Graphics Processing Units (GPUs) have been the subject of considerable attention in recent years. The large number of arithmetic operations per second and the massively parallel architecture provide significant computational power for scientific applications. Molecular dynamics (MD) simulations are particularly amenable to GPU-based computation given the fact that calculation of forces acting on atoms can be easily parallelised. Liquid crystal simulations can benefit greatly from the computing power of GPUs since an accurate description of interactions and a large number of molecules in the simulated fluid are essential to investigate liquid crystal characteristics.
Our custom code for MD simulations of liquid crystals has been redesigned to run entirely on CUDA enabled NVIDIA GPUs. The supported molecular model is semi coarse-grained, with Gay-Berne ellipsoids for aromatic rings, Lennard-Jones united atom description of alkoxy tails and point charges placed on atoms. The bond lengths are kept fixed with a Rattle algorithm, but the molecules are otherwise flexible. A cut-off radius for short-range interactions is used to achieve linear scaling with respect to system size. We chose a smectogenic system, 2-(4-butyloxyphenyl)-5-octyloxypyrimidine, to serve as a test case for our implementation.
Algorithms for efficiently computing Lennard-Jones, Gay-Berne and electrostatic interactions on GPU are presented. We discuss approaches that are used to exploit the processing power of the GPUs and review algorithm bottlenecks. Test calculations on systems that contain from 512 to 8192 molecules show up to 300 fold increase in performance compared to a serial implementation.
The development of ferroelectric liquid crystals for device applications can be experimentally pursued by designing efficient dopants that can be added to modify the properties of an existing liquid crystal phase. However, this is a time-consuming process and may involve difficult syntheses or expensive starting materials. It is here that computer simulations can play an important role, as a complementary technique, for improving the understanding of behaviours, which are not fully accessible to experimental investigation.