Multimodal multi-objective optimization (MMOP) poses significant challenges in multi-objective optimization, as it requires algorithms to obtain high-quality Pareto-optimal solutions in the objective space while identifying multiple diverse yet equivalent solutions in the decision space. However, strong convergence in the objective space often leads to the loss of decision space diversity, causing structural degeneration of the solution set. Moreover, imbalanced interaction mechanisms may lead to population homogenization or a loss of cooperation. To overcome these issues, this paper proposes a dual-population co-evolutionary algorithm that incorporates an enhanced weak interaction mechanism and a Lennard-Jones (LJ) potential-based guidance mechanism. The proposed algorithm first establishes an asymmetric information exchange mechanism, where an elite-guided path is built from the convergence population to the diversity population during the mating and offspring generation stages, effectively balancing convergence and diversity. Then, the environmental selection strategy is changed from parallel to sequential execution, which enhances population heterogeneity, reduces reliance on additional diversity maintenance strategies, and improves stability and robustness. To further improve convergence and diversity across different evolutionary stages, an adaptive candidate solution selection strategy based on the LJ potential model is designed. This strategy re-quantifies the interaction weights among individuals and effectively achieves a dynamic balance between exploration and exploitation. Experimental results on representative MMOP benchmarks show that the proposed algorithm achieves superior performance in diversity, convergence quality, and efficiency, demonstrating strong generalization ability and practical potential.