pith. sign in

arxiv: 1907.00944 · v1 · pith:EXU7UJSDnew · submitted 2019-07-01 · ❄️ cond-mat.stat-mech · physics.data-an

Extremes and Records

Sanjib Sabhapandit This is my paper

Pith reviewed 2026-05-25 11:20 UTC · model grok-4.3

classification ❄️ cond-mat.stat-mech physics.data-an
keywords extreme value statisticsrecord statisticsrandom walksBrownian motionlimit lawsi.i.d. random variables
0
0 comments X

The pith

Limit laws govern the maxima and records for i.i.d. random variables and random walks.

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

These lecture notes establish the limit laws for the sample mean and the maximum of i.i.d. random variables, as well as the density of near-extreme events. They derive the statistics of records both for i.i.d. sequences and for random walks in discrete and continuous time. A sympathetic reader would care because these laws give precise predictions for how often new extremes appear and how the largest values are distributed, which applies directly to modeling rare events in time series.

Core claim

The lectures demonstrate that for a large number of i.i.d. random variables the properly normalized maximum converges in distribution according to limit laws determined by the tail of the parent distribution, while the probability that the nth observation sets a record equals 1/n, so that the expected number of records grows as the harmonic series; analogous but modified record statistics hold when the underlying sequence is generated by a random walk or Brownian motion.

What carries the argument

The central mechanism is the independence assumption, which makes the distribution of the maximum equal to the parent cumulative distribution raised to the nth power and makes each new observation equally likely to be a record so far.

If this is right

  • The expected number of records up to step n grows as the natural logarithm of n.
  • Near-extreme events occur with a density that can be expressed in terms of the parent distribution and the extreme value law.
  • Record statistics for random walks differ from the i.i.d. case because successive steps are correlated.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same counting argument implies that record rates remain distribution-independent for any continuous parent law.
  • The continuous-time results via Brownian motion directly supply predictions for record properties of diffusive physical processes.

Load-bearing premise

The random variables under consideration are independent and identically distributed.

What would settle it

Measurements from repeated trials of many i.i.d. samples in which the suitably scaled maximum fails to approach any limiting distribution would falsify the claimed limit laws.

Figures

Figures reproduced from arXiv: 1907.00944 by Sanjib Sabhapandit.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: The plots of the three extreme value PDFs: Fréchet (on the left), Gumbel (in the middle), and Weibull (on the right). [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: The points (red and blue) represent random observations in a time sequence. The red points are record events, whose values are [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: A realization of a CTRW in the time interval [PITH_FULL_IMAGE:figures/full_fig_p028_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p028_5.png] view at source ↗
read the original abstract

These are lecture notes from a course offered at the Bangalore School on Statistical Physics - X, during 17-28 June 2019, [ https://www.icts.res.in/program/bssp2019 ] at International centre of theoretical physics (ICTS), Bangalore. These pedagogical lectures are at the introductory level, intended mainly for master/Ph.D. students or researchers from outside the field. In these lectures, we discuss about the limit laws for the sample mean and the maximum of a set of independent and identically distributed (i.i.d.) random variables as well as random walks / Brownian motion. The density of near-extreme events is also discussed. Finally, we discuss the statistics of records for an i.i.d. random sequence as well as random walks in discrete and continuous time. Some exercises are provided for the students to work out. The video recording of the lectures are available at https://www.icts.res.in/program/bssp2019/talks

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 1 minor

Summary. These are lecture notes from the Bangalore School on Statistical Physics (June 2019) presenting standard limit laws for the sample mean and maximum of i.i.d. random variables, random walks and Brownian motion, the density of near-extreme events, and record statistics for i.i.d. sequences as well as random walks in discrete and continuous time, together with exercises for students.

Significance. The notes restate well-known textbook results in extreme-value theory and record statistics under the i.i.d. assumption. Their value is pedagogical: they supply an accessible entry point and exercises for master/Ph.D. students entering the field. No new theorems, derivations, or empirical claims are advanced.

minor comments (1)
  1. The abstract states that the lectures cover 'the limit laws for the sample mean and the maximum'; it would be helpful to add a brief sentence in the introduction clarifying which classical theorems (e.g., CLT, Fisher-Tippett-Gnedenko) are treated and at what level of rigor.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment and recommendation to accept. The manuscript is indeed a set of pedagogical lecture notes with no new results claimed.

Circularity Check

0 steps flagged

Lecture notes restate standard i.i.d. limit laws and record statistics with no original derivations

full rationale

The manuscript consists of pedagogical lecture notes presenting textbook-level results on the law of large numbers, extreme value theory for i.i.d. variables, and record statistics for sequences and random walks. No new theorems, fitted parameters, or predictions are advanced; all content follows directly from the standard i.i.d. assumption without self-referential fitting, self-citation chains, or ansatzes that reduce to the inputs. The derivation chain is therefore self-contained against external benchmarks and exhibits no circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The document is a pedagogical summary of standard results; no free parameters, axioms, or invented entities are introduced by the authors.

pith-pipeline@v0.9.0 · 5687 in / 895 out tokens · 19071 ms · 2026-05-25T11:20:20.718135+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 1 internal anchor

  1. [1]

    Mean number of records 18

    work page

  2. [2]

    Joint distribution 20

    work page

  3. [3]

    Statistics of number of records 21

    work page

  4. [4]

    For a sequence generated by random walks 25 C

    Number of cycles in random permutation 24 B. For a sequence generated by random walks 25 C. For a sequence generated by continuous time random walks (CTRW) 27

    work page

  5. [5]

    Extremes and Records

    Statistics of number of records 29 VIII. Summary 32 arXiv:1907.00944v1 [cond-mat.stat-mech] 1 Jul 2019 2 References 32 I. INTRODUCTION How long does it take to go from this institute (say, starting at 9 AM) to the airport? I am sure, this is a question, the front desk of the institute gets, often. An estimate this time is given by the overage over the tim...

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  6. [6]

    Variance:⟨X 2⟩c =⟨[X−⟨ X⟩]2⟩,

    work page

  7. [7]

    Skewness:⟨X 3⟩c =⟨[X−⟨ X⟩]3⟩,

    work page

  8. [8]

    too much

    Kurtosis:⟨X 4⟩c =⟨[X−⟨ X⟩]4⟩− 3⟨[X−⟨ X⟩]2⟩2. Exercise 2. A Gaussian random variable X with a mean µ and a variance σ 2 has the PDF p(X) = 1√ 2πσ 2 e− (X−µ)2 2σ2 . (8) Compute its characteristic function, and consequently, the cumulant generating function, and show that they are respec- tively given by ⟨ eikX ⟩ = eikµ− 1 2 σ 2k2 , g(k) = ikµ− 1 2 σ 2k2. (9...

    work page

  9. [9]

    CDF: F(z) = { exp [ −z−α ] for z≥ 0, 0 for z≤ 0

    Fréchet class: If p(X) has power-law tail, p(X)∼ X−(1+α) with α > 0 . CDF: F(z) = { exp [ −z−α ] for z≥ 0, 0 for z≤ 0. (59) PDF: f (z) = α exp[−z−α ] z1+α , z∈ (0, ∞). [see Fig. 2 (left)] (60)

    work page

  10. [10]

    [e.g., p(X)∼ exp(−X δ )]

    Gumbel class: If p(X) has faster than power-law, but unbounded right tail. [e.g., p(X)∼ exp(−X δ )]. CDF: F(z) = exp [ −e−z] . (61) PDF: f (z) = exp [ −z− e−z] , z∈ (−∞, ∞). [see Fig. 2 (middle)] (62)

    work page

  11. [11]

    CDF: F(z) =    exp [ −(−z)β ] for z≤ 0, 1 for z≥ 0

    Weibull class: If p(X) is bounded from above, p(X)∼ (a− X)β−1 near the upper support a. CDF: F(z) =    exp [ −(−z)β ] for z≤ 0, 1 for z≥ 0. (63) PDF: f (z) = β (−z)β−1 exp [ −(−z)β ] , z∈ (−∞,0). [see Fig. 2 (right)] (64) Note that, for any values of n, F n(z) =    F ( n−1/α z ) for Fréchet class F ( z− ln n ) for Gumbel class F ( n1/β z ) for W...

    work page

  12. [12]

    F n(z) = F(z + dn), and f (z) has support on z∈ (−∞, ∞). 12

    work page

  13. [13]

    F n(z) = F(cnz) with F(0) = 0 and f (z) has support on z∈ (0, ∞)

    work page

  14. [14]

    Let us consider the case 1

    F n(z) = F(cnz) with F(0) = 1 and f (z) has support on z∈ (−∞,0). Let us consider the case 1. Taking a logarithm gives nln F(z) = ln F(z + dn) (74) Since ln F≤ 0, we multiply both sides by−1 and then take another logarithm ln n + ln[−ln F(z)] = ln[−ln F(z + dn)]. (75) This equation is of the form g(z + d) = g(z) +ν. For a monotonic g(z) the solution is gi...

    work page 1952

  15. [15]

    bN→ ∞ as N→ ∞, for δ < 1

    work page

  16. [16]

    bN = O(1), is independent of N, for δ = 1

    work page

  17. [17]

    For the power-law tail p(x)∼ x−(1+α), recall, bN∼ N1/α→ ∞ as N→ ∞ [see Eq

    bN→ 0 as N→ ∞, for δ > 1. For the power-law tail p(x)∼ x−(1+α), recall, bN∼ N1/α→ ∞ as N→ ∞ [see Eq. (52)], whereas, for the bounded tail p(x)∼ (a− x)β−1 bN∼ N−1/β→ 0 as N→ ∞ [see Eq. (57)]. Therefore, the generic behavior of bN can be classified into three categories:

    work page

  18. [18]

    For the pure exponential tail p(x)∼ e−x, bN is independent of N

    work page

  19. [19]

    If the tail of p(x) decays slower than the pure exponential, then bN→ ∞ as N→ ∞

    work page

  20. [20]

    This is responsible for, three generically different limiting form of ρ(r, N)

    If the tail of p(x) decays faster than the pure exponential, then bN→ 0 as N→ ∞. This is responsible for, three generically different limiting form of ρ(r, N). A. Slower than pure exponential tail We make a change of variable x = aN + bNz in Eq. (119), ρ+(r, N) = ∫ ∞ −∞ p ( z− (r− aN)/bN b−1 N ) [ bN pmax(aN + bNz, N− 1) ] dz. (122) Now, in the limit N→ ∞...

    work page

  21. [21]

    (129) 19 l3 n R1 R2 R4 R3 Xn l1 l2 l4 FIG

    Mean number of records Let In be an indicator variable, where In = { 1 if n-th observation is a record, 0 otherwise . (129) 19 l3 n R1 R2 R4 R3 Xn l1 l2 l4 FIG. 3: The points (red and blue) represent random observations in a time sequence. The red points are record events, whose values are greater than that of all the previous events. Ri’s are record valu...

    work page

  22. [22]

    ,M− 1, to be the ages of the records (except for the last one)

    Joint distribution For a given number of records M, in a given sequence of N variables, we define ln with n = 1,2, . . . ,M− 1, to be the ages of the records (except for the last one). These are the time steps between two successive records, and hence, the time steps for which a record survives. Evidently, the minimum value of ln is 1 (at least one time st...

    work page

  23. [23]

    P(M|N) = ∞ ∑ l1=1 ∞ ∑ l2=1 ··· ∞ ∑ lM=1 P(M; l1, l2

    Statistics of number of records The probability distribution of the number of records is obtained by summing over the ages from the joint distribution obtained above. P(M|N) = ∞ ∑ l1=1 ∞ ∑ l2=1 ··· ∞ ∑ lM=1 P(M; l1, l2 . . . ,lM|N). (143) Note that, although the maximum values ofli’s are bounded byN from above, the upper limit of theli’s in the above summ...

    work page

  24. [24]

    Let N (k1, k2,

    Number of cycles in random permutation We know that total number of permutations of N objects = N!. Let N (k1, k2, . . . ,kN|N) be the number of permutations having k1 cycles, each with one element [represent each cycle by a monomer (•)], k2 cycles, each with two elements [represent each cycle by a dimer (•−•)], . . . kN cycles, each with N elements [repr...

    work page 1953

  25. [25]

    Statistics of number of records The probability distribution of the number of records in a given time can be found by integrating over the ages {li} from the joint distribution Eq. (206). The Laplace transform of the probability distribution is given by ∫ ∞ 0 e−st P(M|t) dt = ˜Q(s) [ ˜F(s) ]M−1 = √ 1− ˜ρ(s) s [ 1− √ 1− ˜ρ(s) ]M−1 . (213) The large t behav...

    work page

  26. [26]

    ⟨τ⟩ = ∫ ∞ 0 τρ (τ) dτ =−˜ρ′(0), is finite

    The mean waiting time is finite. ⟨τ⟩ = ∫ ∞ 0 τρ (τ) dτ =−˜ρ′(0), is finite. (214) We set⟨τ⟩ = 1, without loss of generality. Therefore, for as s→ 0, ˜ρ(s) = 1− s +··· (215) This is the case, where the tail of ρ(τ) decays faster than the power-law τ−2. 30

    work page

  27. [27]

    Therefore, as s→ 0, ˜ρ(s) = 1− sα +··· with 0 < α < 1, (216) where again, without loss of generality, we set the coefficient of sα term to be unity

    The mean waiting time is infinite, i.e., ˜ρ′(0) = ∞. Therefore, as s→ 0, ˜ρ(s) = 1− sα +··· with 0 < α < 1, (216) where again, without loss of generality, we set the coefficient of sα term to be unity. This case corresponds to a slower power-law decay ρ(τ)∼ τ−(1+α) for large τ, with 0 < α < 1. Combining both the cases together, we have the small-s behavior,...

    work page

  28. [28]

    R. A. Fisher and L. H. C. Tippett, Limiting forms of the frequency distribution of the largest or smallest member of a sample, Mathematical Proceedings of the Cambridge Philosophical Society 24, 180 (1928)

    work page 1928

  29. [29]

    Gnedenko, Sur La Distribution Limite Du Terme Maximum D’Une Serie Aleatoire, Annals of Mathematics, 44, 423 (1943)

    B. Gnedenko, Sur La Distribution Limite Du Terme Maximum D’Une Serie Aleatoire, Annals of Mathematics, 44, 423 (1943)

    work page 1943

  30. [30]

    E. J. Gumbel, Statistics of Extremes, Columbia University Press, New York (1958)

    work page 1958

  31. [31]

    V . B. Nevzorov, Records: Mathematical Theory, Translations of Mathematical Monographs, V ol. 194, American Mathematical Society, Provtdenre, Rhode Island (2001)

    work page 2001

  32. [32]

    B. C. Arnold, N. Balakrishnan, and H. N. Nagaraja, Records, Wiley, New York, (1998)

    work page 1998

  33. [33]

    Krug, Records in a changing world, Journal of Statistical Mechanics: Theory and Experiment P07001 (2007)

    J. Krug, Records in a changing world, Journal of Statistical Mechanics: Theory and Experiment P07001 (2007)

    work page 2007

  34. [34]

    Sabhapandit and S

    S. Sabhapandit and S. Majumdar, Density of Near-Extreme Events, Phys. Rev. Lett. 98, 140201 (2007)

    work page 2007

  35. [35]

    Perret, A

    A. Perret, A. Comtet, S. N. Majumdar, and G. Schehr, Near-Extreme Statistics of Brownian Motion, Phys. Rev. Lett.111, 240601 (2013)

    work page 2013

  36. [36]

    S. N. Majumdar and R. M. Ziff, Universal Record Statistics of Random Walks and Lévy Flights, Phys. Rev. Lett., 101, 050601 (2008)

    work page 2008

  37. [37]

    Sabhapandit, Record statistics of continuous time random walk, Europhysics Letters 94, 20003 (2011)

    S. Sabhapandit, Record statistics of continuous time random walk, Europhysics Letters 94, 20003 (2011)

    work page 2011

  38. [38]

    Godréche, S

    C. Godréche, S. N. Majumdar, and G. Schehr, Record statistics of a strongly correlated time series: random walks and Lévy flights, J. Phys. A: Math. Theor. 50, 333001 (2017)

    work page 2017

  39. [39]

    S. N. Majumdar, Universal first-passage properties of discrete-time random walks and Lévy flights on a line: Statistics of the global maximum and records, Physica A, 389, 4299 (2010)

    work page 2010