Massachusetts Institute of Technology
Ryan McKinnon's research at the Massachusetts Institute of Technology is concerned with the smallest parts of enormous things.
He studies the physical processes that shape galaxies, the billions of enormous star clumps that comprise the universe. Clues to those activities are available only by capturing and analyzing light from stars in the galaxies.
But gas and dust also swirl through these conglomerations. The tiny dust grains contribute little to galactic mass but absorb 30 percent or more of starlight and modify whatever light and radiation does escape. "It really obscures the view that we have of galaxies and our understanding of the physics within these galaxies," says McKinnon, a Department of Energy Computational Science Graduate Fellowship (DOE CSGF) recipient.
In his research with advisor Mark Vogelsberger, McKinnon develops and applies computational models of how dust grains interact with fluid gas in a galaxy. "If we can come up with predictions of how these dust grains are distributed within a galaxy, we can better understand" how they block light and correct for it when analyzing images of real galaxies.
It's like flying over a city and trying to photograph it through smog, McKinnon says. "You're not going to get a true sense of what the city looks like. A lot of the light is scattered and affected by the smog." If observers know something about the city, such as where factories and parks are, they could subtract the smog's effects. But "we can't directly go to the galaxy and see what it looks like" without dust obscuring and altering the emitted light. "That's why we try to model it on the computer and predict what's going on."
Most astrophysics simulations include dust's effects only after most of the calculating is done, when creating images of computationally evolved galaxies. Those approaches often distribute dust uniformly rather than calculating where grains should be concentrated. McKinnon says he and Vogelsberger are "some of the first people to try to evolve the distribution of dust and the effects of dust directly in our simulations as they're running instead of adding it back in afterward."
It's clear why some models omit dust: Even the largest grains are just a millionth of a meter in diameter and their size affects how they influence starlight. Particle size also is a moving target, as the specks constantly collide, breaking into fragments. Finally, particle distribution affects how quickly they change size and size affects how they move through the galaxy.
McKinnon and Vogelsberger are implementing the dust physics in AREPO, an astrophysics code developed by collaborator Volker Springel of the Heidelberg Institute for Theoretical Studies. They've benchmarked the dust code against standard problems and now are using Harvard University's Odyssey computing cluster to simulate individual galaxies. If that goes well, the researchers will model the evolution of a region of the universe, including the simultaneous development of several galaxies.
"Our work is definitely going to help people who are observing real galaxies address the challenge of understanding what a galaxy really looks like behind layers of dust," McKinnon says, but it also could improve computational science in general. "People doing astrophysical simulations probe the cutting edge of numerical methods and data analysis and algorithms."
McKinnon says the computational aspects interest him as much as the astrophysics. They also were a large part of his 2015 Lawrence Berkeley National Laboratory practicum with astrophysicist Peter Nugent.
Part of Nugent's research is finding transient objects - astrophysical phenomena like exploding stars that suddenly appear - by comparing images of the night sky with earlier images. McKinnon's project targeted the mysterious Planet 9, a distant, faint member of our solar system that telescopes haven't detected but whose presence can be inferred from measurements. Nugent wants to analyze archival data from the Infrared Processing and Analysis Center (IPAC), a sky survey based at the California Institute of Technology, for signs of the elusive planet.
"An object may be too faint to appear on one image, but if we stack these images in a clever manner, we could see some signals of Planet 9 or objects that are quite far out in the solar system," McKinnon says. He helped develop the image-processing pipeline, writing algorithms to pull IPAC data, align it and subtract it from reference images. Much of his work was done on high-performance computers at DOE's National Energy Research Scientific Computing Center.
McKinnon soon will be engaged in a different hunt, after graduating in 2018 or 2019. He's unsure of his next post, but "whatever I choose will be computationally inclined."
Image caption: Density of solid dust grains (colored regions) in a Sedov blast wave, a spherically symmetric point energy injection in a gaseous medium. Hydrodynamical equations are solved on an unstructured three-dimensional computational mesh (two-dimensional slice shown in black), and dust grains are coupled to hydrodynamics via a drag force. Because of drag, dust density peaks behind the gas blast radius (white circle). Image courtesy of Ryan McKinnon.