No label defined (Q5286484)

The author considers continuous probabilities \(F\) on \([0,\infty)\), having as support an interval. \(F_{i:n}\) denotes the distribution of the \(i\)- th order statistic \(X_{i:n}\) in an \(n\)-sample, \(X_{0:n}=0\). \(U_{i:n}= X_{i:n}- X_{i-1:n}\), \(X_ a=\sum a_ i X_{i:n}\) with distribution \(F_ a\), \(U_ A=\sum A_ i U_{i:n}\) with distribution \(\widetilde F_ A\), for nonnegative \(a_ 1,\dots,a_ n\) and \(A_ 1,\dots,A_ n\). \(F<^* G\) means that \(G^{-1} F(x)/x\) is nondecreasing on \(\text{supp }F\), and \(F<^ c G\) the same with \(G^{-1} F\) convex (\(F\) etc. denote also distribution functions, \(F^{-1}\) inverses). Theorem 1 states that if \(F<^*G\) and \(p\in(0,1)\) then \(F^{- 1}_{i:n}(p)/G^{-1}_{i:n}(p)\) is decreasing in \(i\), increasing in \(n\) and that when replacing \(i\) by \(n-i\) the result is decreasing in \(n\). Theorem 2: If \(F<^ c G\) and \(A_ i>0\) for \(i\leq k\) and \(A_ i=0\) for \(i>k\), then the ratio \(x/y\), \(y= F^{-1}_{k:n}(p)/G^{-1}_{k:n}(p)\) and \(x=\) the analogue with \(\widetilde F_ A\) etc., lies in \([c^{- 1},c]\) where \(c=\max A_ i/\min A_ i\), \(i\leq k\). Theorem 4: If \(F<^ cG\) and \((r,s)\) is the smallest interval outside which \(a_ i=0\), then the previous ratio, with \(F^{-1}_ a\) instead of \(\widetilde F_ A\), lies in \([a_ s/\sum a_ i,1]\). Analogous results for \(F<^*G\), or for \(0<\alpha\leq (G^{-1} F)'\leq\beta\) etc. are established. We continue with Theorem 7: if \(F<^* G\) then \(F^{-1}_{i:n}(p)\) lies between \(F^{-1}(p)\), \(G^{-1}_{i:n}(p)\) divided, respectively, by \(G^{-1}_{i:i}(p)\), \(G^{-1}_{i:n-i+1}(p)\). Corollaries deduced with the aid of theorems 2-6 follow. The case when \(G\) is the exponential distribution is considered. Theorem 10: \(F<^* G\) is equivalent to ``for some \(p\in (0,1)\) and a subsequence of \(n\)'s, \(F^{-1}_{i:n}(p)/G^{-1}_{i:n}(p)\) is nonincreasing in \(i\)''. The concepts in the paper are related to those involving the ``failure rate'' of \textit{R. E. Barlow} and \textit{F. Proschan}; especially their paper, Ann. Math. Stat. 37, 1574-1592 (1966; Zbl 0149.154), is quoted.