The International Conference for High Performance Computing, Networking, Storage and Analysis
Scalable Ensemble Ab Initio Calculations Using the Fragment Molecular Orbital Method in GAMESS.
Authors: Yuri Alexeev (Argonne National Laboratory), Maricris Mayes (Argonne National Laboratory), Spencer Pruitt (Iowa State University), Graham Fletcher (Argonne National Laboratory), Dmitri Fedorov (National Institute of Advanced Industrial Science & Technology), Mark Gordon (Iowa State University)
Abstract: A practical solution to achieving scalability on millions of cores is to employ ‘ensemble’ calculations in which multiple related jobs are executed concurrently. Molecular dynamics (MD) is a powerful technique that allows ensemble methods to be employed while providing detailed and accurate information on structural and dynamical properties of chemical systems. However, classical molecular dynamics simulations use empirical force fields that are obtained by parameterization. Thus, the accuracy of predicted properties depends on the system. To solve this problem we use ab initio quantum chemistry dynamics implemented in the software package, GAMESS. In order to make ab initio dynamics computationally tractable, supercomputers and linear scaling methods are necessary. Specifically, we use the Blue Gene/Q supercomputer, and the linear-scaling Fragment Molecular Orbital (FMO) method, together with ensemble methods, to expedite the calculations. In this poster, we describe how the ensemble strategy can be employed with FMO in GAMESS to do MD.