Adsorbate-Directed Surface Phase Separation in Cu-Ag Catalysts

The determination of surface configuration is essential for understanding catalytic reactivity, yet these active states cannot be reliably inferred from bulk phase diagrams. In this work, we develop a machine-learning-accelerated molecular dynamics (ML-MD) framework on the Cu-Ag system to predict and control surface structure, identifying two competing descriptors: surface energy and adsorbate binding energy. We show that Ag's lower surface energy dominates in vacuum, forming a passive sheath, whereas a CO atmosphere reverses segregation and exposes active Cu sites through lateral phase separation. Decoupling the thermodynamic driving force from the kinetic diffusion barrier, we find that stability emerges not at near-equimolar ratios where segregation is strongest but rather at dilute compositions where weak driving forces and high kinetic barriers coincide. This framework offers a generalizable strategy for balancing activity and stability in metastable catalyst design.