The second is currently defined as the time taken to complete 9192 631 770 oscillations between two energy levels in a caesium atom. The current generation of caesium clocks boast an accuracy of one part in 1015 - equivalent to an error of less than one second in 30 million years. However, this accuracy could be improved by several orders of magnitude if optical transitions were used to measure time rather than microwave transitions in caesium. The main problem when building an optical clock is to relate these optical frequencies to the much lower microwave frequencies that are used to define the second - a task that once required an extremely complex "frequency chain".

Last year Udem, then at the Max Planck Institute for Quantum Optics in Garching, near Munich, and colleagues at Garching and Bath University in the UK, developed a "frequency comb" that made this task much easier. To make the comb the output from a pulsed femtosecond laser is sent through a silica fibre with an array of submicron-sized holes running along its core. When the output is measured as a function of frequency it consists of a series of equally spaced spikes that can be used to measure the difference between two widely separated frequencies.

The NIST team used this approach to measure the frequency of an electric quadrupole transition in a single mercury ion in a cryogenic Paul trap, and the frequency of an optical transition in a collection of 10 million laser-cooled calcium atoms in a magneto-optic trap. In addition to leading to a new generation of standards, the experiments also found no evidence for the variation of these frequencies over time - as has been predicted by some unified theories.