MS9: Computational Fluid Dynamics with High-Order Spectral Element Methods on GPUs
Mathis Bode, Jörg Schumacher, Roshan J. Samuel, Christian Hasse, Hendrik Nicolai, Christos E. Frouzakis, and Ananias Tomboulides
Abstract
Turbulent fluid flows have been important use cases for high performance computing (HPC) platforms since the first spectral simulations of the Navier-Stokes equations by Orszag and Patterson in the late 1960’s. Characterized by exponential convergence that provides high accuracy at lower computational cost, spectral-type numerical schemes are well suited for the efficient simulation of turbulence, where the number of grid points grows faster than quadratic with the Reynolds number when all flow features need to be resolved.
Spectral element methods (SEMs) combine the high accuracy with flexibility in terms of flow geometry. A high-order SEM approximates the solution and data in terms of locally structured Nth-order tensor-product polynomials on a set of globally unstructured elements. Thus, in addition to exponential convergence for smooth solutions with increasing polynomial order, it offers flexibility to handle complex geometries via domain decomposition. For the same accuracy, matrix free SEM solvers also offer low storage and computational cost. In order to exploit the performance potential of existing and upcoming GPU-based exascale supercomputers, SEM solvers for CPU-based HPC systems have to either be ported to GPUs, or to be rewritten from scratch. The potential of SEM solvers for exascale computing has been underlined by the two 2023 Gordon Bell Award finalists with applications using nekRS and neko. Furthermore, the recently developed nekCRF reactive flow plugin showcases how SEMs can be efficiently used for computational fluid dynamics (CFD) including multi-physics effects, like combustion, on exascale supercomputers.
The mini symposium covers spectral element CFD solvers for GPUs. Submissions can include contributions to the development of numerical methods and/or physical models in the context of SEM as well as application examples of CFD using SEM on current GPU HPC systems. CFD can refer to fluid dynamics applications of flows with or without multi-physics effects, such as combustion, multiphase or magnetohydrodynamics.
Mathis Bode - Forschungszentrum Jülich, Jülich Supercomputing Centre
Mathis Bode is a researcher at the Jülich Supercomputing Centre, Forschungszentrum Jülich. His research focuses on fluid mechanics with multi-physics effects such as combustion or multiphase, high-performance computing and exascale applications.
Jörg Schumacher - Technische Universität Ilmenau, Institute of Thermodynamics and Fluid Mechanics
Jörg Schumacher is the head of the Fluid Mechanics group at the Institute of Thermodynamics and Fluid Mechanics at TU Ilmenau. His research focus is on direct numerical simulation studies of turbulent convection and mixing processes with and without phase changes. The analysis and reduced-order modeling of these turbulent processes apply data-driven recurrent and generative machine learning algorithms.
Roshan Samuel - Technische Universität Ilmenau, Institute of Thermodynamics and Fluid Mechanics
Roshan Samuel is a post-doctoral researcher at the Institute of Thermodynamics and Fluid Mechanics at TU Ilmenau. His research work focuses on high-performance computing with emphasis on direct numerical simulations of turbulent thermal convection. These high-resolution simulations aim to probe the physics of high Rayleigh number convection typically observed in atmospheric and geophysical flows.
Christian Hasse - Technical University Darmstadt, Simulation of reactive Thermo-Fluid Systems
Christian Hasse is the head of the institute simulation of reactive thermo-fluid systems at TU Darmstadt. Specializing in turbulent reactive flows for both single and multiphase systems, his research employs high-fidelity direct numerical simulations (DNS) with the objective of comprehending the fundamental physics of combustion and translating this understanding into sophisticated mathematical models.
Hendrik Nicolai - Technical University Darmstadt, Simulation of reactive Thermo-Fluid Systems
Hendrik Nicolai is a researcher and groupleader at the institute simulation of reactive thermo-fluid systems at TU Darmstadt. Using high-fidelity direct numerical simulations (DNS), he aims to understand combustion physics and translate it into advanced mathematical models. Coupling these with large eddy simulations (LES) allows practical applications study, bridging the gap between fundamental and applied research in areas like aircraft engines, furnaces, and chemical reactors.
Christos Frouzakis - ETH Zürich, Combustion and Acoustics for Power & Propulsion Systems Laboratory
Christos Frouzakis is a senior researcher and lecturer at the Combustion and Acoustics for Power & Propulsion Systems laboratory at ETH Zurich. His research focuses on the direct numerical simulation of low Mach number reactive flows using spectral element methods on high performance computing systems.
Ananias Tomboulides - Aristotle University of Thessaloniki, Laboratory of Applied Thermodynamics
Ananias Tomboulides is Professor of Mechanical Engineering at the Aristotle University of Thessaloniki. He specializes in high-order numerical methods for the numerical simulation of reactive flows and in high-performance computing (HPC). He is one of the main developers of the open-source code Nek5000, with contributions in the areas of low Mach number combustion and multiphase flows.