Title: On a formula for moments of the multivariate normal distribution generalizing Stein's lemma and Isserlis theorem

Abstract: We prove a formula for the evaluation of averages containing a scalar function of a Gaussian random vector multiplied by a product of the random vector components, each one raised at a power. Some powers could be of zeroth-order, and, for averages containing only one vector component to the first power, the formula reduces to Stein's lemma for the multivariate normal distribution. Also, by setting the said function inside average equal to one, we easily derive Isserlis theorem and its generalizations, regarding higher order moments of a Gaussian random vector. We provide two proofs of the formula, with the first being a rigorous proof via mathematical induction. The second is a formal, constructive derivation based on treating the average not as an integral, but as the action of pseudodifferential operators defined via the moment-generating function of the Gaussian random vector.