How constant are the 'constants' of nature? Ever since Dirac¹ speculated that temporal changes in the values of various dimensionless physical constants might have occurred over the lifetime of the Universe, vigorous but fruitless attempts have been made to detect such variability. The fine-structure constant, =2e ²/hc, which characterizes the strength of the electromagnetic attraction between photons and electrons, has attracted particularly aggressive attention. Now an Australian-British group is claiming in Physical Review Letters ² that it might have detected evidence for a varying , at a level of about one part in 10⁵, over roughly half the lifetime of the Universe. The authors investigated possible time variation in using cosmological observations of the absorption spectra of distant (high-redshift) quasars. The result must still be considered extremely preliminary, because of the very real possibility of systematic errors in this type of measurement. But given its importance if correct, it can be expected to provoke considerable scrutiny.

The modern expectation of variation in the fine-structure constant arises from theories such as those unifying gravity with the other forces of nature. The best terrestrial limit on the time variation of outside the laboratory is based on examination of the decay products of the Oklo phenomenon — an ancient natural fission reactor discovered in 1972 in the Oklo uranium mine in Gabon, West Africa. Analysis³ of the 1.8-billion-year-old decay products gives a range for a fractional change in the fine-structure constant (/) of between 0.9 × 10^-7 and 1.2 × 10^-7 over this period, which would scale linearly to a limit of about one part in 10⁶ over the lifetime of the Universe. But there is no reason to expect the time dependence to be linear (an oscillating is even allowed in some schemes) and longer time baselines afforded by our ability to measure by observing the atomic and molecular wavelengths of cosmologically distant objects offer a powerful alternative route.

Wavelength spectra of cosmologically distant quasars provide a natural laboratory for investigating changes in . Dark, narrow lines in quasar spectra are produced by absorption of radiation in intervening clouds of gas, many of which are enriched with heavy elements. The wavelength separation between two lines produced by absorption from alkaline atoms and ions (an alkaline doublet) is proportional to ², so any small variation in this separation will be roughly proportional to , to a first approximation. Because quasar spectra contain doublet absorption lines at a number of redshifts — and so at different times in the history of the Universe — it is possible to check for time variation in simply by looking for changes in the doublet separation of alkaline-type ions with one outer electron (such as C³⁺ or Si³⁺), as a function of redshift. Although this sounds straightforward, any change in will be very small, and so the accuracy of measurement needs to be high.

Not surprisingly, this method has a long history of null measurements stretching back to its first application by Savedoff⁴ in 1956. We obtained the most accurate estimate⁵ on the fractional change /, prior to the present paper, of 3.5(5.5) × 10^-6. This limit was reached by comparing the absorption lines of neutral hydrogen and carbon in a cloud of intergalactic gas at a redshift (z) of 1.8 lying nearly two-thirds of the way across the Universe. The relative wavelengths of the lines would have been shifted by any change in the fine-structure constant, but no such change was seen. This is not that surprising: theories allow enormous latitude in the actual variation of ,and a detection at the current level of measurement sensitivity will always be unexpected.

Webb et al. ² introduce a new variant of this technique that compares the absorption wavelengths of magnesium and iron atoms in the same absorbing cloud, which they demonstrate in a companion paper⁶ to be an order of magnitude more sensitive than the alkaline-doublet method. They observe a number of intergalactic clouds at redshifts from 0.5 to 1.6, seen in absorption against a background of quasars. For the entire sample they find / = -1.1(0.4) × 10^-5, consistent with a null result (to within three standard deviations), but crucially most of the signal comes from a small subset of quasars near a redshift of 1. Restricting the sample to z > 1 systems gives a significant change of / = -1.9(0.5) × 10^-5. Such a result would be consistent with little time variation in the past 1.8 billion years since the Oklo event, but larger variation at earlier times when z > 1, although it then becomes hard to understand the previous null result⁵ at a redshift of 1.8. Webb et al. ² also seem unimpressed by the redshift dependence they have apparently uncovered, and interpret their results as "stringent upper limits on any possible time variation rather than a positive detection of change", in part because "a genuine physical effect confined to one specific epoch ... does not seem at present to be well motivated by theoretical expectations".

Of course, the real problem (as Webb et al. discuss at length) is the subtle and not-so-subtle systematic errors that, alas, in this type of measurement are potentially legion. The authors mention most of the likely culprits: the wavelength calibration, uncertainties in the laboratory wavelengths of iron and magnesium, and the effect of unresolved velocity substructure in the absorption lines. The redshift dependence of the effect makes the last by far the most likely: it is not hard to imagine that some of their many lines of sight could have hit an absorbing cloud or two in which their assumptions about the relative distribution and kinematics of light and heavy ions were not correct.

Another potential source of systematic error is the use of both very strong and very weak absorption lines of singly ionized iron: any small perturbation would affect the latter much more than the former and could skew the results. Still, for now, and until we can really probe the details of this investigation, subtle effects in the structure of the absorption lines of the two atoms or slight systematic problems in the wavelength calibration remain the most likely reasons. Because of these problems, it is indeed probably best to regard this measurement as an upper limit, albeit a provocative one.

This new technique ultimately could and should be pushed to higher redshift by using shorter-wavelength lines. And the only thing currently limiting the alkaline-doublet method⁵ is the laboratory measurement of the doublet separation of triply ionized silicon lines. The final irony is that both techniques are limited in sensitivity or scope by poorly defined laboratory wavelengths. Webb et al. conclude their paper with a plea for more laboratory work to be done, with which we can only concur.