One of the most profound features of the universe is that it is stratified. Our everyday world depends hardly at all on the details either of atoms or of galaxies, and the feeling is mutual. Were it otherwise, science would not be possible: we could not know anything without knowing everything. Sometimes, though, the levels of reality do get jumbled, with odd effects.

This past March a group of cosmologists--Edward Kolb of Fermi National Accelerator Laboratory and Sabino Matarrese, Alessio Notari and Antonio Riotto of the Italian National Institute of Nuclear Physics--argued that the acceleration of cosmic expansion, among the biggest mysteries of modern science, is one such effect. It could be the most elegant explanation for acceleration yet proposed, requiring neither exotic forms of energy nor new laws of physics, merely a careful accounting of how gravity interconnects structures of vastly different size. Or it could be a case of cosmological cold fusion.

Cosmologists routinely assume that the detailed arrangement of matter plays no role in the grand scheme of things. Their standard model treats the universe as though its density did not vary from place to place but had a uniform, average value of one atom per cubic meter. They solve for the expansion rate of this averaged universe and equate it with the average expansion rate of the actual universe. Individual patches of space may expand faster or slower, but researchers reckon that the discrepancies are localized.

The trouble is, averages can be deceiving. A golf ball on average is a perfect sphere, but it does not fly like one. The dimples on its surface can double or triple the distance the ball travels. Gravity, like the behavior of air flowing over the dimples, is nonlinear, which led cosmologist George Ellis of the University of Cape Town in South Africa to suggest in the 1980s that the fine-scale texture of the universe might affect its large-scale behavior. This phenomenon is known as backreaction. Analogies occur in fields besides cosmology. Sound, for example, is usually stratified: it can be thought of as the sum of waves of various wavelengths, each of which ripples through a room as though the others were not even there. Yet when nonlinear processes operate, different wavelengths can cross-talk and even shift the average air density.

What cosmologists should do is track the gravitational effects of matter in all its irregularity--to take the average after they solve the equations rather than before--but that is a tall order. So although they widely agree that backreaction occurs, they argue over how big it is.

Kolb and his colleagues claimed a huge effect, but it relied on a linkage not only from small to large but also from large to even larger--basically, attributing the acceleration of the observable universe to matter beyond the observable universe. That sounds impossible by definition. Distant matter may have been in contact with our universe long ago before falling out of touch, but critics such as Christopher Hirata of Prince-ton University and Uro Seljak of the International Center for Theoretical Physics in Trieste, Italy, pointed out that it cannot have an ongoing influence on us without violating relativity theory.

Kolb's team acknowledged making errors and is coming out with a new paper going back to a purely small to large backreaction. This approach might explain the so-called cosmic coincidence: why acceleration kicked in around the same time the growth of galaxies became strongly nonlinear. The sharp increase in galaxies' density could have cascaded up the line and produced a decrease in the average cosmic density, which would have accelerated ex-pan-sion.

But earlier calculations by Seljak and others indicated that such an effect would not have been strong enough, and even cosmologists who are sympathetic think there is a long way to go. "At this point, this is an idea of how things could work out," says Sysky Räsänen of the University of Oxford. "It's something to motivate calculations, not something backed up by calculations."