{"id":"6bb6a097-3225-4931-9743-808e559381b3","arxiv_id":"2501.07704","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Rare-event tails of time-integrated velocity powers in Ornstein-Uhlenbeck processes are shown to arise from a single dominant excursion, via an excursion-based random walk mapping.","lead":"This paper shows that rare, huge fluctuations of the time-integrated velocity power A = ∫ v^n dt in an Ornstein-Uhlenbeck process are caused by a single extreme excursion of the velocity, not by many small fluctuations adding up. The authors map the continuous process to a random walk of independent excursions and use the big jump principle to reproduce known rare-event tails from a new physical picture.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The v0→0 limit in Eq. (37), which converts the excursion CTRW into a statement about the actual OU process, is asserted rather than proven; if subleading v0-dependent terms survive in the tail of f_{v0}(A), the cancellation yielding Eq. (17) is not established.","rationale":"The paper's quantitative tail result (17) is consistent with prior instanton calculations and is supported by the numerical agreement in Figure 6, so a REJECT would be unwarranted. However, the derivation of this tail from the big jump principle depends critically on Eq. (37), where the v0→0 regularization of the infinitely many zero crossings is used to define a CTRW with a finite mean renewal time. The cancellation of the linear v0 prefactor in f_{v0}(A) against the v0 in ⟨τ(v0)⟩ is a leading-order statement; the paper gives no uniform-in-v0 control over the large-A tail. If corrections of the type v0^{1+α} A^{β} survive in the joint limit, the prefactor in Eq. (17) would change even though the exponent remains equal to the instanton result. The reader's weakest_assumption identifies exactly this limit interchange, and I agree that it is the most load-bearing unproven step. The concern is about rigor and completeness rather than the physical correctness of the tail, so the appropriate verdict remains CONDITIONAL: the big-jump mechanism is plausible and numerically supported, but the claimed derivation is not yet fully established.","tokens_in":33690,"tokens_out":16094,"duration_ms":169643,"concrete_test":"Simulate the first-passage area f_{v0}(A) for n=7 (γ=1, σ=1) at several small v0 values (10^{-2}, 10^{-3}, 10^{-4}) and evaluate the ratio R(v0,A) = f_{v0}(A) / [v0 exp(-γ^{(n+2)/n} c_n A^{2/n}/σ²)] along a sequence with A = v0^{-p}, p>0 (e.g., p=1 and p=2), so that A→∞ and v0/A^{1/n}→0. If R tends to a finite nonzero constant as v0→0, the v0-cancellation in Eq. (37) is robust; if R diverges or vanishes, the limit interchange fails and the big-jump tail (17) is not established. A complementary check is to compare the T coefficient in P(A,T) from direct simulation of the OU process with the prediction T/⟨τ(v0)⟩ f_{v0}(A) at small v0.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central result (17) rests on Eq. (37): P(A,T) ≍ lim_{v0→0} (T/⟨τ(v0)⟩) f_{v0}(A), with ⟨τ(v0)⟩ ∼ (√π/(σ√γ)) v0 (19) and f_{v0}(A) ∼ v0 exp{ -γ^{(n+2)/n} c_n A^{2/n}/σ² } (34). The v0 factors cancel and the T factor is the only trace of the CTRW. However, the OU process crosses zero infinitely often in any finite interval, as the paper itself states in Section VI, so the 'renewal process' is not literally a sequence of first-passage excursions from v0>0; it exists only as the v0→0 limit of a regularized restart process. The tail (34) is obtained from the large-deviation ansatz (26), which fixes the leading exponential but not the uniformity of the v0→0 limit. If the exact tail has a subleading correction of the form v0^{1+α} A^{β} exp(...), or if the effective number of independent attempts differs from T/⟨τ(v0)⟩ after the limit, Eq. (17) would acquire a prefactor or an altered coefficient. No proof of this limit interchange is supplied; Section V simply asserts that for very large A the excursion area distribution is independent of v0.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the distribution of the time-integrated observable A = ∫_0^T v^n(t) dt for an Ornstein-Uhlenbeck process. The authors construct a CTRW representation based on excursions between zero crossings, regularized by starting each excursion at v0 > 0 and taking v0 → 0. From this representation they derive the Gaussian bulk of P(A,T) via an Einstein relation, compute the diffusion coefficient D_n in closed form, and then analyze the first-passage excursion area distribution f_{v0}(A) through the large-deviation ansatz in Eq. (26). They obtain the rate-function equation (29), identify a critical point associated with the noiseless solution, and extract the anomalous tail f_{v0}(A) ~ v0 exp{-(γ^{(n+2)/n} c_n / σ^2) A^{2/n}}. Applying the single-big-jump principle then yields the central result P(A,T) ≍ T exp{-(γ^{(n+2)/n} c_n / σ^2) A^{2/n}} for n > 2, matching earlier instanton results while offering a big-jump interpretation.","tokens_in":34041,"tokens_out":7901,"duration_ms":80571,"significance":"If the derivation is made rigorous, the paper gives an appealing physical picture: the anomalous stretched-exponential tail of a time-integrated observable is produced by one dominant excursion, not by a collective instanton path. The explicit mapping of the first-passage area rate function to the weakly bound Brownian particle problem of Refs. [86,87] is a useful connection, and the closed-form diffusion constant obtained by three independent methods (renewal theory, Green-Kubo, and the earlier Donsker-Varadhan result) is a clean consistency check. The numerical simulations support the analytical claims. The final tail itself, however, is not new: it reproduces the known results of Refs. [55-57]. The genuinely new content is the excursion-area rate function and the big-jump derivation, together with the statement that the same c_n emerges from the WKB constant of the mapped problem. The paper would be stronger if the authors explicitly acknowledged that Eq. (17) is a re-derivation of a known asymptotic result rather than presenting it as a new prediction.","major_comments":[{"comment":"The limit interchange v0→0 and A→∞ is the load-bearing step for the central result (17), but it is only asserted. The tail (34) is obtained from the large-deviation ansatz (26) by taking A→∞ at fixed ω=v0/A^{1/n} and then sending ω→0; Eq. (37), by contrast, requires f_{v0}(A) for fixed large A to have a well-defined v0→0 limit that factorizes as v0 times the same exponential. Since ⟨τ(v0)⟩∼v0, the cancellation in (37) depends on this exact factorization. Any subleading v0 dependence in the prefactor (for example v0^{1+α}A^β) or any residual dependence of the effective number of renewals on v0 would change the prefactor or the coefficient in Eq. (17). The statement in Section V that for very large areas f_{v0}(A) is independent of v0 is precisely the needed uniformity condition; it should be proved or supported by a controlled asymptotic bound. The numerical collapse in Fig. 5 at finite v0 is suggestive but does not establish the required limit interchange.","section":"Section V, Eq. (37)"},{"comment":"The phrase 'complete distribution' in the abstract overstates what is derived. The paper obtains the Gaussian bulk (8) by the CLT and the far tail (17) by the big-jump principle, but it does not derive the distribution in the intermediate/matching regime, nor does it prove that these two asymptotic regimes exhaust the distribution of A. Please revise the abstract and the corresponding claims in Section II C to say explicitly that the paper characterizes the bulk and the far tail, not the complete distribution.","section":"Abstract and Section I"},{"comment":"The derivation of the big-jump formula neglects the contribution of the final incomplete excursion A* on the ground that the waiting-time density ψ(τ) decays exponentially, while f(A) is subexponential. The relevant question is whether a subexponential jump can occur in the backward recurrence time τ*, and exponential decay of ψ(τ) alone does not exclude this; this is precisely the subtlety for which Ref. [48] is invoked. The paper should either prove that the A* contribution is subleading in the double limit A,T→∞ or state it explicitly as an assumption. This matters because a non-negligible last-excursion term would alter the prefactor T/⟨τ⟩ in Eq. (38).","section":"Section V, Eqs. (9) and (36)"},{"comment":"The constant c_n in Eq. (30) is imported from Refs. [86,87] after a change of variables that involves a sign difference and a shift of the potential. The equality of I(0) after these transformations is asserted rather than derived. Because c_n enters the central tail (17) exponentially, the authors should show explicitly that the WKB result of Refs. [86,87] applies to Eq. (E11) with their sign convention, or derive I(0) directly from Eq. (29).","section":"Section IV, Eqs. (29)-(30) and Appendix E"}],"minor_comments":[{"comment":"There is a duplicated sentence: 'The result for D_n obtained in [57] is valid only for n>2, The result...' should be corrected to a single, grammatically complete statement.","section":"Section II C"},{"comment":"The caption describes the curves as 'dotted curves' and then as a 'dotted continuous blue curve'; please clarify which curve is the numerical data, which is the big-jump estimate, and which is the far-tail asymptote.","section":"Figure 6 caption"},{"comment":"The sentence 'The expression of f_{v0}(A) has been studied in the previous section by fixing A^{1/n}/v0 and taking A large. This means that also v0 is large' can be misread, because the tail (34) applies for large A with ω=v0/A^{1/n}→0, not with ω held fixed. Please distinguish the two limits explicitly.","section":"Section V, after Eq. (34)"},{"comment":"In Eq. (27) the A-dependence of the term δω^{n+1}I(ω) is not displayed, while the other terms carry explicit powers of A; please show the omitted factor so that the balancing argument leading to Eq. (28) is transparent.","section":"Section IV, Eq. (27)"}],"recommendation":"major_revision","confidential_remarks":"The final asymptotic result (17) is not new; it reproduces the predictions of Refs. [55-57]. The novel and potentially valuable contribution is the excursion/big-jump derivation and the first-passage area rate function, including the connection to Refs. [86,87]. If the authors can supply a rigorous or at least well-controlled justification of the v0→0 limit interchange in Eq. (37) and fix the overstatement in the abstract, the paper would be suitable for publication. The reliance on Refs. [86,87] for c_n is properly cited and is a consistency check rather than a circularity, but the sign-shift issue in the mapping should be verified explicitly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this if you care about the big jump principle or large deviations of integrated observables in correlated continuous processes. The genuinely new piece is the first-passage area rate function for OU excursions (Eq. 29): a nonlinear ODE with a critical point at ω_c=(γn)^{1/n}, whose physical branch interpolates between a Brownian small-A regime and the stretched-exponential large-A regime. The authors derive the rate function from a backward Fokker-Planck equation and then use the single big jump principle to recover the known tail P(A,T) ≍ T exp(-γ^{(n+2)/n} c_n A^{2/n}/σ²), matching the instanton results of Nickelsen-Touchette, Meerson, and Smith. That match is a consistency check, and the paper is candid about it. The CTRW derivation of D_n from ⟨A²⟩/2⟨τ⟩ reproduces the earlier Donsker-Varadhan result and extends it to n=1,2; the Green-Kubo equivalence in the appendix is clean. Credit where due: the derivation of the rate function equation, the identification of the connection to the weakly bound particle problem in [86,87], and the physical link between instantons and a dominant excursion are all real contributions.\n\nThe soft spots are real but not fatal. The main one is the v0→0 limit in Eq. (37). The OU path has uncountably many zero crossings, so the renewal construction exists only as a limit of a regularized restart at v0>0. The paper asserts that for large A the excursion area distribution is independent of v0 and that the prefactor v0 cancels against ⟨τ(v0)⟩∼v0, but it does not prove the limit interchange. The numerical agreement is good, but the concern is legitimate: subleading v0 dependence in the tail of f_{v0}(A) would shift the prefactor of P(A,T). Minor: the abstract says 'complete distribution,' which overstates the derivation—the paper gives the Gaussian bulk and the far tail, not the whole crossover. Also, the abstract calls the first-passage rate function's critical point a 'dynamical phase transition,' but the body notes there is no true non-analyticity at that critical point for the OU case; that is a wording mismatch. The constant c_n is imported from [86,87], one coauthored by Barkai, but it is independently obtained there via WKB and matches [55-57], so I do not see circularity.\n\nWho is this for? Statistical physicists working on large deviations, first-passage functionals, or renewal approaches to continuous processes. It deserves a serious referee: the main result is plausible, well-tested numerically, and the conceptual claim is worth defending. I would ask the referee to focus on the v0→0 limit and the exact asymptotic status of Eq. (17), and to request that the abstract be aligned with what is actually proved.","headline":"Solid paper that connects the big jump principle to OU large deviations; the v0→0 limit is the one place to push back.","tokens_in":34551,"tokens_out":2417,"would_cite":true,"duration_ms":23288,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single giant excursion, not many small fluctuations, produces the rarest values of the Ornstein-Uhlenbeck time-integrated observable.","keywords":["Ornstein-Uhlenbeck process","big jump principle","large deviations","excursions","continuous time random walk","first-passage area","stretched exponential tail","dynamical phase transition"],"falsifier":"Compute the tail of $P(A,T)$ for $n=3$ and $n=4$ at fixed large values of $A/T^{n/(2n-2)}$ using a rare-event sampler, and compare $-\\log P(A,T)$ with $\\gamma^{(n+2)/n} c_n \\sigma^{-2} A^{2/n}$ and with the predicted linear dependence on $T$; also measure whether, conditional on a large $A$, a single excursion accounts for almost all of $A$. A slope disagreement for any $n>2$, or the presence of multiple comparable excursions in the conditioning set, would overturn the central claim.","tokens_in":33432,"feed_emoji":"🎲","tokens_out":15489,"duration_ms":132526,"temperature":0.7,"pith_summary":"The paper claims that the rarest large values of the time-integrated observable $A = \\int_0^T v^n(t)\\,dt$ for an Ornstein-Uhlenbeck velocity $v(t)$ are produced by a single exceptionally large excursion away from zero, not by many small fluctuations adding up. For $n>2$ it derives the stretched-exponential tail $P(A,T) \\asymp T\\exp\\{-\\gamma^{(n+2)/n} c_n \\sigma^{-2} A^{2/n}\\}$, with the constant $c_n$ given explicitly, and shows that this tail coincides with the one previously obtained by a weak-noise path-integral (instanton) calculation. The mechanism is made concrete by mapping the continuous process to a continuous-time random walk whose steps are the signed areas of excursions between zero crossings; because the excursion-area distribution is subexponential, the single big jump principle applies. This matters because it gives a physical picture of anomalous dynamical large deviations and ties the far tail to a critical point in the first-passage area statistics. Typical fluctuations are handled in the same formalism and are Gaussian with an explicit diffusion constant valid for every integer $n$.","feed_headline":"One big jump sets the rare-event tail of Ornstein-Uhlenbeck areas","feed_subtitle":"For velocity powers n>2, the rare-event tail comes from one dominant excursion, matching path-integral predictions.","key_machinery":"The load-bearing machinery is the excursion decomposition of the Ornstein-Uhlenbeck path. Zero crossings form a renewal process, but because continuous paths cross zero uncountably often, the paper regularizes by considering excursions that start at $v_0>0$ and end at the first hitting of zero; each renewal step is the pair $(\\tau_i, A_i)$ consisting of the duration and the signed area $\\int v^n(t)\\,dt$ of one excursion. The first-passage area density $f_{v_0}(A)$ satisfies the backward differential equation $v_0^n \\partial_A f - \\gamma v_0 \\partial_{v_0} f + (\\sigma^2/2)\\partial^2_{v_0} f = 0$. Substituting the anomalous rate-function ansatz $f_{v_0}(A) \\asymp v_0\\exp\\{-A^{2/n}\\sigma^{-2} I(v_0/A^{1/n})\\}$ reduces this equation to the nonlinear ODE $\\frac12 (I')^2 + I'(\\gamma\\omega - \\omega^{n+1}/n) + \\frac{2}{n}\\omega^n I = 0$, whose critical point at $\\omega_c=(\\gamma n)^{1/n}$ marks the deterministic noiseless relaxation and whose value at $\\omega=0$ fixes $c_n$. Since $f_{v_0}(A)$ decays subexponentially for $n>2$, the continuous-time random walk big jump estimate $P(A,T)\\approx \\langle N(T)\\rangle f(A)$ applies; the regularizing factors $\\langle\\tau(v_0)\\rangle\\sim \\text{const}\\times v_0$ and $f_{v_0}(A)\\sim v_0\\exp(\\cdots)$ cancel in the $v_0\\to0$ limit, leaving the $T$-linear stretched-exponential tail.","core_discovery":"The central result is that for $n>2$, in the combined limit $A,T\\to\\infty$ with $A/T^{n/(2n-2)}$ fixed, the distribution of $A=\\int_0^T v^n(t)\\,dt$ obeys $P(A,T) \\asymp T\\exp\\{-\\gamma^{(n+2)/n} c_n \\sigma^{-2} A^{2/n}\\}$ (Eq. (17)), where $c_n$ is the explicit constant in Eq. (30). The tail is subexponential in $A$ and linear in $T$, the linear factor being the average number of zero-crossing renewals in time $T$. The route runs through the first-passage area distribution of a single excursion, $f_{v_0}(A) \\asymp v_0 \\exp\\{-A^{2/n}\\sigma^{-2} I(v_0/A^{1/n})\\}$, whose rate function $I$ solves a nonlinear differential equation, has its minimum at the noiseless deterministic value $\\omega_c=(\\gamma n)^{1/n}$, and takes the constant value $I(0)=\\gamma^{(n+2)/n} c_n$. The paper's reading of this chain is that the anomalous scaling previously attributed to an instantonic solution is exactly the single big jump effect, with the largest excursion controlling the far tail.","pith_inferences":["Editorial inference: if the big jump mechanism is correct, rare-event simulations that condition on one large excursion should reproduce the tail far more cheaply than brute-force sampling, and the distribution of the largest observed excursion area should predict the tail of $P(A,T)$.","Editorial inference: the same excursion-to-CTRW strategy should produce stretched-exponential tails for other recurrent one-dimensional diffusions with exponentially decaying correlations whenever the first-passage area distribution is subexponential; the exponent would be set by the tail of that distribution rather than by $n$ alone.","Editorial inference: the paper leaves open whether the $v_0\\to0$ and $A\\to\\infty$ limits can be rigorously interchanged; proving this interchange would turn the big jump principle into a theorem for correlated continuous processes, while a counterexample would show that the tail prefactor depends on the regularization.","Editorial inference: the mapping to weakly binding potential rate functions suggests an equivalence class of observables whose anomalous rate functions coincide after rescaling; testing this equivalence numerically for $n=3,4,5$ would either confirm the universality or expose where it ends."],"forward_implications":["For $n>2$ the entire large-$A$ tail of $P(A,T)$ is fixed by first-passage area statistics, so the anomalous exponent $2/n$ and coefficient $c_n$ need no variational path-integral input.","Typical fluctuations are Gaussian with the diffusion constant $D_n=\\langle A^2\\rangle/(2\\langle\\tau\\rangle)$ given by Eq. (14), reproducing the previously known perturbative result for $n>2$ and extending it to $n=1,2$.","The excursion-area rate function has a critical point at $\\omega_c=(\\gamma n)^{1/n}$ separating a Brownian small-area regime from a subexponential large-area regime, so the dynamical phase transition in $P(A,T)$ has a counterpart in the single-excursion statistics.","Because the big jump tail and the instantonic tail share the same constant $c_n$, the optimal instantonic path and the single dominant excursion are the same asymptotic event described by two methods.","For even $n$, the same formulas apply to the shifted observable $A-\\langle v^n\\rangle_{\\mathrm{eq}} T$, so the results cover time-averaged energy and higher even moments as well as odd moments."],"supporting_citations":[{"why":"It states the single big jump principle for sums of subexponential random variables, which the paper applies to excursion areas to obtain the tail of $P(A,T)$.","marker":"[34]"},{"why":"It supplies the renewal big jump estimate $P(A,T)\\sim \\langle N(T)\\rangle f(A)$ used to put the CTRW mapping into the tail.","marker":"[48]"},{"why":"It introduces the excursions technique that maps a continuous process to a continuous-time random walk via zero crossings, the backbone of the paper's method.","marker":"[64]"},{"why":"It provides the prior instantonic and quantum-mapping result for the anomalous scaling and the diffusion coefficient that the new approach reproduces and extends.","marker":"[57]"},{"why":"It established the anomalous subexponential large-deviation scaling for time-integrated observables of the Ornstein-Uhlenbeck process, the target result rederived here by big jump methods.","marker":"[55]"},{"why":"It supplies the analytic value of the rate-function constant $c_n$ through the solution for weakly bound Brownian particles, fixing the numerical coefficient in the tail.","marker":"[86, 87]"},{"why":"It gives the Brownian first-passage area distribution whose small-area form fixes the exponent of the subleading prefactor in the excursion-area density.","marker":"[88]"}],"fun_headline_variants":["Single big jump explains rare-event tails in Ornstein-Uhlenbeck processes","For n>2, one dominant excursion sets the rare-event tail of OU areas","Big jump effect: largest excursion controls rare events in OU process","Anomalous scaling in OU processes traced to a single big jump","Largest excursion dictates rare-event tail in Ornstein-Uhlenbeck dynamics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the $v_0\\to0$ regularization of the uncountably many zero crossings can be interchanged with the large-$A$ limit: the factor $v_0$ in the excursion-area density cancels exactly against the factor $v_0$ in $\\langle\\tau(v_0)\\rangle$, leaving a tail independent of the regularization.","fun_headline_variants_meta":{"raw":{"variants":["Single big jump explains rare-event tails in Ornstein-Uhlenbeck processes","For n>2, one dominant excursion sets the rare-event tail of OU areas","Big jump effect: largest excursion controls rare events in OU process","Anomalous scaling in OU processes traced to a single big jump","Largest excursion dictates rare-event tail in Ornstein-Uhlenbeck dynamics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00067,"raw_usage":{"total_tokens":3135,"prompt_tokens":1107,"completion_tokens":2028,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":723,"completion_tokens_details":{"reasoning_tokens":1931}},"tokens_in":723,"tokens_out":2028,"duration_ms":12810,"temperature":1.0,"reasoning_tokens":1931,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:36:50.643844+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the tail of $P(A,T)$ for $n=3$ and $n=4$ at fixed large values of $A/T^{n/(2n-2)}$ using a rare-event sampler, and compare $-\\log P(A,T)$ with $\\gamma^{(n+2)/n} c_n \\sigma^{-2} A^{2/n}$ and with the predicted linear dependence on $T$; also measure whether, conditional on a large $A$, a single excursion accounts for almost all of $A$. A slope disagreement for any $n>2$, or the presence of multiple comparable excursions in the conditioning set, would overturn the central claim.","supporting_citations":[{"cited_title":"Vezzani and R","cited_arxiv_id":null,"evidence_quote":"It supplies the renewal big jump estimate $P(A,T)\\sim \\langle N(T)\\rangle f(A)$ used to put the CTRW mapping into the tail."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It established the anomalous subexponential large-deviation scaling for time-integrated observables of the Ornstein-Uhlenbeck process, the target result rederived here by big jump methods."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the Brownian first-passage area distribution whose small-area form fixes the exponent of the subleading prefactor in the excursion-area density."}],"review_version":1}