figr is a GPU-accelerated finite volume solver for compressible multi-component flows using Information Geometric Regularization (IGR). It's a miniapp based on MFC. IGR replaces traditional viscous or numerical shock-capturing methods with an elliptic regularization of the Euler equations, enabling high-order accurate representations of shock-laden flows.
figr discretizes the compressible Euler equations on structured Cartesian grids using a finite volume method. Instead of solving Riemann problems at cell interfaces, IGR introduces a regularization parameter that smooths discontinuities via an elliptic PDE solve at each time step. The method supports IGR orders 3 and 5, Jacobi and Gauss-Seidel iterative solvers, multi-component flows, viscous flows, 1D/2D/3D problems, MPI parallelism, and GPU acceleration via OpenACC and OpenMP target offload.
- R. Cao and F. Schäfer, "Information geometric regularization of the barotropic Euler equation," SIAM Journal on Scientific Computing, 2026. arXiv:2308.14127
- R. Cao and F. Schäfer, "Information geometric regularization of unidimensional pressureless Euler equations yields global strong solutions," 2024. arXiv:2411.15121
- B. Wilfong, A. Radhakrishnan, H. Le Berre, D. J. Vickers, T. Prathi, N. Tselepidis, B. Dorschner, R. Budiardja, B. Cornille, S. Abbott, F. Schäfer, and S. H. Bryngelson, "Simulating many-engine spacecraft: Exceeding 1 quadrillion degrees of freedom via information geometric regularization," Proceedings of SC '25: The International Conference for High Performance Computing, Networking, Storage and Analysis, 14–24, 2025. arXiv:2505.07392
# Build (CPU)
./figr.sh build -j $(nproc)
# Build (GPU, OpenACC)
. ./figr.sh load -c <cluster> -m g
./figr.sh build -j $(nproc) --gpu acc
# Run an example
./figr.sh run examples/2D_IGR_double_mach/case.py -n 4
# Run tests
./figr.sh test -j 8MIT License
