English

The rate of convergence for the renewal theorem in $\mathbb{R}^d$

Probability 2016-07-12 v2 Dynamical Systems

Abstract

Let ρ\rho be a borelian probability measure on SLd(R)\mathrm{SL}_d(\mathbb{R}). Consider the random walk (Xn)(X_n) on Rd{0}\mathbb{R}^d\setminus\{0\} defined by ρ\rho : for any xRd{0}x\in \mathbb{R}^d\setminus\{0\}, we set X0=xX_0 =x and Xn+1=gn+1XnX_{n+1} = g_{n+1} X_n where (gn)(g_n) is an iid sequence of SLd(R)\mathrm{SL}_d(\mathbb{R})-valued random variables of law ρ\rho. Guivarc'h and Raugi proved that under an assumption on the subgroup generated by the support of ρ\rho (strong irreducibility and proximality), this walk is transient. In particular, this proves that if ff is a compactly supported continuous function on Rd\mathbb{R}^d, then the function Gf(x):=Exn=0+f(Xn)Gf(x) :=\mathbb{E}_x \sum_{n=0}^{+\infty} f(X_n) is well defined for any xRd{0}x\in \mathbb{R}^d \setminus\{0\}. Guivarc'h and Le Page proved the renewal theorem in this situation : they study the possible limits of GfGf at 00 and in this article, we study the rate of convergence in their renewal theorem. To do so, we consider the family of operators (P(it))tR(P(it))_{t\in \mathbb{R}} defined for any continuous function ff on the sphere Sd1\mathbb{S}^{d-1} and any xSd1x\in \mathbb{S}^{d-1} by P(it)f(x)=SLd(R)eitlngxxf(gxgx)dρ(g) P(it) f(x) = \int_{\mathrm{SL}_d(\mathbb{R})} e^{-it \ln \frac{ \|gx\|}{\|x\|}} f\left(\frac{gx}{\|gx\|}\right) \mathrm{d}\rho(g) And we prove that, for some LRL\in \mathbb{R} and any t0R+t_0 \in \mathbb{R}_+^\ast, suptRtt01tL(IdP(it))1 is finite \sup_{\substack{t\in \mathbb{R}\\ |t| \geqslant t_0}} \frac{ 1 }{|t|^L} \left\| (I_d-P(it))^{-1} \right\| \text{ is finite} where the norm is taken in some space of h\"older-continuous functions on the sphere.

Keywords

Cite

@article{arxiv.1603.07214,
  title  = {The rate of convergence for the renewal theorem in $\mathbb{R}^d$},
  author = {Jean-Baptiste Boyer},
  journal= {arXiv preprint arXiv:1603.07214},
  year   = {2016}
}