Pith. sign in

REVIEW 2 major objections 5 minor 35 references

Moderately Heavy Extreme Values under Extreme Long Range Dependence

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper proves that under extreme long-range dependence, the empirical sup-measures of stationary infinitely divisible processes with moderately heavy tails converge to a non-Gumbel random limit, with extremal processes converging in…

desk verdict Strong convergence theorems for a hard LRD extreme-value problem, but the advertised non-Gumbel limit property rests on an invalid product/sum identity. read the letter →

arxiv 2505.23103 v2 pith:SP5PIJ5P submitted 2025-05-29 math.PR

Zao-Li Chen This is my paper
classification math.PR MSC 60G7060F0560G5760G5560K0560G18
keywords extremevaluetheorylongrangedependencesubexponentialtailsGumbelmaximumdomainofattractionrandomsup-measuresextremalprocessesnull-recurrentMarkovchainsstableregenerativesets
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper establishes functional limit theorems for the extreme values of a class of stationary infinitely divisible processes whose marginal tails are moderately heavy (subexponential and in the Gumbel maximum domain of attraction) and whose dependence is extremely long-range. For a memory parameter $\beta\in((m-1)/m,\ m/(m+1))$ with $m\geq 2$, the empirical sup-measures, suitably centered and scaled, converge to a non-Gumbel random sup-measure, and the induced extremal processes converge in $D(0,1)$ with the J1 topology. The unusual mechanism is an $(m+1)$-big-jump principle before the limit: extremes are caused by $m$ large jumps plus one extra jump of smaller magnitude, while in the limit only the $m$ large jumps survive. If correct, the results show that standard Gumbel-type approximations misdescribe extreme clusters in this dependence regime, and that different subexponential families within the Gumbel domain produce genuinely different limit laws.

What carries the argument

The load-bearing object is the random sup-measure $M_\beta$ on $[0,1]$, defined by $\eta(t)=\sum_{j\geq 1}-\log\Gamma_j\,\mathbf{1}(t\in R_j)$ when exactly $\ell_\beta=\max\{\ell\in\mathbb{N}:\ell<1/(1-\beta)\}$ of the random sets $R_j=(Q_j+Z_j)\cap[0,1]$ contain $t$, and $\eta(t)=-\infty$ otherwise; here $\Gamma_j$ are arrival times of a unit-rate Poisson process, $Q_j$ has density $(1-\beta)q^{-\beta}$ on $[0,1]$, and $Z_j$ are independent $\beta$-stable regenerative sets (ranges of $\beta$-stable subordinators). Its role is to serve as the universal limit object for the extreme values of the process. The argument connecting the process to $M_\beta$ is the series representation $X_t=\sum_{j\geq 1} V(w_n/\Gamma_j)\,\mathbf{1}(t\in I_{j,n})$, where $I_{j,n}$ are return-time sets to zero of independent null-recurrent Markov chains whose renewal tail is $F_\beta(n)=n^{-\beta}L_\beta(n)$, and $V$ is the quantile function of the L\'evy tail. A transfer theorem (Theorem 4.5) shows that the cardinality, capacity, and waiting-time statistics of intersections of these return-time sets converge to limits involving a Mittag-Leffler process $Z_{\beta_*}(1-Q_{\beta_*})$, the escape probability $p_{\mathrm{escape}}$, and an exponential variable; these limits drive the lower and upper comparisons that yield the main convergences.

What would settle it

Simulate the series representation (5.10) for a moderately heavy L\'evy tail such as $H_\#(x)=\exp(-(\log x)^\gamma)$ with $\beta=0.7$ (so $m=2$), and compare the empirical distribution of $(M_n([0,1])-b_n)/a_n$ with many independent samples of $M_\beta([0,1])$ from (3.11)--(3.13); if the right tails differ systematically, for instance if the simulation shows a Gumbel tail for $e^{(M_n([0,1])-b_n)/a_n}$ or a different exponent than the Poisson-overlap construction predicts, then the identification in Theorem 5.3 is wrong.

Watch

Extended reading notes

Core claim

The central claim is Theorem 5.3--5.4: for $\beta\in((m-1)/m,\ m/(m+1))$ with $m\geq 2$, the empirical sup-measures $M_n(B)=\max_{t\in nB} X_t$ of the stationary process defined through an infinitely divisible random measure over a null-recurrent Markov chain, centered by $b_n=mV(w_n)+V(\vartheta_n)$ and scaled by $a_n=h\circ V(w_n)$, converge weakly in $\mathrm{SM}([0,1])$ to the random sup-measure $M_\beta$ of (3.13). The empirical extremal processes $E_n(t)=\max\{X_i: 0\leq i\leq \lfloor nt\rfloor\}$ converge in $D(0,1)$ with the J1 topology to $E_\beta(t)=M_\beta([0,t])$. The limit is built from a Poisson point process marked by unit-rate arrival times, variables $Q_j$ with density $(1-\beta)q^{-\beta}$, and independent $\beta$-stable regenerative sets $R_j$; its sup-derivative sums $-\log\Gamma_j$ only at points covered by exactly $\ell_\beta=\max\{\ell<1/(1-\beta)\}$ of the sets, and is $-\infty$ elsewhere. For $\beta>1/2$, $M_\beta$ is not of Gumbel type: if it were, $e^{M_\beta}$ would be $1$-Fr\'echet, but $e^{M_\beta}$ has the distribution of the stable-regenerative sup-measure $M_{1,\beta}$, which is strictly non-Fr\'echet because the regenerative sets overlap. The proof replaces the process by a compound-Poisson series over return-time sets of independent null-recurrent Markov chains and sandwiches the empirical measures between lower and upper bounds whose leading terms converge to $M_\beta$.

Load-bearing premise

The proof rests on a uniform bound on how frequently the Markov chain's return probabilities can exceed their tail distribution, namely $\sup_{n\geq 0} n\,P(\varphi=n\mid Y_0=0)/F_\beta(n)<\infty$; if this bound fails, the estimates that keep simultaneous return-time clusters from blowing up break down and the main convergence is unsupported.

Editorial extensions

If this is right

  • For $\beta\in(1/2,1)$, the limit extremal process $E_\beta(t)=M_\beta([0,t])$ is stochastically smaller than the time-changed Gumbel process $E_0(t^{1-\beta})$; the strict time-change identity of the $\beta\leq 1/2$ case fails.
  • Before the limit, extremes are generated by $m+1$ big values, $m$ of order $V(w_n)$ and one of the smaller order $V(\vartheta_n)$, while in the limit only the $m$-big-jump contributions remain, so the phase transitions at $m/(m+1)$ persist in the moderately heavy regime.
  • The normalization $b_n=mV(w_n)+V(\vartheta_n)$ and $a_n=h\circ V(w_n)$ quantifies the two scales: the $m$ main jumps set the center, the extra jump contributes only through the smaller scale $\vartheta_n$, and the slowly varying part of the renewal tail enters through $V(\vartheta_n)$.
  • The empirical extremal processes converge in $D(0,1)$ with the J1 topology, so not only the sizes but also the jump locations of the limiting cluster shapes converge.
  • The results settle the problem left open for $\beta>1/2$: the limit objects for moderately heavy tails are genuinely different from both the Gumbel case and the regularly varying case in the same dependence range.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension the paper leaves implicit is that replacing the indicator integrand in the stochastic integral by a general function $f$ should yield the same $M_\beta$ limit; if a smooth $f$ produced a different sup-measure, the phenomenon would be an artifact of the indicator constraint.
  • Practically, the non-Gumbel limit implies that tail extrapolation for such series from Gumbel quantiles will misrepresent extreme cluster maxima; simulations with log-normal L\'evy tails and $\beta$ near $0.7$ could detect this by comparing empirical cluster maxima against the law of $M_\beta([0,1])$.
  • At the boundary $\beta=m/(m+1)$, where the proof breaks down, the paper conjectures a different clustering mode; identifying that boundary limit could interpolate between the $m$- and $(m+1)$-big-jump regimes and sharpen the phase diagram.
  • Because the limit depends on the slowly varying factor $L_\beta$ only through $\vartheta_n$, the qualitative cluster shape may be universal across the moderately heavy family, with the tail class entering only through the quantile scale; this could be checked by comparing log-normal and super-log-normal examples at the same $\beta$.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper studies stationary infinitely divisible processes built from a null-recurrent Markov chain, with moderately heavy subexponential margins in the Gumbel maximum domain of attraction. For a memory parameter beta in ((m-1)/m, m/(m+1)), it proves functional extremal limit theorems: after centering by b_n = m V(w_n) + V(vartheta_n) and scaling by a_n = h o V(w_n), the empirical sup-measures converge in SM([0,1]) to a random sup-measure M_beta defined through a Poisson process of shifted stable regenerative sets (Theorem 5.3), and the associated extremal processes converge in D(0,1) under the J1 topology (Theorem 5.4). The paper also claims that M_beta is non-Gumbel for beta > 1/2, and the abstract advertises 'non-Gumbel limit objects' as a central contribution.

Significance. If the convergence theorems are correct, they form a substantial extension of Chen--Samorodnitsky [CS22] to the regime beta > 1/2 and provide the Gumbel-MDA analogue of the phase transitions found by Samorodnitsky--Wang [SW19] for regularly varying margins. The proof strategy is a genuine strength: the series representation, the renewal-theoretic capacity estimates, and the systematic lower/upper bound comparisons are developed in detail and the main convergence argument does not depend on the contested non-Gumbel identity. However, the advertised qualitative conclusion that the limit objects are non-Gumbel currently rests on an invalid algebraic identity, so the paper's framing needs repair even though the weak-convergence results appear sound.

major comments (2)
  1. [Section 3.2.2] The proof that H_beta is not of Gumbel type is invalid. The paper claims that by (3.12)-(3.13), e^{M_beta} has the same distribution as M_{1,beta}([0,1]) = sup_t sum_j Gamma_j^{1/alpha} 1(t in R_j) with alpha = 1. For beta > 1/2, l_beta >= 2, so on the event that a point t belongs to exactly l_beta sets R_j, eta(t) = sum_{j: t in R_j} (-log Gamma_j) and therefore e^{eta(t)} = prod_{j: t in R_j} Gamma_j^{-1}. In contrast, M_{1,beta}([0,1]) is the supremum over t of sum_{j: t in R_j} Gamma_j (when alpha = 1). A product of l_beta >= 2 Gamma_j^{-1} variables is not equal in distribution to their sum; for l_beta = 2, the supremum over an overlap region of Gamma_i^{-1} Gamma_j^{-1} and of Gamma_i^{-1} + Gamma_j^{-1} have different laws. Thus the identity e^{M_beta} d= M_{1,beta}([0,1]) is false. The weak-convergence proofs of Theorems 5.3 and 5.4 do not use this identity, so those theorems are not damaged, but the title and abstract's 'non-Gumbel' claim is unsupported by the only proof offered. Proposition 3.4 does not fill the gap: a location-shifted extremal process E_0(t^{1-beta}) - delta with delta > 0 also satisfies the strict inequalities stated there, since P(E_0(t^{1-beta}) - delta <= x) = exp(-e^{-delta} t^{1-beta} e^{-x}) > exp(-t^{1-beta} e^{-x}). A valid proof of non-Gumbelness, or a reformulation of the paper's claims, is required.
  2. [Section 4.3.2] The definition of p_escape in (4.30) depends on the claim that the intersection of m+1 independent renewal ranges is almost surely finite, which is justified by the summability of sum_n u_beta(n)^{m+1}. The manuscript states 'The key is to check that ... We leave it to the reader.' This is a load-bearing step because p_escape enters Theorem 4.5 and hence the centering constants in Theorem 5.3. The verification is elementary under (4.8) -- u_beta(n) ~ n^{beta-1}/C and (m+1)(beta-1) < -1 follows from beta < m/(m+1) -- but it should be written out rather than delegated, since the constant p_escape is part of the limit in (4.23).
minor comments (5)
  1. [Section 5.4.2] After Eq. (5.49), the phrase 'as n → 0' should read 'as n → ∞'.
  2. [Section 4.3.2] The heading 'Simutaneous Renewal' contains a typo; it should be 'Simultaneous Renewal'.
  3. [Proposition 5.8] The notation 'lim_{rho->0} k_0->infty limsup' in the statement of Proposition 5.8 is ambiguous; please specify the order of the iterated limits explicitly.
  4. [Theorem 4.5] The display contains stray '= = = =⇒' formatting artifacts and should be typeset using standard weak-convergence arrows.
  5. [Section 3.2.2] In the comparison with M_{alpha,beta}, the substitution alpha = 1 should be written explicitly as sum_j Gamma_j 1(t in R_j); as printed, the formula 'Gamma_j^{1/alpha}' for alpha = 1 is confusing because Gamma_j are arrival times of a unit-rate Poisson process, not inverse-Gamma or Pareto marks.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Mβ is constructed a priori and the convergence in Theorem 5.3 is proved from independent renewal estimates; self-citations to [CS22] are not load-bearing.

full rationale

Walking the derivation chain from Definition 2.4 through Section 4 and the proofs of Theorems 5.3/5.4, I find no step in which a prediction reduces to an input by construction. The limit Mβ is defined in Section 3.2.2 before any empirical process is introduced, using an independent Poisson point process and stable regenerative sets, so the weak convergence (5.6) is a genuine theorem rather than a definition. The proof uses the series representation (5.10)-(5.11), constructs the lower bound (5.18) and upper bound (5.30), and shows after centering and scaling that the gap between them vanishes; the limit of the lower bound is identified via (5.14), (4.13), and Lemma 4.3, not by assuming Theorem 5.3. No parameter is fitted to force the conclusion, and the normalizations bn=mV(wn)+V(ϑn) and an=h∘V(wn) are standard quantile/auxiliary scalings. The paper does cite the author's own [CS22] (for example, stationarity/self-affinity of Mβ and the wandering-rate asymptotic) and also relies heavily on [SW19]; these are published, parameter-free results with stated assumptions that do not include the present target theorem, so under the stated rules they are independent support and do not raise the circularity score. I do flag, outside the circularity question, a correctness defect in the 'Non-Gumbel Distributions' paragraph: the identity 'by (3.12) and (3.13), e^{Mβ} d= M_{1,β}([0,1])' is algebraically unsupported for β>1/2, since e^{η(t)} is a product of Γ_j^{-1} over the ℓβ sets covering t while M_{1,β} is a supremum of sums of Γ_j. That concerns the advertised non-Gumbel conclusion, not the convergence mechanism, so it does not make the derivation circular.

Assumptions & free parameters 0 free parameters · 6 assumptions · 1 invented entities

The central theorem is conditional on two explicit domain assumptions: the null-recurrent Markov chain with renewal tail (Assumption 4.1) and the infinitely divisible random measure with moderately heavy Lévy tail (Assumption 5.1). The class of moderately heavy tails is newly defined in Definition 2.4 with technical Karamata-style conditions; the paper shows log-normal and super-log-normal tails satisfy it. No fitted parameters appear; the limit object M_β is a constructed entity whose validity is established by the paper's own proofs.

assumptions (6)
  • standard math Karamata's theorem and regular variation theory, including Γ-variation and Π-variation of Gumbel MDA quantile functions.
    Used throughout Section 2 to derive quantile expansions (2.13), (5.14), and the separation estimates in Section 5.
  • standard math Pitman's criterion for subexponentiality (Prop 2.1) and the infinite-divisibility subexponentiality result of Embrechts, Goldie, and Veraverbeke (Prop 2.2).
    Justify that the class of moderately heavy tails is subexponential and can appear as Lévy tails.
  • standard math Theory of stable subordinators and stable regenerative sets: self-similarity, Hausdorff dimension, intersection dichotomy (Lemma 3.1 from SW19).
    The limit object M_β is built from these sets; the dichotomy controls the number of overlapping sets.
  • standard math Renewal theory for infinite-mean renewal processes, including Doney 1997, Whitt 2002, Spitzer capacity theory, and Durrett's subadditive ergodic theorem.
    Underpins the estimates on return-time sets I_{j,n}, the wandering rate w_n, the scalings ϑ_n, and the capacity convergence in Proposition 4.9.
  • domain assumption Assumption 4.1: (Y,μ,τ) is an irreducible, aperiodic, null-recurrent Markov chain with return distribution F_β(n)=n^{-β}L_β(n) and the uniform bound (4.7).
    Defines the canonical LRD dynamics; the theorems are stated only for this class.
  • domain assumption Assumption 5.1: N is an independently scattered infinitely divisible random measure with constant local characteristic triple, and its Lévy tail is moderately heavy as in Definition 2.4.
    Defines the marginal tail class; the new Definition 2.4 is a technical but broad replacement of CS22's Assumption 3.1.
invented entities (1)
  • Random sup-measure M_β for β>1/2 (Eq. 3.13)
    purpose: Candidate limit object for empirical sup-measures of long-range dependent moderately heavy processes; characterized by exactly ℓ_β overlapping stable regenerative sets.
    New object introduced by the paper; there is no external falsifiable handle outside the paper's own theorems. It is defined mathematically, and its non-Gumbel character is proved by contradiction in Section 3.2.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Moderately Heavy Extreme Values under Extreme Long Range Dependence." pith.science (2026). https://pith.science/paper/SP5PIJ5P

@misc{pith2026250523103,
  author       = {Pith},
  title        = {Pith review of: Moderately Heavy Extreme Values under Extreme Long Range Dependence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SP5PIJ5P}},
  note         = {Machine review of arXiv:2505.23103}
}
read the original abstract

We consider stationary sequences whose marginal tail is subexponential and lies in the Gumbel Maximum domain of attraction. Due to the extremely strong dependence, their extreme values are caused by multiple big values and are clustered in the large scale with fractal features. We establish functional extremal limit theorems with non-Gumbel limit objects to characterize these delicate phenomena.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sentence := #2 '...

    work page

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

    work page

  3. [3]

    L\'evy processes , volume 121 of Cambridge Tracts in Mathematics

    Jean Bertoin. L\'evy processes , volume 121 of Cambridge Tracts in Mathematics . Cambridge University Press, Cambridge, 1996

    work page 1996

  4. [4]

    Convergence of Probability Measures (Second Edition)

    Patrick Billingsley. Convergence of Probability Measures (Second Edition) . Wiley Series in Probability and Statistics. John Wiley & Sons, Inc., New York, NY, 1999

    work page 1999

  5. [5]

    A remarkable example on clustering of extremes for regularly-varying stochastic processes

    Shuyang Bai, Kulik Rafał, and Yizao Wang. A remarkable example on clustering of extremes for regularly-varying stochastic processes. arXiv:2409.17966

    work page Pith review arXiv

  6. [6]

    Tail processes for stable-regenerative multiple-stable model

    Shuyang Bai and Yizao Wang. Tail processes for stable-regenerative multiple-stable model. Bernoulli , 29(4):3255--3279, 2023

    work page 2023

  7. [7]

    Phase transition for extremes of a family of stationary multiple-stable processes

    Shuyang Bai and Yizao Wang. Phase transition for extremes of a family of stationary multiple-stable processes. Ann. Inst. H. Poincaré Probab. Statist. , 60(3):2157--2193, 2024

    work page 2024

  8. [8]

    Extreme value theory for long-range-dependent stable random fields

    Zaoli Chen and Gennady Samorodnitsky. Extreme value theory for long-range-dependent stable random fields. J. Theoret. Probab. , 33(4):1894--1918, 2020

    work page 1918

Show all 35 references
  1. [9]

    Extreme value theory for long-range-dependent stable random fields

    Zaoli Chen and Gennady Samorodnitsky. Extreme value theory for long-range-dependent stable random fields. Stochastic Process. Appl. , 145:86--116, 2022

    2022

  2. [10]

    Extreme value theory

    Laurens de Haan and Ana Ferreira. Extreme value theory . Springer Series in Operations Research and Financial Engineering. Springer, New York, 2006. An introduction

    2006

  3. [11]

    R. A. Doney. One-sided local large deviation and renewal theorems in the case of infinite mean. Probab. Theory Related Fields , 107:451--465, 1997

    1997

  4. [12]

    Probability: Theory and Examples , volume 49 of Cambridge Series in Statistical and Probabilistic Mathematics

    Rick Durrett. Probability: Theory and Examples , volume 49 of Cambridge Series in Statistical and Probabilistic Mathematics . Cambridge University Press, Cambridge, fifth edition, 2019

    2019

  5. [13]

    Goldie, and No\"el Veraverbeke

    Paul Embrechts, Charles M. Goldie, and No\"el Veraverbeke. Subexponentiality and infinite divisibility. Z. Wahrsch. Verw. Gebiete , 49(3):335--347, 1979

    1979

  6. [14]

    An introduction to heavy-tailed and subexponential distributions

    Sergey Foss, Dmitry Korshunov, and Stan Zachary. An introduction to heavy-tailed and subexponential distributions . Springer, New York, 2013. Springer Ser. Oper. Res. Financ. Eng

    2013

  7. [15]

    T. E. Harris and H. E. Robbins. Ergodic theory of M arkov chains admitting an infinite invariant measure. Proc. Nat. Acad. Sci. U.S.A. , 39:860--864, 1953

    1953

  8. [16]

    Heavy-tailed time series

    Rafał Kulik and Philippe Soulier. Heavy-tailed time series . Springer, New York, 2020. Springer Ser. Oper. Res. Financ. Eng

    2020

  9. [17]

    Extremes and Recurrence in Dynamical Systems

    Valerio Lucarini, Davide Faranda, Ana Cristina Moreira Freitas, Jorge Milhazes Freitas, Tobias Kuna, Mark Holland, Matthew Nicol, Mike Todd, and Sandro Vaienti. Extremes and Recurrence in Dynamical Systems . John Wiley & Sons, Inc, Hoboken, New Jersey, 2016

    2016

  10. [18]

    Time-changed extremal process as a random sup measure

    C\' e line Lacaux and Gennady Samorodnitsky. Time-changed extremal process as a random sup measure. Bernoulli , 22(4):1979--2000, 2014

    1979

  11. [19]

    Theory of random sets , volume 87 of Probability Theory and Stochastic Modelling

    Ilya Molchanov. Theory of random sets , volume 87 of Probability Theory and Stochastic Modelling . Springer-Verlag, London, second edition, 2017

    2017

  12. [20]

    Max-stable random sup-measures with comonotonic tail dependence

    Ilya Molchanov and Kirstin Strokorb. Max-stable random sup-measures with comonotonic tail dependence. Stochastic Process. Appl. , 126(9):2835--2859, 2016

    2016

  13. [21]

    Extreme Value Theory for Time Series: Models with Power-Law Tails

    Thomas Mikosch and Olivier Wintenberg. Extreme Value Theory for Time Series: Models with Power-Law Tails . Springer, Cham, 2024. Springer Ser. Oper. Res. Financ. Eng

    2024

  14. [22]

    O'Brien, Paul J

    George L. O'Brien, Paul J. J. F. Torfs, and Wim Vervaat. Stationary self-similar extremal processes. Probab. Theory Related Fields , 87(1):97--119, 1990

    1990

  15. [23]

    E. J. G. Pitman. Subexponential distribution functions. J. Austral. Math. Soc. Ser. A , 29(3):337--347, 1980

    1980

  16. [24]

    Pruitt and S

    W. Pruitt and S. Taylor. Sample path properties of processes with stable components. Z. Wahrscheinlichkeitstheorie verw Gebiete , 12:267--289, 1969

    1969

  17. [25]

    Sidney I. Resnick. Extreme values, regular variation and point processes . Springer Series in Operations Research and Financial Engineering. Springer, New York, 2008. Reprint of the 1987 original

    2008

  18. [26]

    Classes of mixing stable processes

    Jan Rosi\'nski and Gennady Samorodnitsky. Classes of mixing stable processes. Bernoulli , 2(4):365--377, 1996

    1996

  19. [27]

    Stochastic processes and long range dependence

    Gennady Samorodnitsky. Stochastic processes and long range dependence . Springer, Cham, 2016. Springer Ser. Oper. Res. Financ. Eng

    2016

  20. [28]

    Lévy Processes and Infinitely Divisible Distributions (Revised Edition)

    Ken-iti Sato. Lévy Processes and Infinitely Divisible Distributions (Revised Edition) . Cambridge University Press, Cambrige, UK, 2013. Cambridge Studies in Advanced Mathematics 68

    2013

  21. [29]

    A note on generalized inverses

    Anthony Saint-Criq. A note on generalized inverses. 2014

    2014

  22. [30]

    Principles of Random Walk

    Frank Spitzer. Principles of Random Walk . Springer, New York, 1964. Graduate Texts in Mathematics 34

    1964

  23. [31]

    Gennady Samorodnitsky and Murad S. Taqqu. Stable non- G aussian random processes . Stochastic Modeling. Chapman & Hall, New York, 1994. Stochastic models with infinite variance

    1994

  24. [32]

    Extremal theory for long range dependent infinitely divisible processes

    Gennady Samorodnitsky and Yizao Wang. Extremal theory for long range dependent infinitely divisible processes. Ann. Probab. , 47(4):2529--2562, 2019

    2019

  25. [33]

    Choquet random sup-measures with aggregations

    Yizao Wang. Choquet random sup-measures with aggregations. Extremes , 25(1):25–54, 2022

    2022

  26. [34]

    Weak convergence of first passage time processes

    Ward Whitt. Weak convergence of first passage time processes. J. Appl. Probab. , 8(2):417--422, 1971

    1971

  27. [35]

    Stochastic-Process Limits

    Ward Whitt. Stochastic-Process Limits . Springer, New York, 2002. Springer Series in Operation Research

    2002

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.