Quantum-Accurate Machine Learning Potentials for Metal-Organic Frameworks using Temperature Driven Active Learning

TL;DR

This paper introduces a DFT-accurate machine learning potential for flexible MOFs, trained with an active-learning approach that significantly reduces computational costs while maintaining high accuracy for structural and vibrational properties.

Contribution

The work develops a spectral neighbor analysis potential for MOFs trained on DFT data and introduces an active-learning algorithm to efficiently reduce training data requirements.

Findings

Accurately models MOF properties at DFT level

Active learning reduces DFT calculations by mapping internal coordinates

Potential enables long-time simulations of flexible MOFs

Abstract

Understanding how structural flexibility affects the properties of metal-organic frameworks (MOFs) is crucial for the design of better MOFs for targeted applications. Flexible MOFs can be studied with molecular dynamics simulations, whose accuracy depends on the force-field used to describe the interatomic interactions. Density functional theory (DFT) and quantum-chemistry methods are highly accurate, but the computational overheads limit their use in long time-dependent simulations for large systems. In contrast, classical force fields usually struggle with the description of coordination bonds. In this work we develop a DFT-accurate machine-learning spectral neighbor analysis potential, trained on DFT energies, forces and stress tensors, for two representative MOFs, namely ZIF-8 and MOF-5. Their structural and vibrational properties are then studied as a function of temperature and tightly compared with available experimental data. Most importantly, we demonstrate an active-learning algorithm, based on mapping the relevant internal coordinates, which drastically reduces the number of training data to be computed at the DFT level. Thus, the workflow presented here appears as an efficient strategy for the study of flexible MOFs with DFT accuracy, but at a fraction of the DFT computational cost.