High-Performance Computing of 3D Magma Dynamics, and Comparison With 2D Simulation Results

The dynamics of magma is often studied through 2D numerical simulations because 3D
simulations are usually complex and computationally expensive. However, magmatic
systems and physical processes are 3D and approximating them in 2D requires an
evaluation of the information which is lost under different conditions. This work presents a
physical and numerical model for 3D magma convection dynamics. The model is applied to
study the dynamics of magma convection and mixing between andesitic and dacitic
magmas. The 3D simulation results are compared with corresponding 2D simulations. We
also provide details on the numerical scheme and its parallel implementation in C++ for
high-performance computing. The performance of the numerical code is evaluated
through strong scaling exercises involving up to > 12,000 cores.