|Is Bigger Really Better? Coupled-Cluster Theory for Larger Molecules|
|Rodney J. Bartlett, Andrew Taube, Tom Hughes, Tomasz Kus, Victor Lotrich, Norbert Flocke|
Quantum Theory Project
University of Florida
Gainesville, FL, USA, 32611-5500
|Today, single reference coupled-cluster theory typically provides the benchmark results for molecules with several atoms. This includes ground state energies, structures, properties, and transition states; and with its EOM-CC extensions, those for excited states. These results are also achieved within a virtually ‘black-box’ form, as CC theory has few choices except basis set, level of correlation, and the mean field reference function, which could be Hartree-Fock, Brueckner, Kohn-Sham, Natural Determinant, or something else. The next frontier for coupled-cluster theory is to make such well characterized and calibrated methods applicable to larger molecules. After all, the rationale for CC theory is its size-extensivity, which enables quantum chemistry to give meaningful results all the way to infinite systems. However, achieving meaningful large molecule CC results depends upon a variety of complementary strategies. |
The simplest is to reduce the virtual space in large CC calculations by using frozen natural orbitals (FNO). Without any meaningful error, such a transformation will save over an order of magnitude in CCSD, CCSD(T), and ΛCCSD(T) calculations, and ~ 2 orders of magnitude in CCSDT. However, to apply this tool effectively requires the development of analytical gradient techniques which makes studies of potential energy surfaces with several minima and transition states amendable to high-level study. An illustration for the unimolecular dissociation of nitroethane illustrates the method.
The second is to develop parallel CC programs that can scale over hundreds and thousands of processors. The newly written ACES III is such a program that shows exceptional performance and functionality, with the same ease of ‘black-box’ application. All memory handling and message passing are achieved with a new super-instruction assembler language (SIAL) separating those details from the quantum chemistry being programmed. Illustrations for some molecules of biochemical interest will be presented.
Finally, our natural linear scaling CC method will be described as a way to provide highly correlated results for very large molecules, from local units. Many variants of the same basic idea have had some success, but our approach differs in some essential elements, and is directed at capturing the ultimate transferability we see in chemistry. We consider transferable energies, densities, and response properties.