VERY BAD NEWS of IBZKPT in vasp 5.3 with IBRION=6

[terencelz]

I'm trying to determine the elastic constants of a cubic cell (rocksalt TiN, Fm-3m). I selected a primitive cell (1 formula unit), KPOINTS with MP mesh of [9 9 9] and ran vasp with ISIF = 3.

The finite difference progress suggested there are 8 degrees of freedom with the default 2 displacements each. However after 3 * 2 = 6 progress steps, VERY BAD NEWS errors kept showing up:

VERY BAD NEWS! internal error in subroutine IBZKPT:
149 Reciprocal lattice and k-lattice belong to different class of lattices. Often results are still useful...

The error persisted for another 3 * 2 = 6 steps, and went away for the last 2 * 2 = 4 steps.

I read the IBRION info page and am aware that vasp 4.6 does not automatically change the k-points set, but vasp 5 should. I switched off symmetry (ISYM=0) and the run went very slowly but no error occurred. Could this error be related to k-points set not changing automatically?

I tried higher KPOINTS divisions, but it did not help. Meanwhile, with these errors, the elastic constants vary a lot with the increase of division values. I'm not sure, but certainly hope that if the error messages are gone, the results will be better converged.

[terencelz]

Did I make an obvious mistake? I searched the whole forum with key words like "elastic" but had no luck...

[admin]

Please attach your input files.

[terencelz]

INCAR:

PREC = Normal
ENCUT = 550
EDIFF = 0.00001
SIGMA = 0.1
NSW = 20
IBRION = 6
ISIF = 3
POTIM = 0.015
EDIFFG = -0.05
NPAR = 12
LCHARG = .FALSE.
LWAVE = .FALSE.
=====================
KPOINTS:

Automatic mesh
0
Monkhorst Pack
9 9 9
0. 0. 0.
=====================
POSCAR:

Ti1 N1
1.00000000000000
2.1268080000000000 2.1268080000000000 0.0000000000000000
0.0000000000000000 2.1268080000000000 2.1268080000000000
2.1268080000000000 0.0000000000000000 2.1268080000000000
Ti N
1 1
Direct
0.0000000000000000 0.0000000000000000 0.0000000000000000
0.5000000000000000 0.5000000000000000 0.5000000000000000

[admin]

The message does not mean an error.
The procedure doing the evaluation of symmetry uses a numerical criterion (SYMPREC=1E-5)
to determine which atoms are symmetrically equivalent. When this criterion is too soft
the programm can find "false" symmetry and reports that "reciprocal lattice and k-lattice belong
to different class of lattices." To get rid of this message you can:
1) Switch off the symmetry (ISYM=0)
2) Make the treshold for detecting the symmetry more severe (SYMPREC=1E-8)

[terencelz]

That solved it. What still is a bit painful is that a very dense k-point mesh is needed. I tried from NFREE = 2 to 4, but little is improved.

This is a plot of k-point division from 7 to 25, of the three elastic constants, C11, C12, C44. Big fluctuation still can be seen.

http://astro1.panet.utoledo.edu/~terencezl/moduli.png

[alex]

Hi terencelz,

I've got two thoughts about your non-divergent elastic moduli:

a) You fell for a trap with rhombohedral cells. Try a gamma centred mesh
=====================
KPOINTS:

Automatic mesh
0
Gamma
9 9 9
0. 0. 0.
=====================

This conserves the symmetry of the k-mesh. As a goodie you might get away with fewer k-points. There is a tutorial made by Andreas Eichler where youmay read the explanation (it should be somewhere on the VASP site - well it's 10+ years old, but hey, some things never change ;-) )


b) Did you consider IBRION = 8 ?
You get rid of all your weird artefacts (if any) created by numerical differentiation.

Cheers,

alex

[terencelz]

Hi alex,

a) From the docs,

"In this case, the third line (again, only the first character is significant) might start with 'G' or 'g' --for generating meshes with their origin at the $ \Gamma $ point (as above)-- or 'M' or 'm', which selects the original Monkhorst-Pack scheme. In the latter case k-point grids, with even ( $ {\rm mod}(N_i,2)=0)$ subdivisions are shifted off $ \Gamma $."

I used all odd division MP meshes, from 9 9 9 to 25 25 25, so I guess it shouldn't matter. In fact, I tried to switch to odd G-centered meshes, and it yielded exactly the same results as odd MP meshes.

b) In the OUTCAR there are a bunch of something like "MODULI FROM IONIC CONTRIBUTION", with tensors all zero, consistent with the IBRION=8 doc, but other contributions are simply not there. I'm not sure why.

[jnokelai]

The message does not mean an error.
The procedure doing the evaluation of symmetry uses a numerical criterion (SYMPREC=1E-5)
to determine which atoms are symmetrically equivalent. When this criterion is too soft
the programm can find "false" symmetry and reports that "reciprocal lattice and k-lattice belong
to different class of lattices." To get rid of this message you can:
1) Switch off the symmetry (ISYM=0)
2) Make the treshold for detecting the symmetry more severe (SYMPREC=1E-8)

Hello everyone,
I have been encountering the error discussed in this topic in different 2D systems and surfaces, but for me neither 1) nor 2) of admin's advices helped. To elaborate, in one of my systems I have a graphene supercell that is larger in x-dimension than in y-dimension and hence I would like to use less k-points in the x-direction, but if I do that, I obtain the error discussed in this topic:

VERY BAD NEWS! internal error in subroutine IBZKPT:
Reciprocal lattice and k-lattice belong to different class of lattices. Often results are still useful... 96

Only if I use equal amount of k-points in both x- and y-directions the simulation runs without error. To see the importance of this error I have conducted tests for a 2x1 graphene supercell test system:

Code: Select all

 graphene 2x1 supercell
 1.0000000000000000
     4.2744055462745010   -2.4678291927672120    0.0000000000000000
     2.1372027731372505    1.2339145963836060    0.0000000000000000
     0.0000000000000000    0.0000000000000000   15.0000000000000000
   4
Cartesian
  0.0000000000000000  0.0000000000000000  0.0000000000000000
  1.4248018487581771  0.0000000000000000  0.0000000000000000
  2.1372027731372505 -1.2339145963836060  0.0000000000000000
  3.5620046218954275 -1.2339145963836060  0.0000000000000000

I tested this structure with 13x15, 14x15, 15x15, 16x15 and 17x15 gamma-centered grids, for example:

Code: Select all

Gamma-centered grid
0
Gamma
  13  15  1
  0   0   0

INCAR used:

Code: Select all

PREC = Accurate
EDIFF = 1.0E-6
ISMEAR = -5
ICHARG = 11
GGA = PE

(The CHGCAR file used was from the 15x15 calculation).

If the error is important one should expect jump at energy at 15x15 when there is no error message, and this is indeed what happens. The jump is about 2meV:

Code: Select all

k-points    energy     NKPTS   error message?
 13x15    -36.928778     98         yes
 14x15    -36.928633    106         yes
 15x15    -36.926471     64         no
 16x15    -36.928581    121         yes
 17x15    -36.928467    128         yes

If I turn the symmetry off (add line "ISYM = 0" to INCAR), I still get the error messages and the energies do not change, not even for 15x15, for which amount of k-points increased from 64 to 113:

Code: Select all

k-points    energy     NKPTS   error message?
 13x15    -36.928778     98         yes
 14x15    -36.928633    106         yes
 15x15    -36.926471    113         no
 16x15    -36.928581    121         yes
 17x15    -36.928467    128         yes

I managed to get rid of the error messages by tilting the z-direction cell vector slightly (without "ISYM = 0"), but this did not fix the energy jump issue either:

Code: Select all

k-points    energy     NKPTS   error message?
 13x15    -36.928765     98         no
 14x15    -36.928624    106         no
 15x15    -36.926457    113         no
 16x15    -36.928577    121         no
 17x15    -36.928462    128         no

Tilted POSCAR:

Code: Select all

 graphene 2x1 supercell, tilted
 1.0000000000000000
     4.2744055462745010   -2.4678291927672120    0.0000000000000000
     2.1372027731372505    1.2339145963836060    0.0000000000000000
     0.0100000000000000    0.0000000000000000   15.0000000000000000
   4
Cartesian
  0.0000000000000000  0.0000000000000000  0.0000000000000000
  1.4248018487581771  0.0000000000000000  0.0000000000000000
  2.1372027731372505 -1.2339145963836060  0.0000000000000000
  3.5620046218954275 -1.2339145963836060  0.0000000000000000

In this test system the errors were about 2meV, but I am afraid this error might be much bigger in some cases. Also I would like to compare some 2D structures in such a situation that 1meV energies are relevant. So is there some way to fix this error altogether? Or should I just simply use N x N meshes even for cells that are much bigger in x direction than in y direction? Or can I just ignore this error and just make sure that I use the same k-point file for two systems whose energies I compare? With this I am meaning that are the errors equal for both systems so that the errors cancel and difference in energy is accurate in meV scale?

Thank you very much in advance.