Asymptotic analysis of distribution of Dixon statistics (Q1822155)

Let \(X_ 1,...,X_ m\) and \(Y_ 1,...,Y_ n\) be two samples such that \(X_ 1,...,X_ m\), \(Y_ 1,...,Y_ n\) are observations of \(m+n\) r.v's which are independent and identically distributed according to a continuous distribution. Let \(X_{(1)}<X_{(2)}<...<X_{(m)}\) be the order statistics from the first sample and \(S_ k\) the number of Y's falling into the interval \([X_{(k-1)},X_ k)\) (assuming \(X_{(0)}=- \infty\), \(X_{(m+1)}=+\infty)\) and put \[ T_{m,n}=\sum^{m+1}_{k=1}a(k,S_ k) \] where a(i,j) is an integer- valued function. The asymptotic behaviour of the probability \(P(T_{m,n}=N)\) is analyzed for the special case \(a(i,j)=j^ 2\), (i.e. for \(T_{m,n}\) being Dixon's statistic). Asymptotic representations of \(P(T_{m,n}=N)\) (including a rest member) are derived under assumptions concerning mutual relations among m, n and N (all of them going to infinity). Correction, in the form of a counterexample, of an earlier- known result together with the discussion of precision of results concludes the paper.