Abstract
First-principles prediction of lattice thermal conductivity of strongly anharmonic crystals is a long-standing challenge in solid-state physics. Making use of recent advances in information science, we propose a systematic and rigorous approach to this problem, compressive sensing lattice dynamics. Compressive sensing is used to select the physically important terms in the lattice dynamics model and determine their values in one shot. Nonintuitively, high accuracy is achieved when the model is trained on first-principles forces in quasirandom atomic configurations. The method is demonstrated for Si, NaCl, and , an earth-abundant thermoelectric with strong phonon-phonon interactions that limit the room-temperature to values near the amorphous limit.
- Received 22 April 2014
DOI:https://doi.org/10.1103/PhysRevLett.113.185501
© 2014 American Physical Society