Car-Parrinello molecular dynamics on massively parallel computers

Abstract

The work of Car and Parrinello created the field of ab initio molecular dynamics. Since then, the CP scheme has been applied to many different simulation methods, but the originally proposed combination with Kohn-Sham-based density functional theory in the pseudopotential/plane-wave framework has proven to be the most successful one. Started in the realm of semiconductor solid-state physics, its combination of accuracy and flexibility allowed the method to have a large impact in many different fields, most noticeably in liquids and solutions, catalysis and enzymatic reactions. Another non-negligible reason for its success is, that the CPMD method could be adapted optimally to the emerging parallel-computer platforms in the 19905. Combining the increase in computer power (about a factor of 300 in the past 10 years) with algorithmic improvements pushed the limits of simulations to larger systems and longer time-scales, therein we have shown parallelization strategies that allow the enhancement of the scalability of the standard Car-Parrinello algorithm on two emerging families of supercomputers, namely, clustered SMP servers and ultra-dense supercomputers. Results on diverse physical systems having sizes ranging from 100-1000 atoms exhibit good scalability to thousands of processors and molecular dynamics throughputs ranging from two to approximately 200 ps per week. These results make us confident that the CPMD method will continue to play an important role in ab initio molecular dynamics simulations. Most noticeably for systems ranging up to 1000 atoms and in connection with multiscale modelling, both for length and time-scales, CPMD will remain a leading method. It will have a continuing impact among others in materials science, simulation of liquids and biological systems. © 2005 Wiley-VCH Verlag GmbH & Co. KGaA.