Your independent source for Harvard news since 1898

Your independent source for Harvard news since 1898

Right Now | Seeing Stars

Perseus in 3-D

May-June 2009

Part of the Perseus molecular cloud, as seen through the Smithsonian’s MMT telescope in Arizona. Self-gravitating structures—dense areas where stars are forming—appear as dark spots. The L1448 region is located in the upper-right quadrant of the image.

Part of the Perseus molecular cloud, as seen through the Smithsonian’s MMT telescope in Arizona. Self-gravitating structures—dense areas where stars are forming—appear as dark spots. The L1448 region is located in the upper-right quadrant of the image.

Image courtesy of Alyssa Goodman

Sitting at home at your computer, you--the armchair traveler--can now tour a region of the Perseus molecular cloud, a dense patch of interstellar medium located 700 light-years from Earth that is a hotbed for star formation. Just as Google's Street View feature lets you pull up an image of your own house and then "walk" down the street, new mapping technology developed by Harvard scientists lets anyone with a computer mouse inspect a piece of the interstellar cloud, moving it from side to side and even turning it completely around.

This technology is actually better than Street View, which doesn't allow looking inside buildings or going around to view them from the back. And it gives a tour of a place that humans have never been, but only observed with telescopes.

The advent of radio astronomy after World War II enabled imaging the optically opaque innards of molecular clouds. In subsequent decades, scientists developed a rough sketch of the star-formation process: turbulence, arising from forces that weren't well understood, moved particles around until small, dense pockets developed in which particles came close enough that gravity took hold and the masses collapsed into themselves to become stars.

An algorithm written in the 1990s to pick out the clouds where stars would form, and map them in relation to each other, turned out to be a blunt instrument for analyzing the cloud innards' structure, says professor of astronomy Alyssa A. Goodman, one of the project leaders: "People were applying this algorithm blindly, using it on a scale that was much too small." Although scientists didn't yet know how to build a better algorithm, they could see that the maps it produced--clusters of dense pockets, with no space in between--did not correctly predict where stars would form. Nor did the maps re?ect the fact that a smaller area of even higher density can exist within a given dense area. The gas formations' structure was hierarchical-like clouds within clouds--but the maps showed only disparate, non-overlapping blobs.


Courtesy of A. Goodman et al, “A role for self-gravity at multiple length scales in the process of star formation.” Nature 457:63-66 (2009).

The image on the left shows data gathered from the L1448 region of Perseus, analyzed using the dendrogram algorithm devised by Erik Rosolowsky, Alyssa Goodman, and colleagues. By taking into account the hierarchical nature of structures in the cloud, the algorithm enables scientists to see, within already dense parts of the cloud, even denser areas. An image generated using an earlier, widely used algorithm (right) depicts the same region as disparate, non-overlapping blobs.

Erik Rosolowsky, then a postdoctoral fellow working with Goodman (now a professor at the University of British Columbia at Okanagan), wrote a new algorithm based on the dendrogram, a hierarchical structure used to analyze natural phenomena in fields from genetics to botany. This algorithm not only produced maps with results that more closely matched actual telescopic images of the Perseus region, but also led to an important discovery. Earlier astronomical simulations of star formation had routinely omitted gravity, presuming its in?uence to be small relative to other forces, except at very high density. The new results, which Goodman and Rosolowsky published in Nature, suggested that gravity is in fact very important in star formation, exerting in?uence across distances much greater than previously thought.

The fact that they checked their work by looking at a photograph reminds us that the human brain is still our best tool for pattern recognition, more powerful than the fastest computer or the most precise algorithm. "Face-recognition software has just now gotten to the point where it can distinguish a face, and maybe tell who it is," says Goodman. "You can see a fuzzy cell-phone picture of somebody you haven't seen in 10 years and say, 'Hey, that's Mike!' A computer cannot come anywhere close to doing that, and probably won't for 10 years."

To create a visual representation of the dendrogram analysis, the scientists used a program adapted from medical-imaging software by Michael Halle, an instructor in radiology at the Harvard-affiliated Brigham and Women's Hospital, and colleagues. But for submission to Nature, they converted their image of the Perseus region, representing four million data points, to a common file format: PDF, viewable with Adobe Reader, a common, free software program. Theirs was the first-ever three-dimensional PDF published in a major scientific journal. The printed copy of the journal had two-dimensional figures, and people who saw that version found it less convincing: "It has a take-my-word-for-it feel," reports Goodman, the former director of the Initiative in Innovative Computing (see "Science's 'Third Branch,'" May-June 2007, page 56). She points to the 3-D PDF on her screen: "This is what convinces people."