Exploring inhomogeneous surfaces: Ti-rich SrTiO₃(110) reconstructions via active learning

Area of Science:

• Materials Science
• Surface Science
• Computational Materials Science

Background:

• Understanding inhomogeneous surfaces with coexisting local structures is vital for technologically relevant interfaces.
• Studying these complex surfaces presents significant experimental and theoretical challenges.

Purpose of the Study:

• To investigate the atomic configurations of (2 × m) Ti-rich surfaces on (110)-oriented strontium titanate (SrTiO3).
• To develop and apply a machine learning-backed approach for predicting complex surface structures.

Main Methods:

• Combined scanning tunneling microscopy (STM) with transferable neural-network force fields and evolutionary exploration.
• Utilized an active learning methodology to iteratively refine training data.
• Employed density functional theory (DFT) for verification.

Main Results:

• Successfully extrapolated from simple known reconstructions to complex, diverse overlayers on inhomogeneous SrTiO3(110)-(2 × m) surfaces.
• Generated several novel candidate surface structures.
• Achieved good agreement between predicted structures and experimental observations.

Conclusions:

• The developed machine learning approach effectively predicts complex surface reconstructions on inhomogeneous metal oxides.
• This methodology can be extended to study other complex metal oxides with large coexisting surface reconstructions.
• The findings advance the understanding of SrTiO3 surfaces and offer a powerful tool for materials discovery.