Estimation of dispersion relations from short-duration molecular dynamics simulations

Abstract

Simulation of the motion of particles in large molecular chains is computationally complex, generally involving the solution of large systems of differential equations employing iterative techniques. Generating solutions for even very short time intervals can require significant amounts of computational effort. Though the particle trajectories are significant, further insight into the molecular dynamics is often gained by attempting to predict experimentally verifiable dispersion relations. By use of conventional techniques, however, the accuracy with which frequencies (as a function of wave number) can be estimated is limited by the amount of data available, and often these techniques are completely inadequate with short-duration trajectories. Herein, a new technique (ESPRIT) for obtaining accurate estimates of the complex poles (frequencies of vibration as well as damping coefficients) associated with dominant modes of oscillation present in short-duration molecular dynamics simulations is described.