AgentRivet: an automated system for producing Rivet routines from journal publications

2026-06-11 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Physical Sciences & Chemistry, Artificial Intelligence & Machine Learning, Research Methodology & Innovation · Depth: Expert, extended

Summary

AgentRivet, an automated system using Large Language Models (LLMs), generates Rivet routines directly from particle physics journal publications. Rivet routines are essential for comparing theoretical models to experimental data, aiding Monte Carlo event generator development and searches for physics beyond the Standard Model. Currently, only 39% of measurements have publicly available Rivet routines. AgentRivet, a Python-based AI workflow, extracts physics analysis information, writes C++ Rivet routines, and incorporates intermediate code and physics reviews for quality control. The system was benchmarked using commercial LLMs (OpenAI's Gpt-5.5, Anthropic's Claude-Opus-4.6, and Google's Gemini-3.5-Flash) on two recent ATLAS and CMS measurements. AgentRivet produced competent routines with few syntax errors and reasonable physics fidelity, though some implementation issues arose from ambiguous definitions in publications or LLM struggles with complex observables. The cost per routine ranged from USD1.20 to USD2.20.

Key takeaway

For particle physicists or research scientists aiming to improve analysis preservation, AgentRivet offers a viable path to automate Rivet routine generation. You should consider integrating LLM-powered agent workflows to address the current 39% coverage gap. Be aware that ambiguous publication definitions can still lead to physics implementation issues, requiring careful human oversight or prompt refinement. Explore using multiple LLM providers for redundancy and cost-effectiveness.

Key insights

AgentRivet automates Rivet routine generation from physics papers using LLM-orchestrated agents, addressing a significant gap in analysis preservation.

Principles

• Modular LLM orchestration improves scientific analysis reinterpretation.
• Iterative review loops are critical for complex code generation.
• Ambiguous source definitions cause physics implementation issues.

Method

AgentRivet employs specialized agents (Analyst, Coder, Code Reviewer, Physics Reviewer) in a Python-based workflow. It extracts structured analysis info, generates C++ code, and refines it through an iterative review loop with shared state.

In practice

• Use Pydantic models for structured LLM outputs.
• Implement retry logic for commercial LLM API calls.
• Decompose review into code and physics concerns.

Topics

• AgentRivet
• Rivet Routines
• Large Language Models
• Particle Physics
• Analysis Preservation
• AI Agents

Code references

• hepcedar/AgentRivet

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Exploring LLM-based Verilog Code Generation with Data-Efficient Fine-Tuning and Testbench Automation — Multi-agent LLMs can automate testbench generation, creating high-quality data for efficient Verilog code generation.
• Context-as-a-Service: Surfacing Cross-File Dependency Chains for LLM-Generated Developer Documentation — CaaS enhances LLM documentation agents by surfacing non-obvious cross-file dependencies, improving accuracy and efficiency.
• AgentForge: Execution-Grounded Multi-Agent LLM Framework for Autonomous Software Engineering — Execution-grounded verification and specialized multi-agent decomposition significantly enhance LLM performance in autonomous software engineering.
• AgentFactory: A Self-Evolving Framework Through Executable Subagent Accumulation and Reuse — AgentFactory uses executable Python subagent code for LLM agent self-evolution, improving reliability and efficiency.
• Alternatives for the EDIT tool of LLM agents — Token-efficient LLM agent editing can be achieved via line-specific checksum tags, improving reliability and speed.
• Can Coding Agents Reproduce Findings in Computational Materials Science? — LLM-based coding agents currently struggle to reproduce computational materials science claims due to procedural and execution challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.