High-Throughput of Measure-Preserving Integrators Derived from the Liouville Operator for Molecular Dynamics Simulations on GPUs

Molecular dynamics simulation is currently a theoretical technique eligible in simulating of a wide range of systems, from soft condensed matter to biological systems. Excellent outcomes have resulted from using this technique; however, the implementation of this approach remains computationally expensive to some extent. Novel computing technologies may help reducing the computational simulation time, particularly, by using Graphical Processing Units (GPUs). Calculations on GPUs make possible to carry out simulations of large and complex molecular systems at a significantly reduced time. In this manuscript, the implementations of measure-preserving geometric integrator in the canonical ensemble coded in Compute Unified Device Architecture (CUDA) language are presented. The performance and validation of our High-throughput Molecular Dynamics (HIMD) code was done by calculating the thermodynamic properties of a Lennard-Jones fluid. From our tested systems, an excellent agreement was achieved with the reported of literature, compared with calculations carried out on Central Processing Units (CPUs). The implementation of the HIMD code performs time integration on Nose-Hoover chains (NHC) faster in comparison to the NHC method implemented in LAMMPS code tested with one CPU vs one GPU. Along this work, the scope and limitations in performing simulations by using our HIMD code under rigorous statistical thermodynamics in the canonical ensemble are discussed and analyzed.