A Bayesian approach to parameter identification with an application to Turing systems

TL;DR

This paper introduces a Bayesian framework for parameter identification in PDE models, enabling uncertainty quantification and sensitivity analysis, demonstrated on reaction-diffusion systems with evolving domains.

Contribution

The paper develops a Bayesian methodology for PDE parameter identification that includes uncertainty quantification and is applicable to complex, real-world problems.

Findings

Provides a rigorous Bayesian approach for PDE parameter estimation

Demonstrates the method on reaction-diffusion systems with evolving domains

Highlights the computational feasibility with parallel algorithms

Abstract

We present a Bayesian methodology for infinite as well as finite dimensional parameter identification for partial differential equation models. The Bayesian framework provides a rigorous mathematical framework for incorporating prior knowledge on uncertainty in the observations and the parameters themselves, resulting in an approximation of the full probability distribution for the parameters, given the data. Although the numerical approximation of the full probability distribution is computationally expensive, parallelised algorithms can make many practically relevant problems computationally feasible. The probability distribution not only provides estimates for the values of the parameters, but also provides information about the inferability of parameters and the sensitivity of the model. This information is crucial when a mathematical model is used to study the outcome of real-world experiments. Keeping in mind the applicability of our approach to tackle real-world practical problems with data from experiments, in this initial proof of concept work, we apply this theoretical and computational framework to parameter identification for a well studied semilinear reaction-diffusion system with activator-depleted reaction kinetics, posed on evolving and stationary domains.