Advanced search
Start date

Cosmological covariance matrices and machine learning methods

Grant number: 19/13108-0
Support type:Scholarships in Brazil - Doctorate
Effective date (Start): September 01, 2019
Effective date (End): August 31, 2022
Field of knowledge:Physical Sciences and Mathematics - Astronomy - Extragalactic Astrophysics
Principal researcher:Luis Raul Weber Abramo
Grantee:Natali Soler Matubaro de Santi
Home Institution: Instituto de Física (IF). Universidade de São Paulo (USP). São Paulo , SP, Brazil


Cosmic acceleration is Cosmology's greatest puzzle: is it due to dark energy or is it a result of modified gravitational theory on large scales? Many large-scale investments in astronomical instruments (DESI, Euclid, PFS, J-PAS, and others), in 2-4 years, will lead to a dramatic increase in the quantity and quality of the cosmological data. These data sets will encompass larger volumes, higher redshifts, and an amazing variety of different types of galaxies, observed in many wavelengths. Given the complexity of these data, how can we explore efficiently the observations, extracting useful information about Cosmology, as well as galaxy formation and evolution? The comparation of data to models assumes that the nature of the uncertainties is already under control and is well understood.The data covariance matrices must be extremely well understood, not only insofar as they reflect the propagation of the statistical errors intrinsic to observations, but mainly because these covariance matrices depend on our assumptions about the underlying fiducial models that we use for the data analysis, i.e., on the theoretical models that we employ in order to reduce the data - although this is often neglected in practical applications. The goal of this project is to compute covariance matrices for this new generation of galaxy surveys, including not only the large volumes and high redshifts, but also the many different tracers of large-scale structure. We will employ a range of mock catalogs of halos and galaxies, generated with techniques ranging from simple lognormal mocks, to dark matter halo simulators, and including full N-body simulations. We will also study different machine learning methods, which will be trained to recompute covariance matrices, more quickly, given small changes in the parameters of the underlying theoretical model. Our final goal is to forecast in a more precise way how these new datasets will impact Cosmology, in terms of measurements of the baryon acoustic oscillations, redshift-space distortions and, ultimately, in the search for the nature of dark energy. (AU)

News published in Agência FAPESP Newsletter about the scholarship:
Articles published in other media outlets (0 total):
More itemsLess items