OPTIMIZED TWO-ELECTRON-INTEGRAL TRANSFORMATION PROCEDURES FOR VECTOR-CONCURRENT COMPUTERR ARCHITECTURES.

Various approaches for the transformation of two-electron integrals are discussed. Using the full index symmetry (FIS) of the array to be transformed for a case with N basis functions, the N-to-N transformation yields an operation count of (25/24)N**5, while transformation to a small set with M orbitals yields (1/8)N**4M if N VM GT TH M. We discuss the vector-optimized implementation of this (FIS) procedure for concurrent architectures. Tests were carried out on an Alliant FX8/8 mini-supercomputer (eight processors sharing one memory) and compared with the performance of DEC VAX scalar machines.