energy deviation in LAMMPS simulation #1190
-
|
I have noticed that in the new v2 version the energy deviation from lammps output file is gone. It was replaced by virial deviation which is undoubtedly useful, but so is the energy deviation. Is there some specific reason for this? I was not able to find any information on this in the respective pull request #594. |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
|
The energy deviation is not as accurate as force deviation or viral deviation -- that's the reason why it's deprecated. |
Beta Was this translation helpful? Give feedback.
-
|
There is an arbitrariness in the energy decomposition. For example, we have molecule AB in the system, adding an arbitrary constant to E_A and subtract the same constant from E_B does not change the total energy of the system. Therefore, the model deviation of atomic energy is not a good indication for the quality of the model. The atomic energy deviation is printed in deepmd-kit v1.x, but it is not intended to be used in any case. That's why it is removed in v2. |
Beta Was this translation helpful? Give feedback.
-
|
Ok, the reasoning is clear. That is what I have suspected. But I could have been mentioned in the docs to avoid confusion. |
Beta Was this translation helpful? Give feedback.
There is an arbitrariness in the energy decomposition. For example, we have molecule AB in the system, adding an arbitrary constant to E_A and subtract the same constant from E_B does not change the total energy of the system. Therefore, the model deviation of atomic energy is not a good indication for the quality of the model. The atomic energy deviation is printed in deepmd-kit v1.x, but it is not intended to be used in any case. That's why it is removed in v2.