Improving Cosmological Distances to Illuminate Dark Energy: Hierarchical Bayesian Models for Type Ia Supernovae in the Optical and Near-Infrared

Meeting: 
ISBA 2012
Meeting Program Chair: 
ISBA Program Chair
Presenter First Name: 
Kaisey
Presenter Last Name: 
Mandel
Presenter's Email: 
k.mandel@imperial.ac.uk
Affiliation: 
Imperial College London
Country: 
United Kingdom
Presentation Type: 
Invited
Abstract: 
Type Ia supernovae, brilliant stellar explosions, are observed in distant galaxies in the Universe. These supernovae can be used as "standard candles": if their luminosity is known, their distances can be estimated from their observed brightness. Observing optical light from faraway supernovae to determine their distances, astronomers found that the expansion of the Universe is accelerating, rather than slowing down under the attractive force of gravity, an astounding discovery that won the 2011 Nobel Prize in Physics. The cosmic acceleration is thought to be caused by a repulsive "dark energy", whose physical nature is the most perplexing mystery in modern cosmology. To shed light on the nature of dark energy, I seek to improve statistical inference of the history of cosmic expansion using supernovae. The major source of systematic error confounding current distance estimates is interstellar dust in the galaxies of the supernovae. Dust dims a supernova's optical light, making it appear farther away. However, in near-infrared (NIR) light, the dust is nearly transparent, and supernovae are even better standard candles. The combination of optical and NIR time series and spectroscopic data mitigates the pernicious effects of dust and can improve inferences in supernova cosmology. To optimize distance estimates, I constructed a principled, hierarchical Bayesian framework, described by a directed acyclic graph, to coherently model the multiple random and uncertain effects underlying supernova time series observations, including measurement error, dust, intrinsic supernova covariances across time and wavelength, galaxy motions and distances. I developed an MCMC code, BayeSN, using a Gibbs sampling structure to efficiently compute probabilistic inferences for the parameters of individual supernovae and the hyperparameters of their population, while dealing with incomplete data. Applying this to nearby supernova data, I demonstrate that the combination of optical and NIR data almost doubles the precision of cross-validated distance predictions, and is a more powerful method to measure the properties of dark energy.
Keywords: 
Cosmology
Keywords: 
Astronomy
Keywords: 
Bayesian hierarchical model
Keywords: 
MCMC
Keywords: 
Gibbs sampling
Keywords: 
Dark Energy Equation of State
Keywords: 
directed acyclic graph
Keywords: 
supernovae
Date/Time: 
Friday, June 29, 2012 - 09:20 - 09:40