{"id":"1e9fbf9e-03a5-4f46-ad31-004d78221314","arxiv_id":"2411.18576","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For a 1D random walk, the paper derives the joint distribution of first return time and number of distinct sites visited, including exact conditional means and variances.","lead":"This paper derives exact formulas for the joint statistics of how long a one-dimensional random walk takes to return to its starting point and how many distinct sites it visits in between. The results give closed-form expectations and variances for both quantities conditioned on the other, which is useful for understanding random search and foraging processes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sec. 7 asymptotic results for E[S|T] and Var(S|T) rest on numerical fitting rather than analytic derivation; the leading constants are plausible, but the subleading terms and the Discussion's variance expansion are unsupported.","rationale":"After independent checks of the exact conditional moments for small s (s=1, 2, 3) and of the normalizations, the combinatorial and generating-function part of the paper is solid. Eq. (7) is a standard bounded-Dyck-path count; the paper also provides U(n,n)=C_n and binomial representations (A.4)-(A.6), so the imported formula is not a real vulnerability. The genuine load-bearing issue is the asymptotic analysis of Section 7. The abstract's asymptotic claims E[S|T=2n] ~ sqrt(pi n) and Var(S|T=2n) ~ pi(pi/3-1)n are leading-order results that are supported by known Brownian excursion statistics, but the paper derives them by numerical fitting of exact sums, not by an analytic limit. Moreover, the subleading expansions in Eqs. (51) and (57) are fit-dependent, and Section 8's variance expansion is inconsistent with those equations: combining Eq. (51) with Eq. (57) cancels the sqrt(t) term, whereas Section 8 keeps -sqrt(pi/2) sqrt(t). This does not invalidate the leading claims, but it means the asymptotic part of the central claim is not fully established. The reader's CONDITIONAL verdict is therefore appropriate; I recommend no change to the verdict.","tokens_in":21233,"tokens_out":29572,"duration_ms":227725,"concrete_test":"Derive E[S|T=2n] and E[S^2|T=2n] analytically from the Theta distribution in Eq. (42) (or via singularity analysis of the generating function Es(y) in Eq. (10)); then compare the resulting coefficients with Eqs. (48, 51, 53, 57) and check whether Var(S|T=2n) has a sqrt(n) term as implied by Section 8. Alternatively, evaluate Eq. (47) and Eq. (52) at n=10^6 and n=10^7 with high-precision arithmetic and compute Var(S|T=2n) - pi(pi/3-1)n: if the remainder contains a term proportional to sqrt(n), the Discussion's expansion is right and Eqs. (51)/(57) are wrong; if not, the Discussion needs correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The exact conditional moments E[T|S=s] and Var(T|S=s) (Eqs. 32, 35) are derived from the generating function and check out for small s; the imported formula Eq. (7) is a standard bounded-Dyck-path count and is cross-checked against Eq. (A.4), so it is not the main vulnerability. The load-bearing weak point is Section 7: the asymptotic forms E[S|T=2n] ~ sqrt(pi n) (Eq. 48), E[S|T=2n] ~ sqrt(pi/2) sqrt(t) - 1/2 (Eq. 51), E[S^2|T=2n] ~ pi^2/3 n - sqrt(pi n) (Eq. 53), and E[S^2|T=t] ~ pi^2/6 t - sqrt(pi/2) sqrt(t) - pi^4/450 (Eq. 57) are obtained by evaluating the exact sums (47)/(52) at finite n and fitting Delta_1 and Delta_2 to assumed decay forms (Eqs. 50, 56). No analytic derivation from the exact generating function or from the Theta distribution (Eq. 42) is given. The leading constants are almost certainly correct (they match Brownian excursion results), but the subleading coefficients are fit-dependent. More seriously, Section 8 states Var(S|T=t) ~ (pi^2/6 - pi/2)t - sqrt(pi/2) sqrt(t) + O(1), which is inconsistent with combining Eqs. (51) and (57): the sqrt(t) terms cancel exactly, leaving no sqrt(t) term in the variance. This internal inconsistency shows the asymptotic expansion is not fully controlled.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the joint distribution of the first-return time T_FR and the number S of distinct sites visited before first return for a symmetric one-dimensional random walk. Starting from a bounded Dyck path count U(n,s), the authors express the joint distribution P(T_FR=2n,S=s), derive the conditional distributions P(T_FR|S=s) and P(S=s|T_FR=2n), and obtain closed-form conditional moments E[T_FR|S=s]=(2/3)(s^2+s+1) and Var(T_FR|S=s)=(4/45)(s-1)(s+2)(s^2+s-1). For the reverse conditioning, they evaluate exact binomial sums numerically and infer the asymptotics E[S|T_FR=2n] ~ sqrt(pi n), E[S^2|T_FR=2n] ~ (pi^2/3)n - sqrt(pi n), and hence Var(S|T_FR=2n) ~ pi(pi/3-1)n, with additional subleading terms. The paper presents simulation comparisons for the conditional distributions and moments.","tokens_in":21622,"tokens_out":12323,"duration_ms":106777,"significance":"If the exact derivations are correct, the paper advances the recent generating-function result of Klinger et al. by providing an explicit joint distribution and finite conditional moments. The closed-form expressions for E[T_FR|S=s] and Var(T_FR|S=s) are clean, check out for small s, and are supported by simulation, and the leading asymptotics for S|T_FR match known excursion and Theta-distribution results. The main caveat is that the subleading asymptotic terms in Section 7 are obtained by numerical fitting rather than by derivation from the exact sums or from Eq. (42), and Section 8 contains an internal inconsistency in the variance expansion. With those issues corrected, this would be a solid contribution to the random-walk and first-passage literature.","major_comments":[{"comment":"The asymptotic results are not derived analytically but are inferred by evaluating the exact sums (47) and (52) at finite n and fitting the residuals Delta_1 and Delta_2 to assumed decay forms (Eqs. (50) and (56)). No derivation is given from the exact combinatorial sums or from the Theta distribution in Eq. (42). The leading constants are plausible and agree with known results, but the subleading coefficients in Eqs. (51) and (57), which are advertised as new, are fit-dependent and not established. These terms should either be derived analytically or presented explicitly as numerical conjectures with a clear statement of their status.","section":"Section 7, Eqs. (48)-(57)"},{"comment":"The displayed asymptotic Var(S|T_FR=t) ~ (pi^2/6 - pi/2)t - sqrt(pi/2) sqrt(t) + O(1) is inconsistent with combining Eqs. (51) and (57). From those equations, E[S|t]^2 = (pi/2)t - sqrt(pi/2) sqrt(t) + 1/4 + ..., so the sqrt(t) terms cancel in Var(S|T_FR=t) = E[S^2|t] - E[S|t]^2, leaving Var(S|T_FR=t) ~ (pi^2/6 - pi/2)t - (pi^4/450 + 1/4) + ... . The stated sqrt(t) term in Section 8 therefore cannot be correct, and this internal inconsistency indicates that the subleading expansion is not fully controlled.","section":"Section 8, variance expansion"}],"minor_comments":[{"comment":"Eq. (39) appears to have an index-shift error: substituting U(n-1,s-1)-U(n-1,s-2) into Eq. (14) yields a prefactor 2^{2n-1} with arguments (n-1,s-1), (n,s-1), (n-1,s-2), (n,s-2), not the displayed F(n,s), F(n+1,s), F(n,s-1), F(n+1,s-1) with prefactor 2^{2n+1}. The current form is not equal to Eq. (37) and can take negative values, for example for n=2, s=1.","section":"Section 6, Eq. (39)"},{"comment":"The statements that for p<1/2 the walk is recurrent with P_R=1 and finite mean first-return time, and that for p>1/2 the return probability is (1-p)/p, are only correct for the process conditioned on the first step being to the right. Unconditioned, a biased one-dimensional random walk with p != 1/2 is transient. Please clarify that the conditioning on x_1=1 is maintained throughout this paragraph.","section":"Section 8, biased random walk discussion"},{"comment":"Eq. (54) cannot be valid for all moments as stated: for r=1 the formula gives zero, whereas E[S|T_FR=2n] ~ sqrt(pi n). The displayed leading-order formula should be restricted to r>=2, or supplemented by the separate r=1 result.","section":"Section 7, Eq. (54)"},{"comment":"Appendix B uses the word 'nodes' in one place where the rest of the paper uses 'sites'; please unify the terminology.","section":"Appendix B"}],"recommendation":"major_revision","confidential_remarks":"The exact conditional moment results are strong and likely sufficient to justify a revision rather than a rejection. The essential blocker is Section 7: the subleading asymptotics need either a genuine analytic derivation or a clear downgrade to numerical conjectures, and the Section 8 variance expansion must be corrected. I would be willing to look at a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on arXiv:2411.18576. The paper's real contribution is the explicit joint distribution P(T_FR=2n, S=s) and the closed-form conditional moments: E[T|S=s] = (2/3)(s^2+s+1) and Var(T|S=s) = (4/45)(s-1)(s+2)(s^2+s-1). Those are derived cleanly from the Dyck-path generating function, they check out for small s, and they match simulations. That's a solid, citable advance over Klinger et al., who stopped at the generating function.\n\nThe exact binomial-coefficient expressions for E[S|T=2n] and E[S^2|T=2n] in Appendix B are also useful and allow high-precision numerical evaluation. The leading asymptotics are plausible and consistent with known Brownian excursion results.\n\nThe soft spot is Section 7. The asymptotic results are obtained by evaluating the exact sums and fitting the deviations to assumed decay forms. That's fitting, not derivation. It's a legitimate way to conjecture the subleading terms, but those terms should be treated as conjectural until derived from the generating function or the Theta distribution. The stress-test note found a concrete inconsistency: combining (51) and (57), the sqrt(t) terms cancel, so the variance should be (pi^2/6 - pi/2) t + O(1), with no sqrt(t) term. Section 8's stated Var ~ (pi^2/6 - pi/2) t - sqrt(pi/2) sqrt(t) + O(1) contradicts that. So the subleading expansion is not controlled. Also, Eq. (54) is stated as valid for all r, but it gives 0 for r=1; it should be restricted to r>=2. Minor but worth fixing.\n\nThe imported formula Eq. (7) is standard and cross-checked in Appendix A; I don't see a circularity problem there.\n\nOverall: the main conditional-moment results are solid and new; the asymptotic section needs an analytic derivation of the subleading terms and a corrected variance statement. A serious referee should engage with it. Send to review, and ask for revision.","headline":"Solid closed-form conditional moments for first return and range in 1D, with a fit-based asymptotic section that needs an analytic fix before the subleading terms can be trusted.","tokens_in":22170,"tokens_out":2974,"would_cite":true,"duration_ms":25414,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G50","05A15","82C41","60J10"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper derives the exact joint distribution of first return time and number of distinct sites visited for a one-dimensional random walk, yielding closed-form conditional means and variances.","keywords":["random walk","first return time","Dyck paths","Catalan number","joint distribution","conditional moments","distinct sites visited","bounded Dyck paths"],"falsifier":"A brute-force enumeration of all first-return trajectories for $n$ up to about 20, tallying each by return time $2n$ and maximum height $s$, and comparing the normalized histogram with Eq. (8), would settle the central count; any mismatch for a single $(n,s)$ pair would falsify the joint distribution. The conditional moments could be checked independently by high-statistics simulation at fixed $s=8$.","tokens_in":20990,"feed_emoji":"🎲","tokens_out":8027,"duration_ms":66305,"temperature":0.7,"pith_summary":"A symmetric random walk on the integers returns to its starting point with probability one, but both the mean first return time and the mean number of distinct sites visited before that return are infinite. This paper shows that conditioning on one of these quantities makes the other finite and explicit: trajectories that visit exactly $s$ distinct sites before returning have a first return time with mean $\\frac{2}{3}(s^2+s+1)$ and variance $\\frac{4}{45}(s-1)(s+2)(s^2+s-1)$. Conversely, a trajectory whose first return occurs at time $2n$ typically visits about $\\sqrt{\\pi n}$ distinct sites, with variance growing linearly in $n$. These results are obtained from an exact formula for the joint distribution built from bounded Dyck paths, and they reproduce the known marginal laws as consistency checks. The practical upshot is that first-return search processes can be described by how much territory is covered before the walker comes back.","feed_headline":"Exact formula ties first return time to sites visited","feed_subtitle":"Conditioning on territory size makes one-dimensional walk returns finite and predicts sqrt-time coverage.","key_machinery":"The load-bearing object is the combinatorial count $T(n,s)=U(n,s)-U(n,s-1)$, where $U(n,s)$ is the number of bounded Dyck paths of length $2n$ that start and end at zero, never go below zero, and never exceed height $s$; a Dyck path is a walk with steps $\\pm1$ that stays nonnegative. Equation (7) expresses $U(n,s)$ as a finite trigonometric sum, and weighting the difference $T(n-1,s-1)$ by the path probability $2^{-(2n-1)}$ produces the joint distribution. The same generating function of $U(n,s)$ supplies the conditional moments through derivatives, and a binomial-coefficient reformulation makes the moments numerically accessible for very large $n$.","core_discovery":"The central discovery is an exact expression for $P(T_{\\mathrm{FR}}=2n,S=s)$, the probability that a simple symmetric nearest-neighbor walk on the integers, starting at the origin, first returns at time $2n$ after visiting exactly $s$ distinct sites. The paper writes this probability as $T(n-1,s-1)\\,2^{-(2n-1)}$, where $T(n,s)=U(n,s)-U(n,s-1)$ is the number of bounded Dyck paths of length $2n$ whose maximum height is exactly $s$. From this joint distribution it derives the conditional distributions $P(T_{\\mathrm{FR}}=2n|S=s)$ and $P(S=s|T_{\\mathrm{FR}}=2n)$, and from those the exact conditional mean and variance $\\mathbb{E}[T_{\\mathrm{FR}}|S=s]=\\frac{2}{3}(s^2+s+1)$ and $\\mathrm{Var}(T_{\\mathrm{FR}}|S=s)=\\frac{4}{45}(s-1)(s+2)(s^2+s-1)$, together with the asymptotic laws $\\mathbb{E}[S|T_{\\mathrm{FR}}=2n]\\simeq\\sqrt{\\pi n}$ and $\\mathrm{Var}(S|T_{\\mathrm{FR}}=2n)\\simeq\\pi(\\pi/3-1)n$. In this way the divergences of the marginal distributions are controlled, and the coupling between the duration and the spatial extent of a first-return excursion is quantified.","pith_inferences":["If the same conditioning strategy is applied to biased or resetting walks, whose joint generating functions are already known from previous work, explicit conditional moments of a similar polynomial or square-root form may follow by the same derivative method; this is a natural extension the paper notes but does not carry out.","The identification of the number of distinct sites with the maximum height is special to one dimension, so in higher dimensions a genuinely new geometric count is needed and the scaling $\\mathbb{E}[S|T]\\sim\\sqrt{T}$ may not survive.","For random-search models in which an agent resets or dies upon returning to the origin, the exact conditional variance quantifies the spread in search outcomes and suggests that the covered territory, rather than elapsed time, is the better control variable."],"forward_implications":["Conditioning on territory size removes the divergence: for every finite $s$, the conditional mean and variance of the first return time are exact polynomial functions of $s$.","For long excursions the number of distinct sites visited before the first return grows as $\\sqrt{\\pi n}$, so the explored region extends like the square root of the return time, with a smaller prefactor than the unconditioned mean because the walk must retrace its steps near the end.","The conditional distribution $P(T_{\\mathrm{FR}}=2n|S=s)$ has an exponential tail whose rate is set by $\\cos^2(\\pi/(s+1))$, so trajectories that cover more sites are not only slower on average but have return times with a wider spread.","The explicit joint distribution reproduces the known marginal laws, including $P(S=s)=1/[s(s+1)]$ and the Catalan-number first-return distribution, which anchors the new formulas to classical one-dimensional random walk results.","The binomial-coefficient representation lets the conditional moments be evaluated numerically up to $n=10^6$, making the asymptotic constants directly checkable."],"supporting_citations":[{"why":"It supplies the generating function of the joint distribution that this paper inverts into an explicit distribution for the first-return case.","marker":"[30]"},{"why":"It is the source of the trigonometric formula for $U(n,s)$, the bounded-Dyck-path count on which the joint distribution is built.","marker":"[45]"},{"why":"It provides a further derivation of the bounded Dyck path enumeration used in the combinatorial factor $T(n,s)$.","marker":"[31]"},{"why":"It supplies the asymptotic analysis of the conditional distribution and the leading moments that the paper refines with subleading terms.","marker":"[32]"},{"why":"It gives the known marginal law $P(S=s)=1/[s(s+1)]$ and the gambler's-ruin estimate used to check the joint distribution.","marker":"[1]"}],"fun_headline_variants":["Exact joint law links return time to distinct sites","Taming divergent walk returns with exact joint distribution","First return and coverage: exact conditional statistics","Exact joint distribution for walk returns and sites","1D walk: exact formula for return time and territory"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the imported count $U(n,s)$ in Eq. (7), namely that the trigonometric sum equals the number of length-$2n$ walks that stay nonnegative, return to zero, and never exceed height $s$.","fun_headline_variants_meta":{"raw":{"variants":["Exact joint law links return time to distinct sites","Taming divergent walk returns with exact joint distribution","First return and coverage: exact conditional statistics","Exact joint distribution for walk returns and sites","1D walk: exact formula for return time and territory"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000344,"raw_usage":{"total_tokens":2071,"prompt_tokens":1310,"completion_tokens":761,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":926,"completion_tokens_details":{"reasoning_tokens":688}},"tokens_in":926,"tokens_out":761,"duration_ms":7689,"temperature":1.0,"reasoning_tokens":688,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:08:13.669574+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A brute-force enumeration of all first-return trajectories for $n$ up to about 20, tallying each by return time $2n$ and maximum height $s$, and comparing the normalized histogram with Eq. (8), would settle the central count; any mismatch for a single $(n,s)$ pair would falsify the joint distribution. The conditional moments could be checked independently by high-statistics simulation at fixed $s=8$.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the generating function of the joint distribution that this paper inverts into an explicit distribution for the first-return case."},{"cited_title":"15 (New York: Academic Pr ess)","cited_arxiv_id":null,"evidence_quote":"It is the source of the trigonometric formula for $U(n,s)$, the bounded-Dyck-path count on which the joint distribution is built."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides a further derivation of the bounded Dyck path enumeration used in the combinatorial factor $T(n,s)$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the asymptotic analysis of the conditional distribution and the leading moments that the paper refines with subleading terms."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the known marginal law $P(S=s)=1/[s(s+1)]$ and the gambler's-ruin estimate used to check the joint distribution."}],"review_version":1}