|
NEW JERSEY INSTITUTE OF TECHNOLOGY |
Qun
Ma
Department
of Computer Science
New
Jersey Institute of Technology
Macromolecules such as proteins, ribonucleic acids (RNAs) and deoxyribonucleic acids (DNAs) are fascinating objects to study. Among the widely used computational techniques, molecular dynamics (MD) is a unique one in which a dynamic picture of the movement of the molecular system is studied by integration of a system of ordinary differential equations governed by Newton's law of motion. However, the current time stepping algorithms are not able to meet the challenges, i.e., long simulations, e.g., those on the order of microseconds, and large systems, e.g., those on the order of millions of atoms.
In this talk, I will show that the time steps allowed are limited by linear and nonlinear numerical resonances. For biological systems, the shortest periods of motion are around 10 femtoseconds. Nonlinear resonances limit the time steps allowed by the Impulse method, the state-of-the art multiple time stepping integrator, to less than 3.3 femtoseconds. This claim is supported by both analytical and numerical results. Furthermore, I will describe some recent advances in lengthening the time steps allowed in those simulations. One technique is called Langevin stabilization in which frictional and Langevin forces are introduced to the system to stabilize the numerical solution. I will illustrate a technique along this line in which the artificial forces are introduced in a pair-wise targeted manner to reduce the numerical artifacts. Preliminary results show that in some applications, the step sizes can be quadrupled than the Impulse method. We call this new multiscale method Targeted MOLLY.