Persistence in a stationary time series

A random walk with catastrophes

Random population dynamics with catastrophes (events pertaining to possible elimination of a large portion of the population) has a long history in the mathematical literature. In this paper we study an ergodic model for random population dynamics with linear growth and binomial catastrophes: in a catastrophe, each individual survives with some fixed probability, independently of the rest. Through a coupling construction, we obtain sharp two-sided bounds for the rate of convergence to stationarity which are applied to show that the model exhibits a cutoff phenomenon.

Persistence of non-Markovian Gaussian stationary processes in discrete time

The persistence of a stochastic variable is the probability that it does not cross a given level during a fixed time interval. Although persistence is a simple concept to understand, it is in general hard to calculate. Here we consider zero mean Gaussian stationary processes in discrete time n. Few results are known for the persistence P_{0}(n) in discrete time, except the large time behavior which is characterized by the nontrivial constant θ through P_{0}(n)∼θ^{n}. Using a modified version of the independent interval approximation (IIA) that we developed before, we are able to calculate P_{0}(n) analytically in z-transform space in terms of the autocorrelation function A(n). If A(n)→0 as n→∞, we extract θ numerically, while if A(n)=0, for finite n>N, we find θ exactly (within the IIA). We apply our results to three special cases: the nearest-neighbor-correlated "first order moving average process", where A(n)=0 for n>1, the double exponential-correlated "second order autoregressive process", where A(n)=c_{1}λ_{1}^{n}+c_{2}λ_{2}^{n}, and power-law-correlated variables, where A(n)∼n^{-μ}. Apart from the power-law case when μ<5, we find excellent agreement with simulations.

Statistics of the number of minima in a random energy landscape

We consider random energy landscapes constructed from d -dimensional lattices or trees. The distribution of the number of local minima in such landscapes follows a large-deviation principle and we derive the associated law exactly for dimension 1. Also of interest is the probability of the maximum possible number of minima; this probability scales exponentially with the number of sites. We calculate analytically the corresponding exponent for the Cayley tree and the two-leg ladder; for two- to five-dimensional hypercubic lattices, we compute the exponent numerically and compare to the Cayley tree case.

Persistence exponents and the statistics of crossings and occupation times for Gaussian stationary processes

We consider the persistence probability, the occupation-time distribution, and the distribution of the number of zero crossings for discrete or (equivalently) discretely sampled Gaussian stationary processes (GSPs) of zero mean. We first consider the Ornstein-Uhlenbeck process, finding expressions for the mean and variance of the number of crossings and the "partial survival" probability. We then elaborate on the correlator expansion developed in an earlier paper [G. C. M. A. Ehrhardt and A. J. Bray, Phys. Rev. Lett. 88, 070602 (2002)] to calculate discretely sampled persistence exponents of GSPs of known correlator by means of a series expansion in the correlator. We apply this method to the processes d(n)x/dt(n)=eta(t) with n>/=3, incorporating an extrapolation of the series to the limit of continuous sampling. We then extend the correlator method to calculate the occupation-time and crossing-number distributions, as well as their partial-survival distributions and the means and variances of the occupation time and number of crossings. We apply these general methods to the d(n)x/dt(n)=eta(t) processes for n=1 (random walk), n=2 (random acceleration), and larger n, and to simple diffusion from random initial conditions in one to three dimensions. The results for discrete sampling are extrapolated to the continuum limit where possible.

Exact occupation time distribution in a non-Markovian sequence and its relation to spin glass models

We compute exactly the distribution of the occupation time in a discrete non-Markovian toy sequence that appears in various physical contexts such as the diffusion processes and Ising spin glass chains. The non-Markovian property makes the results nontrivial even for this toy sequence. The distribution is shown to have non-Gaussian tails characterized by a nontrivial large deviation function that is computed explicitly. An exact mapping of this sequence to an Ising spin glass chain via a gauge transformation raises an interesting question for a generic finite sized spin glass model; at a given temperature, what is the distribution (over disorder) of the thermally averaged number of spins that are aligned to their local fields? We show that this distribution remains nontrivial even at infinite temperature and can be computed explicitly in few cases such as in the Sherrington-Kirkpatrick model with Gaussian disorder.

Persistence of a continuous stochastic process with discrete-time sampling: non-Markov processes

We consider the problem of "discrete-time persistence," which deals with the zero crossings of a continuous stochastic process X(T) measured at discrete times T=nDeltaT. For a Gaussian stationary process the persistence (no crossing) probability decays as exp(-theta(D)T)=[rho(a)](n) for large n, where a=exp(-DeltaT/2) and the discrete persistence exponent theta(D) is given by theta(D)=(ln rho)/(2 ln a). Using the "independent interval approximation," we show how theta(D) varies with DeltaT for small DeltaT and conclude that experimental measurements of persistence for smooth processes, such as diffusion, are less sensitive to the effects of discrete sampling than measurements of a randomly accelerated particle or random walker. We extend the matrix method developed by us previously [Phys. Rev. E 64, 015101(R) (2001)] to determine rho(a) for a two-dimensional random walk and the one-dimensional random-acceleration problem. We also consider "alternating persistence," which corresponds to a<0, and calculate rho(a) for this case.

Statistics of multiple sign changes in a discrete non-Markovian sequence

We study analytically the statistics of multiple sign changes in a discrete non-Markovian sequence psi(i)=phi(i)+ phi(i-1) (i=1,2, em leader,n) where phi(i)'s are independent and identically distributed random variables each drawn from a symmetric and continuous distribution rho(phi). We show that the probability P(m)(n) of m sign changes up to n steps is universal, i.e., independent of the distribution rho(phi). The mean and variance of the number of sign changes are computed exactly for all n>0. We show that the generating function (tilde)P(p,n)= summation operator(infinity)(m=0)P(m)(n)p(m) approximately exp[-theta(d)(p)n] for large n where the "discrete" partial survival exponent theta(d)(p) is given by a nontrivial formula, theta(d)(p)=ln[sin(-1)(square root of [1-p(2)])/square root of [1-p(2)]] for 0< or = p < or = 1. We also show that in the natural scaling limit m-->infinity, n-->infinity but keeping x=m/n fixed, P(m)(n) approximately exp[-n Phi (x)] where the large deviation function Phi(x) is computed. The implications of these results for Ising spin glasses are discussed.