Assistance Required: Adding a New Potential and Moment Tensor Potential

[Priyadarshini]

Hello everyone,  I would like to know the detailed process for the following:

Adding New Potential from the Literature:
I have come across Vashishta potential in a publication (Ref: P. Vashishta, R.K. Kalia, J.P. Rino, and I. Ebbsjö (1990), "Interaction potential for SiO2: A molecular-dynamics study of structural correlations," Physical Review B, 41(17), 12197-12209. DOI: 10.1103/physrevb.41.12197), which is not a predefined potential set in quantumATK, that aligns well with my system's requirements. However, I am unsure of the exact steps needed to integrate it into QuantumATK even after referring this documentation. https://docs.quantumatk.com/tutorials/combining_potentials/combining_potentials.html
Could anyone please provide a detailed guide on the following aspects?

The required format or file type for custom potentials (e.g., parameter files or scripts).
The directory structure and location where such potentials should be placed.
Instructions for ensuring compatibility with the existing workflows, including any scripts or modules that need to be edited.
Any examples or templates that can serve as a reference for adding potentials from external sources.

[Anders Blom]

I couldn't find a matching potential type in our list, so probably it is not easy at all. There are many other SiO2 potentials to choose from, but this functional form is not supported as far as I can tell (the problem is the third term in the two-body potential, we can cover all the other ones).

[Priyadarshini]

"Thanks for checking. If this functional form isn't directly supported, we might need to explore alternative SiO₂ potentials that QuantumATK can handle. Do you think we could approximate the third term using an existing potential or a combination of supported terms? Otherwise, we might need to consider a custom implementation or another workaround. Let me know your thoughts!"

[Anders Blom]

That's what I was checking, it's often the case that the whole potential doesn't exist under a dedicated name, but each term exists already. But I could not find anything resembling the last term. There are Coulomb terms that are proportional to ZiZj and Lennard-Jones terms which can do 1/r^n, but not the combination. You can still create a tabulated potential yourself, but in my own opinion there is bound to be as good (or better) potentials for SiO2 compared to this one from 1990... A lot of articles have been published since, and there are at least 10 (if not 30) other potentials in QuantumATK for Si+O, including the very popular Tersoff ones, or for glassy materials with partial charges you can try Pedone or some COMB potential.

[Priyadarshini]

"Greetings, as you mentioned, I have used the Tersoff potential to model my porous structure, where I aim to control its density by applying pressure. However, instead of reducing the density, I am observing an increase. What could be causing this issue, and could you suggest an effective workflow to achieve the desired lower density?"

[Anders Blom]

Tough question. Since there are many potentials for SiO2 you may want to play around with different ones. Perhaps having partial charges like in COMB will be helpful. Or not. One simple needs to run the simulations to find out.

[Priyadarshini]

Thanks again for the insights!
I now have a related question concerning energy convergence: How to resolve "Geometry optimization not converged. Stopped after 200 steps" during MD simulations of an amorphous bulk system?

I'm running a geometry optimization for a bulk material system in QuantumATK, but the calculation ends with the message:

"Geometry optimization not converged. Stopped after 200 steps."

I’d like to understand:

What are the main causes of this lack of convergence?

Should I increase the maximum number of optimization steps, or is this a sign of deeper issues like a poor initial structure or inappropriate force tolerance?

What are the recommended values for force tolerance, stress tolerance, and max step size in such cases?

Would refining the constraints block help?

Is it advisable to use a different optimizer (e.g., LBFGS vs. FIRE)?

Any tips, diagnostics, or example settings that worked for similar bulk systems would be appreciated

[Anders Blom]

So I guess you are actually not running MD, but a geometry optimization.

It's very difficult to advise without seeing the structure, log file, etc, and even with those it is something that probably requires investigation, and not just a simple fix.

Some systems do need much more than 200 steps - you mentioned it could be an amorphous system, so it can easily have a very "flat" minimum, and many degrees of freedom. Of course, a very low force tolerance will also require more steps, in particular if you also relax the cell size. It may be a good idea to try FIRE for an amorphous system; LBFGS is generally very fast when you are close to the minimum, but maybe FIRE is more robust for getting there.