Carbon capture technology is crucial to reduce carbon dioxide emissions. Previous reports of nanoporous materials, including zeolites and metal organic frameworks (MOFs), have proven their excellent adsorption capabilities for carbon capture applications. A fundamental property that is closely related to their performance is heat capacity. However, heat capacity of nanoporous materials, especially of MOFs, is often assumed to be constant across all materials. This has led to an overestimation of the total energy requirement for the carbon capture process, which ultimately has resulted in an unreliable prediction of the carbon capture performance. Major limiting factors are a poor understanding of the connection between heat capacity and crystal structure, and the absence of a methodology to evaluate the heat capacity of the various existing nanoporous materials. To tackle these issues, Seyed Mohamad Moosavi, Berend Smit, and colleagues recently developed a machine learning framework — trained on a dataset produced using density functional theory (DFT) calculations — that can accurately predict the heat capacity of nanoporous materials.
The authors first analyzed the heat capacity of representative nanoporous materials using DFT calculations. The results confirmed that, while the change in topology has a minor effect on heat capacity, the local chemical environment is important in determining this property. Taking advantage of this, the authors generated a training set composed of representative MOFs with a relatively small number of atoms by DFT calculations. The chemical environment of each atom was described using a feature vector, including elemental properties, chemical descriptors, and geometry descriptors. Then, a gradient-boosted decision tree machine learning model was developed to predict heat capacity from the feature vectors with high accuracy: the relative mean absolute error of the predictions was at 2.89%. One of the key advantages of this methodology is that the framework can be readily updated to consider a new structure with different chemical environments by performing simple additional DFT calculations of a small subset of materials. While the total energy requirement for a carbon capture process also depends on other factors, such as the process design, the presented method has the potential to improve the design of porous materials for carbon capture and other potential applications.
This is a preview of subscription content, access via your institution