Skip to content

Latest commit

 

History

History
executable file
·
111 lines (67 loc) · 5.45 KB

README.md

File metadata and controls

executable file
·
111 lines (67 loc) · 5.45 KB
Raw

PYDEM

python bindings for granular DEM simulations.

Useful Links:

The LAMMPS simulator project.

Pygame, an optional dependency for simple visualisations.

Basics

This is a nearly-stable set of python bindings for LAMMPS, built around the ctypes interface in LAMMPS/src/library.h. It currently requires a few small patches to the LAMMPS code, which I will provide as a patch file here until they are accepted into the stable lammps release.

All versions will operate in approximately the following way:

  • allow you to import a normal python module
  • allow you to initialise data (like interactions, particle positions, boundary conditions, etc) in python.
  • allow you to run the simulation, and retrieve the data in memory afterwards,
  • allow you to save and restore the particles in compressed form. (support for lammps binary restart files will be added, too.)

in fact, it looks like the python/C++ data transfer will be encapsulated within ordinary operations on the interface, with fallback to manual controls if you're feeling technical.

This Version

PROGRESS REPORT Version 2 is nearly ready - most features are implemented - next job to debug a few physics issues using a visualiser which uses pygame. This will not be required in the final build - so no broken imports.

This is now Version 2, in alpha. We can run (and visualise in 2D) a granular simulation of polydisperse spheres, although there are some bugs still to be nailed down (see the TODO file for details).

NOTE this version does NOT support MPI parallel computations - they might work, but I have little desire to test them! This library will be bundled with a tool for running multiple simulations at once, as an alternative to this functionality, at a later date.

Future Versions

CANCELLED Version 1 aims to implement this by generating very simple LAMMPS scripts, calling them using os.system(), and retrieving the data dump files and parsing them for you. Since the ultimate aim is to avoid the domain speific LAMMPS language, this version will not support constraints, tests or loops in LAMMPS; it will run one granular simulation for a specified number of timesteps and then scoop up the data.

NEW TARGET Version 2 will use a simple C interface combined with Cython to expose a well defined set of functionality for passing data between Python and C++ land. It will also provide an interface for using LAMMPS compute functionality, although the results of computes will be passed all the way out to Python for processing and mid-simulation logic, so that one can use rich language features to if and elif, loop and raise, map and reduce. Note that you'll still be able to do this in version 1, it'll just be a lot slower, and you'll have to implement your own computes.

Version 3 will support aspherical particles in a similar way to LIGGGHTS - i.e. via coupled/rigid sets of spheres.

Installation Instructions:

These instruction are for Mac and Linux (tested on a mac, so if you find a Linux bug do report it on github!) If you're on Windows, all I can suggest is that you try installing cygwin.

Building LAMMPS is big and complicated and tricky, they have a whole documentation section devoted to it; but fear not! it is slightly simpler to do, if you only want support as a python library.

  1. check out lammps with git like so (much faster than the old SVN checkouts):

    git clone http://git.icms.temple.edu/lammps-ro.git lammps-checkout

  2. install the GRANULAR module

    cd lammps-checkout/src make yes-granular cd ..

  3. for now apply my patch (old patches here have been merged in a bit.)

    patch -p0 -i pydem-checkout-path/patches/0001-adding-support-for-atom-deletion-by-id.patch

(where pydem-checkout-path is a valid path to a checkout of this project, obviously.)

  1. we now actually build the dynamic library in this step (before it was just to generate some of the class headers.)

    cd src

    e.g. for mac:

    make -f Makefile.shlib mac

    or linux:

    make -f Makefile.shlib serial

    either of these will be followed by loads of output from make and gcc.

(type make for a full list of makefiles included.)

4.1) double check the locations that CDLL will search for shell libraries. On Mac OS X, the system searches /usr/local/lib before /usr/lib; on debian linux it only searches the latter!

man dlopen

4.2) install lammps python module as usual using setup.py (note, I tend to do this in two steps so that the temporary build files are not owned by root, which is annoying to clean up later.)

cd ../python
sudo python install.py /library/load/path

(where /libarary/load/path is something like /usr/lib -- the default for this value is /usr/local/lib/)

  1. Finally, you're ready to install my python bindings! Same drill as the last step:

    cd pydem-checkout-path python setup.py build sudo python setup.py install # (or python setup.py install --user)

You're now ready to test it all using my example script for a 2D simulation with visualisations (which requires pygame, which comes with oneclick installers for most platforms.)

# never run scripts in a module's build directory (because shenanigans...)
cp test_run.py ~/
cd ~
python test_run.py

Take a peek inside test_run.py to see the key classes in action, and an example energy evaluation function.