Persistent random walks. II. Functional Scaling Limits

TL;DR

This paper characterizes the functional scaling limits of persistent random walks, revealing a phase transition based on memory, and introduces generalized drifts and new limit processes, including arcsine Lamperti anomalous diffusions.

Contribution

It provides a unified description of scaling limits for persistent random walks, including the critical Cauchy case, and introduces generalized drifts and new non-Markovian limit processes.

Findings

Limit processes are either Markovian or not, depending on tail decay rate.

Classical stable Lévy processes arise in the memoryless case.

Critical Cauchy case fills a gap in the theory of directionally reinforced random walks.

Abstract

We give a complete and unified description -- under some stability assumptions -- of the functional scaling limits associated with some persistent random walks for which the recurrent or transient type is studied in [1]. As a result, we highlight a phase transition phenomenon with respect to the memory. It turns out that the limit process is either Markovian or not according to -- to put it in a nutshell -- the rate of decrease of the distribution tails corresponding to the persistent times. In the memoryless situation, the limits are classical strictly stable L{\'e}vy processes of infinite variations. However, we point out that the description of the critical Cauchy case fills some lacuna even in the closely related context of Directionally Reinforced Random Walks (DRRWs) for which it has not been considered yet. Besides, we need to introduced some relevant generalized drift -- extended the classical one -- in order to study the critical case but also the situation when the limit is no longer Markovian. It appears to be in full generality a drift in mean for the Persistent Random Walk (PRW). The limit processes keeping some memory -- given by some variable length Markov chain -- of the underlying PRW are called arcsine Lamperti anomalous diffusions due to their marginal distribution which are computed explicitly here. To this end, we make the connection with the governing equations for L{\'e}vy walks, the occupation times of skew Bessel processes and a more general class modelled on Lamperti processes. We also stress that we clarify some misunderstanding regarding this marginal distribution in the framework of DRRWs. Finally, we stress that the latter situation is more flexible -- as in the first paper -- in the sense that the results can be easily generalized to a wider class of PRWs without renewal pattern.