Exact convergence rates in Erdős-Rényi type theorems for renewal processes (Q1091666)

Let \((X_ n)\) be a sequence of i.i.d. non-negative random variables and let \(S_ 0=0\), \(S_ n=\sum^{n}_{j=1}X_ j\), \(n\geq 1\). Define \(N(t)=\max (n\geq 0\), \(S_ n\leq t)\), \(t>0\), \(K_ T=C \log T\), \[ \Delta_ T=\sup_{0\leq t\leq T}(N(t+K_ T)-N(t))\text{ and }\delta_ T=\inf_{0\leq t\leq T}(N(t+K_ T)-N(t)),\quad T>1. \] Suppose that \(E \exp(sX_ 1)<\infty\) for some \(s>0\). Then the author proves that \[ \limsup_{T\to \infty}\text{ (resp. } \liminf_{T\to \infty})(\Delta_ T-CA \log T)/\log \log T=h(C)\text{ (resp. } -h(C))\text{ a.s.,} \] where \(A\), \(C\) and \(h(C)\) are constants which depend on the distribution of \(X_ 1\). Similar results have been obtained for \(\delta_ T\) also. The technique used is to relate the a.s. occurences of events involving \(\Delta_ T\) and \(\delta_ T\) to those of events involving the increments of partial sums and then to appeal to a result on the increments of partial sums due to the author [Tech. Rep. 86-45, L.S.T.A. Univ. Paris-VI, Publ. Inst. Univ. Paris (1986)]. The last author [Ann. Probab. 14, 547-559 (1986; Zbl 0601.60029)] proved that, as \(T\to \infty\), \(\Delta_ T-CA \log T=O(\log \log T)\) a.s.