“Revealing the Dynamic Universe”
When one looks up at the sky, it’s easy to think that the universe doesn’t change much. While the planets and moons move across the sky from night to night, the constellations known to us since ancient Greece haven’t changed much in the last two thousand years. And yet, this is an illusion. In reality, the night’s sky is incredibly dynamic, and every object in the universe—every star, every galaxy, and some of the more exotic objects we know of—varies on some time scales.
Stars show changes in brightness on timescales of minutes to weeks, including the same kinds of flares we observe from our own sun. Many black holes accrete matter, either from companion stars or from the surrounding gas and dust. As this matter falls into the black hole, it produces changes in brightness that can be as fast as milliseconds and can be as big as two hundred times the previous brightness.
At the end of their lifetime, many stars explode in a firework of expelled gas and light. These stellar explosions, also called supernovae, have been known to us since ancient Chinese astronomers observed them as “guest stars”, because they are visible with the naked eye when they occur in our own galaxy. With new instruments, we can peer farther and farther into the distance, finding these explosions in galaxies millions of light years away.
Almost all new telescopes currently being built allow for repeated observations of the same sources over time. This allows researchers to track how the brightness of these sources changes with time, and use these brightness changes to learn about the fundamental physics. For example, as gas and dust slowly fall into a black hole, the light they emit allows us to test fundamental predictions made by General Relativity, as well as study hot gases and plasmas under conditions we cannot reproduce on Earth. Supernovae, on the other hand, provide fundamental constraints on our theories of the cosmos as a whole.
At the same time, the new telescopes we build will produce time series — repeated measurements of the brightness of a source over time — at time scales and data rates that unprecedented. This will allow us to probe new theories, but also comes with unique challenges. We are in an era where we can no longer analyze data by hand, but need a computer to process these huge data sets and classify — did we observe a flaring star, a black hole in a binary or a supernova explosion? — them automatically and in real-time. Building these systems requires implementation state-of-the-art results from statistics and machine learning. The time domain group at DIRAC will use its expertise in data science, statistics and machine learning to build a new classification engine and find innovative ways to use time series for understanding our universe.
- Faculty: Eric Bellm, Andy Connolly, Zeljko Ivezic
- DIRAC Fellows: Jim Davenport, Zach Golkhou, Melissa Graham, Colin Slater