Non-central limit theorems for random selections (Q975304)

For each \(n\), let \[ A_n= (x_{1,n}\leq x_{2,n}\leq\cdots\leq x_{n,n}) \] be a sequence of real numbers and denote by \(X^{(n)}_1,\dots, X^{(n)}_m\) the random variables obtained by drawing, with or without replacement, \(m= m(n)\) elements from \(A_n\). Define the partial sum process \[ Z_{n,m}(t)= \sum^{[mt]}_{j=1} X^{(n)}_j\qquad (0\leq t\leq 1). \] In this paper, without the uniform asymptotic negligibility condition, the authors obtain functional limit theorems for the above partial sum processes as \(m,n\to\infty\) with \(m/n> c> 0\) in order to exclude a Wiener or Brownian bridge component in the limiting process. More precisely, we assume that for each \(n\), \(A_n\) satisfies the conditions \(\sum^n_{j=1} x_{j,n}= 0\), \(x_{j,n}\to y_j\) and \(x_{n-j+1,n}\to z_j\) for any fixed \(j\), \[ \lim_{K\to\infty}\;\limsup_{n\to\infty}\;\sum^{n-K}_{j= K} x^2_{j,n}= 0,\quad \sum^\infty_{j=1} y^2_j< \infty,\quad \sum^\infty_{j=1} z^2_j< \infty. \] If \(X^{(n)}_1,\dots, X^{(n)}_m\) \((m\leq n)\) are drawn from \(A_n\) without replacement, then \[ Z_{n,m}(t)\to R(ct)\quad\text{in }D[0,1]\;(n\to\infty), \] where \[ R(t)= \sum^\infty_{j=1} y_j(\delta_j(t)- t)+ \sum^\infty_{j=1} z_j(\delta^*(t)- t)\qquad (0\leq t\leq 1), \] \(\{\delta_j(t)\}\) and \(\{\delta^*_j(t)\}\) are independent jump processes, each making a single jump from 0 to 1 at a random point uniformly distributed in \((0,1)\). If \(X^{(n)}_1,\dots, X^{(n)}_m\) are drawn with replacement, then the analogous functional limit theorem holds, but in this case \(\{\delta_j(t)\}\) and \(\{\delta^*_j(t)\}\) are independent Poisson processes with parameter 1. Several corollaries to those theorems are given and some applications to bootstrap and permutation statistics are also considered.