Intermolecular potential energy surfaces are usually complicated, multidimensional functions of the coordinates of the interacting molecules. They are used widely in different areas of science, but often in an oversimplified form. The major obstacle to the adoption of more realistic potential energy surfaces is often the problem of fitting or interpolating calculated energies efficiently and accurately, rather than calculating the energies themselves.
This talk will describe recent progress made with Gaussian Process interpolation of intermolecular potential energies. It will include a critical overview of the design of the problem, and of the interpolation and fitting methods, and will show how high-quality predictions can often be made using only a few calculated energies.