[Wien] lapwdm with ROOT option in case.in2

[Masao ARAI]

Dear Dr. P. Novak,

Thank you for your reply.
It seems that I failed to explain my problem clearly.

The non self-consistent calculation shows the same problem.

(1) I used converged charge density and case.dmatup/dn.

(2) I performed lapw0 and lapw1. They produced energies and eigenvalues. 
(They do
not depend on the choice of TETRA or ROOT.) The results are insulator.

(3) If the electronic structure is insulating, the results should not be 
sensitive whether
we use TETRA or ROOT. Indeed, The charge densities from lapw2 were 
almost similar.

(4) On the other hand, the lapwdm produced completely different density 
matrix whether
 we use TETRA or ROOT (from the same eigenvalues and eigenvectors). 
My humble analysis about this problem was summarized in my previous  e-mail.

I will make a simple test case.
Best regards,

--
NIMS
Masao Arai


> Dear Masao Arai,
>
> your result is in fact not too surprising. ROOT option simply fill the
> electron states until the number of electrons is exhausted. If for the
> last electron there are several degenerate states available, it puts in
> each of these states equal fraction of the electron. TETRA procceds
> differently and thus ROOT and TETRA may well finish with different
> distribution of electrons that are close to the Fermi energy. LDA+U then
> basicaly stabilize the states that are occupied and destabilize empty
> states. Different state may therefore be stabilized with TETRA and with
> ROOT. The best you can do is to construct different starting population
> matrices and to run LDA+U several times (with the same option either TETRA
> or ROOT). Etot then decides which of the solutions corresponds to the
> ground state.
>
> Regards
> Pavel
>
> _________________________________________________
> Dr. Pavel Novak
> Department of Magnetism and Superconductivity
> Institute of Physics AS CR
> Cukrovarnicka 10, 162 53 Praha 6, Czech Republic
> tel: +420 2 20 318 532
> e-mail: novakp at fzu.cz
>
> On Thu, 28 Jun 2007, Masao ARAI wrote:
>
>   
>> Dear WIEN2k users and depelopers,
>>
>> I have experienced a strange problem about LSDA+U calculation
>> with ROOT option in case.in2. Is the ROOT option compatible with
>> lapwdm?
>>
>> Let me explain the situation.
>>
>> (1) First, I performed standard LSDA+U calculations with
>> "TETRA" option in case.in2.
>> I obtained well converged insulating ground state.
>>
>> (2) I modified the case.in2 and set "ROOT" option. Then I restarded
>> the LSDA+U calculations with converged charge density and case.dmat* files.
>> However, the calculuation was converged to the completely different state.
>> At the first iteration, the eigenvalue agreed with those from (1). But, the
>> density matrices computed from lapwdm were different from those in (1).
>>
>> I suspect that the lapwdm is not compatible with ROOT option.
>>
>> The lapw2 emits weights to case.weighdn and case.weighup. These weights
>> are then read by lapwdm. When the ROOT option is used, the weights
>> emitted by
>> lapw2  are just proportinal to the multiplicities of k-points.
>> Even states higher than Fermi energy have non zero weights in case.weighdn
>> and case.weighup.
>>
>> Note, the weights in case.weightup/dn (file no.11) are
>> appropriately truncated at EF and the weights of states with higher
>> energy are zero.(around line 455-462 in lapw2.F)
>>
>> 455            DO IE=1,NE(K)
>> 456                wei=WEIGH(k,ie)
>> 457                eee= E(k+(ie-1)*2*nkpt)
>> 458                IF (eee.lt.eeeo) shift=0.
>> 459                eeeo=eee
>> 460                eee=eee+shift
>> 461                IF (eee.gt.EF) wei=0.
>> 462                WRITE(11,79) eee,wei
>> 463            END DO
>>
>>
>> But, the weights in case.weighup (file no.26)and case.weighdn (no.27)
>> are emitted without any special treatment. (line 333 and 351 in lapw2.F)
>>
>> 333                   write(26)(WEIGH(J,N), N=1,NE(J))
>> 351                      write(27)(WEIGH(J,N), N=1,NE(J))
>>
>>
>> Of course, the states higer than EF will not be used in lapw2 because
>> the varibale nemax is appropriately adjusted.
>>
>> In lapwdm, the weights are read from case.weighdn/up.
>> Then all states are used for computation of the density matrices.
>> (There is no trick like nemax in lapw2). So, if the ROOT option is used,
>> the lapwdm seems to give wrong values.
>>
>> Your sincerely,
>>
>> --
>> Masao Arai
>> NIMS(National Institute for Materials Science)
>> CMSC(Computational Materials Science Center)
>> Tsukuba Japan
>>
>> _______________________________________________
>> Wien mailing list
>> 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
>