My Community

Dear VASP users and/or developpers,

I recently encountered multiple errors using the Tkatchenko-Scheffler method with iterative Hirshfeld partitioning (TS\HI, IVDW = 21), which seem to be very much related to the initial geometries of my system. Specifically, I used crystal-structure prediction calculations (with VASP v5.4.1 for the structure relaxations) in the LaOCuS system and then faced for a number of structures a problem that appears to be related to the iterative Hirshfeld partitionning procedure. Note that I only include the IVDW = 21 tag after a first step in which a crude optimization with fixed cell volume is performed. When I include the TS\HI correction, in a second step of the geometry relaxation, the first electronic minimization converges succesfully, but then the calculation stops in the middle of the TS\HI calculation procedure. The error file ends with the following lines :

******************** ERROR file *********************
Hirshfeld-I algorithm:
translations: 2 2 1
SUBROUTINE COMPENSATION_WATSON: ERROR1, unable to reach convergence in 1000 cycles.
(written once per processor)
********************************************************

The POSCAR and INCAR files of a representative example of this problem are given below :
******************** POSCAR **********************
La2O2Cu2S2 - structure generated by a "symmetric random" algorithm within USPEX
1.0000
4.947281 -0.120366 -0.541741
-2.032031 4.004683 0.009609
-0.314816 0.033545 7.542765
La O Cu S
2 2 2 2
Direct
0.345819 0.662108 0.123361
0.663701 0.343225 0.106892
0.996173 0.979656 0.977840
0.974999 0.987101 0.547130
0.989501 0.991909 0.779362
0.992996 0.989935 0.232638
0.325775 0.645705 0.718672
0.662507 0.342007 0.728175

******************** INCAR **********************
SYSTEM = La2O2Cu2S2, variable-cell-fixed-volume optimization, close to minimum
LREAL = .FALSE.
PREC = Normal
IVDW = 21
VDW_C6AU = 800 15.60 253.0 134.0
VDW_R0AU = 4.8 3.194 3.761 3.855
VDW_ALPHA = 74.6 5.40 42.0 19.6
ENCUT = 400
EDIFF = 1e-3
EDIFFG = 1e-2
IBRION = 1
ISIF = 4
NSW = 100
ISMEAR = 1 ; SIGMA = 0.14
POTIM = 0.20
LCHARG = FALSE
LWAVE = FALSE
NCORE=4
SYMPREC = 1e-3
*******************************************************

Increasing HITOLER up to 5e-2 does not solve the problem. I also tried to play with the PREC parameter (setting it to Accurate instead of Normal) and the number of k-points, but without any effect.

Can someone help me understand what is going wrong with the TS\IH correction in this case and how I can possibly solve the problem ?

Thank you very much.
Best regards.

Sylvian.

The problem cause POTCARs constructed for non-integer atomic electron
occupations (La, La_GW are problematic; La_s is OK). Write to the author
of the code (Tomas Bucko - tomas.bucko@univie.ac.at). He will send you the patch
which will correct your code (5.4.1 or 5.4.4).

Thank you very much for your answer. I am glad to hear that this is a known technical issue and that a solution exists. I will contact Tomas Bucko as you suggest.

Is the patch implemented in the most recent versions of VASP ?

Best regards.
Sylvian.

It was not known. You have found it. We have analysed the problem
and offer a quick solution. Later a corrected code will be distributed.