MFC 5.0: An exascale many-physics flow solver
Benjamin Wilfong, Henry A. Le Berre, Anand Radhakrishnan, Ansh Gupta, Daniel J. Vickers, Diego Vaca-Revelo, Dimitrios Adam, Haocheng Yu, Hyeoksu Lee, Jose Rodolfo Chreim, Mirelys Carcana Barbosa, Yanjun Zhang, Esteban Cisneros-Garibay, Aswin Gnanaskandan, Mauro Rodriguez Jr, Reuben D. Budiardja, Ste · 2025
Read MFC 5.0: An exascale many-physics flow solver by Benjamin Wilfong, Henry A. Le Berre, Anand Radhakrishnan, Ansh Gupta, Daniel J. Vickers, Diego Vaca-Revelo, Dimitrios Adam, Haocheng Yu, Hyeoksu Lee, Jose Rodolfo Chreim, Mirelys Carcana Barbosa, Yanjun Zhang, Esteban Cisneros-Garibay, Aswin Gnanaskandan, Mauro Rodriguez Jr, Reuben D. Budiardja, Ste (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-07953.txt.
Introduction
Benjamin Wilfong leads a large team releasing version five of an open flow solver built for exascale machines. Their paper appeared in two thousand twenty-five. MFC simulates flows where many physics meet: bubbles, shocks, and phase change. Version three, published earlier, already served the open-source community. This update is a major step, not a patch. New physical models and numerical methods arrive together.
Graphics processors and modern superchips now accelerate the code. Performance tests show ideal scaling on huge runs. Engineering, medicine, and basic science all use such solvers. The broad feature set covers well-established and novel methods alike. High-fidelity flow simulation gains a faster public workhorse.
Catalog record
| CREATOR | Benjamin Wilfong, Henry A. Le Berre, Anand Radhakrishnan, Ansh Gupta, Daniel J. Vickers, Diego Vaca-Revelo, Dimitrios Adam, Haocheng Yu, Hyeoksu Lee, Jose Rodolfo Chreim, Mirelys Carcana Barbosa, Yanjun Zhang, Esteban Cisneros-Garibay, Aswin Gnanaskandan, Mauro Rodriguez Jr, Reuben D. Budiardja, Ste |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 116 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-07953
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
MFC 5.0: An exascale many-physics flow solver Benjamin Wilfong, Henry A. Le Berre, Anand Radhakrishnan, Ansh Gupta, Daniel J. Vickers, Diego Vaca-Revelo, Dimitrios Adam, Haocheng Yu, Hyeoksu Lee, Jose Rodolfo Chreim, Mirelys Carcana Barbosa, Yanjun Zhang, Esteban Cisneros-Garibay, Aswin Gnanaskandan, Mauro Rodriguez Jr, Reuben D. Budiardja, Stephen Abbott, Tim Colonius, Spencer H. Bryngelson Abstract Many problems of interest in engineering, medicine, and the fundamental sciences rely on high-fidelity flow simulation, making performant computational fluid dynamics solvers a mainstay of the open-source software community. Previous work MFC 3.0 was made a published, documented, and open-source solver via Bryngelson et al. Comp. Phys. Comm. (2021) with numerous physical features, numerical methods, and scalable infrastructure. MFC 5.0 is a significant update to MFC 3.0, featuring a broad set of well-established and novel physical models and numerical methods, as well as the introduction of GPU and APU (or superchip) acceleration. We exhibit state-of-the-art performance and ideal scaling on the first two exascale supercomputers, OLCF Frontier and LLNL El Capitan. Combined with MFC’s single-accelerator performance, MFC achieves exascale computation in practice, and achieved the largest-to-date public CFD simulation at 200 trillion grid points as a 2025 ACM Gordon Bell Prize finalist. New physical features include the immersed boundary method, N N -fluid phase change, Euler–Euler and Euler–Lagrange sub-grid bubble models, fluid-structure interaction, hypo- and hyper-elastic materials, chemically reacting flow, two-material surface tension, magnetohydrodynamics (MHD), and more. Numerical techniques now represent the current state-of-the-art, including general relaxation characteristic boundary conditions, WENO variants, Strang splitting for stiff sub-grid flow features, and low Mach number treatments. Weak scaling to tens of thousands of GPUs on OLCF Summit and Frontier and LLNL El Capitan achieves efficiencies within 5% of ideal to over 90% of their respective system sizes. Strong scaling results for a 16-times increase in device count show parallel efficiencies over 90% on OLCF Frontier. MFC’s software stack has undergone further improvements, including continuous integration, which ensures code resilience and correctness through over 300 regression tests; metaprogramming, which reduces code length while maintaining performance portability; and code generation for computing chemical reactions. † † Code available at: https://github.com/MFlowCode/MFC Program title : MFC 5.0 Developer’s repository link : https://github.com/MFlowCode/MFC Licensing provisions : MIT license (MIT) Programming language : Fortran08 and Python Journal Reference of previous version : doi: 10.1016/j.cpc.2020.107396 Does the new version supersede the previous version? : Yes Reasons for the new version : An abundance of new features and capabilities that are challenging to convey outside of an academic article. Nature of problem : Simulation of compressible flows requires careful physical model selection and treatment of spatial derivatives to keep solutions physically consistent and free of numerical artifacts. The associated methods should be high-order accurate to reduce computational cost. Solution method : The present software uses multiple physical models and numerical schemes for the treatment of multi-phase, multi-component, chemically reacting, and magnetohydrodynamic flows. Additional physical effects and models are included. The numerical method is based on the finite volume method. 1 Introduction MFC was introduced to the literature and open source community via Bryngelson et al. [17] , which described the software design and features of MFC 3.0, a CPU-based compressible multi-component flow code (MFC) for the 5-equation diffuse interface methods of Kapila et al. [49] and Allaire et al. [4] and the six equation diffuse interface model of Saurel et al. [78] . MFC 3.0 included other features that are discussed throughout this manuscript. Since MFC 3.0’s release, the MFC community has added a broad set of physical models, numerical methods, GPU offloading capabilities, and performance optimizations. This paper provides a holistic overview of MFC’s development over the past four years. We present physical models, numerical methods, validity, software robustness and testing, and performance on diverse architecture sets at scale on exascale machines and beyond. 1.1 The history of MFC The MFC codebase archaeology dates back to the mid-2000s. At this time, MFC was an unnamed solver developed by Eric Johnsen. This code represented a multi-component flow using the diffuse interface method, as described in Johnsen [47] and Johnsen and Colonius [46] . Following this effort, Vedran Coralic performed a major code rewrite and adorned MFC with its name. These features adorned MFC with a multi-dimensional numerical reconstruction treatment and are described in Coralic and Colonius [24] . Jomela Meng added cylindrical coordinate systems and corresponding simulations of high-speed shock–droplet interaction [ 56 ] . Kazuki Maeda added an Euler–Lagrange representation of sub-grid bubble dynamics [ 53 ] and acoustic wave generation [ 52 ] . Kevin Schmidmayer and Spencer Bryngelson added the 5-equation model of Kapila et al. [49] and the 6-equation model of Saurel et al. [78] . These efforts were conducted in the research group of Tim Colonius at the California Institute of Technology. Spencer Bryngelson later led a restructuring of MFC at the Georgia Institute of Technology, which included additional modeling and numerical features. This effort culminated in the open-sourcing of MFC 3.0 in 2020 and is largely described in Bryngelson et al. [17] . Modeling, method, and software scalability and portability capabilities were added in subsequent years, which are discussed in this manuscript. 1.2 New MFC capabilities To make high-fidelity CFD a practical instrument for today’s many-physics problems, a solver must span liquids, gases, and solids (at least) and accommodate complex geometries and broad separations of time and length scales. At the same time, this should be accomplished efficiently on modern accelerators, which serve as the backbone of current supercomputing centers. MFC 5.0 is architected around that premise. The new capabilities unify established and novel models. Each of these capabilities is accompanied by state-of-the-art numerics and GPU/APU offloading, allowing a single code to transition from low-Mach, geometry-rich flows to shock- and detonation-dominated regimes at exascale. In this sense, the new capabilities were crafted to establish an infrastructure for credible, end-to-end simulations across engineering, medicine, and the sciences. MFC now includes more capable physical models and numerical methods, software resistance through continuous integration and deployment, code coverage testing, and computation acceleration via GPU and APU devices. With these, MFC conducted the largest publicly known CFD simulation at 200T grid points, or 1 quadrillion degrees of freedom [ 103 ] , and was an ACM 2025 Gordon Bell Prize Finalist. Physical features include six new models for phase change, non-polytropic sub-grid bubble dynamics, treatment for elastic materials, chemical reactions and combustion, and surface tension. An immersed boundary method that supports complex geometries from analytical level sets or STL files. New numerical methods include generalized characteristic boundary conditions, improved shock capturing with TENO and WENO-Z constructions, Strang splitting for stiff sub-grid dynamics, and treatment for low-Mach number flow. Significant software additions have made MFC a portable and performant solver on CPUs, GPUs, and APUs from various vendors, including Intel, AMD, and NVIDIA. Decreases in runtime are, in part, enabled via metaprogramming and static code generation. 1.3 Use of MFC MFC has a history of being used for early access programs on flagship supercomputers. At the time of writing, these programs include JSC JUPITER (JUREAP) and the LLNL El Capitan and OLCF Frontier final and early access systems, including LLNL Tioga (AMD MI300A testbed) and OLCF Spock and Crusher (AMD MI100 and MI250X testbeds). Commodity