|About this Abstract
||Materials Science & Technology 2019
||Late News Poster Session
||P1-92: Crystal Structure Prediction through Density Functional Theory Combined with Unsupervised Machine Learning: A Vitamin B2 Case Study
||Thiago Henrique da Silva, Matthew King
|On-Site Speaker (Planned)
||Thiago Henrique da Silva
First-principles density functional theory (DFT) calculations were performed to predict the unknown crystal structure of riboflavin, also known as vitamin B2. Assessment of the optimal structure of vitamin B2 was done in three steps: initial structure generation, crystal geometry optimizations using DFT and implementation of an unsupervised learning algorithm to select the fittest lowest-energy structures. Structural candidates were created by implementing changes in the rotational and translational elements of the molecule inside of the initial unit cell. The DFT self-consistent field method (SCF) was applied using the CRYSTAL14 software for all structures utilizing an atom-centered basis at the PBE/6-31G(d,p) level. Lowest-energy structures were selected through an unsupervised learning algorithm and the corresponding structures were then submitted to a full geometry optimization. Through this method a structural candidate was obtained that led to an accurate prediction of the crystal structure, verified experimentally by PXRD and low-frequency vibrational THz spectra.