Markov Chain Monte Carlo Inversion for the Rheology of Olivine Aggregates

Abstract:

Many experimental studies have been conducted to derive the constitutive relations that govern the rheology of olivine aggregates. Experimental observations are accompanied by uncertainty that must be properly accounted for while estimating such relations. Korenaga and Karato [2008] was the first study to conduct a rigorous statistical analysis by developing a Markov Chain Monte Carlo (MCMC) inversion code, and they demonstrated the significance of taking into account all kinds of experimental errors, as well as including more than one mechanism in the deformation model and estimating all flow law parameters simultaneously. Their MCMC code was, however, later found to have a potential of introducing some parameter bias [Mullet et al., 2015].

In this study, we utilize an improved version of their MCMC code to revisit the same data set. The newly estimated flow law parameters are distinctly different from those obtained by Korenaga and Karato [2008], as well as older studies from which the experimental data was compiled. Our analysis demonstrates the superiority of the new inversion scheme over simple least square inversion, and underscores the need for a thorough statistical analysis of experimental data. We also discuss how covariance among model parameters could affect the predictions of viscosity based on the newly estimated flow laws.