A new algorithm for the numerical evaluation of gradients in semiempirical
methods is described. The method is approximately twice as fast as the sche
mes currently employed and produces gradients of comparable accuracy. This
method has been tested by comparing the results obtained by the new method
with those of the previous numerical scheme, and also with those calculated
analytically. The results of using the new gradients in geometry optimizat
ions are also presented. (C) 1999 Kohn Wiley & Sons, Inc.