Quantifying model uncertainty for the observed non-Gaussian data by the Hellinger distance

Abstract

Mathematical models for complex systems under random fluctuations often certain uncertain parameters. However, quantifying model uncertainty for a stochastic differential equation with an $\alpha$-stable Lévy process is still lacking. Here, we propose an approach to infer all the uncertain non-Gaussian parameters and other system parameters by minimizing the Hellinger distance over the parameter space. The Hellinger distance measures the similarity between an empirical probability density of non-Gaussian observations and a solution (as a probability density) of the associated nonlocal Fokker-Planck equation. Numerical experiments verify that our method is feasible for estimating single and multiple parameters. Meanwhile, we find an optimal estimation interval of the estimated parameters. This method is beneficial for extracting governing dynamical system models under non-Gaussian fluctuations, as in the study of abrupt climate changes in the Dansgaard-Oeschger events.

References in corpus (9)

• Kernel density estimation via diffusion
• Scaling of the distribution of fluctuations of financial market indices
• Analysis of data sets of stochastic systems
• Estimation of Kramers-Moyal coefficients at low sampling rates
• Densities for SDEs driven by degenerate $\alpha$-stable processes
• Numerical algorithm for the space-time fractional Fokker-Planck system with two internal states
• Tractable diffusion and coalescent processes for weakly correlated loci
• Variational estimation of the drift for stochastic differential equations from the empirical density
• Fat-Tailed Distributions and Levy Processes