[Pw_forum] PAW pseudopotential with gipaw reconstruction

[Giuseppe Mattioli]

Dear all
I've generated a PAW pseudopotential with gipaw reconstruction, by  
following the template found in pslibrary.0.2.4. This is my PP

  &input
    title='C',
    prefix='C_paw_gipaw'
    zed=6,
    rel=0,
    config='1s2 2s2 2p2 3d-2',
    iswitch=3,
    dft='PBE'
  /
  &inputp
    upf_v1_format=.true.,
    lpaw=.true.,
    pseudotype=3,
    file_pseudopw='C.PBE-n-kjpaw-gipaw.UPF',
    author='GM following psl-0.2.4',
    lloc=2,
    which_augfun ='BESSEL',
    rmatch_augfun=1.1,
    nlcc=.true.,
    new_core_ps=.true.,
    rcore=0.8,
    tm=.true.
    lgipaw_reconstruction=.true.,
  /
5
2S  1  0  2.00  0.00  1.00  1.30  0.0
2S  1  0  0.00  0.05  1.00  1.30  0.0
2P  2  1  2.00  0.00  1.00  1.45  0.0
2P  2  1  0.00  0.05  1.00  1.45  0.0
3D  3  2 -2.00  0.10  1.10  1.10  0.0
  &test
  /
3
2S  1  0  2.00  0.00  1.00  1.30  0.0
2P  2  1  2.00  0.00  1.00  1.45  0.0
3D  3  2 -2.00  0.10  1.10  1.10  0.0

pw.x (QE 4.3.2) does not like the above PP... It stops complaining

   
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
      from read_pseudo_nl : error #         3
      Reading pseudo file (QIJ)
   
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

      stopping ...

If I drop the following line from the input file (not suggested by the  
XSPECTRA manual...)

    upf_v1_format=.true.,

then pw.x crashes silently, and my cluster complains

# FROM IOTK LIBRARY, VERSION 1.2.0
# UNRECOVERABLE ERROR (ierr=-1)
# ERROR IN: iotk_scan_begin (iotk_scan.f90:182)
# CVS Revision: 1.23
# Tag not found
namel=PP_GIPAW_ORBITALS

Did I mess up with some of the properties of PAW PPs?

You could say: why don't you try good old rrkjus PPs?

I've tried this one

  &input
    title='C half1s',
    prefix='C-gipaw-1shalf',
    zed=6,
    rel=0,
    config='1s1.5 2s2 2p2 3d-2',
    iswitch=3,
    dft='PBE'
  /
  &inputp
    upf_v1_format=.true.,
    pseudotype=3,
    file_pseudopw='C.pbe-n-rrkjus-half1s-gipaw.UPF',
    zval=4.5,
    author='GM',
    lloc=2,
    which_augfun ='PSQ',
    rmatch_augfun=0.9,
    nlcc=.true.,
    new_core_ps=.true.,
    rcore=0.8,
    tm=.true.
    lgipaw_reconstruction=.true.,
  /
5
2S  1  0  2.00  0.00  1.00  1.30  0.0
2S  1  0  0.00  0.05  1.00  1.30  0.0
2P  2  1  2.00  0.00  1.00  1.45  0.0
2P  2  1  0.00  0.05  1.00  1.45  0.0
3D  3  2 -2.00  0.10  1.10  1.10  0.0
  &test
  /
3
2S  1  0  2.00  0.00  1.00  1.30  0.0
2P  2  1  2.00  0.00  1.00  1.45  0.0
3D  3  2 -2.00  0.10  1.10  1.10  0.0

It contains half-core hole, which I need for XSPECTRA purposes, but it  
is roughly the same PP. The problem is that it is very hard to achieve  
convergence by using this one. Sometimes pw.x converges rather  
smoothly, sometimes diagonalization goes bananas (both cg and david,  
both with very tiny wfc mixing). I could suspect a ghost state, but  
I'm not able to deal with half-core ghosts... So I was looking for  
something different.

Sorry for the long tale and thanks in advance for advice

Giuseppe

Giuseppe Mattioli
ISM-CNR
Italy