Adsorption Isotherm Prediction of Diverse Adsorbates in Metal–Organic Frameworks using Machine Learning

This thesis addresses challenges in grand canonical Monte Carlo (GCMC) simulations for metal–organic frameworks (MOFs) by employing machine learning (ML) techniques in two distinct approaches. In Part I, we utilize the rapid screening capabilities of ML to assess a variety of MOFs across different adsorbates, focusing on calculating Henry’s constants to describe the interactions between MOFs and adsorbates. A novel scheme was developed to allow a single ML model to process multiple MOFs and molecules, enhancing dataset diversity via active learning and uncertainty quantification methods, thus breaking new ground in MOF research. In Part II, ML is utilized to replicate ab initio calculations, focusing on CO2 and H2O for direct air capture (DAC) applications. We introduced the Open DAC 2023 dataset, comprising over 35 million density functional theory calculations, to train ML models that predict adsorption energies. These models are integrated into GCMC simulations via the GraphCMC toolkit, which aims to explore the potential for achieving accurate adsorption isotherms and identifying optimal MOF sorbents for DAC. This dual application of ML demonstrates its potential to transcend traditional limitations in MOF adsorption and separation studies, paving the way for novel discoveries in the field.

Date

2025-01-13

Resource Type

Text

Resource Subtype

Dissertation

Full item page