Large deviations for a random walk in dynamical random environment (Q1265723): Difference between revisions
From MaRDI portal
Created a new Item |
Set profile property. |
||
(3 intermediate revisions by 2 users not shown) | |||
Property / reviewed by | |||
Property / reviewed by: Wolfgang König / rank | |||
Property / reviewed by | |||
Property / reviewed by: Wolfgang König / rank | |||
Normal rank | |||
Property / MaRDI profile type | |||
Property / MaRDI profile type: MaRDI publication profile / rank | |||
Normal rank | |||
links / mardi / name | links / mardi / name | ||
Latest revision as of 02:44, 5 March 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Large deviations for a random walk in dynamical random environment |
scientific article |
Statements
Large deviations for a random walk in dynamical random environment (English)
0 references
26 April 1999
0 references
Consider a \(d\)-dimensional random walk \((X_n)_{n\in\mathbb N_0}\) and a dynamical random environment \(\xi_n=(\xi_n(x))_{x\in\mathbb Z^d}\) which have a mutual interaction with each other described as follows. The pair sequence \((X_n,\xi_n)_{n\in\mathbb N_0}\) is Markov and homogeneous taking values in \(\mathbb Z^d\times S^{\mathbb Z^d}\) where \(S\) is some finite state space. Fix \(n\in\mathbb N_0\) and condition on \((X_n,\xi_n)\). Then the variables \(X_{n+1}\) and \(\xi_{n+1}(x), x\in\mathbb Z^d,\) are conditionally independent. The conditional distribution of \(\xi_{n+1}(y)\) is independent of \(\xi_n(x)\) for \(x\not= y\), and is equal to \(q(\xi_n(y),\cdot)\) if \(X_n\not= y\) (here \(q\) is some irreducible and aperiodic stochastic matrix on \(S\)), and otherwise it is some fixed probability distribution on \(S\). The conditional distribution of \(X_{n+1}\) is independent of \(\xi_n(x)\) for \(x\not= y\), and it is given by some shift-invariant irreducible transition kernel \(p(X_n,\cdot)\) (having all exponential moments finite), perturbed by some uniformly bounded function of \(X_n\) and \(\xi_n(X_n)\). In other words, the processes \((\xi_n(x))_{n\in\mathbb N_0}\) form an i.i.d. collection of irreducible and aperiodic Markov chains on \(S\) which loose their memory when visited by the random walk, and the transition kernel of the random walk is influenced slightly by the value of the environment at the walker's current position. Let \(\mu_{\delta,n}\) denote the distribution of \(X_n/n\), where \(\delta \in\mathbb R\) is the perturbation parameter. (This corresponds to the so-called annealed setting.) The main result of the paper is a large deviation principle for \((\mu_{\delta,n})_{n\in\mathbb N}\) if \(| \delta| \) is small. The rate function \(L_\delta\) is shown to be strictly convex and to have compact level sets. The map \((v,\delta)\mapsto L_\delta(v)\) can be analytically continued to the product of \(\mathbb R^d\) and a small complex neighborhood of zero. The main step of the proof is the analysis of the exponential rate of the moment generating function \(R_n(\delta,\alpha)= E_\delta\exp\langle\alpha,X_n -X_0\rangle\) in terms of the poles of \(z\mapsto \sum_{n\in\mathbb N_0} R_n (\delta,\alpha) z^n\). (Then the Gärtner-Ellis theorem yields the main result more or less immediately.) Using a cluster expansion, the latter function is written as a quotient of analytic functions such that Dyson's equations can be applied to analyse the zeros of the denominator.
0 references
large deviations
0 references
random walk in random environment
0 references
cluster expansion
0 references
Dyson's equations
0 references