Machine Learning-Guided Rational Design of Dual-Site Catalysts for Synergistic CO2-to-CO Photoreduction and Antibiotic Degradation

Abstract

Rising atmospheric CO2 levels and pervasive pharmaceutical wastewater contamination are two pressing global environmental challenges. Integrating CO2 reduction with the organic pollutant degradation in a single photocatalytic system is an attractive strategy for collaborative environmental remediation. Here, we combine an intermediate spillover strategy with machine learning (ML) to rationally construct a dual-site photocatalyst, comprising Au nanoclusters and copper single atoms coanchored on graphitic carbon nitride (denoted CuSAAuC/CN). In-situ spectroscopy and theoretical calculations reveal that oxophilic CuSA sites act as Lewis acid centers that facilitate CO2 adsorption and activation, substantially lowering the energy barrier for *COOH formation. Subsequently, the key *COOH intermediate migrates to adjacent Au nanoparticles, where CO desorption readily occurs. Meanwhile, photogenerated holes effectively degrade tetracycline-type antibiotics in wastewater, achieving synergistic CO2 reduction and pollutant detoxification. Under visible-light irradiation, the catalyst exhibits near-100% CO selectivity for CO2 photoreduction along with excellent antibiotic degradation efficiency. Techno-economic analysis and tests using simulated industrial flue gas further confirm the system’s outstanding stability and its practical potential for cotreating industrial flue gas and pharmaceutical wastewater. This study presents a machine learning-guided design strategy for high-performance dual-function photocatalysts that simultaneously address carbon mitigation and water remediation, offering a feasible route toward practical environmental governance.