[Wien] Calculation of Ueff

[Gavin Abo]

The NiO example [1] with 1 k-point gave about 6 eV with the NiO_+.5-1 
calculation taking about 36 cycles [2,3].  N.B. [4], this is with 2.00 
and not 20.3 in case.in1.  I believe the 20.3 in the paper may be a 
typographical error. At the bottom of [1], there is the statement:

Usually, it is better to just set the d energy 2-4 Ry above the Fermi level.

The Fermi level is not known when starting the calculation, but a small 
value of about 0 is likely a reasonable assumption.  So let Fermi level 
= 0.  Taking the lower value of 2 Ry from the statement above.  So took 
2.00 for case.in1.  I think the Fermi level obtained after the 
calculation was about 0.5. If you round that down, it becomes zero.  So 
it seems the 0 was likely an okay starting assumption.

I suspect 1 k-point and 5 RKmax wouldn't be used if it was a proper 
calculation, but the paper is an example such that it was likely chosen 
for speed [5].

For example, it looks like 100 k-points and 5.7 RKmax (instead of 1 
k-point and 5 RKmax) were used in G. K. H. Madsen and P. Novak 
calculation for Fe3O4 [6].

Keep in mind that the Ueff values obtained from the Constrained method 
are known to be of questionable accuracy [7].

Rather than your two choices below, it is better to do convergence tests 
[8-12].

[1] http://susi.theochem.tuwien.ac.at/reg_user/textbooks/Constraint_U.pdf
[2] 
http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg14460.html
[3] 
http://wien2k-algerien1970.blogspot.com/2016/08/how-to-calculate-u-parameter-in-nickel.html
[4] https://en.wikipedia.org/wiki/Nota_bene
[5] 
http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg15593.html
[6] https://arxiv.org/abs/cond-mat/0412560v1
[7] 
http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg12429.html
[8] 
http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg15596.html
[9] 
http://wien2k-algerien1970.blogspot.com/2016/08/kpoint-convergence-test.html
[10] 
https://www.beilstein-journals.org/bjnano/content/supplementary/2190-4286-2-45-S1.pdf
[11] 
http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg01700.html
[12] 
http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg08186.html

On 3/27/2017 9:50 AM, karima Physique wrote:
> Thank you for your answer but i I noticed that Ueff is very sensitive 
> with the change of RKmax and Kpoint it is for this I evoked this 
> subject abd  for example
> For K = 1 and RKmax 5 i found Ueff = 3.6 eV
> But For K = 100 Points and RKmax 8.5 i found Ueff = 5.4 eV
>
> 5.4 and 3.6 are relatively far
>
> What I really ask is that the choice of one point is it mandatory? (as 
> is indicated in the paper of G. K. H. Madsen and P. Novak) or else I 
> have to force these parameters (a large number of Kpoint and rkmax> 7) 
> which of these two choices is good
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20170327/d816dd63/attachment.html>