Efficient Parallel Implementation of a Full Configuration Interaction Algorithm for Circular Polyenes On a Cray Y-Mp

Abstract

An efficient implementation on the CRAY Y-MP of a new full configuration interaction algorithm is illustrated on a model problem. This shows how a careful optimization can lead to huge speedups in CPU times and interesting speedups in the overall elapsed times. The implementation takes full advantage of the algorithm's characteristics, with a full vectorization of the heaviest innermost loops and parallelization of the immediately outer loops.

The problem of interoperability: A common data format for quantum chemistry codes

“…FullCI [3,11] is a program designed to compute energies and eigenvectors at a FCI level, together with first-and second-order properties (both static and dynamic at real or imaginary frequencies) and transition moments. It requires precomputed integrals in molecular orbital base, and so it has to be interfaced with a zero-level program.…”

“…▪ VEGA: a FCI [3][4][5][6][7][8][9][10][11] package, with calculation of energy and first-and second-order properties; ▪ COLUMBUS [12] package: a general-purpose ab initio chain (SCF, CAS-SCF, CI); ▪ DALTON [13] package: a general-purpose ab initio chain;…”

Full configuration interaction study of the ground state of closed‐shell cyclic PPP polyenes

“…In the case N = 18, the dimension of the FCI vector is quite large: If we take into account, as we did, the spatial symmetry associated with the CIS cyclic group and the spin-reversal symmetry [32], we end up with a vector of about 73 million components. For such a vector, the iterative procedure is very time-consuming, even using an optimized code on a CRAY [33], and we could not obtain the FCI solution with a precision comparable to that obtained for smaller cases. For the same reason, the range of small IpI values (IpI < 2.5) could not be explored.…”