A Local Limit Theorem for random walks conditioned to stay positive

Author:
  • FCFrancesco Caravenna
    Francesco Caravenna

Abstract
We consider a random walk $S_n = X_1 + ... + X_n$, where $X_k \sim X$ are i.i.d. random variables with $E[X]=0$ and $E[X^2]=1$. Let $Z_n^+$ be the random variable $S_n / \sqrt{n}$ conditioned on the event $C_n := \{S_1 > 0, ..., S_n > 0 \}$; from a general result it follows that the sequence $Z_n^+$ converges weakly to the $t=1$ marginal $Z^+$ of the Brownian meander process. In this paper we obtain a local refinement of this result: assuming that the step $X$ has an absolutely continuous law and that for some $n_0$ the density of $S_{n_0}$ is bounded (the same hypothesis of the classical Local Limit Theorem), we prove that the density of $Z_n^+$ converges uniformly to the density of $Z^+$. Unlike the standard proof of the classical LLT, we make no use of characteristic functions, since conditionally on $C_n$ the steps $\{X_k \}$ are no more independent; our techniques are rather taken from the so-called fluctuation theory for random walks.