Jeff Hammond, University of Chicago
The proliferation of parallel computers with thousands of processors has the potential to open up totally new areas of computational chemistry. We report on the marriage of high-level coupled-cluster methods with state-of-the-art supercomputing hardware. The computational chemistry software suite NWChem has been ported to the BlueGene/L and Cray XT4 platforms, allowing the ab initio study of molecules of an unprecedented size. We focus on coupled-cluster properties, including polarizabilities and hyperpolarizabilities for large conjugated molecules at the CCSD level of theory, as well as benchmark CCSDT response property calculations on small molecules. In addition, we compare the performance of bottleneck algorithms on a number of supercomputing platforms. Computational benchmarks suggest new parallel strategies and allow us to evaluate the utility of various types of supercomputing hardware.
Abstract Author(s): Jeff R. Hammond and Karol Kowalski