Planar Graphs (Draft)

(Nov 27, 2015)

Definition of Planar Graphs

In class, we often draw graphs on the whiteboard, representing vertices by circles with labels inside, and
representing edges as lines connecting these circles. For large graphs, these pictures could get messy, especially
if a graph has a lot of intersecting edges. So we would like to draw graphs in such a way that the number of
edge intersections is minimized. In particular, we would like to avoid edge intersections altogether.
This is not just a question of aesthetics; in some applications minimizing the number of intersections is
critical. One such application is circuit routing, where we have gates (vertices) connected with wires (edges).
Intersecting wires need to be routed in different layers, increasing cost. Reducing or eliminating intersections
could significantly reduce this cost.
This motivates the concept of planar graphs, which is probably the most studied class of graphs in mathematics. A graph G = (V, E) is called planar if it can be drawn in the plane such that
Vertices are represented by points,
Each edge (u, v) is represented by a continuous line connecting the points that represent u and v,
Lines representing the edges do not intersect, except possibly for touching at the endpoints.
This definition is slightly informal, as we do not specify what exactly a continuous line is, what we mean
by intersection, etc, but all these concepts should be intuitively obvious. (A fully formal definition requires
the usage of some concepts from topology, and well skip it here.) Also, instead of points, we can represent
vertices by circles, and we will use labels inside to indicate which vertex is represented by the circle. (We
could also use other geometric objects, like squares.)
Example.

Consider the graph G below. Is G planar?

d

c
e

The way we draw it, it does not seem planar. But remember, the definition of planarity does not say that
each drawing of G must be planar; it only says that there must exist a planar drawing of G. So perhaps there
is some way to redraw G to make it planar? Indeed, here it is:

a
f

d
g
b
e
c
We can thus conclude that G is planar.

Characterization of Planar Graphs

Are all graphs planar? The intuition is that no. Let Kn be the complete graph with n vertices, that is a graph
where each pair of edges is connected by an edge. Kn thus has n(n 1)/2 = (n2 ) edges. If n is large, it
seems impossible to draw such Kn without any intersections there are just too many edges to draw.
If not all graphs are planar, then how difficult it is to tell whether a given graph G is planar or not? A
nave way of doing this would be to try all possible ways to draw G, but there are of course infinitely many
choices for the locations of Gs vertices and infinitely many ways to draw edges connecting them. It can be
shown that one can restrict this only to finitely many choices, but there will be so many that even then this
approach will be completely impractical.
Clique K5 .

Lets start by examining K5 , the clique with 5 vertices:

z
u

Is K5 planar? If you try to draw K5 without edge intersections, moving the vertices and modifying the
edges, you will see that you can avoid most of the intersections, but no matter how you try, there will be
always at least one left. So perhaps K5 is not planar.
Indeed, we will show that K5 is not planar. The proof is basically by the analysis of cases, but we will
make it short by taking advantage of K5 s symmetry (which means that we can permute the labels of K5 in
an arbitrary way, without changing the graph).
So here is the proof. Suppose, towards contradiction, that G is planar. Let C be the cycle x, y, z, u, v, x.
Then C must be drawn as a closed curve in the plane:

z
u

This curve contains 5 edges of K5 , so there are 5 other edges left. Then either the interior of C will have
3 or more edges or the exterior of C will have 3 or more edges. By symmetry, lets assume that there are 3
edges in the interior of C. At least two of these edges must share an endpoint. (Otherwise their endpoints
would give us 6 vertices, and we only have 5). By symmetry, we can assume that this endpoint is x. Then the
edges in the interior sharing x must be exactly (x, z) and (x, u), so now the picture looks like this:

z
u

But we have at least one more edge in the interior, and in that picture it is impossible to add any more
edges in the interior. We thus get a contradiction, concluding that K5 is not planar.
Bipartite clique K3,3 .
partition, denoted K3,3 :

The next graph we examine is the complete bipartite graph with 3 vertices in each

Same question, is K3,3 planar? It turns out that, using a similar argument to the one for K5 , it is possible
to prove that K3,3 is not planar.
Of course, for any graph G, if G is planar then each subgraph H of G is also planar, since we can take any
planar drawing of G, remove edges and vertices that are not in H, and this will give us a planar drawing for
H. The two observations above thus give us the following:
Observation 1 If a graph G contains a subgraph isomorphic to K5 or K3,3 then G is not planar.

This observation gives us a necessary condition for planarity, but is it also sufficient? To answer this
question, we look at some other graphs.
Subdivisions.

Consider a graph G obtained from K5 by splitting an edge into two edges with a vertex:

z
u

Is G planar? It cannot be, because if we can draw G without edge intersections, then we can use this
drawing to obtain a planar drawing for K5 : simply remove t and connect x to y by following the curve from
x to t and then to y.
This operation, namely splitting edges into two, is referred to as subdividing. Formally, we will say that a
graph H is a sub-division of G if H can be obtained from G by replacing some edges in G by paths whose all
interior vertices are new vertices (that is, not in G) of degree 2. By the above paragraph, G is planar if and
only if H is planar. This gives us a somewhat stronger condition for planarity.
Observation 2 If a graph G contains a subgraph isomorphic to a sub-division of K5 or K3,3 then G is not
planar.
Now again we can ask the question: is this condition sufficient for planarity? Amazingly, the answer is yes,
as stated in the following theorem.
Theorem 1 (Kuratowski) A graph G is planar if and only if it does not contain a subgraph isomorphic to a
sub-division of K5 or K3,3 .
This is a truly profound statement: it basically says that K5 and K3,3 are the only two bad graphs, that
are obstacles for planarity. As long as a graph does not contain a subdivision of one of these two graphs, it
must be planar!
Kuratowskis theorem can be converted into a polynomial-time algorithm for planarity testing of graphs.
In essence, this algorithm tries all 5-element sets of vertices and verifies if they form a sub-division of K5 , and
then it tries all 6-element sets of vertices and verifies if they form a sub-division of K3,3 . (This is a bit more
complicated, because testing if given 5 vertices form a sub-division of K5 requires a bit of work, but it can




be done efficiently.) There are n5 = O(n5 ) choices of 5-element sets of vertices and n6 = O(n6 ) choices of
6-element sets of vertices. This thus gives a polynomial-time algorithm although very slow. With a much
more clever algorithm, Tarjan showed in the late 1960s that planarity testing can be done much much faster,
in time O(n).
Example. Kuratowskis theorem can be used to prove that some graphs are not planar. Given a graph G,
if we can identify a sub-division of K5 or K3,3 in G, it gives us a certificate that G is not planar. Consider,
for example, this graph G:

Graph G
a

g
c
b

G has a sub-division of K5 , as shown below

d
a

c
b

so G is not planar.

Density of Planar Graphs

A graph G with n vertices has between 0 and n2 = n(n 1)/2 edges. What can we say about the number
of edges in G if G is planar? The intuition is that a planar graph cannot be too dense. We know that, in
particular, it cannot contain K5 . In this section we will show that planar graphs are very sparse; a planar
graph with n vertices has at most 3n 6 edges (as long as n 3).
Trees. Before discussing the density of planar graphs, we need to examine graphs that are called trees. A
tree is defined as a graph that is connected and acyclic (which means that it does not have any cycles). Here
is an example of a tree:

2
1
3

7
4

10

11

Consider a graph G that is a tree with n vertices. You can build G by starting with a single vertex (with no
edges) and repeatedly adding to it leaves (vertices of degree 1) and connecting them to the already constructed
tree. At each step we will add one vertex and one edge, so at the end we will have m = n 1, where m is the
number of edges in G. We thus have:

Observation 3 A tree with n vertices has m = n 1 edges.

We can in fact extend this observation to a theorem that gives several different characterizations of trees.
Theorem 2 Let G be a graph with n vertices and m edges. The following conditions are equivalent:
1. G is a tree.
2. G is connected and m = n 1.
3. G is acyclic and m = n 1.
4. G is connected and removing any edge disconnects G.
5. G is acyclic and adding any edge creates a cycle in G.
Eulers identity. If we draw a planar graph G in a plane without any edge intersections, then the drawing
of G will partition the plane into connected regions. These regions are called the faces of G. Each face can
be identified by a tour in G that forms the boundary of this face. (Recall that a tour is a relaxed notion of a
cycle; in a tour a vertex may be visited multiple times, but each edge appears at most once.)
For example, in the planar drawing of a graph G shown below
a

A
D
F

d
g

B
c

we have several faces identified by capital letters A, B, C, D, F . Face A can be specified by the tour a, d, f, a.
Face B can be specified by the tour a, g, h, g, b, c, f, d, a. Note that node g appears twice on face B. Face C is
called the external face.
The theorem below, attributed to Euler, says that the number of faces in a connected graph is uniquely
determine by n and m (the numbers of vertices and edges). In other words, no matter how you draw a
connected planar graph, as long as the drawing has no edge intersections, the number of faces will be always
the same.
Theorem 3 Let G be a connected planar graph with n vertices and m edges. Then the number of faces in
any planar drawing of G is the same. Further, denoting this number by f , we have that
n m + f = 2.
Proof: (Sketch.) The proof is by induction. The base case is when G is a tree, in which case m = n 1
and f = 1, so the theorem holds. In the inductive step, we reduce G by removing any edge that belongs to
some cycle, and apply the inductive assumption to such obtained graph G0 . When we add this edge back,
we increase the number of edges by 1 and the number of faces by 1, so the formula in the theorem will be
preserved. 2

Eulers inequality. With Eulers identity in hand, we are now ready to prove a bound on the number of
edges in a planar graph. This bound is often referred to as the Eulers inequality.
Theorem 4 Let G be a planar graph with n vertices and m edges, where n 3. Then
m 3n 6.
Proof: We can assume that G is connected, for otherwise we could prove the above inequality for each
connected component of G separately, and by adding these inequalities we will get the same inequality for
the whole graph. Since G is connected, it satisfies Eulers identity, n m + f = 2. Computing f from this
identity, we get f = m + 2 n.
Our goal now is to find some relationship between f and m. To this end, we think of each edge as having
two sides, and we will count the number of sides in two different ways. First, the number of edge sides is of
course 2m. On the other hand, note that each face has at least three edge sides along its boundary and each
edge side belongs to exactly one face. So the number of edge sides is at least 3f . Thus we have
2m = number of edge sides 3f.
This gives us that f 2m/3. Substituting this into Eulers identity, we obtain that
m + 2 n 2m/3
By simple algebra, this equation gives us m 3n 6, which is exactly what we wanted to prove. 2
Application to proving non-planarity. Eulers inequality can be useful for proving non-planarity. For
example, for K5 , we have n = 5 and m = 10, and 10 6 3 5 6. Since K5 violates Eulers inequality, it cannot
be planar.
But we need to be careful: this works only one way, if a graph satisfies Eulers inequality, it does not mean
that its planar. For example, K3,3 satisfies Eulers inequality but it is not planar.