Math 3322: Graph Theory1 Mikhail Lavrov

Lecture 24: Graph coloring and the greedy algorithm

November 7, 2024 Kennesaw State University

1 The chromatic number

Today we will talk about coloring graphs.
Informally, this means painting every vertex of a graph some color. We can formally define a
coloring of G as a function f : V (G) → S, where S is the set of colors: we call f (v) the color
of vertex v. Usually, it will not matter what S is: we can be poetic and pick a set like S =
{red, blue, purple, violet}, or practical and pick a set like S = {1, 2, 3, 4}.
It will matter how large S is: how many colors we use. If |S| = k (in which case, it is convenient
to set S = {1, 2, 3, . . . , k}), we say that f is a k-coloring of G. We like to use as few colors as
possible. . .
. . . but right now, we haven’t said anything that requires us to use more than one color. We say
that a proper coloring of G is a coloring in which adjacent vertices receive different colors. This
goes back to the origins of graph coloring: coloring maps, in which adjacent regions should have
different colors to distinguish them. However, graph coloring shows up in many applications: if
edges represent “conflicts” between vertices (which is a very flexible interpretation), and we want
to partition the vertices into non-conflicting groups, then we’re looking for a proper coloring.
We say that G is k-colorable if it has a proper k-coloring. The chromatic number of G,
denoted χ(G), is the minimum value of k for G is k-colorable. Once again, we’re faced with an
optimization problem, and so we can make some familiar observations:
• If we find any proper k-coloring of G, this is a proof by demonstration that G is k-colorable:
that χ(G) ≤ k.
• Proving that χ(G) ≥ k seems very hard. The brute-force approach is to show that none of
the many many (k − 1)-colorings of G are proper.
I am talking about graph coloring as though it is a new problem, but we have already seen one
aspect of it near the beginning of the semester. A graph G is 2-colorable if and only if it is bipartite,
and a bipartition is very nearly the same thing as a proper 2-coloring: just color vertices by which
side of the bipartition they’re on.

2 Greedy coloring
We are often interested in proving that all graphs of a certain type are k-colorable for some k. Here,
it is not enough to draw a picture, because there could be infinitely many graphs of that type to
1
This document comes from the Math 3322 course webpage: http://facultyweb.kennesaw.edu/mlavrov/
courses/3322-fall-2024.php

1
color. Usually, the way we do this is to find an algorithm that tells us how to color the graphs we
care about, and then prove that the algorithm never uses too many colors.
The simplest graph coloring algorithm is the greedy coloring algorithm. This does the follow-
ing:
1. Number the vertices v1 , v2 , . . . , vn in an arbitrary order. (We will usually illustrate this by
drawing the graph so that the vertices are v1 , v2 , . . . , vn from left to right.)
2. Color the vertices in that order, giving vi the first color2 that does not appear on any of its
neighbors among v1 , v2 , . . . , vi−1 .
This is “greedy” because it does the first thing it can think of without thinking about future
consequences. Sometimes this backfires. The graph below is a tree, so it is bipartite—and 2-
colorable. But if we go from left to right, we get a 4-coloring (labels in red):

1 2 1 3 1 2 1 4

The truth about the greedy algorithm is that the result heavily depends on the order of vertices
v1 , v2 , . . . , vn that we picked at the start. In the example above, I had to carefully order the vertices
to get the tree to be 4-colored. On the other hand, if you want the tree to be 2-colored, you also
have to order the vertices at least a little bit carefully.
The following bound on chromatic number is what we’re guaranteed to get even if we take no care
in ordering the vertices:
Theorem 2.1. For any graph G, χ(G) ≤ ∆(G) + 1.

Proof. We will prove that the greedy algorithm never uses more than ∆(G)+1 colors. (Here, ∆(G) is
the maximum degree of G.) In other words, if our colors come from the set S = {1, 2, . . . , ∆(G)+1},
the greedy algorithm will never get stuck.
When we are coloring vertex vi , it has at most ∆(G) neighbors among v1 , v2 , . . . , vi−1 . (It could
have fewer for two reasons: it’s possible that vi has fewer than ∆(G) neighbor, and it’s possible
that not all of them come before vi in the ordering we picked.)
Among those neighbors, there can be at most ∆(G) colors. Therefore there is at least one color
from the set S = {1, 2, . . . , ∆(G) + 1} that does not show up on any of them. We can continue by
giving vi the first such color.
Since the greedy algorithm never fails to pick a color from S, at the end we will have a proper
(∆(G) + 1)-coloring of G.

In the remainder of the lecture, we will apply the greedy coloring algorithm to two special cases of
the graph coloring problem. By putting some thought into the order we color the vertices, we will
be able to do better than the general bound in Theorem 2.1.
2
The “first color” assuming the colors are numbered 1, 2, 3, . . . , so that they have a definite order.

2
3 Greedily coloring planar graphs
Coloring a map is the origin of graph coloring, and when we color a map, we are usually coloring
a planar graph. (Put a vertex in each region on the map. Draw an edge between vertices if
their regions share a border. If the regions are connected—which is not always true in real-world
maps—then these edges can be drawn without crossing.)
By carefully ordering the vertices of a planar graph, we can get the greedy algorithm to perform
better. Here, the heuristic we will use is that it is good to leave low-degree vertices until
the end. If a vertex v does not have many neighbors, then even if v’s neighbors all get different
colors, there will be a color left for v.
What kind of low-degree vertices will planar graphs have? We can show the following:
Lemma 3.1. Every planar graph G has minimum degree δ(G) ≤ 5.

Proof. This is true for every planar graph with at most 6 vertices because at that point, you can’t
have any degrees bigger than 5.
For planar graphs with n ≥ 3 vertices and m edges, we have the inequality m ≤ 3n − 6. However,
if every vertex had degree 6 or more, we would have m ≥ 12 (6n) = 3n by the handshake lemma,
and we cannot have 3n ≤ 3n − 6. Therefore not all vertices have degree 6 or more: there must be
a vertex with degree 5 or less.

Just having a small minimum degree would not help us. What does help us is that when you remove
a vertex of minimum degree, we are left with a smaller planar graph, for which the following holds.
This lets us pick a good vertex ordering:
Lemma 3.2. Every planar graph G has a vertex ordering v1 , v2 , . . . , vn in which each vertex is
adjacent to at most 5 of the vertices that come before it.

Proof. We will prove this by induction on n. When n is small (say, n ≤ 6), any vertex ordering
will do.
Assume that the lemma is true for all (n − 1)-vertex planar graphs, and let G be an n-vertex planar
graph. By Lemma 3.1, G has a vertex v with deg(v) ≤ 5.
Apply the induction hypothesis to find a vertex ordering v1 , v2 , . . . , vn−1 of G − v. We can extend
it to a vertex ordering of G by setting vn = v. This vertex ordering does what we want for the first
n − 1 vertices by the induction hypothesis, and for the last vertex because deg(v) ≤ 5.
By induction, we can find such a vertex ordering for all n.

Using this lemma, we can get a bound on the chromatic number of arbitrary planar graphs!
Theorem 3.3. Every planar graph G has χ(G) ≤ 6.

Proof. Using the vertex ordering given by Lemma 3.2 and 6 colors, apply the greedy coloring
algorithm to color G. We will show that at every step in the greedy algorithm, one of the 6 colors
is available to color the next vertex.

3
Since each vertex vi is adjacent to at most 5 of the vertices that come before it, there will be at
least one of the colors from S = {1, 2, 3, 4, 5, 6} that does not appear on any of vi ’s already-colored
neighbors. So we can always continue by assigning vi some color from S. When we’re done, we will
have given G a proper 6-coloring.

Much more is true! In fact, there is a famous theorem called the Four Color Theorem that χ(G) ≤ 4
for every planar graph G. However, this is much harder to prove, to the point that all known proofs
require a computer to check many cases.

4 Greedily coloring interval graphs

Recall from the previous lecture that an interval graph is a graph defined by a set of intervals
{[a1 , b1 ], [a2 , b2 ], . . . , [an , bn ]}. For each interval [ai , bi ], there is a vertex vi ; two vertices are adjacent
whenever the intervals overlap.
Here is an example of a set of intervals and their interval graph:
A
B
C A B
D C D F G
E
F
G
E H
H

If planar graphs are the historically important application of graph coloring, then interval graphs
are the modern application. There are two important applications that boil down to coloring an
interval graph:
• Event scheduling. Here, we imagine that each interval [ai , bi ] represents an event starting at
time ai and ending at time bi . We want to put the events in rooms, but we have a limited
supply of rooms available, so we want to use as few as possible. However, if two events are
happening at the same time, they must be in different rooms.
• Register allocation: an application to computer science. Here, each interval [ai , bi ] represents
a variable in a computer program, which is initialized at step ai in the code, and last used at
step bi . There are special parts of computer memory called registers which are particularly
easy to access, but their number is limited. We want to see if the variables can be stored in
registers, so that variables which are in use simultaneously occupy different registers.
With interval graphs, the greedy algorithm turns out to perform well with the natural ordering of
the vertices. We sort the intervals [ai , bi ] by their starting point, so that a1 < a2 < · · · < an . Then,
we color the vertices v1 , v2 , . . . , vn in this order.
How well does this do? Here’s how well:
Theorem 4.1. If G is an interval graph, then χ(G) = ω(G) (ω(G) is the clique number of G).

4
Proof. Suppose we are coloring vertex vi during the greedy algorithm. Which vertices among
v1 , . . . , vi−1 are adjacent to this vertex: which of the intervals that come before [ai , bi ] overlap with
it?
Well, they all start before ai , so to overlap with [ai , bi ], they must end after ai . This means that
they all contain the point ai itself. But if we’re looking at a set of intervals that all contain ai , then
they form a clique, because they all overlap! So, together with vi , we are looking at ω(G) or fewer
vertices: vi has at most ω(G) − 1 neighbors that come before it.
Therefore the greedy algorithm will never use more than ω(G) colors.
(Conversely, it will have to use at least ω(G) colors, because the vertices of a clique will all need
different colors. More on this in the next lecture.)

5
5 Practice problems
1. Draw a tree that, when the vertices are unfortunately ordered, ends up being given 5 different
colors by the greedy algorithm.
(Think about how the 4-color example works; don’t try to reinvent the wheel.)
2. Here is a plane embedding of the icosahedral graph.

(a) Find a proper 4-coloring of this graph.

(b) Prove that there is no proper 3-coloring of this graph: the chromatic number is exactly 4.
3. The lowest degree last vertex ordering of an n-vertex graph G is constructed as follows,
recursively. We choose the last vertex, vn , to be any vertex whose degree is the lowest degree
in G. To find the ordering v1 , v2 , . . . , vn−1 of the other vertices, we find the lowest degree last
ordering of the (n − 1)-vertex graph G − vn .
This is exactly the vertex ordering which we described in Lemma 3.2 for planar graphs.
Prove that if G is a tree, then the greedy coloring algorithm, using the lowest degree last
ordering, will never use more than 2 colors.
4. When we take the union of graphs with the same vertex set, we just keep that set of vertices
and include an edge if it is contained in any of the graphs.
(a) Prove that if G is the union of k forests, then the minimum degree of G is 2k − 1 or less.
(Think about the proof of Lemma 3.1.)
(b) Prove that if G is the union of k forests, then it is 2k-colorable.
(c) Prove that the complete graph K2k is the union of k forests, so we can’t prove any better
result than what we got in part (b).
5. (a) Suppose that the greedy algorithm
uses k colors on a graph G. Prove that G must have
at least 1 + 2 + · · · + (k − 1) = k2 edges.
(Hint: whenever a color other than color 1 is used, there are some edges that have to
exist to cause this.)
√
(b) Prove that if G has m edges, then χ(G) ≤ 1 + 2m.