Question Regarding Dataset Structure Training the DeepLearning Potentials

[WJiangH]

Summary

General Queries on Applying this Deep Learning Potential Framework to a Specific Domain.

DeePMD-kit Version

DeePMD-kit v2.2.4

TensorFlow Version

2.14.0

Python Version, CUDA Version, GCC Version, LAMMPS Version, etc

No response

Details

Dear Authors,

Firstly, I'd like to commend you on the fantastic work presented here. I'm exploring the potential of integrating this deep learning framework into my research on the defective properties of HEA.

I've reviewed your paper, "End-to-end Symmetry Preserving Inter-atomic Potential Energy Model for Finite and Extended Systems (2018)", and noticed the fitting results for HEA data. I observed that the dataset consists of defect-free systems (3x3x5) with 180 atoms.

I have two questions and they may be generic to other applications.
First: Can I clarify if all the frames in this dataset were gathered from the relaxation processes of these structures?
Second: In terms of preparing the dataset, can the model trained on these defect-free structures be directly applied to defective structures, such as those containing vacancies or interstitials? Or would there be a need for preparing datasets with these specific defective structures?

Thank you for your insights.
JJ