{"id":"119e4e75-f132-4fea-8a64-8d7c95529ec9","arxiv_id":"2506.09592","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The extremal process of centered square-root local time on the leaves of a regular tree converges to a decorated Poisson point process with the same cluster law as the Gaussian Free Field.","lead":"This paper proves that the extreme values of the local time of a random walk on a regular tree converge to a decorated Poisson point process, whose cluster decoration law coincides with that of the Gaussian Free Field. The result places the model in the universality class of logarithmically correlated fields and is obtained by a Lindeberg-style swap based on the Ray-Knight isomorphism.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Corollary 1.6's identification of the local-time decoration D with the BRW decoration D_1 is left as a sketch: (5.16) is derived, but the equality of the decoration law ν there with D_1 is not proved.","rationale":"The paper is a serious, carefully written contribution and the main Theorems 1.1 and 1.2 are supported by detailed arguments, including the barrier estimates in Section 6 and the structured-process convergence in Theorem 3.1. I do not regard the pointwise Ray-Knight isomorphism as a paper gap: it is imported from [30] and [47] and is the right tool. The genuinely load-bearing soft spot is the proof of Corollary 1.6, which is the stated 'most important part of our conclusions': it asserts D = D_1, but after deriving (5.16) it jumps to the identification without comparing ν with D_1. This is not a fatal flaw; it is a routine but nontrivial reduction that should be written out. The reader's verdict of CONDITIONAL is therefore appropriate, and my analysis does not change it.","tokens_in":51145,"tokens_out":17305,"duration_ms":203243,"concrete_test":"Complete the omitted step of Corollary 1.6: from (5.16) and Proposition 3.6, write the Laplace functional of the decoration law ν, and compute the corresponding functional for D_1 from the recursion (1.26)-(1.28) using the BRW maximum asymptotics (5.6)-(5.7). Show that the two functionals agree for every continuous compactly supported f:R→[0,∞); if they differ, D ≠ D_1 and the universality claim fails, while if an extra hypothesis is needed, that hypothesis must be stated and proved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central universality claim is D = D_1. In Section 5.1 the authors derive the GFF structured extremal limit (5.16) with an unspecified decoration law ν from Proposition 3.6. They then assert: 'Proceeding along the same argument as in the reduction of Theorem 3.1 to Theorem 1.2 then shows that the cluster process of GFF is that defined in (3.13), thus identifying it with the cluster process of the local time.' This only identifies D with the law of χ_φ for φ ∼ ν; it does not establish that ν equals the D_1 in (1.28). To conclude, one must show that the decoration law extracted from the GFF/BRW extremal process is the same as the cluster law D_1 in the Aïdékon/Madaule limit (1.28), for instance by proving equality of the corresponding Laplace functionals. The manuscript cites [6,38] for the existence of D_1 but neither states nor proves the needed uniqueness/identification. The reader's weakest assumption, the pointwise Ray-Knight lemma, is a cited theorem and is not where the central claim is under-supported; the under-supported step is the omitted reduction in the proof of Corollary 1.6.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the extremal process of the centered square-root local time of a continuous-time simple random walk on a regular rooted b-ary tree of depth n. Two settings are treated: the walk started at a leaf and killed at the root, and the walk started at the root and run until it has spent a prescribed time there. The main results, Theorems 1.1 and 1.2, assert that the extremal process converges weakly to a decorated Poisson point process with a random intensity measure Z(dx) or Z_t(dx) times b e^{-2 sqrt(log b) h} dh, and with i.i.d. decorations sampled from a law D. The intensity is characterized via a decomposition in Corollary 1.3. The paper's central universality claim, Corollary 1.6, states that D equals the cluster law D_1 of the tree-indexed Branching Random Walk / Gaussian Free Field with normal step distribution. The proof strategy is to prove the root-started case via a structured extremal process, swap the last k generations of the local time for a GFF using the Ray-Knight isomorphism, control the resulting barrier estimates, and then derive the leaf-started case by a Markovian decomposition.","tokens_in":51322,"tokens_out":9339,"duration_ms":104430,"significance":"If the main theorems hold, this paper constitutes a substantial advance: it moves the local-time model on the regular tree from the level of the law of the maximum to the full extremal process, and it claims that the local structure of the extremal clusters is universal, coinciding with the BRW/GFF cluster law. The paper contains serious technical contributions, including the coupling in Lemma 4.1, the ballot-type estimates in Proposition 6.1, and the structured extremal-process framework of Theorem 3.1. The proof of the decorated PPP limit with an abstract decoration law appears to be carried out in detail. However, the advertised identification of that decoration law with D_1 is not completed: Corollary 1.6 is justified by a sketch that leaves a load-bearing reduction to the reader. Until that gap is closed, the paper establishes a decorated PPP limit with an unspecified decoration law, rather than the full universality statement.","major_comments":[{"comment":"The identification of the decoration law D with the BRW cluster law D_1 is the central universality claim, but the proof stops at Eq. (5.16). That display produces a limit for the GFF/BRW structured extremal process with an abstract decoration law ν taken from Proposition 3.6. The sentence 'Proceeding along the same argument as in the reduction of Theorem 3.1 to Theorem 1.2 then shows that the cluster process of GFF is that defined in (3.13)' only identifies the cluster process of the GFF with the law of χ_φ for φ sampled from ν; it does not prove ν = D_1, where D_1 is the cluster law appearing in the Aïdékon/Madaule limit (1.28). To close the argument one must prove uniqueness of the cluster law in the BRW extremal limit, or compute the Laplace functional of the cluster process from (5.16) and compare it with that of D_1. This is load-bearing: without it the paper proves a decorated PPP limit with an unspecified decoration law rather than the advertised universality of D.","section":"Section 5.1 (proof of Corollary 1.6)"},{"comment":"The comparison between (5.16) and (1.28) is also not immediate because the two statements describe different objects: (5.16) is a point process on [0,1] x R with a random measure W(dx), while (1.28) is a point process on R with a scalar shift α^{-1} log W. For the BRW on the regular tree one expects W(dx) = W Leb(dx) by symmetry, but this is not stated or proved. The marginalization from (5.16) to (1.28), and the uniqueness of the cluster law in that limit, must be spelled out; as written, this is part of the omitted reduction flagged above.","section":"Section 5.1, Eq. (5.16) and comparison with (1.28)"}],"minor_comments":[{"comment":"The expectation on the left-hand side of (5.16) is written as E_ρ, but the process η̃ is the BRW/GFF process under the measure P̃, not under the local-time measure P_ρ; this appears to be a typo and should be corrected.","section":"Section 5.1, Eq. (5.16)"},{"comment":"Please state the values of c_1, c_2 and α for the step distribution in (1.27), and in particular note that α = 2 sqrt(log b) and that the centering m̃_n in (5.2) is the same as that in (1.28); this is needed for the comparison in Corollary 1.6.","section":"Eq. (1.28)"},{"comment":"The parenthetical '(it appears that a factor 1/2 is missing there)' should be resolved explicitly, either by giving the corrected statement of Zhai's result or by removing the speculation.","section":"Lemma 5.3"},{"comment":"The phrase 'may behave somewhat strange' should read 'may behave somewhat strangely'.","section":"Section 3.1"},{"comment":"The proof that Z is diffuse away from 0 is not written; it follows from the decomposition (1.22) and Theorem 1.2, but should be stated explicitly.","section":"Proof of Theorem 1.1, after (5.24)"}],"recommendation":"major_revision","confidential_remarks":"The main gating issue is the proof of Corollary 1.6; the rest of the paper's technical apparatus appears sound and detailed. If the authors can provide a complete identification of ν with D_1, including the scalar-reduction of W(dx), this would be a strong paper. I would not ask the authors to re-prove the cited Ray-Knight theorem, but I would appreciate a precise statement of which parts of Lemma 4.1 are already in [47] and which parts are new. The remark that the proof of [20, Theorem 1.5] is missing is unusual and should be phrased carefully in the final version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a strong paper and the main theorems hold up on close reading. It proves the full extremal process for the regular-tree local time in both the leaf-started and root-started settings, and it identifies the decorations with those of the BRW/GFF. Earlier work only had the law of the maximum, so the extremal process, the leaf-started case, and the explicit characterization of the intensity via (1.22) are all genuinely new.\n\nThe proof is written at a high level of rigor. Theorem 3.1 and its two technical propositions are proved in detail, and the structured extremal-process machinery is used cleanly. I also credit the paper for noticing that a total-mass convergence claimed in [20] was missing and for supplying a proof in Proposition 5.5. The reliance on cited results from [2], [20], [30], and [47] is standard and not a real weakness.\n\nThe genuine soft spot is Corollary 1.6, and the stress-test note is right about where it sits. In Section 5.1 the authors derive the GFF/BRW structured limit (5.16) with an unspecified decoration law ν, then say that proceeding as in the reduction of Theorem 3.1 identifies the cluster process with the one in (3.13). That establishes that the GFF cluster law is ν. It does not by itself prove ν equals the Aïdékon–Madaule decoration law D1 from (1.28); one needs a uniqueness argument, for instance equality of Laplace functionals. This is a routine step for specialists, and I do not think it is load-bearing in the sense that the rest of the paper collapses without it, but the paper's central universality claim is exactly this equality, so leaving it to the reader is the wrong place to economize. A referee should ask for the missing argument.\n\nNo circularity or post-hoc fitting. The math is careful and the citations are honest. This paper deserves a serious referee and should be published after the Corollary 1.6 gap is filled. I would bring it to a reading group and cite it.","headline":"Strong, credible paper resolving the extremal process for regular-tree local time; the one real gap is that the headline identification of the decoration law with the BRW cluster law is asserted rather than proved.","tokens_in":51950,"tokens_out":2586,"would_cite":true,"duration_ms":33450,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J55","60G70","60F05","60J80"],"pacs":[],"model":"deepseek-v4-flash","headline":"For a simple random walk on a regular tree, the extremal process of centered square-root local time on the leaves converges to a decorated Poisson point process whose decoration law is exactly the cluster law of a branching random walk…","keywords":["extremal process","local time","regular tree","decorated Poisson point process","Ray-Knight isomorphism","branching random walk","cluster law","log-correlated fields"],"falsifier":"On a small tree, simulate the coupling in Lemma 2.3: for one vertex $x$, compare the conditional law of $\\tilde h(x)+\\sqrt{t}$ given the absolute values $\\{|\\tilde h(y)+\\sqrt{t}|\\}_y$ on both sides of the equality (2.13); any discrepancy at a vertex would violate Lemma 2.3 and invalidate Proposition 3.5. A cheaper check is to estimate empirically the cluster law of $\\sqrt{L_t}$ near a local maximum for growing $n$ and compare its Laplace transform to the branching random walk cluster law $D_1$; a difference at the scales $k^{1/3}\\le u\\le k^{12/13}$ appearing in Proposition 3.6 would falsify the claimed universality.","tokens_in":2116,"feed_emoji":"🌳","tokens_out":6781,"duration_ms":124778,"temperature":0.7,"pith_summary":"The paper studies the largest values of the time a random walk spends on the leaves of a regular rooted tree, in two settings: the walk started at a leaf and killed at the root, and the walk started at the root and run until a fixed occupation time. It proves that the entire extremal process—positions, centered values, and the clusters of nearby near-maximal values—converges to a decorated Poisson point process with a random intensity measure. The intensity is specific to the local-time problem, but the decoration law is exactly the cluster law of the tree-indexed branching random walk (equivalently, the Gaussian free field) with normal step distribution. This places the local-time extremes in the same universality class as other logarithmically correlated models, going beyond earlier results that identified only the law of the maximum.","feed_headline":"Local-time extremes on a regular tree match BRW cluster law","feed_subtitle":"The full cluster structure of near-maximal leaf local times matches branching random walks, not just the maximum.","key_machinery":"The central object is the structured extremal process $\\zeta_{n,k}^{(t)}$, which records, for each $k$-local maximum of $\\sqrt{L_t}$ on the leaves, its position, its centered height, and the full shape of $\\sqrt{L_t}$ around it encoded by the map $x\\mapsto\\sqrt{L_t(x)}-\\sqrt{L_t(x\\cdot)}$. The argument is carried by a Lindeberg-style swap: for fixed $t$, the last $k$ generations of $\\sqrt{L_t}$ are replaced by independent Gaussian-free-field increments, justified by the pointwise Ray-Knight isomorphism of Lemma 2.3, namely $L_t(x)+h(x)^2=(\\tilde h(x)+\\sqrt{t})^2$ with $h$ independent of $L_t$. Barrier events $B_{n,k}(x)$ restrict the path value $\\sqrt{L_t(z)}$ so that the swap error is of order $k^2/m_n$ and disappears in the double limit. Extracting the shape near Gaussian-free-field local maxima then yields the decoration law $\\nu$, and hence $D$, via Proposition 3.6 and inhomogeneous ballot estimates.","core_discovery":"For a continuous-time simple random walk on a $b$-ary rooted tree of depth $n$, the centered square-root local time on the leaves, $\\sqrt{\\ell_{\\tau_\\rho}(x)}-m_n$ for the leaf-started walk or $\\sqrt{L_t(x)}-m_n$ for the root-started walk, converges weakly, as $n\\to\\infty$, to a decorated Poisson point process. The intensity measure is random and of the form $Z(dx)\\,b\\,e^{-2h\\sqrt{\\log b}}\\,dh$, together with independent decorations drawn from a deterministic law $D$. The central identification is Corollary 1.6: the decoration law $D$ is the same as the cluster law $D_1$ of the branching random walk with step distribution i.i.d. $N(0,1/2)$; in other words, the shapes of the clusters hanging off each extreme local-time maximum are universal, while the intensity carries the geometry specific to local time. The proof achieves this through a Lindeberg-type swap that replaces the last $k$ generations of $\\sqrt{L_t}$ by increments of a Gaussian free field, with the error vanishing in the limit $n\\to\\infty$ followed by $k\\to\\infty$.","pith_inferences":["A natural testable extension, consistent with the paper's own expectation, is that the same decorated Poisson structure (and the same cluster law $D$) persists for root-started walks when the occupation time $t$ grows with $n$ as $t=o(n^2)$, with the centering adjusted to $\\sqrt{t}+a_n(t)$; this would confirm that the swap mechanism is not an artifact of fixed $t$.","The paper's swap recipe suggests a broader principle: any field that admits a pointwise Ray-Knight-type coupling to a Gaussian field, with matching local maxima, should inherit the branching random walk cluster law while keeping a problem-specific intensity measure.","The cascade representation (1.22) could be used to derive finer distributional information about the maximizer, such as the asymptotic distance from the starting leaf to the location of the maximum, going beyond the current convergence of the embedded position $\\theta_n(Y_n)$.","One could test the universality claim numerically on moderately large trees by estimating the cluster law of $\\sqrt{L_t}$ near local maxima and comparing its Laplace transform to the branching random walk cluster law $D_1$; a discrepancy at the scales $k^{1/3}\\le u\\le k^{12/13}$ of Proposition 3.6 would challenge the identification."],"forward_implications":["The full extremal process of local time has an explicit limit law, not merely the maximum, so questions about the second maximum, spatial clustering, and dependence inside clusters are answered.","The cluster law is universal: Corollary 1.6 identifies the decorations with the branching random walk cluster law $D_1$ for $N(0,1/2)$ steps, so local-time extrema and tree-indexed Gaussian free field extrema have the same local shape.","The random intensity measure $Z$ is characterized by the cascade decomposition (1.22), giving a self-similar description of where the extreme values sit and, via Corollary 1.4, identifying the limiting location of the maximizer.","For the root-started walk, conditioning on the event that the leaves are reached by time $t$ yields the same decorated Poisson structure with a diffuse measure $Z_t$, and the probability of reaching the leaves has the explicit limit $P(Z_t([0,1])>0)$.","The convergence of the total mass of the intensity measure is proved (Proposition 5.5), closing a gap in the earlier derivation of the limit law for the most favorite point."],"supporting_citations":[{"why":"Supplies the baseline limit law for the maximum of the local time and the tail asymptotics (Proposition 2.2) that the extremal-process proof extends.","marker":"[20]"},{"why":"Supplies the Second Ray-Knight isomorphism theorem that gives the coupling identity (2.13) in distribution.","marker":"[30]"},{"why":"Extends the Ray-Knight identity to a pointwise almost-sure coupling by sampling the signs of $\\tilde h(x)+\\sqrt{t}$, which Lemma 2.3 relies on.","marker":"[47]"},{"why":"Provides the barrier estimates and control of extreme local maxima (Lemmas 3.2 and 3.3) used in the swap and tightness arguments.","marker":"[2]"},{"why":"Establishes the convergence of the branching random walk extremal process and the existence of the cluster law $D_1$ quoted in (1.28).","marker":"[6]"},{"why":"Proves the convergence of the branching random walk seen from its tip, another source for the cluster law $D_1$ used in Corollary 1.6.","marker":"[38]"},{"why":"Gives the asymptotic tail (5.6) of the branching random walk maximum needed for the Gaussian-free-field cluster extraction.","marker":"[22]"},{"why":"Provides the clustering property (5.4) of extreme level sets of the branching random walk, used in the GFF analogue of the swap step.","marker":"[40]"},{"why":"Supplies exponential bounds on the branching random walk maximum used in the Section 6 barrier estimates.","marker":"[5]"}],"fun_headline_variants":["Extreme local-time clusters match branching random walk shapes","Local-time extremes: universal cluster law, random intensity","Extremal local time: clusters same as BRW, intensity tree-specific","Extreme leaf local-time clusters: universal shape, local intensity","Local-time extremes on trees: decorations match BRW, intensity random"],"cache_read_input_tokens":54016,"weakest_assumption_plain":"The load-bearing premise is the pointwise Ray-Knight coupling of Lemma 2.3, namely that $L_t(x)+h(x)^2=(\\tilde h(x)+\\sqrt{t})^2$ holds pointwise a.s. with $h$ independent of $L_t$; if this equality held only in distribution, the Lindeberg swap and the identification of $D$ with $D_1$ would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Extreme local-time clusters match branching random walk shapes","Local-time extremes: universal cluster law, random intensity","Extremal local time: clusters same as BRW, intensity tree-specific","Extreme leaf local-time clusters: universal shape, local intensity","Local-time extremes on trees: decorations match BRW, intensity random"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000914,"raw_usage":{"total_tokens":3927,"prompt_tokens":946,"completion_tokens":2981,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":562,"completion_tokens_details":{"reasoning_tokens":2895}},"tokens_in":562,"tokens_out":2981,"duration_ms":23032,"temperature":1.0,"reasoning_tokens":2895,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:45:20.317674+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On a small tree, simulate the coupling in Lemma 2.3: for one vertex $x$, compare the conditional law of $\\tilde h(x)+\\sqrt{t}$ given the absolute values $\\{|\\tilde h(y)+\\sqrt{t}|\\}_y$ on both sides of the equality (2.13); any discrepancy at a vertex would violate Lemma 2.3 and invalidate Proposition 3.5. A cheaper check is to estimate empirically the cluster law of $\\sqrt{L_t}$ near a local maximum for growing $n$ and compare its Laplace transform to the branching random walk cluster law $D_1$; a difference at the scales $k^{1/3}\\le u\\le k^{12/13}$ appearing in Proposition 3.6 would falsify the claimed universality.","supporting_citations":[{"cited_title":"Biskup and O","cited_arxiv_id":null,"evidence_quote":"Supplies the baseline limit law for the maximum of the local time and the tail asymptotics (Proposition 2.2) that the extremal-process proof extends."},{"cited_title":"Eisenbaum, H","cited_arxiv_id":null,"evidence_quote":"Supplies the Second Ray-Knight isomorphism theorem that gives the coupling identity (2.13) in distribution."},{"cited_title":"Zhai (2018)","cited_arxiv_id":null,"evidence_quote":"Extends the Ray-Knight identity to a pointwise almost-sure coupling by sampling the signs of $\\tilde h(x)+\\sqrt{t}$, which Lemma 2.3 relies on."},{"cited_title":"Abe (2018)","cited_arxiv_id":null,"evidence_quote":"Provides the barrier estimates and control of extreme local maxima (Lemmas 3.2 and 3.3) used in the swap and tightness arguments."},{"cited_title":"A ¨ıd´ekon (2013)","cited_arxiv_id":null,"evidence_quote":"Establishes the convergence of the branching random walk extremal process and the existence of the cluster law $D_1$ quoted in (1.28)."},{"cited_title":"Madaule (2017)","cited_arxiv_id":null,"evidence_quote":"Proves the convergence of the branching random walk seen from its tip, another source for the cluster law $D_1$ used in Corollary 1.6."},{"cited_title":"Bramson, J","cited_arxiv_id":null,"evidence_quote":"Gives the asymptotic tail (5.6) of the branching random walk maximum needed for the Gaussian-free-field cluster extraction."},{"cited_title":"Mallein (2016)","cited_arxiv_id":null,"evidence_quote":"Provides the clustering property (5.4) of extreme level sets of the branching random walk, used in the GFF analogue of the swap step."},{"cited_title":"Addario-Berry and B","cited_arxiv_id":null,"evidence_quote":"Supplies exponential bounds on the branching random walk maximum used in the Section 6 barrier estimates."}],"review_version":1}