LoopBench: Discovering Emergent Symmetry Breaking Strategies with LLM Swarms

2025-12-07

AI Summary

This work addresses deadlock arising from symmetry in distributed graph coloring tasks—specifically on odd-length cycles (C₃, C₅, C₁₁)—by non-communicating intelligent agents. Method: We introduce LoopBench, the first benchmark systematically evaluating large language models’ (LLMs) joint capabilities in distributed symmetry breaking and metacognitive reasoning. Our approach designs an LLM-based multi-agent coordination mechanism grounded in policy transfer and shared memory, integrating prompt engineering, graph-theoretic constraint modeling, and a metacognitive reasoning framework to enable collaborative decision-making under ultra-low communication. Contribution/Results: We demonstrate that LLM swarms can spontaneously emerge correct distributed algorithms; notably, advanced reasoning models such as O3 achieve zero-deadlock coloring across all tested cycle graphs—surpassing the limitations of conventional deterministic protocols—and establish LLM swarms as a novel emergent substrate for distributed algorithm execution.

Technology Category

• Multiagent Systems: Distributed Problem Solving
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large Language Models (LLMs) are increasingly being utilized as autonomous agents, yet their ability to coordinate in distributed systems remains poorly understood. We introduce extbf{LoopBench}, a benchmark to evaluate LLM reasoning in distributed symmetry breaking and meta-cognitive thinking. The benchmark focuses on coloring odd cycle graphs ($C_3, C_5, C_{11}$) with limited colors, where deterministic, non-communicating agents fail in infinite loops. A strategy passing mechanism is implemented as a form of consistent memory. We show that while standard LLMs and classical heuristics struggle, advanced reasoning models (e.g., O3) devise strategies to escape deadlocks. LoopBench allows the study of emergent distributed algorithms based on language-based reasoning, offering a testbed for collective intelligence.

Problem

• Evaluates LLM coordination in distributed symmetry breaking tasks
• Tests emergent strategies for escaping infinite loops in graph coloring
• Studies language-based reasoning for collective intelligence in autonomous agents

Innovation

• LoopBench benchmark tests LLM reasoning in distributed symmetry breaking
• Strategy passing mechanism acts as consistent memory for agents
• Advanced reasoning models devise strategies to escape deadlock loops