too slow Hirshfeld charges in iterative Tkatchenko-Scheffler

Problems running VASP: crashes, internal errors, "wrong" results.

Moderators: Global Moderator, Moderator

Post Reply
Message
Author
jcconesa
Newbie
Newbie
Posts: 34
Joined: Mon Jul 03, 2006 10:48 am
Location: Madrid

too slow Hirshfeld charges in iterative Tkatchenko-Scheffler

#1 Post by jcconesa » Mon Dec 05, 2016 8:52 pm

Hi all,
With certain systems, and using Tkatchenko-Scheffler vdW method with iterative Hirshfeld partitioning (IVDW=21), I found the error

Error(hirshfeld_iterative): Hirshfeld-I did not converge in 150 steps

Seemingly, the method convergence criterion is that the change in atomic charges should be |dq|<5E-5. My system was about to reach that convergence level (dq was 6.3E-5) but obviously was not quick enough. Is there a keyword for allowing an increase in the number of iterations allowed in that process, and/or an increase in the dq convergence criterion?
Best regards,
José C. Conesa

admin
Administrator
Administrator
Posts: 2922
Joined: Tue Aug 03, 2004 8:18 am
License Nr.: 458

Re: too slow Hirshfeld charges in iterative Tkatchenko-Schef

#2 Post by admin » Tue Dec 06, 2016 2:45 pm

There are two possibilities.
1. Decrease the convergence criterion via the parameter HITOLER in INPUT (default 5e-5).
2. Redefine maximum number of iterations in subroutine hirshfeld_iterative
(line "INTEGER,PARAMETER :: maxiter=150"
Normally the value of 150 is enough.
Can you display your INCAR? It is likely that SCF is not properly converged.

scadars
Newbie
Newbie
Posts: 3
Joined: Wed Oct 04, 2017 9:53 am
License Nr.: 5-449

Re: too slow Hirshfeld charges in iterative Tkatchenko-Schef

#3 Post by scadars » Tue Oct 10, 2017 11:20 am

Hi,

I have been very much interested by this post because I also got the same error (with VASP v5.4.1) on two systems among a large set of La-containing systems that I explored to test the effects of the La free-atom correction parameters, which are not defined in VASP for Lanthanides (and by the way is there a physical reason for this ?). The problematic systems are La2B3Cl and La2B3Br, which have the peculiarity to include B3 rings with short B-B distances of 1.51 Angstroms. I do not know whether this is the reason, but in those two systems I had to increase the HITOLER value up to 5e-3 to reach convergence (depending on the VDW_C6AU, VDW_R0AU and VDW_ALPHA values used in the calculation).

I have two questions regarding this issue :
1. Is there a clear reason why some geometries should be more problematic than others ? Is the short B-B bond length indeed the source of the problem in this system, and if so why ? Is it somehow related to the free-atom R0 value for B ?

2. What are the possible consequences of increasing the HITOLER value on the results of the calculations (energies, relaxed geometry, other...) ?

Below are one POSCAR and one INCAR files representative of this problem:

***************** POSCAR *************************
B3 Br1 La2
1.0
7.8341999054 0.0000000000 0.0000000000
-3.9170999527 6.7846161364 0.0000000000
0.0000000000 0.0000000000 8.4399995804
La Br B
8 4 12
Direct
0.000000000 0.000000000 0.233009994
0.000000000 0.000000000 0.766990006
0.499949992 0.000920000 0.219290003
0.999080002 0.499029994 0.219290003
0.500970006 0.500050008 0.219290003
0.499949992 0.000920000 0.780709982
0.999080002 0.499029994 0.780709982
0.500970006 0.500050008 0.780709982
0.338999987 0.169430003 0.500000000
0.830569983 0.169569984 0.500000000
0.830430031 0.661000013 0.500000000
0.333333343 0.666666687 0.500000000
0.527000010 0.264999986 0.000000000
0.735000014 0.262000024 0.000000000
0.737999976 0.472999990 0.000000000
0.866999984 0.141000003 0.000000000
0.858999968 0.725999951 0.000000000
0.274000019 0.133000016 0.000000000
0.261000007 0.737999976 0.000000000
0.262000024 0.523000002 0.000000000
0.476999968 0.738999963 0.000000000
0.135000005 0.277999997 0.000000000
0.722000003 0.856999993 0.000000000
0.142999992 0.865000010 0.000000000


***************** INCAR *************************
SYSTEM=La2B3Br, free-cell geom opt with TS-IH disp correction, variable correction parameters on La
PREC=High
ENCUT=520
VDW_C6AU = 400 162.0 99.5
VDW_R0AU = 4.6 3.931 3.893
VDW_ALPHA = 45.2762 20.0 21.0
IVDW=21
HITOLER = 1e-5
LREAL=Auto
IBRION=1
ISIF=3
NSW=100
ISMEAR=-5
ISTART=0
ICHARG=1
NCORE=2
***************************************************

Thank you very much.
Best regards.
Sylvian.

Post Reply