Liquids aren't what they used to be. On the basis that all disorder must presumably lead to the same time-averaged structure, a liquid substance was assumed to be a unique state. But there is no fundamental reason why that should be so, for liquids are not after all structureless over short ranges. There is now plenty of evidence that several substances possess more than one kind of liquid state, which can interconvert through phase transitions1.

Among the materials thought likely to display liquid–liquid phase transitions are silicon, silica, germanium, phosphorus and water. One indicator of this behaviour, evident for silicon and germanium, is the existence of distinct structures in the amorphous, glassy state, a property called polyamorphism. The notion is that the polyamorphs are 'arrested' forms of the liquid phases.

Water is one of the most compelling cases, not least because of its ubiquity and long-standing status as an 'anomalous' liquid. Water's unusual density maximum at 4 °C and expansion on freezing have long been rationalized in terms of the directional nature of the hydrogen bonding between water molecules, which in turn seems to offer the possibility of a low-density, 'ice-like' liquid in which hydrogen bonding is preserved and a high-density liquid in which it is deformed or broken. Two-state models like this have a long history, but it's simplistic to imagine ordinary liquid water as being composed of an intimate mixture of such distinct regimes.

All the same, it's striking that many of the candidate systems for liquid–liquid transitions share with water either tetrahedral coordination or at least some degree of directionality in the associations between the component particles. Phosphorus is a curious case. A first-order liquid–liquid phase transition has been found experimentally2, but this seems to be an unusual transition between a molecular fluid and a polymeric liquid3.

Perhaps the first experimental claim for a liquid–liquid phase transition came from the seemingly exotic system of molten alumina–yttria (Al2O3–Y2O3) (ref. 4). Unlike several putative such transitions that happen in metastable high-pressure or negative-pressure regimes, this one takes place at ambient pressure. All the same, the melt must be supercooled, so it is hard to study experimentally without triggering crystallization.

Neville Greaves of Aberystwyth University in Wales and colleagues have now found5 a clever way to do that: by levitating a molten droplet in a stream of gas inside a furnace, eliminating all contact with a solid container. This has enabled them to verify that the transition is first-order — a point that has been contended for the analogous liquid–liquid transition of water. The researchers use X-ray scattering to identify the two states directly, clarifying how their atomic arrangements differ. The method, they say, might also be used to study other putative instabilities in deeply supercooled water.