Search for higgsinos in compressed mass spectra using low-momentum tracks in $pp$ collisions at $\sqrt{s}=13$ TeV with the ATLAS detector

Technical Summary: Search for Higgsinos in Compressed Mass Spectra using Low-Momentum Tracks in $pp$ Collisions at $\sqrt{s}=13$ TeV

Problem and Motivation
This paper addresses the search for supersymmetric (SUSY) higgsinos in scenarios where the mass spectrum is "compressed." In Minimal Supersymmetric Standard Model (MSSM) scenarios where the higgsino mass parameter $|\mu|$ is much smaller than the bino ($M_1$) and wino ($M_2$) mass parameters, the lightest supersymmetric particles (LSPs) form a nearly degenerate triplet of higgsino-like states ($\tilde{\chi}^0_1, \tilde{\chi}^\pm_1, \tilde{\chi}^0_2$). The small mass splittings ($\Delta m$) between these states, driven by radiative corrections, result in decay products with very low transverse momentum ($p_T$). Consequently, standard trigger thresholds and reconstruction algorithms often fail to detect these events, particularly when the mass splitting is below 1 GeV. Previous searches have covered "disappearing track" signatures (for $\Delta m \lesssim 300-400$ MeV) and prompt lepton signatures (for $\Delta m \gtrsim 1.5$ GeV), but a gap remained in the intermediate and low-mass splitting regions where existing constraints were weaker or non-existent.

Methodology
The analysis utilizes 140 fb$^{-1}$ of proton-proton collision data at $\sqrt{s}=13$ TeV collected by the ATLAS detector. The search strategy relies on identifying higgsino pair production in association with a high-$p_T$ initial-state radiation (ISR) jet, which boosts the higgsino system and generates significant missing transverse momentum ($E^{\text{miss}}_T$) to trigger the event. Two distinct search strategies are employed to cover different mass splitting regimes, both leveraging machine learning (neural networks) to identify soft decay products:

1. Displaced Track Search ($\Delta m \approx 0.3\text{--}1$ GeV):

   • Target: Scenarios where the charged higgsino ($\tilde{\chi}^\pm_1$) has a lifetime of $c\tau \approx 0.1\text{--}10$ mm, decaying into a soft pion ($\pi^\pm$) and the LSP ($\tilde{\chi}^0_1$).
   • Signature: Events with high $E^{\text{miss}}_T$, a high-$p_T$ jet, and at least one low-momentum charged track with a large transverse impact parameter ($d_0$).
   • Technique: Two fully connected neural networks (NNs) are used: an "event-level" NN utilizing global kinematics (jet $p_T$, $E^{\text{miss}}_T$) and a "track-level" NN utilizing track properties ($p_T$, $d_0$, isolation, secondary vertex matching). Signal regions (SRs) are defined in the 2D plane of these NN scores.
   • Background Estimation: Dominated by $Z(\to\nu\bar{\nu})+$jets and $W^\pm(\to\tau\nu)+$jets. A semi-data-driven approach is used, employing control regions (CRs) to derive normalisation factors and track-level reweighting factors to correct for mismodelling of pile-up and track isolation.

2. One-Lepton and One-Track (1$\ell$1T) Search ($\Delta m \approx 1\text{--}3$ GeV):

   • Target: Scenarios where higgsinos decay promptly via off-shell gauge bosons ($\tilde{\chi}^0_2 \to Z^* \to \ell^+\ell^-$ and $\tilde{\chi}^\pm_1 \to W^{*\pm} \to \ell\nu$).
   • Signature: Events with high $E^{\text{miss}}_T$, a high-$p_T$ jet, exactly one standard "tight" lepton, and one additional low-$p_T$ lepton-track candidate (electron or muon).
   • Technique: Dedicated low-$p_T$ lepton-track identification (ID) algorithms based on NNs are developed to identify electrons and muons in the range $p_T \in [0.5, 5]$ GeV, extending beyond standard ATLAS capabilities. A parameterised neural network (pNN) is employed, taking the mass splitting $\Delta m(\tilde{\chi}^0_2, \tilde{\chi}^0_1)$ as an input parameter to dynamically optimize selection criteria across different mass hypotheses.
   • Background Estimation: The dominant background is $W^\pm+$jets with a fake low-$p_T$ lepton-track. A data-driven "transfer factor" method is used, measuring the ratio of passing-to-failing tracks in control regions to estimate the fake background in the signal regions.

Key Contributions

• Novel Identification Techniques: The development and application of NN-based "soft-lepton taggers" and "track-level" classifiers allow for the reconstruction and identification of particles with $p_T$ as low as 0.5 GeV, a regime previously difficult to access with standard triggers and reconstruction.
• Parameterised Machine Learning: The use of a pNN in the 1$\ell$1T search allows a single model to effectively cover a wide range of mass splitting hypotheses ($\Delta m = 1\text{--}5$ GeV) without requiring separate training for each specific mass point.
• Comprehensive Background Modeling: The implementation of sophisticated semi-data-driven methods, including track-level reweighting for the displaced track search and transfer factor methods for the 1$\ell$1T search, provides robust background estimates in regions with low statistics and complex kinematics.

Results

• Observation: No significant excess of events over the Standard Model (SM) predictions was observed in either search. All observed event yields in the signal regions were consistent with background expectations within statistical and systematic uncertainties (deviations were generally within $2\sigma$).
• Exclusion Limits:
  • Displaced Track Search: Excludes chargino masses ($m_{\tilde{\chi}^\pm_1}$) up to approximately 200 GeV for mass splittings around 0.6 GeV.
  • 1$\ell$1T Search: Excludes chargino masses up to approximately 130 GeV for mass splittings around 2 GeV.
  • Combined: The combined results exclude chargino masses below 126 GeV at 95% confidence level for mass splittings between 0.3 and 2 GeV.
• Model-Independent Limits: Upper limits on the visible cross-section for generic beyond-SM processes were set for the 1$\ell$1T search regions optimized for 1, 3, and 5 GeV mass splittings.

Significance
The paper claims that these results represent the first ATLAS constraints in a specific portion of the compressed higgsino parameter space (specifically $\Delta m \approx 0.3\text{--}2$ GeV). Crucially, these new limits surpass the limits previously set by other experiments, including the LEP experiments, which held the strongest constraints for mass splittings in the range of 0.9–1.2 GeV. By extending the exclusion reach to higher masses and covering the gap between the "disappearing track" and "prompt lepton" regimes, this work provides the most stringent constraints to date on higgsino-like charginos and neutralinos with compressed mass spectra in the context of simplified SUSY models.