The negative association property for the absolute values of random variables equidistributed on a generalized Orlicz ball (Q941694): Difference between revisions

A function \(f:\mathbb{R}_+\to\mathbb{R}_+\cup\{\infty\}\) is called a \textit{Young function} if it is convex, \(f(0)=0\) and \(\exists_x:f(x)\neq 0,\exists_x:f(x)\neq \infty\). For \(n\) Young functions \(f_1,\dots,f_n\), the set \(K=\left\{(x_1,\dots,x_n):\sum_{i=1}^nf_i(| x_i| )\leq1\right\}\) is said to be a \textit{generalized Orlicz ball}. A sequence of random variables \((X_1, \dots ,X_n)\) is said to be \textit{negatively associated}, if for any coordinate-wise increasing bounded functions \(f,g\) and disjoint sets \(\{i_1, \dots , i_k\}\) and \(\{j_1, \dots , j_l\}\subset \{1, \dots , n\}\), the inequality \(\text{Cov}(f(X_{i1}, \dots ,X_{ik}),g(X_{j1}, \dots ,X_{jl}))\leq 0\) holds true. The sequence is said to be \textit{weakly negatively associated} if the inequality holds for \(l=1\), and \textit{very weakly negatively associated} if it holds for \(l=k=1\). The main result of the paper is that, for a generalized Orlicz ball \(K\), the sequence of variables \((| X_1| ,\dots,| X_n| )\) is negatively associated, where \(X_i\) are the coordinates of a random vector uniformly distributed on \(K\). The proof is first given for the simpler case of an \(l_p^n\) ball, namely the generalized Orlicz ball defined by the Young functions \(f_i(x) = | x| ^p\), where a stronger result is also achieved. This serves as a good illustration of what happens in general, but it takes advantage of the fact that the \(l_p^n\) balls are the only Orlicz ball whose Young function scales well with the radius, that is, \(f_i(tx_i)=\phi(t)f_i(x_i)\) for some function \(\phi\). Consequently a different approach is developed for generalized Orlicz balls, first proving the weak negative association, and then passing to the general case. Many interesting details and further results are provided, together with several comments which reveal useful in better understanding the organization of the paper and the technical steps of the proofs