Thermodynamic formalism and Uncertainty Quantification
Luc Rey-Bellet
Abstract.
Imagine a stochastic model subject to uncertainty (e.g. model
miss-specifcation or uncertainty in the parameters in the models). We are
particularly interested in large systems in a long-time, steady state,
regime. We provide new uncertainty bounds on statistical estimators
(expected value of observable, variance, probability of rare event) using
relative entropy, relative renyi entropy and Gibbs-type variational principles.