Abstract
We consider one-dimensional discrete-time random walks (RWs) of n steps, starting from \(x_0=0\), with arbitrary symmetric and continuous jump distributions \(f(\eta )\), including the important case of Lévy flights. We study the statistics of the gaps \(\varDelta _{k,n}\) between the \(k\text {th}\) and \((k+1)\text {th}\) maximum of the set of positions \(\{x_1,\ldots ,x_n\}\). We obtain an exact analytical expression for the probability distribution \(P_{k,n}(\varDelta )\) valid for any k and n, and jump distribution \(f(\eta )\), which we then analyse in the large n limit. For jump distributions whose Fourier transform behaves, for small q, as \({\hat{f}} (q) \sim 1 - |q|^\mu \) with a Lévy index \(0< \mu \le 2\), we find that the distribution becomes stationary in the limit of \(n\rightarrow \infty \), i.e. \(\lim _{n\rightarrow \infty } P_{k,n}(\varDelta )=P_k(\varDelta )\). We obtain an explicit expression for its first moment \(\mathbb {E}[\varDelta _{k}]\), valid for any k and jump distribution \(f(\eta )\) with \(\mu >1\), and show that it exhibits a universal algebraic decay \( \mathbb {E}[\varDelta _{k}]\sim k^{1/\mu -1} \varGamma \left( 1-1/\mu \right) /\pi \) for large k. Furthermore, for \(\mu >1\), we show that in the limit of \(k\rightarrow \infty \) the stationary distribution exhibits a universal scaling form \(P_k(\varDelta ) \sim k^{1-1/\mu } \mathcal {P}_\mu (k^{1-1/\mu }\varDelta )\) which depends only on the Lévy index \(\mu \), but not on the details of the jump distribution. We compute explicitly the limiting scaling function \(\mathcal {P}_\mu (x)\) in terms of Mittag–Leffler functions. For \(1< \mu <2\), we show that, while this scaling function captures the distribution of the typical gaps on the scale \(k^{1/\mu -1}\), the atypical large gaps are not described by this scaling function since they occur at a larger scale of order \(k^{1/\mu }\). This atypical part of the distribution is reminiscent of a “condensation bump” that one often encounters in several mass transport models.
Similar content being viewed by others
Data Availability
Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
References
Gumbel, E.J.: Statistics of Extremes. Dover, Downers Grove (1958)
Katz, R.W., Parlange, M.P., Naveau, P.: Statistics of extremes in hydrology. Adv. Water Resour. 25, 1287 (2002)
Embrecht, P., Klüppelberg, C., Mikosh, T.: Modelling Extremal Events for Insurance and Finance. Springer, Berlin (1997)
Majumdar, S.N., Bouchaud, J.-P.: Comment on “Thou shall buy and hold’’. Quant. Financ. 8, 753 (2008)
Bouchaud, J.-P., Mézard, M.: Universality classes for extreme-value statistics. J. Phys. A 30, 7997 (1997)
Mézard, M., Parisi, G., Virasoro, M.A.: Spin Glass Theory and Beyond: An Introduction to the Replica Method and Its Applications. World Scientific Publishing Company, Singapore (1987)
Le Doussal, P., Monthus, C.: Exact solutions for the statistics of extrema of some random 1D landscapes, application to the equilibrium and the dynamics of the toy model. Physica A 317, 140 (2003)
Leblanc, M., Angheluta, L., Dahmen, K., Goldenfeld, N.: Universal fluctuations and extreme statistics of avalanches near the depinning transition. Phys. Rev. E 87, 022126 (2013)
Raychaudhuri, S., Cranston, M., Przybla, C., Shapir, Y.: Maximal height scaling of kinetically growing surfaces. Phys. Rev. Lett. 87, 136101 (2001)
Gyorgyi, G., Holdsworth, P.C., Portelli, B., Racz, Z.: Statistics of extremal intensities for Gaussian interfaces. Phys. Rev. E 68, 056116 (2003)
Majumdar, S.N., Comtet, A.: Exact maximal height distribution of fluctuating interfaces. Phys. Rev. Lett. 92, 225501 (2004)
Majumdar, S.N., Comtet, A.: Airy distribution function: from the area under a Brownian excursion to the maximal height of fluctuating interfaces. J. Stat. Phys. 119, 777 (2005)
Schehr, G., Majumdar, S.N.: Universal asymptotic statistics of maximal relative height in one-dimensional solid-on-solid models. Phys. Rev. E 73, 056103 (2006)
Tracy, C.A., Widom, H.: Level-spacing distributions and the Airy kernel. Commun. Math. Phys. 159, 151 (1994)
Majumdar, S.N., Schehr, G.: Top eigenvalue of a random matrix: large deviations and third order phase transition. J. Stat. Mech. 1, 01012 (2014)
Majumdar, S.N., Pal, A., Schehr, G.: Extreme value statistics of correlated random variables: a pedagogical review. Phys. Rep. 840, 1 (2020)
Schehr, G., Majumdar, S.N.: Exact record and order statistics of random walks via first-passage ideas. In: Metzler, R., Oshanin, G. (eds.) First-Passage Phenomena And Their Applications, vol. 226. World Scientific, Singapore (2014)
Vivo, P.: Large deviations of the maximum of independent and identically distributed random variables. Eur. J. Phys. 36, 055037 (2015)
Sabhapandhit, S., Majumdar, S.N.: Density of near-extreme events. Phys. Rev. Lett. 98, 140201 (2007)
Sabhapandit, S., Majumdar, S.N., Redner, S.: Crowding at the front of marathon packs. J. Stat. Mech. 2008, 03001 (2008)
Arnold, B.C., Balakrishnan, N., Nagaraja, H.N.: A First Course in Order Statistics. Wiley, New York (1992)
Nagaraja, H.N., David, H.A.: Order Statistics, 3rd edn. Wiley, New Jersey (2003)
Feller, W.: An Introduction to Probability Theory and Its Applications, vol. I and II, 3rd edn. Wiley, New York (1968)
Dean, D.S., Majumdar, S.N.: Extreme-value statistics of hierarchically correlated variables deviation from Gumbel statistics and anomalous persistence. Phys. Rev. E 64, 046121 (2001)
Carpentier, D., Le Doussal, P.: Glass transition of a particle in a random potential, front selection in nonlinear renormalization group, and entropic phenomena in Liouville and sinh-Gordon models. Phys. Rev. E 63, 026110 (2001); Erratum-ibid. 73, 019910 (2006)
Fyodorov, Y.V., Bouchaud, J.-P.: Freezing and extreme-value statistics in a random energy model with logarithmically correlated potential. J. Phys. A Math. Theor. 41, 372001 (2008)
Fyodorov, Y.V., Le Doussal, P., Rosso, A.: Duality, freezing and extreme value statistics of 1/f noises. J. Stat. Mech. 10, 10005 (2009)
Majumdar, S.N., Comtet, A.: Airy distribution function: from the area under a Brownian excursion to the maximal height of fluctuating interfaces. J. Stat. Phys. 119, 777 (2005)
Györgyi, G., Moloney, N., Ozogány, G., Rácz, Z.: Maximal height statistics for signals. Phys. Rev. E 75, 021123 (2007)
Majumdar, S.N., Randon-Furling, J., Kearney, M.J., Yor, M.: On the time to reach maximum for a variety of constrained Brownian motions. J. Phys. A Math. Theor. 41, 365005 (2008)
Comtet, A., Majumdar, S.N.: Precise asymptotics for a random walker’s maximum. J. Stat. Mech. Theor. Exp. 06, 06013 (2005)
Schehr, G., Le Doussal, P.: Extreme value statistics from the real space renormalization group: Brownian motion, Bessel processes and continuous time random walks. J. Stat. Mech. 01, 01009 (2010)
Schehr, G., Majumdar, S.N.: Universal order statistics of random walks. Phys. Rev. Lett. 108, 040601 (2012)
Majumdar, S.N., Mounaix, Ph., Schehr, G.: Exact statistics of the gap and time interval between the first two maxima of random walks and Lévy flights. Phys. Rev. Lett. 111, 070601 (2013)
Majumdar, S.N., Mounaix, Ph., Schehr, G.: On the gap and time interval between the first two maxima of long random walks. J. Stat. Mech. 2014, 09013 (2014)
Lacroix-A-Chez-Toine, B., Majumdar, S.N., Schehr, G.: Gap statistics close to the quantile of a random walk. J. Phys. A Math. Theor. 52, 315003 (2019)
Mori, F., Majumdar, S.N., Schehr, G.: Time between the maximum and the minimum of a stochastic process. Phys. Rev. Lett. 123, 200201 (2019)
Mori, F., Majumdar, S.N., Schehr, G.: Distribution of the time between maximum and minimum of random walks. Phys. Rev. E 101, 052111 (2020)
Battilana, M., Majumdar, S.N., Schehr, G.: Universal gap statistics for random walks for a class of jump densities. Markov Process. Relat. Fields 26, 57 (2020)
Pitman, J., Tang, W.: Extreme order statistics of random walks. http://arxiv.org/abs/2007.13991 (2020)
Mori, F., Majumdar, S.N., Schehr, G.: Distribution of the time of the maximum for stationary processes. EPL 135, 30003 (2021)
Pitman, J., Tang, W.: Hidden symmetries and limit laws in the extreme order statistics of the Laplace random walk. http://arxiv.org/abs/2107.05095 (2021)
De Bruyne, B., Majumdar, S.N., Schehr, G.: Expected maximum of bridge random walks & Lévy flights. J. Stat. Mech. 8, 083215 (2021)
De Bruyne, B., Bénichou, O., Majumdar, S.N., Schehr, G.: Statistics of the maximum and the convex hull of a Brownian motion in confined geometries. J. Phys. A Math. Theor. 55, 144002 (2021)
Erdös, P., Kac, M.: On certain limit theorems of the theory of probability. Bull. Am. Math. Soc. 52, 292 (1946)
Darling, D.A.: The maximum of sums of stable random variables. Trans. Am. Math. Soc. 83, 164 (1956)
Pollaczek, F.: Sur la répartition des périodes d’occupation ininterrompue d’un guichet. C. R. Acad. Sci. Paris 234, 2334 (1952)
Pollaczek, F.: Order statistics of partial sums of mutually independent random variables. J. Appl. Probab. 12, 390 (1975)
Wendel, J.G.: Order statistics of partial sums. Ann. Math. Stat. 31, 1034 (1960)
Port, S.C.: An elementary probability approach to fluctuation theory. J. Math. Anal. Appl. 6, 109 (1963)
Dassios, A.: Sample quantiles of stochastic processes with stationary and independent increments. Ann. Appl. Probab. 6, 1041 (1996)
Chaumont, L.: A path transformation and its applications to fluctuation theory. J. Lond. Math. Soc. 59, 729 (1999)
Embrechts, P., Rogers, L.C.G., Yor, M.: A proof of Dassios’ representation of the alpha-quantile of Brownian motion with drift. Ann. Appl. Prob. 5, 757 (1995)
Dassios, A.: The distribution of the quantile of a Brownian motion with drift and the pricing of related path-dependent options. Ann. Appl. Prob. 5, 389 (1995)
Spitzer, F.: On interval recurrent sums of independent random variables. Proc. Am. Math. Soc. 7, 164 (1956)
Evans, M.R., Hanney, T.: Nonequilibrium statistical mechanics of the zero-range process and related models. J. Phys. A Math. Gen. 38, 195 (2005)
Majumdar, S.N., Evans, M.R., Zia, R.K.P.: Nature of the condensate in mass transport models. Phys. Rev. Lett. 94, 180601 (2005)
Evans, M.R., Majumdar, S.N., Zia, R.K.P.: Canonical analysis of condensation in factorised steady states. J. Stat. Phys. 123, 357 (2006)
Majumdar, S.N.: Real-space condensation in stochastic mass transport models. In: Exact Methods in Low-dimensional Statistical Physics and Quantum Computing: Lecture Notes of the Les Houches Summer School, vol. 89 (2010)
Gradenigo, G., Majumdar, S.N.: A first-order dynamical transition in the displacement distribution of a driven run-and-tumble particle. J. Stat. Mech. 5, 053206 (2019)
Mori, F., Gradenigo, G., Majumdar, S.N.: First-order condensation transition in the position distribution of a run-and-tumble particle in one dimension. J. Stat. Mech. 10, 103208 (2021)
Mori, F., Le Doussal, P., Majumdar, S.N., Schehr, G.: Condensation transition in the late-time position of a run-and-tumble particle. Phys. Rev. E 103, 062134 (2021)
Smith, N.R., Majumdar, S.N.: Condensation transition in large deviations of self-similar Gaussian processes with stochastic resetting. J. Stat. Mech. 5, 053212 (2022)
Haubold, H.J., Mathai, A.M., Saxena, R.K.: Mittag–Leffler functions and their applications. J. Appl. Math. 2011, 1 (2011)
Majumdar, S.N., Mounaix, Ph., Schehr, G.: Survival probability of random walks and Lévy flights on a semi-infinite line. J. Phys. A Math. Theor. 50, 465002 (2017)
Acknowledgements
This work was partially supported by the Luxembourg National Research Fund (FNR) (App. ID 14548297).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Communicated by Christian Maes.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Proof of the Schuette–Nesbitt Formula
In this Appendix we provide two different derivations of the Schuette–Nesbitt formula.
1.1 Direct Derivation
In this section, we provide a proof of the Schuette–Nesbitt formula in (33). Let \(A_1,\ldots ,A_n\) be a set of events. We introduce the indicator function I(A) which takes the value 1 if A occurred and 0 otherwise. We wish to compute the counting probability \(\text {Prob.}(\nu =\sum _{m=1}^n I(A_m))\). To do so, let us first work with indicator functions and denote by \(\mathcal {N}=\{1,\ldots ,n\}\) the set of numbers from 1 to n. For exactly \(\nu \) events to happen, we sum over all possible combinations of sets of \(\nu \) events and require that all the events in the set happened, while requiring that the remaining ones did not happen. This reads
where |J| denotes the cardinal number of the set J. By using the rule \(I(B)=1-I({\bar{B}})\), where \(\bar{B}\) is the complementary event of B, we can rewrite the last indicator function as
Making use of the inclusion-exclusion principle and that \(|\mathcal {N}\setminus J|=n-\nu \), we rewrite the last indicator function as
Distributing the product, we obtain
where we have used that \(I\left( \bigcap _{j\in J} A_{j}\right) \times I\left( \bigcap _{k\in K} A_k\right) =I\left( \bigcap _{q\in K\cup J} A_q\right) \). The last double sum enumerates all the sets in \(\mathcal {N}\) with \(\nu +p\) elements. However, due to the double sum, we over-count them by a factor \(\left( {\begin{array}{c}\nu +p\\ p\end{array}}\right) =\left( {\begin{array}{c}\nu +p\\ \nu \end{array}}\right) \), which is the number of ways one can separate a set of size \(\nu +p\) into two sets of sizes \(\nu \) and p. Therefore, it can be written as
Finally, noting that the first term on the rhs of (131) can be written as the \(p=0\) term of the sum over p, we find
which, upon taking the expectation value and taking the generating function with respect to \(\nu \), we get
which recovers the Schuette–Nesbitt formula in (33).
1.2 Alternative Derivation
An alternative proof of this last formula (133) is as follows. The generating function of \(P_{\nu , n} = \mathbb {E}[ I\left( \nu =\sum _{m=1}^n I(A_m)\right) ]\)—with respect to \(\nu =0, \ldots , n\)—can also be written as
Since \(I(A_m)\), for \(m=1, \ldots , n\), is a binary variable that takes only values 0 or 1, it is easy to check the identity
By inserting this identity (135) in (134), one obtains
We then expand the product in (136) to obtain
which is the result given in (133).
Evaluation of the Integral (49) Using the Pollaczek–Spitzer Formula
In this appendix, we compute the integral in (49), which we decompose in two parts
where
Below, in Sects. B.1 and B.2, we show that \(I_1(z,s)\) and \(I_2(s)\) are given by
Inserting these expressions into (138), we recover the expression (52) from the main text.
1.1 Analysis of \(I_1(z,s)\)
To analyse \(I_1(z,s)\) in (139), we introduce an exponential cutoff in the integral which we let go to 0 as
Using the Pollaczek–Spitzer formula in (50) and the expression of the effective distribution (51), we find
We decompose the logarithms in (144) as
to find that (144) becomes
Taking the limit \(\lambda \rightarrow 0\) and simplifying we get (141).
1.2 Analysis of \(I_2(s)\)
To analyse \(I_2(s)\) in (140), we cannot directly use Pollaczek–Spitzer formula in (50) as the initial position in the generating function of the survival probability is negative. Therefore, we first decompose the generating function of the survival probability over the first step as
where \(f(\eta )\) is the jump distribution. We can then replace \({\bar{S}}_0(s,y)\) by Pollaczek–Spitzer formula (50) in (148) as the initial position is now positive, which gives (142). The limit \(s\rightarrow 1\) of \(I_2(s)\) can be obtained by rescaling q by \(\lambda \) and, a, y and \(\lambda \) by \((1-s)^{\frac{1}{\mu }}\), which gives
where we used the expansion \(f(\eta )\sim \frac{c_\mu }{\eta ^{1+\mu }}\) for \(\eta \rightarrow \infty \), where \(c_\mu =\varGamma (1+\mu )\sin (\pi \mu /2)/\pi \), which can be obtained by inverting the small q expression of the Fourier transform \({\hat{f}}(q)\) in (11). Using that \(\int _0^\infty \frac{dq}{\pi }\frac{1}{1+q^2}=\frac{1}{2}\), it gives
One can easily check that the remaining integrals in (150) are well-behaved for \(\mu >1\).
Alternative Derivation of the Expected Gap
As in [33], we start from the celebrated Pollaczek–Wendel identity [47,48,49] which states that the double generating function of the \(k\text {th}\) maximum is given by
Note that the \(k\text {th}\) maximum \(M_{k,n}\) in the Pollaczek–Wendel identity is the \(k\text {th}\) maximum of the set \(\{x_0,\ldots ,x_n\}\), which includes the initial position of the random walk. This is different to the definition given in the main text above (2), which does not include the initial position, but should not matter for the analysis of the expected stationary gap \( \mathbb {E}[\varDelta _{k}]\). As [33] showed, this formula can be more conveniently written as
where PV refers to the principal value of the integral. From this generating function, we can extract the generating function of the mean value \( \mathbb {E}[ M_{k,n}]\) of the \(k^{\text {th}}\) maximum:
Next, as in [33], we notice that (152) is not well suited to for an expansion close to \(s=0\), and we use trick, similar to theirs but adapted to fat-tailed distributions (11), which consists in writing
As in Eq. (45) in [31], we denote
and
Using the decomposition (154) and the definitions (155) and (156), we find that \(\varphi (s,z,\rho )\) writes
As in [31], we take a derivative with respect to \(\rho \) and evaluate at \(\rho =0\), which gives
where we used that \(I_1(s,0)=I_2(s,0)=0\). The derivative of \(I_1(s,\rho )\) at \(\rho =0\) is given by
where we rescaled the integration variable by \(\rho ^\mu \). The derivative of \(I_2(s,\rho )\) at \(\rho =0\) is given by
Inserting the expressions (159) and (160) in (158), we find
Taking the limit \(s\rightarrow 0\) gives
Using Tauberian theorem, we find
To invert the remaining generating function we use the identity
in order to rewrite the term \(\frac{z^{1/\mu }}{(1-z)^{1/\mu }\sin (\frac{\pi }{\mu })}\), which gives
Using the following generating functions
and using the fact that a product of generating functions becomes a convolution, we find
The stationary gap is therefore given by
Upon using the identity (56), we recover the expression (12) given in the introduction.
Analysis of the General Formula in Eq. (98) in the Case of a Double Sided Exponential Jump Distribution
To compute explicitly the generating function of the stationary gap distribution \({\tilde{P}}_z(\varDelta ) = \sum _{k=0} P_k(\varDelta )\) given in Eqs. (97) and (98) for the double exponential jump distribution \(f(\eta ) = (1/2)\,e^{-|\eta |}\), it is useful to recall the remark done at the end of Sect. 3.1. Indeed, as done there (see Eqs. (88)–(90)), it is possible to express \({\tilde{p}}_z^{(1)}(\varDelta )\) in (97) in terms of the solution of an integral equation, similar to what was done in Eqs. (88)–(90). One finds indeed
where \({\bar{F}}_*(x_1|\varDelta )\) satisfies the integral equation
In general, it is very difficult to solve this integral equation (170). However, it is possible to solve it explicitly for the double sided exponential distribution \(f(\eta )=e^{-|\eta |}/2\), whose Fourier transform is
In this case, the effective jump distribution (51) is given by
which, in real space, reads \(\mathrm { f_s}(\eta )= \sqrt{1-s}\exp (-\sqrt{1-s}|\eta |)/2\). Inserting this expression in (92) and (93), and using that in this case \(g(\eta )=0\)—see Eq. (94)—we find
where we used the identities
together with the fact that the Laplace transform of the inverse Laplace transform of a function is the function itself, i.e., \(\int _0^\infty dx e^{-x} \int _{\gamma _B} d\lambda \frac{e^{\lambda x}}{2\pi i} h(\lambda )=h(1)\) for any integrable function \(h(\lambda )\). Inserting (173) and (174) into the integral Eq. (170) gives
One can now take two derivatives with respect to x, using \(\partial ^2_{x} \max (x_1,x) = \delta (x-x_1)\), to get the simple partial differential equation satisfied by F(x), namely
Therefore \({\bar{F}}_*(x|\varDelta )\) is of the form
where \(A(\varDelta )\) and \(B(\varDelta )\) are two unknown integration constants, i.e. independent of x. To determine them, we inject this solution (179) in the original integral Eq. (177), which yields a linear system of two independent equations relating \(A(\varDelta )\) and \(B(\varDelta )\). By solving this linear system, one finds
Substituting these expressions in the one for \({\bar{F}}_*(x|\varDelta )\), we find
Inserting it in (169) and performing the integration, we find a fairly simple expression, namely
On the other hand, the term \({\tilde{p}}_{z,s}^{(2)}(\varDelta )\) in (86) in the limit \(s \rightarrow 1\) reads
where we used (54) for the expected gap in the limit \(s\rightarrow 1\). Finally, combining the results in (182) and (184) one finds
Hence we recover the expression obtained in [33] and [39] [see equation (70) in that reference]. Note that there the generating function of \(P_{k,n}(\varDelta )\) includes the term \(k=0\) (with by convention \(P_{0,n} = 1\)), which is not the case here: this explains the \(-1\) in (185), which is not present in [33, 39]. Note that it is also possible to compute explicitly the expressions of \({\bar{S}}_*(z|x)\) and \({\tilde{E}}_*(z,y|x)\) for the case of the jump distribution \(f(\eta ) \propto |\eta |e^{-|\eta |}\) (see Appendix F).
Small s Limit of \({\bar{S}}_s\left( u|x\right) \) and \({\bar{E}}_s\left( u,y|x\right) \) for \(x<0\) and \(y<0\)
In this Appendix, we analyse the small s limit of \({\bar{S}}_s\left( u|x\right) \) and \({\bar{E}}_s\left( u,y|x\right) \), with \(u=s(z-1)/(1-s)\) for \(x<0\) and \(y<0\).
1.1 \({\bar{S}}_s\left( u|x\right) \)
From the Pollaczek–Spitzer formula [47, 48], we know that for \(x>0\), the generating function of the survival probability is given by (50). In particular, we have the Sparre Andersen result
To obtain an expression for \(x<0\), we expand the survival probability over the first jump, which reads
where \(\textrm{f}_s(\eta )=\int _{-\infty }^\infty \frac{dq}{2\pi }e^{-iq\eta }\,{{\hat{f}}_s}(q)\) is the effective jump distribution (51). The expression (187) states that for the random walk to survive from x during n steps, it must jump to \(y>0\) at the first step and survive \(n-1\) steps from y. Upon taking a generating function of (187) and using that by definition in (41) one has \(S_s(0,x)=1\) for \(x\in \mathbb {R}\), the generating function of the survival probability reads
which is valid for all \(x\in \mathbb {R}\). We now analyse (188) in the limit \(s\rightarrow 1\). To do so, we found it convenient to add and subtract the following terms
Recognizing that the first two terms correspond to the generating function of the survival probability evaluated at \(x=0\), we get
where we used the Sparre Andersen result (186). As \({\bar{S}}_s(z,x')\) in the integral is only evaluated for positive argument \(x'>0\), we can replace it by the Pollaczek–Spitzer formula given in (50). We find
We now evaluate this expression at \(u=s(z-1)/(1-s)\) in the limit \(s \rightarrow 1\). It reads explicitly
Naively, it seems that one can take directly the limit \(s \rightarrow 1\) in (192) by setting \(s=1\) in the integral. This is however too naive, as it can be shown by an explicit calculation for the double exponential case. In fact, this can be seen from the fact that the expansion of the effective distribution as \(s \rightarrow 1\) leads, to leading order
Once inserted in (192), we see that the integral of the term of order \(O(1-s)\) is diverging for \(q \rightarrow 0\) since \(1/(1-{\hat{f}}(q))^2 \propto 1/|q|^{2\mu }\). This signals the fact that one must be careful to take this limit \(s \rightarrow 1\). To analyse this limit, it is convenient to use the following trick which amounts to add and subtract \( {\hat{f}^\mu _s}(q) = 1/(1+|q|^\mu )\) to \({\hat{f}_s}(q)\)
An interesting property of \({\hat{F}_s}(q)\) is that its expansion for \(s \rightarrow 1\), i.e. the analogue of (193) reads, to leading order
Since \({\hat{f}}(q) \sim 1 - |q|^\mu \) for small q we see that the divergence \(\propto 1/q^{2\mu }\) is thus suppressed in the term of order \(O(1-s)\) in (195), while the first term goes to a constant as \(q \rightarrow 0\) – assuming that \({\hat{f}}(q) = 1 - q^\mu + O(q^{2\mu })\) as \(q \rightarrow 0\). In the limit \(s\rightarrow 1\), we find that the Fourier transform of the effective distribution (194) takes the limiting form
which in real space reads
where
and \(\mathcal { F}_\mu (x)=\int _{-\infty }^\infty \frac{dq }{2\pi }e^{-ixu}\mathcal {{\hat{F}}}(u)\), and \(g(x)=\int _{-\infty }^\infty \frac{dq }{2\pi }e^{-ixq}{\hat{g}}(q)\). Upon inserting the scaling (196) in (192), and rescaling all integration variables by \((1-s)^\frac{1}{\mu }\), we get upon letting \(s\rightarrow 1\),
Further letting \(s\rightarrow 1\) in the logarithm, and using that \(\lambda \int _0^\infty \frac{dq}{\pi } \frac{1}{\lambda ^2+q^2}=\frac{1}{2}\), we find
where \(\mathcal {F}'_\mu \) is the derivative of \(\mathcal {F}_\mu \). Keeping only the highest order terms in (200), which means the first and the last terms for \(\mu <2\) and all of them for \(\mu =2\), we find the expression (91) given in the main text. For \(\mu =2\), we used that
which can be easily shown by noting that \(\mathcal {F}_2(x) = e^{-|x|}/2\) and using the fact that the Laplace transform of the inverse Laplace transform of a function is the function itself, i.e. that \(\int _0^\infty dx e^{-x} \int _{\gamma _B} d\lambda \frac{e^{\lambda x}}{2\pi i} h(\lambda )=h(1)\) for a function h.
1.2 \({\tilde{E}}_s\left( u,y|x\right) \)
From the Pollaczek–Spitzer formula we know that for \(x>0\) and \(y>0\), the generating function \({\bar{E}}_s(z,y|x)\) is given by
Note that in our case, we are interested in \({\tilde{E}}_s(z,y|x)=\sum _{n=1}^\infty z^n E_s(n,y|x)\), where the index starts from 1 (hence the different notations \({\tilde{E}}_s(z,y|x)\) and \({\bar{E}}_s(z,y|x)\)). We therefore need to remove the term \(n=0\) in the sum in (202), which gives
where we used that \(E_s(0,y|x)=\delta (y-x)\). In particular for \(x=y=0\), we have (see Appendix E.3)
To obtain an expression for \(x<0\) and \(y<0\), similarly to the previous section, we expand the excursion over the first jump and last jump
where \(\textrm{f}_s(\eta )=\int _{-\infty }^\infty \frac{dq}{2\pi }e^{-iq\eta }\,{{\hat{f}}_s}(q)\) is the effective jump distribution (51). The expression (205) states that for the random walk to make an excursion from x to y during n steps, it must perform a first jump to \(x'>0\) at the first step, make an excursion of \(n-2\) steps to \(y'\), then jump from \(y'\) to y. Upon taking a generating function of (205) and using that by definition in (72) \(E_s(0,y|x)=\delta (y-x)\) and \(E_s(1,y|x)=f_s(y-x)\) for \(x,y\in \mathbb {R}\), we get that the generating function of the survival probability reads
Inserting this in (203), we get
We now analyse the small s limit of (207). As in the previous section, we find it convenient to add and subtract the following terms
Using (207), we recognize that the last line is the generating function evaluated at \(x=y=0\), namely
where we used the expression of \({\tilde{E}}_s(z,0|0)\) in (204). Upon replacing z by \(u=s(z-1)/(1-s)\) and using the same trick as explained before in (194) and using the expansion (196), one finds in the limit \(s\rightarrow 1\) that
Next, we note that in the limit \(s\rightarrow 1\) with \(x=O(1)\) and \(y=O(1)\), the first line becomes
Finally, keeping only the highest order terms in the limit \(s\rightarrow 1\), which corresponds to all the lines in (210) for \(\mu =2\) and all the lines except the terms of order \(O[(1-s)^{\frac{1}{\mu }-\frac{1}{2}}]\), we recover the expression (91) given in the main text. In doing so, we used that, for \(\mu =2\), we have
which can be easily shown by noting that \(\mathcal {F}_2(x) = e^{-|x|}/2\) and using the fact that the Laplace transform of the inverse Laplace transform of a function is the function itself, i.e. that \(\int _0^\infty dx e^{-x} \int _{\gamma _B} d\lambda \frac{e^{\lambda x}}{2\pi i} h(\lambda )=h(1)\) for a function h.
1.3 \({\tilde{E}}_s(z,0|0)\)
We take the limit \(\lambda \rightarrow \infty \) and \(\lambda '\rightarrow \infty \) in (202), which gives
Inverting the Laplace transforms with respect to \(\lambda \) and \(\lambda '\) and inserting it in (203), we recover (204).
The Special Case of the Distribution \(f(\eta ) = \frac{3}{2} |\eta |\,e^{-\sqrt{3}|\eta |}\)
In this Appendix, we focus on the special case of
for which the Fourier transform is given by
In this case, \({\hat{g}}(q)\), which is defined in (198), reads
so that
Inserting this expression (218) in (92) and (93), we find
where the function \(\phi (\lambda , z)\) is given in (95) with \({\hat{f}}(k) = 3(3-k^2)/(k^2+3)^2\) and where we used the fact that the Laplace transform of the inverse Laplace transform of a function is the function itself, i.e. that \(\int _0^\infty dx e^{-x} \int _{\gamma _B} d\lambda \frac{e^{\lambda x}}{2\pi i} h(\lambda )=h(1)\) for a function h, as well as
In particular, the function \(\phi (\lambda ,z)\) admits the small z expansion
from which one obtains the small z expansion of \(S_*(z|x)\) as
Quite remarkably, one observes from (224), that \(S_*(z=0,x)\) coincides exactly with the function \(\sqrt{\pi }\, U_0(-x)\) defined in [65]—see (77). Note that this also holds for the double exponential case in (173). Inserting the small z expansion (223) in (221) one can similarly obtain the small z expansion of \(E_*(z,y|x)\). These expansions, up to order O(z), can eventually be used in Eq. (97) to obtain the small \(\varDelta \) expansion of the distribution of the first gap for this jump distribution \(f(\eta ) = \frac{3}{2} |\eta |\,e^{-\sqrt{3}|\eta |}\). We have carefully checked that this small \(\varDelta \) expansion exactly coincides (up to order \(O(\varDelta ^2)\)) with the small \(\varDelta \) expansion of the exact result for the distribution of the first gap, obtained from a completely different method in Refs. [34, 35].
Scaling Function of the Condensed Phase
In this Appendix, we are interested in the stationary probability distribution of the \(k\text {th}\) gap \(P_k(\varDelta )\) in the limit of large gaps \(\varDelta \). To study this limit, we have found convenient to leave aside “Spitzer’s idea” for a moment, and to write out explicitly the random walk path contributions. We argue that all the contributions can be written as (see Fig. 9):
where \(E_0(n,y|x)\) is the propagator of the random walk starting from \(x>0\) and reaching \(y>0\) after n steps while remaining above the origin. It is given by the Pollaczek–Spitzer formula in (202) with \(s=0\). Using the following expansions
where
and \(c_\mu =\varGamma (1+\mu )\sin (\pi \mu /2)/\pi \). we find that (225) becomes \(\lim _{n\rightarrow \infty }\int _{-\infty }^\infty da p_{k,n}(a,\varDelta )= \int _{-\infty }^\infty da p_{k}(a,\varDelta )\) where
where we used that \(\sum _{j=1}^{n-1} j^{-1/2}(n-j)^{-1/2}\sim \pi \) for \(n\rightarrow \infty \). Rescaling all the integration variables by \(\varDelta \) and using the following expansions
where \(A_\mu =1/[\sqrt{\pi }\,\varGamma (1+\mu /2)]\), \(C_\mu =1/\varGamma (\mu /2)^2\) and \(v_\mu (x,y)= \int _0^{\min (x,y)}dz (y-z)^{\frac{\mu }{2}-1}(x-z)^{\frac{\mu }{2}-1}\), gives
Integrating over \(x_1\) and \(x_{4j}\) gives
where \(D_\mu =\sqrt{\pi } 2^{-\mu } \varGamma \left( \frac{\mu }{2}\right) /\varGamma \left( \frac{\mu +1}{2}\right) \). Formally taking a double derivative with respect to \(\varDelta \), we recover the scaling form (22) announced in the introduction. In the limit of large \(\varDelta \), we only take the term in \(j=1\) in (237) which gives
Using that \(\mathcal {J}_\mu (x_1,x_2)\rightarrow \mathcal {L}_\mu (x_2-x_1)\) for large \(x_1\) and \(x_2\) gives
Changing coordinates \(u=(x_3-x_2)/\sqrt{2}\) and \(v=(x_3+x_2)/\sqrt{2}\), and rescaling u by \(\varDelta k^\frac{1}{\mu }\) it gives
where \(E_\mu =1/(\mu -1)\). Inserting this result in (29), we find the large argument of \(\mathcal {M}_\mu (u)\) in (23).
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
De Bruyne, B., Majumdar, S.N. & Schehr, G. Universal Order Statistics for Random Walks & Lévy Flights. J Stat Phys 190, 20 (2023). https://doi.org/10.1007/s10955-022-03027-w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10955-022-03027-w