Discrepancy in PotEng Between lmp-otf MD and Single-Point OTF/DFT for Water Molecules

[ShuaishuaiYuan]

I am closely following this tutorial: link and performing a calculation on a supercell containing 58 H2O molecules. While tracking the flare sampling process, I observed a noticeable discrepancy between the potential energy (PotEng) values recorded in log.lammps and otf.gan.log. For example, why does the PotEng in the LAMMPS OTF MD simulation is around -640 eV, while the DFT and Sparse GP values are approximately -580 eV?

Issues:

• The single-point otf LAMMPS calculation matched well with DFT single-point calculation (also indicated by very small EMAE)
• But the PotEng energy in log.lammps is not the same as the single-point otf LAMMPS calculation

Details are below:
PotEng in ./log.lammps is -614.59888 eV

1940   1.94           275.34091     -614.59888     -608.44171      97727.866      0.0052391881   98891.122  

PotEng in ./gandft/1940_otf is -576.23253 eV (single-point LAMMPS otf calculation)

0   0              0             -576.23253     -576.23253      18993.738      0              0  

VASP energy in ./gandft/1940_DFT is -576.15148778 eV

-.81644036E+03  + 240.28887221999997 = -576.15148778

240.28887221999997 is the correction of H2O (H116O58)

>>> -1.11728959*116-1.90833241*58 
-240.28887221999997

Why is there a significant discrepancy between the PotEng values from lmp-otf MD and those from single-point OTF and DFT calculations? (This issue seems to occur only for water molecule calculations, not for bulk materials.)