I.

4 Planar Graphs 15

I.4 Planar Graphs

Although we commonly draw a graph in the plane, using tiny circles for the
vertices and curves for the edges, a graph is a perfectly abstract concept. We
now talk about constraints necessary to draw a graph in the plane without
crossings.

Embeddings. Let G = (V, E) be a simple undirected graph. A drawing maps

every vertex u ∈ V to a point ε(u) in R2 , and it maps every edge uv ∈ E to a
path with endpoints ε(u) and ε(v). The drawing is an embedding if the points
are distinct, the paths are simple and do not cross each other, and incidences
are limited to endpoints. Not every graph can be drawn without crossings.
The graph G is planar if it has an embedding in the plane. As illustrated in
Figure I.13 for the complete graph of four vertices, there are many drawings
of a planar graph, some with and some without crossings. Note that a graph

Figure I.13: Three drawings of K4 . From left to right: a drawing that is not an
embedding, and embedding with one curved edge, and a straight-line embedding.

has an embedding in the plane if and only if it has one on the 2-sphere. The
latter is sometimes preferred because there is no outside region that has to be
treated differently from the other regions, which are finite.

Euler’s formula. A face of an embedding ε of G is a component in the

decomposition of the plane defined by ε. We write v = card V , e = card E,
and f for the number of faces. Euler’s formula is a linear relation between the
three numbers.

Euler Relation. Every embedding of a connected graph in the plane sat-

isfies v − e + f = 2.

Proof. Choose a spanning tree (V, T ) of (V, E). It has v vertices, card T = v−1
edges, and one face. We have v − (v − 1) + 1 = 2, which proves the formula
16 I Graphs

if G is a tree. Otherwise, draw the remaining edges, one at a time. Each

edge decomposes one face into two. The number of vertices does not change, e
increases by one, and f increases by one. Since the graph satisfies the claimed
relation before drawing the edge it satisfies the relation also after drawing the
edge.
If the graph consists of c ≥ 1 connected components we have v −e+f = c+1.
Note that the Euler Relation implies that the number of faces is the same
for all embeddings and is therefore a property of the graph. We get bounds
on the number of edges and faces, in terms of the number of vertices, by
considering maximally connected graphs for which adding any one edge would
violate planarity. Every face of a maximally connected planar graph with three
or more vertices is necessarily a triangle, for if there is a face with more than
three edges we can add a path that crosses none of the earlier paths. Let v ≥ 3
be the number of vertices, as before. Since every face has three edges and every
edge belong to two triangles, we have 3f = 2e. We use this relation to rewrite
3f
the Euler Relation: v − e + 2e
3 = 2 and v − 2 + f = 2 and hence e = 3v − 6 and
f = 2v − 4. Every planar graph can be completed to a maximally connected
planar graph, which implies that it has at most these numbers of edges and
faces.
Note
P that the sum of vertex degrees is twice the number of edges, and there-
fore u deg u < 6n. It follows that every planar graph has a vertex of de-
gree less than six. This observation suggests inductive approaches to various
questions about planar graphs, such as coloring the vertices and constructing
straight-line embeddings.

Non-planarity. We can use the Euler Relation to prove that the complete
graph of five vertices and the complete bipartite graph of three plus three
vertices are not planar. Consider first K5 , which is drawn in Figure I.14, left.
It has v = 5 vertices and e = 10 edges, contradicting the upper bound of

Figure I.14: K5 on the left and K3,3 on the right, each drawn with the unavoidable
one crossing.
I.4 Planar Graphs 17

at most 3v − 6 = 9 edges for maximally connected planar graphs. Consider

second K3,3 , which is drawn in Figure I.14, right. It has v = 6 vertices and
e = 9 edges. Each cycle has even length, which implies that each face has four
or more edges. We get 4f ≤ 2e and e ≤ 2v −4 = 8 after plugging the inequality
into the Euler Relation, again a contradiction.
In a sense, K5 and K3,3 are the quintessential non-planar graphs. Two
graphs are homeomorphic if one can be obtained from the other by a sequence
of operations, each deleting a degree-2 vertex and merging their two edges into
one or doing the inverse.

Kuratowski Theorem. A simple graph is planar iff no subgraph is home-

omorphic to K5 or to K3,3 .

The proof of this result is omitted. The remainder of this section focuses on
straight-line embeddings of planar graphs.

Convex combinations. Two points a0 6= a1 define a unique line that passes

through both. Each point on this line can be written as x = (1 − λ)a0 + λa1 ,
for some λ ∈ R. For λ = 0 we have x = a0 and for λ = 1 we have x = a1 .
The point belongs to the line segment connecting a0 to a1 iff 0 ≤ λ ≤ 1. If
we have more that two points we repeat the construction by adding all points
y = (1 − λ)x + λa2 for which 0 ≤ λ ≤ 1, and so on, as illustrated in Figure I.15.
Given k + 1 points a0 , a1 , . . . , ak , we can do the same construction in one step,

Figure I.15: From left to right: the construction of the convex hull of five points by
adding one point at a time.
Pk
calling a point x = i=0 λi ai a convex combination of the ai if λi ≥ 0 for all
0 ≤ i ≤ k and ki=0 λi = 1. The set of convex combinations is the convex hull
P
of the ai .
Let now K be a triangulation of a disk. In other words, the edge-skeleton
consisting of the vertices and edges of K is a planar graph such that each
interior face is bounded by three edges and their union is homeomorphic to
18 I Graphs

B2 . Letting V be the set of vertices, we call f : V → R a (strictly) convex

combination function if for each interior vertex u ∈ V there are real numbers
λuv > 0 satisfying
X
λuv = 1; (I.1)
v
X
λuv f (v) = f (u), (I.2)
v

where both sums are over all neighbors v of u. Similarly, we call ε : V → R2 a

(strictly) convex combination mapping if there are strictly positive real numbers
λuv such that (I.1) and (I.2) hold for ε.

Straight-line embedding. It is not difficult to show that every straight-line

embedding of the edge-skeleton of K defines a convex combination mapping.
We will show that the reverse is also true provided the boundary vertices form
a strictly convex polygon.

Tutte’s Theorem. If ε : V → R2 is a convex combination mapping that

maps the boundary vertices to a strictly convex polygon then drawing the
straight edges between the images of the vertices is a straight-line embedding
of the edge-skeleton of K.

We will give the proof in three steps, which we now prepare with a few ob-
servations. A non-zero vector p ∈ R2 and a real number c define a function
f (x) = hx, pi + c, which is positive on one side of the line f −1 (0) and negative
on the other. Suppose ε is a convex combination
P mapping. Then there are
positive real numbers λuv such that ε(u) = v λuv ε(v) for each interior vertex
u. We therefore have
X X
f (ε(u)) = h λuv ε(v), pi + λuv c
v v
X
= λuv f (ε(v)).
v

In words, f is a convex combination function.

A separating edge of K is an interior edge that connects two boundary ver-
tices. It is convenient to assume there are no separating edges, and if there is
one we can split K into two and do the argument for each piece. Call a path
interior if all its points are interior except possibly its two endpoints. Under
the assumption of no separating edges, every interior vertex u can be connected
I.4 Planar Graphs 19

to every boundary vertex by an interior path. Indeed, since K is finite, we can

find an interior path that connects u to a first boundary vertex w. Let w0 and
w1 be the neighboring boundary vertices. Since none of the edges separate,
the neighbors of w form a unique interior path connecting w0 to w1 . It follows
that there is an interior path connecting u to w0 . By repeating the argument
substituting w0 for w we eventually see that u has an interior path to every
boundary vertex. This can be used to prove the following result.

Maximum Principle. Let K be a triangulation of B2 without separating

edges and f : V → R a convex combination function. If f (u) ≥ f (w) for
an interior vertex u and every boundary vertex w then f (u) = f (v) for every
vertex v ∈ V .

To see this, let u0 be the interior vertex that maximizes f . By assumption,

f (u0 ) ≥ f (v) for all v ∈ V . Since f is a convex combination function, all
neighbors of u0 have the same function value. Repeating this argument for the
neighbors of u0 and their neighbors and so on implies the Maximum Principle.

Proof of Tutte’s Theorem. We now present the proof in three steps. First,
all interior vertices u of V map to the interior of the strictly convex polygon
whose corners are the images of the boundary vertices. To see this, choose
p ∈ R2 and c ∈ R such that the line f −1 (0) defined by f (x) = hx, pi + c
passes through a boundary edge and f (ε(w)) > 0 for all boundary vertices
other than the endpoints of that edge. Then f (ε(u)) > 0 else the Maximum
Principle would imply f (ε(v)) = 0 for all vertices. Repeating this argument
for all edges of the convex polygon implies that all interior vertices u have ε(u)
in the interior of the polygon. This implies in particular that each triangle
incident to a boundary edge is non-degenerate, that is, its three vertices are
not collinear.
Second, letting yuv and zuv be the two triangles sharing the interior edge
uv in K, the points ε(y) and ε(z) lie on opposite sides of the line f −1 (0) that
passes through ε(u) and ε(v). To see this assume f (ε(y)) > 0 and find a strictly
rising path connecting y to the boundary. It exists because f (ε(y)) > f (ε(u))
so one of the neighbors of y has strictly larger function value, and the same
is true for the next vertex on the path and so on. Similarly, find a strictly
falling path connecting u to the boundary and the same for v, as illustrated in
Figure I.16. The rising path does not cross the falling paths, but the two falling
paths may share a vertex, as in Figure I.16. In either case, we get a piece of
the triangulation bounded by vertices with non-positive function values. Other
than u and v all other vertices in this boundary have strictly negative function
20 I Graphs

y
z v

Figure I.16: One strictly rising and two strictly falling paths connecting y, u, and v
to the boundary.

values. If z belongs to the boundary of this piece it has strictly negative

function value simply because it differs from u and v. Else it belongs to the
interior of the piece and we have f (ε(z)) < 0 by the Maximum Principle. We
note that this argument uses f (ε(y)) > 0 in an essential manner. To show
that this assumption is justified, we connect yuv by a sequence of triangles
to one incident to a boundary edge. In this sequence, any two contiguous
triangles share an edge. As observed in Step 1, the image of the last triangle
is non-degenerate. Going backward this implies that the image of the second
to the last triangle is non-degenerate and so on. Finally, the image of yuv is
non-degenerate, as required.
Third, the images of any two triangles in K have disjoint interiors. To get a
contradiction assume x is a point in the common interiors of two such images,
σ and τ . Choose a half-line that emanates from x and avoids all images of
vertices. It defines a sequence of triangles starting with σ and ending at the
triangle υ incident to the boundary edge whose image crosses the half-line.
Similarly, the half-lines defines another sequence of triangles starting with τ
and ending with the same triangle υ. Going back from υ we pass from one
quadrangle to the next. Each step is unambiguous which implies σ = τ .

Bibliographic notes. Graphs that can be drawn in the plane without cross-
ings arise in a number of applications, including geometric modeling, geographic
information systems, and others. We refer to [4] for a collection of mathemat-
I.4 Planar Graphs 21

ical and algorithmic results specific to planar graphs. The fact that planar
graphs have straight-line embeddings has been known long before Tutte’s The-
orem. Early last century, Steinitz showed that every 3-connected planar graph
is the edge-skeleton of a convex polytope in R3 [5]. This skeleton can be pro-
jected to R2 to give a straight-line embedding. In the 1930s, Koebe proved that
every planar graph is the intersection graph of a collection of possibly touching
but not otherwise overlapping closed disks in R2 [3]. We get a straight-line
embedding by connecting the centers of touching disks. This result has been
rediscovered by Andreev and independently by Thurston and is known under
all three names. Probably the simplest proof that every planar graph has a
straight-line embedding uses an inductive argument adding a vertex of degree
at most five at each step [1]. This leads to huge differences in lengths and turns
out to be less useful in practice for this reason. The original theorem by Tutte
is for coefficients λuv equal to one over the degree of u [6]. The more general
version and the proof presented in this section are taken from the more recent
papery by Floater [2]. The theorem can be turned into a linear system with
equally many equations as unknowns. By the Euler Relation this system is
sparse and thus permits efficient solutions.

[1] I. Fáry. On straight line representation of planar graphs. Acta Univ. Szeged.
Sect. Sci. Math. 11 (1948), 229–233.

[2] M. S. Floater. One-to-one piecewise linear mappings over triangulations. Math.

Comput. 72 (2003), 685–696.

[3] P. Koebe. Kontaktprobleme der konformen Abbildung. Ber. Sächs. Akad. Wiss.
Leipzig, Math.-Phys. Kl. 88 (1936), 141–164.

[4] T. Nishizeki and N. Chiba. Planar Graphs: Theory and Algorithms. North-
Holland, Amsterdam, the Netherlands, 1988.

[5] E. Steinitz. Polyeder und Raumeinteilung. In Enzykl. Math. Wiss., Part 3AB12
(1922), 1–139.

[6] W. T. Tutte. How to draw a graph. Proc. London Math. Soc. 13 (1963), 743–
768.