Stats4Grads: Calibrating the Universe: an Uncertainty Analysis for a Galaxy Formation Simulation
8 October 2008 14:15 in CM105
Abstract: Many scientific areas now employ large physical computer simulations in order to understand complex real world systems. Such simulations contain a large number of input and output parameters and this leads to several problems of interest. A subset of the outputs can be compared with observed data of the real system, and a natural question to ask is: which choices of input parameters to the model will give rise to `acceptable' matches between the outputs and the data. This is known as Calibration and is one of the most important problems when dealing with models of this size and complexity.
I will describe a version of Calibration known as History Matching in the context of a Galaxy Formation simulation that attempts to model the evolution of one million galaxies from the beginning of the Universe until the present day. I will describe the process of Emulation (the statistical modelling of a deterministic function) and the method of History Matching via an Implausibility measure. Communicating the results of a successful Calibration is also a non-trivial matter as this involves describing a complex shape in a high-dimensional space, so various visualisation issues and techniques will also be discussed.
This seminar is for Postgraduate Students and Postdoctoral Researchers only
Contact thomai.tsiftsi(@)durham.ac.uk for more information
See the Stats4Grads page for more details about this series.