Date of Award

Spring 1-1-2019

Document Type

Thesis

Degree Name

Master of Science (MS)

First Advisor

Stephen Becker

Second Advisor

William Kleiber

Third Advisor

Charles Musgrave

Abstract

Compared with experiments, molecular dynamics (MD) simulations provide a quick and inexpensive way to study the properties of chemical systems. In many cases, it is necessary to extract spectral data from these simulations, such as infrared or Raman spectra. For instance, to validate that the computational system matches a physical system, the spectral “fingerprints” can be examined. For complicated systems, Raman spectroscopy calculations are computationally expensive, providing an incentive to reduce the amount of data required. Currently, spectral estimation from MD simulations relies on the discrete Fourier transform (DFT); however, alternative methods can more precisely model the spectra using fewer data points. These methods are particularly effective when prior knowledge of the spectral shape is considered. Several methods, including the direct regression, Welch power estimation, the regularized resolvent transform (RRT), and a modified version of the filter diagonalization method (FDM) are compared to the DFT when applied to MD simulations of methanol and sodium chloride. We propose a novel modification of the FDM, including use of the LASSO (least absolute shrinkage and selection operator) to improve the method's accuracy. Moreover, `windowing' present in FDM is modified to produce a significantly more accurate spectrum. The performance of these methods is then compared with each other to determine which methods are prone to include incorrect spectral features or lack correct spectral features. In brief, the modified FDM and RRT far outperformed other methods: the modified FDM produces the lowest rate incorrect spectral peaks while the RRT produces the lowest rate of missing peaks.

Share

COinS