Water freezes when its molecules slow down enough to lock into a rigid, repeating crystal structure held together by hydrogen bonds. At standard atmospheric pressure, this happens at 0 °C (32 °F), but the reality is far messier than that clean number suggests. Pure water with nothing to freeze onto can remain liquid well below zero, and the shape, size, and speed of ice crystal growth depend on a surprisingly long list of factors, from airborne dust particles to bacterial proteins to the dimensions of the container the water sits in.
What Happens to Water Molecules When They Freeze
Every water molecule can form up to four hydrogen bonds with its neighbors, two through the hydrogen atoms and two through the oxygen. In liquid water, these bonds constantly break and reform as molecules tumble past each other. Cooling the liquid slows that tumbling. At the freezing point, molecules settle into a hexagonal arrangement where each oxygen atom is surrounded by four others in a roughly tetrahedral pattern. This is ordinary ice, known to scientists as Ice Ih (the “h” stands for hexagonal), and it is the only form of ice that occurs naturally at Earth’s surface.
The hexagonal lattice is more open and spacious than the jumble of liquid water. That is why ice floats: the crystal structure is less dense than the liquid it came from, a quirk that is rare among substances and enormously consequential for life on Earth.
Within this lattice, the hydrogen atoms are not fixed in one orientation. They obey certain rules about how they sit along the bond between two oxygens, but statistically they can occupy different positions. This “proton disorder” means that even a single ice crystal contains a subtle randomness in its internal arrangement, despite looking perfectly ordered from the outside.
Nucleation, or Why Freezing Needs a Starting Point
Reaching 0 °C is necessary but not sufficient. Water molecules need something to organize around before the first tiny ice crystal can appear. Physicists call this step nucleation, and it comes in two flavors.
In homogeneous nucleation, water molecules spontaneously arrange themselves into an ice-like cluster without any outside help. This requires extreme supercooling, because the energy cost of creating a new crystal surface is high. Experiments on very thin water films cooled at extraordinary rates (billions of degrees per second) have measured the peak homogeneous nucleation rate and found it to be staggeringly large, but only at temperatures far below 0 °C.
In practice, almost all freezing you encounter is heterogeneous nucleation: ice forms on the surface of something else. A scratch on the inside of a glass, a speck of dust, a grain of pollen, or a mineral particle in a cloud droplet all serve as nucleation sites. The surface lowers the energy barrier, making it much easier for ice to begin growing. Research on desert dust particles has shown that the mineral composition matters: quartz and feldspar are particularly effective at triggering ice formation in the atmosphere at temperatures between about −31 and −35 °C.
Not all surfaces are equally good at this. The microscopic features of a particle, including surface defects and chemical groups, partly determine how readily it triggers freezing. Some minerals like microcline (a type of feldspar) can nucleate ice at temperatures as warm as −26 °C, while others require much colder conditions before they become active.
Supercooled Water and Its Strange Behavior
If water is very pure and sits in a smooth container with no dust or rough edges, it can stay liquid far below 0 °C. This is supercooled water, and it is not just a laboratory curiosity. Cloud droplets in the atmosphere routinely exist in a supercooled state, sometimes down to −30 or −40 °C, because they are so small and clean that nucleation sites are scarce.
Supercooled water is metastable, meaning it is liquid only because it has not yet found a way to start freezing. Disturb it (tap the container, drop in a crystal, or introduce a dust particle) and it can flash-freeze almost instantly. The deeper the supercooling, the more dramatic the snap to ice.
Researchers have long debated whether supercooled water undergoes something even stranger before it freezes: a transition between two different liquid forms, one denser and one less dense. Recent experiments using ultrafast heating of amorphous ice observed this liquid-liquid phase transition happening on a timescale of less than 100 nanoseconds, well before ice crystals began forming (which took more than a microsecond). Computer simulations of supercooled water nanodroplets have added to this picture, showing that as droplets get smaller, internal pressure can push the water from its low-density liquid phase into a high-density phase. Whether this two-liquid picture fully explains all of supercooled water’s oddities remains an active area of research.
How Snowflakes Get Their Shapes
Snowflakes are ice crystals that grow directly from water vapor in clouds rather than from liquid water. Because Ice Ih has a hexagonal lattice, all snowflakes share six-fold symmetry, but the details of their shape depend on the temperature and humidity they encounter as they fall.
The relationship between growth conditions and crystal shape was mapped out decades ago by physicist Ukichiro Nakaya in what is now called the Nakaya diagram. At temperatures just below freezing, crystals tend to grow as flat plates. A few degrees colder, they switch to columns and needles. Colder still, they return to plates and eventually to the elaborate branching dendrites people picture when they think of snowflakes. Higher humidity generally produces more complex, branched shapes, while lower humidity yields simpler, more compact crystals. Computational models that simulate crystal growth under different humidity and temperature conditions have successfully reproduced both the basic shapes and the onset of branching instabilities seen in real snowflakes.
Field observations in places like northern Finland have confirmed these relationships by matching the crystal habits found on the ground with radiosonde measurements of temperature and humidity at the altitude where the crystals grew. Above a certain humidity threshold, crystals tend to be large and sometimes coated with frozen cloud droplets (a process called riming). Below that threshold, they stay compact and unrimed.
Secondary Ice Formation in Clouds
Not every ice crystal in a cloud forms by nucleating on a dust particle. A growing body of evidence points to secondary ice formation, where existing ice crystals shatter, splinter, or otherwise produce fragments that seed new crystals. One study analyzed 190 planar, branched ice crystals collected from mixed-phase clouds and melted them to see whether they contained an ice-nucleating particle. Only about one in eight did. The other seven out of eight likely formed through some secondary process, since they contained no particle active above −17 °C. This finding has big implications for how quickly clouds can glaciate (convert from liquid droplets to ice) and, by extension, for precipitation and climate modeling.
Bacteria That Make Ice
Some of the most effective ice nucleators in nature are not minerals but proteins. The bacterium Pseudomonas syringae, commonly found on plant leaves, produces ice-nucleating proteins (INpro) that can trigger freezing at temperatures as warm as −2 to −5 °C, far warmer than any mineral particle. These proteins work by organizing water molecules at their surface into an ice-like arrangement that lowers the energy barrier for nucleation.
High-speed cryo-microscopy has revealed that INpro from sterilized P. syringae cells preferentially trigger freezing at hydrophobic interfaces, such as the air-water boundary of a droplet or the bacterial membrane itself. When cellular fragments were removed by filtration, the proteins nucleated ice even more frequently at the air-water interface. Intact bacteria, by contrast, nucleated ice without a preference for that interface, suggesting the bacterial membrane plays its own role in the process.
Spectroscopy studies have shown how this works at the molecular level. The ice-active repeats of the InaZ protein adopt a specific structure that imposes order on nearby water molecules. As the temperature drops toward freezing, the protein reorients in a way that strengthens its interaction with water, making nucleation progressively more effective. This is not a passive surface waiting for ice to land on it; it is a temperature-activated molecular machine for organizing water.
The practical consequence is real: P. syringae causes frost damage to crops by nucleating ice inside plant tissues at temperatures that would otherwise be safe. A commercial product derived from these bacteria (Snomax) is widely used to make artificial snow at ski resorts, because it dramatically raises the temperature at which snow machines can operate.
Antifreeze Proteins and How Organisms Resist Freezing
If some organisms have evolved proteins that promote ice formation, others have evolved the opposite: antifreeze proteins (AFPs) that bind to ice crystal surfaces and prevent further growth. Fish in polar oceans, overwintering insects, and some plants all produce AFPs, though the proteins vary widely in structure across species.
The mechanism is elegantly simple in concept. AFPs adsorb onto specific faces of a growing ice crystal and physically block water molecules from joining the lattice. Experiments using temperature-controlled microfluidic devices have demonstrated that once a hyperactive AFP from the beetle Tenebrio molitor binds to an ice crystal, the binding is essentially irreversible, and the surface-bound proteins alone are enough to stop crystal growth even if the surrounding solution is flushed clean of dissolved protein.
Not all AFPs work the same way, though. A comprehensive comparison of all major classes of AFPs found that two commonly measured antifreeze activities, thermal hysteresis (the gap between melting and freezing temperatures) and ice recrystallization inhibition (preventing small crystals from merging into large ones), are not correlated with each other. Blocking fast ice growth requires rapid binding to non-basal crystal faces, while inhibiting slow recrystallization depends on binding to the basal face. Different protein structures are better at one or the other, which explains why some AFPs excel at depressing the freezing point while others are better at keeping already-frozen tissues from suffering further crystal damage.
The Quasi-Liquid Layer on Ice
Ice is not truly solid all the way to its surface. Even well below 0 °C, the outermost layer of an ice crystal is disordered, forming what researchers call a quasi-liquid layer (QLL). This thin, mobile film is the reason ice is slippery and the reason snowflakes can sinter together into snowballs.
Molecular simulations combined with surface spectroscopy measurements have mapped out how this layer develops. The topmost molecular layer of ice begins to show disorder at around −90 °C, as molecules with three hydrogen bonds convert to a less-bonded state. The second layer joins in at about −16 °C, and the disordering proceeds in a bilayer-by-bilayer fashion rather than as a smooth gradient. As the temperature approaches 0 °C, the quasi-liquid layer thickens rapidly. Studies on different ice forms have confirmed that the QLL preferentially decorates certain crystal faces and grows thicker as the melting point draws near.
This surface premelting explains a range of everyday phenomena. It is why pressing two ice cubes together can fuse them, why glaciers flow despite being solid, and why ice skating works. The quasi-liquid layer acts as a lubricant, reducing friction in a way that a perfectly crystalline surface would not.
Freezing in Tiny Spaces
Confine water inside a tube only a nanometer or two wide, and the rules of freezing change. Simulations of water inside single-walled carbon nanotubes have revealed at least nine distinct ice structures that do not exist in the bulk, each one maximizing hydrogen bonds within the cylindrical space. The freezing point depends strongly on tube diameter: at certain widths, water freezes at temperatures far above 0 °C. Simulations using a well-established water model found a maximum freezing temperature of about 390 K (117 °C) inside nanotubes of a particular diameter, a result that agrees with experimental measurements.
The ice that forms in these confined spaces includes exotic arrangements like square ice nanotubes, where water molecules line up in four-sided rings instead of the usual hexagons. The global maximum in the melting curve occurs at a tube diameter of roughly 1.1 nanometers, corresponding to a square ice nanotube structure. These findings are not just academic curiosities. They are relevant to understanding water transport through biological channels, the behavior of water in porous materials, and the design of nanofluidic devices.
Food Freezing and Why Speed Matters
Anyone who has frozen strawberries knows the result can be mushy. The reason comes down to ice crystal size and location, both of which depend on freezing speed. When food with a cellular structure freezes slowly, ice tends to form outside the cells first. Water inside the cells, driven by a chemical potential difference, migrates outward to join those growing extracellular crystals, causing the cells to shrink and dehydrate. Freeze the same food rapidly, and ice forms inside the cells instead, but those intracellular crystals can puncture cell membranes. Either way, the cells suffer damage, and the texture degrades.
The ideal approach is to freeze fast enough that crystals stay very small, limiting mechanical damage. This is where ultrasound-assisted freezing has attracted attention. Applying ultrasound during freezing dramatically reduces the degree of supercooling (the temperature overshoot before nucleation kicks in), triggers nucleation sooner, and produces smaller, more uniform ice crystals. One study found that multi-frequency ultrasound at a moderate power level reduced both pore size and ice crystal size in frozen fish, preserving the microstructure better than conventional immersion freezing. Intermittent ultrasound application has also been shown to save energy, with one set of experiments reporting savings of about 10-12% compared to freezing without ultrasound.
Electric Fields and Electrofreezing
Electric fields offer another way to nudge water toward freezing. In electrofreezing, applying a voltage to supercooled water promotes nucleation, sometimes at temperatures just barely below 0 °C. Experiments using aluminum and magnesium electrodes achieved freezing of supercooled water at −0.5 °C, with molecular dynamics simulations suggesting that metal hydroxide species released from the electrodes, under the influence of the electric field, created ice-like molecular arrangements that triggered the process.
Interestingly, the nucleation in electrofreezing appears to be driven primarily by events at the three-phase boundary (where the liquid, solid surface, and air or vapor all meet), not simply at the solid-liquid interface underneath the droplet. This distinction matters for designing surfaces or devices that can control when and where freezing starts, with potential applications in de-icing technology and controlled crystallization processes.
Ice Beyond Earth
Most of the ice in the universe looks nothing like the ice in your freezer. At temperatures below about 30 K (−243 °C), water vapor depositing onto cold surfaces in space forms amorphous ice, a solid without any regular crystal structure. The dominant form in interstellar environments is high-density amorphous ice, which can be thought of as a collapsed version of the more familiar low-density amorphous ice. Electron diffraction experiments and molecular dynamics simulations have characterized its structure, finding that it features oxygen-oxygen distances between 3 and 4 angstroms that do not appear in crystalline ice. This high-density amorphous form dominates the 3.07-micrometer ice absorption band observed in various astronomical environments, meaning it is the primary ice signature astronomers detect when they look at interstellar dust grains and protoplanetary disks.
At the other extreme, ice subjected to enormous pressures, such as those found deep inside giant planets, can form superionic phases where the oxygen atoms lock into a crystal lattice while the hydrogen atoms flow freely through it like a liquid. At pressures around 1.6 terapascals and temperatures near 1,900 K, one such phase transitions to a new structure with different symmetry. These exotic ices are thought to exist in the interiors of Uranus and Neptune, and understanding them helps planetary scientists model these worlds’ magnetic fields and internal heat.
Frost Heaving and Freezing Ground
When the ground freezes, ice does not simply fill the existing pore spaces between soil grains. Instead, thin layers of ice called ice lenses can grow by drawing water upward from unfrozen soil below. As these lenses thicken, they push the soil apart, heaving the ground surface upward. This is frost heaving, and it buckles roads, cracks foundations, and reshapes landscapes across cold regions.
The process depends on a suction gradient: as ice forms, it lowers the local water pressure, pulling liquid water toward the freezing front through the soil’s tiny pore network. Research on how freezing direction affects ice lens growth has shown that top-down freezing (the typical pattern when cold air chills the ground surface) produces more heaving than bottom-up freezing, because gravity helps water migrate toward the lens. Vertical cracks that develop in the soil during ice lens formation further improve the hydraulic pathways, allowing even more water to accumulate and the lenses to grow thicker. This feedback loop is why frost heaving can be so destructive, lifting fence posts, snapping water pipes, and turning smooth pavement into a roller coaster over a single winter.