Time Parallelization Studies in Bio-Molecular Dynamics Simulations

Molecular Dynamics (MD) is an important simulation technique with widespread use in computational chemistry, biology, and materials. An important limitation of MD is that the time step size is limited to around a femto (10-15) second. Consequently, a large number of iterations are required to simulate to realistic time spans. This is a major bottleneck in MD, and has been identified as an important challenge in computational biology and nano-materials. While parallelization has been effective in dealing with the computational effort that arises in simulating large physical systems (that is, having a large number of atoms), conventional parallelization is not effective in simulating small or moderate sized physical systems to long time spans. We recently introduced a new approach to parallelization, where data from prior simulations are used to parallelize a new computation along the time domain. We demonstrated its effectiveness in a nano-materials application, where this approach scaled to a larger number of processors than conventional parallelization. In this thesis, we parallelize a computational biology application – the AFM pulling of a protein – using this approach. The significance of this work arises in demonstrating the effectiveness of this technique in a soft-matter application, which is more challenging than the hard-matter applications considered earlier.