ParSNIP: Using deep learning to identify supernovae and probe dark energy

DiRAC researchers are heavily involved in building the Vera C. Rubin Observatory, a new facility that is currently under construction in Chile. This observatory will feature the 8.4 meter Simonyi Survey Telescope and the world’s largest CCD camera which will scan the entire visible sky every three nights. It will discover and observe millions of supernovae which are powerful explosions of stars that can outshine an entire galaxy for a brief period of time.

A particular type of supernovae called “Type Ia” can be used to map out how the universe has expanded since the big bang. This led to the discovery of dark energy which was awarded the Nobel Prize in 2011. The Rubin Observatory will discover over 100 times as many Type Ia supernovae then have been observed by all surveys to date and will dramatically improve our understanding of the universe.

Extracting scientific results from this large deluge of data is a big challenge. In a paper that was recently published in the Astronomical Journal, DiRAC Fellow Kyle Boone discusses a new statistical model called ParSNIP that can be used to distinguish Type Ia supernovae from others and improve our maps of the universe. This novel work combines recent advances in computer science and deep learning with physics models of how light propagates through the universe. The resulting hybrid model is the first one that can empirically describe how the emitted light spectrum from any kind of supernova evolves over time.

This foundational work has many applications. ParSNIP will be used to identify the different kinds of supernovae that the Rubin Observatory finds, and it can do this with over twice the performance of previous models. It will also be used to hunt for new unknown kinds of supernovae in the large Rubin dataset. ParSNIP will use all of the millions of supernovae that the Rubin Observatory discovers to measure the properties of dark energy in contrast to current methods that can only use less than a tenth of the full sample. This work will transform supernova science with the Rubin Observatory and help to extract the full scientific potential.

ADS Publication: Published October, 2021, ParSNIP: Parametrization of SuperNova Intrinsic Properties


Kyle Boone is DiRAC Postdoctoral Fellow. Kyle’s research focuses on developing novel statistical methods for astronomy and cosmology. He is particularly interested in using Type Ia supernovae to probe the accelerated expansion of the universe that we believe is due to some form of “dark energy”. One aspect of his research focuses on identifying Type Ia supernovae among the millions of astronomical transients that upcoming astronomical surveys such as the Large Synoptic Survey Telescope (LSST) will discover.

Read more here. GitHub here.