Searching for Rare Objects

Astronomers, like other research scholars, specialize in the investigation of objects whose importance they recognize. How, then, are new kinds of objects discovered, ones that were previously unknown or unrecognized? One way is by accident. Today modern telescopes, with their capability of imaging or surveying large regions of the sky in the course of probing particular kinds of celestial bodies, will automatically uncover all sorts of other things. For instance, the Infrared Array Camera (IRAC) on the Spitzer Space Telescope obtained extremely sensitive, multicolor images of the sky across a region as large four full moons in a search for faint galaxies. But among the many galaxies in the image were hundreds of thousands of bright points, mostly stars of all types but also maybe other things, all appearing point-like only because they are far enough away. Discovering the unexpected, as well as understanding the expected, are both parts of the success of large, systematic studies.

Being able to spot those unusual objects, however, is usually not easy. One of the most common techniques in astrophysics is the classification of astronomical objects by their observed, quantified color and brightness,
rather than by parameters like mass or size which are always much harder to determine. A plot of the luminosity versus the redness of each of the many unknown sources, for example, will usually end up with the data points in clusters, each cluster containing objects with similar physical properties - perhaps normal stars, or unusual stars, asteroids, galaxies, or otherwise. Making sense of it is an active area of astronomical research.

SAO astronomer Massimo Marengo, together with his physicist wife, has published a paper on a new technique to classify rare astronomical sources using these kinds of plots without needing to assume anything about the nature of the sources. Their technique is based on statistical methods that measure the distance between neighboring data points in a plot of properties (color versus brightness for example) and calculate whether or not that distance implies the two objects are likely to be physically similar. If there are any relatively isolated points, they are in this way easily identified as being rare, and suited for more detailed followup studies. The plots are powerful in part because they ignore where the sources are in the spatial image, and instead look at how their observed
properties compare to the properties of other objects.

These scientists applied their method to images from the IRAC search for faint galaxies. It turns out that so-called brown dwarf stars - stars that are only slightly larger than planets and barely able to burn nuclear fuel - have an infrared color that is enough different from other stars and galaxies that their data points on these plots usually fall outside the main groupings, and they can therefore be readily spotted amidst hundreds of thousands of other data points. The new technique, which is quite general and can work with any combination of properties and sources, offers a powerful new way to study systematically large quantities of data and search for the rare, perhaps unexpected sources that help to tell the full story of the cosmic zoo.