ASGarD - Adaptive Sparse Grid Discretization

To cite the ASGarD code in your work, please use: doi:10.11578/dc.20201125.5

Papers using ASGarD:

Sparse-grid Discontinuous Galerkin Methods for the Vlasov-Poisson-Lenard-Bernstein Model

The ASGarD project has the goal of building a solver specifically targeting high-dimensional PDEs where the "curse-of-dimensionality" has previously precluded useful continuum / Eularian (grid or mesh based as opposed to Monte-Carlo sampling) simulation. Our approach is based on a Discontinuous-Galerkin finite-element solver build atop an adaptive hierarchical sparse-grid (note this is different from the "combination tecnique" when applied to sparse-grids).

The developer documentation contains information about how to contribute to the ASGarD project.

(TODO) user docs about building/using the code
(TODO) docs about the method

Contact Us

Issues are a great way to discuss all aspects of the ASGarD project, whether it is to ask a general question, request a new feature, or propose a contribution to the code base.

The ASGarD project was initiated by David Green at Oak Ridge National Laboratory.

For technical questions, contact Miroslav Stoyanov (stoyanovmk@ornl.gov) at Oak Ridge National Laboratory.

Automated Test Status

Test	Status (Develop)
format/clang
warnings/clang
unit/g++
unit/clang++
unit/g++/mpi
unit/g++/cuda
unit/g++/io

Dependencies

C++17
cmake 3.19
blas

Optional depedencies

cuda
mpi
HighFive/hdf5
MATLAB
ScaLAPACK

Quickstart

Download and build

git clone https://github.com/project-asgard/asgard.git
cd asgard
mkdir build && cd build
cmake ../
make
ctest
./asgard

For best performance (especially on accelerators) please pass -DCMAKE_BUILD_TYPE=Release to disable asserts when building the code.

For specific platform build instructions, see this wiki page.

Example Execution

Examples demonstrating ASGarD's capabilities are available in the wiki

To see a list of available PDEs, run ./asgard --available_pdes. The listed PDEs can be selected using the -p argument to asgard.

To see the list of all runtime options, run ./asgard --help.

HighFive/HDF5 IO

To enable HDF5 wavelet and realspace output, pass the -DASGARD_IO_HIGHFIVE=ON.

Pass -DASGARD_BUILD_HDF5=TRUE and ASGarD will download and build HDF5 and HighFive libraries. If HIGHFIVE and/or HDF5 are already available, pass -DASGARD_HIGHFIVE_PATH=<path to HighFive> and/or -DASGARD_HDF5_PATH=<path to HDF5>.

MATLAB Interface

Asgard can interface with MATLAB to plot, share data, and run scripts. To enable this, compile with -DASGARD_USE_MATLAB=ON.

If using only for plotting purposes, then Asgard can be run without any additional arguments. However, for sharing data with MATLAB, Asgard will need to connect to a shared session. This can be done in MATLAB by running

matlab.engine.shareEngine

Asgard should automatically connect with this shared session, but the engine name can also be specified directly by giving the result of

matlab.engine.engineName

to Asgard with the --matlab_name option.

Note: do not start MATLAB in the DG-SparseGrid repo when trying to use this for plotting.

ScaLAPACK Integration

The ASGARD_USE_SCALAPACK build option adds the "scalapack" solver option to ASGarD. In the implicit timestep method, this option distributes the matrix solve across multiple processes and nodes and enables larger problems that are too large for a single computer.

This build option is only available when ASGARD_USE_MPI=ON. Unfortunately, setting ASGARD_USE_SCALAPACK alone is not sufficient and doesn't change the ASGARD_USE_MPI setting.

mkstoyanov / asgard