Machine learned force-fields for an Ab-initio quality description of metal-organic frameworks

Abstract Metal-organic frameworks (MOFs) are an incredibly diverse group of highly porous hybrid materials, which are interesting for a wide range of possible applications. For a meaningful theoretical description of many of their properties accurate and computationally highly efficient methods are in high demand. These would avoid compromises regarding either the quality of modelling results or the level of complexity of the calculated properties. With the advent of machine learning approaches, it is now possible to generate such approaches with relatively little human effort. Here, we build on existing types of machine-learned force fields belonging to the moment-tensor and kernel-based potential families to develop a recipe for their efficient parametrization. This yields exceptionally accurate and computationally highly efficient force fields. The parametrization relies on reference configurations generated during molecular dynamics based, active learning runs. The performance of the potentials is benchmarked for a representative selection of commonly studied MOFs revealing a close to DFT accuracy in predicting forces and structural parameters for a set of validation structures. The same applies to elastic constants and phonon band structures. Additionally, for MOF-5 the thermal conductivity is obtained with full quantitative agreement to single-crystal experiments. All this is possible while maintaining a very high degree of computational efficiency. The exceptional accuracy of the parameterized force field potentials combined with their computational efficiency has the potential of lifting the computational modelling of MOFs to the next level.