Abstract: Reactive molecular dynamics (RMD) models bridge quantum-scale and classical MD approaches by explicitly modeling bond activity and redistribution of charges. As such they enable the study of important phenomena which otherwise is impractical using classical or quantum techniques. However, RMD models have a significantly complex formulation, making fast, scalable and high fidelity RMD simulations extremely challenging to achieve. In this talk, I will present our work towards addressing both the scalability and fidelity challenges. I will start by describing the parallel algorithms and numerical techniques that we developed for a fast implementation of the Reax Force Field (ReaxFF), which is used by hundreds of researchers worldwide. Particular emphasis will be on novel solvers we recently developed for the dynamic charge distribution problem that constitutes the most important scalability bottleneck in large RMD simulations. I will conclude the talk by outlining our efforts towards addressing the fidelity challenge, i) through an automated force field framework for RMD models, ii) by developing a novel hybrid ReaxFF/AMBER simulation software in the spirit of QM/MM techniques.

Bio: H. Metin Aktulga received his B.S. degree from Bilkent University in 2004, M.S. and Ph.D. degrees from Purdue University in 2009 and 2010, respectively; all in Computer Science. Before joining the Michigan State University (MSU) in 2014, he was a postdoctoral researcher in the Computational Research Division at the Lawrence Berkeley Lab.