Next: The CASE Approximation
Up: Faster Integral Calculation
Previous: Empirical Performance Analysis
This chapter presents a new path for the construction of (0)(m)s. By performing contraction earlier than was previously possible, large savings in time can be made for highly contracted integrals. This has the effect of moving previously O(N2) work dependent on K into an O(N) part of the code. The algorithm also uses the BLAS matrix algebra, adding to the efficiency of the implementation. The CO path has been implemented in the Q-CHEM program and shows a three-fold speed up for only moderately contracted (ss|ss) integrals. However, the new algorithm shows a steeper increase in cost with momentum than the more traditional methods, and is seldom faster when the total momentum is greater than four.
Ross D. Adamson
1999-01-27