Hi,
A noncollinear job I was running killed itself (after looking very good). The only thing I can see in OUTCAR that might be relevant is a warning:
WARNING: grid for Broyden might be too small
Any ideas guys? There are various parameters controlling Broyden mixing. I’m not sure which I should try adjusting.
Also, looking more at the output it seems that it went through all allowed SC iterations (set to 200) despite reaching the required accuracy (EDIFF set to 1.0e-5) well before that . What reasons coule there be for this?
Thanks
Ben
Two queries: 1. Broyden mixing, 2. EDIFF convergence
Moderators: Global Moderator, Moderator
-
- Newbie
- Posts: 22
- Joined: Tue May 10, 2005 9:57 am
Two queries: 1. Broyden mixing, 2. EDIFF convergence
Last edited by bth20 on Tue Jul 12, 2005 9:56 am, edited 1 time in total.
-
- Administrator
- Posts: 2921
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
Two queries: 1. Broyden mixing, 2. EDIFF convergence
1)the grid on which Broyden mixing is done is reduced compared to the
full FFT grid, for all wavevectors which are not included in this grid,
a simple mixing is performed. Then, the residuals are compared. If
(rms(total) - rms(broyd)) / rms(total) > 0.1,
this warning is written.
to avoid this error, you can do 2 things
--) decrease the mixing parameters from the default settings
(AMIX,BMIX, AMIX_MAG, BMIX_MAG), which may be useful for slabs, clusters and
magnetic systems
--) increase the FFT grid (by increasing PREC)
2) concerning the changes of TOTEN and EDIFF: there is always one more
step done in order to check whether the energy really is converged to within
the limit set by EDIFF. Only if the difference is smaller than EDIFF for this
step as well, the electronic scf loop is terminated.
full FFT grid, for all wavevectors which are not included in this grid,
a simple mixing is performed. Then, the residuals are compared. If
(rms(total) - rms(broyd)) / rms(total) > 0.1,
this warning is written.
to avoid this error, you can do 2 things
--) decrease the mixing parameters from the default settings
(AMIX,BMIX, AMIX_MAG, BMIX_MAG), which may be useful for slabs, clusters and
magnetic systems
--) increase the FFT grid (by increasing PREC)
2) concerning the changes of TOTEN and EDIFF: there is always one more
step done in order to check whether the energy really is converged to within
the limit set by EDIFF. Only if the difference is smaller than EDIFF for this
step as well, the electronic scf loop is terminated.
Last edited by admin on Fri Jul 22, 2005 1:45 pm, edited 1 time in total.
-
- Jr. Member
- Posts: 53
- Joined: Thu Mar 11, 2010 1:13 am
- Location: Raleigh, NC
Two queries: 1. Broyden mixing, 2. EDIFF convergence
I was wondering, 1) how many "WARNING: grid for Broyden might be too small " error messages warrant evasive maneuvers, ie tweaking the INCAR parameters? I have a system in which I only care about convergence of the electronic GS. I receive this WARNING message once or twice; in a total of 4 calculations differing only in magnetic states. The warning appears 10 -12 iterations before convergence. Should I be alarmed. 2) How should one qualify this WARNING? Are there situations in which the WARNING may be disregarded?
<span class='smallblacktext'>[ Edited Mon Feb 07 2011, 02:16AM ]</span>
<span class='smallblacktext'>[ Edited Mon Feb 07 2011, 02:16AM ]</span>
Last edited by jlbettis on Mon Feb 07, 2011 1:14 am, edited 1 time in total.
VASP 5.2.11
Cray XE6
Cray XE6
-
- Jr. Member
- Posts: 53
- Joined: Thu Mar 11, 2010 1:13 am
- Location: Raleigh, NC
Two queries: 1. Broyden mixing, 2. EDIFF convergence
If the Admin has anything to add, it would be greatly appreciated.
Last edited by jlbettis on Fri Feb 11, 2011 2:14 am, edited 1 time in total.
VASP 5.2.11
Cray XE6
Cray XE6