A great deal of modern astronomy and cosmology is based upon experimental observations of clusters of luminous matter.
Typically, one devised a model of how one expects the structure of the universe to develop over time, implements it on a Monte Carlo basis thousands of time, evaluates the results statistically, and compares them to observations. Models that rarely produce results that are statistically similar to our universe are ruled out. Models that often produce results that are statistically similar to our universe are considered to be consistent with the empirical evidence.
But, statistically identifying clusters of luminous matter in astronomy observations is tricky business. A small number of individual stars or galaxies with peculiar velocities (a vector quantity that measures both speed and direction), or a subtle misjudgment of red shift or angle in an observed luminous object, can add a great deal of statistical noise to an effort to accurately identify clusters of stars. These systemic uncertainties in the observations of the real world universe can make it hard to rule out any models, since so many possibilities are consistent with observation to within reasonable margins of error.
A couple of papers over the last few days (here and here) have used a more statistically robust way of measuring many of the phenomena normally studied using analysis of clusters of stars and galaxies. Rather than measuring clusters of stars, the technique measures the location and extent of clusters of cosmic voids - the space between luminous objects in space.
In other words, the astronomers are literally measuring "nothing", rather than measuring the "stuff" itself.
Cosmic voids can be measured using essentially the same data points as traditional cluster analysis, but this approach to the analysis is more statistically robust. Since the "thing" observed in part of a cosmic void is the complete absence of luminous matter, uncertainties regarding its properties and red shift are gone. And, since most of space is empty, cosmic voids are more spatially expansive than stars or galaxies, so one does not have to fix their location so precisely to accurately capture their statistical distribution.
Yet, the clustering of cosmic voids is almost as useful as the clustering of luminous matter in empirically testing questions about the large scale structure of the universe, such as cosmological homogeneity in the universe that has been analyzed traditionally in the past, or traces of "inflation" in the early moments of the universe revealed by "clustering fossils." Any computer model of how large scale structure arises can be quantified in terms of its cosmic void structure just as easily as it can be quantified in terms of its clusters of luminous matter, and these predictions can then be compared to empirical data about cosmic voids which is more statistically robust.
Kudos to the authors for coming up with this clever and non-intuitive means of data analysis.
* Teeraparb Chantavat, Utane Sawangwit, P. M. Sutter, Benjamin D. Wandelt, "Cosmological Parameter Constraints from CMB Lensing with Cosmic Voids" (September 11, 2014).
* Kwan Chuen Chan, Nico Hamaus, Vincent Desjacques, "Large-Scale Clustering of Cosmic Voids" (September 12, 2014)