What is a metal? It depends on who you ask, for like most colloquial words used in science it has meant different things at different times. The etymology from ancient Greek probably refers to any substance mined or quarried. Most folks would expect a metal to be hard and silvery-shiny; some sophisticates may demand electrical conductivity. Some will refer you to the periodic table; physicists may expect a partially filled electron band. (No one else, however, has yet come up with a definition quite as eccentric as that used by astronomers.)

But studies of atomic clusters reveal that the question 'metal or not?' depends not just on chemical composition but on size. The whole concept of electronic bands, on which discussions of metallic behaviour generally rest, is contingent on the overlap of a sufficient number of atomic orbitals to create a continuum of energy states. Or is it? A study by Bowlan et al. suggests that even the smallest sodium clusters can be considered to exhibit metallic behaviour1.

Quantum calculations and many experiments have seemed to imply a transition from molecular-like clusters to truly metallic particles as the number of atoms increases for elements that are bulk metals. Just a few examples: early work on mercury clusters Hgn suggested an onset of metallicity above about n = 40 (ref. 2); Mossbauer spectroscopy shows that the inner core atoms of about 55 gold atoms have a different charge density from those of bulk gold3; and photoelectron spectroscopy of magnesium clusters Mgn shows bandgap closure at around n = 18 (ref. 4).

Such findings raise questions about whether the popular, theoretically tractable jellium model is a good description of the smallest clusters of metal atoms. This more or less insists on regarding the cluster as a metal: a spherical droplet in which the nuclei and core electrons are embedded in a delocalized blob of valence electrons. Nonetheless, jellium seems a good approximation for predicting the electronic properties, such as polarizability5, even of some relatively small metal clusters.

Bowlan et al. now confront that picture with a particularly stringent test. Whereas earlier measurements of polarizability of sodium clusters were performed at temperatures where the clusters are liquid, they have made measurements at 20 K, well below the freezing point.

They argue that the most general definition of a metal in terms of electronic behaviour is that a metallic object cannot sustain an internal electric field. The charge density will always rearrange to eliminate any inhomogeneity, meaning that there can be no electric dipole moment. Quantum-chemical calculations suggest that small sodium clusters should have a non-vanishing dipole moment caused by imperfect screening of the ion cores by valence electrons6. Yet the cryogenic molecular-beam experiments of Bowlan et al. show that, not only are the polarizabilities smaller than reported in previous experiments, but the dipole moments are orders of magnitude smaller too and in fact very close to zero for all clusters, even dimers and trimers. On this measure, at least, sodium is a metal all the way down.