next up previous contents
Next: Theoretical Performance Analysis Up: Faster Integral Calculation Previous: Matrix Equations

General Algorithm

It is helpful to present the entire final algorithm. The six steps involved are:
1.
Construct moments. Loop over all the shell-pairs, generating moments up to and including ninth order, with a separate set for each type of scaling required for that type of shell-pair. Note that this part of the algorithm is only O(N) in cost.
2.
Convergence test. Loop over all www-theors for the present bra, finding for which shell-pairs the expansion will converge, and construct lists of those that pass and fail. This section (and those that follow) is O(N2).
3.
Find geometric parameters. Calculate the values of $\frac{AB \cdot DB}{DB \cdot DB}$, etc.
4.
Form ${\bf H}$ from the recurrence relation. This is performed once per m value of the class. There is a significant time saving if |AB|=0.
5.
Form ${\bf Hv}$. Required once per unique ${\bf Hv}$ product. Again there is a significant time saving if |AB|=0.
6.
Generate all (0)(m)s, until a new ${\bf Hv}$ is required.



Ross D. Adamson
1999-01-27