How could a far smaller sample of data—of comparable quality to the data we had used ourselves—yield such a stringent constraint? It could not. Unfortunately, it turns out that the Chand analysis contained important errors, and the upper bounds on any change in α were subsequently relaxed. But the full story is more interesting than a mere mistake in a complex analysis.
By mid-2010 we completed the analysis of a large amount of new data from the Very Large Telescope (VLT) operated by the European Southern Observatory and obtained 153 new measurements. All the data our group had previously analyzed had come from the Keck telescopes on Mauna Kea in Hawaii. For this new VLT data, everything was different—the telescopes, the spectrograph, the detectors and the software used for the initial stages of the data analysis. This VLT data therefore provided a beautiful cross-check with our results from the Keck telescopes.
We thought it was possible that the new data would show no change in α at all or that they would show the same effect the Keck data did—with α appearing smaller at higher redshifts. What we actually found was truly astonishing and, if correct, will revolutionize some of our most fundamental concepts in physics.
The new VLT data showed not a smaller value of α at high redshift but a larger value, larger by just about the same amount as the Keck data is smaller. How can this be? Our immediate thought was that we were seeing evidence for systematic problems in both data sets. Add the Keck and VLT samples together, and to a good approximation, the combined sample shows no change in α with redshift. Problem solved. The constants are really constant after all.
But if that is the explanation, it requires two different systematic effects, one for each telescope, such that both effects are, independently, of the same magnitude but opposite sign. This is not impossible, although so far we have not managed to identify what this unknown pair of systematic effects could be.
We have discovered another curiosity, however. The Keck data cover a largish portion of the sky in the Northern Hemisphere, large enough to ask whether there is any “preferred direction” for the change in α seen with that sample. Put another way: Could it be that α changes not with redshift but with position on the sky? A simple analysis identified one particular direction for which that might be the case. Surprisingly, when the VLT data are analyzed independently, the same direction pops up. The VLT is in Chile and, on average, points to a very different part of the universe than the Keck telescopes do. Another coincidence? Possibly, but that now makes two coincidences.
What happens when we merge the old Keck and the new VLT samples? The result is positively intriguing: the directional dependence becomes highly significant. Deriving such a result by chance appears to be extremely unlikely. If the result is a fluke, we might expect a subset of the data to be generating a rogue result. With this in mind, we devised a simple test to iteratively reduce the sample, discarding one point at a time, to see how much data we needed to eliminate before the apparent spatial dependence of α vanishes. We found that we needed to throw away half the data before the chance probability reduced to a sufficiently unimpressive level! Again, perhaps this is a fluke. Despite extensive attempts, however, we have yet to find a combination of systematic effects in the data that could mimic a spatial dependence. Alpha appears to change spatially—across, perhaps, the entire observable universe. Any change with time is smaller and is currently below our detection sensitivity.
Reforming the Laws
If our findings prove to be right, the consequences are enormous, though only partially explored. Until quite recently, all attempts to evaluate what happens to the universe if the fine-structure constant changes were unsatisfactory. They amounted to nothing more than assuming that α became a variable in the same formulas that had been derived assuming it was a constant. This is a dubious practice. If α varies, then its effects must conserve energy and momentum, and they must influence the gravitational field in the universe. In 1982 Jacob D. Bekenstein of the Hebrew University of Jerusalem was the first to generalize the laws of electromagnetism to handle inconstant constants rigorously. Bekenstein’s theory elevates α from a mere number to a so-called scalar field, a dynamic ingredient of nature. His theory did not include gravity, however. Ten years ago one of us (Barrow), with João Magueijo of Imperial College London, and Håvard B. Sandvik, then also at Imperial, extended it to do so.