Some new Gaussian product inequalities (Q2157700): Difference between revisions

Using a combinatorial argument, the authors establish some special cases of the Gaussian product inequality. In particular, they show that, for any Gaussian random vector \((X_1,X_2,X_3)\) and \(m_3\in\mathbb{N}\), \[ E[X_1^2X_2^4X_3^{2m_3}]\geq E[X_1^2]E[X_2^4]E[X_3^{2m_3}], \] and \[ E[X_1^2X_2^6X_3^{2m_3}]\geq E[X_1^2]E[X_2^6]E[X_3^{2m_3}], \] with equality if and only if the \(X_i\) are independent. This is shown to be equivalent to a strengthened Cauchy-Schwarz inequality for bivariate Gaussian random variables, which in turn is used to establish moment inequalities for bivariate Gaussian random variables with standard Gaussian marginals. The authors show that for such a random variable \((U,V)\) and \(m_2,m_3\in\mathbb{N}\), \[ E[U^{2m_2}V^{2m_3}]\leq\frac{E[U^{2m_2+2}V^{2m_3}]}{2m_2+1}, \] with equality if and only if \(U\) and \(V\) are independent, and that if \(m_3\leq m_2\) we have \[ \left|E[U^{2m_2+1}V^{2m_3-1}]\right|<\frac{2m_2+1}{2m_3}E[U^{2m_2}V^{2m_3}]. \]