Machine learning and first-principles insights on molecularly modified CH3NH3PbI3 film in water

The molecular approach is effective to improve the optoelectronic properties of the halide perovskite films in ambient and hostile conditions while understanding the interactions between molecules and the halide perovskite surfaces is critical. In this manuscript, we explore the molecularly modified halide perovskite films via the combination of machine learning, photoelectrochemical experiments and first-principles calculations. The optoelectronic properties of the molecularly engineered halide perovskite surfaces in the aqueous solution are comprehensively evaluated with a case study on the prototypical CH3NH3PbI3 film modified by a D-pi-A molecule ethyl red. The accurate machine learning model with an accuracy of 96% is constructed via the ExtraTrees algorithm to classify the photocurrents produced by the ethyl red/perovskite surface systems in water. Chemical insights are obtained by analyzing the molecular descriptors and the experimental features, suggesting the importance of the molecular concentration as well as the molecular shapes and chemical compositions of the solvent environments for the perovskite optoelectronic properties. The first-principles calculations reveal the formation of O center dot center dot center dot Pb bonds and additional octahedrons at the perovskite surface in the presence of the molecular surface modifier and the water molecule. The intricate influences and the double-edged sword effects of the surface molecular modifier and the water molecule on the optoelectronic properties of the halide perovskite surface are further elucidated. The present study demonstrates an effective workflow combining machine learning, high-throughput experiments and first-principles calculations to comprehensively evaluate the halide perovskite films and can be elaborated to other molecularly modified surface systems in complex environments via multiple research paradigms.