Fig. 1: Potential energy surface softening in uMLIPs.
From: Systematic softening in universal machine learning interatomic potentials
Left: schematic representation of the potential energy surface (PES) described in density functional theory (DFT), with two arbitrary coordinate axes. Right: PES described by universal machine learning interatomic potentials (uMLIPs), which well describes the PES regions sampled by near-equilibrium states in the pre-training dataset (orange), but experience larger errors in high-energy regions (red) with underprediction of energies and forces. The softening behavior is largely systematic in a given chemical space and, can therefore be efficiently fixed locally with a small amount of data augmentation. Using a linear correction, we demonstrate the data efficiency of uMLIP fine-tuning.
