Probing Compressed Mass Spectrum Supersymmetry at the LHC with the Vector Boson Fusion Topology

TL;DR

This work investigates probing compressed-mass spectrum SUSY via electroweakino pair production in the LHC using the vector boson fusion topology and a sequential-attention deep learning classifier (TabNet) to enhance signal discrimination across 1–3-lepton final states. It models two SUSY scenarios—Wino-Bino $W^*/Z^*$ and Wino-Bino light-slepton—and demonstrates that TabNet, combined with VBF tagging, extends sensitivity beyond current LHC limits, achieving 95% CL exclusions up to $m(\tilde{\chi}_{1}^{\pm})$ on the order of 0.9–1.1 TeV and 5σ discovery reach at the few-hundred GeV level for favorable $\Delta m$. The study incorporates interference between purely EW and mixed QCD–EW diagrams via an $\text{RKI}$ functional, uses profile-likelihood fits, and projects HL-LHC performance for integrated luminosities of 137, 300, and 3000 fb$^{-1}$ under realistic pileup. The results indicate that VBF with soft leptons, aided by TabNet’s interpretability and performance, can access previously unexplored regions of the compressed MSSM parameter space, motivating experimental implementation with appropriate HL-LHC triggers.

Abstract

We present a phenomenology study probing pair production of supersymmetric charginos and neutralinos ("electroweakinos") with the vector boson fusion (VBF) topology in proton-proton collisions at CERN's Large Hadron Collider (LHC). In particular, we examine the compressed-mass spectrum phase space that has been traditionally challenging due to experimental constraints. The final states considered have two jets, large missing transverse momentum, and one, two, or three light leptons. Different model scenarios are considered for the production and decays of the electroweakinos. A novel high-performance and interpretable sequential attention-based machine learning algorithm is employed for signal-background discrimination and is observed to significantly improve signal sensitivity over traditional methods. We report expected signal significances for integrated luminosities of $137$, $300$, and $3000$ $\textrm{fb}^{-1}$ corresponding to the current data acquired at the LHC, expectation for the end of Run 3, and the expectation for the high-luminosity LHC. Our methodology results in projected 95\% confidence level bounds that cover chargino masses up to 1.1 TeV in compressed-mass spectrum scenarios within the R-parity conserving minimal supersymmetric standard model. This parameter space, currently beyond the reach of ATLAS and CMS searches at the LHC, is traditionally challenging to explore due to significant Standard Model backgrounds and low signal cross-sections.

Figures (17)

• Figure 1: Representative Feynman diagram for VBF $t$-channel chargino-chargino pair production, and subsequent decay to a 2-lepton final state through a $W^*$ boson in the Wino-Bino $W^*/Z^*$ model scenario.
• Figure 2: Representative Feynmann diagram from CMS:2024gyw for the production of $\widetilde{\chi}_{2}^{0}$ and $\widetilde{\chi}_{1}^{\pm}$, with the $\widetilde{\chi}_{2}^{0}$ decaying to either a $Z$ or $H$ boson and a $\widetilde{\chi}_{1}^{0}$, and the $\widetilde{\chi}_{1}^{\pm}$ decaying to a $W^\pm$ boson and a $\widetilde{\chi}_{1}^{0}$ in the Wino-Bino model scenario.
• Figure 3: Wino-bino model exclusion contours targeting the $WZ$ topology for the compressed-mass region studied in this paper (red-dashed), the CMS combined search SUS-21-008 CMS:2024gyw, and the CMS vector boson fusion search SUS-17-007 CMS:2019san.
• Figure 4: Projected cross-section in femtobarns as a function of the $\widetilde{\chi}_{1}^{\pm}$ and $\widetilde{\chi}_{2}^{0}$ masses and the mass difference $\Delta m$ for the Wino-Bino $W^*/Z^*$ scenario.
• Figure 5: Representative Feynman diagram for mixed electroweak-QCD chargino-neutralino pair production with two associated jets, and subsequent decay to a 3-lepton final state through $Z^*/W^*$ bosons in the Wino-Bino $W^*/Z^*$ model scenario.