Abstract:
We say that a random variate on a Euclidean space is marginal infinitely divisible with respect to a class of linear mappings on that space if each of these mappings results in an infinitely divisible random variate. Special cases are applied in a multivariate extension of the concept of type $G$ probability laws. Random nonnegative matrices play a central role.
Keywords:inverse Wishart distribution, matrix inverse Gaussian law, multivariate normal inverse Gaussian law, multivariate stable laws, random positive definite matrices, self-decomposability.