Language Model-Assisted\nMachine Learning, Photoelectrochemical,\nand First-Principles Investigation of Compatible Solvents for a CH<sub>3</sub>NH<sub>3</sub>PbI<sub>3</sub> Film in Water

Machine learning and data-driven methods have attracted\na significant\namount of attention for the acceleration of the design of molecules\nand materials. In this study, a material design protocol based on\nmultimode modeling that combines literature modeling, numerical data\ncollection, textual descriptor design, genetic modeling, experimental\nvalidation, first-principles calculation, and theoretical efficiency\ncalculation is proposed, with a case study on designing compatible\ncomplex solvent molecules for a halide perovskite film, which is notorious\nfor optoelectronic deactivation under hostile conditions, especially\nin water. In the multimode modeling design process, the textual descriptors\nplay the central role and store rich literature scientific knowledge,\nwhich starts from the construction of a high-dimension literature\nmodel based on scientific articles and is realized by a genetic algorithm\nfor materials predictions. The prediction is substantiated by follow-up\nexperiments and first-principles calculations, leading to the successful\nidentification of effective molecular combinations delivering an unprecedented\nlarge aqueous photocurrent (increasing by 3 orders of magnitude compared\nwith that of CH<sub>3</sub>NH<sub>3</sub>PbI<sub>3</sub>) and remarkable\naqueous stability (improving from 36% to 89% after immersion in water)\nunder the hostile condition. This study provides a practical route\nvia multimode modeling for accelerating the design of molecule-modified\nand solution-processed materials in a real scenario.