I am calculating force sets on finite displacements generated by phonopy. My structure is TlNbO3 and phonopy generates 3 displacement poscars. When I used a 2x2x2 supercell and calculated the force sets there were no issues or errors. However, the phonon dispersions were no good and so I increased the size of the supercell to 3x3x3. When I went to calculate the force sets in vasp 2 of the 3 calculations gave me an error while 1 ran just fine.
The error reads: " BRMIX: very serious problems the old and the new charge density differ". The error appears about 10 DAV steps in for one run and about 5 DAV steps in for the other.
The incar I used for the 2x2x2 and what I started with for the 3x3x3 has the following:
PREC = Accurate
ENCUT = 600.0
EDIFF = 1e-8
IBRION = -1
NELMIN = 6
ISMEAR = 0
SIGMA = 0.01
LREAL = FALSE
LWAVE = FALSE
LCHARG = FALSE
Again, this worked just fine for 2x2x2 and for 1 of the displacements for the 3x3x3 supercell. To try and solve the error I searched online and was suggested to include ADDGRID = TRUE and ISYM = 0 in the incar. I tried this and still received the error. I optimized the structe with a 12x12x12 kmesh and am using a 4x4x4 kmesh for the calculation of the force sets on the finite displacements with a 3x3x3 supercell. Any suggestions on how to fix this issue would be appreciated. I can include more information if needed. Thanks.
I have had similar issues in the past where I would get the same error after few steps. ADDGRID and ISYM were helpful for me. I am not sure what else could help but as I was going through your INCAR, I was just thinking as why are you using IBRION=-1. I would say IBRION=2 will be better, unless there is a specific reason.
If change in IBRION does not help then I would suggest to use k-point mesh of 2x2x2 or 3x3x3 to just to see if that helps to get rid of error.
Pulkit Garg thank you for your response. I am using IBRION = -1 from the suggestion in this reference for using vasp with phonopy --> https://atztogo.github.io/phonopy/vasp.html . The goal here is to calculate the force sets so phonopy can create the phonon dispersion and write the force constants. From my understand IBRION = 2 is an ionic relaxation algorithm.
I optimized the structure with a kmesh of 12x12x12 and then now using 3x3x3 supercell i am using 4x4x4 kmesh. Are you suggesting to use even smaller kmesh of 2x2x2 or 3x3x3? I am not sure what you mean in the last line of your reply.
Thanks,
Robert
Yes I am suggesting to use even smaller K-point mesh to check if that works or you still get the same error.
Pulkit Garg So I try using smaller kmesh. For 2x2x2 kmesh It actually worked for one of my jobs. So now I can calculate the force sets from 2 out of 3 of the finite displacements created by phonopy. But for the third finite displacement I still get the BRMIX error. Can you help me make sense of this? What is the connection from the kmesh size and this error? Why would one finite displacement require different kmesh and how do I move forward? Any advice will help.
Thanks.
For those who may face this problem in the future:
BRMIX is a symmetry-related error and some forums say it may/ should disapperar with ISYM=0 (switching off the symmetry) but I found it is not helpful always. Moreover, affordable it could be not afordable for a large system.
Here is what worked for me :
- strick the symmetry criteria, setting SYMPREC=1e-6 (default is 1e10-5)
- increasing the k-mesh and keeping default SYMPREC. Since the mixer has found in-sufficient kpoints in BZ.
I hope this may be helpful.
Best,
Sufyan
I have just found that this error can be avoided in some cases by the following settings:
AMIX=0.02
BMIX=0.2
MAXMIX=40
ALGO=FAST
Especially MAXMIX seems to be very efficient (it increases memory requirements, though).
Follow the instruction given here -
https://dannyvanpoucke.be/vasp-errors-en/ .
It works for me.
On further experimenting, I also found that this error may depend on your choice of KPAR and the number of processors you use.
Check NKPTS in the OUTCAR. Make sure that you KPAR evenly divides that and if you are on a HPC machine (using nodes with many cores per node), choose the number of nodes equal to KPAR.
So, for example, you have NKPTS=80, you can use KPAR=10 and run on 10 nodes (so each node will calculate 8 k-points). If each node has e.g., 96 cores, you may also want to set NPAR to 4 or 8, this will depend on the architecture.
Generally, this is a bug which can be traced back to MPI_REDUCE calls.
Add "export I_MPI_ADJUST_REDUCE=3" in your submit-script or add "LSCALAPACK = .FALSE. " in your INCAR file.
In most cases, this error can be solved by testing ALGO=Normal | VeryFast | Fast.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science