A note on bivariate Dawson-Sankoff-type bounds (Q1897068)

Let \((X_n, Y_m)\) be a random discrete vector with integer values in \(0 \leq X_n \leq n\) and \(0 \leq Y_m \leq m\). Put \(S_{kt} = \left[ {X_n \choose k} {Y_m \choose t}\right]\), \(k \geq 1\), \(t \geq 1\). The authors investigate bounds of the form \(P(X_n \geq 1\), \(Y_m \geq 1) \geq aS_{11} - bS_{12} - cS_{21} + dS_{22}\) and \(P(X_n \geq 1\), \(Y_m \geq 1) \geq aS_{11} - bS_{12} - cS_{21}\) with real \(a\), \(b\), \(c\), \(d\), which inequalities are presumed valid for arbitrary choice of the distribution of \(X_n\) and \(Y_m\). The authors generalize results of the reviewer and \textit{Y. Xu} [Proc. Am. Math. Soc. 117, No. 2, 523-528 (1993; Zbl 0769.60019)] and \textit{M.-Y. Lee} [Aequationes Math. 44, No. 2/3, 220-225 (1992; Zbl 0764.60021)]. In recent years, a large number of papers appeared on this subject matter which results are covered in a book of the reviewer and \textit{I. Simonelli} [``Bonferroni-type inequalities with applications'' (Springer, 1996, to appear)]. Reviewer's remark: Although there is a misprint in the cited work of the reviewer and Xu concerning the definition of optimality of bounds, the bound quoted by the authors is indeed optimal among all those based on \(S_{11}\), \(S_{12}\) and \(S_{21}\). When one constructs an example, then automatically utilizes the whole structure and not just the \(S_{kt}\), and thus no particular example can serve for comparison in such cases.