Flexoelectric Effect from First-Principles and machine-learned interatomic potentials

Flexoelectricity, the polarization response to a strain gradient, is a critical phenomenon in all insulators. Although intrinsic flexoelectric coupling can be calculated in bulk crystals using first-principles methods, it is often much smaller than experimentally observed values, likely due to the influence of defects and surface effects. Standard approaches such as long-wavelength DFPT are computationally demanding, limiting their applicability to large supercells with defects. In this work, we investigate whether fine-tuned machine-learned interatomic potentials can serve as accurate and efficient surrogate models for capturing flexoelectric responses in such systems.