Deciphering borophene growth pathways with data-driven simulations

Deterministic synthesis of borophene remains challenging because many polymorphs compete during nucleation and growth. Here we combine a reactive machine-learned interatomic potential with grand-canonical Monte Carlo simulations and data-driven structural classification to track borophene formation from early nuclei to extended layers on Ag(111) and Ag(100). We build temperature-pressure substrate growth maps and resolve how vacancy motifs, phase intermixing and seed structure govern polymorph selection. The simulations reproduce key experimental trends, including the prevalence of $β_{12}$/$χ_3$ phases and their temperature-dependent competition, while revealing kinetic pathways that connect metastable nuclei to long-range order. We identify conditions that suppress competing motifs and promote targeted phases, providing actionable synthesis windows. These results establish a predictive framework for directing borophene growth and, more broadly, for controlling polymorphism in low-dimensional materials by coupling atomistic simulation with machine-learning-enabled phase recognition.