Extended Geminal Models
The extended geminal model is a particular way of constructing a full
configuration interaction (FCI) wavefunction. Contrary to conventional
approaches where the "building block" is the orbital, i.e. an one-electron
function, the "building block" in extended geminal models is the geminal, i.e.
a two-electron function. For a closed shell 2N-electron system we have the
following ansatz for the wavefunction




In this equation

denotes an antisymmetric product of strongly orthogonal geminals (APSG),

is the APSG approximation of the (2N-2)-electron cluster generated by deleting
the geminal of electron pair K in
 ,

is a two-electron correction function,

are normalization constants and

is the antisymmetrizer. There are analogous definition of the higher order
terms in the equation. In the more compact representation,

represents a single pair correction term ( with respect to

),

a double pair correction term, and so on. The expansion

is identical to a FCI expansion.
The basic approximation, or root function,
 ,
has the form

where

is a normalization constant determined by the equation

The geminals, i.e. two-electron functions,
 ,
are products of an orbital part and a spin function which is singlet coupled,
i.e.

The spatial two-electron geminal can be expressed in terms of natural orbitals
(NOs)

If the number of NOs is greater than 1, i.e.
n ,
then the optimization of

leads to unique geminals. In most cases, the geminals

have a localized character. Hence, in a molecule, the geminals can be
associated with familiar chemical concepts like core electrons, bond pairs
and lone pairs. If
n ,
the optimization of the wavefunction no longer leads to uniquely defined
orbitals. Hovewer, by localizing the orbitals, one can obtain a similar
interpretation of the geminals.
The energy can formally be evaluated within the framework of the method of
moments. Since we by construction we have

it follows


where




In the last equations, H denotes the conventional spin-independent electronic
Hamiltonian, i.e.

The energy expression is in most applications truncated at the double pair
correction level.
|