While computational methods of studying molecular systems have continued to advance over the past years, obtaining a good balance between computational effi- ciency and accuracy in estimations of electronic interactions remains a challenge.
As methods improve in their ability to accurately represent electronic interactions, their rise in computational cost typically limits their use to small systems. Consequently, chemical accuracy1 remains out of reach for condensed systems.
The balance between accuracy and efficiency can be improved for specific systems through the parameterization of efficient force fields and models derived from first principles (also referred to as ab initio models). Once a dataset of properties has been calculated using a high precision method for small systems, more efficient models can be optimized to the dataset by adjusting a set of parameters. Unfortunately, the process of parameterization of a complex model is complicated and involves choosing largely arbitrary functional forms that depend on many parameters, followed by a lengthy and difficult trial and error optimization process. The balance between the number of parameters and the size of the fitted data sets involves decisions that are difficult and subjective, yet critical for obtaining the best result. In this thesis, a general and powerful optimization scheme that provides an efficient way to obtain high-quality parameters, termed "Data Projection onto Parameter Space" (DPPS), is presented.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados