Post date: Nov 22, 2017 2:44:18 PM
Instead of publish a short note somewhere else, this is a good place.
In my recent paper, Synergy of van der Waals and self-interaction corrections in transition metal monoxides, I brought the importance of van der Waals interactions into attention for transition metal monoxides, which had not been expected for a long time. The same importance was reported also for NaCl, and I will not be surprised for more systems, especially ionic systems, which are not regarded as conventional vdW systems.
The final conclusion of this paper is kind of self-evident: A remedy to a semilocal DFT with the known drawbacks of self-interaction errors and missing long-range vdW interactions, is apparently to add corresponding corrections to them. My paper provides a practical recipe, and demonstrates its generality with several combinations of the basis semilocal functional, self-interaction corrections, and vdW corrections. SCAN+rVV10+U turns out to be the best among these combinations and indeed works remarkably well, but I would rather put more emphasis on the self-evident however longly overlooked truth.
There are some discussion I prepared in my initial versions but was finally removed by John for safety or something else. There is a vague concept of "intermediate-range" vdW interaction, one can define it as the vdW interaction which can be roughly included by a semi-local functional. The long-range part has to be a beyond semi-local functional approach. Similarly, there can be a vague concept of "local" self-interaction correction, which I tend to define as that can be done via DFT+U. The local SIC will be mainly illustrated by the renormalization of the charge density and the band structure, while the "global" one mainly by opening the band gap with a rigid shift. Similar arguments have also been mentioned by Takao Kotani. Meta-GGAs such as SCAN, with the generalized Kohn-Sham Scheme implementation, is essentially a beyond semi-local functional. It includes about right intermediate-range vdW interaction as discussed in my SCAN+rVV10 paper, and includes some global SIC as illustrated by a systematical improvement of band gaps as shown in Zeng-Hui's paper. SCAN should also includes some "local" SIC, or less local self-interaction error, as illustrated by the systematically smaller Hubbard U parameters from linear response in this synergy paper. Hence SCAN is a better starting point than other GGAs and meta-GGAs, but all "semilocal" functionals equally need extra components.