NATURE|Vol 440|2 March 2006 NEWS & VIEWS 35 with what we know about the growth of silicon nanowires by molecular-beam epitaxy. This is a technique based on the VLS method, but in which silicon is supplied not as a gas but as a directed beam of atoms. Molecular-beam epitaxy usually requires an ultra-clean envi- ronment, and here the transport of silicon as well as that of gold occurs through diffusion on the silicon surface, not through the gas 9 . Ostwald ripening would make the growth of ordered arrays of millions of essentially iden- tical silicon nanowires — a prerequisite for nanoelectronic applications — exceedingly dif- ficult. Even tiny variations in the size of the gold nucleation droplets would lead to unacceptable variations in the length and diameter of the sil- icon nanowires. There is, however, an escape clause mentioned by the authors that is also in agreement with our own preliminary experi- mental results. A tiny amount of oxygen — as is present under most technological growth con- ditions, but not in the ultra-clean, high-vacuum environment used in the authors’ experiments — might efficiently block the diffusion path of gold on the silicon surface. This would render the gold droplets independent of each other, as has been assumed for the past 40 years. In this way, Hannon and colleagues’ results could resemble the oxygen-in-silicon story of the 1970s. At that time, manufacturers of inte- grated circuits were concerned that silicon crys- tals contained some oxygen atoms from the crystal-growth process, and pushed the manu- facturers of silicon wafers to eliminate these remnants. It was subsequently observed that wafers with negligible oxygen content produced less reliable electronic devices. It turns out that oxygen in fact precipitates onto small regions of the wafers, creating traps for detrimental metal- lic impurities in a process known as gettering 10 . Today, the oxygen content of silicon wafers is exactly specified for the best possible getter- ing performance. Analogously, the indirect conclusion from the fact that Hannon and colleagues 1 observe an effect under highly controlled, low-vacuum conditions that is not observed under less severely controlled condi- tions could be that a little added oxygen impu- rity — although not too much — is beneficial for silicon nanowire growth. That would be an unexpected and useful message: sometimes extremely clean is just too clean. ■ Ulrich Gösele is at the Max Planck Institute of Microstructure Physics, Weinberg 2, D-06120 Halle, Germany. e-mail: goesele@mpi-halle.de 1. Hannon, J. B., Kodambaka, S., Ross, F. M. & Tromp, R. M. Nature 440, 69–71 (2006). 2. Morales, A. M. & Lieber, C. M. Science 279, 208–211 (1998). 3. Wagner, R. S. & Ellis, W. C. Appl. Phys. Lett. 4, 89–90 (1964). 4. Givargizov, E. I. Highly Anisotropic Crystals (Reidel, Dordrecht, 1987). 5. Haraguchi, K. et al. Appl. Phys. Lett. 60, 745–747 (1992). 6. Ostwald, W. Z. Phys. Chem. 34, 495–503 (1900). 7. Wagner, C. Z. Elektrochem. 65, 581–591 (1961). 8. Lifschitz, I. M. & Slyozov, V. V. J. Phys. Chem. Solids 19, 35–50 (1961). 9. Schubert, L. et al. Appl. Phys. Lett. 84, 4968–4970 (2004). 10. Tan, T. Y. & Tice, W. Phil. Mag. 34, 615–631 (1976). The most useful scientific models are those with the fewest parameters and the least- disputable assumptions. By testing real data against these ‘null models’, researchers are insured against over-interpreting their data. If the null model can successfully characterize the data, there is often no compelling reason to go beyond it to seek explanations. In a new look at how species coexist in coral-reef communities, Dornelas et al. (page 80 of this issue) 1 have found a significant departure from the neutral theory of biodiver- sity 2 — community ecology’s most recent and elegant null model. They also provide empiri- cal support for an often overlooked compo- nent of niche theory, one of ecology’s oldest and most influential explanations for the coexistence of species. Neutral theory is one of the most exciting conceptual advances in ecology in decades. It provides a simple tool to assess the degree to which variation in the composition of natural communities can be explained solely by ran- dom demographic processes. There are only two major assumptions. The first is that every individual in a community has the same proba- bility of birth, death, migration and speciation. Differences among species are assumed to be irrelevant when it comes to predicting commu- nity composition. The second is that the com- munity is characterized by a zero-sum game — when one individual emigrates or dies, the space it leaves is immediately occupied by another individual. Everyone, including the theory’s architect Steve Hubbell, knows that species differ in many ways. After all, not all coral species participate in the annual ‘coral- reef orgy’ mass spawning event; some species brood internally every month. The real ques- tion that neutral theory asks is whether or not we need to take these differences into account to predict species-distribution patterns. Recent tests of the neutral theory have cen- tred on the shape of species-abundance curves 3 . As prima facie evidence against non-neutral systems, Hubbell offers the successful fitting of zero-sum multinomial functions, which meet the requirements of neutral theory, to curves from data collected in Barro Colorado Island, Panama 2 . Since then, others have made similar comparisons using these and additional data sets, and have found both agreement with and departures from the predicted curves. More recent work has shown that most of these com- parisons are weak, primarily because more than one ecological mechanism can give rise to very similar theoretical distributions, and the ‘best-fit’ distribution depends on what measure of ‘best-fit’ we use. If any model can fit, the shape that the species-abundance curves ultimately take may not be so important 3 . In a paper that will turn our attention in a completely new direction, Dornelas et al. 1 describe an approach that will join an emerg- ing next generation of tests of the neutral theory. Out go the old equivocal comparisons of the fit of different models to species- abundance distributions, and in come the new — tests that actually relate theoretical assump- tions to the biological reality inherent in ecological communities 4,5 . Dornelas et al. set out to see whether local communities really do show the vast differ- ences in species composition predicted by the demographic randomness inherent in neutral theory. Instead, they found something unex- pected and initially puzzling. Coral communi- ties along a traverse spanning part of the Indian and Pacific oceans are dramatically less similar to each other and even more variable than predicted by neutral theory. But the results are in the opposite direction to those Figure 1 | Corals with a difference. These two shallow-water coral assemblages occur in the same habitat on the Houtman Abrolhos Islands, Western Australia, and are evidently dissimilar. Dornelas et al. 1 find that coral communities along their study transect are much less alike and more variable than predicted by neutral theory — which, they suggest, can be explained by environmental variance. ECOLOGY Corals fail a test of neutrality John M. Pandolfi Ecologists continue to wrestle with a central question in biodiversity studies — the prediction of species’ distributions in various environments. A merger of different theories is the long-term prospect. B. GREENSTEIN Nature Publishing Group ©2006