[Wien] Interstitial N-N pair in GaP supercell

[Kakhaber Jandieri]

Dear Wien2K users,

As I wrote in my previous letters, I am trying to minimize the atom
positions in GaP supercell where one of the P atoms is replaced by N-N
interstitial molecule.
I found some strange (in my opinion) behaviour of  "mini".
Apparently the convergence was achieved after 18 scf cycles with the
output "ec cc and fc_conv 1 1 1". The  distance  between  relaxed  N-N
atoms was 1.57 A.
But the program did not stop and began to recalculate again from cycle 1.
After 16 scf cycles the situation was repeated but now  the  distance 
between  relaxed  N atoms was reduced down to  1.27 A.
And so on without ending.
In my opinion even 1.27 A  is already too small value (this is almost
the interatomic distance in isolated N2 molecule), but it was
decreasing  further and further.
So, finally I had the problems with overlapping spheres in spite of very
small  RMT(N)=1.1 bohr.

In output file of our cluster I found the following (below some fragment
of this file is shown):

..............................................................
..............................................................
in cycle 18    ETEST: .0001117900000000   CTEST: -.0009873
ec cc and fc_conv 1 1 1

>   stop
Fallback to compatibility mode with "old" save_lapw
broyden files deleted, clm*, dmat*, vorb*, vresp*, eece*, scf and struct
files saved under GaPNN_1
20.841u 0.352s 0:21.70 97.6%    0+0k 0+20648io 0pf+0w
0.844u 0.276s 0:01.54 72.0%    0+0k 1272+16896io 8pf+0w
clmextrapol_lapw has generated a new GaPNN.clmsum
ec cc and fc_conv 0 0 0
in cycle 2    ETEST: 0   CTEST: 0
ec cc and fc_conv 0 0 0
in cycle 3    ETEST: 0   CTEST: 0
ec cc and fc_conv 0 0 0
.............................................................
.............................................................


In my opinion 1.57 A  is much reasonable value for relaxed N-N distance
then 1.27 A or even smaller value.
I still wonder why the program begins recalculation after the apparently
successful message: "ec cc and fc_conv 1 1 1"?

Is there some possibility that the reason is the specific position of
N-N interstitial pair?  In my initial structure file  P(0,0,0) atom was
replaced by N-N interstitial pair having the positions symmetrically
shifted from (0,0,0) along  the z-direction. The constrains (0.0, 0.0,
1.0) in my case.inM   file  allow the displacement of N atoms only along
the z-direction. My native thought was that may be (0,0,0) is some
specific position in the supercell. If I know correctly, in usual
minimization tasks with substitution impurity, the atom at (0,0,0)
position should have full constrains in all directions. I tried to
replace  other P atom in the middle of initial 2x2x2 supercell, but "x
sgroup" always returns the converted supercell where N-N pair is near
(0,0,0), (0.5,0.5,0.5), (0,0,0.5)  and so on   positions.


I will be extremely thankful for any suggestion.
Kakha