Constraint satisfaction problems have been widely investigated in artificial intelligence area. This paper investigates whether the coarse-grained maintaining arc is consistent, which is used to solve CSPs. The study has found that during the search for a solution, there are ineffective revisions, which revise the arcs that point to assigned variables. This study has proved that such revisions are redundant and proposed a method to avoid such redundant revisions. The paper gives the improved basic frame for the coarse-grained arc consistency algorithms, named AC3_frame_ARR. The new frame can be applied to improve all the coarse-grained AC algorithms. The experimental results show that the improved algorithms can save at most 80% revisions and 40% CPU time.