Combination

LAMMPS LAMMPS General Discussion
1

Dear Professors,

Good evening. 2 years ago, I did a chemical vapor deposition (CVD) simulation with LAMMPS.

For the bottom substrate, I divided it into three layers:

  • Top layer: NVE
  • Middle layer: NVT
  • Bottom layer: Fixed

However, during my recent pre-defense, one of the professors questioned the validity of this setup, suggesting that such a combination might not be reasonable. I was asked to provide a theoretical basis that supports this combination — not just examples where others have used it, but actual literature that explains why it is theoretically valid.

I have searched extensively but have not been able to find a suitable reference. I would greatly appreciate it if any of you could provide some guidance or point me in the right direction.

Thank you very much,
Li

2

There is no theory needed, just common sense:

  • you need an immobile chunk of atoms at the bottom to avoid unphysical reconstruction at the bottom surface. That layer needs to be big enough for the cutoff of the interaction model you are using
  • you need a thermalized chunk of atoms above that to model the exchange of kinetic energy with the bulk that you have cut away. this layer has to be large enough to have reasonably small total thermal fluctuations for that.
  • you need a non-thermalized chunk of atoms at the top so you get the correct behavior with respect to any kinetic energy that is transferred from the deposited atoms. You only want that to be transferred to the bulk through the “natural” thermal conductivity of the surface medium. Thus this layer has to be large enough to allow for that transfer to happen.

My colleagues and I have been recommending this kind of setup for many years. Many people make the mistake of choosing the immobile bottom too small and have the thermostat go all the way to the surface (which is bad, too, for the reasons mentioned above). So I don’t know what kind of alternative would be reasonable.

I would be curious to know the justification for the concern of that professor and specifically which alternative would be suggested.

1 Like
3

It’s worth thinking about whether your model has predictive value instead.

Does it prescribe how CVD progresses as a function of the surface composition, temperature, or other parameters?

Does it make those prescriptions with statistical clarity? In other words, do you think you might get an opposite result simply by starting from another initial configuration?

If the model makes clear, practical predictions, which you can explain as a function of some feasible theory about the molecular dynamics you are seeing, then it is a valid computational study. Even if it is wrong, it is wrong in an interesting way (that is, some assumption you used in your methodology was not justified, and it is worth understanding why so that future modellers will be more cautious making the same choices).

As a practical LAMMPS comment – if you have some structural descriptor of the NVT region (such as g(r)) and you can show that a bulk, fully periodic, NVT simulation of that material has similar structure, that would show that your simulation protocol hasn’t destroyed the physical characteristic of that bulk region.

Equally well, consider that the professor may just be interested in seeing how you defend your idea, and being overly harsh or broad in how they are phrasing it.

1 Like