[Wien] LDA+U changed between the versions?

[mazin]

Unfortunately this conversation is branching out to a domain that is is 
interesting but does not belong to the mailing list. To keep within the 
computational issues, my experience, and this is not necessarily 
everybody's opinion, is that it is often not possible to decide without 
experimental insight whether you are close to a mixed valence state. I 
am not sure what Ce127 is, but if you can physically meaningfully treat 
Se 4f as an open core in this material, you should not have problems 
with convergence. Sometimes it helps to converge first LDA+so (even 
though the f-band will be cutting the Fermi), and then turn on a U, and 
sometimes when you have a physical idea of what state you actually want 
to reproduce, to create a vorb by hand and then run LDA+U without 
recalculating vorb (-orbc), and then run (-orb) to get the fully 
self-consistent solution.

Igor

On 8/3/2010 11:20 AM, Jian-Xin Zhu wrote:
> Igor,
>
> Thanks for the suggestion.
> For the compound I am faced with, Ce127, I treated 4f as localized
> states first by following the instruction given in
>
>
>   WIEN2k-FAQ: "open core" treatment of 4(5)f electrons
>
> and the procedure worked fine.
> But I also noticed the comment
> *"A better approach would be to use LDA+U, which is available in WIEN2k*."
> I then had to fight with the whole business of the convergence problem.
>
> As for Ce in the compound being mixed valence state or not, how can we
> decide a priori?
> The conventional LDA calculation always puts the Ce 4f close to the
> Fermi energy,
> yielding a fractional occupation.
> I appreciate if you can share more of your experience.
>
> You are right. The LDA+U to iron pnictides is unphysical because they
> are metallic.
> I simply used it to find out why the LDA+U does not work for Ce systems.
>
>
> Laurence,
>
> I just copied the two files you pointed from 10.1 into 9.2 SRC_mixer
> folder, and will report again what I can obtain.
>
>
> Best regards,
>
> Jianxin
>
>
>
>
>
>
>
>
>
> On Aug 3, 2010, at 8:52 AM, mazin wrote:
>
>> Laurence,
>>
>> Thanks for your comments. It is a good idea to download a newer
>> version anyway, but unfortunately my problem is of different sort. I
>> can converged to the machine accuracy both calculations, WIEN2k_07 and
>> WIEN2k_09, and the results are physically different. I have no problem
>> with convergence in either case, and can verify manually that both
>> calculations are fully converged. The problem is therefore unrelated
>> to a possible bug in the mixing scheme.
>>
>> Jianxin,
>> Ce in some compounds is in mixed valence state or close to that. In
>> those cases, you may have problems with convergence, and you should be
>> careful fighting these problems, because even if you manage to
>> converge the calculations, that may be a rather unphysical solution.
>> In CeTe3 Ce is not close to mixed valency so there is no problem with
>> convergence. Neither should there be any with pnictides, assuming you
>> are talking about U on Fe, except that LDA+U is completely unphysical
>> for these materials.
>>
>> Igor
>> --
>> **********************************************************************
>> Igor Mazin, NRL, code 6390, 4555 Overlook Ave SW, Washington, DC 20375
>> Phone: (202) 767-6990; Fax: (202) 404-7546; e-mail mazin at nrl.navy.mil
>> <mailto:mazin at nrl.navy.mil>
>> Home Page http://cst-www.nrl.navy.mil/users/mazin
>> **********************************************************************
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at <mailto:Wien at zeus.theochem.tuwien.ac.at>
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
>
>
>
>
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 
**********************************************************************
Igor Mazin, NRL, code 6390, 4555 Overlook Ave SW, Washington, DC 20375
Phone: (202) 767-6990; Fax: (202) 404-7546; e-mail mazin at nrl.navy.mil
Home Page http://cst-www.nrl.navy.mil/users/mazin
**********************************************************************