by Aries Keck for NCSA
Champaign, IL — A biologist studying frogs can just pick one up by its clammy skin and flip it over to get a good look. But studying galaxies takes more than touch. Cosmologists can’t grab a few galaxies and hold them up to the light. And they can’t slice a galaxy open to see what’s inside.
Cosmologists are limited to only one point of view of their star-filled subjects, that of the earth and relatively near-earth space. Often they know only three pieces of information about location of the galaxies – a vertical coordinate, a horizontal coordinate, and, if the stars move in any of these directions over time, speed. At best, cosmologists are working with flat pictures of galaxies that may be faintly embossed with mere hints of depth.
John Dubinski is hoping to change all this. An astrophysicist and numerical theorist at the University of Toronto, he’s created a massive computer simulation that lets scientists manhandle not only individual galaxies, but whole clusters of them. It even crashes clusters into each other – simulating the collisions that produce merged galaxies. His nine simulations of individual galaxy clusters contain nine to twelve million particles each, and creating them required 60,000 CPU hours on the Alliance’s SGI Origin2000 supercomputer at NCSA.
Then, the make-believe galaxies were compared to observations of the real thing. “We brought the two together. The reality and the simulations,” says cosmologist Margaret Geller of the Harvard-Smithsonian Center for Astrophysics in Cambridge, Mass. The goal is to create the most realistic simulations possible – exposing the secrets that are obscured from our limited point of view.
Comparing simulations to observations is usually like comparing apples to oranges because most simulations are vastly oversimplified, according to Geller. With huge amounts of data and the computing power available at NCSA, however, the results and the comparisons that can be made to them are vastly more complex. Because the models are detailed enough to include information such as an inordinate number of individual galaxies, Dubinski says they are now operating in an environment that’s much closer to the true physics of reality.
“We are mainly interested in examining the evolution of a population of well-formed spiral galaxies as they fall into a forming cluster,” Dubinski says. “This approximation allows us to avoid the messy details of galaxy formation and focus on the stellar dynamics.”
In all, he replaced 300 dark halos with spiral galaxy models. These models then interacted and merged as they fell into the forming clusters. This is a different approach than usually used, Dubinski says, and it has given the model a rich complement of stars and a mass that’s more like reality.
Dubinski’s massively complex simulations let cosmologists like Geller and her doctoral student Daniel Koranyi feel around the galaxies, divining the masses of entire clusters and the distribution of mass across a cluster. In doing so, they’re hoping to answer one of the biggest questions in science – what is dark matter?
The mysterious missing mass of the universe, dark matter was first imagined in 1933. “It’s been with us ever since and we still don’t know what it is or where it is,” Geller says. Only 10 percent of matter is “stuff” that’s made up of protons, neutrons, and electrons. The other 90 percent is dark matter. Forming in halos around galaxies, it’s what keeps galaxies and clusters of galaxies from flying apart. “Galaxies are never alone,” Geller says. “They always have neighbors. And the masses of these clusters are the basis for our knowledge of dark matter in the universe.”
But determining the masses of galaxy clusters isn’t easy.
“Galaxy clusters may contain anywhere from a few dozen to thousands of galaxies in orbit around the common center,” Dubinski says. In building his models, he selected nine galaxy clusters from another large dark matter simulation. Each of these clusters had a range of properties similar to the observed data. Then he ran the simulation looking backward in time to see each of these nine clusters before it formed. Running the simulation backward to the beginning let Dubinski identify halos of dark matter forming around the galaxies. According to theories, galaxies are born in these dark matter halos. But modeling the creation of galaxies out of each halo requires incredibly difficult hydrodynamical calculations that eat up computer resources. To conserve computer resources and increase the amount of data that could be considered, Dubinski took a shortcut. He replaced each halo with a well-resolved model of a spiral galaxy.
The sophistication of Dubinski’s model let Koranyi see if two standard ways of measuring the mass of the clusters gave true results. To test, he simply worked out the mass using the standard method and compared that figure to the mass of the simulated cluster. While the method for figuring out the total mass of a cluster worked fine, he was surprised to find that a common way of figuring out the cluster’s mass distribution was far off the mark. “People said, ‘We know this method isn’t perfect,'” Koranyi says. “Turns out this method doesn’t really work at all.”
Geller says that astronomers have other, more accurate ways of computing mass distribution across a galaxy cluster, but they’re not as simple as the method Koranyi disproved.
It just goes to show, she says, that the sophistication of computer simulations drive astronomers to demand more exacting knowledge than they did in the past. “I mean, when people first started doing this, you wanted to know the mass to a factor of a few, now people want to know it to a few tenths of a percent.”
Watching galaxies grow and collide in the model uncovered another secret – one that gets to the center of the matter. In Dubinski’s simulations, all galaxies start out as spirals, spinning like pinwheels across the sky. But lurking in the center of these clusters is something else: massive football-shaped galaxies called ellipticals. Dubinski found he could form these blobs in his model of colliding galaxies.
“The simulations produce many remnants of these mergers that closely resemble real elliptical galaxies,” Dubinski says. It’s a result, he says, of having massive amounts of galaxies in the simulation. “Where previous work may have a dozen, we have a couple of hundred.”
The mountains of data that have led to such grand results present another dilemma. It will take some time for observational astronomers to catch up with all the information churned out by the models. “During the past few years, there has been a paradigm shift in supercomputing with the movement from vector machines to the new massively parallel machines,” Dubinski says. That has greatly increased not only how much data the simulations handle, but also their complexity. “With our simulations you essentially have complete knowledge,” he adds, “you have the true mass, the three-dimensional structure.”
And that’s letting astronomers finally get their hands dirty. “One of the things that makes astronomy different is that you can’t really experiment with the stuff you’re working on,” Koranyi says.
Geller believes that the models will shed light on the mysteries of dark matter. “I think the thing that’s important here is that simulations have come to the point where you can actually ask questions about galaxies – where you can really make fairly direct comparisons with observations. And you can learn physics from that,” Geller says. “It’s a profound change.”
This research is supported by the Smithsonian Institution.