Entropy Maximization Alone Can Improve LLM Reasoning Performance？