{"id":"8b8dfbbc-45c8-4e9d-b81e-ea8bb5d841ac","arxiv_id":"2509.08559","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For Brox's diffusion in a Brownian environment, the heat kernel satisfies two-sided Gaussian-type bounds at small times and the annealed on-diagonal kernel decays like (log t)^{-2}, up to log-log factors.","lead":"This paper proves new heat kernel bounds for Brox's diffusion, a model of a particle diffusing in a one-dimensional Brownian random landscape. It establishes small-time estimates for each fixed environment and shows that the large-time return probability decays roughly as one divided by the square of the logarithm of time.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.8's proof has a swapped case analysis, so the quenched upper bound in Theorem 1.1 is not rigorously established as written.","rationale":"I read the paper in good faith. Lemma 2.5, which the reader identified as the weakest assumption, is actually correct: the self-referential appearance of V(S(x),R) on both sides of (2.22) is harmless because in every application the argument of ξ is fixed by the time t (e.g., (2R V)^{1/2} = (t/2)^{1/2} when t = 4 R V). The principal load-bearing concern is instead the proof of Lemma 3.8: the case analysis after Eq. (3.15) is genuinely swapped, so the paper does not establish (3.13) as written. Since Proposition 3.9 and Theorem 1.1 upper bound rely on (3.13), this is a real gap. I also noticed a minor overclaim in Lemma 4.7, where P(Γ_t ∩ ∩Θ_i^t) ≥ 1/(2 log t) is impossible because P(Γ_t) = 1/(1+2 log t) < 1/(2 log t); however, this only affects constants and can be repaired by choosing the error probabilities smaller, so the annealed lower bound rate survives. The central claims are likely correct and the issues appear fixable, so the reader's CONDITIONAL verdict is appropriate and no verdict change is needed.","tokens_in":36135,"tokens_out":35043,"duration_ms":486617,"concrete_test":"Rewrite the case analysis in Lemma 3.8 with the corrected regime split: if c0 R^2/t ≥ R Υ(1+R+|x|)^{1/α}, then N = c0 R^2/t and A = c4 sqrt(c0) − c5 ≤ −c5/2; if c0 R^2/t ≤ R Υ^{1/α}, then N = R Υ^{1/α} and A = c4 t^{1/2} Υ^{1/(2α)}/R^{1/2} − c5. Verify that the claimed bound (3.13) follows in both cases, adjusting C9 and C10 if necessary, and then recheck the constant in Proposition 3.9. If the corrected proof requires a different exponent in the positive term, the statement of Proposition 3.9 and Theorem 1.1 upper bound must be revised accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 3.8 (Eq. 3.13) is essential for Proposition 3.9 and therefore for the upper bound in Theorem 1.1. In the proof, after Eq. (3.15) one needs to bound A := c4 N^{1/2} t^{1/2}/R − c5 with N = max(c0 R^2/t, R Υ(1+R+|x|)^{1/α}). The paper claims that A ≤ −c5/2 when c0 R^2/t ≤ R Υ^{1/α}, and A ≤ c6 t^{1/2} Υ^{1/(2α)}/R^{1/2} − c5 when c0 R^2/t ≥ R Υ^{1/α}. This regime split is reversed. In the first case, N = R Υ^{1/α}, so N^{1/2} t^{1/2}/R = t^{1/2} Υ^{1/(2α)}/R^{1/2}, and the condition c0 R^2/t ≤ R Υ^{1/α} implies t^{1/2} Υ^{1/(2α)}/R^{1/2} ≥ sqrt(c0), giving A ≥ c4 sqrt(c0) − c5 = −c5/2, not ≤ −c5/2. The correct split is the opposite. Since the displayed case analysis is the only derivation of (3.13), and Proposition 3.9 invokes (3.13) directly, the proof as written has a genuine gap. An independent re-derivation suggests (3.13) may still be true with the cases swapped and adjusted constants, but the paper does not supply that argument.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper establishes quenched short-time heat kernel bounds and annealed large-time on-diagonal estimates for Brox's diffusion, the one-dimensional diffusion in Brownian environment introduced by Brox (1986). The main results are Theorem 1.1, giving two-sided quenched Gaussian-type bounds for p^X(t,x,y,ω) for t∈(0,1] with environment-dependent exponential corrections, and Theorem 1.3, giving annealed bounds on the on-diagonal kernel p(t,x,x) of order (log t)^{-2} times powers of log log t. The proofs use Brox's scale-transformation and time-change representation, volume estimates for the speed measure, and the theory of resistance forms for strongly recurrent Markov processes. A central ingredient is Lemma 2.5, which provides a self-referential volume estimate using Brownian oscillation, and Lemma 3.8, which gives an exponential exit-time estimate used for the quenched upper bound. The annealed arguments involve a decomposition of the environment according to valleys of the Brownian potential and several Brownian hitting/local-time estimates.","tokens_in":36398,"tokens_out":21881,"duration_ms":159879,"significance":"If correct, these are the first heat kernel bounds for Brox's diffusion, a model for which volume doubling fails at both small and large scales, so existing methods for diffusions in ergodic media do not apply. The annealed leading order (log t)^{-2} is consistent with Sinai-type localization and is new, and the explicit log log corrections are a useful refinement. The quenched bounds with environment-dependent corrections give a detailed picture of the heat kernel for fixed environments. The paper is well structured, uses appropriate tools (Brox's representation, resistance forms, Fernique's theorem, Brownian excursion estimates), and does not fit constants to the target estimates. The self-referential volume inequality in Lemma 2.5 is resolved by monotonicity of the oscillation function and does not appear circular. The paper would be a valuable contribution once the technical gap described below is repaired.","major_comments":[{"comment":"The case analysis for the quantity A := c4 N^{1/2} t^{1/2}/R - c5, with N = max(c0 R^2/t, R Υ(1+R+|x|)^{1/α}), is reversed. Since N^{1/2} t^{1/2}/R = max(√c0, t^{1/2} Υ^{1/(2α)}/R^{1/2}), one has A = c4 √c0 - c5 = -c5/2 when c0 R^2/t ≥ R Υ^{1/α}, and A = c4 t^{1/2} Υ^{1/(2α)}/R^{1/2} - c5 when c0 R^2/t ≤ R Υ^{1/α}. This is the opposite of the split displayed in the proof. Because this display is the only derivation of (3.13), and Proposition 3.9 invokes (3.13) directly, the quenched upper bound in Theorem 1.1 is not rigorously established as written. The lemma is plausibly repairable by swapping the two cases and adjusting constants, and the corrected split would still lead to a bound of the form (3.13), but the paper does not supply that argument and the constants need to be re-verified.","section":"Section 3, Lemma 3.8 (display after (3.15))"}],"minor_comments":[{"comment":"The derivation of the random constant C5(ω) in the lower bound of Theorem 1.1(i) is not spelled out. In particular, it is not immediate from (3.8) and (3.19) that the term t Υ^{2/α} log(t^{1/2} Υ^{1/α}) can be bounded by C5(ω) t^2 [log(2+|x|+|y|)]^{2/α} uniformly in t∈(0,1] and x,y∈R; please add the details of this comparison.","section":"Section 3, proof of Theorem 1.1"},{"comment":"The definition of ξ0(a,b) writes 'sup a≤s≤t≤b', but the intended quantity is the oscillation sup_{a≤s,t≤b} |W(s)-W(t)|; please fix this notation.","section":"Section 2, after (2.4)"},{"comment":"The proof states that the V^- version of (2.15) follows by exactly the same way, but no indication is given for the case x≠0 or for the interval (x-R,x]; a brief sentence confirming the extension would improve readability.","section":"Section 2, proof of Lemma 2.2"},{"comment":"There are several typos: 'some some notations' in Section 2; 'Winner medium' in reference [15] should be 'Wiener medium'; 'entironments' in reference [32] should be 'environments'; 'Cauchy-Schwartz' in the proof of Lemma 2.5 should be 'Cauchy-Schwarz'; 'consequenece' near the end of Section 4.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a significant open problem and the overall strategy is sound. The only major obstacle is the reversed case split in Lemma 3.8; I see no grounds for rejection, as the lemma is likely true with a corrected split. I recommend major revision so the authors can repair this step and re-verify the constants that feed into Proposition 3.9 and Theorem 1.1."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is genuinely new: it gives the first quenched heat kernel estimates for Brox's diffusion at small times, plus annealed on-diagonal bounds with the (log t)^{-2} leading order. The strategy is sound — time-change representation plus resistance-form machinery — and the volume estimates in Lemma 2.5, though self-referential, are handled without fitting constants. The main theorems are plausible and the proof structure is mostly coherent.\n\nThe problem is Lemma 3.8. The display after (3.15) claims two regime bounds on A, but the regimes are swapped. When N = R Υ^{1/α}, the condition implies A ≥ -c5/2, not ≤. When N = c0 R^2/t, A is identically -c5/2. So as written, the derivation of (3.13) fails. This is load-bearing: Proposition 3.9 invokes (3.13) for the quenched upper bound in Theorem 1.1, so that bound is not rigorously established in the manuscript. The fix is straightforward — swap the cases and adjust constants; then (3.13) follows with the same N and a Young-type argument. This is a genuine gap, but not a fatal one.\n\nThe reference list includes [38] (Perkowski–Willem) but the body never cites it; either cite it or drop it. It may overlap with the claimed novelty for heat kernels with distributional drift, so the authors should address that directly.\n\nFor the rest: the annealed lower bound uses an event decomposition with valleys and hitting probabilities that looks correct, and the log-log factors are not optimized, which is fine. The volume-doubling obstruction is real, and the oscillation-based technique is the right idea.\n\nI would send this to a serious referee. The main results are important and likely correct, but the paper needs revision — at minimum the Lemma 3.8 fix and citation cleanup.","headline":"Solid new heat kernel estimates for Brox's diffusion, but the case split in Lemma 3.8 is reversed and needs fixing before the quenched upper bound is rigorous.","tokens_in":36981,"tokens_out":4817,"would_cite":true,"duration_ms":36030,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G51","60G52","60J25","60J75"],"pacs":[],"model":"deepseek-v4-flash","headline":"Brox's diffusion has a Gaussian-type heat kernel at short times and an annealed on-diagonal decay of order 1/(log t)^2 up to log-log factors, despite its reference measure failing volume doubling at every scale.","keywords":["Brox diffusion","heat kernel estimates","quenched and annealed bounds","scale transformation","time change","resistance forms","Brownian potential","volume doubling failure"],"falsifier":"Take a fixed Brownian path $W$, a grid of points $x$ and radii $R$, compute $V(S(x),R)$ directly from $\\mu^Y$, and compare it with the two-sided bound of Lemma 2.5; any pair $(x,R)$ violating the inequality would overturn the quenched theorems. For the annealed claim, Monte-Carlo estimates of $\\mathbb{E}[p(t,0,0)]$ at $t=10^4,\\dots,10^8$ should remain inside the $(\\log t)^{-2}(\\log\\log t)^{-11}$ lower and $(\\log t)^{-2}(\\log\\log t)^{4+1/(2\\alpha)}$ upper envelopes.","tokens_in":35853,"feed_emoji":"📉","tokens_out":11868,"duration_ms":93613,"temperature":0.7,"pith_summary":"The paper seeks heat-kernel estimates for Brox's diffusion, the one-dimensional diffusion in a Brownian potential that is the continuous counterpart of Sinai's random walk. It proves that for every fixed environment and for short times $t\\le 1$, the transition density is bounded above and below by Gaussian-type expressions $t^{-1/2}e^{-C|x-y|^2/t}e^{W(y)}$ times random factors growing like $\\exp(\\pm C(\\omega)t^2[\\log(2+|x|+|y|)]^{2/\\alpha})$. It also proves that the annealed on-diagonal heat kernel obeys $(\\log t)^{-2}(\\log\\log t)^{-11}\\le \\mathbb{E}[p(t,x,x)]\\le (\\log t)^{-2}(\\log\\log t)^{4+1/(2\\alpha)}$ for large $t$. The interest is that the reference measure $e^{-W(x)}dx$ fails volume doubling at every scale, so standard machinery for diffusions in ergodic media cannot be used, and the proof introduces oscillation-based estimates for Brownian paths plus a valley decomposition for the large-time annealed behaviour.","feed_headline":"Brox diffusion: heat kernel decays like 1/(log t)^2","feed_subtitle":"Quenched Gaussian bounds hold for short times; annealed on-diagonal density is 1/(log t)^2 up to log-log factors.","key_machinery":"The Brox construction $X(t)=S^{-1}(B(T^{-1}(t)))$ with scale $S(x)=\\int_0^x e^{W(z)}dz$ and time change $T(t)=\\int_0^t e^{-2W(S^{-1}(B(s)))}ds$, together with the heat-kernel identity $p^X(t,x,y)=p^Y(t,S(x),S(y))$ for the time-changed Brownian motion $Y$. On $Y$, resistance-form estimates (Lemma 2.2 and Lemma 2.4) convert volume bounds into on-diagonal heat-kernel bounds, and Lemma 2.5 supplies the load-bearing self-referential volume bound $2Re^{-2W(x)}e^{-2\\xi(x,(2RV)^{1/2})}\\le V\\le 2Re^{-2W(x)}e^{2\\xi(x,(2RV)^{1/2})}$, with $V=V(S(x),R)$, where $\\xi$ is the oscillation of $W$ on an interval. The Hölder-type coefficient $\\Xi(x,r;\\omega)=\\sup|W(s)-W(t)|/|s-t|^\\alpha$ controls these oscillations and feeds both the short-time quenched bounds and the annealed valley analysis.","core_discovery":"The central claim is that Brox's diffusion, despite singular drift and the absence of volume doubling, has a heat kernel of Gaussian form at short times in every fixed environment, and an annealed on-diagonal density that decays like $(\\log t)^{-2}$ with explicit log-log-correction factors. The proof passes through the identity $p^X(t,x,y)=p^Y(t,S(x),S(y))$, where $Y$ is Brownian motion time-changed by the speed measure $\\mu^Y(dx)=e^{-2W(S^{-1}(x))}dx$, and then applies resistance-form theory for strongly recurrent Markov processes. The entire quenched argument hinges on the self-referential volume estimate of Lemma 2.5, which bounds $V(S(x),R)$ by $2Re^{-2W(x)}e^{\\pm 2\\xi(x,(2RV(S(x),R))^{1/2})}$; annealed large-time bounds come from a decomposition of environments into valleys of the Brownian potential and explicit Brownian hitting-time estimates.","pith_inferences":["The gap between the lower correction $(\\log\\log t)^{-11}$ and the upper correction $(\\log\\log t)^{4+1/(2\\alpha)}$ suggests the exact large-time annealed asymptotics may involve a central-limit order of $(\\log\\log t)^\\gamma$; refining the valley decomposition with local-time estimates could pin $\\gamma$.","The same scale/time-change and self-referential volume estimate should extend to one-dimensional diffusions in random potentials with Hölder regularity of order $\\alpha$, such as fractional Brownian motion with Hurst index $H<1/2$, predicting heat-kernel bounds with $\\alpha$-dependent powers of $\\log\\log t$.","A numerical check of the quenched bounds is directly feasible: for a discretized Brownian potential, pathwise heat-kernel ratios should fluctuate within the random envelopes $\\exp(\\pm C(\\omega)t^2[\\log(2+|x|+|y|)]^{2/\\alpha})$, making the oscillation effect observable in simulations.","The annealed upper bound's dependence on $1/(2\\alpha)$ suggests the method's cost grows as the Hölder exponent approaches $1/2$; this is likely an artifact of the technique rather than a feature of the diffusion, and a proof with a uniform exponent might exist."],"forward_implications":["For every fixed $\\alpha\\in(0,1/2)$, the quenched short-time bounds are two-sided and Gaussian, so the $t^{-1/2}e^{-C|x-y|^2/t}$ factor is sharp in every environment.","For finite time and large separation, Corollary 1.2 removes the random prefactors and yields two-sided Gaussian bounds $C_{13}t^{-1/2}e^{-C_{14}|x|^2/t}\\le p^X(t,0,x,\\omega)\\le C_{15}t^{-1/2}e^{-C_{16}|x|^2/t}$.","The annealed on-diagonal density is comparable to $(\\log t)^{-2}$ up to powers of $\\log\\log t$, matching the $(\\log t)^2$ displacement scale of Brox/Sinai diffusion.","The $e^{W(y)}$ factor in the quenched bounds is forced by symmetry with respect to $\\mu^X(dx)=e^{-W(x)}dx$, so the transition density is not symmetric in its spatial arguments."],"supporting_citations":[{"why":"Introduces Brox's diffusion as a continuous analogue of Sinai's random walk and gives the scale-transformation/time-change representation (1.2) that the whole paper is built on.","marker":"[15]"},{"why":"Supplies the resistance-form heat-kernel estimates for strongly recurrent Markov processes that convert volume bounds into on-diagonal heat-kernel bounds.","marker":"[10]"},{"why":"Provides the theory of resistance forms and heat-kernel estimates used for the time-changed Brownian motion Y.","marker":"[34]"},{"why":"Supplies the Brownian hitting-time and excursion identities used in the annealed upper and lower bounds and in the probability estimates for the valley decomposition.","marker":"[13]"},{"why":"Gives Dirichlet-form and time-change facts, including invariance of Green functions under time change, used to transfer estimates from Y back to X.","marker":"[23]"},{"why":"Provides the time-change theorems for symmetric Markov processes used to realize Y as a time change of Brownian motion.","marker":"[18]"},{"why":"Gives the energy estimate E(p,p)≤p/t used for off-diagonal and continuity bounds in the quenched estimates.","marker":"[5]"}],"fun_headline_variants":["Brox diffusion heat kernel: quenched Gaussian, annealed (log t)^-2","Quenched Gaussian and (log t)^-2 annealed: Brox heat kernel","Brox heat kernel: Gaussian short-time, (log t)^-2 large-time","Annealed (log t)^-2 decay and quenched Gaussian bounds for Brox","Brox diffusion heat kernel bounds without volume doubling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is Lemma 2.5's self-referential volume estimate: for every fixed Brownian potential, the volume $V(S(x),R)$ is trapped between $2Re^{-2W(x)}e^{-2\\xi(x,(2RV)^{1/2})}$ and $2Re^{-2W(x)}e^{2\\xi(x,(2RV)^{1/2})}$, with $\\xi$ the potential's oscillation at the radius $(2RV)^{1/2}$; if that inequality fails at any scale, the resistance-form machinery cannot be applied and both main theorems collapse.","fun_headline_variants_meta":{"raw":{"variants":["Brox diffusion heat kernel: quenched Gaussian, annealed (log t)^-2","Quenched Gaussian and (log t)^-2 annealed: Brox heat kernel","Brox heat kernel: Gaussian short-time, (log t)^-2 large-time","Annealed (log t)^-2 decay and quenched Gaussian bounds for Brox","Brox diffusion heat kernel bounds without volume doubling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001365,"raw_usage":{"total_tokens":5528,"prompt_tokens":927,"completion_tokens":4601,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":543,"completion_tokens_details":{"reasoning_tokens":4497}},"tokens_in":543,"tokens_out":4601,"duration_ms":29274,"temperature":1.0,"reasoning_tokens":4497,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:02:41.733429+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a fixed Brownian path $W$, a grid of points $x$ and radii $R$, compute $V(S(x),R)$ directly from $\\mu^Y$, and compare it with the two-sided bound of Lemma 2.5; any pair $(x,R)$ violating the inequality would overturn the quenched theorems. For the annealed claim, Monte-Carlo estimates of $\\mathbb{E}[p(t,0,0)]$ at $t=10^4,\\dots,10^8$ should remain inside the $(\\log t)^{-2}(\\log\\log t)^{-11}$ lower and $(\\log t)^{-2}(\\log\\log t)^{4+1/(2\\alpha)}$ upper envelopes.","supporting_citations":[{"cited_title":"Brox: A one-dimensional diffusion process in a Winner medium,Ann","cited_arxiv_id":null,"evidence_quote":"Introduces Brox's diffusion as a continuous analogue of Sinai's random walk and gives the scale-transformation/time-change representation (1.2) that the whole paper is built on."},{"cited_title":"Barlow, T","cited_arxiv_id":null,"evidence_quote":"Supplies the resistance-form heat-kernel estimates for strongly recurrent Markov processes that convert volume bounds into on-diagonal heat-kernel bounds."},{"cited_title":"Kumagai: Heat kernel estimates and parabolic Harnack inequalities on graphs and resistance forms,Publ","cited_arxiv_id":null,"evidence_quote":"Provides the theory of resistance forms and heat-kernel estimates used for the time-changed Brownian motion Y."},{"cited_title":"Borodin and P","cited_arxiv_id":null,"evidence_quote":"Supplies the Brownian hitting-time and excursion identities used in the annealed upper and lower bounds and in the probability estimates for the valley decomposition."},{"cited_title":"Fukushima, Y","cited_arxiv_id":null,"evidence_quote":"Gives Dirichlet-form and time-change facts, including invariance of Green functions under time change, used to transfer estimates from Y back to X."},{"cited_title":"Chen and M","cited_arxiv_id":null,"evidence_quote":"Provides the time-change theorems for symmetric Markov processes used to realize Y as a time change of Brownian motion."},{"cited_title":"Barlow:Diffusions on Fractals, Lectures on Probability Theory and Statistics (Saint-FLour, 1995), in: Lecture Notes in Math., vol","cited_arxiv_id":null,"evidence_quote":"Gives the energy estimate E(p,p)≤p/t used for off-diagonal and continuity bounds in the quenched estimates."}],"review_version":2}