I have found some unexpected behavior of ArcGIS union tool. My assumption was this tool will never add any new vertex to the output layer but this is not the case.
I have run the ArcGIS union tool on only one layer GADM (world admn01) layer and found that in output layer there are around 866 vertex new. I think this tool make tiles of data and then process. After processing this extra vertexes which are basically created because of tiles are not deleted.
Is there any body else who have same problem/experience?
I am using ArcGIS 10.0 on Windows7 64bit
Tile processing of large datasets was introduced in ArcGIS 9.2 (well, it could be done in 9.1, but it was more manual and not integrated or automatic). Even without looking at the Help, the progress window for the Union tool shows there is tiling occurring:
Maybe there were not enough qualifiers on your statement about the Union tool and new vertices because the Union tool has to add new vertices in many cases:
If your statement was meant for self unions, then I agree that new vertices seem spurious.
What steps are you using to find the extra vertices?
I am running ArcGIS 10.2.2 with all but the last couple patches, and I did find extra vertices. I only closely looked at a handful of the extra vertices, but they were all duplicate vertices. If it turns out that all of the extra vertices are duplicates, then there is no difference in the geometries of the polygons, well the perimeter and such.
The reason I asked what steps you took to find extra vertices is that I used the Feature Vertices to Points tool. Without doing more digging, I can't say whether the issue resides in the Union tool or Feature Vertices to Points tool.
Beyond the tools, have you checked how clean the polygons are? Looking at a few of the duplicate vertices, they seem to be near areas where the polygons seem to be self intersecting, or really close.
Try a Repair or Check Geometry to see if the result is producing short segments or other features that can be fixed. I see this is an advantage in some instances since one may be interested in determining the locations of those extra points.
Let suppose I have two seamless layers A and B (topologically corrected). I run union on Layer_A only and produce Layer_A_Union. Now the Layer_A_Union is not seamless with Layer_B any more.
I am still thinking how this is an advantage or I am missing some thing?
The complexity, size, or both of the dataset is driving whether or how tiling is being used. If you have access to ArcGIS 10.1 or higher, you could try using 64-bit Background Geoprocessing. The 64-bit Background Geoprocessing tools allow for much more RAM to be utilized, which could change or eliminate tiling and get you the results you are after.