Explicitly correlated local coupled‐cluster methods using pair natural orbitals

Abstract

Recently developed explicitly correlated local coupled-cluster methods [PNO-LCCSD(T)-F12] are reviewed. Extensive benchmarks for reaction energies and intermolecular interaction energies are presented, in which the convergence of the results with respect to all local approximations is studied. The explicit correlation treatment (F12) is shown to be essential to minimize basis set incompleteness errors, as well as errors caused by domain approximations. Generally, the errors of relative energies due to local approximations can be reduced to below 1 kcal/mol. The methods are well parallelized, and using small computer clusters with 100-200 computing cores, calculations for systems with 100-200 atoms using augmented triple-ζ basis sets can be carried out within a few hours of elapsed time. Recommendations are made on how such calculations should be carried out, how the accuracy can be tested, and which computational resources are required. This article is categorized under: Electronic Structure Theory> Ab Initio Electronic Structure Methods Software> Quantum Chemistry K E Y W O R D S explicit correlation, coupled cluster, local correlation, pair natural orbitals 1 | INTRODUCTIONThe coupled-cluster method with single and double excitations and a perturbative treatment of triple excitations [CCSD(T)] is considered the gold standard of quantum chemistry for computing energies and other properties of molecules. Often, the accuracy is comparable or even better than that of experimental data, and chemical accuracy (1 kcal/mol or better for relative energies) can be achieved, provided that the system under consideration is of single-reference character, that is, the Hartree-Fock (HF) Slater determinant dominates the total wave function expansion. However, in its traditional canonical form, the application of the CCSD(T) method is limited to rather small systems (20-30 atoms) due to the steep increase of the computational effort with system size: The central processing unit (CPU) time scales as O N 7 el

Reactive dynamics and spectroscopy of hydrogen transfer from neural network-based reactive potential energy surfaces

“…49 As such, TL is a valuable tool to bypass the high computational cost of modern electronic structure calculations: First, an NN is pre-trained on a large number of low level ab initio data and then re-trained with a smaller number of high level ab initio data to slightly adjust its parameters. 64 Here, the model trained on the MP2 data is the base model and the pair natural orbital (PNO-)-LCCSD(T)-F12/cc-pVTZ-F12 method 65,66 is the higher level of theory. A new data set at this level of theory is generated by extracting structures from MD simulations run with the NN trained on the MP2 reference data.…”

The performance of explicitly correlated methods for the computation of anharmonic vibrational frequencies

“…Explicitly correlated methods (F12) have promised greater accuracy for similar computational cost especially when paired with CCSD(T) implementations . The two‐electron terms incorporate more correlation by their very definition.…”

“…The tradeoff for the reduction of time, naturally, is typically in terms of accuracy.Explicitly correlated methods (F12) have promised greater accuracy for similar computational cost especially when paired with CCSD(T) implementations. [23][24][25][26][27] The two-electron terms incorporate more correlation by their very definition. Such has been proposed as a means of…”

Machine learning the computational cost of quantum chemistry

“…perf allows profiling of the kernel and user code at run time with little CPU overhead and can give FLOP counts with reasonable accuracy. FLOP count is an adequate measure of the computational cost when the program execution is CPU bound by numerical operations, which is given in the PNO-LCCSD(T)-F12 implementation [48][49][50]70 in Molpro.…”