[Pw_forum] Iron pseudopotential (NC, fully relativistic, with semicore states in valence)

[Carsten Fortmann]

Hi,
I'm trying to generate a fully relativistic Iron NC
PP with semicore states in valence with 'atomic' i.e. ld1.x

This is my input deck:

!Input file for quantumEspresso's 'atomic' pseudopotential generator.
&input
    iswitch=3,
    rel=2,
    dft='PBE',
    atom='Fe',
    config='[Ar] 3d6 4s0 4p-1 4d-1',
    nld=9,
    rlderiv=5.000000,
    eminld=-10.0,
    emaxld=2.0,
    deld=0.01d0,
/
&inputp
    file_pseudopw='Fe.upf'
    pseudotype=2,
    tm=.true.,
    nlcc=.true.,
    lloc=2 rcloc = 1.0
/
9
3S  1  0  2.000000  0.000000  1.10000  1.10000  0.500000
3P  2  1  2.00000   0.000000  1.10000  1.10000  0.500000
3P  2  1  4.00000   0.000000  1.10000  1.10000  1.500000
4P  3  1  -1.000000  -0.10000  2.860000  2.86000  0.500000
4P  3  1  -1.000000  -0.10000  2.860000  2.86000  1.500000
3D  3  2  4.000000  0.000000  0.80000  0.80000  1.500000
3D  3  2  2.000000  0.000000  0.80000  0.80000  2.500000
4D  4  2  -1.000000  -0.10000  4.370000  4.370000  1.500000
4D  4  2  -1.000000  -0.10000  4.370000  4.370000  2.500000
/

!&test
nconf=1
configts(1)='3d6 4s2'
/

The cutoff radii are taken from a scalar relativistic PP that I found
through the QE PP search engine,
http://www.quantum-espresso.org/wp-content/uploads/upf_files/Fe.pbe-sp-mt_gipaw.UPF

First question regards the &test configuration: I discarded the 4s state
in the PP generation and included it in the test, in accordance with
the following note in INPUT_LD1.txt:

    For PP generation you do not need to specify namelist &test, UNLESS:

    1. you want to use a different configuration for unscreening wrt the
    one used to generate the PP. This is useful for PP with semicore
    states: use semicore states ONLY to produce the PP, use semicore
    AND valence states (if occupied) to make the unscreening.

Is what I did correct? I don't see a 4S BETA or CHI in the upf file.
Where and how do I have to include it? If i add a 4S orbital with 0
occupation to the pp generation card, the calculation aborts with
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     Error in routine run_pseudo (1):
     Errors in PS-KS equation
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

I also set the 4p level to bound (occupation 0, energie 0), since it
appears as a bound state in the AE spectrum, in that case, the
calculation aborts with "Errors in PS-KS equation."

For now, I proceeded without the 4S orbital.
Looking at the output from ld1.x, I get a warning like this:

       Warning: n=3, l=2 expected 0 nodes, found 1
       Setting wfc to zero for this iteration
       (This warning will only be printed once per wavefunction)

I tried varying the cutoff radii, use a different local component, to no
avail. At least one orbital always has the wrong number of nodes.

Furthermore, the 3p eigenvalues differ significantly from the AE
calculation:

n l  j  nl             e AE (Ry)       e PS (Ry)    De AE-PS (Ry)
     1 0 0.5 3S   1( 2.00)       -8.26193       -8.26196        0.00003
     2 1 0.5 3P   1( 2.00)       -5.84251       -4.44750       -1.39501  !
     2 1 1.5 3P   1( 4.00)       -5.72956       -4.42853       -1.30103  !
     3 2 1.5 3D   1( 4.00)       -1.89476       -1.89435       -0.00041
     3 2 2.5 3D   1( 2.00)       -1.88318       -1.88326        0.00008


Does someone have a suggestion or sees what I'm doing wrong?

In earlier posts to this list, I read statements similar to
"multiple projectors in NC pseudos are not supported in this version."

Does this statement still reflect the current status?

What does "not supported" actually
mean?

On the same token, in pseudo-gen.pdf, Section 2.2.1, I read
    Note however that the implementation
    of multiple-projector PP's is correct for US pseudization: NC
pseudization
    is not properly done (a generalized norm-conservation requirement is
not accounted
    for).

What is behind this "generalized norm-conservation requirement"? Is it
related
(or identical) to the issues discussed e.g. in Morrison et al. (PRB_47_,
6728 (1993)) and recently picked up by Hamann, PRB_88_, 085117 (2013)?
I.e. the projector overlap (coupling matrix) becomes non-diagonal.

As far as I understand, this issue could easily be resolved
by diagonalizing the coupling matrix and
rotating the projector functions accordingly.

I am very interested in any comments on this issue.

Carsten

--
Carsten Fortmann, Ph.D.
Scientific Software Developer
QuantumWise A/S
Lersoe Parkalle 107
2100 Koebenhavn
Denmark


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: OpenPGP digital signature
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20150209/09ef92c9/attachment.sig>