Organizations: School of Physics, Peking University, Beijing 100871, China · Center for High Energy Physics, Peking University, Beijing 100871, China · School of Physics, Zhengzhou University, Zhengzhou 450001, China · State Key Laboratory of Nuclear Physics and Technology, Peking University, Beijing 100871, China · Beijing Computational Science Research Center, Beijing 100193, China · Institute of Particle Physics and Key Laboratory of Quark and Lepton Physics (MOE), Central China Normal University, Wuhan, Hubei 430079, China
We present, to our knowledge, the first language-driven agent system capable of executing end-to-end collider phenomenology tasks, instantiated within a decoupled, domain-agnostic architecture for autonomous High-Energy Physics phenomenology. Guided only by natural-language prompts supplemented with standard physics notation, ColliderAgent carries out workflows from a theoretical Lagrangian to final phenomenological outputs without relying on package-specific code. In this framework, a hierarchical multi-agent reasoning layer is coupled to Magnus, a unified execution backend for phenomenological calculations and simulation toolchains. We validate the system on representative literature reproductions spanning leptoquark and axion-like-particle scenarios, higher-dimensional effective operators, parton-level and detector-level analyses, and large-scale parameter scans leading to exclusion limits. These results point to a route toward more automated, scalable, and reproducible research in collider physics, cosmology, and physics more broadly.
Figures & tables
Figure 1: Decoupled architecture of ColliderAgent . A hierarchical cognitive reasoning engine assigns phenomenological tasks to specialized sub-agents equipped with dedicated Agent Skills , which communicate through a standardized CLI with Magnus , the unified execution backend hosting the underlying HEP tools. The framework translates natural-language input and standard physics notation into final phenomenological outputs such as kinematic distributions and exclusion limits.
Figure 2: Representative literature reproductions by ColliderAgent . (a) mej distribution for pp→LQ→ej at the LHC in the minimal scalar LQ model. (b) Normalized missing transverse energy distribution for pp→aW±γ with W±→ℓ±ν at the LHC in the ALP EFT. (c) Expected 2σ exclusion contours for a Z′ model from Drell-Yan production at the LHC. (d) 2σ exclusion contour for the U1 LQ from the LHC mono- τ search. Panels (a)-(d) reproduce Refs. [ 18 , 17 , 3 , 31 ] , respectively. See text for details.
Figure S1: Additional literature reproductions by ColliderAgent . (a) Production cross section for pp→μ±N at the LHC as a function of mN . (b) Dilepton production cross section for a general Z′ benchmark at the 13TeV LHC as a function of MZ′ for the SSM and Zψ′ scenarios. (c) Cross section for e+e−→μ+μ− as a function of s in the Randall–Sundrum model, showing the resonant structure of the KK graviton tower. (d) 95% CL exclusion and 5σ discovery contours for the U1 leptoquark benchmark from μ+μ−→bbˉ at a muon collider. Panels (a)-(d) reproduce Refs. [ 27 , 49 , 24 , 11 ] , respectively. See text for details.
Autonomous research agents use language models and tools to search literature, design and run experiments, analyze results, and support scientific discovery.
University of Chicago, Enrico Fermi Institute, 60637 IL, USA · Istituto Nazionale di Fisica Nucleare, Bari Division, BA 70126, Italy · Vanderbilt University, Nashville, Tennessee, USA +1