Moment ratio inequality of bivariate Gaussian distribution and three-dimensional Gaussian product inequality (Q6111126)

For a three-dimensional centred Gaussian random vector \((X_1,X_2,X_3)\) and positive integers \(m_2\) and \(m_3\), the authors prove the Gaussian product inequality \[ E[X_1^2X_2^{2m_2}X_3^{2m_3}]\geq E[X_1^2]E[X_2^{2m_2}]E[X_3^{2m_3}]\,, \] with equality holding if and only if the \(X_i\) are independent. This follows from an upper bound on the moment ratio \[ \frac{|E[X_2^{2m_2+1}X_3^{2m_3+1}]|}{E[X_2^{2m_2}X_3^{2m_3}]} \] which is also established in the present paper. The authors include a thorough discussion of their results, including on those which continue to hold in the case where \(m_2\) and \(m_3\) are not integers, and on the role of computing as part of their proofs.