"atomic energy" are not same as I defined in input?

[jtyang-chem]

I did a train with atom_ener specified, then I freeze and got a model, but the strange thing is, if I make an AB 2 atom system and let the distance be like 20.0 (out of cut-range), the Ea, Eb I got from the model are not equal to the values I specified, and total energy is Ea+Eb but not what I expect.

Why?

[njzjz]

Did you set descriptors' set_davg_zero to true?

[jtyang-chem]

Even though, how to explain the big difference?

[njzjz]

avg and std are computed based on training data. They are not fixed values.

[jtyang-chem]

Sure, they are not, but for calculation of no-neighbor system, the avg and std are still from the training with-neighbor training data?

[njzjz]

They are average of all your data. Do you only provide no-neighbor system?

[jtyang-chem]

No. That's explained the difference, thx for your patience.