Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.596904
Title: Statistical methods in cosmology
Author: Bridges, M.
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 2008
Availability of Full Text:
Full text unavailable from EThOS. Please contact the current institution’s library for further details.
Abstract:
We outline the application of a new method of evidence calculation called nested sampling Skilling (2004). We use a clustered ellipsoidal bound to restrict the parameter space sampled, that is generic enough to be used for even complex multimodal posteriors. We demonstrate our algorithms, COSMOCLUST makes important savings in computational time when compared with previous methods. The study of the primordial power spectrum, which seeded the structure formation observed in both the CMB and large scale structure, is crucial in unravelling early universe physics.  In this thesis we analyse a number of spectral parameterisations based on both physical and observational grounds. Using the evidence we determine the most appropriate model in both WMAP 1 year and WMAP 3 year data (including additionally a selection of high resolution CMB and large scale structure data). We conclude that currently the evidence does suggest the need for a tilt in the spectrum, however the presence of running of the spectral index is dependent on the inclusion of, specifically Ly-α data. Bayesian analysis in cosmology is computationally demanding. We have succeeding in improving the efficiency of inference problems for a wide variety of cosmological applications by training neural networks to ‘learn’ how observables such as the CMB spectrum change with input cosmological parameters. We demonstrate that improvements in speed of several orders of magnitude are possible using our algorithm COSMONET.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.596904  DOI: Not available
Share: