Comparing transferability in neural network approaches and linear models for machine-learning interaction potentials


Atomic simulations using machine learning interatomic potential (MLIP) have gained a lot of popularity owing to their accuracy in comparison to conventional empirical potentials. However, the transferability of MLIP to systems outside the training-set poses a significant challenge. Here, we compare the transferability of two MLIP approaches: i) Neural Network Potentials (NNP) and ii) Physical LassoLars Interactions Potential (PLIP), trained over a small but diverse configuration of zinc oxide polymorphs. We obtained good agreement between both MLIPs and density functional theory reference results for physical properties including bulk lattice parameters, surface energies, and vibrational density of states. However, the NNP model performed poorly when compared to the linear PLIP model for the structural optimization of nanoparticles and molecular dynamics simulation of liquid phases, which are systems outside the training-set. While generally providing less accurate results, the studied linear model appears more reliable than NNP when it comes to transferability in this specific example