Atomic-scale surface reconstructions critically affect the performance of ${AB\text{O}}_3$ perovskites in catalysis and electronic devices. However, their structural complexity, together with the limited predictive power of existing theoretical models, makes it difficult to determine the true atomic arrangements and therefore hinders rational materials design. Here, using ${\mathrm{SrTiO}}_3\left(001\right)$ [STO(001)] as a model system, we develop a deep potential (DP)-based global optimization framework that efficiently screens candidate reconstructions in large supercells. Within this framework, we identify a series of Sr-adatom-stabilized surface structures for the ($2\times 2$), $c$($4\times 2$), and ($\sqrt{5}\times \sqrt{5}$)$R26.6^{\circ }$ reconstructions, all of which have substantially lower surface energies than conventional models, thereby helping resolve longstanding discrepancies between theory and experiment. We further reveal that Sr adatoms act as structural bridges that interconnect ${\mathrm{TiO}}_5$ units, triggering a pronounced “bonding reinforcement” characterized by enhanced orbital resonance and a significant downward shift of electronic energy levels. To examine the transferability of this $A$-site-adatom stabilization principle, we fine-tune the STO training set to rapidly obtain an accurate DP model for ${\mathrm{BaTiO}}_3\left(001\right)$ [BTO(001)]. Applying the same framework to BTO(001), we predict a previously unreported Ba-adatom ($\sqrt{5}\times \sqrt{5}$)$R26.6^{\circ }$ reconstruction, indicating that the mechanism can be transferred from STO to the related BTO surface. Overall, this work identifies $A$-site adatoms as a stabilizing motif for the related ${AB\text{O}}_3\left(001\right)$ surfaces examined here and provides an extendable, high-accuracy strategy for predicting complex oxide surface and interface structures.

Close

PrevNext

View figure in article

• Electronic structure
• Phase diagrams
• Surface reconstruction
• Perovskites
• Surfaces
• Transition metal oxides
• First-principles calculations
• Machine learning