Developments in parallel electronic structure theory

A primary goal of parallel electronic structure theory is the exploitation of large-scale distributed memory platforms. This paper describes scalable distributed data algorithms that target the increasingly large aggregate memories available with such architectures. The design of distributed data algorithms is enhanced considerably by the availability of one-sided forms of communication that facilitate remote data access in a shared-memory style of environment. One example, a distributed-data Moller-Plesset second-order (MP2) gradient [1] algorithm, which achieves good scalability from 1 to 256 nodes of a Gray T3E, is described. The calculation of a numeric MP2 Hessian and a large-basis-set geometry optimization demonstrate the effectiveness of the approach in real applications.