A covariance inequality under a two-part dependence assumption (Q1359746): Difference between revisions

Suppose \((\Omega,{\mathcal F}, P)\) is a probability space, \(\mathcal A\) and \(\mathcal B\) are \(\sigma\)-fields \(\subset\mathcal F\), \(\alpha, \lambda\in [0,1]\), \(p, q\in (1,\infty)\) such that \(1/p+1/q=1\), and one has that \[ |P(A\cap B)-P(A)P(B)|\leq\alpha +\lambda [P(A)]^{1/p}[P(B)]^{1/q} \] for every \(A\in\mathcal A\) and \(B\in\mathcal B\). Then the inequality \[ |EXY-EX EY|\leq 4\smallint_0^{\alpha}Q_{|X|} (z)Q_{|Y|}(z) dz+C\lambda (1-\log\lambda)(E|X|^p)^{1/p}(E|Y|^q)^{1/q} \] holds for any \(\mathcal A\)-measurable random variable \(X\) such that \(E|X|^p<\infty\) and for any \(\mathcal B\)-measurable random variable \(Y\) such that \(E|Y|^q<\infty\), where \(Q_W\) denotes the quantile of \(W\) and \(C\) is a constant depending on \(p, q\). This result combines and generalizes several known related inequalities. The proof is detailed and direct.