{"id":"7a234214-0bd3-4d98-9ff9-cdf993459d03","arxiv_id":"2412.05247","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Derives Feynman-Kac equations for occupation time statistics of continuous-time random walks with arbitrary waiting times, recovering arcsine and Lamperti distributions and adding resetting.","lead":"This paper derives equations for how long a non-Markovian random walker spends inside a region, with and without stochastic resetting. The framework unifies known occupation-time laws and can be applied to animal foraging and other search processes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central FK equations are derived in the small-jump (diffusion) limit, not for arbitrary CTRW; validity for jump length comparable to interval width is unstated and untested.","rationale":"The paper is a serious theoretical contribution: the memory-kernel representation of the CTRW, the backward Feynman-Kac equation, and the solved characteristic functions are internally consistent and correctly reproduce the Brownian arcsine law and known fractional Feynman-Kac results in the appropriate limits. The reader's weakest assumption, the diffusion approximation, is indeed the single most load-bearing concern. Although the paper mentions the continuum limit at the start of Section II, the abstract and introduction claim generality for 'generic random walkers' and 'arbitrary rest periods' without stating that the jump-length distribution is restricted to its second moment. Since Eq. (19) is derived by expanding in sigma up to order sigma^2, all subsequent formulas inherit this limitation. The practical risk is that practitioners applying these results to lattice CTRWs with jump lengths not negligible compared to the domain size may obtain incorrect occupation-time statistics. The proposed test directly checks whether this concern is realized. I also examined the reader's other reservations: Eq. (52), when read as Q(Ta,t) ~ Aalpha (t-Ta)^{alpha/2}(2+Ta/(t-Ta))/Ta^{1+alpha}, does reduce to Eq. (48) in the limit Ta -> t, so that concern appears to be a misreading of the typeset formula. The resetting central-limit statement is asserted rather than proved, but the numerical evidence and the renewal argument in Section VI.C make it plausible; it is not as load-bearing as the diffusion approximation. Overall, the reader's CONDITIONAL verdict remains appropriate: the paper is correct in its stated diffusion-limit regime, but the scope of the central claim should be clarified and tested near the limit of validity.","tokens_in":29426,"tokens_out":37858,"duration_ms":327748,"concrete_test":"Simulate the exact CTRW used in Fig. 1 with the fat-tailed waiting-time PDF Eq. (10) for alpha = 0.4, sigma = 0.01, and t ~ 10^7, then compare predictions with three interval half-widths: a = 0.055 (sigma/a ~ 0.18, as in the paper), a = 0.01 (sigma/a = 1), and a = 0.005 (sigma/a = 2). If the mean occupation time Eq. (34) or the PDF Eq. (52) deviates by more than a few percent as sigma/a approaches unity, the diffusion approximation is the operative restriction on the central claim and the paper should explicitly qualify its scope to the continuum limit.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that Eqs. (19), (23), and (57) hold for a CTRW with arbitrary waiting-time distribution is only established in the continuum limit where the jump length distribution is expanded to second order. In Section II, Eq. (6) uses Psi(k) ~ 1 - sigma^2 k^2/2, and Appendix A, Eq. (A4), expands Q(p,s|x0 ± sigma) to O(sigma^2). This requires sigma to be much smaller than all relevant length scales, especially the interval half-width a. For the half-occupation problem, x0 = 0 is the boundary, and the second-order expansion is applied at a point where U(x0) is discontinuous, so lattice-scale effects such as finite jump-over probability and finite residence at x = 0 are neglected. The paper does not state a validity condition such as a >> sigma, and its own simulations use sigma/a ~ 0.18, which is not asymptotically small. If sigma becomes comparable to a, the predicted occupation-time statistics need not match exact CTRW statistics. This is the most load-bearing limitation because it affects the derivations of every subsequent moment, PDF, and ergodicity statement. A secondary issue is that Eq. (74) is misprinted: the correct coefficient multiplying <Z(s+r)>^2 is 2r(s+r)/s, not 2r/[s(s+r)], although later results in Section VI.A appear to use the correct form.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a backward Feynman-Kac formalism for occupation-time statistics of continuous-time random walks (CTRWs) with arbitrary waiting-time distributions, using the generalized master equation (GME) with a memory kernel K(s). The authors derive characteristic functions for the occupation time in a finite interval (Eq. 23) and for the half-line occupation time (Eq. 57), then compute moments, limiting PDFs, and ergodicity-breaking parameters for normal, power-law, and Mittag-Leffler waiting times. They also extend the results to Poissonian stochastic resetting, obtaining long-time Gaussian and delta-peak limits for the occupation-time PDFs. All analytical predictions are compared with numerical simulations for several parameter regimes.","tokens_in":29667,"tokens_out":4447,"duration_ms":44210,"significance":"If correct, the paper provides a unified generating-function framework for occupation-time statistics of non-Markovian random walks, connecting the memory kernel to the Feynman-Kac equation and recovering known results (arcsine law, Lamperti distribution) as special cases. The derivation is self-contained and the paper ships explicit, machine-checkable analytic formulas for moments in Laplace space, together with extensive numerical verification for multiple waiting-time distributions. The novel resetting results and the universality claims for finite-mean waiting times are of interest to the statistical-physics community. The main limitation is that the derivation is performed in a small-jump (diffusion) approximation, which is not clearly stated as a validity condition.","major_comments":[{"comment":"The central derivation is not valid for an arbitrary CTRW, but only in the continuum limit where the jump-length distribution is expanded to second order, Ψ(k) ≈ 1 − σ²k²/2, and Q(p,s|x0±σ) is expanded to O(σ²). This requires σ to be much smaller than the interval half-width a and, for the half-occupation problem, requires the starting point x0 = 0 not to lie exactly on the discontinuity of U(x0). The paper does not state this validity condition, and its own simulations use σ/a ≈ 0.18 (a = 0.055, σ = 0.01), which is not asymptotically small. Please state explicitly that Eqs. (19), (23), and (57) hold for σ/a → 0 (and for x0 not at a boundary), and discuss the expected error at finite σ/a, ideally with a numerical test for larger σ.","section":"Section II, Appendix A, Eqs. (6) and (A4)"},{"comment":"The claim that Eq. (52) reduces to Eq. (48) when Ta ∼ t is incorrect. Setting Ta = t − ε with ε ≪ t in Eq. (52) gives Q ∼ Aα/(ε^{1+α/2} t^α), whereas Eq. (48) gives Q ∼ Aα/(ε^{1−α/2} t^α). The discrepancy is a factor (t−Ta)^{−α/2} (or equivalently the exponents of ε differ by α). Thus the 'central bulk' formula does not match the right-edge formula in the overlapping regime, and the statement 'if Ta ∼ t ... Eq. (52) reduces to Eq. (48)' is false. This must be corrected: either Eq. (52) is wrong, or the reduction claim is wrong, and the numerical agreement shown in Figure 5 should be re-examined in the right-edge region.","section":"Section IV.B, Eqs. (48) and (52)"},{"comment":"The left-edge PDF (43) is asserted to hold for Ta ≪ (a/√Kα)^{2/α}, which is a very narrow region near y = 0. However, the derivation of Eq. (43) uses the intermediate formula Q(p,s) ≃ 1/p + 2 e^{−a p^{α/2}/√Kα}/(s^{1−α/2} p^{α/2}) and then Laplace inverts term by term. The validity of this double-Laplace inversion approximation, and its consistency with the normalization of Q(Ta,t), are not discussed. Please provide a more precise statement of the error terms or a numerical check of the normalisation of Eq. (43).","section":"Section IV.B, Eq. (43) and Appendix D"}],"minor_comments":[{"comment":"Equation (74) contains a typographical error: the coefficient multiplying ⟨Z(s+r)⟩² should be 2r(s+r)/s, not 2r/[s(s+r)]. The subsequent long-time result in Eq. (75) and the variance formula (76) are consistent with the corrected coefficient, so this appears to be a simple misprint, but it should be fixed.","section":"Section VI.A, Eq. (74)"},{"comment":"The jump-length distribution is expanded to second order in Fourier space, but the notation σ is introduced as a 'characteristic dispersal distance' without discussing the case of asymmetric jump distributions. For an asymmetric distribution with nonzero first moment, the expansion should include a drift term; please clarify that the analysis assumes symmetric jumps (or state the generalization).","section":"Section II, Eq. (6)"},{"comment":"The Lamperti distribution in Eq. (59) is written with a denominator y^α + (1−y)^α + 2y^{α/2}(1−y)^{α/2} cos(απ/2), while the standard Lamperti form often appears with a factor sin(απ/2)/π times y^{α/2−1}(1−y)^{α/2−1} divided by the same denominator. Please verify the normalization and the convention for α; the current expression appears to be missing a factor of 2 in the denominator when compared to Ref. [14].","section":"Section V, Eq. (59)"},{"comment":"The series expansion in Eq. (D3) has a summand (−a T_a^{−α/2} a√Kα)^n, which appears to contain a repeated factor a; please check whether this should be (−a T_a^{−α/2}/√Kα)^n or a different expression.","section":"Appendix D, Eq. (D3)"}],"recommendation":"major_revision","confidential_remarks":"The paper is substantial and likely publishable after revision, but the diffusion-limit caveat and the bulk/right-edge mismatch in Section IV.B are load-bearing issues that need to be resolved. The self-citation of Ref. [28] is not problematic because the present derivation does not rely on it as an input. If the authors can fix Eq. (52) or remove the incorrect reduction claim, and add a clear validity condition for σ/a, the paper would be a strong contribution to the field."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid, workmanlike paper. The main object is Eq. (19), a backward Feynman-Kac equation written through the memory kernel K(s)=sφ(s)/(1−φ(s)) for a functional of a CTRW in the small-jump/diffusion limit. Solving it for interval and half-line occupation times and adding Poissonian resetting produces a lot of formulas, and the simulations back the main asymptotic moments and PDFs. I believe the central formulas are correct in the regime the derivation actually assumes: σ small compared with a.\n\nThe reader's concern about the half-line boundary is worth taking seriously. Appendix A expands in σ around the starting point assuming x0 ≫ σ, then the paper evaluates at x0 = 0, where U is discontinuous. The known arcsine and Lamperti recoveries reassure me, but the abstract's promise about generic non-Markovian random walks should be qualified as a continuum/diffusion limit. That is the main caveat, and it is a stated-assumption problem rather than a fatal numerical error.\n\nWhat is genuinely new: expressing occupation-time statistics through K(s) unifies fractional Feynman-Kac and resetting cases; the interval characteristic function and the resetting variance formulas are not in the older literature. The self-citation to Ref. [28] is used only for comparison, so that does not bother me.\n\nSoft spots, in decreasing order. (1) Eq. (52) does not reduce to Eq. (48) when Ta ∼ t; the promised matching is off by powers of Ta/(t−Ta), and the scaling form (53) does not follow from (52) as printed. This is a localized but real inconsistency in the bulk/right-edge section. (2) Eq. (74) has a typo in the cross term; the later long-time limit and the simulations confirm the correct coefficient, so this is a typo, not a conceptual error. (3) The resetting CLT argument is heuristic — the paper itself says a rigorous proof is beyond scope — yet the conclusions say it was proved. The Gaussian PDF is well supported by simulations, so this is an overstatement of rigor, not a wrong result.\n\nNo code or data are included, but the methods are detailed enough for re-implementation.\n\nBottom line: the paper deserves a serious referee. It will be useful to people working in CTRW occupation statistics and movement ecology. The fixes needed are: qualify the diffusion limit, correct or adjust Eq. (52), and soften the CLT claim. I would accept after those revisions and would cite it for the K(s) machinery.","headline":"A useful memory-kernel toolkit for occupation times in the continuum CTRW limit, with real but localized holes around the right-edge asymptotics and the half-line boundary expansion.","tokens_in":30238,"tokens_out":6842,"would_cite":true,"duration_ms":67999,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K40","82C41","60J60"],"pacs":["05.40.Fb"],"model":"deepseek-v4-flash","headline":"A single memory kernel determines occupation-time statistics for arbitrary non-Markovian random walks.","keywords":["occupation time","continuous-time random walk","Feynman-Kac equation","non-Markovian random walks","stochastic resetting","arcsine law","memory kernel","anomalous diffusion"],"falsifier":"Simulate a CTRW with a jump-length distribution of infinite second moment, such as Cauchy jumps, and compare the interval-occupation-time PDF with Eq. (23); the predictions should fail near the interval edges because the $\\Psi(k)\\simeq 1-\\sigma^2 k^2/2$ closure no longer holds. A sharper check is to take a finite-variance jump distribution with strongly non-Gaussian shape and verify that Eq. (23) still matches the simulation once $a\\gg\\sigma$, which would confirm the second-moment closure is the operative condition.","tokens_in":29204,"feed_emoji":"⏱️","tokens_out":6755,"duration_ms":64694,"temperature":0.7,"pith_summary":"This paper aims to give a unified description of occupation time—the total time a random walker spends inside a given region up to time $t$—for continuous-time random walks with arbitrary waiting-time distributions between jumps. It derives a backward Feynman-Kac equation for the characteristic function of any time-integrated functional and solves it for two observables: the occupation time in a symmetric interval and the half-occupation time on the positive line. The resulting formulas, Eqs. (23) and (57), are written entirely in terms of a memory kernel $K(s)=s\\varphi(s)/(1-\\varphi(s))$, so Markovian, power-law, and Mittag-Leffler waiting times all fall out of one expression. From these formulas the paper obtains moments, probability-density tails, limiting distributions such as the arcsine and Lamperti laws, ergodicity-breaking parameters, and the modifications produced by stochastic resetting.","feed_headline":"Occupation-time law for arbitrary waiting times derived","feed_subtitle":"One backward Feynman-Kac equation delivers moments, arcsine law, and resetting behavior for Markovian and anomalous walkers.","key_machinery":"The load-bearing object is the memory kernel $K(s)=s\\varphi(s)/(1-\\varphi(s))$, the Laplace-domain encoding of the waiting-time distribution. The derivation starts from the Montroll-Weiss propagator, closes the jump distribution at second order in Fourier space ($\\Psi(k)\\simeq 1-\\sigma^2 k^2/2$), and converts the exact renewal equation into the backward Feynman-Kac equation (19): $sQ-1=(\\sigma^2/2)K(s+pU(x_0))\\,\\partial^2 Q/\\partial x_0^2-pU(x_0)Q$. Solving that ordinary differential equation in the spatial variable with the indicator $U$ of the interval or the half-line produces the characteristic functions (23) and (57), and the same kernel appears in the resetting renewal relation (68).","core_discovery":"The paper's central discovery is that for a continuous-time random walk whose jumps have finite second moment $\\sigma^2$, the generating function $Q(p,s)$ of the occupation time is determined by the memory kernel alone: for the interval $[-a,a]$, $Q(p,s)$ solves the boundary-value problem built on Eq. (19) and yields Eq. (23), and for the half-line it yields Eq. (57). The paper then claims that all long-time statistical properties follow from these closed forms. In particular, the half-occupation time has mean $t/2$ for every isotropic walk, its limiting distribution is the arcsine law (the classic U-shaped distribution for the fraction of time spent on one side) whenever waiting times have finite first moment, and the Lamperti distribution when the power-law exponent satisfies $0<\\alpha<1$. Interval-occupation moments show distinct regimes in $\\alpha$, and under Poissonian resetting the same kernels predict that the occupation-time fraction becomes Gaussian at intermediate times and collapses to a delta function at its mean at long times, restoring ergodicity at the level of the occupation time even though the reset-free walk is non-ergodic.","pith_inferences":["The same backward equation should apply to other additive functionals such as local time, area under the path, or first-passage functionals by changing $U$, suggesting the paper's machinery is a template for a wider class of non-Markovian functional statistics.","The paper notes that near $\\alpha=1$ its moment results diverge and leaves that case aside; a separate boundary-layer analysis could reveal logarithmic corrections in the occupation-time moments.","The authors conjecture that any time-integrated functional under Poissonian resetting converges to a delta at its mean; testing this with nonlinear or unbounded functionals would show whether the mechanism is the renewal structure or a special property of bounded occupation indicators.","Since the diffusion approximation requires $a\\gg\\sigma$, the formulas are inherently long-wavelength results; lattice or strongly intermittent walks may need finite-$\\sigma$ corrections."],"forward_implications":["For any waiting-time distribution with finite first moment, the half-occupation-time PDF approaches the arcsine law; only when the first moment diverges ($0<\\alpha<1$) does the Lamperti distribution replace it.","The mean interval-occupation time in the $1<\\alpha<2$ power-law regime matches normal diffusion, but the second moment separates off until $\\alpha>3/2$, so the universality class depends on which moment one measures.","Reset-free occupation times are non-ergodic for every waiting-time distribution considered, with explicitly computed ergodicity-breaking parameters.","With Poissonian resetting, occupation-time fluctuations become Gaussian with variance linear in time, and the long-time PDF is a delta function centered on the mean, making the time-averaged occupation ergodic.","Because all results are expressed through $K(s)$, inserting any other waiting-time PDF into Eqs. (23) and (57) immediately yields that model's occupation-time statistics."],"supporting_citations":[{"why":"Supplies the Montroll-Weiss propagator from which the memory-kernel form and the generalized master equation are obtained.","marker":"[29]"},{"why":"Prior fractional Feynman-Kac equations for subdiffusive functionals that this work generalizes to arbitrary non-Markovian waiting times and that provide the interval-occupation comparison.","marker":"[13,14]"},{"why":"Establishes the Brownian-functional Feynman-Kac framework and the half-occupation-time setting the paper extends.","marker":"[6]"},{"why":"Gives the renewal relation (68) linking the reset-free generating function to the generating function under Poissonian resetting.","marker":"[8]"},{"why":"Provides Levy's arcsine law, the limiting distribution the paper recovers for finite-mean waiting times.","marker":"[15]"},{"why":"Supplies the Lamperti distribution used for the half-occupation-time limit in the $\\alpha<1$ power-law regime.","marker":"[39]"},{"why":"Recent results for Brownian occupation time under resetting that the paper reproduces and generalizes as special cases.","marker":"[28]"}],"fun_headline_variants":["Non-Markovian walks: occupation time law solved","Feynman-Kac unlocks occupation time for arbitrary waits","Arcsine law revisited for non-Markovian and reset walks","Occupation time statistics unified for all waiting times","Resetting restores ergodicity in occupation time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The formulas assume jumps have finite second moment and that the interval width is large compared with a typical jump, so the expansion of the jump distribution to second order in Fourier space is legitimate; if that expansion fails, the characteristic functions no longer describe the exact continuous-time random walk.","fun_headline_variants_meta":{"raw":{"variants":["Non-Markovian walks: occupation time law solved","Feynman-Kac unlocks occupation time for arbitrary waits","Arcsine law revisited for non-Markovian and reset walks","Occupation time statistics unified for all waiting times","Resetting restores ergodicity in occupation time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000212,"raw_usage":{"total_tokens":1405,"prompt_tokens":917,"completion_tokens":488,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":533,"completion_tokens_details":{"reasoning_tokens":408}},"tokens_in":533,"tokens_out":488,"duration_ms":4885,"temperature":1.0,"reasoning_tokens":408,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:49:56.266876+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a CTRW with a jump-length distribution of infinite second moment, such as Cauchy jumps, and compare the interval-occupation-time PDF with Eq. (23); the predictions should fail near the interval edges because the $\\Psi(k)\\simeq 1-\\sigma^2 k^2/2$ closure no longer holds. A sharper check is to take a finite-variance jump distribution with strongly non-Gaussian shape and verify that Eq. (23) still matches the simulation once $a\\gg\\sigma$, which would confirm the second-moment closure is the operative condition.","supporting_citations":[{"cited_title":"Sadhu, M","cited_arxiv_id":null,"evidence_quote":"Supplies the Montroll-Weiss propagator from which the memory-kernel form and the generalized master equation are obtained."},{"cited_title":"Campos and V","cited_arxiv_id":null,"evidence_quote":"Gives the renewal relation (68) linking the reset-free generating function to the generating function under Poissonian resetting."},{"cited_title":"Singh and A","cited_arxiv_id":null,"evidence_quote":"Provides Levy's arcsine law, the limiting distribution the paper recovers for finite-mean waiting times."},{"cited_title":"Abramovitz and I","cited_arxiv_id":null,"evidence_quote":"Supplies the Lamperti distribution used for the half-occupation-time limit in the $\\alpha<1$ power-law regime."},{"cited_title":"Singh and A","cited_arxiv_id":null,"evidence_quote":"Recent results for Brownian occupation time under resetting that the paper reproduces and generalizes as special cases."}],"review_version":1}