Scalable Electron Correlation Methods. 2. Parallel PNO-LMP2-F12 with Near Linear Scaling in the Molecular Size

We present an efficient explicitly correlated pair natural orbital local second-order Moller-Plesset perturbation theory (PNO-LMP2-F12) method. The method is an extension of our previously reported PNO-LMP2 approach [Werner et al. J. Chem. Theory Comput. 2015, 11, 484]. Near linear scaling with the size of molecule is achieved by using domain approximations on both virtual and occupied orbitals, local density fitting (DF), and local resolution of the identity (RI), and by exploiting the sparsity of the local molecular orbitals (LMOs) as well as of projected atomic orbitals (PAOs). All large data structures used in the method are stored in distributed memory using Global Arrays (GAs) to achieve near inverse-linear scaling with the number of processing cores, provided that the GAs can be efficiently and independently accessed from all cores. The effect of the various domain approximations is tested for a wide range of chemical reactions. The PNO-LMP2-F12 reaction energies deviate from the canonical DF-MP2-F12 results by <= 1 kJ mol(-1) using triple-zeta (VTZ-F12) basis sets and are close to the complete basis set limits. PNO-LMP2-F12 calculations on molecules of chemical interest involving a few thousand basis functions can be performed within an hour or less using a few nodes on a small computer cluster.