Ice is hexagonal. Its crystal structure has six-fold symmetry, visible in every snowflake. So you'd expect surfaces with hexagonal symmetry to be the best templates for ice nucleation — the matching symmetry should guide water molecules into the ice lattice like a key into a lock.
Molecular simulations of heterogeneous ice nucleation on model substrates (arXiv:2601.05758) show this expectation is wrong. Hexagonal symmetry does not guarantee better ice nucleating ability. Some cubic lattices outperform hexagonal ones, despite lacking the symmetry that matches ice.
The explanation: ice nucleation depends on the three-dimensional arrangement of surface atoms, not just the two-dimensional symmetry visible from above. A substrate with the “right” surface symmetry but the “wrong” layer spacing can actually inhibit nucleation by forcing the first ice layer into a strained configuration. Meanwhile, a substrate with “wrong” symmetry but compatible layer spacing can nucleate ice efficiently by accommodating the full 3D crystal structure, not just its surface projection.
The paper also finds that more efficient nucleants induce a more structured liquid layer adjacent to the substrate. The ordering happens in the liquid before it freezes — the substrate doesn't template ice directly but rather pre-organizes water into a configuration from which ice crystallization becomes kinetically favorable.
This reframes nucleation from template-matching to pre-organization. The best nucleation surface isn't the one that looks like ice — it's the one that makes liquid water behave more like ice. The geometry matters, but not the geometry you'd predict from staring at a snowflake.
The broader lesson: in any process where a template guides self-assembly, the relevant structural match may be between the template and the precursor state, not between the template and the final product.