginal trait-transmission probabilities $e_{ji}$, and $\\eta^\\xi$ is the unique solution of the neutral sojourn-time system (35). The degree $D$ of the process is the largest set size appearing in those coefficients. If $D$ is fixed, computing the first-order effect requires solving only $O(N^{D+1})$ linear equations rather than the exponential number of states in the full Markov chain.","pith_inferences":["The polynomial-size formulation suggests a practical screening strategy for large structured populations: for any fixed-degree update rule, solve Eq. (35) once and rank structures by the size of the first-order effect; the paper demonstrates such a comparison but stops short of proposing it as a general search over structures.","If the smoothness premise fails only mildly—say a kink at $\\delta=0$ with one-sided derivatives—one could still obtain a formal expansion using one-sided coefficients, but the paper's uniqueness and $O(\\delta^2)$ error bounds would need re-examination; this is an extension, not a claim of the paper.","The paper notes that the $N\\to\\infty$ and $\\delta\\to0$ limits do not commute; an inference is that Eq. (34) is best suited to finite-$N$ analyses, and a large-population diffusion analogue would require either bounding $D$ as $N$ grows or a new averaging argument."],"forward_implications":["For fixed degree $D$, the first-order weak-selection expansion is computable in $O(N^{3(D+1)})$ time (or $O(N^{2.373(D+1)})$ with fast matrix multiplication), eliminating exponential state enumeration for this class of models.","Proposition 4 gives the donation-game fixation probability from any initial configuration on any weighted connected graph, unifying previous regular-graph and uniform-initialization formulas from the literature.","Because the magnitude of the first-order coefficient is available, population structures can be compared by how much they boost cooperation at a given benefit-to-cost ratio, not only by the critical ratio; the paper shows these two rankings can disagree.","For uniform mutant appearance, the quantities $\\eta^{\\mu_A}_I$ equal $1/N$ times the expected coalescence time of the set $I$, connecting the expansion to standard coalescent branch lengths.","The framework also covers stochastic mutant-appearance distributions and relative fixation conditions for $A$ versus $B$, so the expansion applies beyond a fixed initial configuration."],"supporting_citations":[{"why":"supplies the general modeling framework of replacement rules in which all results are stated.","marker":"Allen and Tarnita (2014)"},{"why":"supplies the reproductive-value equations and the neutral martingale property used in the proof.","marker":"Allen and McAvoy (2019)"},{"why":"provides the theorem that identifies the stationary limit of the amended chain with fixation probabilities.","marker":"Fudenberg and Imhof (2006)"},{"why":"gives the uniqueness result for the pseudo-Boolean representation used to define the coefficients $c_{ji}^I$.","marker":"Hammer and Rudeanu (1968)"},{"why":"justifies the improved matrix-multiplication exponent quoted for solving the linear system.","marker":"Le Gall (2012)"},{"why":"is the arbitrary-configuration regular-graph donation-game formula that Corollary 4 reproduces.","marker":"Chen et al. (2016)"},{"why":"is the uniform-initialization donation-game result that Proposition 4 extends to arbitrary graphs.","marker":"Allen et al. (2017)"},{"why":"is the death-Birth constant-fecundity expansion under uniform initialization that the paper's formulas generalize.","marker":"Allen et al. (2020)"},{"why":"provides the single-mutant regular-graph result that the paper's weak-selection formulas extend.","marker":"Chen (2013)"}],"fun_headline_variants":["Weak selection fixes fixation odds with polynomial speed","For weak mutations, fixation probability becomes polynomial","Polynomial-time fixation probabilities under weak selection","Weak selection: exponential to polynomial in one step","Fixation probabilities simplified: weak selection polynomial"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The expansion rests on the assumption that every replacement-event probability is a smooth function of the selection intensity in a neighborhood of $\\delta=0$; if a model has a kink or threshold at zero, the first-order coefficients $c_{ji}^I$ do not exist and the formula cannot be applied.","fun_headline_variants_meta":{"raw":{"variants":["Weak selection fixes fixation odds with polynomial speed","For weak mutations, fixation probability becomes polynomial","Polynomial-time fixation probabilities under weak selection","Weak selection: exponential to polynomial in one step","Fixation probabilities simplified: weak selection polynomial"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000182,"raw_usage":{"total_tokens":1315,"prompt_tokens":952,"completion_tokens":363,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":296}},"tokens_in":568,"tokens_out":363,"duration_ms":4156,"temperature":1.0,"reasoning_tokens":296,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:01:22.494505+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a small smooth model (say a 4-cycle with death-Birth updating and the donation game), solve the exact $2^N$-state Markov chain for $\\rho_A(\\xi)$ at several small $\\delta$, compute the right side of Eq. (34) by solving the $\\eta$-system, and check that the difference is $O(\\delta^2)$; a difference linear in $\\delta$ would refute the expansion.","supporting_citations":[],"review_version":1}