Hamiltonian equation energy gradient calculations

Table 1 shows the accuracy of the divide-and-conquer approach for different Hamiltonian matrix cutoff distance, / h, and different buffer size, R, in self-consistent semiempirical MOPAC calculations of RP71955, a small protein molecule. Data are taken from the work of Lee et al. Each subsystem consists of a single amino acid. If we define an accuracy criterion as 5.0 x 10" kcal mol" per atom in energy calculations and 0.1 kcal mol" A" in gradient calculations, then the buffer size Rb should be no less than 6.0 A to meet this accuracy criterion. For the cutoff distance in density and Hamiltonian matrices, a value of 7.0 A is appropriate to save memory use and CPU time without loss of accuracy. For the inverse temperature in equation (32) we chose 39.3 eV", which corresponds to 298 K. 

---