Search

Machine learning has demonstrated great power in materials design, discovery, and property prediction. However, despite the success of machine learning in predicting discrete properties, challenges remain for continuous property prediction. The challenge is aggravated in crystalline solids due to crystallographic symmetry considerations and data scarcity. Here, the direct prediction of phonon density‐of‐states (DOS) is demonstrated using only atomic species and positions as input. Euclidean neural networks are applied, which by construction are equivariant to 3D rotations, translations, and inversion and thereby capture full crystal symmetry, and achieve high‐quality prediction using a small training set of ≈103 examples with over 64 atom types. The predictive model reproduces key features of experimental data and even generalizes to materials with unseen elements, and is naturally suited to efficiently predict alloy systems without additional computational cost. The potential of the network is demonstrated by predicting a broad number of high phononic specific heat capacity materials. The work indicates an efficient approach to explore materials' phonon structure, and can further enable rapid screening for high‐performance thermal storage materials and phonon‐mediated superconductors., A machine‐learning‐based predictive model allows a direct output of phonon density‐of‐states (DOS) by inputting atomic coordinates. The implemented Euclidean neural network preserves crystallographic symmetry and enables high‐quality prediction with small training data. The model allows for fast DOS calculation, reaching ab initio accuracy for 70–80% materials with much lowered computational cost, facilitating materials design especially in alloy systems.

Polarization of ionic and electronic defects in response to high electric fields plays an essential role in determining properties of materials in applications such as memristive devices. However, isolating the polarization response of individual defects has been challenging for both models and measurements. Here the authors quantify the nonlinear dielectric response of neutral oxygen vacancies, comprised of strongly localized electrons at an oxygen vacancy site, in perovskite oxides of the form ABO3. Their approach implements a computationally efficient local Hubbard U correction in density functional theory simulations. These calculations indicate that the electric dipole moment of this defect is correlated positively with the lattice volume, which they varied by elastic strain and by A‐site cation species. In addition, the dipole of the neutral oxygen vacancy under electric field increases with increasing reducibility of the B‐site cation. The predicted relationship among point defect polarization, mechanical strain, and transition metal chemistry provides insights for the properties of memristive materials and devices under high electric fields.

Phonon density‐of‐states is a key property that governs materials thermal properties but is nontrivial to compute or measure. A neural network that carries full crystal symmetry allows a prediction of phonon density‐of‐states using a small volume of training data, approaching ab initio accuracy but with significantly increased efficiency, as demonstrated in article number 2004214, by Mingda Li, Zhantao Chen, Nina Andrejevic, Tess Smidt, and co‐workers. This work enables direct structure‐property design of materials with superior thermal properties. [Image: see text]