For simplicity the special case of
is considered first. As well as being conceptually simpler it also produces computational savings which are useful, so this special case has been coded into the Q-CHEM program.
The first (and most drastic) approximation is to assume that the Bra and Ket are well separated, something equivalent to assuming that all Gaussians on a shell-pair are point charges.
Rij is generated from the length DB, which is the same for all elements of the contracted shell-pair and therefore only needs to be calculated once per (0)(m). Note that yj is the fraction that Qj is along DC. Computational time can be saved if the formula generates the reciprocal of Rij, avoiding a costly divide. Now, by simple vector addition:
where
therefore,
Using a binomial expansion and the Legendre polynomials Pk, the square root is replaced by an infinite series
with
The magnitude of the kth term is bounded by
because
and
.
Therefore the series will converge whenever the ratio of |DC| to |DB| is less than one. The coefficients a0k can also be generated recursively (providing easier computation):
The power of this group of formulae is that Rijs can be calculated for the entire (0)(m) using only one measurement that involves both shell-pairs (the distance DB). The other information required (the distance DC and the fractional lengths yj) is all within the www-theor shell-pair, and can therefore be constructed before pairing all shell-pairs, thus removing a large fraction of the O(N2) work, at the cost of a little extra O(N) work. The general case is now presented.