Portal de Periódicos da CAPES

Estimating full configuration interaction limits from a Monte Carlo selection of the expansion space

1995; American Institute of Physics; Volume: 103; Issue: 5 Linguagem: Inglês

10.1063/1.469756

1520-9032

James C. Greer,

Machine Learning in Materials Science

Full configuration interaction (FCI) calculations are useful as benchmarks for approximate techniques used in quantum chemistry: they are indeed the desired goal for all energy and wave function calculations in that they are the best solution to the Schrödinger equation within a finite basis Ansatz. Application of the method is limited due to the rapid increase in the number of configurations as the basis set size is increased. Many means have been applied to limit the number of terms in the expansion with the best known method being the singles and doubles expansion CI(SD). A Monte Carlo algorithm is proposed here whereby a CI expansion is allowed to expand by randomly including new terms which interact with those terms already present in the expansion. Solution of the variational problem is then performed for these randomly chosen configurations and a selection criterium for the resulting CI coefficients is applied. Repeated application of this method allows for estimates of the FCI energy. Calculations for the water molecule are performed to demonstrate the method.