Ab initio predictions of adsorption in flexible metal-organic frameworks for water harvesting applications.

Veronique Van Speybroeck,Ruben Goeminne

DOI: https://doi.org/10.26434/chemrxiv-2024-lpm8c

2024-11-20

Abstract:Recently, metal-organic frameworks like MOF-303 and MOF-LA2-1 have demonstrated exceptional performance for water harvesting applications. To enable a reticular design of such materials, an accurate prediction of the adsorption properties with chemical accuracy and fully accounting for the flexibility is crucial. The computational prediction of water adsorption properties of metal-organic frameworks (MOFs) has become standard practice. However their predictive power to design new materials is hindered by the limited accuracy of the used interatomic potential and the limitations on how to account for the framework flexibility. In this work, we showcase a methodology to obtain chemically accurate adsorption isotherms that fully account for the framework flexibility. The method is founded on very accurate and efficiently trained machine learning potentials and transition matrix Monte Carlo simulations to account for framework flexibility. By first benchmarking the reference electronic level of theory used for the training, quantitatively accurate adsorption isotherms are obtained for MOF-303, a highly topical MOF being investigated for its potential use in water harvesting applications. We show that both an accurate level of theory and a proper inclusion of local and global framework flexibility is vital in the prediction of the adsorption properties of MOF-303. The broader applicability of our methodology is demonstrated through the study of related linker-exchanged materials, MOF-333 and MOF-LA2-1. Analyses of the density profiles of water adsorbed in these MOFs yields deeper insight into the origins and differences of the observed isotherms. An optimal water harvester should have initial seeding sites with intermediate adsorption strength, to prevent detrimental low-pressure water uptake. To increase the working capacity, linker extension strategies can be used while maintaining the initial seeding sites, as was done in the MOF-LA2-1. The proposed methodology is applicable to other guest molecules and MOFs, paving the way to future rational design of MOFs with specific adsorption properties for the application at hand.

Chemistry

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the accuracy of water adsorption prediction in flexible metal - organic frameworks (MOFs). Specifically, the existing prediction methods have limitations in describing the interaction between water and MOF and considering the framework flexibility, which leads to a large deviation between the predicted water adsorption isotherms and the experimental results. The paper proposes a new method, aiming to overcome these limitations by using machine - learning potential (MLP) and transition - matrix Monte Carlo (TMMC) simulations, so as to achieve chemical - accuracy prediction of MOF water - adsorption properties and comprehensively consider the framework flexibility.

### Main problems solved:
1. **Improve prediction accuracy**: Due to the limited inter - atomic potential used and the insufficient handling of framework flexibility, the existing methods result in inaccurate predicted water - adsorption isotherms. The paper improves the prediction accuracy by using high - precision machine - learning potential and TMMC simulations.
2. **Comprehensively consider framework flexibility**: The traditional rigid - framework assumption is not applicable in some MOFs because the adsorption of water molecules will cause changes in the framework structure. The method in the paper can comprehensively consider the local and global flexibility of the framework, thus predicting the adsorption behavior more accurately.
3. **Optimize the design of water - adsorption materials**: By providing more accurate adsorption isotherm predictions, this method can help researchers design MOFs with specific adsorption properties, especially materials for water - collection applications.

### Method overview:
- **Machine - learning potential (MLP)**: Train MLP through an efficient active - learning process to represent the full flexibility of the host framework in the presence of guest adsorbents, which requires thousands of DFT reference calculations.
- **Transition - matrix Monte Carlo (TMMC) simulations**: Combine molecular - dynamics (MD) simulations of the NPT ensemble, including the framework flexibility when adsorbing guest species. Evaluate the adsorption characteristics through snapshots generated by MD simulations, thereby generating reliable adsorption isotherms.

### Application examples:
- **MOF - 303**: The water - adsorption isotherm predicted by this method is quantitatively consistent with the experimental results, verifying the effectiveness of the method.
- **MOF - 333 and MOF - LA2 - 1**: Further apply this method to predict the adsorption characteristics of other relevant link - exchange materials, revealing the origin and differences in the adsorption behaviors of different MOFs.

In conclusion, this paper solves the limitations of existing methods in predicting the water - adsorption properties of flexible MOFs by proposing a new computational method, providing a powerful tool for designing MOFs with specific adsorption properties in the future.