FFT grid resolution convergence problem

[guoh]

I tested convergence of the Bader charge analysis with respect to the FFT grid resolution. The model I used is bulk diamond, 2 atoms each unit cell. PAW; GGA; KPOINTS 8x8x8. The default NG(X,Y,Z)F is 28. I included core electrons as suggested.

NG(X,Y,Z)F atom1 atom2
-------------------------------
28 4.1020 3.8980
36 4.0741 3.9259
48 4.0525 3.9475
56 4.0435 3.9565
72 4.0322 3.9678
96 4.0222 3.9778
150 4.0126 3.9874
---------------------------------

I think numbers of electrons are approching 4, which should be the exact value. The problem is twice the default grid size still can not produce satisfactory results in my case. Even with 4 times the grid size, or 64 times the sample points, there is still an annoying error, which would result in wrong conclusions. I think the resolution is fine for this small box. Did I do something wrong? I appreciate your advice.

The version of Bader is v0.25b 08/07/08.
Thanks.

[andri]

Have you read through viewtopic.php?f=1&t=368 ? It's a similar system.

[guoh]

The problem with diamond seems to be less severe than that with graphite since the charge assigned to a carbon atom in diamond is not oscilating. But the common problem in both systems (diamond and graphite) is, the charge of each carbon atom should be the same. I think the problem may be related to the partition to the delocalized electrons in C-C covalent bonds. I am not sure, just doubt. Twice the default FFT grid resolution can not reduce the error to an acceptable level for diamond.

[andri]

Can you post your error?

[guoh]

Please see the first message of this thread.

The error is that the two C atoms in diamond have different charges.