No label defined (Q4036731)

[For the entire collection see Zbl 0755.00022.] Let \(A_ 1\), \(A_ 2,\dots,A_ n\) and \(C_ 1\), \(C_ 2,\dots,C_ N\) be two sequences of events and put \(m_ n(A)\) and \(m_ N(C)\), respectively, for the number of those \(A_ j\) and \(C_ j\) which occur. The aim is to extend known univariate Bonferroni-type bounds to estimating \(P(m_ n(A)\geq r\), \(m_ N(C)\geq u)\) by linear combinations of the binomial moments \(s_{k,t}=E\left[{m_ n(A)\choose k}{m_ N(C)\choose t}\right]\). The investigation centers on bounds of the form \(B_{r,u;n,N}(a,b)=\sum^ a_{k=0}\sum^ b_{t=0}(- 1)^{k+t}b_{k,t}\) where \(b_{k,t}={k+r-1\choose r-1}{t+u-1\choose u- 1}s_{k+r,t+u}\). Based on the classical method of inclusion and exclusion (in the univariate case), one would expect that, depending on the parity of \(a\) and \(b\), \(B_{r,u;n,N}(a,b)\) is either an upper or lower bound on the cited distribution. It is indeed an upper bound if both \(a\) and \(b\) are even. However, it is shown, that whatever the parity of \(a\) and \(b\), \(B_{r,u;n,N}(a,b)\) cannot serve as a universal lower bound (i.e. a bound valid for all choices of the events \(A_ j\) and \(C_ j)\). The main result is then to show that, for odd \(a\) and \(b\), \(B_{r,u;n,N}-ds_{r+a+1,u+b+1}\) is a universal lower bound. An improvement in the upper bound also is obtained.