[Wien] Inconsistency between energy and forces ?

Stefaan Cottenier Stefaan.Cottenier at fys.kuleuven.ac.be
Mon Nov 24 10:21:51 CET 2003


Dear Wien users,

I observe an inconsistency between total energy and forces, which I do not
understand. The situation is as follows:

A material with space group I4/mmm and atoms at Wyckof positions 4e,  2a, 4c
and 4e is relaxed for internal coordinates (see attach: original.struct).
Then a supercell is made with twice the volume of the original cell, which
has space group Cmca, atoms at 8f, 4a, 8e and 8f (see attach: super.struct).
When doing a single calculation of the supercell, I observe several strange
things:

1) The difference between the total energy of the supercell and twice the
total energy of the original cell, is 0.83 Ry. This is much more than what I
expect for identical Rkmax and a good k-mesh in both cells. For comparison:
the total energy of a supercell which is 4 times as large (space group
I41/acd, atoms at 16d, 8a, 16f and 16d) is only 0.0045 Ry different from 4
times the original cell's energy.

2) The original cell has 2 position components with forces. The forces are
relaxed to below 1 mRy/au before constructing the Cmca supercell. The
supercell has the same 2 degrees of freedom, plus 3 additional position
components that are allowed to feel a force. The force in Cmca on the latter
three components is negligibly small (which is to be expected if this
material has I4/mmm as ground state structure). But suprisingly, the 2
original free positions now feel large forces. This is unexpected (and even
unphysical), as the original cell should have allowed for full relaxation of
these positions. In the larger I41/acd supercell, all forces remain small.

3) Finally, I relax the 2 internal positions with large forces in the Cmca
supercell (even after this relaxation, sgroup still detects I4/mmm, as an
additional proof that this relaxation could/should have been possible also
in the smaller cell). As one of the 2 forces is much larger than the other
one (45 compared to 3), one should expect that the total energy goes down
simultaneously with the large force. However, what I observe is this
(position and force are for the atom with the large force):

:POS            :FOR            :ENE
0.33690       -45.104        -44772.149943
0.33504       -26.434        -44772.151366
0.33350       -11.760        -44772.150417
0.33226           0.395        -44772.148290

One clearly sees that the force becomes zero at a position very different
from the one where the total energy becomes zero.

Summarized, I feel like there is something wrong both with the forces and
the total energy of this Cmca cell. I do not see errors or abnormal things
in any of the calculations, and I did check that the above observations
occur both with normal (6.5) and large (8.0) Rkmax, and with normal (about
40) and large (150-200) k-meshes.

Can somebody give a hint about what is going wrong? Both case.struct's and
scf with info of the last iteration are attached.

Thanks,
Stefaan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: original.scf
Type: application/octet-stream
Size: 17045 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20031124/ab98b440/original.obj
-------------- next part --------------
A non-text attachment was scrubbed...
Name: original.struct
Type: application/octet-stream
Size: 2743 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20031124/ab98b440/original-0001.obj
-------------- next part --------------
A non-text attachment was scrubbed...
Name: super.scf
Type: application/octet-stream
Size: 18424 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20031124/ab98b440/super.obj
-------------- next part --------------
A non-text attachment was scrubbed...
Name: super.struct
Type: application/octet-stream
Size: 2606 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20031124/ab98b440/super-0001.obj


More information about the Wien mailing list