Towards high-performance discontinuous Galerkin simulations of reacting flows using legion
Abstract/Contents
- Abstract
- High-fidelity scale-resolving computational fluid dynamics (CFD) simulations may provide a path towards predictive capabilities for many engineering-relevant applications, in particular those involving the interaction between multiple physics such as turbulent multi-component reacting flows. However, their routine use is hindered by their cost. The throughput of the fastest supercomputers is fortunately still increasing but this comes at the cost of specialized devices and hardware heterogeneity. Modern supercomputing thus imposes new requirements onto numerical algorithms and their implementation for leveraging the available compute while significantly complicating the design, development, and maintenance of massively parallel computational physics software. This dissertation focuses on addressing numerical and implementation challenges of CFD of multi-component reacting flows. It considers high-order discontinuous Galerkin (DG) discretization methods as a way towards a more efficient use of modern compute architectures for performing high-fidelity simulations of configurations involving complex geometries. Both the high accuracy requirement and the ability to accommodate unstructured meshes are key motivations for investigating this class of numerical schemes. The entire content of this manuscript is based on DG-Legion, a novel unstructured DG CFD solver for the compressible Navier-Stokes equations written on top of the Legion system for targeting distributed heterogeneous architectures. The main challenge related to the computer science aspect was to scale the code on GPU supercomputers. An explicit ghost formulation is proposed which results in performance competitive with state-of-the-art solvers based on the message-passing interface (MPI) system while the application code remains free of explicit communication and synchronization, i.e. it has sequential semantics. The contribution related to the numerics addresses two important questions. First, the issue of spurious pressure oscillations generated by fully-conservative schemes for thermally perfect mixtures is analyzed in detail. A generic framework is developed in order to generate systematic variations of the baseline scheme with improved properties. When applied to the problem at hand, several techniques that were previously proposed independently are recovered as special cases in a rigorous way. Finally, a hybrid strategy for preserving the positivity of DG solutions for the composition-describing scalars is introduced. The method based on the composition of subcell finite volume schemes and linear-scaling limiters leads to an overall positivity-preserving scheme in the inviscid limit, compatible with higher-order quadrature rules, and better preserving the subcell resolution compared to only using rescaling techniques.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Bando, Kihiro |
|---|---|
| Degree supervisor | Aiken, Alexander |
| Degree supervisor | Lele, Sanjiva K. (Sanjiva Keshava), 1958- |
| Thesis advisor | Aiken, Alexander |
| Thesis advisor | Lele, Sanjiva K. (Sanjiva Keshava), 1958- |
| Thesis advisor | Alonso, Juan José, 1958- |
| Degree committee member | Alonso, Juan José, 1958- |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Aeronautics and Astronautics |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Kihiro Bando. |
|---|---|
| Note | Submitted to the Department of Aeronautics and Astronautics. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/pm373kw0126 |
Access conditions
- Copyright
- © 2023 by Kihiro Bando
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...