Experimental investigation of first-order hyperpolarizability dispersion leveraging language models for molecular discovery

• Ano de defesa: 2024
• Autor(a) principal: Sciuti, Lucas Fiocco
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Chalcone molecules; First-order hyperpolarizability; Hiperpolarizabilidade de primeira ordem; Large language model; Modelo grande de linguagem; Moléculas chalconas; Nonlinear optics; Óptica não linear
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/76/76132/tde-04102024-083530/

First-order hyperpolarizability is a molecular property of great importance for the development of sensors, optical switches, and second-order generation applications. Due to electronic effects, its dispersion makes it crucial to evaluate first-order hyperpolarizability as a function of the wavelength of the incident electric field. In this thesis, we studied such effects in chalcone- derived molecules. Initially, we demonstrated the experimental setup for measuring the dispersion of first-order hyperpolarizability using wavelength-tunable Hyper-Rayleigh scattering technique. For validation, we measured the dispersion of first-order hyperpolarizability in pNA molecules. Through one- and two-photon absorption measurements combined with Hyper-Rayleigh scattering results in chalcone-based molecules, we showed that the red-shift of electronic absorption bands and the presence of multiple absorption bands influence and enhance the nonlinear scattering effect. Finally, we applied a language modeling machine learning technique to create novel molecules biased by the difference between HOMO and LUMO. We demonstrated that 100% of the created molecules were unseen during training, with 96% of them valid for laboratory synthesis. We hope that this model will accelerate and support experimental development in discovering new molecules for use in the mentioned applications.