Strong limit theorems for increments of renewal processes (Q2567793)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Strong limit theorems for increments of renewal processes
scientific article

    Statements

    Strong limit theorems for increments of renewal processes (English)
    0 references
    0 references
    13 October 2005
    0 references
    Let \(X_i\), \(i\geq 1\), be i.i.d. nonnegative nondegenerate random variables with \(\text{ess\,inf\,}X_1= 0\), \(0< EX_1=\mu<\infty\). Put \(S_0= 0\), \(S_n= X_1+\cdots+ X_n\), \(N(t)= \max\{n\geq 0: S_n\leq t\}\), \(u_T= \sup\{N(t+ a_T)- N(t): 0\leq t\leq T-a_T\}\), \(W_T= \sup\{N(t+ s)- N(t)- s/\mu: 0\leq s\leq a_T\), \(0\leq t\leq T- a_T\}\), where \(a_T\leq T\) and \(a_T\uparrow\infty\) as \(T\to\infty\). A well-known problem is finding conditions such that \(\limsup(u_T- a_T/\mu)/b_T= 1\), a.s. for a suitable norming function \(b_T\). If \(a_T= O(\log T)\), then \(b_T\) depends essentially on the distribution of \(X_1\). If \(a_T/\log T\to\infty\) and \(EX^2_1<\infty\), then \(b_T\) depends only on \(a_T\) and the first two moments of \(X_1\). The paper derives a general norming function \(b_T\) that applies in both (and other) cases. It is given by \(c_T- a_T/\mu\) and by \(c_T\) for \(u_T/b_T\) . Here \(c_T\) is the unique nondecreasing solution of an equation \(c_T q(c_T)= a_T\) where the function \(q\) is defined in terms of \(\varphi\) with \(\varphi(h)= E\exp(h\mu- hX_1)\), \(h> 0\). Then \(T/c_T\) is nondecreasing for large \(T\) and \(\mu c_T> a_T\). The main theorems show that under moderate conditions \(\limsup W_T/b_T= \limsup(u_T- a_T/\mu)= \limsup u_T/b_T= 1\), a.s. and even lim instead of lim\,sup. The lim\,sups mentioned earlier are derived as corollaries, as also the case \(a_T =o(\log T)\). Other corollaries give the limits when \(EX^2_1= \infty\) and \(\mu- X_1\) is attracted by a stable law of index \(\alpha\) with \(1<\alpha < 2\) or \(\alpha= 2\). The strong law and the loglog law for renewal processes also follow.
    0 references
    renewal theory
    0 references
    strong limits for increments
    0 references
    loglog law
    0 references
    stable attraction
    0 references

    Identifiers