[Wien] Lapw1 error

[EB Lombardi]

Dear Lyudmila and Wien authors.

Regarding the previous 'ghost band' errors I mentioned in the previous 
e-mails on this topic:

1.  There has been an additional error in the APW calculation (using 
unmodified .in1c) which I did not notice originally. In the scf files 
the following suspicious lines occur:
QTL-B VALUE .EQ. **********   in Band of energy    0.00000
QTL-B VALUE .EQ. **********   in Band of energy    0.00000
Similarly for a mixed APW/LAPW run (see below):
QTL-B VALUE .EQ.    2.35833   in Band of energy    0.00000
QTL-B VALUE .EQ.    2.35913   in Band of energy    0.00000

The problem is  that there is no band of energy 0.00000.


2.   I tried your suggestion of using a value of -0.2 Ry for the P_s LO 
(using APW). This results in a similar crash, with a ghost band at 
-1.307337 Ry.   (the original unmodified APW run had a ghost band at 
-3.34806 eV)

3.   I also tried several other combinations, none of which worked, as 
follows:

3.1:   When using LAPW rather than APW for all atoms (default .in1c), 
the SCF cycles converge for the initial geometry structure only. 
However, when using "min -p -sp -s 1" for a geometry optimization, lapw2 
crashes again in the first SCF cycle of the new geometry, with many 
ghost bands at -0.39479, -0.38044, +0.26593, +0.28952, +0.33388, 
+0.37419, +0.41470, ..........  Ry

3.2:   When mixing LAPW for the P-s orbitals and APW for everything 
else, the results are similar to 3.1 with the SCF cycles converging for 
the initial geometry, but crashing in the first SCF cycle of the second 
geometry optimization step with ghost bands at +0.20051, +0.52656, 
+0.79082, +1.25045 Ry.

3.3   Changing sphere sizes: When using plain LAPW with default input, 
but changing the sphere sizes (Rmt_C from 1.30au to 1.37au,   Rmt_P from 
1.42au to 1.40au), this results in a crash in the 5th SCF cycle of the 
initial geometry with a message of "FORTRAN STOP L2main - QTL-B Error".
This is an "improvement" since that lapw2 is able to stop with an error 
message, rather than just crashing on the ******* in the *.help_ files. 
The QTL errors are not so large either. This case is rather perplexing 
since the sphere size of P was reduced, while the sphere sizes of the C 
atoms were increased.

For the 5th SCF cycle, the scf files contained:
phos.scf2dn:   QTL-B VALUE .EQ.    6.99287   in Band of energy    0.15828
phos.scf2dn_1:   QTL-B VALUE .EQ.    4.61602   in Band of energy    0.07087
phos.scf2dn_2:   QTL-B VALUE .EQ.    4.55460   in Band of energy    0.63894
phos.scf2dn_3:   QTL-B VALUE .EQ.    3.96843   in Band of energy    0.30243
phos.scf2dn_4:   QTL-B VALUE .EQ.    6.99287   in Band of energy    0.15828
phos.scf2p:   QTL-B VALUE .EQ.    6.99287   in Band of energy    0.15828
phos.scf2up_1:   QTL-B VALUE .EQ.    5.72604   in Band of energy    0.05845
phos.scf2up_2:   QTL-B VALUE .EQ.    5.74549   in Band of energy    0.62481
phos.scf2up_3:   QTL-B VALUE .EQ.    4.91945   in Band of energy    0.28777
phos.scf2up_4:   QTL-B VALUE .EQ.    7.45940   in Band of energy    0.14799

Note: calculations of other impurities with the same small C spheres had 
no problems.

Any suggestions would be welcome

Best regards

Enrico


Lyudmila Dobysheva wrote:

>Dear Enrico,
>
>So, the APW+lo makes the first density not good, as there appeared the level 
>with -3.34 Ry. I have looked in my old calculations (pure phosphorus and 
>alloys with phosphorus) and found that I was completely wrong: these 
>phosphorus s-electrons should be in the valence band. I mixed them with P2s.
>The APW+lo calculation in the first iteration is wrong,  the -0.79 which comes 
>from the atomic density estimate is more correct. Now I think that it is just 
>opposite to my first guess - too low, in my calculations the valence band 
>starts at -0.2 Ry and these P s-electrons should lie there. The wrong 
>linearization energy (too low!) gave a ghost at -3.34 Ry. The LAPW scheme has 
>apparently a wider interval of energy where linearization works well, and 
>does not give the ghost, and the APW+lo fails. 
>So, I think that now, if you can calculate with the LAPW scheme, you can 
>obtain a convergent calculation in it, and then try to continue with APW+lo. 
>And as a test it would be very interesting to make a calculation APW+lo at 
>-0.2 Ry or near, and look at the result, please write me the result - my 
>settings of the system do not allow me to repeat calculation of your system 
>quickly.
>By the way, I was using then:
>  0.30    3  0      (GLOBAL E-PARAMETER WITH n OTHER CHOICES, global APW/LAPW)
> 0   -0.79      0.010 CONT 0
> 0    0.30      0.000 CONT 0
> 1    0.30      0.000 CONT 1
>
>This is a good observation, I'll remember that sometimes APW+lo and LAPW may 
>give such different results at the first stages.
>
>Best regards
>  Lyudmila Dobysheva 
>------------------------------------------------------------------
>Phys.-Techn. Institute of       | Tel.(home):   7 (3412) 442118
>Ural Br. of Russian Ac. of Sci. | Tel.(office): 7 (3412) 218988
>426001 Izhevsk, ul.Kirova 132   | Fax:          7 (3412) 250614
>RUSSIA                          | E-mail: lyu at otf.fti.udmurtia.su
>------------------------------------------------------------------
>http://fti.udm.ru/ltt/personals/dobysh.htm
>------------------------------------------------------------------
>
>_______________________________________________
>Wien mailing list
>Wien at zeus.theochem.tuwien.ac.at
>http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
>  
>