Site-Selective Initial Hydration of Boron at Multicomponent Borosilicate Glass–Water Interfaces Revealed Using Machine Learning Molecular Dynamics
Ryuki Kayano, Jean-Marc Delaye, Stéphane Gin, Faijan, Takahiro OhkuboAbstract
Understanding the hydration reactions at borosilicate glass–water interfaces is essential for predicting long-term chemical durability. However, atomistic simulations of multicomponent glass corrosion require both chemical accuracy and nanosecond-scale sampling, which remain difficult to achieve using conventional force fields or ab initio molecular dynamics. This paper describes the development and validation of a machine learning potential (MLP) for multicomponent borosilicate glass–water interfaces containing Si, Al, B, Zr, Na, Ca, Li, Mg, O, and H. This MLP enables nanosecond-scale machine learning molecular dynamics (MLMD) simulations with accuracy close to that of density functional theory. Benchmark simulations of sodium aluminosilicate and calcium aluminosilicate glass–water interfaces demonstrate that the MLP reproduces hydration-induced changes in Al coordination more accurately than available classical potentials. The validated MLP is subsequently applied to multicomponent (ISG-1) glass–water interface models to investigate the initial dissolution behavior of boron. Surface boron atoms primarily react with water through the formation of hydroxylated species. A reaction network analysis of three- and four-coordinated B species, B[3] and B[4], based on oxygen species labeling, reveals that B[3]–OH and B[4]–OH form through distinct pathways: the B[3]–OH species are mainly formed through ion exchange reactions involving modifier cations, whereas B[4]–OH species form through direct water attachment to B[3] followed by deprotonation. The reactivity of B[3] sites is strongly dependent on the type and number of second-nearest-neighbor cations, with boron sites linked to Al or Zr exhibiting markedly lower reactivity than those linked to Si or B[3], indicating site-selective and incongruent initial boron dissolution. These results provide a quantitative atomistic basis for understanding how glass composition controls boron dissolution at multicomponent glass–water interfaces.