We present a numerical methodology for the construction of molecular potential energy surfaces. This method is based upon a cluster-weighted modeling scheme that uses an expectation/maximization algorithm to represent an ab initio potential energy data set in terms of adaptively parameterized Gaussian clusters and local basis functions. The methodology is illustrated through benchmark studies for simple 1-D potential curves and 2-D/3-D surfaces for triatomic systems. The computational benefits and caveats of the approach are further discussed in the context of more complex molecular systems with larger dimensionality.

Cluster-weighted modeling approach to potential energy surface fitting

251st National Meeting (2016)

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