Independent Poisson processes generated by record values and inter-record times (Q1066550): Difference between revisions

Let X(1), X(2),... be i.i.d. random variables with continuous distribution F and for \(k\leq n\) let \(X_ K(n)\) be the kth decreasing order statistic among X(1),...,X(n). For an integer m fixed throughout, define \(N(1)=1\) and \(N(n+1)=N_ m(n+1)=\min \{k>N(n):\) \(X(k)>X_ m(N(n))\}\); the N(n) are m th-order record epochs (at which \((X_ 1(n),...,X_ m(n))\) changes). Further, define \(\nu (n)=N(n+1)-N(n)\) and \(Y(n)=X_ m(N_ m(n))\); the latter are m th-order record values. Let \(S_ r\) be the point process on \({\mathbb{R}}\) with points \(\{X_ r(N_ r(n)):n\geq 1\}\), which constitute a subset of those of \(S_{r+1}\). Finally, define \(\sigma (n)=k\) if \(Y(n)\in S_ k\setminus S_{k-1}\), i.e., if Y(n) is a k th-order record value but not a j th-order record value for any \(j<k.\) The main result is that the point process with points \(\{\) (Y(n),\(\nu\) (n),\(\sigma\) (n)):n\(\geq 1\}\) is Poisson, with mean measure \(\Lambda ((- \infty,x)\times \{j\}\times \{k\})=(1/j)F^ j(x)\). It follows that the point processes \(S_ k\setminus S_{k-1}\) are independent, a property established originally by \textit{Z. Ignatov} [Point processes and queueing problems, Keszthely 1978, Colloq. Math. Soc. János Bolyai 24, 109-116 (1981; Zbl 0458.60049)]. Related results for record epochs after a prescribed time r are given as well; in this case the main point process is a Cox process whose directing measure depends in an explicit fashion on \(X_ m(r)\).