The meta-GGA SCAN functional has been there for a while. Despite the lack of SCAN (ultrasoft/PAW) pseudopotentials, various works used it in plane-wave pseudopotential calculations with pseudopotentials generated using PBE (in particular, using VASP, such as this work and this work), and reported improvements compared to GGA.

I am wondering how much impact does the use of inconsistent pseudopotentials have, and whether the results are still reliable?

PS: I am aware that rSCAN has enabled the pseudopotential generations, and there is a recent paper proposed the r2SCAN function aiming to improve the accuracy of rSCAN. So hopefully SCAN pseudopotentials will arrive soon.


It is well known that using pseudopotentials generated with a functional which is inconsistent with the functional used in the plane-wave calculation can lead to serious problems, see here for example.

It was also shown that for a potential-only meta-GGA (TB), the choice of pseudopotential makes a huge difference in the band gap, and in particular using PBE pseudopotentials lead to severely underestimated band gaps compared to the all-electron calculations with the TB potential.

Another study showed that PBE pseudopotentials do lead to inaccuracies when combined with SCAN in the plane-wave calculation, although it is somewhat less severe overall - but keep in mind that this study was not too comprehensive.

I would say that as a minimum you must state that this is what you did and indicate that this is an uncontrolled approximation and could be a cause of a range of errors. You may be able to get an idea about the error due to the inconsistency, for your particular system, by comparing to all-electron results, even as a spot-check or for a proxy system.

The fact that a lot of studies are conducted with inconsistent pseudopotentials does not mean that such usage is ideal or even correct.

  • 1
    $\begingroup$ +1. Welcome to our community, and thank you for contributing your excellent answer here! We hope to see much more of you !!! $\endgroup$ – Nike Dattani Sep 21 '20 at 16:14

New functionals often use PBE as a base. This is due to the benefit that everyone is using the same slightly incorrect psuedopotential, meaning it is reproducible. In VASP the psuedopotentials cannot be generated by the user but codes like CASTEP and GPAW can generate them on the fly as needed.

Using PBE as a base for SCAN is not ideal, but its consistent and easy. I don't think anyone would consider it incorrect to do since this is common practice. Your concern is valid though

  • $\begingroup$ I think most people do consider it incorrect to do this, but accept that sometimes it is necessary in practice. SCAN is ill-conditioned in vacuum which makes it difficult to generate pseudopotentials (hence rSCAN), but in principle it can be done. Hybrid functionals are much trickier to deal with. $\endgroup$ – Phil Hasnip Nov 13 '20 at 11:23
  • $\begingroup$ Yeah maybe this should be edited, common practice does not mean correct. VASP is the biggest offender of this mentality towards psuedopotentials. $\endgroup$ – Tristan Maxson Nov 13 '20 at 13:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.