Machine learning approach to reconstruct dephasing time from solid HHG spectra

The dephasing time T₂ is a fundamental parameter that characterizes the coherence of electronic states and electron–phonon interactions in condensed matter physics. Accurate measurement of T₂ is essential for elucidating ultrafast electronic and phononic processes, which are crucial for the development of advanced electronic, optoelectronic, and quantum devices. However, due to the complexity of solid-state systems with their intricate band structures and strong many-body interactions, reconstructing T₂ remains a long-term challenge for both condensed matter physics and optical science. In this work, we introduce a machine learning (ML) approach to retrieve T₂ from the high-order harmonic generation (HHG) spectrum resulting from the interaction between a strong infrared (IR) laser pulse and solid-state material. The consistency between the experimental and reconstructed HHG spectra validates the efficiency of our scheme. Our ML method offers two key advantages: first, it does not require stringent experimental conditions, and second, the optimization process is fully automated and more reliable than empirical selection of dephasing time values. The ability of our method to reconstruct dephasing time from solid HHG spectra provides a powerful tool for probing the intrinsic properties of materials under extreme conditions. Besides, our method provides another significant advantage, which offers a direct approach to calculating the quantum tunneling time of carriers between different energy bands under light-induced excitation.