Enabling Machine Learning-Assisted Discovery of Polyamines for Solid-State CO₂ Capture

Necessitated by the impact of global climate change, the efficient direct air capture (DAC) of CO₂ is one of the technologies with the potential to contribute to the goal of net zero emissions. Solid amine adsorption has shown most promise among existing approaches due to its energy efficiency and scalability. To estimate adsorption for these polyamines, we introduce a computational framework that combines fragment-based polymer generation with Density Functional Theory, molecular dynamics relaxations, and grand canonical Monte Carlo sampling. Through this efficient workflow, we generated a large library of polymers with absorption data — potentially supporting machine learning models for inverse design of polymers with optimized CO₂ adsorption. Experimental results showed that Bayesian optimization can further accelerate the process by efficiently identifying high-performing candidates. In summary, our integrated approach bridges atomistic simulation with data-driven optimization, enabling accelerated screening of polymer sorbents for DAC applications.