Partenaires

CNRS
IN2P3
Sorbonne Universite
Universite de Paris
UPMC

Rechercher

Sur ce site

Accueil > Thèses, Stages, Formation et Enseignement > Propositions de thèses 2026 > Search for axion-like particles decaying into collimated photon pairs using machine learning with the ATLAS detector at the LHC

Search for axion-like particles decaying into collimated photon pairs using machine learning with the ATLAS detector at the LHC

par Tristan Beau - 18 novembre 2025

Titre : Search for axion-like particles decaying into collimated photon pairs using machine learning with the ATLAS detector at the LHC

Directrice/directeur de thèse : José Ocariz

Groupe d’accueil : MIF - ATLAS

Webpage du projet : https://atlas.cern/ http://lpnhe.in...

Collaboration : ATLAS

Description :

Light and weakly-interacting particles are predicted by a wide range of theoretical models and are of particular interest as potential dark matter (DM) mediators. Among these, axion-like particles (ALPs) would couple predominantly to gluons and photons. As a result, they could be produced abundantly in proton-proton collisions, yielding a distinctive experimental signature : narrow resonances decaying into photon pairs. This final state benefits from the outstanding performance of the ATLAS liquid-argon (LAr) electromagnetic calorimeter.
The ATLAS and CMS experiments have conducted searches for new diphoton resonances across a broad mass range, both above and below the Higgs boson mass. In 2023, an ATLAS publication led by the LPNHE team extended the diphoton mass search down to 10 GeV for the first time, using 138 fb⁻¹ of Run-2 proton-proton collision data at 13 TeV and focusing on boosted diphoton pairs. In early 2025, CMS presented a preliminary result based on a similar search using 58 fb⁻¹ of 2018 data. More recently, LHCb submitted for publication the results of a search for diphoton resonances in the forward pseudorapidity range, probing masses as low as 4.9 GeV. In all three cases, no evidence of new physics was observed. For ATLAS, the primary limitations arose from energy thresholds in the online diphoton trigger chains used during Run-2 and from reduced selection efficiency for low-energy photons.
This PhD proposal aims to significantly enhance the search potential for ALPs in the diphoton channel during Run-3 by addressing these two limitations. It will leverage the exceptional data-taking conditions of LHC Phase-I, which is expected to conclude in 2026 with an ATLAS dataset exceeding 500 fb⁻¹. The proposal builds on innovative machine learning (ML) techniques to redesign the offline ATLAS diphoton reconstruction and selection criteria, specifically optimized for closely collimated diphotons.
Currently, photons in ATLAS are identified using information from the calorimetric and tracking systems, with criteria trained on single, isolated photon candidates. These criteria are not optimized for collimated photon pairs, where signatures overlap. However, a 2025 PhD thesis from the LPNHE team demonstrated that a ML-based approach, integrating multivariate information, can substantially improve photon and diphoton identification at multiple stages of the ATLAS reconstruction chain. This translates into a major sensitivity boost for searches targeting narrow diphoton resonances, reaching previously unexplored, very low-mass ranges.
The ATLAS LPNHE team has long-standing expertise in electromagnetic calorimetry, dating back to the R&D and construction of the ATLAS LAr calorimeter. Since the LHC startup, the team has contributed to nearly all aspects of electron and photon triggering, reconstruction, and identification. Team members played a key role in the 2012 Higgs boson discovery in the diphoton channel and have led several ATLAS physics analyses involving photons and diphotons, including searches for dark matter produced in association with a Higgs boson decaying to photons, as well as searches for new diphoton resonances across a wide mass range. The team is also an active participant in DMwithLLPatLHC, an ANR-funded project involving three French laboratories focused on exploring dark matter signatures at the LHC.
Currently, the LPNHE ATLAS team is heavily involved in two major upgrade projects for the High-Luminosity LHC (HL-LHC or Phase-II) : the development of a new silicon pixel tracker (ITk) with extended acceptance and a high-granularity timing detector (HGTD) for tracks at large pseudorapidities. The group also maintains a productive collaboration with the Brazilian ATLAS cluster, focusing on improving electron/photon triggers using ML techniques. This includes the design of new trigger chains based on neural networks, intended for implementation in the Phase-II ATLAS trigger system.
This PhD proposal includes contributing to the LPNHE upgrade activities and the development of a new trigger chain dedicated to strongly collimated, low-mass diphotons. The chain will fully exploit ML algorithms optimized for HL-LHC detector conditions, ensuring maximal sensitivity to potential ALP signals.

Lieu(x) de travail : LPNHE

Déplacements éventuels : Missions régulières au CERN, Genève ; un séjour de courte/moyenne durée au Brésil peut être envisagé ; un séjour de courte/moyenne durée au CERN peut être envisagé

Stage proposé avant la thèse : Oui

Facebook

Dans la même rubrique :

Enregistrer au format PDF