Presentation Information

[7p-N201-6]Analysis of Electric Field Response of Ferroelectric Materials via Machine Learning Potentials

〇(D)Poyen Chen1, Teruyasu Mizoguchi1,2 (1.Univ. of Tokyo, 2.IIS, UTokyo)

Keywords:

machine learning potential,ferroelectric and dielectric property,molecular dynamic simulation

With the advancement of machine learning techniques, machine learning potentials (MLPs) have emerged as powerful tools for accelerating molecular dynamics (MD) simulations while maintaining accuracy comparable to density functional theory (DFT). These potentials enable efficient simulations of large-scale systems that would otherwise be computationally prohibitive using conventional first-principles methods. However, most existing MLPs are primarily trained on structural and energetic information, without explicitly incorporating electronic interactions. As a result, their applicability to the study of electric or dielectric properties, particularly in ferroelectric materials, is significantly limited. To address this limitation, a recent approach leveraging Born effective charge (BEC) prediction, exemplified by the Equivar model, has shown promise in overcoming this limitation. BECs capture the polarization induced by atomic displacements and thus provide a direct link between atomic motion and dielectric response under an applied electric field. Incorporating BECs into the force field formulation allows MD simulations to explicitly capture electric field effects.
In this presentation, we demonstrate the electric field response of representative ferroelectric materials by employing MLPs. We developed material-specific MLPs by fine-tuning the pretrained mace-mp-0 model and then incorporated the external electric force acting on each atom into NpT ensemble molecular dynamics simulations. This approach enables the direct investigation of electric properties in ferroelectric materials. We believe that this electric-field-coupled MD framework not only reproduces key ferroelectric phenomena but also offers a powerful tool for probing complex dielectric responses, ultimately offering a powerful computational strategy for accelerating the discovery and optimization of next-generation ferroelectric and dielectric materials.