A Multiple Prior Monte Carlo Method for the Backward Heat Diffusion Problem

TL;DR

This paper introduces a multiple prior Monte Carlo approach for solving the nonlinear inverse problem of reconstructing heat conductivity in a 2D steady-state model, leveraging different priors to improve solution accuracy.

Contribution

It develops a novel multiple prior framework using Metropolis Hastings MCMC for heat conductivity reconstruction, enhancing prior incorporation in inverse problems.

Findings

Multiple priors improve reconstruction quality.

Different priors target specific parts of the conductivity.

Preliminary results show promise for combined priors.

Abstract

We consider the nonlinear inverse problem of reconstructing the heat conductivity of a cooling fin, modeled by a 2-dimensional steady-state equation with Robin boundary conditions. The Metropolis Hastings Markov Chain Monte Carlo algorithm is studied and implemented, as well as the notion of priors. By analyzing the results using certain trial conductivities, we formulate several distinct priors to aid in obtaining the solution. These priors are associated with different identifiable parts of the reconstruction, such as areas with vanishing, constant, or varying slopes. Although more research is required for some non-constant conductivities, we believe that using several priors simultaneously could help in solving the problem.