Advancements in particle tagging accelerate the search for new particles

5 November 2024 | By

At experiments on the Large Hadron Collider (LHC), physicists are searching for new physics phenomena that could explain unresolved mysteries, like the nature of dark matter and the matter-antimatter asymmetry of the universe. The ATLAS Collaboration is advancing this effort with innovative techniques to identify — or “tag” — the known particles produced in their experiment. By efficiently tagging particles based on their distinctive characteristics, researchers can zoom into specific interactions that may reveal undiscovered phenomena.

Physicists have been using particle tagging techniques to identify beauty quarks since the 1990s, with the first developed for experiments at the Large Electron-Positron (LEP) Collider. Recent advancements in algorithm development, often utilising deep learning techniques, have enabled researchers to tag other “flavours” of quarks with more subtle features, such as top and charm quarks. Since top and charm quarks decay before reaching the detector material, these algorithms rely on the experimental signatures of their decay products. Top quarks decay into a W boson and beauty quark, while charm quarks produce a spray of particles called “jets”. The W boson can further decay into leptons or lighter quarks, which also leave behind a jet signature.

The ATLAS Collaboration has released three major searches benefiting from advancements in particle tagging: the first exploring final states with top quarks and missing transverse momentum (MET); the second focusing on top quarks, charm quarks, and MET; and the third examining solely charm quarks and MET. In all three, MET plays a crucial role. Before a collision, the incoming protons have zero total transverse momentum, since their momentum is directed along the beamline. After the collision, conservation laws require the transverse momentum of the outgoing particles to also sum to zero. By measuring all outgoing particles, any imbalance in this plane indicates the presence of invisible particles, like neutrinos or potential dark matter particles, which escape detection and cause the “missing” momentum.


Recent advancements in algorithm development have enabled researchers to tag new “flavours” of quarks, such as top and charm quarks.


Tops, charms and a dash of MET

In the first result, researchers searched for dark matter and supersymmetric particles in collision events involving a pair of top quarks, where each quark decays in different ways (see event display). The W boson from the first top quark decays into a lepton and a neutrino (leptonic decay), while the W boson from the second decays into two quarks, which appear as jets (hadronic decay). Neural network-based classifiers were used extensively in this analysis, significantly boosting its sensitivity. They were employed to tag and reconstruct the properties of the top quark that decays hadronically, aiding in the understanding of the event’s kinematic properties. They also helped distinguish signals of potential new physics from Standard-Model backgrounds. Researchers categorised events based on the kinematics of the final state, allowing them to increase the generality of their search.

The second ATLAS result focused on collision events where both a top quark and a charm quark have been produced. This is the first time at the LHC that this mixed final state has been investigated for the presence of new particles – an accomplishment facilitated by recent advancements in charm-tagging techniques by the ATLAS Collaboration. This search was motivated by a non-traditional supersymmetry scenario, in which quarks can transform from one flavour to another through weak interactions (non-minimal flavour mixing). Identifying new physics in such events is particularly challenging, especially when dealing with low-momentum objects, as they can closely resemble Standard-Model particles. To address this, researchers used a multi-class neural network that provided exceptional discrimination power in these cases.

Motivated by supersymmetric and leptoquark models, the third ATLAS search focused on collision events where only charm quarks are produced. This was the first study of events with two charm-tagged jets, as researchers took advantage of the improved charm-tagging algorithms and the large dataset collected during Run 2 of the LHC. In addition, they used an advanced analysis technique called “Recursive Jigsaw Reconstruction”, which incrementally pieces together decay patterns to reconstruct particle kinematics, increasing the results’ sensitivity in cases where the produced particles have relatively low momentum.

Physics,ATLAS
Figure 1: Exclusion limits at the 95% confidence level on the supersymmetric-top quark mass. The BR scan interprets different levels of flavour mixing between the second and third quark generations in Supersymmetry models. (Image: ATLAS Collaboration/CERN)
Physics,ATLAS
Figure 2: Exclusion limits at the 95% confidence level on the supersymmetric-charm or -top quark mass. Improved charm-tagging algorithms and the Recursive Jigsaw Reconstruction technique allowed higher experimental reach by ~100 GeV for the compressed Supersymmetric mass spectra. (Image: ATLAS Collaboration/CERN)

These major searches highlight the power of advanced particle tagging techniques and machine-learning methods in the ongoing search for new-physics phenomena.


Setting new limits

Physics,ATLAS
Figure 3: Exclusion limits at 95% confidence level for up-type scalar leptoquarks coupled to first-generation leptons. This high BR and low mass (thus high production cross-section) region is excluded for the first time at the LHC. (Image: ATLAS Collaboration/CERN)

The ATLAS Collaboration has set new limits on potential dark matter, supersymmetry and leptoquark models, ruling out higher mass ranges for certain hypothetical particles than were previously possible. In the two top-quark searches, the masses of supersymmetric-top quarks were constrained above 1.2 TeV, while those of dark matter mediator particles were restricted to above 350 GeV for specific benchmark scenarios. In the mixed top and charm quark search with maximal flavour mixing, supersymmetric-top quarks were excluded up to 800 GeV and neutralinos up to 400 GeV. Results are also provided for different levels of mixing by analysing a range of possible decay probabilities or “Branching Ratios” (BR), as shown in Figure 1. Lastly, in the charm-quark-pair search, ATLAS has improved the experimental reach by ~100 GeV for supersymmetry with a compressed mass spectra, as shown Figure 2, and, for the first time at the LHC, scalar leptoquarks with masses below 900 GeV have been excluded, as shown Figure 3.

These major searches highlight the power of advanced particle tagging techniques and machine-learning methods in the ongoing search for new-physics phenomena. By leveraging improved charm- and top-tagging algorithms, as well as innovative methods like Recursive Jigsaw Reconstruction, ATLAS researchers are pushing the boundaries of their experimental reach.


About the event display: A reconstruction of a proton-proton collision at 13 TeV from 2018, classified as very “signal-like” by neural networks. Orange lines trace particle trajectories while green and yellow blocks show calorimeter energy deposits. The blue line marks the muon’s path and yellow cones represent four reconstructed jets. The three closest jets form the hadronic top-quark candidate, identified by the neural-network classifier, while the remaining jet is a b-tagged jet which, together with the muon, form the leptonic top-quark candidate. The white dashed line indicates a missing momentum of 544 GeV. (Image: ATLAS Collaboration/CERN)

Learn more