{"id":"ce1fd033-d1ba-4d35-8c54-312ae427f27d","arxiv_id":"2501.05585","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a quenched trap model, the disorder-averaged particle packet decays with a universal Laplace-like tail set by the first moments of the escape rates.","lead":"This paper shows that a particle diffusing through a frozen random environment spreads with a universal exponential-like tail, using a one-dimensional trap model. The result connects decades-old observations of non-Gaussian diffusion in glasses and cells to the statistics of energy traps.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (12)'s equality rests on an unproved upper-bound monotonicity ⟨P⟩<P_ordered; without it only the '≥1' lower-bound statement is established.","rationale":"The reader's weakest assumption and mine coincide: the upper-bound monotonicity is the single unproved link in the two-sided limit. I agree with the CONDITIONAL verdict: the lower-bound derivation is solid and the numerical evidence is persuasive, so rejection would be too harsh; but the paper should either supply a comparison theorem (e.g., a coupling showing the time-changed ordered process dominates |X_t| for large x) or explicitly state Eq. (12) as a lower bound plus numerical conjecture. The two-rate calculation in the proposed test indicates the upper bound is very likely true, since any rare-environment advantage is exponentially suppressed by p^x while P_ordered decays superexponentially in x; this supports keeping the claim as conditional rather than rejecting it. The direct-path dominance for finite-x corrections is a separate, less central issue: Eq. (16)'s agreement with simulations is strong independent support, but it is not a proof of the equality in Eq. (12).","tokens_in":18130,"tokens_out":22895,"duration_ms":250578,"concrete_test":"Test the two-rate thermal model R∈{r,ε}, P(R=r)=p, ε→0. Compute ⟨P(x,t)⟩ to leading order in large x and compare with P_ordered(x,t). The dominant candidate to violate the upper bound is the 'fast corridor' environment R_0=...=R_{x-1}=r, R_x=ε, whose contribution is p^x(1−p) times the ordered hitting probability (i.e., p^x(1−p)e^{rt}P_ordered in the direct-path limit). If this worst-case class stays below P_ordered for all p<1 at large x, the monotonicity is supported; if it exceeds P_ordered, Eq. (12)'s equality fails because the liminf drops below 1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline equality, Eq. (12), is established only if both sides of the sandwich hold. The lower-bound side (limsup ≤ 1) follows from Eqs. (7)-(10). The upper-bound side (liminf ≥ 1) rests on the unproved sentence after Fig. 2: 'Therefore, for large x, ⟨P(x,t)⟩ < P_ordered(x,t)'. No stochastic-ordering or coupling argument is given; the preceding 'a particle diffuses to larger distances faster' is physical intuition. That missing step is load-bearing because P_ordered's own ratio in the denominator of Eq. (12) approaches 1 from below as x→∞, so the upper bound is what converts the stated '≥ 1' into the claimed identity. The paper acknowledges the direct-path dominance used to justify the identity only as 'we argue' / 'we expect'; the numerical evidence and 1/x corrections make it plausible but do not prove the upper bound. If the monotonicity failed in some regime, the universal limit could be strictly less than 1 and the central claim would reduce to an inequality, not the Laplace-law identity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"Using a one-dimensional quenched trap model with independent, identically distributed rates drawn from a density of states ρ(E), the paper studies the large-displacement tail of the disorder-averaged propagator ⟨P(x,t)⟩. It derives a lower bound from directed paths, whose disorder-averaged Laplace transform is evaluated by a saddle-point method, yielding the Laplace-like form in Eq. (10). It proposes an upper bound by comparison with the ordered system in which all rates equal the maximal rate r. Single-turn backtracking paths are then added to produce the improved lower bound in Eq. (16), with explicit 1/x corrections depending on the first three moments of the rates. Numerical simulations for exponential, Gaussian, and uniform densities of states are compared with Eq. (16). The central claim is Eq. (12): the ratio -ln⟨P(x,t)⟩/[|x| ln(2|x|/(e⟨R⟩t))] tends to 1, so the packet tail is a universal Laplace-like law whose constants are set by moments of the escape rates.","tokens_in":18396,"tokens_out":18861,"duration_ms":184770,"significance":"If fully established, the result would extend Laplace's first law of errors from CTRW mean-field models to quenched, spatially correlated disorder, and would connect the phenomenon to the density of states through moments of the escape rates. The paper's strengths are its explicit saddle-point calculations in the Supplemental Material, the absence of fitted parameters in the theoretical curves, and numerical tests across several disorder families, including the anomalous T<Tg exponential case. The improved lower bound in Eq. (16) is a practically useful finite-x handle for comparing with experiments. The main caveat is that the exact equality in Eq. (12) is not fully proven; the lower-bound and upper-bound arguments establish the two sides of the sandwich only if the unproved ordering against the ordered system holds.","major_comments":[{"comment":"Equation (12) states the limit is ≥1 and attributes this to Eq. (10). Because Eq. (10) is a lower bound on ⟨P(x,t)⟩, it gives -ln⟨P⟩ ≤ -ln⟨Prob(→)⟩, so the ratio in Eq. (12) is asymptotically ≤1, not ≥1. The ≥1 direction is what would follow from the upper bound against P_ordered. The inequality direction, or its attribution, should be corrected; as printed, the logical support for Eq. (12) is inverted.","section":"Lower bound, Eq. (12)"},{"comment":"The assertion 'for large x, ⟨P(x,t)⟩ < P_ordered(x,t)' is load-bearing: it supplies the liminf ≥1 half of Eq. (12) and hence the claimed equality. No coupling, stochastic-ordering, or other rigorous argument is given; the physical intuition about short waiting times is not a proof, especially because a deep trap at the destination can locally increase the occupation probability. Please provide a rigorous argument, for example a per-site Poisson-thinning coupling followed by a large-deviation estimate, or explicitly weaken the central claim to a conjecture stated as such.","section":"Upper bound, paragraph after Fig. 2"},{"comment":"The statement that Eq. (12) is an identity because the direct path dominates for large x is presented only as 'we argue' and 'we expect'. This is a separate, unproved asymptotic equivalence. The single-turn corrections and numerical evidence make it plausible, but they do not exclude the possibility that the accumulated contributions of paths with many backtracking events alter the leading exponent. Either prove the dominance by bounding the total contribution of all non-direct paths, or state Eq. (12) as a conjecture supported by the bounds and numerics.","section":"Identity claim, after Eq. (12)"}],"minor_comments":[{"comment":"There is an inconsistency in the denominator of the single-turn contribution: main-text Eq. (15) has (2⟨R⟩|x|)^2, while the Supplemental Material Eq. (SM69) appears to have (2⟨R⟩x^2)^2. The O(1/x) correction in Eq. (16) confirms the main-text version; please correct the SM typo.","section":"Eq. (15) and SM Eq. (SM69)"},{"comment":"The indicator in Eq. (6) has the upper limit Σ_{i=0}^x τ_{i+1}; the correct condition is t_x = Σ_{i=0}^x τ_i, which is what is used in Eq. (SM14). Please fix the typo in the displayed expression.","section":"Eq. (6) and SM Eq. (SM10)"},{"comment":"There are several typos: 'isordered' in the abstract, 'Bertheir' for 'Berthier' in the second paragraph, and 'appoach' in reference [64].","section":"Abstract and references"},{"comment":"The notation '[Prob(→)' for the Laplace transform is introduced without a definition in the main text; please define the bracket notation before first use, or replace it with an explicitly defined symbol.","section":"Eq. (9) and SM notation"},{"comment":"The phrase 'the lower bound always saturates' is unclear; it presumably means that the simulated packet approaches the lower bound at large x, but the wording should be made precise.","section":"Fig. 2 caption"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is self-contained and the numerical work is honest, so I see no integrity or scope problem. The obstacle to acceptance is the rigor gap around Eq. (12): the inequality direction in Eq. (12) is misattributed, and the upper-bound ordering is asserted rather than proved. Both are fixable within the manuscript's scope by either supplying a proof or by carefully restating the identity as a conjecture with the proven inequalities presented as the main rigorous results. With that revision I would be supportive."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a real step forward for the quenched trap model, with a careful lower bound and genuinely useful backtracking corrections, but the paper's headline equality (Eq. 12) still leans on an unproved monotonicity assertion. I'd send it to review, with a clear request to either prove or label that step.\n\nWhat's new: the direct-path saddle-point calculation leading to Eq. (10) is straightforward CTRW large-deviation machinery, but the authors extend it with explicit single-turn backtracking paths and show they give 1/x corrections that match simulations across exponential, Gaussian, and uniform disorder densities. That is real work, and the comparison is convincing. The universality claim — that the tail constant is 1, independent of ρ(E), with prefactors set by the moments of the escape-rate distribution — is a nice result, and the numerics support it over the range shown.\n\nThe soft spot is exactly where the reader put it. The upper bound ⟨P⟩ < P_ordered for large x is asserted on physical intuition after Fig. 2, not proved. The stress test's worry that P_ordered's own ratio approaches 1 from below is not the real obstacle — the constant shift in the logarithm is subdominant, so that part is fine. What is missing is a coupling or stochastic-ordering argument that each realization with rates ≤ r has a large-x propagator no larger than the ordered one. That inequality is load-bearing: the lower-bound argument only establishes '≥ 1' in Eq. (12), and without the upper bound you cannot promote it to the identity. The paper does confess it is arguing/expecting, and the numerical evidence is good, but the abstract and conclusion state the identity as established. A referee should ask the authors to either supply a proof or clearly mark Eq. (12) as a conjecture with a proven lower bound and numerical support.\n\nThe rest holds up. The treatment of disorder correlations in the backtracking paths is careful, the moments of ρ(E) are computed consistently, and I see no circularity — no fitting of the target tail, just parameters set by the model. The citation pattern is reasonable.\n\nWho should read this: anyone working on anomalous diffusion, trap models, or large deviations in quenched disorder. It deserves a serious referee, but with a request to tighten the upper-bound step. My own verdict would be conditional acceptance along those lines.","headline":"Solid lower-bound derivation and convincing numerics, but the claimed identity in Eq. (12) needs a proof of the upper-bound monotonicity or a softer claim.","tokens_in":18859,"tokens_out":6153,"would_cite":true,"duration_ms":62954,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K37","82C31","82C44"],"pacs":["05.40.Fb","05.60.-k"],"model":"deepseek-v4-flash","headline":"Disorder can't erase Laplace's exponential diffusion tails","keywords":["Laplace's first law","exponential tails","trap model","quenched disorder","large deviations","random walk","anomalous diffusion","disorder-averaged propagator"],"falsifier":"Numerically compute the ratio in Eq. (12) for exponential trap-energy disorder with $T$ just above $T_g$, pushing to very large $|x|$ with rare-event sampling or by exact evaluation of path sums; a limiting ratio strictly below 1 while the limsup is 1 would falsify the identity. Alternatively, test the asserted inequality $\\langle P(x,t)\\rangle < e^{-rt}I_{|x|}(rt)$ directly at $T=T_g/2$; a single violation would show the upper-bound argument, and therefore the derivation of the liminf, needs another proof.","tokens_in":17949,"feed_emoji":"📉","tokens_out":9354,"duration_ms":85279,"temperature":0.7,"pith_summary":"This paper tries to prove that Laplace's first law of errors—the exponential decay of error frequencies—survives when diffusion happens in a frozen, disordered environment, not just in renewal or mean-field models. Working in a one-dimensional trap model where every site has its own fixed escape rate, the authors show from below and above that the disorder-averaged packet $\\langle P(x,t)\\rangle$ decays, for large displacement $|x|$ at fixed time, exactly like the ordered lattice: exponentially, with a logarithmic correction. The decay rate is set only by the mean and variance of the random rates, so the result is the same for exponential, Gaussian, or uniform trap-energy densities. If correct, this supplies the missing bridge between the widely observed exponential tails in single-particle tracking and a genuinely quenched random environment.","feed_headline":"Disorder can't erase Laplace's exponential diffusion tails","feed_subtitle":"In a random trap lattice, the tail shape is fixed by just two moments of the local escape rates.","key_machinery":"The load-bearing mechanism is the direct-path probability $\\mathrm{Prob}(\\to)$, the probability that a walker reaches $x$ by moving right at every step without ever turning back. Because a direct path visits each site once, its waiting times are independent even in a quenched environment, so its disorder-averaged Laplace transform factorizes as $\\langle\\widehat{\\psi}(s)\\rangle^{x}(1-\\langle\\widehat{\\psi}(s)\\rangle)/s$, where $\\langle\\widehat{\\psi}(s)\\rangle=\\langle R/(R+s)\\rangle$ is the Laplace transform of the averaged waiting-time density. Evaluating this by saddle point at $s^*=x/t$ converts the product into the exponential tail with moments $\\langle R\\rangle$ and $\\langle R^2\\rangle$. The paper then adds the single-turn paths, whose Laplace transforms contain $\\langle\\widehat{\\psi}(s)^2\\rangle$ because a revisited site must reuse the same random rate; these are down by one power of $1/x$ and give the correction series in Eq. (16).","core_discovery":"The paper's central claim is that the quenched disorder average satisfies\n$$\\lim_{|x|\\to\\infty}\\frac{-\\ln\\langle P(x,t)\\rangle}{|x|\\ln(2|x|/(e\\langle R\\rangle t))}=1,$$\nso the packet has a Laplace-like tail independent of the density of states $\\rho(E)$. The lower-bound half is derived: the direct, never-backtracking path contributes\n$$\\langle \\mathrm{Prob}(\\to)\\rangle \\sim \\frac{$e^{{-\\langle R^2\\rangle t/\\langle R\\rangle}}$}{\\sqrt{2\\pi|x|}}\\left(\\frac{e\\langle R\\rangle t}{2|x|}\\right)^{|x|},$$\nfrom a saddle point on the disorder-averaged waiting-time Laplace transform; this requires only $\\langle R\\rangle$ and $\\langle R^2\\rangle$. Single-turn backtracking paths add $1/|x|$ corrections involving $\\langle R^3\\rangle$, which capture the correlations created when a particle revisits a trap. Against this, the ordered infinite-temperature system provides an exponential upper bound, and the two bounds squeeze the packet into the same universal tail.","pith_inferences":["Beyond the paper: the same direct-path lower bound should hold in higher dimensions and for biased hopping, since the factorization argument only needs the path not to revisit sites; I would expect the same log-normalized limit with $\\langle R\\rangle$ replaced by the directional average rate.","Beyond the paper: the appearance of only $\\langle R\\rangle$ and $\\langle R^2\\rangle$ suggests the tail can be read as a large-deviation rate function for the empirical mean of $\\ln R$ along the path; proving this would connect Eq. (12) to standard large-deviation theory for products of random variables.","Beyond the paper: if the ordered-system upper bound can be replaced by a stochastic-ordering proof, the identity in Eq. (12) becomes a theorem rather than a physically argued identity; testing the inequality for intermediate temperatures and short times would show where such a proof must work."],"forward_implications":["For any trap-energy density $\\rho(E)$ with finite $\\langle R\\rangle$ and $\\langle R^2\\rangle$, the large-$|x|$ tail has the same functional form; temperature and disorder change only the prefactor through $\\langle R^2\\rangle/\\langle R\\rangle$ and the logarithmic scale $\\langle R\\rangle t$.","Because the tail shape is insensitive to $\\rho(E)$, exponential displacement tails in experiments cannot by themselves identify the microscopic trap statistics.","Backtracking paths contribute only at order $1/x$, so direct paths dominate the far tail; the $1/x$ corrections carry the quenched-disorder correlation signature in the packet.","The result holds even in the anomalous-diffusion regime of exponential disorder with $T<T_g$, where the mean-square displacement is $\\langle x^2\\rangle\\sim t^{T/T_g}$; the Laplace tail and the central Gaussian-like region coexist."],"supporting_citations":[{"why":"It documents exponential displacement tails near glass and jamming transitions, the experimental motivation for Laplace-like packets.","marker":"[1]"},{"why":"It supplies the ordered-lattice master equation and its modified-Bessel solution, Eq. (2), used for the upper bound.","marker":"[7]"},{"why":"It provides the trap-model and quenched-disorder background, including the density-of-states formulation and anomalous long-time limits.","marker":"[8]"},{"why":"It is the original 1774 memoir stating Laplace's first law of errors, the historical claim being revived.","marker":"[39]"},{"why":"It recent applies Laplace's first law to diffusive motion and frames the paper's target phenomenon.","marker":"[41]"},{"why":"It establishes universal exponential tails for packets in continuous-time random walks, the mean-field result this paper extends to quenched environments.","marker":"[42]"},{"why":"It introduces the jumping-time construction used to write the direct-path probability and its Laplace transform.","marker":"[43]"},{"why":"It provides the trap model with exponential density of states and glass phenomenology used in the examples.","marker":"[49]"},{"why":"It gives the large-order modified-Bessel asymptotics behind the ordered-system exponential tail.","marker":"[68]"},{"why":"It supplies the saddle-point approximation method used to invert the Laplace transforms for the bounds.","marker":"[69]"}],"fun_headline_variants":["Two moments fix Laplace tails in random trap lattices","Universal exponential tails despite arbitrary disorder","Laplace's first law revives in quenched random environments","Random trap disorder still yields Laplace exponential tails"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's upper bound assumes, without a proof, that for large $x$ the disordered packet is smaller than the ordered one, $\\langle P(x,t)\\rangle<P_{\\mathrm{ordered}}(x,t)$, because every rate is at most $r$; if this monotonicity ever fails, the identity in Eq. (12) is only a lower bound.","fun_headline_variants_meta":{"raw":{"variants":["Two moments fix Laplace tails in random trap lattices","Universal exponential tails despite arbitrary disorder","Laplace's first law revives in quenched random environments","Random trap disorder still yields Laplace exponential tails"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000315,"raw_usage":{"total_tokens":1765,"prompt_tokens":904,"completion_tokens":861,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":802}},"tokens_in":520,"tokens_out":861,"duration_ms":7578,"temperature":1.0,"reasoning_tokens":802,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:14:22.514143+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically compute the ratio in Eq. (12) for exponential trap-energy disorder with $T$ just above $T_g$, pushing to very large $|x|$ with rare-event sampling or by exact evaluation of path sums; a limiting ratio strictly below 1 while the limsup is 1 would falsify the identity. Alternatively, test the asserted inequality $\\langle P(x,t)\\rangle < e^{-rt}I_{|x|}(rt)$ directly at $T=T_g/2$; a single violation would show the upper-bound argument, and therefore the derivation of the liminf, needs another proof.","supporting_citations":[{"cited_title":"Chaudhuri, L","cited_arxiv_id":null,"evidence_quote":"It documents exponential displacement tails near glass and jamming transitions, the experimental motivation for Laplace-like packets."},{"cited_title":"Haus and K","cited_arxiv_id":null,"evidence_quote":"It supplies the ordered-lattice master equation and its modified-Bessel solution, Eq. (2), used for the upper bound."},{"cited_title":"Bouchaud and A","cited_arxiv_id":null,"evidence_quote":"It provides the trap-model and quenched-disorder background, including the density-of-states formulation and anomalous long-time limits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It is the original 1774 memoir stating Laplace's first law of errors, the historical claim being revived."},{"cited_title":"Hamdi, S","cited_arxiv_id":null,"evidence_quote":"It recent applies Laplace's first law to diffusive motion and frames the paper's target phenomenon."},{"cited_title":"Barkai and S","cited_arxiv_id":null,"evidence_quote":"It establishes universal exponential tails for packets in continuous-time random walks, the mean-field result this paper extends to quenched environments."},{"cited_title":"Monthus and J.-P","cited_arxiv_id":null,"evidence_quote":"It provides the trap model with exponential density of states and glass phenomenology used in the examples."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the large-order modified-Bessel asymptotics behind the ordered-system exponential tail."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the saddle-point approximation method used to invert the Laplace transforms for the bounds."}],"review_version":1}