Thermal Transport in SrSnO3 Revealed by First-principles Theory, Raman Thermometry and Machine Learning

Abstract

The thermal conductivity (κ) of SrSnO₃ is systematically investigated using a comprehensive approach integrating Raman thermometry, laser flash analysis (LFA), first-principles calculations based on density functional theory (DFT) combined with the semi-classical Boltzmann transport theory (BTE), and machine learning methodologies. Raman thermometry, modelled using the Balkanski framework, reveals strong cubic and quartic phonon-phonon interactions, providing a physical basis for the observed low lattice thermal conductivity (κ_L). First-principles calculations indicate that the Sn-O framework, due to its strong bonding, promotes dispersive electronic states and high-velocity phonon propagation, while the weaker Sr-O bonds act as effective phonon scattering centers. This mass- and bonding-driven phonon separation, reinforced by a distinct phonon gap, intrinsically limits κ_L. A comprehensive BTE analysis reveals that long-wavelength acoustic phonons with high velocities and extended lifetimes dominate heat transport, while optical modes contribute minimally due to strong anharmonic scattering and reduced phonon dispersion. The κ_L of SrSnO₃ decreases monotonically with temperature, following an intrinsic Umklapp-limited T⁻¹ trend characteristic of pronounced lattice anharmonicity. Excellent agreement between first-principles predictions and Raman-validated refined Slack modelling confirms that intrinsic three-phonon scattering within the Sn-O framework governs thermal transport in SrSnO₃. However, experimental LFA measurements reveal a T^−0.42 power-law dependence, with additional temperature-independent scattering mechanisms, such as point-defect and grain-boundary scattering, superimposed on intrinsic Umklapp processes. Cumulative mean-free-path analysis further highlights opportunities for nanostructuring-driven thermal suppression, as a substantial fraction of κ_L arises from phonons extending to micrometer scales. Finally, a machine-learning stacking architecture achieved superior performance, with a Root Mean Square Error (RMSE) of 0.456 Wm^-1K^-1, representing a 16.5 % improvement in accuracy over a standalone artificial neural network model. These findings reveal the fundamental vibrational transport mechanisms in SrSnO₃, establishing its viability for high-temperature electronics, thermoelectric energy conversion, and perovskite-based thermal management.