Massively Parallel and GPU-accelerated Implementation of the Third-Order Cluster Perturbation Method for Excitation Energy Calculations
Ashleigh Barnes1 , Dmytro Bykov1 , Dmitry Lyakh1 , Pablo Baudin2,3 , Filip Pawłowski3,4 and Poul Jørgensen3
1National Center for Computational Sciences, Oak Ridge National Laboratory (ORNL), 1 Bethel Valley Rd, Oak Ridge, TN 37831, USA
2Laboratory of Computational Chemistry and Biochemistry, Institute of Chemical Sciences and Engineering, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland
3Department of Chemistry, Aarhus University, Langelandsgade 140, DK-8000 Aarhus C, Denmark
4Department of Chemistry and Biochemistry, Auburn University, Auburn, Alabama, 36849-5312, USA
The coupled cluster singles and doubles (CCSD) method has been shown to be a reliable means of calculating molecular energies and properties, including excitation energies [1,2]. However, the calculation of excitation energies requires solving the CCSD response eigenvalue equation via an iterative method which scales as N6, often making these calculations prohibitively expensive for larger molecular systems. Cluster perturbation (CP) theory [3] seeks to overcome this scaling by avoiding explicitly solving the doubles eigenvalue problems required to obtain the full spectrum of excitation energies. Instead, excitation energies from coupled cluster singles (CCS) calculations are perturbatively corrected one at a time in orders of the fluctuation potential to achieve CCSD quality. This series of corrections has been termed the CPS(D) series and has been applied out to the 6th order correction. Calculations have shown that the third-order correction, CPS(D-3), results in energies of CCSD quality in a fraction of the computational time. Here we present a massively parallel implementation of the CPS(D-3) method within the LSDalton program [4]. We find good agreement with previously reported small molecule benchmark studies of CCSD excitation energies. Additionally, we report excitation energies for various configurations of retinal calculated with a triple-ζ basis as well as efforts to implement efficient GPU-acceleration of expensive tensor contractions within the CPS(D-3) calculation.
ACKNOWLEDGMENTS: This research used resources of the Oak Ridge Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC05- 00OR22725.
[1] G. D. Purvis and R. J. Bartlett, J. Chem. Phys., 76, 1910 (1982).
[2] H. Koch, H. J. A. Jensen, P. Jørgensen, and T. Helgaker, J. Chem. Phys., 93, 3345 (1990).
[3] F. Pawłowski, J. Olsen, and P. Jørgensen, “Cluster perturbation theory: II. Excitation energies for a coupled cluster target state,” (2018) unpublished work.
[4] K. Aidas et al., WIREs Comput. Mol. Sci., 4, 269 (2014).