Quantum informed machine-learning potentials for modeling CO₂ adsorption in metal organic frameworks

Abstract

Global warming caused by excessive emission of greenhouse gases $ (mainly \ {CO_2)} $ into atmosphere imposes profound changes in environment. For curbing the global temperature increase, effective approaches for carbon capture are needed. As porous sorbents, metal organic frameworks (MOFs) are promising candidate-materials that potentially combine high $ {CO_2} $ uptake and $ {CO_2/N_2} $ selectivity. However, it is still challenging to computationally identify the best suited species within the hundreds of thousands of MOF structures known today. First-principles-based simulations of $ {CO_2} $ adsorption in MOFs would provide the necessary accuracy, however, they are impractical for screening purpose due to the high computational cost. Classical-force-field based simulations would be computationally feasible, however, they do not provide sufficient accuracy. Here, we report the quantum-informed machine-learning force fields (QMLFF) for atomistic simulations of $ {CO_2} $ in MOFs. We demonstrate that the method has a much higher computational efficiency (~1000 times) than first-principles one while maintaining quantum-level accuracy. As a proof of principle, we show that the QMLFF-based atomistic simulations can yield various physical quantities comparable to experimental results. The combination of machine learning and atomistic simulation paves the way for modeling $ {CO_2} $ capture by MOFs both accurately and efficiently.