Can I safely replace the calculation of SCAN+rVV1O with R2SCAN+rVV10?

[chenggoj]

Dear Administrator,

I have been utilizing the meta-GGA functional (SCAN) + rVV10 in VASP 5.4.4 for simulating complex materials. During these simulations, I frequently encounter convergence issues where the energy and force oscillate near their extremal points without truly converging. I've noticed that the R2SCAN functional offers significant improvements in terms of numerical stability and efficiency, while maintaining a comparable accuracy to SCAN. Consequently, I have integrated the patch to support R2SCAN and recompiled VASP. With this, I am now equipped to use R2SCAN+rVV1O for material simulations.

My primary concern is the transition: can I seamlessly shift from SCAN+rVV10 to R2SCAN+rVV10? Specifically, if I employ R2SCAN+rVV10 for future calculations, can their outcomes be directly compared with the results from previous SCAN-based calculations? Furthermore, is there a necessity to redo the vast amount of data I've computed using R2SCAN+rVV10? Undertaking such a task seems daunting and perhaps impractical for me.

Thanks for your time and look forward to your reply.

Best regards,

Chenggong

[andreas.singraber]

Dear Chenggong,

In principle, R2SCAN was constructed in such way that its results are (supposedly) close to the original SCAN. However, there can still be some non-negligible differences depending on the system you study. Therefore, it is impossible for us to give you a general answer about how comparable the results will be. For sure, you should never compare directly total energies but only physical properties like geometry or binding energy. You will have to carry out your own tests for your systems and establish a well-founded argument if you want to combine data sets for SCAN and R2SCAN. In a publication you should make very clear when you chose which functional.

Please have a look at this publication which provides a refit resulting in an updated value for the b parameter (INCAR tag BPARAM) for the R2SCAN+rVV10 combination (see also here: wiki/index.php/Nonlocal_vdW-DF_functionals).

All the best,
Andreas Singraber and Fabien Tran

[chenggoj]

Hi Andreas Singraber and Fabien Tran,

Thanks a lot for your valuable reply. It is very helpful. Yes, I mainly compare some properties of materials, such as binding energies and some electronic structure information. I will select some cases and do some tests to compare the differences between the two functionals. However, in principle, should I expect that R2SCAN+rVV10 and SCAN+rVV10 should produce results with very small relative errors?

In addition, do you have any suggestions for improving the convergence issue of SCAN+rVV10? I've tried the following before:
(1) Near the extreme point (usually dE<-E-03) I change IBRION=2 to IBRION=1, with smaller POTIM (usually 0.01~0.1).
(2) Change the calculation method of ALGO. Change from the recommended ALGO=A to ALGO=Fast, with LMAXMIX = 4 (for d-group elements) and LMIXTAU = T, or vice versa.
In some cases they did work, but in many cases, they didn't. I don't know if I'm doing the right things, do you have any better suggestions?

Thanks for your time and look forward to your reply.

Chenggong

[fabien_tran1]

However, in principle, should I expect that R2SCAN+rVV10 and SCAN+rVV10 should produce results with very small relative errors?

We have not done ourselves any comparison between SCAN+rVV10 and R2SCAN+rVV10, therefore we don't really know more than what has been reported in the literature. Either search for such comparison in the literature or do calculations yourself, which is possible since both functionals are available in your VASP version.

[chenggoj]

Thanks a lot.