MFEM

MFEM
Jakub Cerveny · CC BY-SA 4.0 · source
Name	MFEM
Developer	Lawrence Berkeley National Laboratory
Released	2010
Programming language	C++
Operating system	cross-platform
Genre	numerical analysis, finite element library
License	BSD-3-Clause

MFEM is a lightweight, flexible C++ library for finite element discretizations of partial differential equations. It provides tools for high-order finite element methods, parallel linear algebra, and mesh handling used in scientific computing projects supported by national laboratories and academic institutions. MFEM interoperates with solvers, mesh generators, and visualization packages to enable simulations in computational physics, engineering, and geosciences.

Overview

MFEM targets high-performance computing applications requiring Lawrence Berkeley National Laboratory-grade numerical libraries and interfaces with MPI, PETSc, Hypre, Trilinos, SuiteSparse, and CUDA ecosystems. It supports high-order finite elements, mixed formulations, discontinuous Galerkin methods, and spectral element approaches compatible with meshes produced by Gmsh, Cubit, Netgen, and MeshAdapt. MFEM's design emphasizes extensibility for research groups at institutions like University of California, Berkeley, Massachusetts Institute of Technology, and Stanford University collaborating on multiphysics problems.

History and Development

Development began at Lawrence Livermore National Laboratory-adjacent projects and continued under sponsorship from U.S. Department of Energy programs including initiatives at Exascale Computing Project and collaborations with National Energy Research Scientific Computing Center. Key contributors include teams associated with Berkeley Lab, Argonne National Laboratory, and university groups from University of Michigan, Princeton University, and University of Texas at Austin. MFEM's roadmap reflects advances from conferences such as SIAM Conference on Computational Science and Engineering, International Conference on Spectral and High Order Methods, and workshops held by ACM/IEEE. Over successive releases, MFEM integrated capabilities originating in projects at Oak Ridge National Laboratory and code patterns influenced by libraries like deal.II and FEniCS.

Features and Architecture

MFEM implements finite element discretizations with abstractions for meshes, finite element spaces, bilinear forms, linear solvers, and preconditioners. Core modules interoperate with external packages such as BLAS, LAPACK, MKL, and accelerator runtimes including OpenMP and OpenCL. The library supports high-order curved elements, adaptive mesh refinement workflows influenced by techniques from DUNE and libMesh, and time integration approaches employed in projects at Los Alamos National Laboratory and Sandia National Laboratories. MFEM's object model enables coupling to solvers in Trilinos packages like Amesos and Ifpack2 and to multigrid frameworks such as BoomerAMG from Hypre.

Applications and Use Cases

Researchers employ MFEM in simulations spanning computational fluid dynamics studied at California Institute of Technology, seismic wave propagation investigated by teams at United States Geological Survey, and electromagnetics projects aligned with Naval Research Laboratory priorities. MFEM has been used in magnetohydrodynamics studies related to Princeton Plasma Physics Laboratory, climate modeling efforts with collaborators at National Center for Atmospheric Research, and structural mechanics applications within European Organization for Nuclear Research. Projects integrating MFEM often interface with visualization tools like ParaView and VisIt and couple to data frameworks such as HDF5 and ADIOS for I/O at leadership computing facilities like Argonne Leadership Computing Facility and Oak Ridge Leadership Computing Facility.

Performance and Scalability

MFEM emphasizes performance portability on architectures ranging from commodity clusters at National Science Foundation shared systems to GPU-accelerated platforms deployed at Oak Ridge National Laboratory and Argonne National Laboratory. Benchmarks compare MFEM-based solvers against implementations using PETSc and Trilinos on problems from exascale workflows discussed at International Supercomputing Conference and SC Conference. Scalability strategies include algebraic and geometric multigrid techniques pioneered in research at Lawrence Livermore National Laboratory and optimized matrix-free operator applications compatible with TensorFlow-adjacent hardware for hybrid workflows. MFEM supports performance tuning with vendor libraries such as NVIDIA CUDA libraries and Intel performance tools used at Los Alamos National Laboratory.

Installation and Support

MFEM is distributed with build configurations for compilers from GNU Compiler Collection, Intel Compiler, and Clang, and provides CMake-based integration similar to ecosystems used by Boost and Eigen. Prebuilt binaries and source packages are deployed by institutions like Berkeley Lab and mirrored in repositories utilized by GitHub workflows and continuous integration services such as Travis CI and GitLab CI/CD. Users obtain community and developer support through mailing lists, issue trackers inspired by practices at Apache Software Foundation, and documentation modeled after resources at National Institute of Standards and Technology.

Licensing and Community

MFEM is released under a permissive BSD-style license compatible with software policies at Lawrence Berkeley National Laboratory and broadly adopted by academic and industrial partners including teams at General Electric, Siemens, and Raytheon Technologies. The project maintains collaborations with initiatives like Exascale Computing Project and academic consortia at Imperial College London and ETH Zurich. Community engagement occurs via workshops at SIAM, code sprints hosted by Argonne National Laboratory, and contributions from students and researchers affiliated with institutions such as University of Illinois Urbana-Champaign and University of Cambridge.

Category:Numerical libraries