from the instrument to the data analysis.
Faustine Cantalloube (MPIA) on High-contrast imaging of exoplanets and circumstellar disks
The Imprint of Cosmic Web Quenching on Galaxy Evolution
Nico Winkel
Thu, 11 Mar 2021, 11:00
MPIA travel analysis 2018-2020
Sustainability group
Thu, 11 Mar 2021, 11:00
MPIA business trip analysis 2018-2020
Jan Rybizki
Fri, 12 Mar 2021, 10:00

Boosting Monte Carlo sampling with a non-Gaussian fit

Adriŗ Gůmez-Valent , ITP Heidelberg
Monte Carlo analyses are a key ingredient in many branches of natural and social sciences. Also in cosmology. They are typically used to sample posterior distributions (built from data) in high-dimensional parameter spaces and infer the confidence regions of the parameters that enter the model under study. When the evaluation of the likelihood is computationally expensive, Monte Carlo analyses can demand prohibitive computational times, even with the use of powerful clusters. In this talk I will describe a new method, called Monte Carlo Posterior Fit, which allows to reduce in some cases an order of magnitude the time spent in the Monte Carlo sampling process. The idea is to approximate the posterior function by an analytical multidimensional non-Gaussian fit. The many free parameters of this fit can be obtained by a smaller sampling than is needed to derive the full numerical posterior, and the evaluation of the resulting analytical distribution can be quite faster than the original one. I will show some examples of the performance of this method in cosmology, based on supernovae and cosmic microwave background data. The method was recently introduced by Prof. Amendola and me in arXiv:2007.02615 [Mon.Not.Roy.Astron.Soc. 498 (2020) 1, 181-193]. Finally, I will present some preliminary results obtained also in collaboration with Dr. Marco Baldi by applying our method to cases in which N-body simulations are required to evaluate the likelihood.
Kosmologie und Elementarteilchenphysik
19 Jan 2021, 15:15
Institut für Theoretische Physik, Online

Add to calendar Add to calendar