Abstract

Direct air capture (DAC) using solid sorbents is limited by energy demand and productivity, which depend strongly on adsorbent properties. Cooperative adsorbents with step-shaped isotherms are promising for capturing CO2 from ultra-dilute air, but the link between isotherm characteristics and process performance remains unclear. This work couples a validated packed-bed temperature–vacuum swing adsorption (TVSA) model with an artificial neural network (ANN) surrogate to explore the multi-dimensional design space of cooperative sorbents, using mmen–Mg2(dobpdc) as a benchmark. The surrogate model is trained on 2866 simulation data points to enable efficient sensitivity analysis and optimisation.

Results reveal a clear hierarchy of control: step pressure (Pstep0), step enthalpy (ΔHstep), and high-loading enthalpy (ΔHHigh) dominate specific energy consumption (SEC), productivity, and purity, whereas kinetics show minor influence under the studied regime. Significant second-order interactions were identified between Pstep0 and ΔHstep, highlighting the coupled role of step position and energetics. Surrogate-assisted design exploration identifies an optimal design region, characterised by low step pressure, moderate step enthalpy, and high adsorption capacity. A representative optimal isotherm improves CO2 recovery from 53% to 76% and increases productivity by approximately 38%, while reducing energy consumption to 2.9 MJ/kg CO2 at ∼98% purity. These improvements arise from aligning the adsorption step with the DAC operating window while maintaining favourable adsorption energetics. Importantly, these findings establish process-informed design principles, enabling targeted tuning of material properties for improved DAC performance.