When you assign a resource to a task, you probably assume that the resource will work continuously on the task for the entire task duration. But in reality a resource may work on a task on and off, stopping and starting several times before completing the task. You can accurately represent discontinuous work on a task by splitting the task.
When you split a task, you divide it into portions, each portion represented by a separate bar in the Gantt Chart view. The beginning of the first portion shows when work starts on the task. The beginning of each successive portion shows when work resumes. Between each portion is a gap that indicates a period of time when the task isn’t being worked on.
Split a task when work on that task will stop for a known amount of time. That way, you can get a more accurate picture of how much time a resource will actually spend on the task. If the resource is overallocated, you might reduce the amount of the overallocation or even eliminate it altogether. You can also split a task for which some work has already been done, to indicate when work will begin again on the remaining portion of the task.