Exploring Foundational Machine Learned Potentials for Treating the High Temperature Dynamics of Metal‐Organic Frameworks

Abstract Metal‐organic framework (MOF) derived materials, formed through high temperature processes, show great potential as catalysts, but structure‐property relationships between initial MOFs and final catalysts remain poorly understood due to characterization challenges. Current simulation methods like ab initio molecular dynamics are too computationally intensive for pyrolysis studies. Neural network approaches for learning interatomic potentials from density functional theory (DFT) offer a solution. Pyrolysis of CALF‐20 and ZIF‐8 is explored using machine learned potentials (MLPs) that can simulate high‐temperature decomposition with near‐DFT accuracy. Random sampling and two biased sampling techniques are tested to sample zinc coordination number and bond length phase space. Biased methods significantly outperformed random sampling, successfully recreating DFT simulation environments. Using this model, nanosecond‐scale quenches of both MOFs are simulated at high temperatures, revealing atomistic details of gas formation, zinc coordination changes, and linker decomposition. This demonstrates the potential for MLPs simulating complex high‐temperature MOF processes to understand reactivity and predict features for new catalytic materials.