STKFMM-examples

Examples of use of STKFMM library (https://github.com/wenyan4work/STKFMM) via the python interface.

• Focus on Stokes equation in 3D and Stokes equation with a boundary wall.

Installation of STKFMM

• Installed on Ubuntu 16.04 system.
• Steps: install dependencies via conda, install PVFMM, install SKFMM.

1. Install prerequisites in a conda virtual environment (for reproducibility: I use conda-forge as the main channel)

conda create -n stkfmm python=3.7 # create environment named stkfmm
conda activate  stkfmm            # switch to the new environment
conda install mkl-devel scipy numpy openmpi mpi4py gcc_linux-64 gxx_linux-64 gfortran_linux-64 boost
pip install pybind11-cmake # will be used to bind STKFMM and python
conda install cmake

Optional:

conda install jupyter # to run notebooks
conda install matplotlib # for visualization
conda install numba # to run examples.py and kernels.py
conda install tbb # for numba + threading

2. Install PVFMM -> I followed steps in https://github.com/wenyan4work/Environment

   git clone https://github.com/wenyan4work/Environment
   mv Environment STKFMM-env # rename
   cd STKFMM-env
   

   1. Comment out in Compile.py everything, except the part related to PVFMM, and the partt which updates git submodules (TODO: test if Eigen and gcc are needed).
   2. Set variables

   export MKL_THREADING_LAYER=GNU         # necessary because conda uses gcc
   export MKL_INTERFACE_LAYER=GNU,LP64    # necessary because conda uses gcc
   export MKLROOT=/home/icemtel/miniconda3/envs/stkfmm/ # SET THE PATH TO THE CODNA ENVIRONMENT
   

   3. Run installation script

   python Compile.py 2>&1 | tee Compile.log # 
   

3. Install STKFMM

git clone https://github.com/wenyan4work/STKFMM.git
cd STKFMM

1. Adjust parameters in the cmake script

   • PyInterface=ON
   • BUILD_DOC=ON - a doxygen (documentation utility) html file will be generated (to edit and change execution rights)

   nano do-cmake.sh
   chmod 777 do-cmake.sh
   

2. Run cmake and make mkdir build cd build ../do-cmake.sh make

3. Download precomputed data from https://users.flatironinstitute.org/~wyan/pdata.7z and unzip to $PVFMM_DIR/pdata -> weights >3.5 GB Q: where is "$PVFMM_DIR"? I don't have such variable. [needs to be fixed to solve with periodic boundary conditions]

   • Alternative: install Eigen, and set BUILD_M2L=ON in STKFMM.

4. File /build/Python/PySTKFMM.so is the python bindings - make sure that they are visible to python (copy this file to the directory with python scripts and jupyter notebooks, or to the path with python libraries, or add /build/Python/ to the path).