Installation & Setup Guide
System Requirements
| Component | Requirement |
|---|---|
| OS | Linux (tested on Ubuntu 22.04) |
| GPU | NVIDIA GPU (Ampere or newer recommended) |
| CUDA | 12.x |
| Compiler | NVIDIA HPC SDK 25.5 or newer (mpif90) |
| MPI | OpenMPI bundled with HPC SDK |
| CMake | 3.18 or newer |
| fypp | any version (pip install fypp) |
| Python | 3.10 or newer (optional, for Python API and tests) |
1. Install NVIDIA HPC SDK
The solver requires mpif90 from the NVIDIA HPC SDK. Download the installer from:
After installation, verify the tools are on your PATH:
2. Install CMake
Most Linux distributions include a recent CMake. If yours does not:
Verify:
3. Install fypp
CMake calls fypp to expand the .f90.fypp Fortran templates at build time:
4. Clone the Repository
5. Install the Python Package (optional)
Install the ouxsbli Python package and test dependencies:
This installs:
- ouxsbli — the Python API for parametric runs (editable install)
- numpy — array operations
- scipy — analytical reference solutions (Blasius, oblique-shock, Sod)
- vtk — VTK file reading
- pytest — for running the test suite
The install is editable on purpose: the solver sources stay in the checkout and are found through the
package location, so import ouxsbli and the ouxsbli command work from any directory. To use another
checkout without reinstalling, set OUXSBLI_ROOT=<checkout>. If an older, non-editable ouxsbli is
already in the environment, remove it first (pip uninstall ouxsbli).
Tip: Use a virtual environment to keep dependencies isolated:
6. Build a Test Case
Each case directory contains a CMakeLists.txt. To build the NS Taylor-Green vortex case:
CMake runs fypp on all .f90.fypp templates, then compiles with mpif90. The executable a.out lands in build/.
7. Run the Simulation
VTK output files (Q00000.vtr, Q00001.vtr, …) appear in data/. Open them in ParaView as a time series.
Available Cases
2D Solver (2D_solver/)
| Directory | Physics | Description |
|---|---|---|
BL/ |
Navier-Stokes | Supersonic laminar boundary layer (M=2) |
DSL/ |
Euler | Double shear layer |
EVC/ |
Euler | Euler vortex convection (grid-convergence study) |
OS/ |
Euler | 2D oblique shock (M=2, θ=8°) |
SBLI/ |
Navier-Stokes | 2D shock-boundary layer interaction |
ST/ |
Euler | Sod shock tube |
3D Cartesian Solver (3D_solver/)
| Directory | Physics | Description |
|---|---|---|
BL/ |
Navier-Stokes | Quasi-2D laminar flat-plate boundary layer (extrudes 2D_solver/BL; validated vs. Blasius) |
DHIT/ |
Navier-Stokes | Decaying homogeneous isotropic turbulence |
ETGV/ |
Euler | Entropy-preserving Taylor-Green vortex |
EVC/ |
Euler | Quasi-2D Euler vortex convection (extrudes 2D_solver/EVC; grid-convergence study) |
IVST/ |
Euler | Inviscid vortex smooth test case |
KHI/ |
Euler | Kelvin-Helmholtz instability |
NSTGV/ |
Navier-Stokes | NS Taylor-Green vortex (Re=1600, M=1.25) |
OS/ |
Euler | Quasi-2D oblique shock + wall reflection (extrudes 2D_solver/OS; validated vs. Rankine-Hugoniot) |
SBLI/ |
Navier-Stokes | Shock-boundary layer interaction (RESCALE=True) |
STZ/ |
Navier-Stokes | z-direction MPI halo exchange validation |
TBL/ |
LES | Turbulent boundary layer (Hybrid scheme) |
Curvilinear Solver (3D_solver_curv/)
| Directory | Physics | Description |
|---|---|---|
NACA/ |
Navier-Stokes | NACA 0012 airfoil on an O-grid |
CORN/ |
Navier-Stokes | Compression corner (M=2, θ=8°) |
Note: The curvilinear solver also uses CMake. Build the same way:
cmake -B build && cmake --build build -j
Configuration Reference
Compile-time scheme and physics choices are set in config.fypp, not in mod_globals.f90. See Configuration for details.
At a glance, the dispatch encoding used internally by mod_constant.f90:
| Parameter | kind=2 | kind=4 | kind=8 |
|---|---|---|---|
id_visc |
Euler | Navier-Stokes | LES |
id_accuracy |
2nd order | 4th order | 6th order |
id_tvd |
no limiter | tvd | hybrid |
id_scheme |
Type | Scheme |
|---|---|---|
integer(2) |
KEEP | Kinetic Energy & Entropy Preserving |
real(2) |
SLAU | Simple Low-dissipation AUSM |
real(8) |
Hybrid | KEEP ↔ SLAU via Ducros sensor |
These Fortran parameters are generated automatically from config.fypp — you do not edit them directly.