Solid Calculations with Meta-GGA Accuracy at Little more than GGA Cost*

The recent revision [1] of the strongly constrained and appropriately normed (SCAN) meta-generalized gradient approximation (GGA) exchange-correlation functional into r²SCAN reduces the numerical instabilities and integration grid sensitivities exhibited by SCAN very substantially. Similar numerical difficulties plus high iteration counts impaired the utility of the otherwise successful deorbitalization of SCAN to SCAN-L, SCAN with density Laplacian dependence [2]. The suspected cause was the Laplacian dependence.
However, use of the identical de-orbitalization to generate r²SCAN-L [3] from r²SCAN is successful both in preserving the accuracy of r²SCAN and in reducing, drastically, the SCAN-L technical problems. The result is much faster calculations. On a set of 660 solid calculations with VASP, r²SCAN-L needed only 20% more time than PBE and only 30% of the time needed for r²SCAN. We provide validation data and discuss the remaining Laplacian-dependence issues, including the use of r²SCAN-L in orbital-free DFT calculations.

[1] J. Phys. Chem. Lett. 11, 8208 (2020).
[2] Phys. Rev. B 98, 115161 (2018)
[3] Phys. Rev. B 102, 121109(R) (2020).