On the coverage of electroweak-inos within the pMSSM with SModelS -- a comparison with the ATLAS pMSSM study

The ATLAS collaboration has recently performed a vast scan of the phenomenological Minimal Supersymmetric Standard Model (pMSSM) with a focus on the electroweak-ino sector, and analysed how their Run 2 searches for electroweak production of supersymmetric (SUSY) particles constrain this dataset. All the SLHA files from the scan as well as the constraints from the eight individual searches considered by ATLAS were made publicly available. We use this material to study how well the ATLAS constraints can be reproduced with SModelS v3.0. Moreover, we explore how the picture changes when also including CMS results, and what can be gained by the statistical combination of analyses. Finally, we discuss the part of parameter space with light electroweak-inos that remains valid despite the stringent LHC limits. Our results underscore the need of a broad, multifaceted approach for maximising sensitivity and closing loopholes in the extensive SUSY parameter space.

💡 Research Summary

This paper presents a detailed validation of the public reinterpretation tool SModelS v3.0 against the extensive ATLAS pMSSM electroweak‑ino (EW‑ino) study that was released together with the full set of SLHA spectra and the eight Run‑2 electroweak SUSY searches. The authors start from the ATLAS “EWKino” scan, which sampled the 19‑dimensional pMSSM parameter space with a focus on the bino (M₁), wino (M₂) and higgsino (μ) mass parameters, while decoupling the first‑two‑generation sfermions. After applying the ATLAS‑provided filters (χ₁⁺ > 1 TeV, insufficient event yield, Higgs invisible width, heavy‑A limits) and removing pathological points with an unrealistically small chargino‑neutralino mass splitting that would lead to long‑lived charged tracks, the authors retain 8 953 viable model points for comparison.

Using SModelS v3.0, the authors first reproduce the ATLAS “single‑analysis” approach, i.e. for each point they consider only the most sensitive ATLAS analysis (the one with the largest expected r‑value, r ≡ 1/μ₉₅). Six of the eight ATLAS EW‑ino analyses are fully implemented in the SModelS database; the remaining two are only available for the 36 fb⁻¹ dataset, which slightly weakens the comparison. With a signal‑cross‑section cut of 10⁻³ fb and a minimum mass gap of 10 GeV (to improve coverage of compressed higgsino scenarios), SModelS excludes about 85 % of the points that ATLAS excludes. The residual differences are traced to (i) missing efficiency maps for the two incomplete analyses, (ii) the limited treatment of very compressed spectra where the 10 GeV mass‑gap requirement discards some relevant topologies, and (iii) the absence of CMS constraints in the first step.

In the second step the authors augment the SModelS database with all available CMS electroweak‑ino searches and with gluino‑pair production limits. This “full‑database” run raises the overall exclusion fraction by roughly 5 % relative to the ATLAS‑only case. Notably, CMS analyses such as SUS‑19‑006 (multilepton + MET) and SUS‑19‑008 (1‑lepton + 2 b‑jets) provide complementary sensitivity to wino‑like χ₁⁺‑χ₁⁰ compressed scenarios that are only weakly constrained by ATLAS, thereby closing a small but phenomenologically interesting loophole.

The third and most powerful step is the statistical combination of (approximately) uncorrelated signal regions across all analyses. By enabling the “combineSRs” option, SModelS builds an approximate χ²‑based likelihood that combines the contributions of several signal regions, selecting the set that maximises the combined expected r‑value (r_comb_exp). This procedure yields an average improvement of about 30 % in sensitivity compared with the single‑analysis approach. In practice, the combination of the 3‑lepton on‑shell and 2‑lepton + jets searches leads to the near‑complete exclusion of wino‑like χ₁⁺ masses below ≈200 GeV, a region that remained partially open when only the most sensitive single analysis was considered.

Having established the performance of SModelS, the authors turn to the residual viable parameter space. Three distinct classes survive the combined ATLAS + CMS constraints: (1) ultra‑compressed spectra with χ₁⁺–χ₁⁰ mass splittings below ≈5 GeV, where current public efficiency maps are insufficiently granular; (2) higgsino‑dominant LSP scenarios with χ₁⁰ ≲ 100 GeV and χ₁⁺–χ₁⁰ splittings of 10–15 GeV, which produce very soft leptons that evade existing lepton‑based searches; and (3) mixed wino‑higgsino states where the branching fractions to the canonical multilepton final states are suppressed, making ISR‑driven “soft‑lepton + MET” searches the most promising avenue. Moreover, in the ultra‑compressed case the chargino lifetime can reach the centimetre scale, leading to disappearing‑track signatures that are only weakly constrained by the current ATLAS disappearing‑track analysis (which was not fully implemented in SModelS). The authors therefore advocate dedicated long‑track searches and refined soft‑object analyses for Run‑3 and the High‑Luminosity LHC.

In summary, the paper demonstrates that SModelS v3.0 can faithfully reproduce the ATLAS pMSSM electroweak‑ino limits, and that the inclusion of CMS results together with a statistical combination of analyses significantly strengthens the overall exclusion power. The remaining allowed regions are highly compressed or feature mixed electroweak‑ino compositions, highlighting the need for a diversified experimental programme—including low‑pT lepton triggers, ISR‑based selections, and disappearing‑track searches—to fully close the electroweak‑ino sector of the pMSSM.