Is that the shortest road path? He uses geodesic in his post. I thought it might be explained because of the projection used - projected distance measurements can differ from geocoordinate measurements. So I messed around in ArcGIS for a bit, getting WGS 1984 Web Mercator (auxilliary) projection length measurements first - that's the one used by Openstreetmaps, and I assume Google Maps uses the same. Then vanilla geocoordinate system. There wasn't a big difference between planar and geodesic. But I did find your proof that the line is longer:
The other length measurement was 565 miles. Am I missing something or is he wrong?
The Pleasant Prarie - Aurora - Highland Park - Country Club Hills portion is more obviously sub-optimal.
Then near New York City, the path passes within about three miles of three stations without stopping, then doubles back more than 15 miles to connect them.
- Am I missing something or is he wrong?
The Concorde tool he used "can find the exact optimal path" but I can't tell from the documentation what it returns if it cannot definitely find the shortest path; perhaps it does the best it can. We don't see how Mr. Mehyar used it because he skips a step in his wonderful report:
[13] # create input file for Concorde TSP solver
[14] # after running the Concorde executable, parse the output file- There wasn't a big difference between planar and geodesic
Apparently TSPs with a large number of nodes can be definitively solved, and this has been true for some time. I would be surprised if the Concorde program would report a path as the definite solution when there is still some doubt, but I can't believe that what looks like the long way around New Mexico is actually the shortest path.