(193x) Improved Nonlinear Models for the Refractive Index (RI) of Organic and Inorganic Materials

The identification of optimal values of thermophysical, mechanical and biological properties is a prime objective in the development of new materials for chemical industries. Computer-assisted molecular design (CAMD) techniques are being employed increasingly to develop novel molecular structures possessing desired properties. The CAMD algorithms help expedite the design process by predicting the behavior of promising molecules through the use of reliable property models. In the material sciences, refractive index is a widely-used property for the evaluation of both novel and targeted optical materials.

 In this work, high quality non-linear molecular models have been developed for the prediction of the refractive index. These predictive models include (a) an improved computational algorithm, which incorporates binary particle swarm optimization and support vector regression and (b) quantitative structure-property relationship models employing wrapper-based evolutionary algorithms and artificial neural networks. Data from the DIPPR database were used to develop the models, which were capable of prediction within the experimental error of the data. The root-mean squared error for the multivariate and support vector regression based models were 0.0314 and 0.0311, respectively.