REVIEW 2 major objections 5 minor 47 references
Extremal process of the local time of simple random walk on a regular tree
T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read 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…
desk verdict 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. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
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.
What would settle it
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.
Extended reading notes
Core claim
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 $k\to\infty$.
Load-bearing premise
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.
Editorial extensions
If this is right
- 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.
Reading between the lines
- 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.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
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.
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 (2)
- [Section 5.1 (proof of Corollary 1.6)] 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 5.1, Eq. (5.16) and comparison with (1.28)] 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.
minor comments (5)
- [Section 5.1, Eq. (5.16)] 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.
- [Eq. (1.28)] 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.
- [Lemma 5.3] 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 3.1] The phrase 'may behave somewhat strange' should read 'may behave somewhat strangely'.
- [Proof of Theorem 1.1, after (5.24)] 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.
Circularity Check
No significant circularity: the local-time decoration law is derived from the Ray-Knight isomorphism and a Lindeberg swap; the compressed identification with the BRW cluster law is an unproved detail, not a circular reduction.
full rationale
The paper does not fit any parameter to its target conclusion and then rename the fit as a prediction. The decoration law D is constructed as the law of chi_phi for phi sampled from the limit measure nu produced by Proposition 3.6 via Riesz representation; no equation defining D uses D_1 or the BRW cluster law as an input. The intensity measures Z and Z_t are characterized by the independent decomposition (1.22) and by convergence of the explicit measure (1.18); the proof of Theorem 1.1 reduces to Theorem 1.2 through Lemma 5.2 and external results [20,30,47]. The self-citations to [2] and [20] are load-bearing but not circular: they are prior published theorems about the maximum and barrier estimates, not about the decoration law or the extremal process. In addition, Section 5.3 explicitly notes that the proof of [20, Theorem 1.5] 'appears to be missing' and then supplies a proof in Proposition 5.5. The only real weakness is in Corollary 1.6: after deriving (5.16), the paper asserts '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 compresses the explicit comparison of nu (or the induced D) with the Aidekon/Madaule D_1 from (1.28), and the text says 'Leaving the details to the reader' before (5.16). That is a missing detail in the proof, not a circularity: D is not defined in terms of D_1, no fitted constant is relabeled as a prediction, and no self-citation is used to forbid alternatives.
Assumptions & free parameters
assumptions (4)
- domain assumption Pointwise Ray-Knight isomorphism: for each n and t there is a coupling of L_t with GFFs h and h~ such that h is independent of L_t and L_t(x)+h(x)^2 = (h~(x)+sqrt(t))^2 pointwise (Lemma 2.3).
- domain assumption Uniform tail asymptotic of the maximum local time: P(max_x sqrt(L_t(x)) - sqrt(t) - a_n(t) > u) = c_* (1+eps_{n,t,u}) u e^{-2 sqrt(log b) u} with lim_m sup_{t,u>=m} limsup_n |eps_{n,t,u}| = 0 (Prop 2.2, restated from [20, Prop 3.5]).
- domain assumption Clustering of the high level set: for each lambda > 0, lim_k limsup_n P( exists x,y in Gamma(lambda) with y in B_{n-k}(x) \ B_k(x) ) = 0 (Lemma 3.3, restated from [2, Prop 4.1]).
- domain assumption Upper tail bound for the maximum: P(max sqrt(L_t) >= sqrt(t)+a_n(t)+u) <= c(1+u) e^{-2 sqrt(log b) u - c' u^2/n} (Lemma 3.2, restated from [2, Prop 3.1] and [20, Thm 2.1]).
Cite this review
Pith. "Pith review of Extremal process of the local time of simple random walk on a regular tree." pith.science (2026). https://pith.science/paper/POVZDGGM
@misc{pith2026250609592,
author = {Pith},
title = {Pith review of: Extremal process of the local time of simple random walk on a regular tree},
year = {2026},
howpublished = {\url{https://pith.science/paper/POVZDGGM}},
note = {Machine review of arXiv:2506.09592}
}
abstract
We study a continuous-time simple random walk on a regular rooted tree of depth $n$ in two settings: either the walk is started from a leaf vertex and run until the tree root is first hit or it is started from the root and run until it has spent a prescribed amount of time there. In both cases we show that the extremal process associated with centered square-root local time on the leaves tends, as $n\to\infty$, to a decorated Poisson point process with a random intensity measure. While the intensity measure is specific to the local-time problem at hand, the decorations are exactly those for the tree-indexed Markov chain (a.k.a. Branching Random Walk or Gaussian Free Field) with normal step distribution. The proof demonstrates the latter by way of a Lindeberg-type swap of the decorations of the two processes which itself relies on a well-known isomorphism theorem.
Reference graph
Works this paper leans on
-
[20]
M. Biskup and O. Louidor (2024). A limit law for the most favorite point of simple random walk on a regular tree.Ann. Probab.52, no. 2, 502–544
work page 2024
-
[2]
Y. Abe (2018). Extremes of local times for simple random walks on symmetric trees.Electron. J. Probab. 23, paper no. 40, 41 pp
work page 2018
-
[30]
N. Eisenbaum, H. Kaspi, M. B. Marcus, J. Rosen, and Z. Shi (2000). A Ray-Knight theorem for symmet- ric Markov processes.Ann. Probab.28, 1781–1796
work page 2000
-
[47]
A. Zhai (2018). Exponential concentration of cover times.Electron. J. Probab.23no. 32
work page 2018
-
[1]
Y. Abe (2015). Maximum and minimum of local times for two-dimensional random walk.Electron. Commun. Probab.20, 14 pp
work page 2015
- [3]
-
[4]
Y. Abe, M. Biskup and S. Lee (2023). Exceptional points of discrete-time random walks in planar do- mains.Electron. J. Probab.28, paper no. 137, 1–45
work page 2023
-
[5]
L. Addario-Berry and B. Reed (2009). Minima in branching random walks.Ann. Probab.37, no. 3, 1044– 1079
work page 2009
Show all 47 references
-
[6]
A ¨ıd´ekon (2013)
E. A ¨ıd´ekon (2013). Convergence in law of the minimum of a branching random walk.Ann. Probab.41, no. 3A, 1362–1426
2013
-
[7]
A ¨ıd´ekon, J
E. A ¨ıd´ekon, J. Berestycki, E. Brunet and Z. Shi (2013). Branching Brownian motion seen from its tip. Probab. Theory Rel. Fields157, 405–451
2013
-
[8]
Arguin, A
L.-P . Arguin, A. Bovier, and N. Kistler (2011). Genealogy of extremal particles of branching brownian motion.Commun. Pure Appl. Math.64, 1647–1676
2011
-
[9]
Arguin, A
L.-P . Arguin, A. Bovier, and N. Kistler (2012). Poissonian statistics in the extremal process of branching brownian motion.Ann. Appl. Probab.22, no. 4, 1693–1711. 56 Y. ABE, M. BISKUP
2012
-
[10]
Arguin, A
L.-P . Arguin, A. Bovier, and N. Kistler (2013). The extremal process of branching Brownian motion. Probab. Theory Rel. Fields157, 535–574
2013
-
[11]
Barashkov, T.S
N. Barashkov, T.S. Gunaratnam and M. Hofstetter (2023). Multiscale coupling and the maximum of Ppφq2 models on the torus.Commun. Math. Phys.404, 833–882
2023
-
[12]
Bauerschmidt and M
R. Bauerschmidt and M. Hofstetter (2022). Maximum and coupling of the sine-Gordon field.Ann. Probab.50, no. 2, 455–508
2022
-
[13]
Belius, J
D. Belius, J. Rosen, and O. Zeitouni (2019). Barrier estimates for a critical Galton-Watson process and the cover time of the binary tree.Ann. Inst. Henri Poincar´ e Probab. Statist.55127-154
2019
-
[14]
Biskup (2020)
M. Biskup (2020). Extrema of the two-dimensional Discrete Gaussian Free Field. In: M. Barlow and G. Slade (eds.): Random Graphs, Phase Transitions, and the Gaussian Free Field. SSPROB 2017. Springer Proceedings in Mathematics & Statistics, 304 (2020) 163–407. Springer, Cham
2020
-
[15]
Biskup (2025)
M. Biskup (2025). Extremal properties of the random walk local time. arXiv:2502.09853
2025 arXiv
-
[16]
Biskup and H
M. Biskup and H. Huang (2023). A limit law for the maximum of subcritical DG-model on a hierarchi- cal lattice. arXiv:2309.09389
2023 arXiv
-
[17]
Biskup and O
M. Biskup and O. Louidor (2016). Extreme local extrema of two-dimensional discrete Gaussian free field.Commun. Math. Phys.345271–304
2016
-
[18]
Biskup and O
M. Biskup and O. Louidor (2020). Conformal symmetries in the extremal process of two-dimensional discrete Gaussian free field.Commun. Math. Phys.375, no. 1, 175–235
2020
-
[19]
Biskup and O
M. Biskup and O. Louidor (2018). Full extremal process, cluster law and freezing for two-dimensional discrete Gaussian free field.Adv. Math.330589–687
2018
-
[21]
Bramson (1978)
M. Bramson (1978). Maximal displacement of branching Brownian motion.Commun. Pure Appl. Math.31, no. 5, 531–581
1978
-
[22]
Bramson, J
M. Bramson, J. Ding, and O. Zeitouni (2016). Convergence in law of the maximum of nonlattice branch- ing random walk.Ann. Inst. Henri. Poincar´ e Probab. Stat.521897–1924
2016
-
[23]
Bramson, J
M. Bramson, J. Ding and O. Zeitouni (2016). Convergence in law of the maximum of the two- dimensional discrete Gaussian free field.Commun. Pure Appl. Math69, no. 1, 62–123
2016
-
[24]
X. Chen, C. Garban, and A. Shekhar (2023). The fixed points of branching Brownian motion.Probab. Theory Relat. Fields.185, 839–884
2023
-
[25]
The structure of extreme level sets in branching Brownian mo- tion
A. Cortines, L. Hartung, and O. Louidor (2019). Decorated random walk restricted to stay below a curve (supplementary material to “The structure of extreme level sets in branching Brownian mo- tion”.)
2019
-
[26]
Cortines, O
A. Cortines, O. Louidor and S. Saglietti (2021). A scaling limit for the cover time of the binary tree.Adv. Math.391, Paper No. 107974, 78 pp
2021
-
[27]
Dembo, Y
A. Dembo, Y. Peres, J. Rosen, and O. Zeitouni (2001). Thick points for planar Brownian motion and the Erd˝os-Taylor conjecture on random walk.Acta Math.186239-270
2001
-
[28]
Ding (2014)
J. Ding (2014). Asymptotics of cover times via Gaussian free fields: Bounded-degree graphs and gen- eral trees.Ann. Probab.42, 464–496
2014
-
[29]
J. Ding, R. Roy, and O. Zeitouni (2017). Convergence of the centered maximum of log-correlated Gauss- ian fields.Ann. Probab.45, no 6A, 3886–3928
2017
-
[31]
Erd˝os and S.J
P . Erd˝os and S.J. Taylor (1960). Some problems concerning the structure of random walk paths.Acta Math. Acad. Sci. Hungar.11, 137–162
1960
-
[32]
Hofstetter (2021)
M. Hofstetter (2021). Extreme local extrema of the sine-Gordon field. arXiv:2111.04842
2021 arXiv
-
[33]
Jego (2020)
A. Jego (2020). Thick points of random walk and the Gaussian free field.Electron. J. Probab.25(2020), paper no. 32, 39 pp
2020
-
[34]
Jego (2021)
A. Jego (2021). Critical Brownian multiplicative chaos.Probab. Theory Rel. Fields180, 495–552
2021
-
[35]
Kabluchko (2012)
Z. Kabluchko (2012). Persistence and equilibria of branching populations with exponential intensity. J. Appl. Probab.49, no. 1, 226–244
2012
-
[36]
Louidor and S
O. Louidor and S. Saglietti (2024). Tightness for the cover time of wired planar domains. arXiv: 2406.11034 LOCAL-TIME EXTREMAL PROCESS 57
2024 arXiv
-
[37]
Madaule (2015)
T. Madaule (2015). Maximum of a log-correlated Gaussian field.Ann. Inst. H. Poincar´ e Probab. Statist.51, no. 4, 1369–1431
2015
-
[38]
Madaule (2017)
T. Madaule (2017). Convergence in law for the branching random walk seen from its tip.J. Theor. Probab. 30, 27–63
2017
-
[39]
Maillard and B
P . Maillard and B. Mallein (2021). On the branching convolution equationE“ZfE.Electron. Com- mun. Probab.26, article no. 59, 1–12
2021
-
[40]
Mallein (2016)
B. Mallein (2016). Asymptotic of the maximal displacement in a branching random walk.Grad. J. Math.1, no.2, 92–104
2016
-
[41]
Paquette and O
E. Paquette and O. Zeitouni (2025). The extremal landscape for the CβE ensemble.Forum of Math., Sigma13, 1–126
2025
-
[42]
Rosen (2023)
J. Rosen (2023). Tightness for thick points in two dimensions.Electron. J. Probab.28, article no. 18, 1–45
2023
-
[43]
Schweiger (2020)
F. Schweiger (2020). The maximum of the four-dimensional membrane model.Ann. Probab.48, no. 2, 714–741
2020
-
[44]
Schweiger, W
F. Schweiger, W. Wu and O. Zeitouni (2024). Tightness of the maximum of Ginzburg-Landau fields. arXiv:2403.11500
2024 arXiv
-
[45]
Schweiger and O
F. Schweiger and O. Zeitouni (2024). The maximum of log-correlated Gaussian fields in random envi- ronments.Comm. Pure Appl. Math.77, no. 5, 2778–2859
2024
-
[46]
Wu and O
W. Wu and O. Zeitouni (2019). Subsequential tightness of the maximum of two dimensional Ginzburg- Landau fields.Electron. Commun. Probab.24, 1–12
2019
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.