Redefining tropical photovoltaics: an AI–DFT-driven physics-constrained framework linking electronic structure, climate response, and device-level performance in CsPbI3 perovskites
TL;DR
Overall, the proposed AI–DFT framework provides a physically interpretable computational approach for climate-aware prioritization of perovskite photovoltaic materials by connecting quantum-derived descriptors, ML predictions, environmental factors, sensitivity-based robustness assessment, and device-level performance estimation.
Abstract
Climate-induced temperature and humidity variations pose significant challenges to the development of stable and efficient perovskite photovoltaic materials suitable for tropical environments. This study presents a climate-aware artificial intelligence–density functional theory (AI–DFT) computational framework for the accelerated screening and optimization of caesium lead iodide (CsPbI3)-based perovskite absorbers by integrating first-principles calculations, machine learning (ML), atomistic stability analysis, and reduced-order device modelling. DFT calculations were employed to evaluate the structural, electronic, optical, thermodynamic, and defect-related characteristics of pristine and compositionally modified CsPbI3 systems. The optimized cubic CsPbI3 structure exhibits a lattice constant of 6.28 Å, a direct bandgap of 1.48 eV, a formation energy of −1.238 eV atom−1, and an energy above hull of 0.025 eV atom−1, indicating favourable characteristics relevant to photovoltaic applications. Finite-temperature ab initio molecular dynamics, phonon calculations, and reactive molecular dynamics simulations provide complementary insights into lattice behaviour and initial surface–water interaction mechanisms under the investigated conditions. The atomistic line graph neural network was trained and independently evaluated using a labelled dataset of 1024 structures, achieving a test mean absolute error of 0.0189 eV and an R2 value of 0.97 for bandgap prediction. A random forest model subsequently integrated intrinsic material descriptors with temperature, relative humidity, and solar irradiance to generate a climate suitability ranking for candidate materials under tropical operating conditions. Perturbation-based sensitivity analysis revealed that temperature (24.5%), defect formation energy (21.8%), and ion-migration barrier (18.6%) were among the dominant factors affecting the predicted suitability index. AI-guided screening identified Cs0.1FA0.9PbI3 and CsPbI2Br as promising absorber compositions with complementary efficiency–stability characteristics. A DFT-informed reduced-order photovoltaic model estimated device-level parameters, with predicted efficiencies of 21%–26% interpreted as physics-informed screening indicators rather than experimentally validated device efficiencies. Overall, the proposed AI–DFT framework provides a physically interpretable computational approach for climate-aware prioritization of perovskite photovoltaic materials by connecting quantum-derived descriptors, ML predictions, environmental factors, sensitivity-based robustness assessment, and device-level performance estimation. The framework is intended to guide future experimental validation, advanced device simulations, and climate-specific photovoltaic material development under tropical operating conditions.