Towards Out-of-Distribution Uncertainty Quantification for Machine-Learned Interatomic Potentials via Pre-Training

Informative uncertainty quantification (UQ) is essential for the practical deployment of machine-learned interatomic potentials (MLIPs), as it provides insight into their reliability across chemical space. Although numerous UQ architectures have recently been proposed, most have been trained in low-data regimes, restricting their applicability to narrow domains of interest. Inspired by the emergence of foundational MLIP, we propose pre-training uncertainty quantifiers on large, diverse datasets to enable broader applicability. We pre-train several architectures on large materials science datasets to quantify the uncertainty of different MLIPs, and evaluate their performance on both in-distribution and out-of-distribution adsorption-energy predictions. We find that pre-trained uncertainty quantifiers provide viable zero-shot uncertainty estimates across multiple datasets, while fine-tuning consistently yields positive correlations between predicted uncertainty and prediction error, even on the most challenging datasets. These results lay the groundwork for foundational UQ models for MLIPs, facilitating their deployment by providing a priori information about model reliability and applicability.