Quantum-accurate machine learning potentials for metal-organic frameworks using temperature driven active learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Accurate simulation of metal-organic frameworks (MOFs) is vital for designing advanced materials.
• Molecular dynamics (MD) simulations are essential for understanding MOF structural flexibility.
• Current methods like Density Functional Theory (DFT) are computationally expensive, while classical force fields lack accuracy for coordination bonds.

Purpose of the Study:

• To develop a computationally efficient yet accurate method for simulating the structural flexibility of MOFs.
• To create machine-learning potentials that achieve DFT accuracy for MD simulations of MOFs.
• To reduce the computational burden of simulating MOFs, facilitating the design of new materials.

Main Methods:

• Development of DFT-accurate machine-learning spectral neighbor analysis potentials for two representative MOFs.
• Utilizing an active-learning algorithm based on mapping relevant internal coordinates to minimize DFT training data.
• Studying structural and vibrational properties of MOFs and comparing them with experimental data.

Main Results:

• Successfully developed machine-learning potentials that reproduce DFT accuracy for MOF simulations.
• Demonstrated significant reduction in required DFT computations through an active-learning strategy.
• Validated the accuracy of the developed potentials by comparing simulated structural and vibrational properties with experimental data.

Conclusions:

• The presented workflow offers an efficient strategy for studying flexible MOFs with DFT accuracy.
• The active-learning approach drastically reduces the computational cost associated with DFT calculations.
• This method enables the study of flexible MOFs at a fraction of the standard DFT computational expense, paving the way for improved MOF design.