Molecular mechanism of heterogeneous ice nucleation in the atmosphere

Abstract

Mineral dust aerosols strongly influence Earth's climate by acting as ice-nucleating particles (INPs). Feldspar minerals, particularly K-feldspar, are recognized as dominant INPs, and a previous study attributed this behavior to (100) surfaces exposed at defects. Using machine-learning molecular dynamics simulations, we systematically investigate ice nucleation on all K-feldspar surfaces. We identify the (110) surface, exposed at defects such as steps, as the most active plane for ice formation. This surface uniquely structures interfacial water into an arrangement resembling that on the (110) surface of cubic ice, providing an optimal template for nucleation. Using advanced sampling, we directly observe the formation of clusters with cubic-ice structure and their orientation agrees with experiment. These results provide a molecular-level explanation of how ice forms in our planet's atmosphere.

Figures (4)

• Figure 1: K-feldspar serves as an effective ice-nucleating particle in the atmosphere owing to structural defects that promote ice nucleation.(A) Schematic illustration of the heterogeneous ice nucleation at K-feldspar defects with exposed non-perfect cleavage planes. The pink and green dashed lines highlight the first and second prismatic surfaces of ice, which are approximately parallel to the (100) and (110) planes of feldspar, respectively. (B) The triclinic crystal structure of microcline K-feldspar. (C) Crystallographic planes of K-feldspar: (001), (010), (100), (110), ($\bar{1}$10), and ($\bar{2}$01). (D) Configuration for molecular dynamics simulations with one termination of the K-feldspar (110) surface in contact with water above.
• Figure 2: Comparison of the water distribution at feldspar–water interfaces and ice–water interfaces at 18 K of supercooling.(A) Oxygen (O) density (molecules/nm$^{3}$) profile perpendicular to the feldspar (110)-$\alpha$ interface with water. (B) O density profile perpendicular to the cubic ice I$_{\mathrm{c}}$ (110) interface with water. (C) Density difference of O between water on the feldspar (110)-$\alpha$ surface and that on the I$_{\mathrm{c}}$ (110) surface. (D) Mean squared error (MSE) between water density profiles at the feldspar–water and ice–water interfaces in the region $z < 15~\text{\AA}$. The 13 K-feldspar surfaces are labeled according to their Miller indices and different terminations are differentiated using Greek letters. The seven ice surfaces are labeled according to the polymorph (hexagonal ice I$_{\mathrm{h}}$ or ice I$_{\mathrm{c}}$) and the Miller indices. The white rectangle highlights the location of the lowest MSE. (E, F) Water-density isosurfaces at 70 molecules/nm$^{3}$ in the $x$–$y$ plane, corresponding to the first peak of the density profiles shown in panels A and B, respectively. (G) The surface of the feldspar (110)-$\alpha$ termination. Black lines highlight the pattern of water distribution in panels E–G.
• Figure 3: Formation of ice at the feldspar (110)-$\alpha$ surface.(A) Snapshots of ice nucleation on the feldspar surface over the $Q_6$ order parameter from MD simulations with a bias along the $Q_6$. (B) Number of ice-like water molecules and $Q_6$ versus simulation time. (C) Water-density isosurfaces of the first ice layer at 70 molecules/nm$^3$. (D) Ice-like water molecules over time at different supercooling, ${\Delta}T$. (E) Critical cluster size, $N^*$, versus ${\Delta}T$ with a classical nucleation theory (CNT) fit (dashed lines). Data for homogeneous nucleation is taken from Ref. piaggi2022homogeneous. (F) Interfacial free energy, $\gamma$, versus ${\Delta}T$. The dashed lines represent a fit based on CNT. (G) Solid lines show the CNT-calculated free-energy barriers for heterogeneous (${\Delta}G^*_\mathrm{het}$) on the feldspar surface and homogeneous (${\Delta}G^*_\mathrm{hom}$) nucleation. The heterogeneous nucleation data point (${\Delta}G^*_\mathrm{het-US}$) comes from umbrella sampling (panel H), and the homogeneous point is from Ref. chen2025exploring. (H) Free-energy profile along $Q_6$ obtained from umbrella sampling simulations. (I) Heterogeneous nucleation rates ($J$) versus ${\Delta}T$ with CNT fit and experimental estimates. Further details of the ice-structure identification, CNT calculations, and estimation of nucleation rates are provided in the Methods section.
• Figure 4: Orientation relationship between feldspar and the ice cluster formed in our simulation.(A) Orientation and lattice of ice formed on the K-feldspar (110)-$\alpha$ surface. Visualizations of the ice structure along the $x$, $y$, and $z$ axis are shown. Characteristic patterns on each ice plane are depicted using black dashed lines. (B) Illustration of the ice basal-plane axis orientation relative to the feldspar (001), (010), and (110) planes. Note the basal (0001) plane of hexagonal ice (I$_\mathrm{h}$) is equivalent to the (111) plane of cubic ice (I$_\mathrm{c}$).