Performance optimization of solving methods in ELEFANT and development of simPyFEM: An efficient educational tool for geodynamic simulations using Python

Abstract

Numerical models are used to develop a better physical understanding of small and large scale geodynamic processes over geological time scales. Through continuous development of computational techniques and methods, the accuracy and resolution of the models increases each year. Simulating geodynamic processes such as mantle convection or subduction requires the solve of a linear system in which efficient solvers and preconditioners can be used to yield low computation times and thus optimal performance. Many state of the art codes are fully parallel but even then can require multiple days or weeks to run large simulations. Implementing an efficient solver and preconditioner into ELEFANT is the first step into optimizing its performance. A free and open source package for scientific computing (PETSc) is used to facilitate this process. The second part of this thesis consists of the development of an educational tool for geodynamics called simPyFEM. By translating and optimizing a pre-exisiting code (simpleFEM) from Fortran to Python, its usability and versatility are improved. This modernization is inviting to students who are new to geodynamic modelling and want to experience the effect, on an educational level, of changing parameter such as viscosity, density and gravity.