This program is used as a benchmark to demonstrate the benefits of parallelization, hp-adaptive methods and matrix-free methods combined.
All results of the following paper have been generated with this program.
@article{fehling_bangerth_2023,
author = {Fehling, Marc and Bangerth, Wolfgang},
title = {Algorithms for Parallel Generic hp--adaptive Finite Element Software},
journal = {ACM Transactions on Mathematical Software},
year = 2023,
volume = 49,
number = 3,
pages = {1--26},
doi = {10.1145/3603372},
url = {https://dl.acm.org/doi/10.1145/3603372}
}
This program requires a deal.II library built from the current master branch (version 9.4.0-pre). It needs to be configured with both p4est and LAPACK.
Further, your deal.II library has to be configured with either PETSc or Trilinos for algebraic multigrid (AMG) methods. The geometric multigrid (GMG) implementation currently requires Trilinos and is not compatible with PETSc.
To generate a makefile for this code using CMake, create a build directory to your liking and type the following command into the terminal from the build directory:
cmake /path/to/hpbox -DDEAL_II_DIR=/path/to/deal.II
To compile the program with all of the applications, call:
make
An executable named hprun will be created. Running the application
will automatically generate a default parameter file in the folder where
the executable is located, and uses the default parameter set for the
current run.
A selection of parameter files for different scenarios is located in the
examples folder. In addition, you will also find the parameter files
that were used for data generation in the above mentioned paper.
The author would like to thank Peter Munch (@peterrum) for lots of detailed discussions on the multigrid topic.