Meet a RC researcher- Greg Salvesen


          Greg Salvesen is a graduate student studying astrophysics and planetary sciences at the University of Colorado, Boulder. When Salvesen began graduate school, he knew that he wanted to research black holes. Because black holes do not allow light to escape, they are impossible to directly observe. Fortunately, black holes are surrounded by discs of gas that orbit very rapidly, at approximately half the speed of light. As they orbit, these discs become so hot that they radiate x-ray light, which astronomers can observe and infer the black hole’s presence. Salvesen’s research focuses on using computer models to simulate the structure, physics and behavior of these discs.

            Disc formation begins with two differently-sized stars in a binary orbit. If one of those stars is especially massive, it will die quickly and leave behind a black hole, around which the smaller star will continue to orbit. Late in its life, that star will become a Red Giant, a puffed-up older star made up of loosely-bound gas. In the same way that the moon’s gravity tugs at the on Earth as it orbits, drawing water towards the side to which it is nearest to create tides, the black hole’s gravity pulls on the sides of the star as it orbits. When the star becomes a Red Giant, its gas is bound loosely enough for the black hole to tug some of it away from the star completely. That gas begins to orbit the black hole, spiraling towards its center. As gas is sucked in, it is replaced in the disc by more matter drawn from the star. This relationship persists throughout the Red Giant phase of a star’s life, which can last millions of years.


Black Hole

Black hole siphoning gas off a red giant

            Astronomers know from observation that gas in the disc spirals inward toward the black hole as it orbits, but the reason for this phenomenon is not immediately apparent. A molecule of gas should theoretically continue on a circular orbit indefinitely, like the Earth orbiting the sun, rather than moving gradually closer to the center. Salvesen explains, “It turns out that when the gas is perturbed by a magnetic field, it goes crazy. That’s how molecules move from the outer part of the disc to the center.” That perturbation has become the focus of Salvesen’s research, which he uses computer simulations to conduct.

            Because the disc is too large for all of its physics to be simulated, Salvesen’s simulation zooms in on a small, box-shaped patch of gas. The box is divided into 20 million cells, each of which is initially stable. Salvesen perturbs the gas, imitating the effects of a magnetic field passing through it, and collects data on the series of reactions that follow. He says, “We compute how the gas behaves in all 20 million cells, which all have to communicate with each other. Every step you take forward in time has 20 million parts.” Although the equations that describe the reactions in each cell are straightforward, enormous computational power is necessary to run the simulation due to the sheer number of cells. Salvesen explains, “If you gave your laptop computer to Abe Lincoln and said, ‘Hey, run this simulation for me,’ back in 1860, it would just now be finishing because it requires that much computational power. But if you have a supercomputer, it’s a different story.”

            Salvesen runs his simulation on Janus, the supercomputer on campus. Janus “does the grunt work to perform the simulation at every one of those 20 million little zones, and then to analyze the data.” For a recent job, Salvesen used 288 nodes, each of which contains 12 processor cores, creating a total of 3,456 processors. A standard laptop, by comparison, has 4 processors. Even with this amount of computational power, the longest period of time Salvesen has been able to simulate is 0.01 seconds, a task that requires one million processor-hours on Janus. “It’s rare to have that kind of resource completely available to you. Usually you have to write a very serious proposal and submit it to a supercomputing facility everybody in the country is competing for,” Salvesen says. “If you say, hey, I’d like to have 10 million hours on the supercomputer, maybe they’ll give you one. That completely changes what your plans were.” Because Janus is local, Salvesen is able to use it with far fewer restrictions with the support of Research Computing, which oversees the supercomputer.

            “Research Computing is easy to work with,” Salvesen says. “It’s a very reasonable process for requesting and then being granted computer time, and it’s quick. They really try to work with you.” When he encounters technical difficulties, Salvesen turns to Peter Ruprecht, a senior high-performance computing analyst, for help. “Pete’s great. He’s always worked quickly with me. He’ll come to my office, sit down with me, and help me figure out what’s happening. You’d never get that at a national supercomputer facility. To have somebody local who’ll just say, ‘I’ll swing by your office and we’ll work this out,’ is really nice.”