The use of semiempirical model parameters to reduce the cost of electronic structure calculations will be presented.  The approach embeds parameters in a low-level  method  (minimal basis Hartree Fock) and adjusts these to reproduce the results of a high-level method (double-zeta or larger basis Hartree Fock). First, a large set of ab initio data is generated for molecules in a range of geometries and electrostatic environments. Agreement between low-level and high-level methods is then sought for both the total energy and expectation value of individual operators. The parameters interpolate between integrals evaluated in a narrow and a diffuse basis, with parameters depending on both operator and atom type. The interpolation parameters may depend on charge and other features of the current state of the atoms, and machine learning techniques are used to select optimal functional forms. The ability to transfer parameters between different classes of molecules will be discussed.

Embedding empirical parameters in ab initio models to reduce the cost of electronic structure calculations

244th National Meeting (2012)

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