<p>As an essential extension of classical rough sets, multigranulation rough sets have emerged as a powerful tool for data processing and analysis. Among them, optimistic multigranulation rough sets, a vital type of multigranulation rough sets, have garnered substantial research attention. The focus of this paper is an investigation of attribute reduction in optimistic multigranulation rough sets. To address the issue that restrictions on relations may limit the applicability of reductions, the optimistic lower approximation distribution reduction is first extended to arbitrary relations in relation decision systems and a corresponding discernibility matrix-based reduction algorithm is developed. Subsequently, the optimistic upper approximation distribution reduction is extended to relations without any restrictions and a corresponding discernibility matrix-based reduction algorithm is proposed. To evaluate the effectiveness of the proposed algorithms, comparative experiments were conducted on 12 public datasets. The results demonstrated that the selected classification algorithms achieved higher classification accuracy on the reduced datasets obtained by the proposed algorithms with optimal granular structure configurations than on the raw datasets.</p>