Global optimization of SixHy at the ab initio level via an iteratively parametrized semiempirical method

Previously we searched for the ab initio global minima of several SixHy clusters by a genetic algorithm in which we used the AMI semiempirical method to facilitate a rapid energy calculation for the many different cluster geometries explored. However, we found that the AM1 energy ranking significantly differs from the ab initio energy ranking. To better guarantee locating the ab initio global minimum while retaining the efficiency of the AM1 method, we present an improved iterative global optimization strategy. The method involves two separate genetic algorithms that are invoked consecutively. One is the cluster genetic algorithm (CGA), mentioned above, to find the semiempirical SixHy cluster global minimum. A second and separate parametrization genetic algorithm (PGA) is used to reparametrize the AM1 method using some of the ab initio data generated from the CGA to form a training set of different reference clusters but with fixed SixHy stoichiometry. The cluster global optimization search (CGA) and the semiempirical parametrization (PGA) steps are performed iteratively until the semiempirical GA reparametrized AM1 (GAM1) calculations give low-energy optimized structures that are consistent with the globally optimized ab initio structure. We illustrate the new global optimization strategy by attempting to find the ab initio global minima for the Si6H2 and Si6H6 clusters. (C) 2003 Wiley Periodicals, Inc.