Peculiar velocity estimation from Sunyaev-Zel'dovich effect simulated signal using deep learning

Abstract

Cosmic Microwave Background photons are inverse Compton scattered by energetic electrons in galaxy clusters and distorted by its bulk motion in a process known as the SunyaevZel’dovich (SZ) effect. Conventional methods that calculate cluster peculiar velocities require additional information like the thermal component (tSZ) or underlying electron density ne to estimate each cluster’s optical depth τ, however observational measurements contain large errors and biases. This work studies the feasibility of using deep learning regression algorithms for estimating individual cluster peculiar velocities from the simulated kinematic (kSZ) component signal, thus exempting the need of τ. This formalism is tested using the Magneticum cosmological hydrodynamical simulation. Both tSZ and kSZ, along with ne maps were generated at z = [1:04-1:32]. Trials with a simple convolutional neural network yielded prediction error standard deviations as low as σe = 11:019 km s^-1 ; whilst both methods implementing τ’s explicit calculation yielded standard deviations of 81:569 km s^-1 using tSZ and 87:527 km s^-1 for ne. The neural network provided reliable predictions for studying LSS velocity fields using the pairwise velocity estimator v12, in addition to improved correlation compared to conventional methods.