Abstract: The search for weakly interacting matter particles (WIMPs) is one of the main objectives of the High Luminosity Large Hadron Collider (HL-LHC). In this work we use Machine-Learning (ML) techniques to explore WIMP radiative decays into a Dark Matter (DM) candidate in a supersymmetric framework. The minimal supersymmetric WIMP sector includes the lightest neutralino that can provide the observed DM relic density through its co-annihilation with the second lightest neutralino and lightest chargino. Moreover, the direct DM detection cross section rates fulfill current experimental bounds and provide discovery targets for the same region of model parameters in which the radiative decay of the second lightest neutralino into a photon and the lightest neutralino is enhanced. This strongly motivates the search for radiatively decaying neutralinos which, however, suffers from strong backgrounds. We investigate the LHC reach in the search for these radiatively decaying particles by means of cut-based and ML methods and estimate its discovery potential in this well-motivated, new physics scenario.
Introduction
In the search for Supersymmetry (SUSY), a leading theory for physics beyond the Standard Model, the "lightest neutralino" ($\tilde{\chi}^0_1$) is a prime candidate for Dark Matter. However, experimental challenges arise in "compressed" scenarios where the mass difference between the lightest neutralino and the second-lightest neutralino ($\tilde{\chi}^0_2$) is small. In these cases, the resulting decay products have very low energy ("soft" signals), making them difficult to distinguish from Standard Model backgrounds using traditional "cut-and-count" methods.
This study specifically targets the radiative decay signature ($\tilde{\chi}^0_2 \to \tilde{\chi}^0_1 + \gamma$), which has been historically overlooked due to high background noise. By combining advanced Machine Learning (ML) techniques with high-energy physics theory, the researchers demonstrate a significant enhancement in discovery potential at the HL-LHC.
Signal Signature and ISR Boosting
To make the soft decay products visible, the analysis focuses on events where the entire SUSY system recoils against a high-energy "Initial State Radiation" (ISR) jet. This "boosts" the system, giving the resulting photon and missing energy enough momentum to be detected above the noise floor. The primary signal process involves the production of a chargino-neutralino pair ($\tilde{\chi}^\pm_1 \tilde{\chi}^0_2$) alongside an ISR jet.
Methodology and Benchmark Scenarios
The researchers employed a multi-step computational pipeline using MadGraph5_aMC@NLO for event generation, Pythia8 for particle showering, and Delphes for detector simulation. The study evaluated 15 "Benchmark Points" (BPs) representing different mass configurations within the Minimal Supersymmetric Standard Model (MSSM). These points were selected because they align with current cosmological observations of dark matter density and the experimental discrepancy in the muon's magnetic moment ($g-2$).
An XGBoost classifier was trained using 22 kinematic variables, ranging from low-level detector data to high-level "physics-aware" variables like transverse mass ($m_T$) and $E_T^{miss}$ significance. The model captures multi-dimensional correlations between variables that human-designed cuts often miss.
Key Results and Statistical Innovation
The XGBoost classifier achieved an impressive Area Under the Curve (AUC) of 0.94, indicating excellent separation between signal and background. Furthermore, the study compared two statistical approaches:
- Binned Likelihood (BL): The traditional method of dividing ML output into a histogram.
- Machine-Learned Likelihood (MLL): An unbinned method using Kernel Density Estimators (KDE) to fit the ML output distribution directly.
The MLL method consistently outperformed the binned approach, preserving more information and contributing to a 3–6x improvement in discovery significance compared to baseline cut-based methods. The study projects that the HL-LHC could probe neutralino masses up to 365 GeV with mass splittings as small as 10–30 GeV.
Conclusion
This research provides a proof-of-concept for using ML to explore challenging regions of the SUSY parameter space. By targeting radiative decays and utilizing unbinned likelihood fits, the discovery reach for Dark Matter candidates at the LHC is substantially expanded. The next step involves implementing these ML-based triggers and filters in actual search pipelines for upcoming LHC runs.
Summary updated