{"id":"8bac642b-9186-4eae-89bd-9b009808b517","arxiv_id":"2505.12170","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A combinatorial model of Pólya's walker yields recurrence formulas that extend to complex edge weights, with explicit effective bounds in two dimensions.","lead":"This paper recasts Pólya's random walk recurrence as a combinatorial counting problem and extends it to walks with complex weights, where probabilities become sums of complex numbers. A generalist might read it to see how a 1921 probability theorem can be pushed outside probability using generating functions.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorems 5.21 and 5.22 need Bh(1)≠0 but prove it only via an unstated hypothesis Ch(1) exists; the v-recurrence extension is incomplete.","rationale":"The reader's verdict of CONDITIONAL is appropriate. My independent read finds the same soft spot: the v-recurrence theorems are not fully proved because Bh(1)≠0 is derived from an unstated hypothesis, namely the existence of Ch(1). The central 0-recurrence theorem (5.18) is solid: it explicitly assumes Ch(1) exists, and the proof uses Corollary 5.15 correctly. The conditional nature of the existence of boundary sums is stated honestly as a hypothesis, not hidden, so it is a scope limitation rather than a flaw. The square-root ambiguity in v-recurrence is transparent and acknowledged by the statement '∈ sqrt(...)'. The Bh(1)≠0 gap is a genuine proof gap, but it is localized and likely fixable by adding an assumption or supplying a new argument that does not require Ch(1). Therefore the paper should remain CONDITIONAL rather than being rejected or accepted unconditionally. I agree with the reader's weakest_assumption, which explicitly identifies the gap in Theorems 5.21 and 5.22.","tokens_in":27102,"tokens_out":15470,"duration_ms":140713,"concrete_test":"Attempt to re-derive the step 'Bh(1)≠0' in Theorems 5.21 and 5.22 using only the stated hypotheses (existence of Bh(1), B0,h(1), C0,h(1), plus v-transitivity and convexity/lightness). If the derivation requires Ch(1) to exist, add 'Ch(1) exists' or D|h|(1)<∞ to the hypotheses and re-check the proof; this determines whether the defect is a missing hypothesis or a false inference. As a secondary check, try to construct a v-transitive convex light weight with Bh(1)=0 and B0,h(1)=0 satisfying the other assumptions; if one exists, the theorems as stated are false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Both Theorem 5.21(2) and Theorem 5.22(2) divide by Bh(1), so they require Bh(1)≠0. The proof asserts this by citing 'the proof of part 1 of Theorem 5.17'. That proof derives Bh(1)≠0 from Corollary 5.15, which needs both U(1) and V(1) to exist for the formal identity V(x)=1-1/U(x). In that application U=Bh and V=Ch, so the citation requires Ch(1) to exist. However, Theorem 5.21(2) only assumes Dh(1), C0,h(1), B0,h(1), Bh(1), A0,h(1) exist, and Theorem 5.22(2) only assumes Bh(1), B0,h(1), C0,h(1) exist; neither includes Ch(1). Thus the cited argument does not apply. If Bh(1)=0, the equation Bh(1)=B0,h(1)+C0,h(1)^2 Bh(1) would force B0,h(1)=0, which is not contradictory for complex weights (the constant term 1 can be cancelled by later terms). The paper itself proves Proposition 5.20 at the 'high level of SFA' with fewer details, so this is not a purely cosmetic gap. The 0-recurrence extension (Theorem 5.18) is unaffected because it explicitly assumes Ch(1) exists, but the v-recurrence extension is not established as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a combinatorial model of Pólya's random walker on Z^d, counting walks rather than assigning probabilities, and then extends the classical recurrence theorems to a model W_C in which edges carry complex weights. After proving the classical 0- and v-recurrence theorems in generating-function form (Theorems 3.9 and 3.11) and giving an effective version for d=2 in Section 4, the paper introduces in Section 5 the notions of light, convex, and v-transitive weights on the complete graph on N. The main results are Theorem 5.18, which states that for a convex light complex weight with Bh(1) and Ch(1) existing one has lim_n h(W_rec(n)) = 1 - 1/Bh(1), and Theorems 5.21 and 5.22, which give set-valued or square-root formulas for v-recurrence under analogous existence assumptions. The proofs rely on a semi-formal approach to generating functions using absolute convergence, grouping of series, and Abel's theorem.","tokens_in":27382,"tokens_out":13288,"duration_ms":130155,"significance":"If the results hold, the paper offers an original combinatorial framing of Pólya recurrence and a genuinely non-probabilistic generalization to complex weights. The 0-recurrence extension appears sound, and the effective bounds of Section 4 are a concrete contribution. The paper is also careful to state the conditional nature of the complex-weight results. However, the v-recurrence theorems 5.21 and 5.22 are not established as stated: their proofs use an argument for Bh(1) ≠ 0 that requires an existence assumption — Ch(1) — not present in the theorem statements. This gap is load-bearing because the square-root formula and the limits in those theorems require division by Bh(1). The overall framework and the 0-recurrence results are nevertheless defensible, so the issue is repairable within the manuscript's scope.","major_comments":[{"comment":"The proofs of Theorems 5.21(2) and 5.22(2) both assert that Bh(1) ≠ 0 and divide by Bh(1). In each case the proof refers to the proof of Theorem 5.17: Theorem 5.22 says 'part 2 of Theorem 5.17', which is presumably a typo for 'part 1'. The proof of Theorem 5.17(1) obtains Bh(1) ≠ 0 from Corollary 5.15, and Corollary 5.15 requires both U(1) = Bh(1) and V(1) = Ch(1) to exist. Theorem 5.21(2) assumes only Dh(1), C0,h(1), B0,h(1), Bh(1) and A0,h(1) exist, and Theorem 5.22(2) assumes only Bh(1), B0,h(1) and C0,h(1) exist; neither assumes Ch(1). The cited argument is therefore not applicable. Moreover, if Bh(1) = 0, the identity in Proposition 5.20(2) gives only B0,h(1) = 0, which is not contradictory for complex weights. The conclusion Bh(1) ≠ 0, and hence the square-root formula and the resulting limit statements, are unproven. The authors should either add Ch(1) to the hypotheses (the sufficient condition D|h|(1) < +∞ already guarantees it) or supply a direct proof of Bh(1) ≠ 0 from the stated hypotheses.","section":"Theorems 5.21(2) and 5.22(2), with proofs"},{"comment":"Proposition 5.20 is proved at a deliberately low level of detail, with the proof saying that it argues 'on the high level of SFA'. This proposition supplies the three generating-function identities on which Theorems 5.21 and 5.22 depend after taking boundary values. Given that the rest of Section 5 proves the analogous identities in Proposition 5.8 with full details, please expand the proof of Proposition 5.20 to the same level, or at least state explicitly which applications of Propositions 5.2, 5.4 and 5.5 are used in each decomposition. As written, the reader cannot fully verify the formal identities for general complex weights.","section":"Proposition 5.20"}],"minor_comments":[{"comment":"The proof refers to 'part 2 of Theorem 5.17' when deriving Bh(1) ≠ 0; it should refer to 'part 1 of Theorem 5.17'.","section":"Proof of Theorem 5.22(2)"},{"comment":"The abstract and the preview on page 6 state that Theorems 5.21 and 5.22 extend Theorem 3.11 to W_C. In light of the missing Ch(1) hypothesis, it would be prudent to phrase this claim conditionally on the additional hypothesis or on an independent proof of Bh(1) ≠ 0.","section":"Abstract and introduction preview"},{"comment":"Corollary 4.6 asserts an effective constant N0 without giving its value. Since Propositions 4.3 and 4.5 give explicit numerical bounds, it would be useful to state, at least in principle, how N0 is computed, or to include the resulting explicit constant.","section":"Section 4"},{"comment":"The notation U(R) is used for both the nonnegative series evaluation in Definition 3.2 and the complex evaluation in Definition 5.11; the two definitions allow different values (+∞ only in the former). A sentence noting the intended meaning at each use would reduce ambiguity.","section":"Definitions 5.11 and 5.6"}],"recommendation":"major_revision","confidential_remarks":"The central 0-recurrence results appear sound and the overall framework is interesting, but the v-recurrence extension currently contains a proof gap that affects two of the four main theorems. I would not recommend rejection, because adding Ch(1) to the hypotheses or closing the gap with a direct argument is likely feasible. Please also ask the authors to expand the proof of Proposition 5.20 so that the dependence on the semi-formal method is explicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: worth refereeing. The reader's conditional verdict is about right, and the stress-test note lands. This is not a takedown: the combinatorial reformulation of Pólya is clean, and the effective two-dimensional bounds in Section 4 are a genuine plus. The complex-weight model in Section 5 is a plausible extension, and the 0-recurrence theorems (5.17, 5.18) are proven under explicit existence assumptions. The generating-function decompositions in Propositions 5.8 and 5.20 are correct in spirit; the SFA label is softer than a formal method, but the specific bijections are visible.\n\nThe soft spot is Theorem 5.21(2) and Theorem 5.22(2). Both divide by Bh(1) and claim Bh(1)≠0 by citing the proof of part 1 of Theorem 5.17. That proof uses Corollary 5.15, which needs both U(1) and V(1) to exist for the identity V=1−1/U. Here U=Bh and V=Ch. Neither 5.21 nor 5.22 assumes Ch(1) exists, so the citation does not do the work. It is possible that convexity plus v-transitivity rules out Bh(1)=0 by a different argument, but the paper doesn't give one. Since the formal relation Bh=B0,h+C0,h^2Bh would only force B0,h(1)=0 when Bh(1)=0, that outcome is not contradicted by anything in the hypotheses. This is a real gap, not a cosmetic one. The 0-recurrence theorems are unaffected because 5.18(1) explicitly assumes Ch(1) exists.\n\nMinor things: Proposition 5.20 is sketched at the high level of SFA rather than proven; I think it can be formalized with the same weight-preserving bijections, but as written it is less than a full proof. The promised follow-up papers are fine to cite but shouldn't be load-bearing. The citation pattern is otherwise unremarkable, and there are no fitted constants or circular steps.\n\nFor whom: people interested in combinatorial proofs of Pólya recurrence and in how far the generating-function machinery goes when probabilities are replaced by complex weights will get value from this. I would send it to a serious referee. A good referee will ask for a repair of 5.21/5.22 — either add Ch(1) as a hypothesis or prove Bh(1)≠0 directly. If that cannot be done, the v-recurrence extension should be downgraded to a conjecture or restricted. The rest of the paper can likely be published after modest revision.","headline":"Genuinely useful combinatorial framework and effective d=2 bounds, but the complex v-recurrence theorems carry an unproved Bh(1)≠0 step that the current hypotheses don't support.","tokens_in":27919,"tokens_out":5703,"would_cite":true,"duration_ms":55935,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G50","60J10","05A15","30B10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Pólya's recurrence theorem is not a probabilistic fact: the same limits hold when edges carry arbitrary complex weights, with the answer read off from the generating function of closed walks.","keywords":["Pólya's random walk","recurrence to a vertex","complex weights","light weight","convex weight","generating functions","semi-formal symbolic method","random walk on Z^d"],"falsifier":"Take the one-dimensional birth-death chain on positive integers with complex step weights $p=0.75+0.1i$ to the right and $q=0.25-0.1i$ to the left, so $p+q=1$; set the start vertex's only outgoing weight to $1$. This weight is convex and light. Compute the weights $a^h_n$ of length-$n$ walks that revisit the start by dynamic programming for $n$ up to $10^5$. Theorem 5.18 predicts $a^h_n\\to 1-1/B_h(1)=q/p\\approx 0.310-0.175i$; if the computed sequence converges to any other value or fails to converge, the central claim is refuted.","tokens_in":26893,"feed_emoji":"🎲","tokens_out":14010,"duration_ms":131484,"temperature":0.7,"pith_summary":"This paper asks which part of Pólya's classic walker result is genuinely probabilistic. In a combinatorial model that counts length-$n$ walks on $\\mathbb{Z}^d$, it first turns the recurrence theorems into exact formulas: for $d\\ge 3$ the limiting proportion of walks returning to the origin is $1-1/B(1)$, where $B(x)$ is the generating function of closed walks, and visits to a vertex $v$ give $\\sqrt{1-B_0(1)/B(1)}$. The main step then passes to a model $\\mathrm{W}_{\\mathbb{C}}$ in which every edge of a countable graph carries an arbitrary complex weight, subject only to lightness: at each length $n$, the total weight of all length-$n$ walks converges absolutely. For convex light weights (each vertex's outgoing weights sum to $1$), the recurrence limit is still $1-1/B_h(1)$, and for nonnegative weights with $B_h(1)=\\infty$ it is $1$. The upshot is that recurrence is a property of the walk-counting structure itself, not of the axioms of probability.","feed_headline":"Pólya's walker returns even with complex weights","feed_subtitle":"Replacing edge probabilities by complex weights still yields the classical Pólya limits, via generating functions.","key_machinery":"The machinery is the model $\\mathrm{W}_{\\mathbb{C}}$ of light weights $h\\colon \\mathbb{N}^2\\to\\mathbb{C}$ (for each $n$, the total weight of all length-$n$ walks converges absolutely), together with a semi-formal approach to generating functions that extends the symbolic method from finite to countable sets. The argument runs on two combinatorial identities, $A_h(x)=C_h(x)D_h(x)$ and $B_h(x)=1/(1-C_h(x))$, where $B_h$ counts closed walks and $C_h$ counts closed walks whose interior avoids the start. For convex weights, $D_h(x)=1/(1-x)$, and Abel's theorem turns the formal identities into boundary-value formulas such as $\\lim_n a^h_n=1-1/B_h(1)$.","core_discovery":"The central discovery is that the recurrence of Pólya's walker is a formal-power-series identity that survives arbitrary complex edge weights. In the complex-weighted model, the weight $a^h_n$ of length-$n$ walks that revisit their starting point satisfies the same combinatorial decompositions as the classical counts: $A_h(x)=C_h(x)D_h(x)$ and $B_h(x)=1/(1-C_h(x))$, with $B_h$ the generating function of closed walks. A semi-formal extension of the symbolic method to countable sets, justified by absolute convergence and Abel's theorem, converts these identities into evaluated limits. Convexity—each vertex's outgoing weights sum to $1$—makes $D_h(x)=1/(1-x)$ and yields $\\lim_n a^h_n=1-1/B_h(1)$ whenever $B_h(1)$ exists; for nonnegative $h$ with $B_h(1)=\\infty$ the limit is $1$. For visits to a vertex $v\\neq 1$, the same machinery locates the limit among the two square roots of $1-B_{0,h}(1)/B_h(1)$, with $B_{0,h}$ the closed-walk generating function avoiding $v$.","pith_inferences":["Because the limits are boundary values of generating functions, one can tune complex weights to interpolate between recurrent and transient behavior: a finite $B_h(1)$ acts as a transience parameter, and signed weights allow cancellations with no probabilistic counterpart.","The two-valued square root in the $v$-recurrence theorem suggests an underlying phase or topological index in the complex weights; selecting the correct root will likely require extra hypotheses such as nonnegativity or a fixed argument, beyond the paper's present assumptions.","The semi-formal approach to generating functions may transfer to other countable combinatorial classes (trees, lattice configurations, self-avoiding walks) wherever absolute convergence lets finite product rules be replaced by countable ones.","A direct numerical check on a biased birth-death chain with complex step weights would test the formula: with step weights $p$ (right) and $q=1-p$ (left), the predicted limit is $q/p$ whenever $|q/p|<1$, which can be compared with dynamic-programming computation of $a^h_n$."],"forward_implications":["For the classical lattice walker, the strengthened formulas give explicit limits in $d\\ge 3$: the return proportion tends to $1-1/B(1)$, and the visit proportion for a vertex $v$ tends to $\\sqrt{1-B_0(1)/B(1)}$.","In two dimensions the return theorem becomes effective: for all sufficiently large $N$, the deficit from the limit $1$ is sandwiched between $(0.9\\log N)^{-1}$ and $(0.1\\log N)^{-1}$.","The combinatorial and Markov-chain models record the same probabilities: $\\Pr(\\bigcup_{n\\ge1} X_n=v)=\\lim_n(2d)^{-n}|W_d(v,n)|$.","Any convex complex weight with finite $B_h(1)$ yields a well-defined recurrence limit $1-1/B_h(1)$; nonnegative convex weights with infinite $B_h(1)$ are recurrent with limit $1$.","The $v$-recurrence limit for general complex weights is only determined up to sign, as one of two square roots, a genuinely new ambiguity that cannot appear in the probabilistic setting."],"supporting_citations":[{"why":"The original 1921 theorem on the probability that the walker ever visits a given node; the result being extended to complex weights.","marker":"[22]"},{"why":"Supplies Abel's theorem for complex power series, used to pass from formal generating-function identities to evaluated limits.","marker":"[28]"},{"why":"The analytic-combinatorics treatise whose symbolic method the paper extends from finite to countable sets.","marker":"[9]"},{"why":"Provides the formal product formula for generating functions that the semi-formal approach generalizes.","marker":"[11]"},{"why":"Gives the asymptotic $\\|v\\|^{2-d}$ for the return probability in dimension $d\\ge3$, identifying the limiting constants in the strengthened theorems.","marker":"[19]"},{"why":"Supplies the effective Stirling bounds used in the two-dimensional effective version of the return theorem.","marker":"[26]"}],"fun_headline_variants":["Complex weights don't break Pólya's walker recurrence","Random walk recurrence survives complex edge weights","Pólya's walker: complex weights, same returns","Beyond probability: Pólya recurrence with complex weights","Complex-weight Pólya walker still returns"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The formulas require the total complex weight of all closed walks to be a well-defined finite or infinite sum; with arbitrary complex weights this sum can diverge or depend on how the walks are ordered, and the paper's conclusions then fall back to weaker statements.","fun_headline_variants_meta":{"raw":{"variants":["Complex weights don't break Pólya's walker recurrence","Random walk recurrence survives complex edge weights","Pólya's walker: complex weights, same returns","Beyond probability: Pólya recurrence with complex weights","Complex-weight Pólya walker still returns"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000593,"raw_usage":{"total_tokens":2774,"prompt_tokens":938,"completion_tokens":1836,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":1760}},"tokens_in":554,"tokens_out":1836,"duration_ms":10911,"temperature":1.0,"reasoning_tokens":1760,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:39:41.781355+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the one-dimensional birth-death chain on positive integers with complex step weights $p=0.75+0.1i$ to the right and $q=0.25-0.1i$ to the left, so $p+q=1$; set the start vertex's only outgoing weight to $1$. This weight is convex and light. Compute the weights $a^h_n$ of length-$n$ walks that revisit the start by dynamic programming for $n$ up to $10^5$. Theorem 5.18 predicts $a^h_n\\to 1-1/B_h(1)=q/p\\approx 0.310-0.175i$; if the computed sequence converges to any other value or fails to converge, the central claim is refuted.","supporting_citations":[{"cited_title":"Polya, ¨Uber eine Aufgabe der Wahrscheinlichkeitsrechnung betreffend die Irrfahrt im Strassennetz, Math","cited_arxiv_id":null,"evidence_quote":"The original 1921 theorem on the probability that the walker ever visits a given node; the result being extended to complex weights."},{"cited_title":"Tenenbaum, Introduction to Analytic and Probabilistic Number Theory","cited_arxiv_id":null,"evidence_quote":"Supplies Abel's theorem for complex power series, used to pass from formal generating-function identities to evaluated limits."},{"cited_title":"Flajolet and R","cited_arxiv_id":null,"evidence_quote":"The analytic-combinatorics treatise whose symbolic method the paper extends from finite to countable sets."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the formal product formula for generating functions that the semi-formal approach generalizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the asymptotic $\\|v\\|^{2-d}$ for the return probability in dimension $d\\ge3$, identifying the limiting constants in the strengthened theorems."},{"cited_title":"Robbins, A remark on Stirling’s formula, Amer","cited_arxiv_id":null,"evidence_quote":"Supplies the effective Stirling bounds used in the two-dimensional effective version of the return theorem."}],"review_version":1}