Ohio State University
As a child in California’s Bay Area, Thomas Holoien used a small telescope to pick out planets and stars from the night sky.
Holoien (pronounced ho-LOIN) is still at it, but on a scale that spans the night sky.
Holoien, an astronomy doctoral student, is part of the All-Sky Automated Survey for SuperNovae (ASAS-SN, pronounced “assassin”) at Ohio State University. Using small telescopes based in Hawaii and Chile, ASAS-SN automatically images the entire sky about once every three days. Working with the project’s founders, astronomy professors Krzysztof Stanek and Chris Kochanek, Holoien uses these pictures to find supernovae – exploding stars – and other transient objects.
Analyzing the nightly images is a big data problem. Solving it lets Holoien, a Department of Energy Computational Science Graduate Fellowship (DOE CSGF) recipient, merge his astronomy interests with his computational science skills.
It’s also a big change in direction. Holoien, fascinated by Japanese culture, initially earned a degree in East Asian Studies with a minor in Computer Science from Stanford University. “I didn’t have a lot of time to experiment much in college until my senior year,” he says, when he took astronomy courses and discovered a deeper interest.
Holoien and his wife moved to New Jersey so she could attend graduate school. There were fewer job prospects for an East Asian expert on the East Coast than there were on the West Coast, so he worked in information technology. Computer programming was appealing, but not “programming for the sake of programming,” Holoien says. He wanted a science application he loved, so he earned a second bachelor’s degree in Astronomy from Rutgers University before moving to Ohio State.
Most supernova-hunting projects use large telescopes that capture small sky sections at great depth. ASAS-SN’s small telescopes take in a wide field at a limited depth, so it can more rapidly find nearby objects. Because they’re close, researchers can more quickly confirm and study ASAS-SN findings by gathering light spectra and other information.
“The things we find can be easily observed with a 1- or 2-meter telescope, which there are a lot of, versus needing an 8-meter or 10-meter telescope or the Hubbell Space Telescope to get follow-up data” on distant objects, Holoien says. The goal is to “focus on the best and brightest, basically – the things that we can really study in great detail.”
Identifying the best and brightest, however, requires massive data sifting. Holoien is tuning a software pipeline that finds them with a minimum of human interaction.
The program automatically chooses the sky sections that have gone longest since their last analysis. It removes artifacts generated by the camera and other artificial sources, then aligns the image with a previous one. The old image is subtracted from the new one. Ideally, “anything that’s bright or dark in that (remaining) image is something that has gotten brighter or faded since the last observation.”
There still could be false positives, however. Each sky field is captured three times in one visit, so an object found in only one image is unlikely to be real. The algorithm and subsequent filtering removes nearly 90 percent of bright and dark points, discarding them because the shape is wrong, their location indicates they’re false, or due to other factors.
Objects that get through the pipeline’s screens are farmed out to a network of amateur astronomers who confirm or negate the sightings. In about three and a half years, ASAS-SN has found around 500 supernovae and thousands of other transient objects, such as variable stars. Stanek’s group publicly releases all the data and chooses interesting finds for continued study.
Besides his data pipeline work, Holoien studies some of these discoveries, including tidal disruption events, in which a star is destroyed in a supermassive black hole, creating a luminous outburst akin to a supernova.
Effective data processing will become even more important as the number of ASAS-SN telescopes triples, letting the researchers image the entire sky every night. Such surveys are “a big area of astronomy now. Going forward, developing those skills and finding new computational techniques to make this all more efficient is going to be really important.”
Holoien helped advance such techniques for one upcoming project, the Large Synoptic Survey Telescope (LSST), during his 2016 DOE CSGF practicum at SLAC National Accelerator Laboratory in California. The survey, starting in a few years, will employ an enormous telescope to deeply survey the sky, gathering around 200 petabytes of data.
Automatically processing that much information is vital to LSST’s success, so Holoien developed machine learning and training tools to help find supernovae amid the observations. With physicists Phil Marshall and Risa Weschler of SLAC and Stanford, he built XDGMM, a tool that uses machine learning to mathematically model multi-dimensional data sets, can condition the results on known variables, and generates a model for unknown variables. The researchers fed supernovae data into XDGMM to create empiriciSN, a supernova predictor.
LSST researchers use empiriciSN to generate realistic supernovae models and plant them in simulated data used to train the LSST pipeline. Holoien also is using skills developed on his practicum to add a machine-learning algorithm to the ASAS-SN data program.
Holoien graduates in August, but will continue working with ASAS-SN as a Carnegie Postdoctoral Fellow at the Carnegie Observatories in Pasadena, California. With access to the Carnegie telescopes in Chile, he also will embark on new research into tidal disruption events, supernovae, and other subjects.
Image caption: Thomas Holoien in Arizona with the telescopes of the Kitt Peak National Observatory behind him. Image courtesy of Thomas Holoien.