MAC-Reasoner: A Multi-Agent Collaborative Framework for Enhancing Logical Reasoning in Large Language Models

Large language models (LLMs) face challenges in logical reasoning where correctness requires strict deductive procedures. Purely modelbased approaches often suffer from hallucinations, while neuro-symbolic methods typically delegate deduction to external solvers, reducing the LLM to a mere translator. To address this, we propose MAC-Reasoner, a multi-agent framework that constructs a Logic-Augmented Context to enhance LLMs' reasoning. In this framework, a translator agent converts problems into executable symbolic programs. Symbolic information from solver execution is transformed into the Logic-Augmented Context, serving as a verification reference where logical conflicts trigger heightened attention to violated constraints. We evaluate MAC-Reasoner with three backbone LLMs on four challenging benchmarks. Results show consistent and robust improvements over baselines. Furthermore, reasoning traces from MAC-Reasoner can be used for supervised finetuning of LLMs to achieve more accurate and efficient logical reasoning. 1 * Corresponding author. 1 Code is available at https://github.com/l-yh0/ MAC-Reasoner. # Declarations ... ForAll([c:colors], Implies(used(c), And(used_in_rug(c) == rug1, used_in_rug(c) == rug2, used_in_rug(c) == rug3))) ....

(a) CoT (b) Neuro-symbolic method ... Scenario D: Olive (O) is the unused color. ... * Resulting Rugs: F, P, W, T, Y.