Lunch talk on Sep. 21, 2022

Predicting Cosmology with Galaxy Clusters Based on Machine Learning

Speaker: Lanlan Qiu (SYSU)

Venue: Video Conference

Time: 12:30 PM, Wednesday, Sep. 21, 2022

Abstract: 

Galaxy clusters can be used to constrain cosmology with the cluster mass function but with non-negligible inaccuracy in estimating masses which overall limits their constraining ability. We propose a Machine Learning method making use of a more extensive variety of observables or "features", such as the gas mass, gas bolometric luminosity, gas temperature, stellar mass, radius, redshift, etc., to derive more unbiased constraints on cosmology. As the first step, we have used more than 400k clusters from 14 different cosmologies, from Magneticum simulations, to find a relatively best machine learning pipeline to predict the universe’s most likely cosmology. In an ideal ensemble of measurements in simulations, we show that machine learning can predict true cosmology with an accuracy of almost 100%. In the case of measurements with some realistic scattered noise ranging from 0 to 20%, the accuracy remains around 60%. This is a very promising approach to exploit multi-wavelength observations from current and future surveys like CSST in optical bands, EUCLID in NIR bands, and eROSITA in X-rays.