At about 2,350°C—hot enough to melt iron at ordinary pressure—water squeezed to more than two million atmospheres can remain partly crystalline instead of boiling. Its oxygen atoms lock into a lattice while hydrogen nuclei flow freely through it, forming superionic ice; researchers have now reported the first direct observation of a previously predicted hexagonal variety.

At ordinary pressure, water would not remain recognisably icy at 2,350°C. Iron melts at about 1,538°C, and liquid water boils at 100°C. But pressure rewrites the phase diagram. Squeeze water beyond two million atmospheres and its oxygen atoms can stay locked into a crystal even while its hydrogen nuclei move through that framework.

A new experiment reports the first unambiguous diffraction evidence for a long-predicted hexagonal version of this state. At 219 gigapascals and 2,630 kelvin, equal to about 2.16 million atmospheres and 2,357°C, a hexagonal close-packed oxygen lattice dominated the X-ray pattern. The result was published in Physical Review Letters on September 9.

This is one study, not settled consensus. There is also an important limit to the word “direct.” The experiment directly resolved the hexagonal arrangement of oxygen atoms through five classes of diffraction peaks. It did not directly watch hydrogen nuclei moving, nor did it measure the phase’s electrical conductivity. Its identification as superionic rests on the conditions in which the structure formed and on a lattice-expansion signature that the authors associate with hydrogen diffusion.

A crystal that is only partly fixed

Ordinary ice is a molecular solid. Oxygen and hydrogen atoms occupy ordered positions, joined into water molecules and hydrogen-bonded to their neighbours. Superionic ice is different. Its heavier oxygen atoms form a solid lattice, while hydrogen nuclei become mobile and move through it in a liquid-like way. The material can therefore be crystalline and highly conductive at the same time.

The newly observed structure is called hexagonal close-packed, or HCP. It should not be confused with ice Ih, the familiar hexagonal crystal responsible for snowflakes. HCP superionic ice exists at extreme pressure, and “hexagonal” describes the packing of its oxygen sublattice. The hydrogen component is disordered and mobile in the superionic interpretation.

A close-packed oxygen lattice can be assembled in more than one stacking sequence. Face-centred-cubic, or FCC, stacking follows an ABCABC pattern. HCP stacking alternates ABAB. The local environment can look very similar in both, which makes the distinction difficult to establish from a tiny, hot sample. Earlier work had firmly identified FCC superionic ice, also known as ice XVIII, while theory predicted HCP should become stable at greater pressure.

How to squeeze and heat a microscopic sample

Alexis Forestier and colleagues placed water between the tips of two diamond anvils. The sample was enclosed by boron-doped diamond absorbers that converted laser light into heat, with alumina providing thermal insulation. Above 200 gigapascals, the water sample was only about 12 micrometres across.

The team examined it at the European Synchrotron Radiation Facility using a focused X-ray beam about 0.5 by 0.8 micrometres wide. As the beam passed through the compressed, laser-heated water, the crystal scattered X-rays at angles set by its atomic spacing. The resulting diffraction peaks acted as a structural fingerprint. Full experimental details and figures are available in the authors’ open manuscript.

Diamond anvil cells create enormous static pressure, allowing a sample to be heated and cooled while researchers repeatedly collect patterns. That differs from shock-compression experiments, which reach extreme conditions for a very short time. The static approach helped the team follow how the oxygen lattice changed with pressure and temperature instead of capturing only a fleeting endpoint.

Five peak families reveal the hexagonal lattice

In the highest-pressure run, heating at 155 gigapascals and around 2,000 kelvin produced both FCC and HCP signatures. At 197 gigapascals and 2,250 kelvin, peaks from both structures were again visible. When the sample reached 219 gigapascals and 2,630 kelvin, the FCC peaks had almost vanished and the five HCP peak families dominated.

That progression is central to the result. One unfamiliar bump in a diffraction trace can come from the heater, insulation, diamond anvils or a chemical contaminant. Five mutually consistent reflections, changing together as conditions change, offer a much stronger structural identification. The measured ratio between the HCP lattice dimensions was also close to the ideal value expected for tightly packed spheres.

The team saw another route to HCP stacking while cooling a lower-pressure sample. Below about 1,400 kelvin, new HCP reflections emerged alongside FCC and body-centred-cubic signals. This mixture suggests that stacking faults and metastable arrangements can survive as the structure transforms. It also indicates that the route through pressure and temperature matters, not just the final numbers on the instruments.

What was direct, and what was inferred

X-ray diffraction is far more sensitive to oxygen than to hydrogen in this experiment. The direct observation is therefore the HCP oxygen lattice. Calling the hot phase superionic adds an interpretation about the hydrogen nuclei.

The authors support that interpretation with an anomaly in the lattice’s thermal expansion. As temperature increased, the HCP structure’s c-axis followed an S-shaped curve rather than expanding smoothly. Similar behaviour has been connected in simulations and earlier measurements to the onset of rapid hydrogen diffusion. The team places that onset near 1,700 kelvin at the relevant pressures.

That is persuasive but not equivalent to tracking protons or measuring current through the HCP phase. The study does not establish how conductive this structure is, whether its conductivity is direction-dependent, or exactly how quickly hydrogen moves. Those questions require experiments that combine structural measurements with electrical or diffusion probes.

Why earlier experiments saw a blurrier picture

Superionic water has a long history of prediction preceding observation. In 2018, researchers reported experimental evidence for superionic conduction in shock-compressed water ice, confirming that the exotic state could be made in the laboratory. Subsequent diffraction experiments established ordered oxygen frameworks under planetary-interior conditions.

A 2025 shock-compression study found mixed close-packed stacking above about 150 gigapascals and 2,500 kelvin. Its patterns were compatible with intergrown FCC and HCP layers, but the authors could not determine whether the disorder was an intrinsic equilibrium structure or a product of rapid shock loading.

The new static experiment helps separate those possibilities. Forestier and colleagues argue that HCP becomes the thermodynamically preferred oxygen arrangement above roughly 200 gigapascals. Across about 130 to 200 gigapascals, FCC and HCP can coexist through a martensitic transition. In this kind of transformation, layers shift cooperatively without first melting into an ordinary liquid, so faults and mixed stacking are expected.

A possible ingredient in unusual planetary magnetism

The pressure and temperature range overlaps with conditions expected deep inside Uranus and Neptune. Both planets possess unusual, strongly tilted magnetic fields, and electrically conducting material moving in their interiors is one candidate source. A different oxygen stacking arrangement could change conductivity, viscosity, plastic deformation and the way heat is transported.

Those consequences remain possibilities, not measurements from this experiment. The study characterises pure water in a minuscule laboratory sample. Ice-giant mantles are likely mixtures containing water, ammonia, methane and other components, with gradients and chemical reactions that a pure-water phase diagram cannot capture. Even the planets’ basic heat budgets remain active research territory, as a recent ScienceBlog report on Uranus’s slow cooling illustrates.

The authors suggest that an HCP lattice could produce direction-dependent ion transport, because the hexagonal structure is not equivalent along every axis. They did not measure that effect. Direct conductivity and mechanical tests at the same pressures will be needed before planetary models can assign the new phase a specific role in convection or magnetic-field generation.

A precise discovery, with a precise boundary

The experiment closes an important structural gap. It shows that water’s oxygen atoms can adopt a clearly resolved HCP lattice above 200 gigapascals and at thousands of kelvin, just as high-pressure theory had anticipated. The strongest pattern came from water at 219 gigapascals and 2,630 kelvin, conditions far beyond anything found naturally at Earth’s surface.

What it does not show is equally important. The diffraction images are not pictures of hydrogen flowing through a crystal, and they do not prove that hexagonal superionic ice governs Uranus or Neptune. They provide direct evidence for the predicted hexagonal oxygen framework, plus indirect structural evidence that it occupies the superionic regime. That narrower conclusion is still remarkable: under sufficient pressure, water can preserve crystalline order at a temperature that would melt iron, while its lightest nuclei are free to move.