California State University, San Bernardino

CSUSB ScholarWorks

Electronic Theses, Projects, and Dissertations Office of Graduate Studies

12-2016

PLANAR GRAPHS, BIPLANAR GRAPHS AND GRAPH THICKNESS

Sean M. Hearon
California State University - San Bernardino

Follow this and additional works at: https://scholarworks.lib.csusb.edu/etd

Part of the Discrete Mathematics and Combinatorics Commons

Recommended Citation
Hearon, Sean M., "PLANAR GRAPHS, BIPLANAR GRAPHS AND GRAPH THICKNESS" (2016). Electronic
Theses, Projects, and Dissertations. 427.
https://scholarworks.lib.csusb.edu/etd/427

This Thesis is brought to you for free and open access by the Office of Graduate Studies at CSUSB ScholarWorks. It
has been accepted for inclusion in Electronic Theses, Projects, and Dissertations by an authorized administrator of
CSUSB ScholarWorks. For more information, please contact [email protected].
Planar Graphs, Biplanar Graphs and Graph Thickness

A Thesis

Presented to the

Faculty of

California State University,

San Bernardino

In Partial Fulfillment

of the Requirements for the Degree

Master of Arts

in

Mathematics

by

Sean Michael Hearon

December 2016
 Planar Graphs, Biplanar Graphs and Graph Thickness

A Thesis

Presented to the

Faculty of

California State University,

San Bernardino

by

Sean Michael Hearon

December 2016

Approved by:

Dr. Jeremy Aikin, Committee Chair Date

Dr. Cory Johnson, Committee Member

Dr. Rolland Trapp, Committee Member

Dr. Charles Stanton, Chair, Dr. Cory Dunn

Department of Mathematics Graduate Coordinator,
Department of Mathematics
 iii

Abstract

A graph is planar if it can be drawn on a piece of paper such that no two edges
cross. The smallest complete and complete bipartite graphs that are not planar are K5
and K3,3 . A biplanar graph is a graph whose edges can be colored using red and blue such
that the red edges induce a planar subgraph and the blue edges induce a planar subgraph.
In this thesis, we determine the smallest complete and complete bipartite graphs that are
not biplanar.
 iv

Acknowledgements

I would like to thank a few people for making this thesis possible. First, my
loving wife Jessica. Without your support and motivation I don’t think I would of com-
pleted this task. Second, to my parents for supporting me and allowing me a place to live
during my entire masters and thesis process. Finally, Dr. Aikin for all his time, effort
and dedication to helping me through this entire process. Without your encouragement,
patience and guidance I would not have nearly as polished of a thesis. It was your support
that has granted me this opportunity to grow as a Mathematician.
 v

Table of Contents

Abstract iii

Acknowledgements iv

List of Figures vi

1 Introduction 1

2 Graphs and Planar Graphs 5

2.1 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Planar Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Biplanar Graphs 19
3.1 Complete Biplanar Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Complete Bipartite Biplanar Graphs . . . . . . . . . . . . . . . . . . . . . 33

4 Graph Thickness 37

5 Conclusion 39

Bibliography 40
 vi

List of Figures

1.1 An example of a transportation network. . . . . . . . . . . . . . . . . . . 1

1.2 Two models of the same circuit. . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Biplanar model of the electrical circuit in Figure 1.2. . . . . . . . . . . . . 3

2.1 The bridges of Königsberg [CLZ10]. . . . . . . . . . . . . . . . . . . . . . 5

2.2 A graph representaion of Königsberg. . . . . . . . . . . . . . . . . . . . . 6
2.3 A simple graph. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 A picture of the subgraph H. . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Structure of Case 1 of Kuratowski’s Theorem. . . . . . . . . . . . . . . . . 17
2.6 Structure of Case 2 of Kuratowski’s Theorem. . . . . . . . . . . . . . . . . 17
2.7 Structure of Case 3 of Kuratowski’s Theorem. . . . . . . . . . . . . . . . . 18
2.8 Structure of Case 4 of Kuratowski’s Theorem. . . . . . . . . . . . . . . . . 18

3.1 A biplanar decomposition of K8 . . . . . . . . . . . . . . . . . . . . . . . . 20

3.2 Structure of G in Case 3.1. . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Core structure of Case 1.3. . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Structure of Case 1.3 with path P. . . . . . . . . . . . . . . . . . . . . . . 24
3.5 Structure of Case 1.5 with path P. . . . . . . . . . . . . . . . . . . . . . . 26
3.6 Structure of Subcase 2.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.7 Structure of Case 2.1.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.8 Structure of Case 2.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.9 Structure of Case 2.2.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.10 Structure of Case 2.2.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.11 Structure of degree five cycle. . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.12 Structure of degree sequence Case 1. . . . . . . . . . . . . . . . . . . . . . 31
3.13 Structure of degree sequence Case 2. . . . . . . . . . . . . . . . . . . . . . 32
3.14 End structure of degree sequence Case 2. . . . . . . . . . . . . . . . . . . 32
3.15 Structure of degree sequence Case 3. . . . . . . . . . . . . . . . . . . . . . 33
3.16 A biplanar decomposition of K9 − {2, 4}. . . . . . . . . . . . . . . . . . . 34
3.17 A biplanar decomposition of K6,8 . . . . . . . . . . . . . . . . . . . . . . . 36
 1

Chapter 1

Introduction

Within the branch of combinatorics exists a field of mathematics called graph

theory. Graphs can be used to model many everyday circumstances, such as transporta-
tion networks and electrical circuits. Consider a map of a given region of the world. There
exist man made cities with transportation pathways such as roads, freeways, or railways
between the cities. Through graph theory, such a map can be modeled in a very simple
way. A city could be represented by a circle, called a vertex, with lines, called edges,
drawn between the circles to represent transportation pathways between different cities.
Consider Figure 1.1.

Figure 1.1: An example of a transportation network.

Note that not every city is directly connected by a road to every other city, such
as Cities 1 and 2. Thus to travel from one city to another, a person may need to take a
path through multiple cities to get to their destination. In order to reach the destination
 2

of City 2 from City 1, a person would need to travel to City 3 first then to City 2, or
travel from City 1 to City 3 to City 4 to City 2. Another interesting observation is if
transportation through City 3 was blocked, then City 1 would be, in a sense, isolated from
Cities 2 and 4. In this way, graphs can be used to encode important information from
a transportation network, giving us a relatively simplified perspective of an otherwise
complicated system.
Graphs can also be used to model electrical circuits in a computer or even the
electrical circuit within a single computer chip. In these cases we would model separate
electrical components as circles with lines drawn between them to represent electrical
pathways. This is a very similar model to that of the model of transportation networks.
Consider Figure 1.2.

Figure 1.2: Two models of the same circuit.

Although both models are equally valid and encode the same information about
components and electrical pathways, there exists a significant difference in structure. In
the model on the left in Figure 1.2, the electrical pathway connecting Components 1
and 4 crosses the electrical pathway connecting Components 2 and 3. In contrast, the
model on the right in Figure 1.2 does not contain any pathways that cross. This is a
key structural difference that was not important within the transportation network. On
a circuit board,if electrical pathways cross the communication between the components
will not function properly. With this in mind, the graph on the right in Figure 1.2
 3

would provide a better model for an electrical circuit. A graph that can be drawn on
a piece of paper without having any two lines cross is called a planar graph. Modeling
electrical circuits as planar graphs provides our motivation for studying planar graphs.
In particular, we are interested in being able to determine precisely when a given graph
is planar. For, if it is, such a graph could potentially serve as a model for an electrical
circuit embedded on a circuit board.
Intuitively, an example may exist in which the number of electrical pathways is
to great in comparison to the number of electrical components preventing us from placing
the circuit on a single circuit board. We might view such a circuit as one that requires
the use of the front and back sides of a circuit board to place the electrical pathways in
a way that no two pathways cross. Modeling these more complex circuits would require
graphs that can be decomposed in some way to two planar graphs. We call a graph
biplanar when such a decomposition is possible. Hence, biplanar graphs model electric
circuits which require the use of at most two sides of a circuit board. Figure 1 provides
an example of a biplanar model of the planar model in Figure 1.2.

Figure 1.3: Biplanar model of the electrical circuit in Figure 1.2.

As electrical circuits become more complex there would become a point at which
the graph model of the circuit would require even more than two sides of a circuit board.
An electrical circuit of significant complexity may require more than two layers in order
to connect all components in a prescribed way. The graph modeling such a circuit would
then require a decomposition consisting of more than two planar graphs. The number of
planar graphs required to model such a circuit is a parameter we will investigate in this
thesis and is called the thickness of the graph that models the circuit.
 4

Given a graph that serves as a model for an electrical circuit, determining the
thickness of a graph will tell us the minimum number of layers needed in a computer chip
in order to successfully build the circuit. Assuming more layers implies an increased cost
in production, we are in some sense minimizing the cost of microchip production. This
application is one factor that motivates the study of biplanar graphs.
 5

Chapter 2

Graphs and Planar Graphs

2.1 Graphs

The concept of a graph grew from a problem Leonhard Euler had developed
concerning the city of Königsberg.

Figure 2.1: The bridges of Königsberg [CLZ10].

The city of Königsberg, at the time when Euler contemplated it, was located
in Prussia. The city was divided into four sections with seven bridges as seen in Figure
2.1. Euler’s original problem asked the inhabitants of Königsberg whether or not it was
possible to walk around the city and cross each bridge exactly once. Euler was not
concerned with the starting and ending point to be the same. Although the problem
was quite simple, Euler needed to derive a new concept in order to generate a proof that
would be rigorous enough to withstand mathematical scrutiny.
Thus Euler developed the graph as a tool to solve his problem. A graph G is a
 6

finite nonempty set V of objects, called vertices together with a possibly empty set E of
2-element subsets of V called edges. Graphs are commonly viewed as drawings where the
vertices are points or circles and the edges as lines occurring between two vertices. Two
distinct vertices u and v are adjacent if the edge {u, v} is in contained in the edge set E
of G. Rather than denoting the edges as a two element subset of the vertex set, we will
shorten the notation to the edge uv for any two connected vertices u and v.

Example 2.1. Let G have the vertex set V = {v1 , v2 , v3 , v4 } and the edge set E =
{{v1 , v2 }, {v1 , v2 }, {v2 , v3 }, {v2 , v3 }, {v3 , v4 }, {v1 , v4 }, {v2 , v4 }}. The graph of G can be
drawn in the following way:

Figure 2.2: A graph representaion of Königsberg.

Euler constructed the graph in Figure 2.2 to model the bridges of Königsberg.
The vertices represent the land masses separated by the river Pregel, and the edges
represent the bridges. By developing some of the first theorems about graphs, Euler
was able to give a rigorous proof that no such walk around Königsberg was possible. In
particular, Euler proved that there was no such way to traverse the corresponding graph
model unless there were either an even number of edges at all vertices, or there were
exactly two vertices at which there were an odd number of edges.
Notation should be clarified at this point. Let G be a graph with edge set E
and vertex set V . We will use |E| = m to denote the number of elements in the edge set
E. Additionally, |V | = n will denote the number of elements in the vertex set V . Finally
we will denote the number of 2-element sets of E containing a vertex v an an element as
deg(v). A graph G0 = (V 0 , E 0 ) is a subgraph of a graph G = (V, E) if V 0 is a subset of V
 7

and E 0 is a subset of E. Additionally for any vertex v in V but not in V 0 , no 2-element

set of E 0 can contain v. A graph G0 = (V 0 , E 0 ) is a proper subgraph of a graph G = {V, E}
if V 0 is a proper subset of V and E 0 is a proper subset of E such that E 0 . Throughout
this thesis we will denote the subgraph of G = (V, E) obtained by removing a single edge
e as G − e. Additionally, a proper subgraph of G may be obtained by removing a vertex
v from V and removing all elements that include v from E. This is denoted by G − v.
A graph is simple if its edge set has no repeated elements and every element of
the edge set consists of two distinct vertices of the set V . Although there are other types
of graphs, it is important to note that for the remainder of the thesis.
Since the graphs we study here are simple there exists a maximum number of
edges a graph could have. A graph G is a complete graph if for any two distinct vertices
v, u ∈ V then the edge {v, u} is contained in the edge set E. A complete graph with n
distinct vertices is denoted Kn . Additionally, for a complete graph Kn , it is important
to note there are exactly n(n − 1)/2 edges and deg(v) = n − 1 for ever v ∈ V .
A graph G is complete bipartite if the vertex set V can be partitioned into two
parts V1 and V2 such that the edge {u, w} ∈ E if and only if u and w are in different parts.
Complete bipartite graphs are denoted by Ka,b where |V1 | = a and |V2 | = b. Additionally,
for a complete bipartite graph Ka,b has exactly a · b edges.
It is important to properly define complete and complete bipartite graphs be-
cause they play a vital roll throughout graph theory and especially within this thesis.

2.2 Planar Graphs

As mentioned in Chapter 1, it may be possible to draw a graph on a piece of

paper in such a way that the edges do not cross, although this is not always possible. A
graph G is planar if there exists a way to draw the graph without edges crossing. Such
a drawing is called a planar embedding of G.
A path in a graph G is a sequence of vertices (v1 , v2 , ..., vn ) such that vi and vi+1
are adjacent for 1 ≤ i ≤ n and each vertex may only be only appear once in the sequence.
Such a path starting at a vertex vi and terminating at a vertex vj is called a vi , vj -path.
A cycle in a graph G is a path that begins and ends at the same vertex.
A graph G is connected if between any two vertices u, v ∈ V , there exists a
u, v-path. A graph G has components if G is disconnected. That is, G is disconnected if
 8

there exist at least two proper subgraphs H1 and H2 of G such that for any v ∈ H1 and
u ∈ H2 then there does not exist a v, u-path.
Let G = (V, E) be a graph with κ components. The graph G contains a cut
vertex if there exists a vertex v such that the number of components in the proper
subgraph G − v increases. A graph G is n-connected if n vertices must be removed in
order to increase the number of components of G.
In order to better understand the introduced definitions, consider the following
example:

Figure 2.3: A simple graph. .

Between any two vertices i and j in Figure 2.3 there exists an i, j-path. For
example there exists a 1, 7-path by traveling along the edges {1, 3},{3, 4},{4, 7}. Thus,
this graph is connected and has exactly one component. Additionally Vertex 4 is a cut-
vertex in G since G − 4 results in components G1 with the vertex set V1 = {1, 2, 3} and
G2 with the vertex set V2 = {5, 6, 7} Since G contains a cut vertex, G is only 1-connected.
The simplest form of a planar graph is a tree or a forest. A graph G is a tree if
it does not contain any cycles. A graph G is a forest if each component of G is a tree.‘A
region of a graph is a section of a graph that is enclosed by a cycle. Additionally there
exists an unbounded region on the exterior of the graph. The vertices and edges that
create the cycle enclosing the region are said to be incident with the region.

Theorem 2.2 (Euler’s Polyhedral Formula). Let G be a connected graph with n vertices,
m edges, and r regions. The graph G obeys the equality n − m + r = 2.
 9

Proof. We proceed by induction on m. Let m = 0. Since G is a connected graph, n = 1

and since the only region of G that exists is the exterior region, then r = 1. We obtain
the equality 1 − 0 + 1 = 2 and so the formula holds for m = 0.
Let m = 1. The only edge must be between two vertices. Thus n = 2. Since G
can not contain any cycles, then the only region is the exterior region. Thus r = 1. We
obtain the equality 2 − 1 + 1 = 2 and so the formula holds for m = 1.
Now consider m ≥ 2. The graph G is either a tree or contains a cycle. If G is
a tree, then m = n − 1 and since the only region is the exterior region then r = 1. The
formula holds true since n − (n − 1) + 1 = 2. If G contains a cycle, then there exists
an edge e such that e is incident with two regions. Consider the subgraph G − e having
n0 vertices, m0 edges and r0 regions. The new graph of G − e, n0 = n, m0 = m − 1 and
r0 = r − 1. By the induction hypothesis, m0 ≤ m so the equality n0 − m0 + r0 = 2 holds
true. We obtain n − (m − 1) + (r − 1) = 2 implies n − m + r = 2 is true. That is, Euler’s
polyhedral formula holds for G, and so holds in general.

Euler’s Polyhedral Formula is the first important tool in which to study planar
graphs. Through this identity, an upper bound on the number of edges in a planar graph
G with n vertices can be found.

Theorem 2.3. If G is a planar graph with n ≥ 3 vertices and m edges then m ≤ 3n − 6.

Proof. Let G be a planar graph. Since a simple graph with three vertices cannot have
more than three edges the result holds when n = 3. Suppose n ≥ 4. Since G is a planar
graph than it obeys Euler’s Identity, having n vertices, m edges and f faces, n−m+f = 2.
Since every region of G is composed of at least 3 edges, and every edge is incident with
exactly two regions we see that inequality 3r ≤ 2m holds true. By multiplying both sides
of Euler’s Polyhedral Identity we see that 6 = 3n−3m+3r. Since we know that 3r ≤ 2m,
then 3n − 3m + 3r ≤ 3n − m. By combining the equations we obtain 6 ≤ 3n − m or
m ≤ 3n − 6.

A graph G is maximal planar if G is planar and the addition of an edge e to the

edge set E of G produces a nonplanar graph.
 10

Corollary 2.4. If G is a maximal planar graph with n ≥ 3 vertices and m edges, then
m = 3n − 6

We will now establish some lemmas for maximal planar graphs that we will
utilize later on in chapter 3.

Lemma 2.5. Let G be a maximal planar graph. Every region of G is of length three.

Proof. Let G be a maximal planar graph. Suppose the G had a region R of length greater
than three. Since R is bounded by a cycle, then there exists a path P of length three.
Let v1 , v2 and v3 be the vertices of P . At least two among v1 , v2 and v3 are nor adjacent
or else P is part of a cycle of length three. Without lost of generality let v1 and v3 not
be adjacent. Adding the edge {v1 , v3 } to the edge set of G, G would still be planar. But
this is a contradiction to the definition of G being maximal planar.

Lemma 2.6. Let G be a maximal planar graph of order n ≥ 4

Proof. Let G = (V, E) be a maximal planar graph. Suppose there exists a vertex v such
that deg(v) ≤ 2. Let deg(v) = 0 than clearly the edge {v, u} can be added to G for
any vertex u distinct from v and the resulting graph will still be planar. Hence G is not
maximal planar which is a contradiction. Suppose deg(v) = 1. Thus v is adjacent to
exactly one vertex, say u ∈ V . Then u must be adjacent to some other vertex say w.
Then the edge {v, w} can be added to G. since u and w are adjacent, adding the edge
{v, w} would create a u, w-path through v which is homeomorphic to the edge {u, w}
[GM12]. Therefore G is not maximal planar which is a contradiction.
Suppose deg(v) = 2. Let vi and vj be the vertices adjacent to v. At most vi and
vj are adjacent. There must exists a face containing v, vi and vj plus at least another
vertex. Then there exists some vertex vk adjacent to vi or vj . Then the edge vi , vk can
be added to E. Thus G is not maximal planar. Thus for a maximal planar graph, every
vertex must be of at least degree three.

Lemma 2.7. Let G be a maximal planar graph with a vertex v of degree three. Let v1 , v2
and v3 be the three vertices adjacent to v. Then v1 , v2 and v3 are pairwise adjacent.
 11

Proof. Let G be a graph with a vertex v of degree three. Suppose at least two of the
three vertices are not adjacent say v1 and v2 . Then the vertices v, v1 and v2 are all part
of the same region. Since G is maximal planar, by Lemma 2.5 every region is a cycle of
length three. Then v, v1 , and v2 are a cycle of length three. Thus v1 and v2 are adjacent
which is a contradiction. Therefore all three vertices adjacent to v are pairwise adjacent.

Theorem 2.8. K5 is not planar

Proof. Suppose that K5 is planar. Then K5 has exactly five vertices and ten edges. Then
K5 satisfies the inequality m ≤ 3n − 6. But 10 ≤ 9, which is false. Thus K5 is not planar.

We will now prove that K3,3 is not planar. However, K3,3 has six vertices and
nine edges which does not contradict Theorem 2.3. Hence we will need to employ a
different technique.

Theorem 2.9. K3,3 is not planar.

Proof. Suppose K3,3 is planar. We obtain the equality, by Euler’s Polyhedral Formula,
n − m + r = 2 where n = 6 and m = 9. It follows that r = 5 regions. Additionally, since
K3,3 contains no cycles of length three, then each region must be incident with at least
four edges. By summing the edges over each face, we count each edge twice and each
face at least four times. Hence, 4r ≤ 2m. Since r = 5 and m = 9, we obtain 20 ≤ 18, a
contradiction. Thus K3,3 is cannot be planar.

Since neither K5 nor K3,3 are planar, it is clear that any graph G that contains
either K5 or K3,3 as a subgraph, cannot be planar. Thus, we have a necessary but not
sufficient condition for a graph to be planar.
A graph H is a subdivision of a graph G if H is obtained from G by replacing
any of the edges of G by arbitrarily long paths. Two graphs H and H 0 are homeomor-
phic if there is some graph G from which H and H 0 are each obtained from a series of
subdivisions.

Theorem 2.10. Let G be a plane graph. Then for any region R having boundary B,
there exists an embedding such that B is on the boundary of the exterior region.
 12

Proof. Let G be a given plane graph with regions R1 , R2 , ...Rn . Then the boundary of
a region Ri is the collection of edges incident with the region. Let B1 , B2 , ..., Bn be the
boundaries of R1 , R2 , ...Rn , respectively. We construct a planar embedding of G having
B1 on the boundary of the exterior region. We may assume R1 is not already the exterior
region otherwise we are done. If G is not connected, we construct the desired embedding
working only in the component of G containing R1 . Thus we lose no generality in assuming
G is connected. We begin by embedding the boundary B1 in the plane. Next we identify
a region Ri with boundary Bi that is incident with at least one vertex of B1 . We then
embed the boundary Bi on the interior of boundary B1 . Next we identify a region Rj
with boundary Bj which is incident to at least one vertex of either R1 or Ri and embed
the boundary Bj on the interior of B1 .
If at some point in this process we are unable to embed any additional boundaries
on the interior of B1 , then either no remaining boundaries have a vertex in common with
any previously embedded boundaries or no matter how we try to embed another boundary
Bk on the interior of B1 , planarity is violated. In the first case, we have a collection of
boundaries that are disjoint from embedded boundaries. Then G must be disconnected,
which is a contradiction to our assumption that G is connected. In the second case, since
every edge of G that is not a bridge is incident with exactly two regions, all remaining
boundaries can only share edges with boundaries that are not already in common to
two embedded boundaries. Thus the set of edges that are contained in exactly one
embedded boundary bound a region R in the current partial embedding of G. It follows
that all remaining boundaries must be on the interior of R. Hence, any of the remaining
boundaries may be embedded without violating planarity.

From Theorem 2.11, it is easy to see that for any vertex v or edge e of a planar
graph G, there exists a planar embedding of G such that v or e, respectively, is may also
be positioned incident to the exterior region. This fact allows us to construct new planar
graphs from existing planar graphs. Consider two planar graphs G1 and G2 with given
planar embeddings. We construct a planar gluing by identifying either a vertex or edge
in each of G1 and G2 , making the edge or vertex incident with the exterior region in each
graph and taking the union of the planar embeddings of G1 and G2 by overlapping the
planar embeddings precisely at the vertex or edge. Through planar gluing we are able to
 13

construct larger planar graphs by gluing smaller planar graphs together.

Theorem 2.11 (Kuratowski’s Theorem). A graph G is planar if and only if G contains

no subgraph that is homeomorphic to K5 or K3,3

Proof. By Theorem 2.5 and Theorem 2.6, if G is a planar graph, then G cannot contain
a subgraph homeomorphic to K5 or K3,3 . Thus we will verify that if a graph contains no
subgraph homeomorphic to K5 or K3,3 , then G is planar.
To the contrary, suppose there exists a nonplanar graph that contains no sub-
graph homeomorphic to K5 or K3,3 . We may assume that G is an example of such a
graph that is minimal with respect to the cardinality of its edge set. We lose no gener-
ality in assuming G is connected, for if G is disconnected, our argument applies to each
connected component of G. If G is not 2-connected, then there exists a vertex v in G
whose deletion results in a disconnected graph. Let C1 be a component of G − v and let
C2 be the union of the remaining components of G − v. By the minimality of G, we see
that the subgraph of G obtained by deleting all the vertices of C1 is planar. Similarly, the
subgraph of G obtained by deleting all the vertices of C2 is planar. That is, the induced
subgraph G1 on V (C1 ) ∪ {v} is planar, as is the induced subgraph G2 on V (C2 ) ∪ {v}.
Thus, we can construct a planar glueing of G1 and G2 at the vertex v to produce G. As
G is nonplanar, this contradiction implies G must be 2-connected. We now show G is
3-connected.
Suppose G is not 3-connected. Then there exist two vertices v1 and v2 such that
G − {v1 , v2 } is a disconnected graph. Consider the partition {H1 , H2 } of the edge set of G
such that H1 is one of the components of G − {v1 , v2 } and H2 the union of the remaining
components of G − {v1 , v2 }.
Suppose that v1 and v2 are adjacent. Consider the partitions F1 = H1 ∪ {v1 , v2 }
and F2 = H2 ∪ {v1 , v2 }. By the minimality of G, each of F1 and F2 induce planar
subgraphs of G. Thus we can construct a planar gluing of F1 and F2 at v1 v2 producing
the graph G. Thus G is planar, a contradiction.
Now suppose v1 and v2 are not adjacent. Since G is nonplanar, the graph
G0 = G ∪ v1 v2 is also nonplanar. Moreover, since G − {v1 , v2 } is disconnected, so is
G0 − {v1 , v2 }. Let H1 be a connected component of G − {v1 , v2 } and let H2 be the
collection of the remaining components of G − {v1 , v2 }. Let F1 = (G − H2 ) + v1 v2 and
F2 = (G − H1 ) + v1 v2 be subgraphs of G0 . Since G0 is not planar, either F1 or F2 is not
 14

planar. Indeed, suppose both F1 and F2 are planar, then we may construct an embedding
of F1 with v1 v2 on the exterior region by Theorem 2.11 and also construct an embedding
of F2 with v1 v2 as an exterior region. Hence we may assume F1 is not planar.
Since F1 − v1 v2 is a proper subgraph of G then, by minimality, F1 − v1 v2 is
planar. Since G is 2-connected, then there exists some v1 , v2 -path contained F2 other
than the edge v1 v2 . Thus, in G there exists a subgraph homeomorphic to F1 . Therefore
G has a subgraph that is not planar, contradicting the minimality of G. We conclude
that G is 3-connected.
The minimality of G guarantees that there exists an edge e = uv such that the
subgraph H = G − e is planar and 2-connected. For the remainder of the proof, we
assume H is embedded in the plane. Since H is 2-connected, there exists a cycle in H
containing u and v. Among all such cycles, let C be one of maximum length. We let
C = v0 v1 v2 ...vk v0 , and we may assume u = v0 and v = vl for some l with 0 ≤ l ≤ k.
We will call an x, y-path in G that only has vertices x and y in common with C
an x,y-chordal path of C. If for all s and t with 0 < s < l and l < t ≤ k, there does not
exist a vs , vt -chordal path of C in H, then e can be added to H while preserving planarity,
a contradiction. Therefore, there must exist such a vs , vt -path in H. We illustrate this
structure in Figure 2.2.

Figure 2.4: A picture of the subgraph H.

 15

We will now develop a case analysis with the cases being determined by where
cordial paths terminate on C.
Case 1: Let the va ,vb -chordal path of C exist such that neither va nor vb are
any of v0 , vs , vl or vt . Suppose va and vb were between v0 and vl . Then the va ,vb -chordal
path could be embedded parallel to both the v0 ,vl path and the edge e. Thus such a path
would not cross the edge e and so e could be added to G preserving planarity. Similarly,
if va and vb are between the vertices from vi for l < i ≤ k, then we again are able to
generate a planar embedding of G.
Suppose va and vb are between the vertices from vi for s < i < t. Such a va , vb -
chordal path would be able to be drawn on the exterior of of our cycle in such a way
that it was parallel to to the cycle. Thus such an embedding of the chordal path would
not affect the planarity of G. Similarly, if va and vb are between the vertices from vi for
t < i ≤ k or 0 < i < s then the va ,vb -chordal path could be placed parallel to the cycle
and the vs ,vt -path in such a way to not affect the planarity of G when e is embedded.
Thus finally, without lost in generality consider if va is one of the vertices vi such
that 0 < i < s and vb is one of the vertices vj for l < j < t. Such a va ,vb chordal path
must intersect the embedding of the edge e as seen in Figure 2.5. Thus if such a chordal
path existed G is not planar. But such a graph G contains a subgraph homeomorphic
to K3,3 . This is a contradiction to our assumption G has no subgraph homeomorphic to
neither K3,3 nor K5 . Thus such a chordal path can not exist. A similar argument holds
for if va was one of the vertices vi such that s < i < l and vb was one of the vertices vj
for t < j ≤ k.
Case 2: Let the va ,vb -chordal path of C exist such that one of va or vb is either
v0 , vs , vl or vt . Without lost of generality let va be v0 , and let vb be one one the cycle
vertices but not vs , vl , or vt . Regardless of our choice of vb , we are able to embed the
va vb chordal path in such a way that it is parallel to e and thus does not intersect it.
Such a chordal path results in G being planar which is a contradiction to our assumption
that G is not planar. So consider if vb was one of the vertices vi such that l < i < t and
there existed a vertex vc such that s < c < l. Then we know from Case 1 a vb ,vc -chordal
path is planar. Then consider if the va ,vb -chordal path and the vb vc chordal path shared
at least a vertex w. We are then no longer able to embed the chordal paths in such a
way that G is planar as seen in Figure 2.6. But such chordal paths result in G having a
 16

subgraph that is homeomorphic to K3,3 which is a contradiction to our assumption G has

no such subgraph. Regardless of our choices for va and vb , we are always able to identify
a vc such that this argument holds.
Case 3: Let the va ,vb -chordal path of C exist such that va is either v0 , vs , vl or
vt and vb is either v0 , vs , vl or vt but not the same as va . Without lost of generality let
va be vs and vb be vt . If this was the only va vb -chordal path in G then we would be able
to embed the chordal path parallel to the vs vt -path in such a way that the chordal path
does not intersect the edge e and thus G would be planar, which is a contradiction to
our assumption that G is not planar. A similar argument holds true for if va is v0 and
vb is vl . Thus consider if two such chordal paths existed the first being a v0 ,vl -chordal
path and the other being a vs ,vt -chordal path. Since we know if the chordal paths do not
intersect we are able to embed them in such a way that G is planar, let the two chordal
paths intersect.
Suppose the v0 ,vl -chordal path and the vs ,vt -chordal path intersect and share at
least two common vertices. As demonstrated in Figure 2.7, such a structure must in fact
intersect some part of G, mainly it must intersect with some part of the edge e. Thus
G is not planar. But, G would have a subgraph that is homeomorphic to K3,3 which
is a contradiction to our assumption that G has no such subgraph. Thus suppose that
v0 ,vl -chordal path and the vs ,vt -chordal path intersect at exactly one vertex, namely w.
The structure would again intersect e and thus G is not planar as seen in Figure 2.8. But
G would have a subgraph that is homeomorphic to K5 which is a contradiction to our
assumption that G has no such subgraph.
From the case analysis. it becomes clear that with any such chordal path, either
we are able to embed the chordal path in such a way that G is planar, or anytime G is
not planar, G has a subgraph that is homeomorphic to either K3,3 or K5 . Thus since no
such chordal path exists then there is no such graph G.
 17

Figure 2.5: Structure of Case 1 of Kuratowski’s Theorem.

Figure 2.6: Structure of Case 2 of Kuratowski’s Theorem.

 18

Figure 2.7: Structure of Case 3 of Kuratowski’s Theorem.

Figure 2.8: Structure of Case 4 of Kuratowski’s Theorem.

 19

Chapter 3

Biplanar Graphs

Our study of biplanar graphs is inspired by the model of an electrical circuit

that was given in Chapter 1, in which it was necessary to utilize two sides of a chip. A
graph G = (V, E) is biplanar if there exists a partition of E into two parts E1 ,E2 such
that subgraphs G1 = (V, E1 ) and G2 = (V, E2 ) are planar graphs. If a graph G is bipla-
nar then we will state that G has a biplanar decomposition into the subgraphs G1 and
G2 . The ultimate goal in our study of biplanar graphs is to provide a characterization
of forbidden subgraph analogous to Kuratowski’s Theorem. Unfortunately, there exist
example of nonbiplanar graphs that are homeomorphic to biplanar graphs (See Corol-
lary 3.10). Hence, a completely analogous characterization is impossible. With this in
mind, we dedicate this chapter to an investigation of the smallest complete and complete
bipartite graphs that are not biplanar.
The goal of this chapter is to prove the following Theorems.

Theorem 3.1. The graph K9 is the smallest complete graph that is not biplanar.

Theorem 3.2. The graphs K7,7 , K6,9 and K5,12 are the smallest complete bipartite graphs
that are not biplanar.

3.1 Complete Biplanar Graphs

We will first show that the complete graph K9 is the smallest complete graph
that is not biplanar.

Lemma 3.3. The complete graph K8 is biplanar.

 20

Proof. As seen in Figure 3.1 K8 has a biplanar decomposition.

Figure 3.1: A biplanar decomposition of K8 .

In order to show that K9 is not biplanar, then we will use Lemmas, 2.5, 2.6 and
2.7 from chapter two. The original inspiration for the proof can be found here [BHK62].

Lemma 3.4. The complete graph K9 is not biplanar.

Proof. Suppose that K9 is biplanar. Then K9 has a biplanar decomposition into G =

(V, E1 ) and G = (V, E2 ). We first show that both G and G must be connected. Suppose,
to the contrary, that either G or G is not connected. With no loss in generality, we may
assume G is not connected. Consider the following cases determined by the number of
components in G:
Case 1: Suppose G has at least four components. Then there exists a partition
of the components into two parts P1 and P2 such that each part has at least three vertices.
Then, in G, every vertex in P1 is adjacent to every vertex in P2 . Since both parts are of
size at least three, G has a K3,3 subgraph. By Kuratowski’s Theorem, G is not planar,
which is a contradiction to our assumption that K9 is biplanar.
Case 2: Suppose G has exactly three components C1 , C2 and C3 . If each
component has more than one vertex, then we can partition the components into two
parts P1 and P2 such that each part has at least three vertices. Since each of P1 and P2
have at least three vertices, G has a K3,3 subgraph. By Kuratowski’s Theorem, G is not
planar, which is a contradiction to our assumption that K9 is biplanar.
 21

Now suppose that exactly one component, say C1 , of G contains exactly one
vertex. Further assume that C3 has at least as many vertices as C2 . Since there are
exactly nine vertices, it follows that C3 must have at least three vertices. Consider a
partition of vertices of G into two parts P1 and P2 , such that P1 contains the vertex of C1
and vertices of C2 , and P2 contains the vertices of C3 . Since both P1 and P2 contain at
least three vertices, G has a K3,3 subgraph. By Kuratowski’s Theorem, G is not planar
which again is a contradiction to our assumption that K9 is biplanar.
Finally, suppose that two components, say C1 and C2 , have exactly one isolated
vertex. Let C3 contain the remaining vertices. This is actually a proper subgraph of Case
3 in which the vertices in C1 and C2 would be adjacent. So instead consider Case 3.
Case 3: Suppose G has exactly two components. Let C1 and C2 be the compo-
nents of G and assume C2 has at least as many vertices as C1 . Suppose C1 has at least
three vertices in it, then it follows C2 must also have at least three vertices. Then, G
contains a K3,3 . By Kuratowski’s Theorem, G is not planar, which is a contradiction to
our assumption that K9 is biplanar.
Now suppose C1 has exactly two vertices, say v1 and v2 . Then, in G, v1 and v2
are not adjacent but are both adjacent to all the vertices v3 , ..., v9 . In order to minimize
the edges in G, we will let G be maximal. It follows that v3 , ..., v9 are degree at most
four. If vi , 3 ≤ i ≤ 9 is degree at least 5 then G contains a K3,3 as a subgraph making
G not planar by Kuratowski’s Theorem which is a contradiction to our assumption K9 is
biplanar. Thus the vertices v3 , ..., v9 form a cycle. Then G contains the graph as in Figure
3.2. Thus G contains a subdivision of K5 and is not planar by Kuratowski’s Theorem,
which is a contradiction to our assumption that K9 is biplanar.
Suppose C1 has exactly one vertex, say v1 . In order to reduce the number of
edges in G, suppose that G is maximal planar. Then, by Corollary 2.4, G has exactly
3n − 6 = 21 edges. Since we know eight of the edges are incident with v1 there exist
thirteen edges between v2 , ..., v9 . In order for G to be planar,by Theorem 2.3, it must
have m ≤ 3n − 6 edges. Since we know v1 is an isolated vertex, all edges must be between
v2 , ..., v9 except for the thirteen edges in G. Thus G must have less than m ≤ 3(8)−6 = 18
edges. But G must have all the remaining edges except the thirteen in G and thus has
87̇/2 − 13 = 19 edges. Thus G is not planar by Theorem 2.3 which is a contradiction to
our assumption that K9 is biplanar.
 22

Figure 3.2: Structure of G in Case 3.1.

From Cases 1, 2, and 3, it is clear that G and G must both be connected if K9

biplanar. Let d1 , d2 , ..., d9 be the degrees of the vertices of G such that d1 ≤ d2 ≤ ... ≤ d9 .
With no loss in generality, we may assume G is maximal planar so that the cardinality
of E2 is minimized.
If any vertex has degree eight in G, then that vertex must be an isolated vertex
in G. Since G is connected, this is impossible. Additionally since G is maximal planar, by
Lemma 2.6, every vertex has degree at least three in G. Thus 3 ≤ di ≤ 7 for i = 1, 2, ..., 9.
In the case analysis that follows, we narrow down the possibilities of the degree sequence
for G. In Case 1, we will show no two vertices may have degree seven. In case two we
show that there must exist at most one vertex of degree three. In case three we show
there must exist at most one vertex of degree four. Then finally, we show the only degree
sequence possible does not have a corresponding planar graph.
Case 1: Suppose that G has a degree sequence such that d1 = d2 = 7. Let the
two vertices of degree seven be labeled v1 and v2 . Suppose v1 and v2 are not adjacent in
G. Then v1 and v2 are adjacent in G. Then v1 and v2 together with the edge {v1 , v2 }
are a component in G. Thus G has two components which is a contradiction to G being
connected. Thus v1 and v2 are adjacent in G.
Case 1.1: Suppose there exists a vertex v9 in G that is not adjacent to v1 and
is not adjacent to v2 . Then since v1 and v2 are adjacent in G and of degree seven in G,
then v1 and v2 are adjacent to vertices v3 , v4 , ...v8 . Suppose that v9 was adjacent to three
 23

or more of v3 , v4 , ...v8 . Then the graph G contains a K3,3 subgraph since v1 , v2 and v9
are adjacent to at least three common vertices. Therefore, v9 must have degree less than
three. But this contradicts Lemma 2.6 and thus there every vertex must be adjacent to
either v1 , v2 , or both.
Hence there must exist two vertices v3 and v4 such that v3 is adjacent to v1
but not v2 in G and v4 is adjacent to v2 but not v1 in G. Suppose that either v3 or v4
has degree greater than four. We may assume v4 is this vertex. Then v4 is adjacent to
v2 and at least four other vertices. Thus v4 is adjacent to at least three vertices among
v5 , v6 ...v9 . We may assume that v4 is adjacent to v5 , v6 and v7 . It follows G contains a
K3,3 subgraph having a partite sets {v1 , v2 , v4 } and {v5 , v6 , v7 }. We may conclude that
v3 and v4 have degree at most four.
Suppose v4 has degree four. Since v4 is adjacent to v2 and not v1 , then v4 is
adjacent to three other vertices among v3 , v5 , v6 , ..., v9 . If v4 is adjacent to three vertices
among v5 , v6 , ..., v9 , then G once again contains a K3,3 subgraph. This implies that G
would not be planar by Kuratowski’s Theorem which is a contradiction to our assumption
that K9 is biplanar. We may conclude that v4 is adjacent to v3 and two other vertices,
say v5 and v6 .
Suppose that v3 is either not adjacent to v5 or not adjacent to v6 . We may
assume v3 is not adjacent to v5 . Then v3 , v4 , v5 are incident with some region R. But
since v3 and v5 are not adjacent then there exists some vertex vi also a incident with R
for i ∈ 7, 8, 9. Thus R is a region of length at least four in the maximal planar graph
G. But by Lemma 2.5, every region of a maximal planar graph is of length three. With
this contradiction, we conclude that v3 and v4 are both adjacent to v5 and v6 . Figure 3.1
illustrates the structure present in G within our subcase.
Suppose that none of v7 , v8 and v9 are adjacent to v5 . In G, none of v7 , v8 and
v9 are not adjacent to v3 or v4 since each of v3 and v4 are degree four. Then in G,
the vertices v3 , v4 , and v5 are adjacent to all the vertices v7 , v8 , and v9 creating a K3,3
subgraph in G. Thus, by Kuratowski’s Theorem, G is not planar which is a contradiction
to our assumption that K9 is biplanar. A similar argument holds when v7 , v8 and v9 are
all not adjacent to v6 .
The other possible case is if there exists a path P of length three consisting of
two vertices among v7 , v8 and v9 with one of the end points being either v5 or v6 . Without
 24

Figure 3.3: Core structure of Case 1.3.

loss of generality, let P = v9 v8 v5 . Then v7 must be adjacent to v6 . Then we obtain the

graph in Figure 3.4.

Figure 3.4: Structure of Case 1.3 with path P.

In G the vertices v3 and v4 are adjacent to v7 , v8 and v9 . Additionally, v6 is

adjacent to v8 and v9 . But v5 must be adjacent to v6 and v7 . Then through a sub
divided edge, v6 is adjacent to v7 . Thus G contains a subgraph homeomorphic to K3,3
making G and thus, by Kuratowski’s Theorem, G nonplanar which is a contradiction to
our assumption that K9 is biplanar. Note that regardless of choices for the vertices for
P , the same argument can be applied. The degree of vertices v3 and v4 must be equal to
three.
 25

Since both v3 and v4 are of degree three, then by Lemma 2.7, v3 and v4 may
have exactly zero, one or two neighbors in common. These three possibilities are the
following cases.
Case 1.4: Let v3 and v4 be of degree three and be adjacent to exactly zero
common vertices. Then both v3 and v4 are adjacent to four distinct vertices, say v5 , v6 , v7
and v8 . Let v3 be adjacent to v5 and v6 , and let v4 be adjacent to v7 and v8 . Then v9 is
not adjacent to v3 or v4 . Additionally, v9 is adjacent to either one or two of v5 , v6 , v7 , or
v8 .
Suppose that v9 is adjacent to one of v5 , v6 , v7 , or v8 , say v5 . Then v9 , v5 , v3
are not adjacent to v4 , v7 or v8 in G. Thus in G, v9 , v5 , v3 are adjacent to v4 , v7 and v8
creating a K3,3 subgraph in G. Then G is not planar which is a contradiction. Then v9
must be adjacent to two of v5 , v6 , v7 , or v8 .
Suppose that v9 is adjacent v5 and v6 . Then v1 , v3 and v9 are all adjacent to
v2 , v5 and v6 in G. Thus G has a K3,3 subgraph and, by Kuratowski’s Theorem, G is
not planar which is a contradiction to our assumption that K9 is biplanar. A similar
argument holds if v9 is adjacent to v7 and v8 .
Thus v9 must be adjacent to one of v5 or v6 and one of v7 or v8 . Let v9 be
adjacent to v5 and v7 . Then v5 and v7 are adjacent to exactly two vertices of v5 , v6 , ...v9 .
If either v5 or v7 were adjacent to any additional vertex of the set v6 , v8 , v9 , then G would
have a K3,3 and, by Kuratowski’s Theorem, G is not planar which is a contradiction to
our assumption that K9 is biplanar. So v5 is not adjacent to v4 , v7 nor v8 and v7 is not
adjacent to v3 , v5 nor v6 . Then, in G, v3 and v5 are adjacent to v4 , v7 and v8 . Also v6 is
adjacent to v4 and v7 in G. Additionally, v9 is adjacent to v6 and v8 in G. Thus G has a
subgraph homeomorphic to K3,3 and, by Kuratowski’s Theorem, is not planar which is a
contradiction to our assumption that K9 is biplanar.
Case 1.5: Let v3 and v4 be of degree three and be adjacent to exactly one common
vertex. Let v5 be that common vertex, let v6 be adjacent to v3 and v7 be adjacent to v4 .
Suppose that there existed a path P of length three consisting of v8 and v9 with either
v6 or v7 . Let P = v6 , v8 , v9 . Then G has the graph seen in Figure 3.5.
Then in G, vertices v3 and v6 are adjacent to vertices v4 , v7 and v9 . Additionally
v8 is adjacent to v4 and v7 . Finally, since v5 is adjacent to v8 and v9 , then G has a subgraph
homeomorphic to K3,3 . Then G is not planar which is a contradiction.
 26

Figure 3.5: Structure of Case 1.5 with path P.

Since there does not exist a path P of length three, then vertices v8 and v9 must
not be adjacent. Then v8 must be adjacent to v6 or v7 , and v9 must be adjacent to the
opposite vertex. Without loss of generality, let v8 be adjacent to v7 and v9 be adjacent to
v6 . Then vertices v3 , v6 and v9 are not adjacent to v4 , v7 nor v8 . Then in G, vertices v3 , v6
and v9 are adjacent to v4 , v7 and v8 . Then G has a K3,3 subgraph and, by Kuratowski’s
Theorem, G is nonplanar which is a contradiction to our assumption that K9 is biplanar.
Case 1.6: Let v3 and v4 be of degree three and be adjacent to exactly two
common vertices. This case follows a similar argument from that in case 1.2 in which
there either exists a path P of length three resulting in G has a subgraph homeomorphic
to K3 , 3 and, by Kuratowski’s Theorem, G is nonplanar which is a contradiction to our
assumption that K9 is biplanar.
From Cases 1.1 through 1.6 we conclude that, there does not exist a maximal
planar graph G with an edge set E1 with two vertices are of degree seven such that G is
planar.
Case 2: Let G have a degree sequence such that d8 = d9 = 3. Since G has two
vertices of degree three let them be labeled v8 and v9 . By Lemma 2.7, v8 and v9 may
have either zero, one, or two common neighbors. We will consider these possibilities in
the following subcases.
Subcase 2.1: Suppose v8 and v9 have exactly two common vertices. Let the
two common vertices be v6 and v7 . Let v5 be the other vertex adjacent v8 and let v4
to be the vertex adjacent to v9 . Since every vertex adjacent to v8 is adjacent and every
 27

vertex adjacent to v9 is adjacent, then v6 and v7 are of at least degree five. Since we are
considering the decomposition of K9 , then there exists three other vertices v1 , v2 and v3 .
A graph of Subcase 2.1 can be seen in Figure 3.6.

Figure 3.6: Structure of Subcase 2.1.

By Case 1, it is clear that no two vertices may be of degree seven. In particular,

v7 and v8 may not both be of degree seven. Additionally, since no vertex may be of degree
eight, v1 , v2 and v3 may not all be adjacent to either v6 nor v7 .
Suppose that v1 , v2 and v3 were not adjacent to neither v6 nor v7 . Then in G,
v6 , v8 and v9 would all be adjacent to v1 , v2 and v3 . Thus k3,3 would be a subgraph of
G making G not planar. But this is a contradiction, so at least one of v1 , v2 or v3 is
adjacent to v6 .A similar argument can be made for v7 , and so at least one of v1 , v2 or v3
is adjacent to v7 . Without lost of generality, let the vertices v1 and v6 be adjacent and
let the vertices v3 and v7 be adjacent. We will now consider two subcases for if the edge
v4 , v5 is in the edge set E1 or if v4 , v5 is in the edge set E2 .
Subcase 2.1.1: Suppose the edge v4 , v5 was in the edge set E1 . Then without
lost of generality, we may consider the edge v4 , v5 to be part of the region with vertices
v7 , v4 and v5 . Since v7 must be adjacent to at least one of v1 , v2 or v3 then without lost
of generality let v3 be adjacent to v7 . Suppose v3 was the only vertex of v1 , v2 or v3 to
be adjacent to v7 . Then a graph of subcase 2.1.1 can be seen in Figure 3.7.
The vertices v8 and v9 are not adjacent to v1 , v2 or v3 , and vertices v3 and v7
are not adjacent to vertices v1 nor v2 . Then all of these vertices must be adjacent in G.
Then G contains a subgraph homeomorphic to K5 and, by Kuratowski’s Theorem, is not
planar. This is a contradiction to our assumption that K9 is biplanar.
This is the only case to consider with the edge v4 v5 being in the edge set E1 .
 28

Figure 3.7: Structure of Case 2.1.1.

If two vertices of v1 , v2 or v3 were adjacent to v7 and the last vertex was adjacent to v6 ,
then the edge going around v7 may instead be drawn around v6 creating an isomorphic
graph.
Subcase 2.1.2: Suppose the edge v4 , v5 was not in the edge set E1 . Suppose that
v2 was not adjacent to v6 nor v7 . Then in a very similar argument to Case 2.1.1, G has
a subgraph isomorphic to K3,3 making G not planar which is a contradiction.
Subcase 2.2: Suppose v8 and v9 have one common vertex. Let v7 be that common
vertex. Let v5 and v6 be the other two vertices adjacent to v8 and let v3 and v4 be the
vertices adjacent to v9 . Let v1 and v2 be vertices of at least degree three. Then Case 2.2
has a graph like below.

Figure 3.8: Structure of Case 2.2.

Since v7 is already of degree six, then v7 can at most be adjacent to either v1 or

v2 . Then consider the following cases
Subcase 2.2.1 Suppose neither v1 nor v2 are adjacent to v7 . Suppose further
 29

that the edge v5 , v4 is not in the edge set E1 . Then in the graph G, v5 , v4 and v7 are part
of some face. But since v5 , v4 is not in E1 then there must exist some other vertex vi part
of the same face. Then G has a face of size greater then three which is a contradiction
since G is maximal. Then the edge v5 , v4 is in E1 . Similarly, v3 , v6 is in E1 .

Figure 3.9: Structure of Case 2.2.1.

Subcase 2.2.2 Suppose either v1 or v2 is adjacent to v7 . Without lost of gener-

ality, let v1 be adjacent to v7 . Similar to in Case 2.2.1, then either the edge v6 , v3 or the
edge v4 , v5 is in E1 . Without lost of generality, let the edge v4 , v5 be in E1 .

Figure 3.10: Structure of Case 2.2.2.

Suppose that v1 was not adjacent to v3 . Then v1 , v7 and v3 are part of some
 30

face with at least one more vertex. Then either v1 is adjacent to v2 , or one of v4 or v5 . If
v1 was adjacent to v2 then v2 is adjacent to some vi . In order to make G maximal planar,
every face must be of size three. Thus there exists an edge that can be added to E1 . But
since the edge v2 , v7 makes v7 a degree eight vertex, then v1 must be adjacent to v3 . A
similar argument holds for v1 and v6 to be adjacent.
Subcase 2.3 Suppose v8 and v9 have zero common vertices. Then v8 is adjacent
to three vertices that are adjacent. Let these three vertices be v5 , v6 , v7 . Then v9 is
adjacent to three vertices that are adjacent. Let these three vertices be v2 , v3 , v4 . Then
v1 is the remaining vertex. Not from Cases 2.1 and 2.2, it is not possible for v1 to be of
degree three or else it would fall into one of the previous cases of degree three adjacency.
Thus v1 must be of degree greater then three. Additionally, v1 is not adjacent to v9 nor
v9 and thus could only be of max degree six.
Case 3: Let G have a degree sequence such that di = dj = 4. By Case 1 and
Case 2, then G must have at most one degree seven vertex, and one degree three vertex.
Let v8 and v9 be the vertices of degree four. For each construction of the maximal planar
graph G with having two vertices of degree four, every case ends up with G having at
least two vertices of degree three. Thus such a case is not possible
From the Case 2 and Case 3 it follows that G must have exactly one vertex of
degree three and exactly one vertex of degree four. Then the degree sequence of G must
take on the form 5, 5, 5, 5, 5, 5, 5, 4, 3. It is left to show that this degree sequence does not
have a planar graph representation. Suppose that the degree sequence did have a planar
graph representation. Then there would exist a degree five vertex that was adjacent to
five other degree five vertices. Each of these vertices would then be at least adjacent to
two other degree vertices in the set of five degree five vertices. This would form a cycle
like construct seen in Figure 3.11.
Since there is a degree three vertex, it must be adjacent to exactly two adjacent
of the outer degree five vertices. If the degree three vertex was adjacent to three adjacent
then the center of the three degree three vertices would be a degree four. A similar
argument holds true for the degree four vertex. Then there exist exactly three cases, if
the degree three and degree four vertex are adjacent to exactly zero one or two of the
same vertices.
Case 1: The degree three and degree four vertex are adjacent to exactly zero of
 31

Figure 3.11: Structure of degree five cycle.

the same vertices. Then the graph G has the following structure:

Figure 3.12: Structure of degree sequence Case 1.

The degree four vertex would have to be adjacent to at least one of the other
vertices in the structure. If the degree four vertex was adjacent to one of the other
vertices, then it would form a cycle of length three. One of the vertex would then be
locked into an interior region and would not be able to have any additional edges adjacent
to it forcing the degree to be less then five, a contradiction.
Case 2: The degree three and degree four vertex are adjacent to exactly one of
the same vertices. Then the graph G has the following structure:
Since the vertex with all edges complete is degree five, and since every region
must be of length three, then the degree three and degree four vertices are adjacent. In
a similar fashion, the degree four and the degree five vertex that is also adjacent to the
 32

Figure 3.13: Structure of degree sequence Case 2.

degree three vertex are adjacent creating the structure seen in Figure 3.14.

Figure 3.14: End structure of degree sequence Case 2.

From this it is clear that the last vertex could only be of maximum degree three.
Then the graph G is not maximum planar which is a contradiction.
Case 3: The degree three and degree four vertex are adjacent to exactly two of
the same vertices. Then the graph G has the structure seen in Figure 3.15.
The vertices in this graph labeled with a and b are of degree five. Similar to in
Case 2, the two vertices adjacent to the degree four are already degree five, so in order
 33

Figure 3.15: Structure of degree sequence Case 3.

to keep each region a cycle of length three, then the degree four must be adjacent to a
and b making it a degree five, a contradiction.
From these cases it becomes clear that it is not possible to construct a max-
imal degree sequence of 5, 5, 5, 5, 5, 5, 5, 4, 3. So there does not exist a degree sequence
correlating to a maximal planar graph G. Thus K9 is not planar.

Lemma 3.5. The graph K9 is minimal non-biplanar.

Proof. As seen in Figure 3.16, it is possible to remove a single edge e = {2, 4} and generate
a biplanar decomposition. Thus the graph K9 − e is biplanar.

3.2 Complete Bipartite Biplanar Graphs

Although the proof that K9 is not biplanar is very technical, it follows directly
from expanding Euler’s Polyhedral Identity to show the complete bipartite graphs K7,7
and K6,9 are not biplanar. First we will expand Euler’s Polyhedral Identity to biplanar
bipartite graphs.
 34

Figure 3.16: A biplanar decomposition of K9 − {2, 4}.

Theorem 3.6. A the size and order of a biplanar bipartite graph must satisfy the in-
equality m ≤ 4n − 8.

Proof. Let G be a planar bipartite graph. By Euler’s Polyhedral Identity, in order for
G to be planar it must satisfy the equation n − m + r = 2 or r = 2 + m − n. Since
G is bipartite, G contains no cycles of length three. Then every region of G must be
incident with at least four edges and since every edge is incident with exactly two regions
we obtain the following inequality 2m ≥ 4r. By substitution, we obtain the inequality
2 + m − n ≤ m/2, which simplifies to m ≤ 2n − 4. Since a biplanar decomposition of
a bipartite graph results in two planar graphs, the number of edges and vertices in a
biplanar bipartite graph must satisfy the inequality m ≤ 4n − 8.

Lemma 3.7. The complete bipartite graph K7,7 is not biplanar.

Proof. Suppose the complete bipartite graph K7,7 is biplanar. By Theorem 3.2, the size
and order of K7,7 must satisfy the inequality m ≤ 4n − 8. By substituting 49 for m and
14 for n, we obtain 49 ≤ 4(14) − 8, or 49 ≤ 48, which is a contradiction. Thus K7,7 is not
biplanar.

Lemma 3.8. The complete bipartite graph K6,9 is not biplanar.

Proof. Suppose that the complete bipartite graph K6,9 is biplanar. By Theorem 3.2, the
size and order of K6,9 must satisfy the inequality m ≤ 4n − 8. By substituting 54 for m
 35

and 15 for n we obtain 6(9) ≤ 4(15) − 8 or 54 ≤ 52 which is a contradiction. Thus K6,9

is not biplanar.

Lemma 3.9. The complete bipartite graph K5,13 is not biplanar.

Proof. Suppose that the complete bipartite graph K5,13 is biplanar. By Theorem 3.2, the
size and order of K5,13 must satisfy the inequality m ≤ 4n − 8. By substituting 60 for m
and 18 for n we obtain 60 ≤ 4(18) − 8 or 65 ≤ 64 which is a contradiction. Thus K5,13 is
not biplanar.

Lemma 3.10. The graphs K7,7 , K6,9 and K5,13 are the smallest complete bipartite graphs
that are not biplanar .

Proof. As seen in Figure 3.2, the complete bipartite graph K6,8 is biplanar.
Since K6,7 is a proper subgraph of K6,8 , it follows that K6,7 is biplanar as seen
in Figure 3.17.. Additionally, K5,12 has a biplanar decomposition as seen in Figure 3.18.

Corollary 3.11. Biplanar graphs are not closed under subdivision of an edge.

Proof. Let G be the graph K9 − {v2 , v4 } as in Lemma 3.5. Let the edge e = {v2 , v4 } be
subdivided with a vertex v such that the resulting edges are {2, v} and {v, 4}. By placing
the edge {2, v} into one edge set and {v, 4} into the other edge set, each decomposition
of G is still is still planar. However, since K9 is not biplanar, biplanarity is not closed
under subdivisions of edges.

Note that Corollary 3.10 shows biplanar graphs are not closed under subdivision
of an edge, then biplanar graphs are not closed under graph homeomorphism.

Theorem 3.12. The graphs K9 , K7,7 , K6,9 and K5,13 are the smallest complete and
complete bipartite graphs that are not biplanar.

Theorem 3.12 follows directly from Lemmas 3.5, 3.5, and 3.12. As with Kura-
towski’s Theorem, we have identified the smallest complete and complete bipartite graphs
that are not biplanar. Thus for any graph G to be biplanar, G must not have one of K9 ,
K7,7 , K6,9 nor K5,13 as a subgraph.
 36

Figure 3.17: A biplanar decomposition of K6,8 .

 37

Chapter 4

Graph Thickness

As with planar graphs, biplanar graphs can only have a certain number of edges
in comparison to the number of vertices before the decomposition becomes impossible.
This was shown through the cases of K9 , K7,7 and K6,9 . The thickness of a graph G
is smallest number partitions of the edge set E of G such that each subgraph G1 =
(V, E1 ), G2 = (V, E2 )...Gn = (V, En ) is a planar graph. The thickness of a graph will be
denoted by θ(G)
From the previous section, it is clear that θ(K8 ) = 2 since K8 is biplanar.

Theorem 4.1. For a graph G, θ(G) ≥ dm/(3n − 6)e

Proof. Proof: Let G be a given graph. A maximal planar graph has 3n − 6 edges. By
m/(3n − 6) would give a minimum number of planes in which it would take to embed
G. Since m/(3n − 6) may be rational and not an integer, it is sufficient to consider
dm/(3n − 6)e as the minimum number of planes needed to planar embed G. Thus θ(G) ≥
dm/(3n − 6)e.

It directly follows from Theorem 4.2 that a complete graph has a minimum
thickness.

Theorem 4.2. For a complete graph Kn , then θ(Kn ) ≥ b(n + 1)/6c + 1

Proof. Let G = Kn for some n. By the Theorem 4.2, θ(G) ≥ dm/(3n − 6)e. Since G
is complete, then it has exactly n(n − 1)/2 edges. Then θ(G) ≥ dn(n − 1)/2(3n − 6)e
 38

or θ(G) ≥ dn(n − 1)/6(n − 2)e. Then by dividing with remainder, θ(G) ≥ d(n + 1)/6 +
2/6(n − 2)e ≥ b(n + 1)/6c + 1. Thus θ(Kn ) ≥ b(n + 1)/6c + 1.

In a very similar manor of expanding maximal planar, we are able to expand

the result of Eulers Polyhedral Identity for biplanar graphs to give a minimum for graph
thickness.

Theorem 4.3. For all simple bipartite graphs G, θ(G) ≥ dm/(2n − 4)e

Proof. Let G be a bipartite graph. As seen before, m ≤ 2n − 4. Then for any given
partition of the edges of a bipartite graph, there may exist at most 2n − 4. Then by
considering dm/(2n − 4)e would give a minimum number of parts needed for the partition
of the edges of G. Thus for all bipartite graphs G, θ(G) ≥ dm/(2n − 4)e

Although these are theorems only develop the foundations for studying graph
thickness, they create a very powerful tool set. From these theorems, exact formulas
for graph thickness have been determined for all complete graphs and most complete
bipartite graphs. More information that we did not have time to investigate in this thesis
is available in [Bei97].
 39

Chapter 5

Conclusion

Through this thesis, we were able to develop the required tools from introducing
a graph to major theorems that play a key role in determining which graphs are planar
and biplanar along with graph thickness. By building up to the proof that K9 is not
biplanar, we introduced many structural theorems that create an idea of what maximal
planar graphs can look like. Although this thesis is self contained, there is still a lot of
work to be done in the topic of biplanar graphs.
Within future research, we would further investigate more classifications of bipla-
nar graphs including multipartite. Additionally, we would add edges to smaller biplanar
bipartite graphs to determine if there exist any other cases of graphs that are not com-
plete and are not biplanar. After these cases are studied, we would be able to generate
all forbidden subgraphs for biplanar graphs.
In an attempt to still expand Kuratowski’s Theorem, we would also further
investigate additional structures within biplanar graphs. Through these structures we
would hope to find some function on graphs that would assist in determining if a given
graph is biplanar or not.
 40

Bibliography

[Bei97] L.W. Beineke. Biplanar graphs: A survey. Computers and Mathematics with
Applications, 34(11):1–8, 1997.

[BHK62] Joseph Battle, Frank Harary, and Yukihiro Kodama. Every planar graph with
nine points has a nonplanar complement. Bulletin of the American Mathemat-
ical Society, 68(6):569–571, 1962.

[CLZ10] G. Chartrand, L. Lesniak, and P. Zhang. Graphs and Digraphs, Fifth Edition.
A Chapman and Hall book. Taylor and Francis, 2010.

[GM12] Gary Gordon and Jennifer McNulty. Matroids: A Geometric Introduction.

Cambridge University Press, 2012.