The Generative Inverse Design of High-Performance Porous Carbons for CO2 Capture

A Deep Generative Framework for Tailored Carbon Sorbents

Abstract

The discovery of novel materials with tailored properties is essential for technological progress, yet traditional trial-and-error approaches remain slow and resource-intensive. Inverse design offers a transformative paradigm by identifying structures that meet predefined performance targets. In this work, we present a deep generative framework for the inverse design of porous carbons optimized for CO2 capture. Using a database of over 20,000 virtual carbon structures, we trained a 3D convolutional Generative Adversarial Network (GAN) capable of learning a compact and continuous representation of the structure-property landscape. A surrogate predictive model based on gradient boosting enables efficient latent-space optimization, guiding the generator toward high-performance morphologies. The resulting AI-designed material, CG-005, features layered graphene-like sheets forming uniform slit-micropores centered at 0.6 nm. Molecular simulations confirm a CO? adsorption capacity of 6.2 mmol/g at 298 K and 1 bar, outperforming benchmark sorbents such as Zeolite 13X and MOF-177. This study demonstrates the power of deep generative models in accelerating the discovery of next-generation, high-performance carbon sorbents for environmental and energy applications.