My conjecture is false. Following counterexample is given on pp. 234 of this paper:
As pointed out by uncombed_coconut, the correct theorem is:
Theorem (Berlekamp-Gilbert-Sinden, 1965). For almost all simple polygons there exist a smallest natural number such that after iterations of midpoint polygon procedure, we obtain a convex polygon.
The proof of this theorem is very interesting. Till now I thought that proving euclidean geometry theorems using complex numbers was an overkill. But using an N-tuple of complex numbers to represent the vertices of a closed polygon (in given order), , we can restate the problem in terms of eigenvectors (referred to as eigenpolygons) and eigenvalues. Following are the crucial facts used is the proof:
- An arbitrary N-gon (need not be simple) can be written as a sum of regular N-gons i.e. the eigenvalues are distinct.
- The coefficient of eigenvector (when N-gon is written as linear combination of eigenpolygons) is the centroid of the polygon obtained by “winding” times.
- All vertices of the midpoint polygons (obtained by repeating the midpoint polygon procedure infinitely many times) converge to the centroid.
- The sum of two convex components of is a polygon. This polygon is the affine image of a regular convex N-gon whose all vertices lie on an ellipse. (as pointed out by Nikhil)
- A necessary and sufficient condition for to have a convex midpoint polygon (after some finite iterations of the midpoint polygon procedure) is that the ellipse circumscribing the sum of two convex components of is nondegenerate. (The degenerate form of an ellipse is a point. )
For a nice outline of the proof, please refer to the comment by uncombed_coconut on previous post.
Since I didn’t know that this is a well studied problem (and that too by a well known mathematician!) I was trying to prove it on my own. Though I didn’t make much progress, but I discovered some interesting theorems which I will share in my future posts.