A family of semi-implicit hybrid FV/FE methods for computational fluid dynamics using an efficient MPI parallel implementation
A family of semi-implicit hybrid FV/FE methods for computational fluid dynamics using an efficient MPI parallel implementation
“A family of semi-implicit hybrid FV/FE methods for computational fluid dynamics using an efficient MPI parallel implementation” impartido Laura del Río Martín. Laboratory of Applied Mathematics, University of Trento (Italia). Departamento de Matemáticas, Universidade da Coruña.
Fecha: 29 de junio
Hora: 11:30 h.
Lugar: en liña (MS Teams)
Duración: 40 min
Abstract:
In this talk, I will overview the family of semi-implicit hybrid finite volume/finite element (FV/FE) schemes for computational fluid dynamics (CFD). In particular, I will focus on approximating the solution of the incompressible Navier-Stokes equations on staggered unstructured meshes. The key feature of the proposed method is using both finite element and finite volume methods when each is more appropriate: the discretization of the nonlinear convective terms will be done using explicit finite volumes, while the pressure terms are discretized implicitly, using classical continuous finite elements.
The computational cost of realistic simulations with very complex geometries and sophisticated physical phenomena is very high due to the use of fine meshes and complex geometries with millions of degrees of freedom, so the use of high-performance computing (HPC) is essential. Then, to improve the scheme's efficiency, the algorithm has been parallelized using the Message-Passing Interface (MPI) standard in combination with the spatial domain decomposition. The high-performance of the MPI implementation of the hybrid FV/FE approach is illustrated by using a speed-up analysis and comparing the obtained numerical results with some classical benchmarks in 2D and 3D and with other numerical reference solutions, getting excellent results in all cases and showing the suitability of the proposed algorithm for the resolution of large CFD problems on modern massively parallel distributed memory supercomputers.
Actividad co-financiada con la colaboración de la Consellería de Cultura, Educación, Formación Profesional e Universidades