{"id":"f25250b3-f85c-4917-be62-7262ddb0b848","arxiv_id":"1908.10975","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The far tails of the position distribution in generalized Lévy walks are described by the single-big-jump principle, with non-universal exponents set by α, ν, and η.","lead":"This paper derives the rare-event, far-tail statistics of the position of a generalized Lévy walk, a random motion with power-law waiting times and accelerating or decelerating steps. It shows the tails are set by a single big jump and depend on the microscopic step dynamics through three exponents, giving explicit scaling formulas for extreme fluctuations and their moments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central tail formulas (8)-(10) depend entirely on the heuristic single-big-jump ansatz Eq. (3), which the paper itself leaves unproved; the paper does not bound multi-jump paths or the transient error from replacing the renewal density by its asymptotic form.","rationale":"The reader's weakest assumption is exactly the heuristic status of the big-jump ansatz, Eq. (3), and the stress-test does not find a more serious internal inconsistency. The tail formulas (8)-(10), the non-analytic scaling function F(x), and the anomalous moment exponents (11) all depend on that ansatz. The paper is explicit that the rate approach is heuristic, so the concern is acknowledged in the manuscript rather than hidden. The SI calculations are internally consistent, and the simulations show agreement in representative regimes, which is real evidence but not a proof of asymptotic dominance. I therefore keep the reader's CONDITIONAL verdict unchanged: the central claim is plausible and well supported, but the word 'exact' in the abstract is stronger than the derivation. The concrete test of deriving the tail from the exact renewal equations without the ansatz would settle whether the heuristic is actually exact in the claimed regimes.","tokens_in":17961,"tokens_out":19265,"duration_ms":199550,"concrete_test":"Derive the far tail of P(R,T) directly from the exact renewal representation of the SI, Eqs. (18)-(23), using the standard infinite-density/Tauberian method applied to standard Levy walks in Refs. [37,39], without inserting Eq. (3). Do this for the case alpha<1, eta<nu, and compare the resulting R-exponent and T-prefactor with Eq. (10), including the constant D_alpha evaluated from the exact renewal density. Agreement would confirm the single-big-jump ansatz; disagreement would invalidate the central claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. (3) is the sole input for B(R,T) in every regime summarized in Table 1, yet the Discussion explicitly calls the rate approach 'heuristic' and the Methods section calls P(R|T,t,Tw) a 'heuristic expression'; no rigorous derivation is supplied. The load-bearing condition is that for R much larger than the scaling length, all paths involving more than one jump contribute negligibly to P(R,T). The paper asserts this but does not quantify the error. In a renewal process the exact probability of a single jump starting at Tw with duration t is m(Tw)lambda(t), where m is the renewal density; the paper instead uses n_R(Tw)=d<N>/dT and inserts asymptotic forms (constant for alpha>1, C_alpha T_w^{alpha-1} for alpha<1) over the full integration range 0<Tw<T. Transient and boundary corrections are never estimated. Competing paths with two or more moderately long jumps in the same direction can reach the same R, and their relative weight is not bounded by any argument in the paper. If the ansatz fails in any listed regime, Eqs. (8)-(10) and the moment exponents (11) do not follow. The simulations provide useful support, but they do not settle the asymptotic dominance for all regimes.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies generalized Lévy walks, in which step durations are drawn from a power-law distribution λ(t) ∼ t^{-1-α} and the intra-step motion is r(T) = r(T_i) + c_i t_i^{ν-η} (T − T_i)^η. It derives the bulk scaling of the position PDF by Fourier-Laplace methods, obtaining the four regimes in Eq. (7), and then uses the single-big-jump rate ansatz, Eq. (3), to predict the far tail B(R,T) for R ≫ ℓ(T). The main results are the scaling form Eq. (8) with a non-analytic scaling function F(x), the pure power-law tails in Eqs. (9) and (10), the regime classification in Table 1, and the anomalous moment exponents in Eq. (11). All analytical predictions are compared with numerical simulations.","tokens_in":18170,"tokens_out":5918,"duration_ms":63161,"significance":"If the tail formulas are correct, the paper establishes a genuinely non-universal structure for rare events in a broad class of Lévy walks: the far tail depends on all three exponents α, ν, and η, and exhibits non-analytic behavior at R = cT^ν. This goes beyond earlier results for standard Lévy walks and for the η=ν case, and the moment analysis connects the single-big-jump mechanism to strong anomalous diffusion. The manuscript's strengths include explicit, parameter-free calculations in the SI, correct reduction to known limits (standard Lévy walks and the η=ν case), and systematic numerical tests with no fitted parameters. Its main weakness is that the central single-big-jump ansatz is stated heuristically and is not derived or bounded; this limits the claimed 'exactness' of the tail predictions.","major_comments":[{"comment":"The load-bearing step is the ansatz B(R,T) = ∫dt∫dT_w n_R(T_w)λ(t)P(R|T,t,T_w), in which all jumps before and after the big jump are neglected. The manuscript explicitly calls the rate approach heuristic (Discussion, p. 9) and calls P(R|T,t,T_w) a heuristic expression (Methods, p. 10), yet the Abstract and Section 3 describe the resulting tails as 'exact'. No argument is supplied that multi-jump paths contribute subdominantly for R ≫ ℓ(T), nor is the error from replacing the exact renewal density by its asymptotic form bounded. Since Eqs. (8)–(10), Table 1, and the moment exponents (11) all inherit their validity from Eq. (3), the authors should either provide a derivation or a quantitative bound for the ansatz, or consistently present the tail results as heuristic predictions rather than exact forms.","section":"Results §1, Eq. (3); Discussion, p. 9"},{"comment":"The derivations of the tail formulas replace the jump-attempt rate n_R(T_w) = d⟨N(T_w)⟩/dT_w by its asymptotic form over the entire integration range 0 < T_w < T. For α > 1, ptot(t,T_w) = λ(t)/⟨t⟩ is used even for T_w near 0 and near T, where the renewal density has transients and boundary effects. For α < 1, Eq. (37) uses n_R(T_w) = C_α T_w^{α-1}, which is singular at T_w = 0. These transient and boundary contributions are never estimated. The authors should show that the boundary terms are subleading in the regime R ≫ ℓ(T), or explicitly state that the leading-order tail has not been rigorously established.","section":"SI Eqs. (32)–(37); main Eqs. (9)–(10)"},{"comment":"The prediction that moments with q > α/(ν−η) diverge relies on the tail being exactly a pure power law B(R,T) ∼ T^{...} R^{-1-α/(ν−η)} for arbitrarily large R. If the single-big-jump ansatz receives corrections at very large R from multi-step paths, such as the coherent many-step processes mentioned in the text for η ≥ ν, the divergence could be modified or cut off. The NR dependence in Fig. 5 is consistent with divergence, but it does not by itself establish the asymptotic claim. The manuscript should state precisely what assumption about the far tail is needed for the infinite-moment result, and should soften the claim if that assumption is only heuristic.","section":"Results §4, Eq. (11); Fig. 5"}],"minor_comments":[{"comment":"Panel (b) of Figure 2 lists the same parameters as panel (a), α = 1.6 and ν = 0.7, yet the two panels are described as showing qualitatively different scaling regimes (Gaussian versus superdiffusive Lévy scaling). This appears to be a typo; please correct the parameter values in the caption.","section":"Fig. 2 caption"},{"comment":"The expression for B0(R,T) in the η > ν case contains the factor '( R/ctν )', which is ambiguous: it should presumably be (R/(cT^ν))^{1/ν} or a similarly explicit combination of R, T, and ν. Please fix the typographical error and verify the resulting x-dependence of F(x).","section":"SI Eq. (35)"},{"comment":"The Abstract states that the big jump principle gives 'the exact form of the tails', while the Discussion states that the scheme is heuristic and a rigorous derivation is open. These statements should be harmonized so that the reader is not misled about the status of the results.","section":"Abstract vs. Discussion"},{"comment":"The sentence 'Clearly these processes are exponentially suppressed and very difficult to be observed' asserts a quantitative claim about multi-step coherent paths without proof or citation. If this exponential suppression is important for justifying the regime where Eq. (3) does not apply, it should be substantiated.","section":"Results §3, p. 8"},{"comment":"In the definition of γ(k,s), the integrand uses λ(t') but the integration variable in the inner integral is t2; the variable t' is not defined. Please use t2 consistently.","section":"SI Eq. (23)"}],"recommendation":"major_revision","confidential_remarks":"The main risk of this paper is that the 'exact' language outsizes the proof: the central ansatz is acknowledged as heuristic, and no bounds on multi-jump or transient corrections are provided. I would be comfortable with publication after a major revision that either supplies a rigorous justification for the ansatz in the relevant asymptotic regime or consistently reframes the tail results as heuristic predictions supported by simulations and known limiting cases."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper does something concrete and useful: for generalized Lévy walks with step durations λ(t) ~ t^{-1-α} and intra-step motion r(T) = r(T_i) + c_i t_i^{ν-η}(T-T_i)^η, it derives the far-tail of the position PDF as B(R,T) = T^{-(α-1+ν)} F(R/cT^ν) (Eq. 8), with F non-analytic at x=1 and, in the η<ν regimes, a pure power law ~ x^{-1-α/(ν-η)}. It also gives the moment exponents γ(q), including diverging moments for q beyond a critical order. Prior work only had the MSD; the η=ν case was treated separately. So the full parameter map and the non-universal scaling functions are genuinely new.\n\nThe SI is the best part. The integrals leading to F(x) are explicit, the expressions reduce correctly to known limits (standard Lévy walks, η=ν), and no parameters are fitted. The simulation comparisons in Figs. 3–6 are convincing at the level of asymptotic scaling, including the divergent moments depending on realization number. The authors should be credited for flagging regimes where the big jump principle does not apply (η≥ν cases) rather than overclaiming.\n\nThe soft spots are mostly around the foundation. Eq. (3), the single-big-jump ansatz, is taken from the group's prior work and is not derived here. The paper itself calls the scheme heuristic and leaves a rigorous derivation open. For a physics paper that is acceptable, but the abstract's word 'exact' overstates the status: these are exact results conditional on the ansatz, not on a proven theorem. The stress-test worry—no bound on multi-jump paths that reach the same R—is legitimate but not specific to this paper; it applies to the entire big-jump approach. Given the subexponential tail of the step lengths, the single-jump dominance is physically expected, and the simulations support it. I do not see a regime where the ansatz is likely to fail; the authors are careful about the ones where it cannot apply.\n\nMinor issues: C_α in Eq. (10) is left implicit; there are typos in the Figure 2 captions (the same parameters labeled as both Gaussian and Lévy-stable) and in the SI text around Eqs. (34)–(35), where the support of F(x) for η<ν vs η>ν is described backwards relative to the formulas. No code or data are provided, and the simulation plots lack error bars, but the trends are clear enough.\n\nWho should read this: anyone working on Lévy walks, anomalous transport, or rare-event statistics in renewal processes. It is a solid incremental advance, not a revolution. I would send it to peer review; the right referee will ask for the 'exact' language to be toned down and the typos fixed, but the science is in good shape.\n\nRead it critically but accept it conditionally.","headline":"Solid extension of the big jump heuristic to generalized Lévy walks with explicit tail formulas and good simulation support; the main caveat is that the central principle remains unproved.","tokens_in":18748,"tokens_out":6734,"would_cite":true,"duration_ms":67122,"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 big jump determines the full rare-event tail of generalized Lévy walks.","keywords":["big jump principle","Lévy walks","rare events","heavy-tailed distributions","anomalous diffusion","strong anomalous diffusion","non-analytic tails","infinite densities"],"falsifier":"Direct simulation can settle it: for fixed $\\alpha,\\nu,\\eta$ in a predicted big-jump regime, plot $R^{1+\\alpha/(\\nu-\\eta)}P(R,T)$ against scaled $R/T^\\nu$ for several large $T$; the curves should show the predicted power-law prefactor, and the scaling function must have the predicted cusp at $R=cT^\\nu$. Observing a clean power-law tail in a regime where the paper says the big jump does not apply, or a smooth tail where a cusp is predicted, would refute the claim.","tokens_in":17713,"feed_emoji":"📈","tokens_out":9330,"duration_ms":90360,"temperature":0.7,"pith_summary":"For generalized Lévy walks, in which step durations are drawn from a power law $\\lambda(t)\\sim t^{-1-\\alpha}$ and a single step moves as $c_i t_i^{\\nu-\\eta}(T-T_i)^\\eta$, the paper argues that every rare large displacement is produced by one dominant step, not by the accumulation of many small ones. Applying the single-big-jump estimate to this model, it obtains the full tail of the position distribution at distances far beyond the bulk scaling length. The central result is that the tail depends on all three exponents $\\alpha, \\nu, \\eta$, taking the scaling form $B(R,T)=T^{-(\\alpha-1+\\nu)}F(R/(cT^\\nu))$ with a non-analytic scaling function in the main regime, and reducing to pure power laws in the regimes of Eqs. (9)--(10). Because the bulk is universal while the tail is not, the tail of rare events becomes a window onto the microscopic acceleration and deceleration inside a single step. The same calculation explains why high moments diverge and why averages over few trajectories depend on the number of realizations.","feed_headline":"One big jump sets the rare-event tail of Lévy walks","feed_subtitle":"Tail shape carries the step's internal acceleration exponent, not just the jump-length statistics.","key_machinery":"The central object is the generalized Lévy walk, defined by step durations $\\lambda(t)\\sim t^{-1-\\alpha}$ and intra-step motion $r(T)=r(T_i)+c_i t_i^{\\nu-\\eta}(T-T_i)^\\eta$, whose two exponents $\\nu$ and $\\eta$ respectively set how step length grows with duration and how the walker accelerates or decelerates inside the step. The engine of the argument is the big-jump rate formula, Eq. (3), which splits the far tail into the rate $p_{\\rm tot}(t,T_w)$ at which a jump is attempted and the single-jump propagator $P(R|T,t,T_w)$; two geometric contributions are summed, one where the walker is still inside the big jump at the observation time and one where the jump was completed earlier. This machinery produces the scaling function $F(x)$ and the power-law tails, and it identifies the light-cone boundary $R=cT^\\nu$ as the origin of the non-analyticities.","core_discovery":"The paper claims that for $R\\gg\\ell(T)$ the tail $B(R,T)$ of the position PDF of a generalized Lévy walk is exactly the single-big-jump integral $B(R,T)=\\int dt\\int_0^T dT_w\\, p_{\\rm tot}(t,T_w)P(R|T,t,T_w)$, in which $p_{\\rm tot}(t,T_w)=n_R(T_w)\\lambda(t)$ is the rate of attempts to make a jump of duration $t$ and $P(R|T,t,T_w)$ propagates the walker during that one jump. In the regime $\\alpha>1$, $\\nu>1/2$ this yields $B(R,T)=T^{-(\\alpha-1+\\nu)}F(R/(cT^\\nu))$, with $F$ continuous but non-differentiable at $x=1$ for $\\eta\\neq\\nu$ and discontinuous for $\\eta=\\nu$. In the regimes $\\alpha>1,\\nu<1/2,\\eta<\\nu$ and $\\alpha<1,\\eta<\\nu$ the tail is a pure power law $B(R,T)\\sim T^{\\alpha\\eta/(\\nu-\\eta)+1}/R^{1+\\alpha/(\\nu-\\eta)}$, so the single-step dynamics exponent $\\eta$ enters the rare-event tail explicitly. When no single jump can reach beyond the bulk scaling length, the paper argues that the big-jump estimate does not apply and the tail is either zero at the light cone or exponentially suppressed.","pith_inferences":["Beyond the paper: the tail exponent $1+\\alpha/(\\nu-\\eta)$ suggests that rare-event measurements could be inverted to read out the microscopic acceleration exponent $\\eta$ once $\\alpha$ and $\\nu$ are known from the bulk, turning tail shape into a dynamical probe.","Beyond the paper: the criterion distinguishing big-jump from non-big-jump regimes, namely whether one step can exceed the scaling length, likely generalizes to other heavy-tailed observables such as released energies or financial losses wherever a single draw dominates the extreme.","Beyond the paper: finite-time simulations should show the cusp at $R=cT^\\nu$ rounding over a width controlled by $T$; measuring that rounding is a testable finite-time correction absent from the asymptotic $T\\to\\infty$ formulas.","Beyond the paper: because moments diverge for $q>\\alpha/(\\nu-\\eta)$, empirical averages over few realizations are not reproducible, so median or typical-value estimators would remain stable and could serve as better risk measures."],"forward_implications":["The far tail no longer shares the universality of the bulk: it depends explicitly on all three exponents $\\alpha$, $\\nu$, and $\\eta$.","For $\\eta<\\nu$ in the $\\alpha>1,\\nu<1/2$ and $\\alpha<1$ regimes, the tail is a pure power law, so rare events are scale-invariant.","In the $\\alpha>1$, $\\nu>1/2$ regime, the scaling function $F(x)$ is non-analytic at $x=R/(cT^\\nu)=1$, meaning the finite-velocity horizon of the walk creates a cusp or discontinuity in the tail.","For moment order $q>\\alpha/(\\nu-\\eta)$, the moments diverge and empirical averages depend on the number of realizations, producing strong anomalous diffusion with a piecewise-linear exponent $\\gamma(q)$.","Where a single jump cannot reach beyond the bulk scaling length, the big-jump tail formulas fail and deviations are exponentially suppressed or absent."],"supporting_citations":[{"why":"Introduces the single-big-jump rate approach on which the paper's tail calculations are built.","marker":"[15]"},{"why":"Shows the big-jump approach applied to Lévy walks in disordered transport, the immediate precedent for this extension.","marker":"[17]"},{"why":"Provides the standard Lévy walk framework and the step-first and wait-first limits recovered by extreme values of $\\eta$.","marker":"[21]"},{"why":"Defines the generalized Lévy walk with exponents $\\nu$ and $\\eta$ and supplies the model and earlier mean-square-displacement results.","marker":"[28, 29]"},{"why":"Establishes non-normalizable infinite densities, framing why $B(R,T)$ can diverge at $R=0$ yet still control high moments.","marker":"[37]"},{"why":"Computes rare-event densities for the coupled case $\\eta=\\nu$ via moment resummation, the special case that the present $F(x)$ generalizes.","marker":"[39]"},{"why":"Supplies the Lévy-stable and CTRW scaling functions used to classify the universal bulk behavior.","marker":"[40]"}],"fun_headline_variants":["Single big jump dictates Levy walk rare tails","Step dynamics alone sets Levy walk rare tails","Big jump principle reveals non-universal Levy tails","Exact rare tails for generalized Levy walks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the big-jump ansatz, Eq. (3): for $R\\gg\\ell(T)$ the tail of the distribution is equal to the integral over all single jumps, with the motion before and after the big jump neglected; the paper states this scheme is heuristic and leaves its rigorous derivation open.","fun_headline_variants_meta":{"raw":{"variants":["Single big jump dictates Levy walk rare tails","Step dynamics alone sets Levy walk rare tails","Big jump principle reveals non-universal Levy tails","Exact rare tails for generalized Levy walks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001198,"raw_usage":{"total_tokens":4973,"prompt_tokens":1009,"completion_tokens":3964,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":625,"completion_tokens_details":{"reasoning_tokens":3916}},"tokens_in":625,"tokens_out":3964,"duration_ms":24730,"temperature":1.0,"reasoning_tokens":3916,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:28:49.114095+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Direct simulation can settle it: for fixed $\\alpha,\\nu,\\eta$ in a predicted big-jump regime, plot $R^{1+\\alpha/(\\nu-\\eta)}P(R,T)$ against scaled $R/T^\\nu$ for several large $T$; the curves should show the predicted power-law prefactor, and the scaling function must have the predicted cusp at $R=cT^\\nu$. Observing a clean power-law tail in a regime where the paper says the big jump does not apply, or a smooth tail where a cusp is predicted, would refute the claim.","supporting_citations":[{"cited_title":"Geluk, Q","cited_arxiv_id":null,"evidence_quote":"Introduces the single-big-jump rate approach on which the paper's tail calculations are built."},{"cited_title":"Burioni, L","cited_arxiv_id":null,"evidence_quote":"Shows the big-jump approach applied to Lévy walks in disordered transport, the immediate precedent for this extension."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the standard Lévy walk framework and the step-first and wait-first limits recovered by extreme values of $\\eta$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes non-normalizable infinite densities, framing why $B(R,T)$ can diverge at $R=0$ yet still control high moments."},{"cited_title":"Filiasi, G","cited_arxiv_id":null,"evidence_quote":"Computes rare-event densities for the coupled case $\\eta=\\nu$ via moment resummation, the special case that the present $F(x)$ generalizes."},{"cited_title":"Corberi, Development and regression of a large fluctuation , Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the Lévy-stable and CTRW scaling functions used to classify the universal bulk behavior."}],"review_version":1}