Geometry optimization using local-density functional methods: Numerical aspects

The need to perform a numerical integration of the exchange-correlation functional because of its non-analyticity severely complicates the accurate application of local-density functional methods to molecules and clusters. The optimal choice of grid points for this integration and the estimation of the error made by the choice are subtle considerations. In particular, because the position and/or weighting of each grid point must change when the nuclear positions change, these errors are most noticeable when different geometries are compared. We have determined a method of grid point selection and weighting that reduces these errors. We have also determined a simple method of estimating the extent of the error made in the particular density of points used for the grid. These results are illustrated for a selection of small molecules.