We trained a machine learning interatomic potential to sample the binding energies of a water molecule on bare and ice covered (crystalline and amorphous) silicate and graphene dust grains. Three ice coverages - cluster, monolayer, and bilayer of water were investigated for the binding of a probe water molecule. 

Silicate grains showed high binding energies up to the monolayer water coverage regime due to the presence of surface magnesium atoms. In contrast, on graphene, interactions between water molecules dominate, while the surface itself plays a minimal role. 

The resulting binding energy distributions could be incorporated into astrochemical models, where they influence adsorption, diffusion, and desorption processes governing chemical evolution. 

See paper