Superposition of renewal processes with heavy-tailed interarrival times (Q2489864): Difference between revisions

Let \(X(i,j)\), \(i=1,\dots,m\), \(j=1,2,\dots\), be i.i.d. positive random variables with \(P(X(i,j)>t)\) varying regularly of index \(-\alpha\), \(0< \alpha<1\), as \(t\to\infty\). Let \(S(i,n)=X(i,1)+\cdots+X(i,n)\), \(i=1,\dots, m\), \(n>0\), and \(Z(0)=0\), \(Z(n)\), \(n>0\), the combination of the renewal processes \(S(i,.)\), \(i=1,\dots,m\), i.e., \(Z(n)\), \(n>0\), is the \(n\)th nondecreasing order statistic of the sample \(S(i,r)\), \(i=1,\dots,m\), \(r>0\). The process \(Z(n)\) generally is not a renewal process. Let \(M(t)= \#\{n>0:Z(n)\leq t\}\). The spent life \(\Sigma (t)\) at \(t\) associated with the process \(Z(n)\) is defined as \(t-Z(M(t))\). Now let \(m\to\infty\) and \(t\to\infty\) so that \(mt^{\alpha-1}\to C\). It is shown that for \(0<C< \infty\) the distribution of \(\Sigma(t)\) converges to a Weibull distribution. When \(C=0\) and \(C=\infty\), respectively, \(\Sigma(t)\to\infty\) and \(\Sigma(t)\to 0\) in probability. In the latter cases, when \(mt^{\alpha-1}\sim A(t)\) with \(A(t)\to 0\) or \(A(t)\to\infty\), respectively, a limit theorem for \(A(t)^{1/(1-\alpha)}\Sigma(t)\) with Weibull limit is proved. Similar theorems are proved for \(Z(M(t)+1)-t\). Should the \(X(i,j)\) be nonlattice?