Three previously reported algorithms for the evaluation of the fourth-order triple excitation energy component in many-body perturbation theory have been compared. Their implementation on current Intel distributed memory parallel computers has been investigated. None of the algorithms, which were developed for shared memory computer architectures, performed well since they lead to prohibitive IO demands. A new algorithm suitable for distributed memory machines is suggested and its implementation on two Intel i860 supercomputers is described. A high level of parallelism is obtained.
- Fourth-order triple excitation energy
- Intel distributed memory parallel computers
- Many-body perturbation theory