My understanding is that, at the borders, master is constantly changing between its states. A boiling pot has water molecules turning to steam, as well as vice versa. It just has more of the first probabilistically.
That's more or less it, as far as I understand. There's the hexagonal lattice of solid water ice, where hydrogen bonds form a 3D tetrahedral structure which is very stable. At the boundary, there are no molecules "above" the solid latice, so those water molecules are held less tightly. The exact dynamics are challenging to predict because we can't exactly see what's going on (most simulation based).
Maybe the same happens with ice and water.