Wiley eBooks
Iterative Coupled‐Cluster Methods on Graphics Processing Units
A. Eugene DePrince, Jeff R. Hammond, C. David Sherrill
Abstract
This chapter presents algorithms for iterative coupled-cluster calculations in heterogeneous computing environments. It first reviews strategies for implementing iterative coupled-cluster methods in heterogeneous computing environments to overcome the challenges of efficiently using graphics processing units (GPUs) for such calculations. The chapter then discusses more recent work that has emphasized the role of approximate tensor decompositions in the design of an efficient, production-level implementation of singles and doubles coupled cluster (CCSD). Next, the chapter describes an implementation of GPU-accelerated CCSD that makes use of density fitting (DF) or Cholesky decomposition (CD) approximations in the construction and contraction of all two-electron repulsion integrals (ERIs). These approximations reduce the storage requirements for the ERI tensor and hence mitigate the cost of data transfers to the device. Finally, the performance of the GPU-accelerated DF/CD-CCSD implementation for systems with as many as 822 active basis functions is demonstrated.