Combining Machine-Learning and Classical Potentials for Grand Canonical Monte Carlo Simulation of Chemisorption

Binquan Luan,Carine Ribeiro dos Santos dos Santos,Bowen Zheng,Felipe Lopes Oliveira,Rodrigo Neumann Barros Ferreira,Mathias Steiner,Grace X. Gu

DOI: https://doi.org/10.26434/chemrxiv-2024-zdw0l

2024-10-14

Abstract:Chemisorption, characterized by the strong and highly selective chemical bonding formation between a gas molecule and a solid sorbent, plays a critical role in various applications such as surface functionalization, carbon capture, and hydrogen storage. Sorption processes are typically modeled by means of Grand Canonical Monte Carlo (GCMC) simulation. However, GCMC methods are limited to physisorption processes described by molecular-mechanics, or, classical, interatomic potentials. Therefore, a computationally efficient GCMC method that accounts for chemisorption is needed. Here, we report hybrid interatomic potentials that are specifically designed for GCMC simulations of chemisorption. Namely, the sorbate-sorbate interaction is described by classical potentials and the sorbate-sorbent interaction is rendered by a machine-learning interatomic potential (MLIP) trained on simulation data extracted from quantum calculations. To prove the principal, we have applied the method to the simulation of \ce{CO2} chemisorption in the porous metal-organic framework Mg-MOF-74. By comparing the GCMC simulated \ce{CO2} adsorption isotherms and isobars, respectively, with the experimental data, we obtain better agreement compared to other classical and quantum-chemical simulation methods while keeping computational cost at a minimum. With the rise of transferable MLIP, we expect that the method could enable computationally efficient and highly accurate sorption studies in reticular materials.

Chemistry

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Develop an efficient and accurate Grand Canonical Monte Carlo (GCMC) simulation method to simulate the chemisorption process**. Specifically, this research aims to overcome the limitations of existing GCMC methods in describing chemisorption.

### Problem Background:

1. **Importance of Chemisorption**:
- Chemisorption involves the formation of strong and highly selective chemical bonds between gas molecules and solid adsorbents, and is widely used in fields such as surface functionalization, carbon capture, and hydrogen storage.

2. **Limitations of Existing GCMC Methods**:
- Current GCMC methods mainly rely on classical molecular mechanics or classical interaction potentials (such as UFF, DREIDING, etc.). These potential functions are suitable for describing physisorption, but are insufficient in describing chemisorption.
- Although first - principles methods (such as density functional theory, DFT) can accurately describe chemisorption, the computational cost is too high to be applied in large - scale GCMC simulations.

3. **Requirement**:
- A GCMC method that can both maintain computational efficiency and accurately describe chemisorption is needed.

### Solution:

- **Hybrid Machine Learning/Molecular Mechanics (ML/MM) Method**:
- For the interaction between sorbate - sorbent, use the machine - learning interaction potential (MLIP), which is trained with quantum - computing data.
- For the interaction between sorbate - sorbate, continue to use the classical molecular force field (such as TraPPE).

### Application Verification:

- **Simulation of Chemisorption of CO₂ in Mg - MOF - 74**:
- The researchers applied this method to the simulation of chemisorption of CO₂ in the porous metal - organic framework Mg - MOF - 74.
- By comparing with experimental data, it was found that this method can more accurately predict the adsorption isotherm and isobar of CO₂ while maintaining a lower computational cost.

### Conclusion:

- This research proposes a new GCMC simulation method that can accurately describe the chemisorption process under the premise of efficient computation, providing a powerful tool for future material screening and optimization.