AN EFFICIENT EVALUATION OF ATOMIC PROPERTIES USING A VECTORIZED NUMERICAL-INTEGRATION WITH DYNAMIC THRESHOLDING

A new efficient algorithm for the evaluation of atomic properties by means of numerical integration over atomic basins is described. The fully vectorized algorithm takes advantage of a dynamic thresholding technique which neglects the primitive Cartesian Gaussian functions that do not contribute significantly to the properties of a given atom. This results in a procedure with a computational effort that scales only linearly with the size of the molecule in question, as opposed to the quadratic size dependence of the traditional approach. The ensuing savings in CPU time, which vary between 90% and 99%, make the determination of the properties of atoms in such molecules as C60 and C60H60 routinely feasible. In the latter molecule, it is found that on average only almost-equal-to 8% of the primitives survive the thresholding algorithm capable of reproducing the exact properties within 10(-6) au.