Research & Publications

Scanning the potential energy surface for a given compositional space via Ehull  analysisis not sufficient to comment on their thermodynamic stability, since the contribution stemming from the vibrational free energy is typically ignored in high-throughput searches of compositional spaces for stable compounds. The calculation of vibrational free energy through first principles can be computationally very expensive owing to the complexity of the structures, which is directly proportional to the number of symmetrically non-unique terms to be evaluated for the creation of the dynamical matrix. In this work, we use machine learning (ML) to predict the free energy of a given compositional space (ternary perovskite compounds belonging to different symmetric structures) using the elemental and structural descriptors as fingerprints. The temperature dependence of the free energy is modeled using 3rd -order polynomial fit, where the coefficients are learned and predicted using ML. Thereby, a highly accurate model is built for the zero point energy (with an RMSE of 18.9 meV/atom), which is further improved by employing a symbolic regression technique-SISSO, giving a very low RMSE of 8 meV/atom. This model, while providing computationally inexpensive means for predicting the harmonic vibrational free energy of compounds, also provides an aid to get the free energy and hence assess the thermodynamic stability of a given composition at any temperature. This work also provides important insights on how the elemental and compound properties are related to the vibrational free energy and hence, may aid in its prediction.

Predicting the lattice thermal conductivity (κL) of compounds prior to synthesis is an extremely challenging task because of complexity associated with determining the phonon scattering lifetimes for underlying normal and Umklapp processes. An accurate ab initio prediction is computationally very expensive, and hence one seeks for data-driven alternatives. We perform machine learning (ML) on theoretically computed κL of half-Heusler (HH) compounds. An exhaustive descriptor list comprising elemental and compound descriptors is used to build several ML models. We find that ML models built with compound descriptors can reach high accuracy with a fewer number of descriptors, while a set of a large number of elemental descriptors may be used to tune the performance of the model as accurately. Thereby, using only the elemental descriptors, we build a model with exceptionally high accuracy (with an R2 score of ∼0.98/0.97 for the train/test set) using one of the compressed sensing techniques. This work, while unfolding the complex interplay of the descriptors in different dimensions, reveals the competence of the readily available elemental descriptors in building a robust model for predicting κL.