We develop and validate the XYGJ-OS functional, based on the adiabatic connection formalism and Görling-Levy perturbation theory to second order and using the opposite-spin (OS) ansatz combined with locality of electron correlation. XYGJ-OS with local implementation scales as N 3 with an overall accuracy of 1.28 kcal∕ mol for thermochemistry, bond dissociation energies, reaction barrier heights, and nonbonded interactions, comparable to that of 1.06 kcal∕mol for the accurate coupled-cluster based G3 method (scales as N 7 ) and much better than many popular density functional theory methods: B3LYP (4.98), PBE0 (4.36), and PBE (12.10).btaining chemical accuracy (∼1 kcal∕mol) to quantify key chemical quantities (e.g., heats of formation, bond dissociation energies, and reaction barrier heights) using quantum mechanics (QM) has been a major focus in the development of the theory. This has led to, for example, the Gn method (1, 2) that approaches this chemical accuracy. Because G3 is a coupled-cluster based method, it scales on the order of N 7 , where N measures the system size, limiting to fairly small species for routine use.The desire to predict unique physiochemical phenomena (e.g., solvation, catalysis, self-assembly, and drug design) in practical (large) systems has brought about a second major focus of theoretical development, leading, for example, to divide-and-conquer formulations to attain more efficient scaling (3), but with much lower accuracy than Gn.Density functional theory (DFT) in the framework of KohnSham (KS) scheme (4, 5) provides a "shortcut" to the many-body problem. Many density functional approximations (DFAs), provide typical scaling of N 3 ∼ N 4 , while yielding significantly more accurate results than Hartree-Fock (HF) theory, the lowest level wave function based method with similar scaling, but they still lead to significant errors for some systems. For example, current DFAs lead to a poor description of London dispersion (van der Waals attraction), which is essential to predict the packing of molecules into solids, and the binding of drug molecules to proteins. These DFAs are also poor in predicting the magnitude of reaction barriers.In this article, we develop and present a unique functional, XYGJ-OS, that provides a good combination of high accuracy and speed. XYGJ-OS involves a doubly hybrid density functional (DHDF), containing both a nonlocal orbital-dependent component in the exchange term (HF-like exchange), and also information about the unoccupied KS orbitals in the electron correlation part (PT2, perturbation theory up to second order) using the opposite-spin (OS) ansatz to include the locality of electron correlation. XYGJ-OS provides accuracy comparable to that of Gn for the test datasets and speed with N 2 ∼ N 3 for the local implementation. Hence XYGJ-OS is both accurate and fast.
TheoryThe Holy Grail in KS-DFT is to find the exact, yet unknown, exchange-correlation functional E xc ½ρ using density ρ as the basic variable (4, 5). In practice, an approximate E xc must be ado...