ERROR in rs_pw_transfer during cell optimization

[Andrzej]

On Mar 4, 5:33 pm, Teodoro Laino <teodor... at gmail.com> wrote:
> Andrzej,
>
> therefore you won't get anything meaningful even without the above error
> message.
> Why don't you try with plain GPW?
>

Thank you very much for a response.
Why I decided to use all electron/GAPW,
I would prefer to use pseudopotentials and GPW, but this approach led
me to wrong results during geometry optimization.
Here are a summary of my tests. All files I have loaded as
geo_opt_tests_andrzej.tar.gz

I used three approaches during crystal structure optimization
(GEO_OPT):
1. all-electron/GAPW
- results seems to be correct, here is the optimized structure:
i =       20, E =      -254.4397076770
  H         0.0000408810        2.4704770383       14.8136893102
  H        -0.9689532669        3.0298933452        1.4034103139
  H         2.1394587515        1.2352435453        0.1863173192
  H         3.1084573793        0.6758216607       13.5966012290
  H         0.0000004580        1.3516209597        1.4034332796
  H         2.1394994962        2.3541034972       13.5965784978
  H         0.9689524974        3.0298923370        1.4034718821
  H         1.1705433143        0.6758226686       13.5965389830
  B         0.0000057953        2.4704748301        1.0102779165
  B         2.1394940538        1.2352442572       13.9897224598
 Mg         0.0000006098        0.0000060417        0.0001054861

and Mullikan charges:
    9  B    -0.182266
   10  B    -0.182275
   11  Mg    0.384598
Due to crystal symmetry, Mg atom should be at 0.0 0.0 0.0 and B atoms
should be equivalent. It seems, that the program preserves all these
features during the optimization.

2. pseudopotential: GTH-BLYP  // GAPW
Symmetry seems also to be preserves, BUT the optimization goes in
wrong direction. here is an energy listing:
 i =        1, E =       -73.5175896722
 i =        2, E =       -73.5181501663
 i =        3, E =       -73.5169493546
 i =        4, E =       -73.5146991276
 i =        5, E =       -73.5120392333
 i =        6, E =       -73.5092432038
 i =        7, E =       -73.5064694489
 i =        8, E =       -73.5037695513
 i =        9, E =       -73.5011718324
 i =       10, E =       -73.4986892533
...................................................
 i =       50, E =       -73.5064816396
 i =       51, E =       -73.5069568554
 i =       52, E =       -73.5067869799
 i =       53, E =       -73.5065578295
 i =       54, E =       -73.5066356661
 i =       55, E =       -73.5066449967
 i =       56, E =       -73.5066449962 (final)

3. pseudopotentials: GTH-BLYP// GPW
total mess in the results, symmetry is destroyed at once, here is an
example of a structire at one of the optimization step:
 i =      315, E =       -73.6283687957
  H        -0.0033299605        2.4685447504       14.9168795057
  H        -0.9730619996        3.0348140179        1.4858979291
  H         2.1319315210        1.2308569967        0.1017540811
  H         3.1129620505        0.6785024920       13.5228836249
  H         0.0022477584        1.3454147948        1.4856679690
  H         2.1439900361        2.3566698080       13.5228376561
  H         0.9557248302        3.0221676817        1.5307358159
  H         1.1633693810        0.6717413040       13.5297681616
  B        -0.0065932634        2.4665767592        1.1133874591
  B         2.1349713767        1.2326496289       13.8992265268
 Mg        -0.1065214033       -0.0618890343        0.0417902798

Mg is far from 0 0 0
and charges:
     9  B     0.041713
    10  B    -0.057872
    11  Mg   -0.258309
show that B atoms are lo longer equivalent.
Energy is oscillating strongly:
i =      289, E =       -73.6282904477
 i =      290, E =       -73.4773966228
 i =      291, E =       -73.6282970478
 i =      292, E =       -73.4774074691
 i =      293, E =       -73.6283035434
 i =      294, E =       -73.4774181459
 i =      295, E =       -73.6283099418

Of course, I could use CG and force convergence, and end up in an
artificial structure with broken symmetry, but it can't be a solution.

Playing with cutoff, basis sets and pseudopotentials I got results
where Mg atom wandered even more in the cell.
Only very large cut-offs can reduce (but not remove) these
oscillations.

To sum up, I was able to get results which seem reasonable only when
all electron/GAPW was applied.
What should I do then to get correct results with pseudopotential/GPW?

regards
Andrzej