Monte Carlo configuration interaction

A procedure for solving quantum many-body problems is presented and is shown to have properties which make it well suited for parallel computer architectures. The underlying method is an application of the linear variational principle using many-body expansion functions and is known as the configuration interaction or superposition of configurations method. By repeatedly generating expansion vectors using a Monte Carlo technique for configuration generation, a sequential improvement in the variational energy can be achieved. By performing independent samples of the expansion space concurrently on different processors, the results may be combined after a variational calculation to form an improved expansion vector. This sequence of steps is repeated until a desired level of convergence in the wavefunctions or energies is achieved. Analysis of the method is given within a parallel environment: efficiency, scaling, and a two-tiered approach to parallelism with the algorithm are discussed. (C) 1998 Academic Press.