[Wien] in1new problem

Mon Jun 6 11:26:33 CEST 2005

I guess I issued this warning before: the   -in1new   switch is NOT
necessarely more save (or better) than the default WIEN2k input. It depends
on the case AND IN PARTICULAR how the seperation energy between semicore and
valence is defined and found (check case.scf).

The La 5d states are pretty empty, thus do not play a significant role
for any occupied part and thus (independend on E-seper) the 4d character
is found ONLY at the proper states leading to E-parameters for these states.

However, Sb has real 5p valence states and 4p semicore! The 5p states may
now hybridize not only for the true valnece states, but also have some
contributions at states at eg. -1 Ry. It now depends whether these states at
-1 Ry are found as "valence" or semicore" and of course on the amount of
hybridization, but this admixture at eg. -1 Ry may "artificially" raise
the 4p "semicore energy", so that it is not put at the true -6 Ry, but at
-5.95, which is too high.

Thus one must select the E-seper parameters in case.in2 properly, but 
there may be cases, where a single "semicore/valence seperation energy
is not sufficient and will not work.

A way out of the dilemma might be: Change write_in1 such that it
does NOT fix the E-parameter for these low-E states, in other words, it
should still "search" (with the original algorithm, setting a delta-E in 
case.in1) for the E-parameter of all states below eg.  -3 Ry; 
but set properly and automatically good
values for the valence states. This way one would keep the original 
settings for the deep sc states, but set proper E-parameters for the
valence states (the "default 0.3" is the real cause of problems in most
cases. )

I include the modified (untested) script.
Let me know, if this suggestion works.

Regards

> I observe a problem when using in1new, and I'm not sure whether this is due to
> my misunderstanding or due to a problem with the method:
> 
> The compound is LaSb in the NaCl structure, using spin-orbit coupling. If
> in1new is used (eseper0=0.40 is needed), the scf-cycle stops with the famous
> QTL-B error, in a band with energy close to -6 Ry. There are two states with
> such an energy: La-4d and Sb-4p. These are the corresponding eigenvalues at
> gamma, without and with spin-orbit:
> 
> no spin-orbit :
>       EIGENVALUES ARE:
>        -6.0760613   -6.0760009   -6.0759824   -5.9252884   -5.9252770
>        -5.9252356   -5.9248027   -5.9247402
> 
> with spin-orbit :
>     EIGENVALUES ARE:
>          -6.4754331   -6.4754331   -6.0550927   -6.0550927   -6.0550443
>          -6.0550443   -5.8783683   -5.8783683   -5.8783219   -5.8783219
>          -5.8389144   -5.8389144   -5.8386363   -5.8386363   -5.8385865
>          -5.8385865
> 
> If the old linearization scheme is used, acceptable energies are found:
> 
> La-4d:
>          E( 2)=   -5.9275   E(BOTTOM)=   -5.930   E(TOP)=   -5.925
> 
> Sb-5p:
>          E( 1)=   -6.0800   E(BOTTOM)=   -6.090   E(TOP)=   -6.070  However,
> in1new finds quite different values:
> 
> La-4d: -5.9193, which is quite correct :
> 
> :QTL001: 1.9855 5.268411.0825 0.1647 1.7559 3.5126 0.0000 2.2848 2.2855 2.1701
> 4.3420 0.0000
>        Q-s-low E-s-low   Q-p-low E-p-low   Q-d-low E-d-low   Q-f-low E-f-low
> :EPL001:  1.8705 -1.2546    4.5593 -0.1360   10.0095 -5.9193    0.0057 -0.6566
> 
> Sb-5p: -5.9529, which seems too low to me. Note also the 6.126 p-electrons.
> 
> :QTL002: 1.2628 7.706610.0104 0.0443 2.5679 5.1387 0.0000 2.0061 2.0061 1.9993
> 3.9989 0.0000
>        Q-s-low E-s-low   Q-p-low E-p-low   Q-d-low E-d-low   Q-f-low E-f-low
> :EPL002:  0.2081 -0.2120    6.1269 -5.9529    9.9223 -1.0701    0.0154 -0.2356
> 
> A calculation using the old linearization scheme converges smoothly, when the
> in1new is used (even when starting from the converged case), then a QTL-B
> error shows up almost immediately.
> 
> If no spin-orbit coupling is used, then convergence is much better even with
> in1new (still hindered by a QTL-B of about three at the fermi energy). Now,
> in1new identifies the correct energies:
> 
> La :
> 
> :QTL001: 1.9856 5.267911.0853 0.1653 1.7554 3.5125 0.0000 2.2862 2.2870 2.1701
> $
>        Q-s-low E-s-low   Q-p-low E-p-low   Q-d-low E-d-low   Q-f-low E-f-low
> :EPL001:  1.8625 -1.2606    0.0147 -1.0864   10.0033 -5.9217    0.0029 -1.1257
> 
> Sb (note also the Q-p, which is much closer to 6.0 now) :
> 
> :QTL002: 1.2637 7.707210.0091 0.0442 2.5677 5.1395 0.0000 2.0059 2.0059 1.9990
> $
>        Q-s-low E-s-low   Q-p-low E-p-low   Q-d-low E-d-low   Q-f-low E-f-low
> :EPL002:  0.0061 -1.2675    6.0049 -6.0659    9.8659 -1.0725    0.0016 -1.2955
> 
> My hypothesis is that in1new is not able to deal with the large spin-orbit
> splitting in the Sb-4d. I'm not sure whether this is connected by the presence
> of La states at nearly the same energy or not.
> 
> Is this analysis correct? And how can this problem be circumvented? I wanted
> to use in1new in this case because I want to cover a broad volume range, over
> which the Fermi energy changes a lot. I can of course put proper energies by
> hand for every volume, but that's just what in1new was designed for to
> avoid...
> 
> Thanks,
> Stefaan
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> 

                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------