Potential parameterization from proxy systems

We give an analysis and critique of a procedure for sequential multi-scale parameterization of so-called ab initio potentials. Such potentials are based upon a small selection of reference data generated at relatively low levels refinement of quantum mechanical (QM) methodology. Two closely related potentials for silica, BKS and TTAM, are explored in some depth because both have found frequent use in the literature of molecular dynamics simulation. The use of a single small molecular cluster as a proxy for the behavior of larger silica structures is shown to be perilous. Choices of ionic charge, basis set size, inclusion of electron correlation at some prescribed level of approximation, and parameterization data are each shown to have a substantial impact on the final potential parameterization. The analysis and critique provide the logical foundation for the different approach to the problem of parameterization and use of classical potentials that is treated in detail elsewhere in this collection. The use of extended model systems, force data calculated from QM methods, and a focused (limited) range of application are shown to be essential for a consistent (and ultimately, predictive) embedding approach to concurrent multi-scale materials simulation.