A STATISTICAL TREATMENT OF LARGE CONFIGURATION-INTERACTION EIGENVECTORS - TESTS ON MODEL HAMILTONIANS AND ON CLASSICAL MO-CI CALCULATIONS

A statistical treatment for very large configuration interaction (CI) calculations is presented. The energy is written as a sum of elementar y contributions, one per determinant, and the statistical choice is ma de among these elementary contributions. Two types of practical tests determine what conditions must be verified to get reliable results. It is verified that, if N is the dimension of the CI problem, the size o f the statistical samples must grow as square-root N to keep the same accuracy in the results.