This study introduces an improved neighbor search algorithm for efficiently locating material points in numerical simulations using the Material Point Method (MPM). In MPM, determining which elements or nodes transmit and receive information from material points is critical for accuracy and performance, particularly in large-scale dynamic geotechnical problems. Traditional methods rely on mesh structures, where structured meshes simplify this process, but unstructured meshes complicate it, leading to longer computational times[1]. To address these challenges, this research proposes a space-partitioning technique using an auxiliary grid that reduces the number of elements requiring checks during simulations. By employing vectorization and organizing semi-structured meshes into sub-meshes with different spacings, the algorithm significantly improves efficiency [2]. Validation results show that the method enhances computational speed, particularly for large models, without compromising accuracy. This approach offers a promising solution for handling complex, unstructured meshes in MPM simulations and is planned for integration in upcoming releases of the Anura3D software.