MPDATA: positive definite transport equation solver

An independent implementation of MPDATA solver by Piotr K. Smolarkiewicz and Len G. Margolin (paper) used in EULAG model. Our method is designed with built-in grid decomposition for multithread/multiprocess parallelism. We develop CUDA kernels for calculating concentrations and pseudo-velocities on GPUs. The source code includes multiple simple physical test cases.

Prerequisites

• CMake building tool:

$ sudo apt-get install cmake

• NETCDF headers & libraries:

$ sudo apt-get install libnetcdf-dev

• GRADS - 2D visualization tool often used in meteorology:

$ sudo apt-get install grads

• VAPOR - 3D visualization tool: download v2.3.0 (other versions might be incompatible!) from the website and install to /opt/vapor

Building on Linux

$ git clone https://github.com/apc-llc/mpdata.git
$ cd mpdata
$ mkdir build
$ cd build
$ cmake ..
$ make -j12

Testing ring2d on Linux

The ring2d test shows aerosol transfer from the source point in the rounded flow. Ideally, aerosol blob should rotate in the flow without form change and mass dissipation. In practice, finite difference approximation always has accuracy limitations resulting into deformation and energy dissipation. Such diagnostic test is often used for finite difference method verification.

$ ./ring2d 128 1024
$ perl ring2d.pl

Individual visualization frames will be save to rind2d_frames subfolder. They could we futher rendered into a movie:

$ sudo apt-add-repository ppa:mc3man/trusty-media
$ sudo apt-get update
$ sudo apt-get install ffmpeg
$ ffmpeg -i ring2d_frames/step%04d.png -c:v libx264 -pix_fmt yuv420p ring2d.mp4

See the resulting video.

Testing ring3d on Linux

$ source /opt/vapor/bin/vapor-setup.sh
$ ./ring3d 128 128

Use VAPOR to make 3D visualization (Data -> Load a Dataset into Current Session -> ring3d.vdf):

$ vaporgui

Frames captured from VAPOR could be also rendered into a movie