Information-geometric regularization for Euler flows
Get started · Methods · Examples · Validation
IGRlab explores how information-geometric regularization (IGR) shapes compressible Euler flows, from smooth transport to strong shocks. Built on MFEM, it brings DG, DGSEM, and HDG into a shared research codebase for studying numerical behavior and computational cost on CPUs and GPUs.
Experimental research software. Capabilities and validation vary by method, polynomial order, and hardware; see the method guide for scope.
| Method | Implementation |
|---|---|
| DG — discontinuous Galerkin | Euler and IGR operators |
| DGSEM — DG spectral element method | Reference and packed operators |
| HDG — hybridizable discontinuous Galerkin | Euler and IGR solvers with local elimination |
CPU and CUDA implementations support comparisons across polynomial orders. Specialized tensor-core kernels are available for selected configurations. Explore the methods and their supported configurations →
From a source checkout, with a C++20 compiler, MPI, and Spack:
git submodule update --init
source /path/to/spack/share/spack/setup-env.sh
./build.sh --cpu-only --hdg-runtime -j 4Then run a short DG Sedov blast with IGR, using the app_build path printed
by the build:
APP_BUILD=/path/printed/as/app_build
"$APP_BUILD/examples/sedov_taylor_blast/sedov_taylor_blast" \
--device cpu --t-final 0.001 -rs 0 -rp 0 -o 1 -c 0.25Full setup and CUDA builds · PACE · Delta · Fast application-only rebuilds
| Flow problem | Example |
|---|---|
| Strong blast wave | Sedov–Taylor blast |
| Shock structure and interaction | Stationary shock · Shock–vortex interaction |
| Supersonic flow around a body | Cylinder |
| Smooth transport and method comparisons | Periodic advection |
Browse all examples; each executable provides its options with --help.
Accuracy, conservation, stability, and cost per degree of freedom guide development. The current A100 results report field comparisons across orders, targeted CUDA/MPI checks, and DOF-normalized end-to-end timings for the recorded source and configurations.
- Large DG/IGR checkpoint — 6.19× measured application speedup on the fixed 512² p5 GH200 Sedov case, with exact controls and tranche attribution.
- Numerical validation — how changes are assessed.
- GPU studies — reproducible comparisons and Nsight profiling.
- Validation records — selected CPU and SM120 checks.
- Method scope and limitations — what the evidence covers.