Thanks to authors :)

It became such when true claster analysis began
or inner structure of each claster is forgotten and clasters  are simple poins again.
Some time I was tried to trace complicated inner structure but failed.

is there simpler solution?? I use brute_force search....

Yes, O(n^3) solution (maybe can be optimized to O(n^2*log(n)) with nice tree merging) based on MST.
Great problem!!!

yes. love the problem.

some tests:

5   0 0     0 1     0 2     0 4     0 5
5   0 0     0 1     0 3     0 4     0 5
0