Levy diffusion: the density versus the trajectory approach

We discuss the problem of deriving Levy diffusion, in the form of a Levy walk, from a density approach, namely using a Liouville equation. We address this problem for a case that has already been discussed using the method of the continuous time random walk, and consequently an approach based on trajectory dynamics rather than density time evolution. We show that the use of the Liouville equation requires the knowledge of the correlation functions of the fluctuation that generates the Levy diffusion. We benefit from the results of earlier work proving that the correlation function is not factorized as in the Poisson case. We show that the Liouville equation generates a long-time diffusion whose probability distribution density keeps a memory of the detailed form of the fluctuation correlation function, and not only of its long-time inverse power law structure. Although the main result of this paper rests on an approximate expression for higher-order correlation functions, it becomes exact in the longtime limit. Thus, we argue that it explains the failure to derive Levy diffusion from the Liouville equation, thereby supporting the claim that there exists a conflict between trajectory and density approaches in this case.