Critical assessment of the performance of the semiempirical divide and conquer method for single point calculations and geometry optimizations of large chemical systems

We present a detailed analysis of the performance of the semiempirical divi de and conquer method as compared with standard semiempirical MO calculatio ns. The influence of different subsetting schemes involving dual buffer reg ions on the magnitude of the errors in energies and computational cost of t he calculations are discussed. In addition, the results of geometry optimiz ations on several protein systems (453 to 4088 atoms) driven by a quasi-New ton algorithm are also presented. These results indicate that the divide an d conquer approach gives reliable energies and gradients and suggest that p rotein geometry optimization using semiempirical methods can be routinely f easible using current computational resources. (C) 2000 American Institute of Physics. [S0021-9606(00)30146-5].