Deep-learning enabled atomic insights into the phase transitions and nanodomain topology of lead-free (K,Na)NbO3 ferroelectrics

Lead-free K_xNa_1−xNbO₃ (KNN) perovskites have garnered increasing attention due to their exceptional ferropiezoelectric properties, which are effectively tuned via polymorphic structures and domain dynamics. However, atomic insights into the underlying nanomechanisms governing the ferroelectricity of KNNs amidst varying factors such as composition, phase, and domain are still imperative. Here, we perform molecular dynamics simulations of phase transitions and domain dynamics for KNNs with various K/Na ratios (x = 0.25∼1.0) by using ab-initio accuracy deep learning potential (DP). As a demonstration of its transferability, the newly developed DP model shows quantum accuracy in terms of the equation of states, elastic constants, and phonon dispersion relations for various KNbO₃ and K_0.5Nb_0.5O₃. Furthermore, intricate temperature-dependent phase transitions and domain formation of KNNs are extensively and quantitatively captured. Simulations indicate that for KNNs with compositions x ranging from 0.25 to 1.0, the paraelectric-to-ferroelectric phase transition of KNNs is driven primarily by the order-disorder effect, while the displacive effect is dominant in the subsequent ferroelectric phase transitions. Specifically, flux-closure or herringbone-like nanodomain patterns arranged with 90° domain walls formed close to the experimental observations. Detailed analyses reveal that favorable 90° domain wall formation becomes more challenging with increasing Na content due to distinct oxygen octahedron distortion arising from the different ionic radii of K/Na atoms. It is envisioned that the combination of unified DP and atomistic simulations will help offer a robust solution for more accurate and efficient in silico explorations of complex structural, thermodynamic, and ferroelectric properties for relevant energy storage and conversion materials.