Reduced oxide surfaces often undergo complex reconstructions in which atomic rearrangements and reduction-induced electronic redistribution are strongly coupled, obscuring the electronic principles that govern their stability. Here, we address this problem on the prototypical rutile ${\mathrm{TiO}}_2(110)$ surface by combining machine-learning-accelerated global structure search, first-principles calculations, simulated scanning tunneling microscopy, and data-driven electronic-structure analysis. We identify previously unreported ${\text{Ti}}_3{\mathrm{O}}_2\text{−}(2\times 1)$ and ${\text{Ti}}_3{\mathrm{O}}_4\text{−}(2\times 1)$ reconstructions, with ${\text{Ti}}_3{\mathrm{O}}_2\text{−}(2\times 1)$ being more stable than the previously proposed ${\mathrm{Ti}}_3{\mathrm{O}}_2\text{−}(1\times 2)$ model by up to 0.67 eV per minimal reconstruction unit and reproducing the experimentally observed rosettelike STM motifs. Analysis of 80 ${\mathrm{Ti}}_3{\mathrm{O}}_2$ reconstruction configurations reveals that their thermodynamic stability is governed by the integrated hybridization between occupied Ti $3d$ states associated with ${\mathrm{Ti}}^{3+}$ small polarons and $\mathrm{O}2p$ states of the reconstructed oxide lattice. This polaronic lattice state hybridization distinguishes low-energy reconstructions from the broader configurational ensemble and explains the stabilization of the newly found ground-state structure. The ${\mathrm{Ti}}^{3+}$-derived states are also present near the Fermi level, which facilitates electron transfer to adsorbates, allowing the reconstructed surface to combine thermodynamic stability with chemical activity. These results establish polaronic lattice state hybridization as a microscopic mechanism for stabilizing reduced ${\mathrm{TiO}}_2$ surface reconstructions and suggest a broader electronic principle governing reconstruction of defect-rich reducible oxide surfaces.

Close

PrevNext

View figure in article

• Electronic structure
• Polarons
• Surface reconstruction
• Thermodynamics
• Data mining
• Machine learning