Simulation of a Gaussian random vector: A propagative version of the Gibbs sampler

Starting from the Gibbs sampler, an iterative algorithm is designed for simulating a gaussian random vector, that requires neither the inversion nor the factorization of a covariance matrix, without resting on a markovian assumption. A brief survey is given of the various ways to implement it. An example illustrates its feasibility, and a theoretical result is stated about its rate of convergence.