The ABINIT Users Mailing List ( CLOSED )

[Marc Torrent <marc.torrent@cea.fr>]

Dear Kris,

If you want to have a "perfect" restart when using PAW you not only need 
to use the density file but also the wave-functions file.
So add
 prtwf1 1
 getwfk2 -1
in your input file...
... and your "rhoij" and eigenvalues will have the expected values.
This should be documented (I know)...

The reason of this is the following:
 - you need the density to compute the potential and all quantities 
expressed on the plane waves basis
 - you need the wave-functions to compute the rhoij terms governing the 
spherical part.

Some outputs should be suppressed for the non-self-consistent part of 
the run:
 - the comparison of the compensation density between the two grids
 - the pseudopotential strengths Dij
They are not updated... so do not take them into account.

Best wishes,

Marc


kdelaney@mrl.ucsb.edu a écrit :
> Hi Marc,
>
> The problem exists at least with versions 4.6.5 and 5.2.2; I tried with the 
> precompiled binary distributions generated with IFC.
>
> I tried to use five different PAW data with the same result. Three I generated myself, the other two were 6c_lda.paw and 8o.paw found in the Psps_for_tests directory of the abinit distribution. 
>
> The problem is more severe for potentials with large augmentation 
> contributions, hence C and O show the error but it is very subtle. The error 
> is best visualized by running an SCF calculation followed by a non-SCF 
> calculation with a subset of the original k-points. Then, comparing the 
> eigenvalues directly, one can see differences that are too large for a 
> convergence error. For potentials with large augmentation contributions 
> (transition metals, rare earths) the band structure generated by a non-SCF 
> calculation becomes very messy, while the eigenvalue spectrum from the SCF 
> calculation looks sensible.
>
> At the end of this post is an example input file that demonstrates the 
> problem. It is a very basic input modified from the PAW1 tutorial on the 
> abinit website. With this input, I used 6c_lda.paw and obtained the following 
> eigenvalues for kpoint #2 (-0.125, 0.50, 0.00):
>   
>   band  | SCF eval (Ha) | non-SCF eval (Ha)
> --------------------------------------------
>     1   |  -0.39181     | -0.39786
>     2   |  -0.31293     | -0.31258
>     3   |  -0.11604     | -0.11586
>     4   |  -0.09376     | -0.09357
> ...
>
> As I commented in my previous post, the issue appears to stem from the 
> augmentation data not being correctly initialized in the second calculation.
>
> Thanks,
> Kris.
>
> --
>
> ndtset 2
>
> ecut 15.0
> pawecutdg  35
> nstep 30
>
> acell 3*10.6
> rprim 0.0 1/2 1/2
>       1/2 0.0 1/2
>       1/2 1/2 0.0
>
> ntypat 1
> znucl 6
>
> natom 2
> typat 1 1
> xred
> 0.0 0.0 0.0
> 1/4 1/4 1/4
>
> nband 22
> prtwf 0
>
> #SCF calculation
> iscf1 13
> occopt1 7
> tsmear1 5.0d-3
> prtden1 1
> kptopt1 1
> ngkpt1 4 4 4
> nshiftk 4
> shiftk 1/2 1/2 1/2
>        1/2 0.0 0.0
>        0.0 1/2 0.0
>        0.0 0.0 1/2
> toldfe1 1.0d-7
>
> # non-SCF calculation 
> iscf2 -2
> getden2 -1
> kptopt2 0
> nkpt2 3
> kpt2  -0.125 -0.25 0.0
>       -0.125  0.50 0.0
>       -0.250 -3/8  0.0
> tolwfr2 1.0d-10
> enunit2 1
> prteig2 1