Clarification on energy and force loss coefficients used for MACE-POLAR-1

[Alexandrina-Chen]

Hello, and thank you for releasing MACE-POLAR-1 and its training configurations.

I noticed a discrepancy between the preprint and the released reproduction scripts concerning the energy and force loss coefficients:

• Table 6 of the MACE-POLAR-1 preprint (https://arxiv.org/html/2602.19411#S5.SS3) lists an energy coefficient of 1 and a force coefficient of 10.

• The released medium (https://github.com/ACEsuit/mace-foundations/blob/main/mace_polar_1/mace-polar-1-medium.sh) and large (https://github.com/ACEsuit/mace-foundations/blob/main/mace_polar_1/mace-polar-1-large.sh) training scripts both specify:

--loss='l1l2energyforces'
--energy_weight=10.0
--forces_weight=10.0

The scripts state that their flags reproduce the distributed training runs.

Could you please clarify which energy and force coefficients were actually used to train the released MACE-POLAR-1-M and MACE-POLAR-1-L?

Thank you for your help.