Charge-Derived Polarization Fingerprints from Charge-Aware Molecular Dynamics: A Descriptor Framework for BaTiO 3

Charge-aware molecular dynamics provides access to dynamically evolving effective atomic charges alongside conventional structural quantities. Here, we introduce a charge-derived polarization fingerprint framework for characterizing local polarization-active environments in ${\text{BaTiO}}_3$ using trajectories generated with a charge-aware neuroevolution potential ( $q$ -NEP). Local ${\text{TiO}}_6$ environments are described using structural quantities, including Ti off-centering and Ti-O bond asymmetry, together with charge-derived descriptors such as the local ${\text{TiO}}_6$ dipole magnitude, charge asymmetry, and oxygen-charge fluctuations. Correlation analysis, density mapping, event-aligned trajectory analysis, and principal component analysis reveal coupled variations of structural and charge-derived descriptors across the sampled local environments. In particular, oxygen-charge fluctuations and local dipole magnitudes provide charge-response signatures that complement geometric descriptors in distinguishing transient polarization-active configurations. Independent ${\text{r}}^2\text{SCAN}$ /Bader calculations for representative configurations further show that the $q$ -NEP effective charges capture aspects of the configuration-dependent charge response, while emphasizing that they should not be interpreted as direct equivalents of electronic-structure partitioned charges. These results demonstrate that charge-aware machine-learning potentials can provide useful model-derived descriptors for characterizing coupled structural and charge-response behavior in complex materials.