A position map (sometimes called a world position pass), is an image where each pixel's R, G, B colour values represent the x, y, z coordinates of the corresponding vertex, in 3D world space. In Action, a position map works as a type of 3D displacement, that is, it displaces every pixel in that image to the actual location in space of the vertex it represents, allowing you to:
A typical workflow would be to import (from a 3D application) the beauty pass, the position pass (as a 16-bit floating point OpenEXR clip), the normal pass, and the FBX camera that rendered the scene. Then, apply the position and the normal map to the beauty pass image surface, and set the scale of the position map so that it aligns properly with the FBX camera frustum.
When looked from the original camera POV, your scene looks just like the rendered image. But since pixels are accurately located in space relative to the original 3D scene with the attached position map, you can correctly position 2D layers in the same perspective as objects of the 3D scene.
The position object is added to the schematic with its own parent axis. The new axis is the child of the selected surface. In Schematic view, the number in brackets next to the name of the Position node indicates the media used for the position map.