Free energy calculations using deepmd model and fix adapt/fep command

[dkussainova]

Hi,

I am trying to perform free energy calculations of ions in water using the deepmd model and 'fix adapt/fep N pair deepmd scale' command in lammps. However, looks like all interactions are scaled even though in fix adapt/fep function I am choosing only atom types related to ions. For example, when I try to scale interactions of ions by 0, I am getting 0 potential energy, even though I did not include atom types of water in fix adapt/fep. I tried also to include interactions of water in fix adapt/fep with a scaling factor of 1, however, this didn't help. Do you know what can be a reason for this? Any help is appreciated.

Thanks,
Dina

[wanghan-iapcm]

DP is a multi-body potential, so it is not easy to separate and to switch-on/off a certain pair of interaction.

What I would suggest for your case is training two DPs one including ion-water interaction and one only for water, and you may use FEP to calculate the free energy between the two DPs.

Please note that always set the type of H and O to 0 and 1, and any ion type to >= 2. Running lammps with the water-only model on a water-ion configuration will ignore the types >= 2.

The exclude_type may also work. Assume we have one ion type 2. One may set exclude_types to [[0, 2], [1, 2], [2, 2]], and set the atom_ener to [0, 0, 0] and switch on set_davg_zero. @njzjz plz check if I am correct.

[dkussainova]

Thank you a lot for your reply! I am gonna try this and see how it goes.

[njzjz]

One may set exclude_types to [[0, 2], [1, 2], [2, 2]], and set the atom_ener to [0, 0, 0] and switch on set_davg_zero.

If I understand correctly, the water-only model should set the atom_ener to [null, null, 0]. Then only ion atom energy will be zero.

[wanghan-iapcm]

Thanks!

[dingye18]

You can also refer the alchemical protocol used in OpenMM Deepmd Plugin.
We implement the alchemical protocol within this OpenMM plugin. DP 2. version model can be integrate with this plugin directly.