[Pw_forum] Bands.x not ordering bands correctly

stefano de gironcoli degironc at sissa.it
Wed Jan 13 23:29:36 CET 2016


Hi Hank,
   bulk PbS has 2 atom per cell. why do you need 20-30 ?

   bands.x "follows" a band based on the overlap of the wfcs in two 
neighboring points..
   it accepts the band on the basis of a rather small overlap threshold

... somewhere around line 350 in $espresso_root/PP/src/bands.f90
pro=cgracsc(nkb,becp%k(:,jbnd),becpold%k(:,idx(ibnd)), &
                          nhm, ntyp, nh, qq, nat, ityp, ngm, NEW, old, upf)
                  ENDIF
!                 write(6,'(3i5,f15.10)') ik,idx(ibnd), jbnd, abs(pro)
                  IF (abs (pro) > 1.d-2 ) THEN
                     il (idx(ibnd),ik) = jbnd
                     GOTO 10
                  ENDIF
...

    this is probably ok in most cases when there are few bands to chose 
from and only one is the "right one". In your case there are many bands 
and maybe there are many not too wrong bands that fulfill the 
threshold..:-(
    I would try to
    1) uncomment the write statement and see what are the typical values 
of abs(pro) you get.
      if everything is fine there should be only one jbnd band for each 
idx(ibnd) band that is significantly different from zero (much more than 
1.d-2 i would expect).. If you have more than one jbnd band that exceeds 
the threshold for a given examined idx(ibnd) then the code may do the 
wrong assignment
    2) if this is the origin of the problem i would try to play with the 
threshold to induce the code to make the proper assignment

    using a finer k point grid should also work but I would say that 15 
points in a 20-30 atom cell should already be enough !

   good luck,

   stefano

On 13/01/2016 21:52, Henry J Seeley wrote:
> Hello everyone,
>
> I'm using bands.x to assign eigenvalues to specific bands of PbS. I have
> fairly large supercells (20-30 atoms), so my bands are dense with lots
> of crossings. When I run bands.x it incorrectly assigns bands resulting
> in erratic band diagrams (e.g., bands suddenly jump across the bandgap
> to connect conduction and valence eigenvalues). What are the potential
> causes of this error?
>
> Perhaps I am not sampling my k-points densely enough; I have
> approximately 10-15 points per sampling line. I plan on increasing this
> density to 50 kpts per line, and will report back if this solves the
> problem. I am somewhat in a rush to get these results though, so if
> anyone has some insight it would be much appreciated.
>
> Also, assuming there is a solution to this problem, is band.x capable of
> sorting a 2D grid of kpoints into bands? I am interesting in generating
> contour maps.
>
> Thank you,
> Hank Seeley
> Chemistry PhD student
> University of Oregon
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum




More information about the users mailing list