Modified information criteria for a uniform approximate equivalence of probability distributions (Q1821437)

In a previous paper [ibid. 34, 1-17 (1982; Zbl 0486.62007)] the author has suggested a uniform \(\phi\)-equivalence of probability distributions based on Kullback-Leibler information and related measures of discrepancy. Nevertheless, results in that paper concerning lower and upper bounds of the supremum of the absolute difference between two probability distributions were not necessarily satisfactory, and the present paper deals with the improvement of such results. Thus, this paper presents improved lower and upper bounds of almost the same order of magnitude, leading us to a two-sided uniform \(\phi\)-equivalence improving that in the previous paper. As an example of the last equivalence, the approximation errors between samplings with and without replacement from a finite population are then analyzed. Some useful modified information criteria for a broad class of asymptotic problems are later discussed. Finally, these last criteria are applied to examine the asymptotic distributions of quasi-extreme order statistics.