A quadratically convergent reference state optimization procedure

A quadratically convergent procedure for the optimization of multiconfigurational reference states is presented and analyzed. The optimal reference state is determined from the energy variation principle by means of a sequence of unitary transformations of the form exp(i$\text{Λ}$exp(i$\text{k}$, where the hermitean operators $\text{Λ}$ and $\text{k}$ generate transformations of orbitals and correlation coefficients, respectively. Sample calculations are carried out within the Pariser—Parr—Pople model using a correlated reference state suggested by Clementi and Veillard. For the systems considered so far, convergence is achieved after 2–4 transformations.