How to control the temperature fluctuation during MD simulations with ReaxFF

LAMMPS
,
1

Hello!
Recently I’ve been interested in MD simulations on Ni-YSZ (ZrO2 doped with Y element). I started to reproduce the results in some references. Now I have successfully reproduced the results from using the BMH and Morse hybrid potentials and similar models of them (dx.doi.org/10.1021/jp310920d).
For some reasons, I want to use Reaxff (dx.doi.org/10.1021/jz501891y) to simulated the models, but thing not going well … I found that the temperatures of the systems in NVT ensemble sharply raised at the first few steps (thousands of Kevin when my set is 300 K), and the trajectory showed that the structure expanded severely. I think it’s unnormal because it wasn’t observed when I used BMH and Morse hybrid potentials. And I don’t think the structures have any problems because they are similar to those in references and were pre-relaxed by DFT calculations (Redirecting).
Could anyone please give me some advises about how to improve the calculations, like parameter setting? I also uploaded the input set, structure and log of calculations for details.
structure.data (97.8 KB)
input (866 Bytes)
log (16.0 KB)

THANKS!
John.

2

Just thinking loudly: You are using metal units in the attached script. Most ReaxFF parameterizations are available for real units. Have you checked that your reaxff parameter file is suitable for metal units?

Evangelos

3

For what reasons? When people don’t say why, it is often an indication that their choices are not made for a good reason.

I agree with @evoyiatzis that using metal instead of real units with ReaxFF is a mistake.

4

THANKS a lot !!!
It’s very helpful to point out that. I checked the reference and authors indeed used real units, unfortunately I didn’t notice that. And now I change the units in my script and it works.
Again, thanks !
John

5

Actually, it’s because that I want to simulate the Ni-YSZ behaviors in different atmospheres like H2O, SO2. And the former BMH/Morse potentials aren’t able to do so. I have to say that I’m a beginner in lammps and MD simulations, so it’s difficult to develop new potentials for me. Actually I have learned to develop Machine Learning Force Field using DeepMD-kit, but it’s time-consuming.
I’ll be appreciated it if there are better ideas.
John