Machine-Learning-Guided Design of Incommensurate Antiferroelectrics via Field-Driven Phase Engineering

Antiferroelectrics, defined by antiparallel polarization configurations, have emerged as promising dielectric materials for high-performance energy storage applications. The energy storage density and efficiency of antiferroelectrics are governed by the regulation of electric field-driven antiferroelectric-ferroelectric phase transitions (and their reversibility). Herein, we propose a machine learning-guided phase-field simulation framework to accelerate the design of energy storage ceramics. Taking PbZrO₃-based incommensurate antiferroelectrics as an example, we establish quantitative mappings between energy storage density/efficiency and key features (point defect, antiphase boundary energy, grain size, and strain) using XGBoost. The model exhibits high accuracy for energy storage performance prediction, with R² values of 0.99. SHAP interpretable machine learning analysis further deciphers the correlations between multiple features and energy storage properties, while phase-field simulations clarify the mechanisms. Guided by this framework, we achieve a high energy storage density of 22.1 J/cm³ with 96.1% efficiency. This work provides a theoretical foundation and a generalized approach for regulation of antiferroelectric energy storage properties and mechanisms of phase transition.