Bounds on colorings of hypercube graphs

August 2020

Registration number: Math-040

Name: Aaryan Sukhadia
School: British School Jakarta
Country: Indonesia
Abstract

Graph theory is an active field of mathematics about the study of vertices

and edges, and is used in endless applications concerning the pairwise connec-
tions between entities, from modelling computer networks to postal service
routes. In this paper we examine combinatorial optimizations on the graphs
of hypercubes, and in particular colorings on them, inspired by a question
in Stanford University Math Camp 2019 Admission Exam. Numerous work
on similar coloring problems has been done, but we will be looking at a spe-
cific one. An investigation is carried out on the minimum number of vertices
needed to be colored on an uncolored n-dimensional hypercube graph in order
for every vertex to be adjacent to at least one colored vertex. In addition, we
look at a similar but converse problem of the maximum number of vertices
that can be colored on an uncolored hypercube graph such that no vertex is
adjacent to more than one colored vertex. Graphs will be visualised using the
NetworkX Python library, and finally, potential applications of these results
in information theory will be discussed.

Keywords: Graph theory, Coloring, Hypercubes, Combinatorics, Python,

NetworkX

Acknowledgements: Firstly I would like to thank Stanford Univer-

sity and in particular the team behind the Stanford University Math Camp

1
(SUMAC) for the problem that inspired this paper. I would also like to
extend gratitude towards Mr Mark Watson, my teacher and supervisor for
helping guide me through mathematics. Finally, to my teacher Mr Sunil
Kumar for providing the coding skills to create the graph visualisations

2
Commitments of Academic Honesty and Integrity

We hereby declare that we:

1. are fully committed to the principle of honesty, integrity and fair play
throughout the competition.

2. actually perform the research work ourselves and thus truly understand
the content of the work.

3. observe the common standard of academic integrity adopted by most

journals and degree theses.

4. have declared all the assistance and contribution we have received from
any personnel, agency, institution, etc. for the research work.

5. undertake to avoid getting in touch with assessment panel members in

a way that may lead to direct or indirect conflict of interest.

6. observe all rules and regulations of the competition.

7. agree that the decision of YHSA(Asia) is final in all matters related to

the competition.

3
Signatures

Aaryan Sukhadia

Mark Watson

Noted and Endorsed by:

David Butcher
British School Jakarta Principal

4
Contents

1 Introduction to Problem 6

2 Background Information 8

3 Application to the problem 10

4 Conclusion and Potential Applications in Information The-

ory 18

5
1 Introduction to Problem

The following is the statement of the original problem.

Suppose n is a positive integer. The (imaginary) sea of Babab has islands

each of which has an n-letter name that uses only the letters “a” and “b,”
and such that for each n-letter name that uses only the letters “a” and “b,”
there is an island. For example, if n = 3, then Aaa, Aab, Aba, Baa, Abb,
Bab, Bba and Bbb are the islands in the sea of Babab. The transportation
system for Babab consists of ferries traveling back and forth between each
pair of islands that differ in exactly one letter. For example, there is a ferry
connecting Bab and Bbb since they differ only in the second letter.

Babab does not have much in the way of natural resources or farm land
so nearly all food and supplies are provided by the Babab All Bulk Company
(BABCO). The people of Babab (Bababians) desire easy access to a BABCO
store, where “easy access” means there is a BABCO store on their own island
or on one that they can get to with a single ferry ride. However, BABCO
finds it uneconomical to give the people on one island easy access to two
different BABCO stores, and BABCO is willing to deny some Bababians
easy access to a BABCO store in order to meet this restriction.

• In the cases n = 3, n = 4, and n = 5, what is the maximum number

of stores that BABCO can build while satisfying the restriction than no

6
 one has easy access to more than one BABCO store?

• Now suppose BABCO changes its strategy and decides it wants to be

sure every Bababian has access to a BABCO store even if it means
some Bababians have easy access to two stores. What is the minimum
number of stores needed to satisfy this condition in the cases n = 3, = 4,
and n = 5?

7
2 Background Information

We will frame this problem using graph theory. To do this we will have to
formalise many notions.

Definition. A graph G is a pair {V, E} where V is the set of elements

vertices and E is a set of elements known as edges, each edge being an
unordered pair of (not necessarily distinct) vertices. [3]

We will label elements of V as v1 , v2 , and so forth, and the elements of E

ad < vi vj > for some vi , vj ∈ V . Unless otherwise stated, it is assumed that
V and E are part of the same pair in G.

When representing a graph visually, the vertices are typically depicted as

points on a plane and the edges as lines between those points.

Example 2.0.1. Here is a visualisation of the graph G = {{v1 , v2 , v3 , v4 }, {<

v1 v3 >, < v1 v2 >}:

The following is more useful terminology:

• The degree of a vertex vi is the number of edges that contain vi as an

element.

• Two vertices vi , vj are said to be adjacent if < vi vj > is an edge.

Another important concept we will use is from information theory con-

cerning strings:

8
Figure 1: A graph made using Networkx library in python. Note that not
every vertex requires an edge connected to it

Definition. The Hamming distance between two strings of equal length

is the number of positions at which the characters differ.

Example 2.0.2. The Hamming distance between ’BABBAB’ and ’AAB-

BAA’ is 2, since the two strings differ in 2 positions, the first character and
the last one.

Hamming distance can also be thought of as the minimum number of

changes or substitution required to rename a string to another.

9
3 Application to the problem

The first step to abstract the problem is to rename the islands. Instead of
a and b we will use the digits 1 and 0 respectively. That way, the names
of the islands in an n-letter Babab Ocean can now be represented as binary
integers of length n rather than alphabetical strings.
We consider each of the islands (binary integers) in as vertices in a graph.
The edges represent ferry routes between the islands, and so each vertex is
connected to n other vertices. Then, for every n = 2, 3, 4.., we denote the
graph of all the islands with names of length n as Gn For example, below is
a visualisation of G2

Figure 2: A graph of the islands when n = 2

[. The number of vertices in Gn is 2n and the number of edges is 2n−1 n.

Proof. We know each vertex of Gn is a unique binary string of length n.

10
For each character, we have 2 choices: a 1 or 0. Thus, the total number of
possible strings is simply 2n .

The degree of each vertex in Gn is n. This is because if 2 vertices differ

by one character they are adjacent, and there are n possible characters to
differ by, which means each island has a ferry route with n other islands.
This gives us 2n n edges in total. However, we have double counted, since
each edge is connected to 2 different vertices. Thus, the actual number of
edges is 2n−1 n.

A graph such as this with 2n vertices and n×2n−1 edges, with each vertex
connected to n others can also be seen as the graph of an n-dimensional
hyper-cube. (I.e n=2 is a square, n=3 is a cube, n=4 is a tesseract, and so
on).

Figure 3: Graphs of G1 , G2 and G3 as hypercubes. [2]

All of the graphs we’ve seen so far can be described as uncolored. What

11
this means is that no vertex is different from any different vertex. However,
to simulate the situation in the question, we can start assigning vertices a
color. A color is essentially a label that marks a vertex out, in this case a
colored vertex represents an island with a store on it.
Thus, we have the following analogy: Each island is initially an uncolored
vertex of the graph, and adding a store to the island is analogous to coloring
the vertex. This provides a useful abstraction of the problem.
We can now construct the hyper-cube slightly differently, in layers. With-
out loss of generality, we set the first vertex to be colored (i.e the first island
that we put a store on) as v1 . We call this layer x0 . After that, every vertex
in Gn with a Hamming distance of 1 from v1 is put in layer x1 , a Hamming
distance of 2 in layer x2 , and so on until xn , which is a single vertex that is
the complete opposite of every character of v1 .
This is because the islands in layer xi all differ from x0 in i places, and
there’s ni ways to choose this. Another way to say this is the Hamming

distance from each node in xi to the initial island is i.

Maximum number of stores

For every vertex we color, we need to rule out or close off all other is-
lands/vertices with Hamming distance 2 or less from that island, else the
nodes with Hamming distance 1 would have easy access to more than one
store. Let us examine each graph for each case.

12
n=3

The graph for n = 3 is composed of 4 layers as such: We notice that coloring

Figure 4: Graph of n=3

a vertex prevents us from coloring any vertex of a distance 2 or less away from
it, as otherwise the nodes adjacent to that node would have easy access to
more than one store. Thus, we see the only other place we can place another
store on the graph is if it is on the opposite vertex of our initial island. Thus,
the maximum number of stores we can place here is 2.

n=4

Again, drawing out the graph and placing a store on the initial island: x1
and x2 are ruled out. The only layers left are x3 and x4 , and coloring any
vertex in them any of them closes off any other islands. Thus, the maximum
for n = 4 is also 2 stores

13
 Figure 5: Edges in middle layers of graph excluded for simplicity

n=5

The graph for n = 5 has 6 layers. After coloring v1 , a store can’t be placed
on layers x1 and x2 , else layers in x1 would have easy access to more than one
store. Thus, the next vertex we color must be in x3 . Note that the Hamming
distance between any pair of vertices in the same layer is always even, and
less than 5 (since 5 is the maximum hamming distance for a string of length
5) For any island that we choose in x3 , there will be 6 other islands in x3 of
Hamming distance 2 or less. We can see this because in x3 there will be 3 bs
and 2as. To get to the same layer with 2 characters different, we must switch
a 1 to an 0 and vice versa. There are 3 × 2 = 6 ways to do this. Thus, we
are left with 3 vertices for which the Hamming distance is 4 from the colored
vertex we chose in x3 (since it has to be even, and less than 5). Thus, each of

14
Figure 6: Layers x2 and x3 in the graph. The unmarked islands on x3 all
have Hamming distance 2 from the chosen island

these have Hamming distance 2 from one another and coloring any of them
rules out the rest in x3 .
Coloring a vertex in x3 also rules out x5 , and some vertices in x4 . Each vertex
in x3 will share edges with 2 vertices in x4 . We have added 2 stores in x3 ,
and thus we will rule out at most 4 out of 5 vertices, and at least 3.Thus we
can color one more vertex, bringing our maximum up to 4. Note also that
coloring a vertex in x4 rules out all other vertices in x4 .

Finding minimum number of stores

n=3

Looking at our graph for n=3 before, we see that it also satisfies the condition
of every island having easy access to an store, and thus the minimum is also
2. It cannot be one, since that would only cover a total of 4 vertices out of
the 8.

15
n=4

Our initial vertex v1 gives easy access to all nodes in x1 . x2 contains 6

vertices. The islands in x1 and x3 each have 3 connections to the islands in
x2 . We pick an island from x1 , and a corresponding island from x3 such that
the Hamming distance between the 2 colored vertices is 4. This will cause
them each to connect to 3 different islands in x2 , which gives all islands in x2
easy access to a store. Finally, we color a vertex in x4 to give the remaining
islands in x3 easy access. This is optimal, since we cannot have less than 2
islands which connect to all 6 islands in x2 , since each only connects to 4.

n=5

We again start with our initial island v1 , which gives easy access to all vertices
in x1 . Our last two store placements were symmetrical, so symmetry seems
to be the way to go here. We color a vertex in x5 to give easy access to all
the vertices in x4 . Now we have x2 and x3 to worry about.
We need to try and connect to as many uncolored vertices as possible in x2
with as few colored vertices as possible. Adding a store to a vertex of x2
only rules out that vertex in x2 , since all other vertices in the same layer are
at least Hamming distance 2 away. Adding a store in x3 rules out 3 vertices
in x2 , since there’s 31 ways to swap one of the character repeated 3 times.

Adding a store in x1 , however, connects to 4 vertices in x2 . Let’s go with

this.
Suppose we color 2 different vertices in x1 . This means that 7 vertices in

16
x2 will be ruled out (not 8, because they have Hamming distance 2 and
thus must have an island in x2 in common). The 3 remaining in x2 will
have Hamming distance 4 from that common island, since each island on x2
either has distance 4 or 2, and those with 2 are already accounted for by the
connections with the stores on x1 . This means that those 3 all connect to
the island with Hamming distance 5 from the common island on x2 , so we
color that vertex that to ensure all islands on x2 have easy access to a store.
We can repeat the same process with x3 and x4 to get a total of 8 stores

Figure 7: Connections between x1 and x2 and x3

required.

17
4 Conclusion and Potential Applications in

Information Theory

We saw in this paper how we could represent hypercubes as binary strings of

constant length, each string being a vertex of a graph representing a hyper-
cube. This allowed us to abstract the problem in the beginning of the paper
and solve it using graph theory and combinatorics methods.
These types of methods could also prove useful in computer science and
information theory. In particular, [1] has done work on how we can construct
fault-tolerant computer systems and error correcting codes. The concept of
Hamming distance is a useful abstraction for transmission errors in data
transmission. Thus, this problem can be seen to have real-life applications
as well.

18
Bibliography

References

[1] Jehoshua Bruck and Ching-Tien Ho. “Fault-tolerant cube graphs and
coding theory”. In: IEEE Transactions on Information Theory 42.6
(1996), pp. 2217–2221.

[2] Prateek Kunwar. “Embedding of hypercube graphs on orientable sur-

faces”. In: (Dec. 2016).

[3] Keijo Ruohonen. Graph Theory. Online Publishing, 2013.

19