Bonferroni-type inequalities; Chebyshev-type inequalities for the distributions on \([0, n]\) (Q1207629)

Let \(A_ 1,A_ 2,\dots,A_ n\) be events on a probability space and let \(K=K_ n\) be the number of those \(A_ j\) which occur. Put \(S_ t\) for the \(t\)-th binomial moment of \(K\), which can also be expressed as the sum \(\sum P(A_{i_ 1}\cap A_{i_ 2}\cap \dots\cap A_{i_ t})\) where the summation is extended over all subscripts \(1\leq i_ 1<i_ 2<\dots<i_ t\leq n\). If a bound on the distribution of \(K\) by means of linear combinations of the \(S_ t\) is universally valid, it is called a Bonferroni-type inequality. Such inequalities are in fact Chebyshev-type inequalities because, as first observed by \textit{T. F. Móri} and \textit{G. J. Székely} [J. Appl. Probab. 22, 836-843 (1985; Zbl 0579.60017)], any probability distribution \(p_ j\), \(0\leq j\leq n\), on the integers 0 through \(n\) can be represented by \(n\) events such that \(p_ j=P(K=j)\), and linear combinations of \(S_ t\) are linear combinations of moments of \(K\). The author fully utilizes this relation to Chebyshev-type inequalities which enables him to find the most stringent Bonferroni-type inequalities for any given set of \(S_ t\). The method for proof, resulting from this equivalence of types, is the majorant-minorant method, in which one is seeking polynomials \(u(x)\) which stay above (below) the indicator, \(d_ r(x,n)=1\) if \(x=r\) and \(=0\) otherwise, at integers \(x\) in the interval \(0\leq x\leq n\). This is essentially the method of indicators, or the method of polynomials of the reviewer and \textit{R. Mucci} [Publ. Math. 27, 263-268 (1980; Zbl 0455.60014)] but with a different emphasis. This new emphasis enables the author to generate new inequalities and to establish their optimality. Several known inequalities are reproduced and their optimality is studied.