Turning the monojet’s most stubborn background into the widest net in the dark-matter hunt
A PPC spotlight on JHEP 11, 153 (2021) — CMS, “Search for new particles in events with energetic jets and large missing transverse momentum in proton-proton collisions at √s = 13 TeV” (arXiv:2107.13021, analysis CMS-EXO-20-004)
For years, the recipe has been the same: one energetic jet recoiling against a large imbalance of momentum—the classic “mono-X” signature that could reveal dark matter escaping the CMS detector unseen. The problem is that nature has its own way of producing exactly the same event. When a Z boson decays into two neutrinos, it also vanishes without a trace, creating an almost indistinguishable signature far more often than any hypothetical dark matter particle.
This analysis flips that obstacle into an advantage. Instead of simply estimating the irreducible Z→νν background, CMS uses the Z boson’s visible decay modes to measure it directly with unprecedented precision. In doing so, the dominant background becomes the analysis’s greatest asset, allowing a single search to probe an entire spectrum of dark matter and other beyond-the-standard-model scenarios with exceptional sensitivity.
The idea: Measure the cousins, constrain the ghost
You cannot see Z → νν. But you can see Z → ℓ⁺ℓ⁻, W → ℓν, and γ + jets, and the standard model relates all of them with a good precision. The analysis exploits this through a simultaneous maximum-likelihood fit across the signal region and five dedicated control regions: dielectron and dimuon (enriched in Z → ℓℓ), single-electron and single-muon (enriched in W → ℓν), and a photon region (γ + jets). Transfer factors tie the dominant Z → νν and W → ℓν backgrounds to these well-measured visible processes, so the ghost in the signal region is anchored by data rather than by simulation alone. In HEP jargon this is called a data-driven analysis.
Two main upgrades give the search its teeth:
- Machine learning does the tagging. Events are classified using the DeepAK8 neural network into three mutually exclusive categories: monojet, low-purity mono-V, and high-purity mono-V, according to whether a large-radius jet is consistent with the hadronic decay of a W or Z boson. Compared with the previous N-subjettiness-based approach, DeepAK8 suppresses the misidentification of QCD jets as vector bosons by a factor of five to ten while maintaining the same signal efficiency.
- The QCD multijet fakes are exorcised. Additional topological requirements on the angular separation between the jets and the missing transverse momentum suppress events in which detector mismeasurements mimic genuine missing energy. Cross-checks comparing particle-flow and calorimeter-based missing momentum, together with charged-particle recoil information, provide a powerful safeguard against events where a mismeasured jet merely imitates the signature of invisible particles.
This is the same control-region scaffolding that underpins the group’s precision Z → νν measurement; here deployed not to extract a cross section, but to sharpen the broadest BSM search the signature can support.
The result: no excess in data, and limits that lead the world
Working with 101 fb⁻¹ collected in 2017–2018, and combining with the earlier 2016 result for a total of 137 fb⁻¹, the data sit squarely on top of the fitted standard-model expectation. No significant excess appears in any category. That null result, because the background is so well constrained, translates into an unusually wide and stringent set of limits:
B(H → invisible) < 27.8% (25.3% expected) — the most stringent limit from the combined gluon-fusion and V(qq)H channels at the time.
The reach extends well beyond the Higgs portal:
- Spin-1 (vector / axial-vector) DM mediators: excluded up to 1.95 TeV (2.2 TeV expected); couplings probed down to
g_q = 0.018andg_χ = 0.070. - Pseudoscalar mediators: excluded below 470 GeV.
- Fermion-portal (colored scalar) mediators: excluded up to 1.5 TeV.
- Large extra dimensions (ADD): fundamental Planck scale excluded from 10.7 TeV (d = 2) down to 5.2 TeV (d = 7).
- First-generation scalar leptoquarks: couplings from 0.5 to 1.8 excluded for masses of 1.0–2.0 TeV.
Several of these — the spin-1 mediators, the pseudoscalar, the colored mediator, and the leptoquark — were the most restrictive constraints of their kind at publication. One signature, one fit, a whole landscape of models.
The most significant plot
The figure that captures the essence of the analysis is the missing-transverse-momentum spectrum in the monojet signal region. It compares the observed data with the background prediction after the simultaneous fit to all control and signal regions, while representative signal models illustrate how the distribution would have changed in the presence of new physics.
Figure 4 (JHEP 11, 153 (2021)): Missing transverse momentum in the monojet signal region, comparing data to the post-fit standard-model background. Signal templates for a Higgs-portal (B = 25%) and an axial-vector mediator hypothesis are overlaid. The lower panels show the data-to-prediction ratio and the pulls.
Why this one? Because it is the search made visible. The data span several orders of magnitude and track the tamed background across the entire spectrum, with the signal curves peeling away only in the high-momentum tail, exactly the corner of phase space where new physics would first appear. Every world-leading limit in the paper is, in the end, a statement about how well this spectrum agrees with a background that was measured rather than assumed.
The PPC angle
This is a broad search for new physics built on an exceptionally precise understanding of its own most challenging background—and it carries the unmistakable hallmarks of the PPC group. Zeynep Demiragli led the analysis effort, developing the six-region simultaneous fit and the machine-learning event categorization that underpin its sensitivity. Guillelmo Gomez-Ceballos provided critical guidance throughout the review process, helping to refine the analysis strategy and strengthen its presentation. Christoph Paus contributed the central conceptual insight: that the dominant Z → νν background should not simply be estimated, but constrained using the visible decays of the Z boson, transforming the analysis’s principal limitation into its greatest strength.
The same data-driven control-region strategy that powers the group’s precision Z → νν measurement is at the heart of this search. A method developed to understand the standard model background with unprecedented accuracy is now transformed into the engine driving sensitivity to new physics.
We could not see the signal, and we could not pin down the main background either. So we measured everything around them until the invisible had nowhere left to hide.
// Christoph Paus / the PPC, MIT
References
- CMS Collaboration, “Search for new particles in events with energetic jets and large missing transverse momentum in proton-proton collisions at √s = 13 TeV,” JHEP 11 (2021) 153. DOI: 10.1007/JHEP11(2021)153. arXiv: 2107.13021. Report nos. CERN-EP-2021-136, CMS-EXO-20-004.
- Public figures and data: CMS-EXO-20-004 public results; HEPData: 10.17182/hepdata.106115.
- Companion measurement using the same control-region strategy: CMS Collaboration, “Measurement of the Z boson differential production cross section using its invisible decay mode (Z → νν̄),” JHEP 05 (2021) 205, arXiv:2012.09254.
- Predecessor search combined here: CMS Collaboration, “Search for new physics in final states with an energetic jet or a hadronically decaying W or Z boson and transverse momentum imbalance at √s = 13 TeV,” Phys. Rev. D 97 (2018) 092005, arXiv:1712.02345.
- Machine-learning tagging: CMS Collaboration, “Identification of heavy, energetic, hadronically decaying particles using machine-learning techniques” (DeepAK8), JINST 15 (2020) P06005, arXiv:2004.08262.
- V+jets background theory: J. M. Lindert et al., Eur. Phys. J. C 77 (2017) 829, arXiv:1705.04664.
Figure reproduced from CMS-EXO-20-004 under the CMS public-results terms. Spotlight prepared for the MIT Particle Physics Collaboration (PPC).