Galaxy Formation: Bayesian History Matching for the Observable Universe

Understanding the evolution of the Universe from the Big Bang to the current day is the fundamental goal of cosmology. A major part of this is the problem of structure formation: understanding the formation, growth and subsequent evolution of galaxies in the presence of Dark Matter. The world leading Galform group, based at the Institute of Computational Cosmology, Durham University, has developed a state of the art model of Galaxy formation know as Galform. However they face a critical problem. Galform requires the specification of many input parameters and takes a significant time to complete one simulation, making comparison between the model’s output and real observations of the Universe extremely challenging. Here we describe the analysis of this problem using Bayesian history matching methodology, highlighting why the problem itself can only be sensibly formulated within a subjective Bayesian context, and demonstrating the use of Bayesian emulators within an iterative history matching strategy. This work represents the most detailed uncertainty analysis of a Galaxy formation simulation yet performed, and, to our knowledge, the most detailed history match, with the most number of iterations completed, for any model in the scientific literature. This methodology is widely applicable across any scientific discipline that uses computer simulations of complex physical processes. We discuss galaxy formation in section 2, the Bayesian history matching methodology in section 3 and the application and results in section 4. For a more detailed account of this ongoing project see Vernon, Goldstein and Bower (2010).