Predicting Raman Spectra Using Density Functional Theory

Abstract

Accurately computing molecular Raman spectra would enable rapid development of inexpensive and extensive Raman libraries. This is especially beneficial for chemicals that are regulated, toxic, or otherwise difficult to handle. Numerous quantum mechanical methods have been developed that enable computation of Raman spectra. Here, we study the B3LYP exchange correlation functional with various combinations of basis sets, polarization functions, and diffuse functions to determine which combination best computes the Raman spectra for explosive and nonexplosive molecules. In comparing spectra, three metrics were utilized: the root mean square error, the earth mover's distance, and the weighted cross-correlation average. The earth mover's distance and weighted cross-correlation metrics are shown to have significantly greater power at detecting spectral similarities and differences than the root mean square error. Across all methods and molecules examined, B3LYP/6-311++G(d,p) was found to provide the best match between measured and computed Raman spectra. Spectra generated at the B3LYP/6-311++G(d,p) level were found to be accurate enough to correctly identify each molecule out of a set of measured molecular spectra.