Bayesian Optimization for Efficient Prediction of Gas Uptake in Nanoporous Materials

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Nanoporous materials (NPMs), including Metal-Organic Frameworks (MOFs) and Covalent Organic Frameworks (COFs), are vital for addressing global challenges.
• Traditional experimental methods for NPM optimization are slow and costly.
• Efficient strategies are needed to accelerate the discovery and design of NPMs for applications like gas storage.

Purpose of the Study:

• To present a Bayesian optimization (BO) strategy for efficient navigation of NPM design spaces.
• To quantitatively evaluate various surrogate model and acquisition function combinations for BO in MOF optimization.
• To identify optimal machine learning models for use as surrogates in BO for materials discovery.

Main Methods:

• Employed machine learning (ML) techniques for regression analysis on a MOF dataset.
• Evaluated multiple ML models, selecting the top three accurate models as surrogates.
• Utilized Gaussian Process (GP) as a surrogate model with Expected Improvement (EI) as the acquisition function.

Main Results:

• Gaussian Process (GP) with Expected Improvement (EI) and without a gamma prior outperformed other surrogate models.
• The best performing ML model for prediction may not be the most suitable surrogate for BO.
• BO significantly reduces reliance on computationally intensive methods like GCMC and DFT.

Conclusions:

• Bayesian optimization provides an efficient framework for accelerating the discovery and optimization of NPMs.
• The choice of surrogate model and acquisition function is critical for effective BO in materials science.
• This approach enhances computational efficiency, aiding in energy storage and environmental sustainability efforts.