We consider the (minimal) distance
function of a point in the plane to a set
P of
N points in the plane. The
locus of non-differentiability of
this distance function consists (besides of the points of
P) exactly of the
Voronoi diagram of
P. We show that the number of
minima (
m), maxima (
M) and "saddle points" (
s) of the distance function satisfy:
m -
s + M = 1
This is similar to the Morse type of statements for differentiable
functions. The saddle points occur exactly where a Delaunay edge cuts
the corresponding Voronoi edge in its interior. The set of those edges
form a subgraph of the
Delaunay graph, which connects all minima and saddle points. This graph
devides the plane into regions. In each of
the compact regions, there is exactly one maximum, the non compact
regions don't contain a local maximum.
At the end we classify all those graphs if
P contains of 3 or 4 points.
Summary
NB: Use of some terms is from FFL (e.g., generators).
Introduction
One can study either the distance,
d(X)=min{d_i(X)},
from point generators,
Pi, or
the distance squared functions,
D(X)=min{d_i^2(X)}:
they share similar level curves, maxima, minima and saddles.
D however gives nicer formula for
the gradient (and differentiability at the generator loci).
The restriction of
D to the
interior of Voronoi cells is differentiable, and the loci where
D is not differentiable is
precisely the Voronoi Diagram (
VD(P)).
Level curves of
d can be
considered as wavefronts,
{d = u},
which bounds sets
{d x u},
and which start from the generators
P.
It is the
change of topology of these
sets {d x u} which is
studied here.
Consider, as a 1st example, three point generators (in a general
configuration; e.g., not co-linear) in the plane:
- Initially, each generator is surrounded by a circular wavefront,
and hence, the Euler characteristic is Euler = 3.
- Next two wavefronts meet at a point and we get 2 contractible
sets; hence Euler = 2.
- Later the 3rd wavefront meets the newly formed (combined) set and
we get only one resulting set, hence Euler
= 1.
- For case A (obtuse triangular configuration of
generators), the final region gets larger and larger, and Euler = 1 throughout (Figure 1).
- For case B (acute triangular configuration of
generators), all angles are sharp, and the wavefronts meet one more
time enclosing a region in the middle. The set {d x u} is "circular" and no longer
contractible, such that Euler = 0
(Figure 2). As the wavefronts keep propagating, the enclosed region
vanishes, and Euler = 1,
there is only one set left, which is contractible, and there are no
more topological changes if u
keeps increasing.
NB: One need more refined information than just the Voronoi Diagram to
understand the topology dynamics of wavefronts.
Behaviour of D on the
plane A^2
Consider a point
A on the
interior of a Voronoi edge
eij:
- If A is different from
the center of the line PiPj
(midpoint Qij between the
pair of generators), then there is no change in the topology of the
lower level sets {d x u} since
the level curves of D are
homeomorphic to parallel lines, and thus A is a topological regular point of D (Figure 4).
- If A coincides with Qij,
then the local map corresponds to a differentiable saddle point, and A is called the topological saddle point of D (Figure 5).
We are now left with
A being
a
vertex of the Voronoi
diagram:
- If A is a maximum of
each edge intersecting at the vertex locus, then A is in fact also a local maximum
of D (Figure 7) ---
necessarilly an interior (circumcenter) point to the triangle with
three generators as vertices.
- If A is a maximum on
all but one of the adjacent edges, then, locally, D is equivalent to a linear
function and A is again a
topological regular point of D
(Figure 8) --- necessarilly an exterior (circumcenter) point to the
triangle with three generators as vertices.
- There is no other (possible) case to consider.
Main Theorem
- Theorem: Let m, s, M be the number of
topological minima, saddle points and maxima of the distance function D, then:
m - s + M = 1
- Definition (index of critical
pts.): m, s, M are
called critical points of index 0,
1, 2. with corresponding (distance) value called critical value.
- Definition (regular pts):
All other points of D are
called (topologically) regular;
they come in 3 flavors:
- Differentiable regular points in the interior of Voronoi cells.
- Topological regular points in the interior of certain Voronoi
edges.
- Topological regular points which are multiple points.
Proof:
The proof involves defining a
vector
field which flows continuously everywhere except at the critical
points. A natural candidate is to use the gradient of
D. Special care is only needed near
the Voronoi diagram where
D
is not differentiable.
- Build a vector field on the interior
of Voronoi cells by considering the gradient direction of a
differentiable function f (= D):
- Near Voronoi edges eij
but away from saddles and Voronoi vertices (multiple pts), define w to be parallel to an edge eij (Figure 12).
Figure 12. Near regular points on a
Voronoi edge.
- Near multiple pts (Voronoi vertices) which are regular, define w
to be parallel to the outgoing Voronoi edge (Figure 13).
Figure 13. Near a Voronoi vertex which is topologically regular.
- Near each generator, local minimum of D, the topology is that of a disk
for a small enough distance encircling a 0-cell.
- In a small neighborhood near a saddle point, A, a 1-cell (interval Z1Z2) gets attached (Figure 15).
- In a small neighborhood surrounding a local maxima, A, a 2-cell (filled disc) gets attached
(Figure 16).
Graphs
The Delaunay graph related to the set of points P
The vertices of the Delaunay graph are the generators themselves. The
graph contain an edge connecting a pair of generators iff their Voronoi
cells share an edge.
The sm-graph related to
the set of points P
The vertices of the
sm-graph
are the generators themselves, local minima of
D. The edges are those line
segments
PiPj which intersect
a Voronoi edge
eij in its
interior, which are 1-to-1 correspondance with the saddles,
s, of
D, via the intersection
(mid-)points
Qij. This graph
is called the saddle-minima-graph of
P
or
sm(P) for short; it
coincides with the
Gabriel graph
of
P.
- sm(P) is a sub-graph of
the Delaunay graph.
- sm(P) is connected.
- sm(p) coincides with the
Gabriel graph of P (or is included in it if right triangular
configuration are considered or not as part of the Gabriel set).
- Proposition: The number
of bounded regions defined by sm(P)
is equal to M, the number of
maxima of D, such that for
each region there is exactly one maximum in its interior.
Proof: Apply Euler's formula
for a graph:
V - E + R = 1,
where the number of vertices
V = m,
the number of generators (minima of
D),
the number of edges
E = s,
the number of saddles of
D,
and
R is the number of
bounded regions of the graph. Then
R
= 1 - m + s = M, due to the main Theorem above.
Enriched Voronoi Diagram
The Voronoi Diagram can be considered as a planar graph with extra
infinite edges --- but one can compactify this situation by adding one
(virtual) generator at "infinity" and get in that way a graph on the
sphere. By adding as nodes of the graph the saddle points, one can also
add arrows to the graph in the directions of increasing distance
values. The resulting directed graph, the "enriched Voronoi Diagram,"
contains all the topological information studied here (Figure 20).
Figure 20. The enriched Voronoi Diagram.
Configurations with 3 or 4 generators
The Delaunay edges which are not included in
sm(P) are dashed in Figure 21.
Figure 21. Generic configurations with 3 or 4 generators.
The notation is defined as
(m, s, M).
Remarks and Questions
Generalizations to other metrics and higher dimensions.
Johnson-Mehl models
Weighted distance functions.
Convex sites
Piece-wise linear distance functions
Generalization to higher dimensions
NB: For the existence of critical points of the function
D it seems to be important if the
Delaunay face "cuts" its dual Voronoi face in its interior, in a
boundary point or not.