Inequalities for a class of positively dependent random variables with a common marginal (Q1822826)

For fixed \(n\geq 2\), \(Y=(Y_ 1,...,Y_ n)\) and \(Z=(Z_ 1,...,Z_ n)\) are random vectors. If \(Y_ 1,...,Y_ n\), \(Z_ 1,...,Z_ n\) have a common marginal distribution and if \(E[f(Y_ 1)...f(Y_ n)]\geq E[f(Z_ 1)...f(Z_ n)]\) for all nonnegative Borel-measurable functions f for which the expectations exist, we write \(Y\geq_{pd+}Z\). An n-dimensional vector \(k=(k_ 1,...,k_ r,0,...,0)\) of nonnegative integers with \(1\leq r\leq n\), \(k_ j\geq 1\) for \(j\leq r\) and \(k_ 1+...+k_ r=n\) will be said to satisfy condition C. Suppose \((U_ 1,...,U_ n)\) are iid random variables, \((V_ 1,...,V_ n)\) are iid random variables, and W is a random variable independent of the others. Suppose k is a fixed vector satisfying condition C, and g is a real-valued function of three variables. Define the vector \(X=(X_ 1,...,X_ n)\) by \[ X_ 1=g(U_ 1,V_ 1,W),...,X_{k_ 1}=g(U_{k_ 1},V_ 1,W),\quad X_{k_ 1+1}=g(U_{k_ 1+1},V_ 2,W),..., \] \[ X_{k_ 1+k_ 2}=g(U_{k_ 1+k_ 2},V_ 2,W),...,\quad X_{k_ 1+...+k_{r-1}+1}=g(U_{k_ 1+...+k_{r- 1}+1},V_ r,W),...,\quad X_ n=g(U_ n,V_ r,W). \] Denote the vector X by X(k). Suppose that k and \(k^*\) are two vectors satisfying the condition C. It is shown that if k majorizes \(k^*\), then \(X(k)\geq_{pd+}X(k^*)\). Various corollaries and applications are given.