what energy dpdata reads in the VASP OUTCAR file

[chandagsin]

Dear Developers,

I practised with a dataset downloaded from the official website,
https://tutorials.deepmodeling.com/en/latest/Tutorials/DeePMD-kit/learnDoc/Handson-Tutorial%28v2.0.3%29.html

$ wget https://dp-public.oss-cn-beijing.aliyuncs.com/community/CH4.tar

This is a VASP OUTCAR file. The energy of each frame is given in this format (e.g.),

alpha Z PSCENC = 0.24272866
Ewald energy TEWEN = 126.94348121
-Hartree energ DENC = -282.92879547
-exchange EXHF = 0.00000000
-V(xc)+E(xc) XCENC = 24.66067699
PAW double counting = 456.76874710 -457.55018458
entropy T*S EENTRO = 0.01836410
eigenvalues EBANDS = -89.19239622
atomic energy EATOM = 197.04175453
Solvation Ediel_sol = 0.00000000

free energy TOTEN = -23.99562368 eV

energy without entropy = -24.01398778 energy(sigma->0) = -24.00174505

I found that dpdata reads the "TOTEN" energy, which includes the entropy term "entropy T*S EENTRO". From what I understand, AIMD does not need this term to solve the equations of molecular motions. (although sometimes this term is indeed zero) Not sure this would be a bug; just let you know.

Thanks,

chandagsin

[njzjz]

energy without entropy is not consistent with forces. You can see the previous discussion in ASE. To train a machine learning model, we hope that energies are consistent with forces.