Initial datasets for dpgen

[XiaMinqi]

I'm going to use DPMD to investigate a water-graphite interface system, therefore I prepared structures of water, bulk carbon and water-carbon interface as training set.
As the lattice parameters for these 3 kinds of structures are not the same, using the same k-points set may introduce some error for dp training, I suppose.
So in order to keep the results at the same level of precision, is it right to control the kspacing and ENCUT as the same for different structures in training set?

[njzjz]

Actually, different elements will not share the same neural networks.

[XiaMinqi]

I'm afraid I didn't comprehend your reply entirely. Do you mean there's no need to keep parameters for different structures in training set as the same value?

[njzjz]

You may need to keep the parameters the same, but this rule does not affect different atomic types. Of course, you don't need to make C and O the same.

[XiaMinqi]

Oh I see, thank you for your patience!