Multi-Agent Debate (MAD) has emerged as a promising inference scaling method for Large Language Model (LLM) reasoning. However, it frequently suffers from belief entrenchment, where agents reinforce shared errors rather than correcting them. Going beyond merely identifying this failure, we decompose it into two distinct root causes: (1) the model's biased $\textit{static initial belief}$ and (2) $\textit{homogenized debate dynamics}$ that amplify the majority view regardless of correctness. To address these sequentially, we propose $\textbf{DReaMAD}$ $($$\textbf{D}$iverse $\textbf{Rea}$soning via $\textbf{M}$ulti-$\textbf{A}$gent $\textbf{D}$ebate with Refined Prompt$)$. Our framework first rectifies the static belief via strategic prior knowledge elicitation, then reshapes the debate dynamics by enforcing perspective diversity. Validated on our new $\textit{MetaNIM Arena}$ benchmark, $\textbf{DReaMAD}$ significantly mitigates entrenchment, achieving a +9.5\% accuracy gain over ReAct prompting and a +19.0\% higher win rate than standard MAD.

翻译：多智能体辩论（MAD）已成为一种有前景的、用于提升大型语言模型（LLM）推理能力的推断扩展方法。然而，该方法常受信念固守问题困扰，即智能体倾向于强化共享的错误而非纠正它们。我们不仅识别了这一失效模式，更进一步将其分解为两个不同的根本原因：（1）模型存在偏差的 $\textit{静态初始信念}$，以及（2）$\textit{同质化辩论动态}$，该动态会放大多数观点而无论其正确与否。为依次解决这些问题，我们提出了 $\textbf{DReaMAD}$ $($$\textbf{D}$iverse $\textbf{Rea}$soning via $\textbf{M}$ulti-$\textbf{A}$gent $\textbf{D}$ebate with Refined Prompt$)$。我们的框架首先通过策略性的先验知识引导来修正静态信念，然后通过强制视角多样性来重塑辩论动态。在我们新构建的 $\textit{MetaNIM Arena}$ 基准测试上的验证表明，$\textbf{DReaMAD}$ 显著缓解了信念固守问题，相较于ReAct提示方法实现了+9.5%的准确率提升，并且比标准MAD获得了+19.0%的更高胜率。