"Acceleration of molecular dynamics simulations using the fast multipole method and graphics processing units"
Marin D.F., Malyshev V.L., Moiseeva E.F., Gumerov N.A., Akhatov I.Sh., Mikhailenko K.I.

This work contains the speed-up results for the molecular dynamics simulations of electrostatic potentials in the application to water molecules. High-performance computations are achieved via the fast multipole method (FMM) for force calculation and the use of heterogeneous architectures, consisting of central and graphics processing units (CPU and GPU, respectively). FMM allows one to speed up the calculations of the far-field interactions by reducing the computational complexity from square complexity to the linear one. The use of the GPUs allows one to significantly speed up the computations. The realization of the FMM on GPU provides an opportunity to implement the computation experiments for large data sets. This paper shows that the method described has a great scalability and can be used to simulate the water dynamics for domains of 60 nm or for the number of molecules of over 10 million on personal workstations equipped with a single GPU.

Keywords: molecular dynamics, electrostatic interaction, fast multipole method, FMM, heterogeneous computations, GPU

Marin D.F., e-mail: marcom.dimar@gmail.com;   Malyshev V.L., e-mail: victor.l.malyshev@gmail.com;   Moiseeva E.F., e-mail: elena.f.moiseeva@gmail.com;   Mikhailenko K.I., e-mail: const.mkh@gmail.com; – Center for Micro- and Nanoscale Dynamics of Dispersed Systems, Bashkir State University; ulitsa Validi 32, Ufa, 450076, Russia
Gumerov N.A., e-mail: gumerov@umiacs.umd.edu; – University of Maryland, Institute for Advanced Computer Studies; Williams Building, College Park, MD 20742, USA
Akhatov I.Sh., e-mail: e-mail: Iskander.Akhatov@ndsu.edu – North Dakota State University, Department of Mechanical Engineering and Applied Mechanics; NDSU Dept 2490, 210 Dolve Hall, Fargo, ND 58108, USA