This is not checksum-based but rather on the delta of the file versions (unless the checksums are absolutely identical, i.e. a rename without additional changes in the file).
The used SHA-1 (a hash actually, not only a checksum) was designed for security and therefore does not expose any hints to similarity as far as research goes.
That said, you might also look into git's source code to be sure how it works. Maybe start with the piece of front-end code that shows that a rename has taken place, and work your way backwards to the locations where the detection is done.