ArgoLOOM: agentic AI for fundamental physics from quarks to cosmos

TL;DR

The paper addresses the need for cross-frontier computational workflows in fundamental physics and introduces ArgoLOOM, an agentic AI framework that orchestrates cosmology, collider physics, and nuclear QCD analyses. Using a GPT-based backbone with a curated FAISS-enabled knowledge base and run-card based reproducibility, ArgoLOOM can generate end-to-end workflows across domains, producing cosmology outputs such as angular power spectra $C_\ell$ and derived $D_\ell$, as well as collider observables and DIS mappings. Two case studies—top-down sterile-neutrino scenarios and bottom-up high-$x$ DIS to TeV-scale models—demonstrate cross-domain planning and the system's extensibility. The work outlines the architecture, knowledge-base design, and domain-specific modules, underscoring a path toward scalable, interpretable, and reproducible cross-domain discovery pipelines in fundamental physics.

Abstract

Progress in modern physics has been supported by a steadily expanding corpus of numerical analyses and computational frameworks, which in turn form the basis for precision calculations and baseline predictions in experimental programs. These tools play a central role in navigating a complex landscape of theoretical models and current and potential observables to identify and understand fundamental interactions in physics. In addition, efforts to search for new fundamental interactions increasingly have a cross-disciplinary nature, such that understanding and leveraging interoperabilities among computational tools may be a significant enhancement. This work presents a new agentic AI framework, which we call ArgoLOOM, designed to bridge methodologies and computational analyses across cosmology, collider physics, and nuclear science. We describe the system contours, key internal aspects, and outline its potential for unifying scientific discovery pipelines. In the process, we demonstrate the use of ArgoLOOM on two small-scale problems to illustrate its conceptual foundations and potential for extensibility into a steadily growing agentic framework for fundamental physics.

Figures (3)

• Figure 1: A depiction of the ArgoLOOM agentic workflow, focusing on the main components: a primary backbone model to orchestrate calculations and information flow in the cosmology, particle, and nuclear physics domains, as well as interaction with a specialized knowledge base.
• Figure 2: Once ArgoLOOM is directed toward a standard CLASS build, it can agentically setup run cards and executions base on a range of cosmology scenarios and microphysics assumptions. These may be steered based on initial dialogue with the backbone agent and consultation of the knowledge base. In this example, we illustrate a set of power spectra (left) generated by ArgoLOOM as well as families of deviations from these defaults based on sterile neutrino scenarios (right).
• Figure 3: Mirroring the CLASS setup, ArgoLOOM can also invoke and steer computations relevant for particle-nuclear physics. For instance, these can include the assumption of a definite SM or BSM theory baseline ( i.e., corresponding to a particular Lagrangian), from which collider simulations and kinematical matchings to observables relevant to nuclear science might be performed. We illustrate this with a rapidity distribution in $y_{\ell^+\ell^-}$ for 13 TeV neutral-current Drell-Yan (left) and corresponding kinematical matchings relevant for DIS observables with modeled detector-smearing effects (right), both produced in automated fashion within ArgoLOOM.