Understanding the Information Propagation Effects of Communication Topologies in LLM-based Multi-Agent Systems
Xu Shen, Yixin Liu, Yiwei Dai, Yili Wang, Rui Miao, Yue Tan, Shirui Pan, Xin Wang
TL;DR
This work analyzes how communication topology shapes information propagation in LLM-based MAS, introducing CAPE and TCTE to quantify agent-level and topology-level influence. It shows that moderately sparse topologies best balance suppression of erroneous signals with diffusion of beneficial insights. The authors propose EIB-Learner, a dual-view GNN-based topology learner that adaptively fuses sparse and dense connectivity under a query, optimized via policy gradients. Empirical results across six benchmarks demonstrate improved accuracy, reduced communication cost, and enhanced robustness, highlighting the practical value of balanced topology design for scalable, reliable MAS.
Abstract
The communication topology in large language model-based multi-agent systems fundamentally governs inter-agent collaboration patterns, critically shaping both the efficiency and effectiveness of collective decision-making. While recent studies for communication topology automated design tend to construct sparse structures for efficiency, they often overlook why and when sparse and dense topologies help or hinder collaboration. In this paper, we present a causal framework to analyze how agent outputs, whether correct or erroneous, propagate under topologies with varying sparsity. Our empirical studies reveal that moderately sparse topologies, which effectively suppress error propagation while preserving beneficial information diffusion, typically achieve optimal task performance. Guided by this insight, we propose a novel topology design approach, EIB-leanrner, that balances error suppression and beneficial information propagation by fusing connectivity patterns from both dense and sparse graphs. Extensive experiments show the superior effectiveness, communication cost, and robustness of EIB-leanrner.