unexpected MD results

[maria_dresser]

Hi,

I am working on an MD simulation of a layered transition metal oxide to try and observe how H2O molecules might dissociate within the intercalation layer. I heated the structure up to 500 K and all seemed well. I then proceeded to run the structure at 500K and the job stops after ~1100 steps. There is an error at the end:

ERROR FEXCF: supplied exchange-correlation table
is too small, maximal index : 5043

Despite this error, I suspected that the first 1100 steps could still contain important information. So I made an animation by uploading XDATCAR into OVITO, and the results seem strange. My main question is, why do the positions of the atoms just oscillate from one position to another over and over? I showed the animation to a senior student in my group who thinks something does not look right because of this oscillation back and forth. I have gone over all of my inputs, and I don't see anything obviously wrong. Thus, I was hoping someone could take a look at the XDATCAR and provide any insight as to whether this behavior seems unusual. If it is unusual, any advice about how I may need to adjust my inputs? Thank you for any help. (I was unable to attach the OUTCAR or vasp.out because they are too large, even in zipped format, but I have included other relevant files).

Best,
Maria

[alex]

Hello Maria,

I've checked your files and found some points to think about:
- 5fs timestep with O-H bonds in the system seems rather large. Water in Zeolite goes often with ~1fs timestep.
- 520 eV cutoff is high if you don't do NpT, so there is room for savings.
- set a minimum number of electronic steps, which helps to improve forces, e.g. set it to 3 for starters.

The error appears in my calculations if the setup (INCAR) is and the structure goes weird.

Good luck,

alex

[alexey.tal]

Dear Maria,

It would be very helpful to see the energies from this run. Could you please provide OUTCAR from the first steps of this simulation?

Also, I see in the POSCAR file that one of the atoms has an extremely large velocity compared to the others. Has this structure been properly equilibrated at 500K? Perhaps you can try restarting this calculation without initial velocities in the POSCAR file and see if the error persists.

[maria_dresser]

Hi,

Thank you for your reply. When you ask if it has been properly equilibrated at 500K, I don't entirely know what this means. I heated the structure from 0K to 500K with 2fs time step over 500 steps. Once the 500 steps were completed, I took the CONTCAR and copied to POSCAR and used an updated INCAR (which is the INCAR I uploaded that holds the structure at 500K) to get the data I sent you. Is this what is meant by properly equilibrated?

If I start without initial velocities in the POSCAR file, wouldn't this sort of negate the initial heat up to 500K?

The zipped OUTCAR file is too large, and the system won't allow me to upload it. Is there another way I can send it to you?

Re: to Alex, thank you for these pointers. I will consider this as well.

Thank you for your help,
Maria

[alexey.tal]

When you ask if it has been properly equilibrated at 500K, I don't entirely know what this means. I heated the structure from 0K to 500K with 2fs time step over 500 steps. Once the 500 steps were completed, I took the CONTCAR and copied to POSCAR and used an updated INCAR (which is the INCAR I uploaded that holds the structure at 500K) to get the data I sent you. Is this what is meant by properly equilibrated?

Yes, the procedure is fine. But such a big difference between velocities of hydrogen atoms indicates that perhaps the final structure was not converged. Furthermore, the initial structure and the choice of the coupling to the heat bath affect how quickly the systems attains the equilibrium temperature, so 500 steps may not be enough.

If I start without initial velocities in the POSCAR file, wouldn't this sort of negate the initial heat up to 500K?

I was suggesting that to see if your problem was indeed caused by the initial velocities stored in the POSCAR file.

The zipped OUTCAR file is too large, and the system won't allow me to upload it. Is there another way I can send it to you?

Could you extract the first 50-100 MD steps from the OUTCAR file?

[maria_dresser]

How would I know if convergence is reached? When I have previously done relaxation with VASP, the end of the vasp.out file has said something like "convergence reached". Should I be looking for a similar statement in the vasp.out for the MD simulation where I heat from 0K to 500K? If so, that statement is not there. The end of the vasp.out file says:
bond charge predicted
prediction of wavefunctions
wavefunctions rotated
TACC: Shutdown complete. Exiting.

Additionally, how will I know what temperature the system has reached? I thought the temperature information should be held in OSZICAR, but at each step the temperature is not given, it says T= ******.

I have attached the first 48 steps from the OUTCAR as well.

Thank you so much for your help.

[maria_dresser]

Also, I did attempt to remove the velocities from the POSCAR as you suggested, and the run stops after an hour with the following error:
ERROR in EDDIAG: call to ZHEEV/ZHEEVX/DSYEV/DSYEVX failed! error code was 1
I did not change any other initial parameters besides removing the velocities as suggested.

[alexey.tal]

The stars in the temperature mean that the value is too large to be printed with 6 characters. Also, the initial temperature, which is calculated with initial velocities, in the OUTCAR file is already too high. So I suspect that something went wrong in the calculation where you relaxed your structure at 500K.

Regarding the relaxation procedure, the most basic tests would be to check if the temperature is sensible and the temperature fluctuations are stable over time.

Could you please provide the files from the relaxation run at 500K?

[alex]

Hi Maria,

I did a bunch of runs with different time steps (POTIM). 1 fs seems to work fine (no jumping energy issues anymore). However, it's a steady decline in energy and T is 0K at some point.. Weird. This is sth. worth looking into.

Btw. set LDAUPRINT = 0 for far (!) less output.

Hth,

alex

[maria_dresser]

Thank you for this information. I have attached the relevant files for the run where I was trying to heat from 0K to 500K. Thank you for the help!

[alexey.tal]

As Alex correctly suggested you should decrease the time step (1 fs or even lower).

In the calculation where you heat up your system to 500K you used SMASS=-1 with the Nose-Hoover thermostat (MDALGO=2), but in this case SMASS is the effective mass of the heat bath, so velocities were not rescaled.

Also, the initial velocities in your POSCAR are not correct and you should remove them. To equilibrate your system at 500K you can run the same NVT MD simulation with a smaller time step until it converges to the right temperature. But be aware that this equilibration might take a while as the structure is quite far from the equilibrium.

[maria_dresser]

Thank you for your response. I will definitely try reducing the time step. I had a few questions regarding your other suggestions.
1. Regarding SMASS, I don't exactly follow what you are suggesting. Should I change SMASS to another value?
2. When you say 'the original velocities in your POSCAR' are you referring to the first POSCAR I sent for the run that starts at 500K or the second POSCAR I sent for the run from 0K to 500K (in this second case I don't think there are velocities).
Thank you for your help.

[alexey.tal]

1. In the MD simulation with the Nose-Hoover thermostat SMASS should be >=0, so your value is fine.
2. I meant the first POSCAR. The second one does not have velocities.

[alexey.tal]

I tested this calculation with the initial velocites and I also see that the temperature drops to 0K. So it is clearly the result of unreasonable velocities.

[maria_dresser]

Thank you for all of your help! I will implement these changes.