Delaunay Triangulation Splitter

By Alan Pierce

Add red and blue points to the area below, and then hit enter and follow the instructions to step through an algorithm that finds the Delaunay triangulation of just the red points and the Delaunay triangulation of just the blue points.

You need to have Java enabled to use this.

Algorithm Explanation

The basic idea of the algorithm is to use the known Delaunay triangulation to speed up the point location step. Since the point location is in a certain sense the hardest part of the incremental Delaunay triangulation algorithm, the speedup allows the algorithm to run in linear time. The algorithm is recursive, and proceeds as follows:

  1. Pick two points at random (they may or may not be the same color).
  2. From each chosen point, independently, run a search for the nearest neighbor of the same color. Stop when either of the two searches succeeds, and use that result. This search can be done by running a Dijkstra-like algorithm on the given Delaunay triangulation. It can be shown that this step takes expected constant time, in part because two points were chosen for the search instead of one. Once the search finishes, ignore the other point; we now have a single randomly chosen point and we have its nearest neighbor. Call this chosen point p and its nearest neighbor q.
  3. Remove p, and compute the Delaunay triangulation of the new set of red and blue points. This can be done in time O(deg(p)) time, which is expected O(1).
  4. Recursively split the remaining triangulation in linear time.
  5. We know that q is in the same triangulation where we want to insert p, because q and p are the same color. Search all triangles adjacent to q, and test to see if p is in each one. This takes time proportional to the degree of q, which can be shown to be constant in expectation, so this step takes expected constant time.
  6. Insert p into this triangle, and apply Delaunay flips until the triangulation is Delaunay, using the usual method of the incremental algorithm. This step also takes expected constant time.

The complete algorithm, and its analysis, are described in the paper "Splitting a Delaunay Triangulation in Linear Time", by B. Chazelle, O. Devillers, F. Hurtado, M. Mora, V. Sacristan, and M. Teillaud.