Accelerated Python Approaches: Speeding Up A Lebwohl-Lasher Simulation

This repository contains seven methods to accelerate a serial Python implementation of the Lebwohl-Lasher model; simulation of liquid crystal behaviour. Each method implements a different optimisation strategy with combined approaches also explored.

---

Project Structure

1. Serial Version

A basic serial implementation, provided by Dr. Simon Hanna.

python LebwohlLasher.py <ITERATIONS> <SIZE> <TEMPERATURE> <PLOTFLAG>

2. Numba

Uses the Numba just-in-time compiler, converting the script to machine code at compilation time.

python LebwohlLasher_numba.py <ITERATIONS> <SIZE> <TEMPERATURE> <PLOTFLAG>

3. Cython

Compiles critical sections (inefficient loops) using Cython.

Install Cython:

pip install Cython

Compile:

CC=gcc-15 python setup_LebwohlLasher_cython.py build_ext -fi

Run:

python run_LebwohlLasher_cython.py <ITERATIONS> <SIZE> <TEMPERATURE> <PLOTFLAG>

4. NumPy

Vectorised implementation using NumPy arrays for efficient operations.

python LebwohlLasher_numpy.py <ITERATIONS> <SIZE> <TEMPERATURE> <PLOTFLAG>

5. OpenMP (via Cython)

Parallelised using OpenMP within Cython.

Install OpenMP:

brew install llvm
brew install libomp

Compile:

CC=gcc-15 python setup_LebwohlLasher_openmp.py build_ext -fi

Run:

python LebwohlLasher_openmp.py <ITERATIONS> <SIZE> <TEMPERATURE> <PLOTFLAG> <THREADS>

6. MPI (via mpi4py)

Distributed memory parallelisation using mpi4py.

Install MPI:

brew install open-mpi

Install mpi4py:

pip install mpi4py

Run:

mpiexec -n <TASK_COUNT> python LebwohlLasher_mpi4py.py <ITERATIONS> <SIZE> <TEMPERATURE> <PLOTFLAG>

7. Combined Approaches

Implementations combining multiple acceleration approaches:

• LebwohlLasher_numpyxmpi4py.py — NumPy + MPI
• LebwohlLasher_numpyxcython.py — NumPy + Cython

Usage mirrors the respective base methods.

Input Parameters

Argument	Description
ITERATIONS	The number of MC steps for a given simulation
SIZE	Lattice size, integer input for a side length
TEMPERATURE	Reduced temperature, T*, for a simulation
PLOTFLAG	0 for no plot, 1 for energy plot and 2 for angle plot
THREADS	(OpenMP only) Number of threads
TASK_COUNT	(MPI only) Number of parallel processes - must be >1

---

Data

Data was gathered with bash scripts that ran simulations for multiple input parameters. Data has been processed using the data_processing.ipynb script.

---

Notes

• All implementations assume Python 3.
• Plotting requires matplotlib.
• Task count, for mpi implementations must be 2 or greater, as the master task is not a worker task.