My Community

Dear all,

I have to complain about the GGA_PW91 potential.
From the manual, we know that LDA or GGA_PBE is recommended. The PW91
description is not strict, so-called"VASP bug".
However, GGA_PW91 potential, instead of PBE, is the default GGA potential!
Look at the title of the potential,
LDA: PAW
GGA_PW91: PAW_GGA
GGA_PBE: PAW_PBE

Recently, I have used bulk bcc Fe for testing the difference.
Comparing the magnetic energy (E(FM)-E(NM)),
For Ferromagnetism:
GGA_PW91(VOSKOWN = 0), the mangetic energy is much smaller.
GGA_PW91(VOSKOWN = 1) is consistent with GGA_PBE.
However for antiferromagnetism:
Even GGA_PW91(VOSKOWN = 1) is quite different to GGA_PBE. The difference is more than 15%.


So it seems that GGA_PW91 cannot give the good magnetic energy. Unfortunately, I have used the GGA_PW91 doing lots of magnetic calculation, including magnetic phase transition calculation.



Conclusion: use LDA or GGA_PBE, don't use the GGA_PW91.


Zhicheng Zhong

A few questions and remarks since this is a quite broad statement considered it is just concluded from one type of system:

1. I don't know if I would say that PW91 is the default xc-potential. My guess is that the denotation arises from that you first had LDA and PW91 in VASP and then later the PBE was added. I have never got the impression that the PW91 is the default by any means since you have to make an active choice when you copy the potentials from the directories.

2. When you calculate the magnetic energy, do you use the same lattice parameters for both the systems? or do you use the equilibrium values for each system? Are all magnetic energies converged w.r.t. cutoff, smearing, k-points for both PW91, PBE and for the different states (FM, NM, AFM)?

3. VOSKOWN = 1 is the default setting for GGA_PBE so that seems good that this is consistent with GGA_PW91 for the E(FM)-E(NM).

4. How do you define a good magnetic energy? What do you compare with in order to conclude that GGA_PW91 is not good? Experimental data?

5. How large of the difference arises from that the PW91 and PBE is not equal (even when the PW91 is implemented in a strict way)?

You might very well be correct in your conclusions for the Fe-PW91 in your specific situation, but if it's possible I would like to have more specific information.

Best regards,
/Dan Fors

<span class='smallblacktext'>[ Edited Tue Oct 28 2008, 08:01AM ]</span>

I would not call PW91 a "VASP bug", the implementation was done according to the literature reference, and, after all, gives very good results for most applications. You might want to check which potential you use before starting a calculation, and then use it according to the hints given in the manual. Also, please qantify your complaints (by providing your inputs and the data which you obtained.)