Deep learning-assisted double strong coupling between multi-order anapoles and excitons

High-order anapoles present strongly concentrated energy and narrow resonant widths and thus become burgeoning candidates in nonlinear optical conversion and strong coupling fields. Nevertheless, existing studies, fully depending on intricate physical principles and computational methods, are rarely focused on strong coupling simultaneously involving multi-order anapoles. Herein, we theoretically construct a neural network via deep learning (DL) and design a three-layer stacked hybrid system with a Si nanodisk sandwiched by the MoSe2 and MoTe2 nanodisks with the same radius. First-order and second-order anapoles are excited simultaneously in this system and couple with excitons of bulk MoTe2 and MoSe2 nanodisks, respectively. Double strong coupling behaviors are achieved in between first-order anapole and exciton of MoTe2 as well as second-order anapole and exciton of MoSe2, resulting in four energy branches with large Rabi splitting of 100.6 meV and 118.2 meV, respectively. Our work provides an effective approach for light-matter interaction involving multi-order anapoles and multiple excitons and demonstrates the feasibility of designing complex structures based on DL algorithms.