Global optimization of passivated silicon clusters at the ab initio level via semiempirical methods

New strategies based on the evolutionary theory for globally optimizing binary molecular systems have been developed. We first determined the global minima of several SixHy clusters using a genetic algorithm (GA) coupled with the fast AM1 semiempirical method to evaluate the cluster energies. However, we found that the AM1 and the ab initio energy rankings differ significantly. Therefore we proposed an improved iterative global optimization strategy which involved two separate genetic algorithms. The first algorithm is the cluster GA (CGA) that finds the SixHy cluster global minimum. The second algorithm is the parametrization GA (PGA) that reparametrizes the AM1 method against ab initio data to produce the GA-optimized AM1 (GAM1) parameters. Convergence is considered achieved when the CGA stops producing new low energy clusters. However this method is very computationally demanding. Therefore we further examined whether the GAM1 parameters obtained for a small SixHy stoichiometry could be successfully transferred to larger clusters. We have found that the GAM1 parameters derived from the Si7H14 training set can be transferred to larger Si clusters such as Si14H 20 with a similar level of H passivation. Unfortunately we still could not locate the diamond-lattice like Si14H20 global minimum since our original choice of genetic operators produce drastic structural changes which makes production of the lowest energy offspring difficult. Therefore we developed new genetic operators such as SiH2 insertion/removal and H-shift operators to enhance the efficiency at finding low energy clusters. The CGA using the new genetic operators enabled us to locate the Si14 H20 and other global minima that we had missed before. We found that fully passivated SixH y clusters prefer diamond-lattice like structures while slightly under-passivated SixHy -2 global minimum clusters adopt significantly different Si core. We also developed a GAM1* energy evaluation scheme to evaluate the relative energies for the SixFy clusters, which enabled us to search the SixF y global minimum efficiently. Surprisingly the Si xFy clusters and Si xHy clusters prefer completely different Six core structures. In summary, we have successfully determined the global minima of various SixH y and SixFy clusters, which prepares the ground for further investigation of the optoelectronic properties of passivated Si clusters.