Parameter Fitting of Cosmological Models using Evolutionary Strategies

The current tools used for approximating cosmological parameters are Markov Chain Monte Carlo (MCMC) utilities using Metropolis-Hastings and Nested Sampling as the main sampling methods. These tend to have low sampling efficiency as many samples are wasted in trying to find good proposals of points at high regions of likelihood. In the data-rich era in which cosmology is entering, imaging the entire sky and generating PetaBytes of data, this is unsatisfactory. In this project we propose Evolutionary Strategies as a better sampling method. The algorithm works by mutating populations of samples across iterations of the algorithm in order to adapt its behaviour over the current sample’s location in parameter space. The developed method, ESNested, integrates Nested Sampling with Evolutionary Strategies, such that the method is able to approximate the Bayesian Evidence of a distribution. Results show that ESNested returns good parameter fits with a high sampling efficiency. Other methods tend to make a Gaussian assumption on the search space, that can generate a bad parameter fit if the data does not follow this distribution in real life. ESNested does not make this assumption and therefore works well in most search spaces. Using the developed method, various parameter fits, using different combinations of parameters are tested on Cosmological models. By analysing parameter fits, confidence intervals and the importance and correlations of parameters in Cosmological models, this project also looks at what theories in Cosmology do generated posterior samples support or reject.