Quadratically convergent algorithm for orbital optimization in the orbital-optimized coupled-cluster doubles method and in orbital-optimized second-order Moller-Plesset perturbation theory


Creative Commons License

Bozkaya U., Turney J. M. , Yamaguchi Y., Schaefer H. F. , Sherrill C. D.

JOURNAL OF CHEMICAL PHYSICS, cilt.135, sa.10, 2011 (SCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 135 Konu: 10
  • Basım Tarihi: 2011
  • Doi Numarası: 10.1063/1.3631129
  • Dergi Adı: JOURNAL OF CHEMICAL PHYSICS

Özet

Using a Lagrangian-based approach, we present a more elegant derivation of the equations necessary for the variational optimization of the molecular orbitals (MOs) for the coupled-cluster doubles (CCD) method and second-order Moller-Plesset perturbation theory (MP2). These orbital-optimized theories are referred to as OO-CCD and OO-MP2 (or simply "OD" and "OMP2" for short), respectively. We also present an improved algorithm for orbital optimization in these methods. Explicit equations for response density matrices, the MO gradient, and the MO Hessian are reported both in spin-orbital and closed-shell spin-adapted forms. The Newton-Raphson algorithm is used for the optimization procedure using the MO gradient and Hessian. Further, orbital stability analyses are also carried out at correlated levels. The OD and OMP2 approaches are compared with the standard MP2, CCD, CCSD, and CCSD(T) methods. All these methods are applied to H(2)O, three diatomics, and the O(4)(+) molecule. Results demonstrate that the CCSD and OD methods give nearly identical results for H(2)O and diatomics; however, in symmetry-breaking problems as exemplified by O(4)(+), the OD method provides better results for vibrational frequencies. The OD method has further advantages over CCSD: its analytic gradients are easier to compute since there is no need to solve the coupled-perturbed equations for the orbital response, the computation of one-electron properties are easier because there is no response contribution to the particle density matrices, the variational optimized orbitals can be readily extended to allow inactive orbitals, it avoids spurious second-order poles in its response function, and its transition dipole moments are gauge invariant. The OMP2 has these same advantages over canonical MP2, making it promising for excited state properties via linear response theory. The quadratically convergent orbital-optimization procedure converges quickly for OMP2, and provides molecular properties that are somewhat different than those of MP2 for most of the test cases considered (although they are similar for H(2)O). Bond lengths are somewhat longer, and vibrational frequencies somewhat smaller, for OMP2 compared to MP2. In the difficult case of O(4)(+), results for several vibrational frequencies are significantly improved in going from MP2 to OMP2. (C) 2011 American Institute of Physics. [doi:10.1063/1.3631129]