Devices made out of organic polymers are attracting significant interest in the optical and optoelectronic applications due to their flexibility, better mechanical stability on impact, light-weight, and lower manufacturing cost. However, harnessing their potential in these industries has been quite limited because these applications require materials to have optical properties comparable to conventional metal oxides. For example, carbon-based polymers inherently have a refractive index (RI) between 1.3 and 1.5, while most of the applications require RI greater than 1.7. The RI value of polymers can be improved upon by controlling the chemical structure which allows for precise tuning of optical properties. Using our in-house molecular generator, we generated a molecular library of 1.51 million potential organic structures to identify the most promising candidates with high RI values. We recently developed an accurate RI prediction model for organic polymers based on a synergistic combination of first-principles calculations and molecular dynamics (MD) simulations. However, performing first-principles calculations and MD simulations for 1.51 million molecules is not a viable option. Therefore, we evaluated the density and polarizability values of a subset, 100 thousand molecules, from the library. With this data as a training set, we developed a deep neural network model to correlate the molecular structure with the properties. The resultant correlation is shown to be exceedingly efficient with an R2 value of 0.977. Using this model, we were able to quickly compute the RI values of 1.51 million organic molecules. The developed machine learning approach is a powerful tool and has shown to be highly promising for rapidly identifying molecular candidates with exceptional RI values.

Machine learning approach for fast and accurate prediction of optical properties of organic molecules

254th National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.