Name: Philipp Edelmann
Pronouns: he/him/his
Biography:
My background is in computational stellar astrophysics. I have developed and worked on several codes for large scale simulations of the interiors of stars. I am one of the main developers of the SLH code, which was one of the first to achieve fully implicit three-dimensional simulations of a star at scale, which including other relevant physics such as radiation and nuclear reactions. I have been at LANL since 2020 and a staff member since 2022. I am currently working on developing multiphysics codes using modern techniques from computer science, such as task-based parallelism.
Institution/Lab: Los Alamos National Laboratory
Website: https://slh-code.org/
SRP Collaboration Topic/Title: A modern approach to implicit fluid dynamics
Field or research area: Computational Fluid Dynamics
Please select all the topical areas that apply to your project:
Computational Science Applications (i.e., bioscience, cosmology, chemistry, environmental science, nanotechnology, climate, etc.); Computer Science (i.e., architectures, compilers/languages, networks, workflow/edge, experiment automation, containers, neuromorphic computing, programming models, operating systems, sustainable software); High-Performance Computing
Brief Abstract:
Computational fluid dynamics (CFD) is used every day in many fields of science from biology to astrophysics. Yet simulating slow flows (compared to the speed of sound) is a hard problem both in terms of numerical accuracy and in terms of efficiency, in particular when explicit time stepping is used. Recent work from stellar astrophysics has shown that it is feasible to run three-dimensional (3D) CFD simulations with fully implicit time stepping. This is significantly more complex and involves the use of linear and nonlinear iterative solvers. The trends of the past years and especially the new supercomputers of the exascale era have shown that support for GPUs is essential. At the same time traditional methods of parallelization using the Message Passing Interface (MPI) have trouble coping with heterogeneous (from different physics modules) workloads on large systems, which is why task-based approaches to parallelization have gained popularity. LANL developed a modern C++ framework FleCSI to help writing multiphysics codes while providing abstractions for details of the task runtime and GPU vendor library. We propose to implement a prototype 3D CFD code using FleCSI and our own solvers library FleCSolve and test it on LANL’s new supercomputer Vendado.
Desired relevant skills, background, or interests:
The most important thing for this project is an interest and enjoyment of programming and numerical simulations. This is an exploratory project that can be focused on different aspects of computer science, applied math, or physics depending on the interests of the student. A general background in programming is required, but other skills can be picked up during the project as needed. Other useful (but not required) skills are: modern C++, (non)linear iterative solvers, computational fluid dynamics methods, high-performance computing
Other comments:
Do any special requirements apply? Permanent Resident OK; International OK
Other, specify:
Keywords:
high-performance computing; linear solvers; computational fluid dynamics; C++
Lightning Talk Title: A modern approach to implicit fluid dynamics