Casting a graph net to catch dark showers

Elias Bernreuther,Michael Krämer,Thorben Finke,Felix Kahlhoefer,Alexander Mück

doi:10.21468/scipostphys.10.2.046

Abstract

Strongly interacting dark sectors predict novel LHC signatures such as semi-visible jets resulting from dark showers that contain both stable and unstable dark mesons. Distinguishing such semi-visible jets from large QCD backgrounds is difficult and constitutes an exciting challenge for jet classification. In this article we explore the potential of supervised deep neural networks to identify semi-visible jets. We show that dynamic graph convolutional neural networks operating on so-called particle clouds outperform convolutional neural networks analysing jet images as well as other neural networks based on Lorentz vectors. We investigate how the performance depends on the properties of the dark shower and discuss training on mixed samples as a strategy to reduce model dependence. By modifying an existing mono-jet analysis we show that LHC sensitivity to dark sectors can be enhanced by more than an order of magnitude by using the dynamic graph network as a dark shower tagger.

Highlights

The huge wealth of data taken at the LHC offers a unique opportunity to explore the properties of dark sectors and uncover the nature of dark matter (DM)
Dark sectors with new strong dynamics may reveal themselves at the LHC in the form of dark showers resulting from the fragmentation and hadronisation of dark quarks
If some of the dark mesons in the shower are stable on cosmological scales, while other dark mesons decay on collider scales, such dark showers lead to semi-visible jets

Summary

Introduction

The huge wealth of data taken at the LHC offers a unique opportunity to explore the properties of dark sectors and uncover the nature of dark matter (DM). We explore the model dependence of the semi-visible jet classification with the DGCNN, i.e. we study how the performance changes as we vary the parameters of the strongly interacting dark sector This sheds light on how much a specific network generalises to other dark shower scenarios, but it provides some indication of which signal features the network may learn. This way the network is forced to learn features common to the different samples instead of learning to reconstruct, for example, one specific dark meson mass The performance of such a more general classifier is significantly better than that of a classifier trained on specific values of rinv and mmeson when both are applied to a wider range of model parameters, see figure 6 and table 2.3 A significant improvement is present for dark meson masses that were not included in the mixed training sample, as the results for mmeson = 15 GeV show. Accurate simulations of this particular background are very challenging, and we leave a study of the potential sensitivity of such a search to future work

Conclusions

Findings

A Workshop on the Implications of HERA for LHC Physics