Upper bounds on attractor dimensions for lattices of arbitrary size
Résumé
By using the link between extreme value theory and Poincaré recurrences, we compute the attractor dimension D - namely the effective number of degrees of freedom - of systems whose dynamics consist of n non-interacting particles with Gaussian underlying distributions. We also derive a theoretical expression for D in terms of finite size limits and test its validity and applicability with numerical experiments. We find that D in a non-interacting lattice is noticeably lower than the lattice size itself. We then estimate the attractor dimension of a collection of time-series issued from conceptual dynamical systems, finance and climate datasets. We find that spatial correlation within the particles reduce the attractor dimension. We also derive numerically the upper boundary for D of non-Gaussian systems, as their dimension can exceed the Gaussian theoretical limit.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...