A0630
Title: Bayesian mixture models for scientific discovery in astrophysics
Authors: David van Dyk - Imperial College London (United Kingdom) [presenting]
Abstract: Mixture data are ubiquitous in astrophysics. Photons originating from different sources or from different physical processes within a single source are often indistinguishable. Point sources (such as stars) may be physically embedded in an extended source (such as nebulae), or sources may simply appear to overlap with their foreground or background from the vantage point of Earth. Astrophysicists inevitably aim to use photons from each source to understand its individual physical properties but only have a mixture of photons from all sources. Even with well-separated sources, photons originate from multiple physical processes. A star that appears as a point source, for example, may contain both turbulent/flaring regions and quiescent regions, and even in a single region, the star's plasma may exhibit multiple physical processes with different but mixed spectral signatures. A population of sources may contain subclasses that are difficult to distinguish, yet researchers wish to analyze them separately. Instrumental effects such as point-spread functions, photon-redistribution matrices, and pile-up also effectively mix photons. Several specific examples are discussed where the creative deployment of Bayesian mixture models in astronomy has enabled overlapping sources/processes to be disentangled, scientific parameters of individual sources to be estimated, and uncertainty to be properly quantified.
