⚛️ high-energy experiments

Search for a light pseudoscalar Higgs boson in final states with boosted muon pairs and tau lepton pairs in proton-proton collisions at $\sqrt{s}$ = 13 TeV

Using 138 fb $^{-1}$ of proton-proton collision data at $\sqrt{s}$ = 13 TeV collected by the CMS experiment, this study presents a search for a light pseudoscalar Higgs boson in the $H \to aa \to \mu\mu\tau\tau$ final state employing novel boosted ditau reconstruction techniques, finding no significant excess over the Standard Model background and setting the first LHC upper limits on such decays for a heavy scalar boson up to 1 TeV.

Original authors: CMS Collaboration

Published 2026-09-14

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: CMS Collaboration

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

At the heart of modern physics lies a fundamental question: what gives the universe its substance? In the standard model of particle physics, the prevailing theory that describes how matter and forces interact, this property is granted by a field that permeates all of space. Particles moving through this field acquire mass, much like an object moving through thick syrup encounters resistance. The existence of this field was confirmed in 2012 with the discovery of the Higgs boson, a particle that acts as a ripple in the field. However, the standard model is known to be incomplete; it cannot explain gravity, dark matter, or why the universe contains more matter than antimatter. This has led physicists to suspect that the Higgs boson might not be alone. Theories suggest there could be a hidden sector of particles, perhaps including lighter cousins of the Higgs that interact with the matter we can see. Finding these hidden particles would be a monumental step toward a more complete understanding of reality.

To hunt for these elusive particles, researchers at the Large Hadron Collider in Switzerland smash protons together at nearly the speed of light, recreating the intense energy conditions of the early universe. When these protons collide, they can produce heavy particles that instantly decay into lighter ones. The CMS experiment, one of the massive detectors surrounding the collision point, acts as a high-speed camera capturing the debris of these collisions. In a recent analysis, the CMS team focused on a specific scenario where a Higgs boson, or a similar heavy particle, splits into two lighter particles called pseudoscalars. These light particles then decay further: one turns into a pair of muons (heavy cousins of electrons), and the other turns into a pair of tau leptons (even heavier cousins of electrons). The challenge is that if the light pseudoscalar particles are very light, the tau leptons they produce are moving so fast that they are squeezed together, appearing almost as a single object rather than two distinct particles.

The researchers examined data collected between 2016 and 2018, corresponding to an integrated luminosity of 138 fb⁻¹. They were looking for a specific signature: two muons and two tau leptons emerging from a single collision point. Because the tau leptons from a light pseudoscalar are so tightly packed, standard methods of identifying them often fail, causing the signal to be missed. To solve this, the team developed new, sophisticated techniques. They created a specialized deep neural network, a type of computer program trained to recognize the unique, compressed pattern of a pair of tau leptons that are moving in nearly the same direction. They also refined how they identified tau leptons that decayed into other particles, ensuring they could spot these tightly packed pairs even when they were buried in background noise from other common particle interactions.

After applying these new tools to the massive dataset, the researchers found no significant excess over the standard model background. The number of events they observed was consistent with the predictions for known standard model processes, allowing for expected statistical fluctuations. There was no unexpected spike in the data that would indicate the presence of a new light pseudoscalar boson. Consequently, the team set strict upper limits on how often such a decay could occur. They determined that if a Higgs boson with a mass of 125 GeV were decaying into these light particles, the probability would range from 5 × 10⁻⁵ to 2.7 × 10⁻⁴, and for heavier versions of the Higgs boson, the probability is even lower. These results effectively rule out a wide range of theoretical models that predicted these particles would be common enough to be seen with the current data.

This work is significant not only because it narrows the search for new physics but also because it demonstrates a new way of looking at the data. By successfully identifying the tightly packed tau lepton pairs, the team opened a window into a region of particle masses that was previously difficult to explore. While the search did not find the new particles, the absence of a discovery is a powerful result in itself. It tells theorists that if these hidden particles exist, they are either much rarer or behave differently than the most popular models predicted. The analysis extends the reach of previous searches, covering a broader range of masses and providing the most stringent constraints to date on how the Higgs boson might decay into these lighter, hidden states. The search for the hidden sector of the universe continues, but with every negative result, the path to the truth becomes slightly clearer.

Technical Summary: Search for a Light Pseudoscalar Higgs Boson in Boosted $\mu\mu\tau\tau$ Final States

Problem and Motivation
This paper presents a search for a light pseudoscalar Higgs boson ( $a$ ) produced in the decay of a scalar Higgs boson ( $H$ ), specifically the process $H \to aa \to \mu\mu\tau\tau$ . The analysis targets scenarios where $H$ is either the Standard Model (SM) Higgs boson ( $m_H = 125$ GeV) or a heavier scalar state ( $m_H = 250\text{--}1000$ GeV) predicted by extensions of the Higgs sector, such as the Next-to-Minimal Supersymmetric Standard Model (NMSSM) and Two-Higgs-Doublet Models plus a singlet (2HDM+S).

A primary challenge in this search is the kinematic regime: due to the large mass difference between $H$ and the light pseudoscalar $a$ , the decay products of $a$ are highly Lorentz boosted and collimated. This topology causes the decay products of the tau lepton pairs to overlap significantly, often leading to inefficiencies in standard CMS tau lepton reconstruction algorithms. Previous searches focused on resolved topologies or specific decay channels; this analysis aims to cover the full range of kinematically accessible $a \to \tau\tau$ decays, including those where the tau leptons are highly boosted and reconstructed as single jets.

Methodology
The analysis utilizes proton-proton collision data collected by the CMS experiment at $\sqrt{s} = 13$ TeV from 2016 to 2018, corresponding to an integrated luminosity of 138 fb $^{-1}$ . The search targets the final state $\mu\mu\tau\tau$ , where one pseudoscalar decays to a dimuon pair ( $a \to \mu\mu$ ) and the other to a tau pair ( $a \to \tau\tau$ ). The $\tau\tau$ system is analyzed in four decay channels: $\tau_\mu\tau_e$ , $\tau_\mu\tau_h$ , $\tau_e\tau_h$ , and $\tau_h\tau_h$ (where $\tau_h$ denotes a hadronic tau decay).

Key methodological developments include:

Dedicated Reconstruction Algorithms: To address the collimation of boosted tau pairs, the analysis employs modified reconstruction techniques for $\tau_\mu\tau_h$ and $\tau_e\tau_h$ channels. These involve modifying the jet seed by removing overlapping muons or electrons before applying the standard Hadron-plus-Strips (HPS) algorithm.
Deep Neural Network (DEEPDITAU): A significant innovation is the development of a deep neural network tagger, DEEPDITAU, specifically designed to identify collimated $\tau_h\tau_h$ pairs that are reconstructed as single AK4 jets. This network classifies jets into light quarks/gluons, bottom quarks, and $\tau_h\tau_h$ pairs, achieving a 70% tagging efficiency for signal with a 1% mistag rate for QCD jets.
Event Selection and Background Estimation: Events are selected requiring two isolated muons and a tau pair candidate. The dominant background, Drell-Yan $\mu\mu$ production with jets misidentified as taus, is estimated directly from data using a "tight-to-loose" ratio method derived from control regions.
Statistical Analysis: A two-dimensional unbinned maximum likelihood fit is performed on the distributions of the dimuon invariant mass ( $m_{\mu\mu}$ ) and the four-body visible mass ( $m_{\mu\mu\tau\tau}$ ). The signal is modeled as a narrow resonance in $m_{\mu\mu}$ and a broader distribution in $m_{\mu\mu\tau\tau}$ .

Key Contributions

First Limits on Boosted $\tau_h\tau_h$ in this Context: This is the first LHC analysis to successfully include the $\tau_h\tau_h$ final state for boosted pseudoscalar searches using a dedicated deep learning tagger, thereby covering 94% of the theoretically accessible $a \to \tau\tau$ decays and setting upper limits for the first time in this channel.
Extended Mass Reach: The analysis extends previous searches to lower pseudoscalar masses ( $m_a$ as low as 3.6 GeV) and investigates heavy scalar masses ( $m_H$ ) up to 1 TeV, a range not previously probed in the $\mu\mu\tau\tau$ channel for boosted topologies.
Improved Efficiency: The modified reconstruction techniques for $\tau_\mu\tau_h$ and $\tau_e\tau_h$ channels significantly improve signal efficiency (by factors of up to 2–4) compared to standard HPS algorithms, particularly for low $m_a$ where the boost is highest.

Results
No significant excess over the Standard Model background was observed in the data. Consequently, model-independent 95% confidence level (CL) upper limits were set on the ratio of the production cross section times branching fraction to the SM cross section: $\sigma_H \times \mathcal{B}(H \to aa \to \mu\mu\tau\tau) / \sigma_{\text{SM}}$ .

For $m_H = 125$ GeV: Limits range from $5 \times 10^{-5}$ to $2.7 \times 10^{-4}$ for pseudoscalar masses between 3.6 and 21 GeV.
For $m_H = 1$ TeV: Limits range from $3.5 \times 10^{-4}$ to $1.5 \times 10^{-3}$ for pseudoscalar masses between 3.6 and 50 GeV.
Model Interpretation: Limits were interpreted within the context of 2HDM+S models (Types I, II, III, and IV). For Type-I and Type-II models, the results provide the most stringent limits to date on $\mathcal{B}(H \to aa)$ in the range $3.6 < m_a < 4.2$ GeV. For Type-III models, the limits improve upon previous results in the range $13 < m_a < 15$ GeV.

Significance
The paper claims that these results significantly extend the exclusion limits obtained in earlier CMS and ATLAS searches, particularly at low $m_a$ where the boosted topology is dominant. By combining multiple tau decay channels and introducing the DEEPDITAU tagger, the analysis achieves a sensitivity improvement over previous boosted searches (e.g., Ref. [47]) by factors of 2–5 for $m_H = 125$ GeV. Furthermore, this work represents the first search for the decay of a heavy neutral scalar ( $m_H \le 1$ TeV) into two promptly decaying pseudoscalars in the full Run 2 dataset, providing crucial constraints on extended Higgs sector models that predict light pseudoscalars.

More like this