Efficient potential energy surfaces from partially filled ab initio data over arbitrarily shaped regions

Potential energy surfaces (PES) constructed from partially filled ab initio data were investigated. The partially filled interpolated data, evaluated over arbitrarily shaped regions, were analyzed by reproducing the kernel Hilbert space (RKHS) method. Surfaces implemented on arbitarilly shaped partially filled data were found to be as good as the full grids data. Clustering of ab initio points was a feature of the PES to vary the density parameters. The process of ignoring the irrelevant regions of surface did not affect the accuracy of the system.