Machine learning perspective on electron-lattice dynamics

The interplay between electrons and the crystal lattice forms the backbone of many properties in condensed matter systems. Here, we present a time-dependent machine learning approach to explore the diversity of dynamics arising from the electron-lattice interaction, building on a coherent state representation of lattice vibrations. Taking advantage of clustering techniques, we generate a phase diagram of electron dynamics from the electron wavefunction, revealing various regimes of dynamical behavior, including transient Anderson localization, where strong lattice vibrations temporarily confine and subsequently release electronic wavepackets. While our focus has been on the electron-lattice interaction, our approach provides a general framework for analyzing complex dynamical behavior in quantum systems employing machine learning.