Recent Advances in Bayesian Inference in Cosmology and Astroparticle Physics Thanks to the Multinest Algorithm
Roberto Trotta1, Farhan Feroz2, Mike P. Hobson2, Roberto Ruiz de Austri3
1Physics Dept, Astrophysics, Imperial College London, London, United Kingdom; 2Astrophysics Group, Cavendish Laboratory, Cambridge University, Cambridge, United Kingdom; 3Instituto de Fisica Corpuscular, IFIC University of Valencia, Valencia, Spain

Obtaining accurate posterior samples from multi-modal distributions is a challenge for Markov Chain Monte Carlo (MCMC) methods, which are widely used in the context of Bayesian inference in a number of fields. Furthermore, Bayesian model comparison requires the computation of the Bayesian evidence (or model likelihood), which is difficult to obtain with standard MCMC methods.

In this talk I will present a new algorithm, called MultiNest, which implements nested sampling to sample efficiently from highly multi-modal distributions in parameter spaces of moderate dimensionality (up to about 100 dimensions). MultiNest also returns accurate estimates of the model likelihood, with a computational effort that is typically 2 orders of magnitude smaller than conventional MCMC-based methods such as thermodynamic integration.

I will illustrate the capabilities of MultiNest with a number of applications in astroparticle physics and cosmology, including: Bayesian object detection, parameter inference and model comparison for supersymmetric theories, cosmological model building and the assessment of the compatibility of different observations.

Keywords: Cosmology; Bayesian model comparison; Multi-modal likelihoods; Bayesian evidence

Biography: Dr Roberto Trotta is a lecturer in astrophysics at Imperial College London, after having been the Norman Lockyer Research Fellow of the Royal Astronomical Society at Oxford University and a Junior Fellow of St Anne's College, Oxford. Roberto graduated in theoretical physics from ETH Zurich and obtained a PhD in physics from the University of Geneva. His research focuses on advanced statistical methods for the interpretation of cosmological observations, and for the prediction and optimization of future missions. He is an expert in the development and application of Bayesian methods to cosmology and astroparticle physics.