Facilities and practices for linear response Hubbard parameters U and J in Abinit

Abstract: Members of the DFT+U family of functionals are increasingly prevalent methods of addressing errors intrinsic to (semi-) local exchange-correlation functionals at minimum computational cost, but require their parameters U and J to be calculated in situ for a given system of interest, simulation scheme, and runtime parameters. The SCF linear response approach offers ab initio acquisition of the U and has recently been extended to compute the J analogously, which measures localized errors related to exchange-like effects. We introduce a renovated post-processor, the lrUJ utility, together with this detailed best-practices guide, to enable users of the popular, open-source Abinit first-principles simulation suite to engage easily with in situ Hubbard parameters and streamline their incorporation into material simulations of interest. Features of this utility, which may also interest users and developers of other DFT codes, include $n$-degree polynomial regression, error analysis, Python plotting facilities, didactic documentation, and avenues for further developments. In this technical introduction and guide, we place particular emphasis on the intricacies and potential pitfalls introduced by the projector augmented wave (PAW) method, SCF mixing schemes, and non-linear response, several of which are translatable to DFT+U(+J) implementations in other packages.