Dielectric Tensor Prediction for Inorganic Materials Using Latent Information from Preferred Potential
Abstract
Dielectrics are crucial for technologies like flash memory, CPUs, photovoltaics, and capacitors, but public data on these materials are scarce, restricting research and development. Existing machine learning models have focused on predicting scalar polycrystalline dielectric constants, neglecting the directional nature of dielectric tensors essential for material design. This study leverages multi-rank equivariant structural embeddings from a universal neural network potential to enhance predictions of dielectric tensors. We develop an equivariant readout decoder to predict total, electronic, and ionic dielectric tensors while preserving O(3) equivariance, and benchmark its performance against state-of-the-art algorithms. Virtual screening of thermodynamically stable materials from Materials Project for two discovery tasks, high-dielectric and highly anisotropic materials, identifies promising candidates including Cs2Ti(WO4)3 (band gap $E_g=2.93 \mathrm{eV}$, dielectric constant $ arepsilon=180.90$) and CsZrCuSe3 (anisotropic ratio $\alpha_r = 121.89$). The results demonstrate our model's accuracy in predicting dielectric tensors and its potential for discovering novel dielectric materials.
AI-Generated Overview
Overview of the Research: Dielectric Tensor Prediction for Inorganic Materials
-
Research Focus: The study investigates the prediction of dielectric tensors for inorganic materials, addressing the limitations of existing models that primarily predict scalar values without considering the directional nature of dielectrics.
-
Methodology: The authors developed a Dielectric Tensor Neural Network (DTNet) that utilizes multi-rank equivariant structural embeddings from a universal neural network potential (PreFerred Potential, PFP) to enhance predictions of dielectric tensors, while employing a virtual screening approach across various material structures.
-
Results: The model, DTNet, was benchmarked against existing methods and demonstrated superior performance in predicting total, electronic, and ionic dielectric tensors, with significant accuracy improvements noted during virtual screening for high-dielectric and highly anisotropic materials.
-
Key Contribution(s): DTNet effectively integrates higher-order tensorial information for dielectric properties using pretrained graph neural network features, representing a novel application of transfer learning in materials science that allows accurate predictions with limited datasets.
-
Significance: The successful prediction of dielectric tensors has crucial implications for the design and discovery of new materials with tailored properties for applications in electronics, energy storage, and other technologies, enhancing the material development process.
-
Broader Applications: The techniques and findings from this research can be applied to other tensorial property predictions in materials science, opening avenues for advancements in areas such as polarizability and elasticity, ultimately accelerating the material discovery process for diverse applications.