QuantumATK Forum

General => Future Releases => Topic started by: Lim changmin on November 3, 2025, 06:32

Title: Machine Learning potential that can also predict density of state
Post by: Lim changmin on November 3, 2025, 06:32

I understand that machine learning potentials have many advantages and are receiving a lot of attention. However, in the current version of ATK, the provided potentials can only compute force, stress, and energy. Meanwhile, several recent studies have shown the possibility of predicting the density of states (DOS) using machine learning. Therefore, it would be great if ATK could be updated to include ML potentials capable of calculating DOS as well.

Title: Re: Machine Learning potential that can also predict density of state
Post by: Anders Blom on November 4, 2025, 20:39

I have yet to see a predefined ML model (we should not call them potentials) that can accurately predict DOS across many applications, but training a specific one for a given problem space is something we are working on.