Using GNNs for mass reconstruction of light pseudoscalars decaying to merged photon-pairs in the CMS ECAL endcaps

Abstract

The Standard Model (SM) has been thoroughly tested in experiments conducted over the past forty-fifty years, It has successfully explained the phenomena occurring with probabilities over eight orders of magnitude. The discovery of the Higgs boson by ATLAS and CMS in 2012, has opened up new avenues for the search of Beyond Standard Model (BSM) physics. An example of such BSM scenarios is the exotic decays of the SM Higgs boson into a pair of light pseudoscalars (as), with each pseudoscalar decaying into two photons (H −→ aa −→ 4γ ). For such pseudoscalars with low masses (ma < 2 GeV), the two photons from its decay can be merged as single photon object. The project focuses on development of a novel mass reconstruction technique using Graph Neural Networks (GNNs) for such highly boosted pseudoscalars. Using GNNs allows us to train the model on low level input features from different subdetectors with varying geometries, that is sparse and irregular data, without the need for zero padding.