Machine learning potential for modelling dynamic hydrogen bond networks in MOF MIL-120

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Metal-organic frameworks (MOFs) show promise for gas adsorption and separation.
• Accurate simulation of gas adsorption in MOFs using DFT methods is computationally challenging.
• MOFs often exhibit dynamic structural behavior influencing their adsorption properties.

Purpose of the Study:

• To develop a computational workflow for accurate simulation of gas adsorption in flexible MOFs.
• To investigate the interplay between CO2 adsorption and the dynamic pore environment in MIL-120.
• To establish a generalizable strategy for simulating adsorption in dynamic MOF systems.

Main Methods:

• Fine-tuning a pre-trained MACE potential for accurate interatomic interactions in MOFs.
• Developing a computational workflow integrating machine learning potentials with molecular simulations.
• Utilizing accelerated sampling with ML potentials to study CO2 adsorption dynamics.

Main Results:

• Created accurate machine-learning interatomic potentials for MIL-120, capturing its dynamic structural behavior.
• Uncovered a strong coupling between CO2 adsorption and the hydrogen-bond network on the MIL-120 pore surface.
• Observed that CO2 adsorption induces local rearrangements, reshaping the MOF's pore environment.

Conclusions:

• The developed computational workflow enables accurate simulation of gas adsorption in flexible MOFs.
• CO2 adsorption significantly impacts the dynamic hydrogen-bond network and pore structure of MIL-120.
• This strategy offers a generalizable approach for studying adsorption phenomena in other dynamic MOF materials.