Math 3322: Graph Theory1 Mikhail Lavrov

Lecture 19: Tournaments

(canceled) Kennesaw State University

1 Definitions
A tournament is a special kind of directed graph. It has no loops, and for every pair of vertices
v and w, exactly one of the possible arcs (v, w) or (v, w) exists.
You should imagine an actual (round-robin) tournament in which V is the set of contestants. If
v, w ∈ V , then at some point v plays against w. If v beats w, then we record that information with
the arc (v, w), and if w beats v, then we record that information with the arc (w, v). This is only
one of the ways tournaments can come up in applications, but it’s good for intuition.
In some ways, the world of tournaments is just as rich as the world of undirected graphs. If we
n
fix a vertex set V = {v1 , v2 , . . . , vn }, then there are 2( 2 ) possible graphs with that vertex set: for
n
each of the n2 pairs {vi , vj }, choose whether edge vi vj is present or not. Similarly, there are 2( 2 )

possible tournaments with that vertex set: for each of the n2 pairs {vi , vj }, choose whether arc

(vi , vj ) or arc (vj , vi ) is present.

1.1 Transitive tournament

In the world of undirected graphs, there are two “extreme cases”: the empty graph, and the
complete graph. In the world of tournnaments, there is just one “extreme case” (or n!, depending
on how you count). Here it is:
Theorem 1.1. Up to isomorphism, there is only one acyclic tournament.

Proof. We know that every acyclic digraph has a topological ordering: we can rename its vertices
v1 , v2 , . . . , vn such that only arcs (vi , vj ) with i < j exist.
If we do this to an acyclic tournament, we know exactly which arcs exist: all the possible arcs
(vi , vj ) with i < j. For example, when n = 5:

So there is only one possible acyclic tournament we could get. This is “up to isomorphism”,
because the step where we renamed the vertices to respect the topological ordering is exactly an
isomorphism of directed graphs.
1
This document comes from the Math 3322 course webpage: http://facultyweb.kennesaw.edu/mlavrov/
courses/3322-fall-2024.php

1
There are n! possible ways to order n vertices, so if we don’t say the magic words “up to isomor-
phism”, there are n! possible acyclic tournaments on a fixed vertex set of size n. For every possible
ordering of the n vertices, direct all the arcs to go from the earlier vertex to the later vertex in the
ordering.
An acyclic tournament is called a transitive tournament. If we interpret arc directions as com-
parisons, then a transitive tournament is one where all comparisons satisfy the transitive property:
if there are arcs u → v and v → w, there is also an arc u → w. (This is analogous to “if u > v and
v > w, then u > w”.)
This is not obviously equivalent to “there are no cycles”. We could prove that the two properties
are equivalent with a long argument now, or we could wait until later in this lecture when we’ll be
able to do it in one line.

1.2 Score sequences

(This section is in the notes aspirationally, but probably we won’t actually have time to talk about
most of it in class.)
When we discussed indegrees and outdegrees in class, the question was asked: do directed graphs
have degree sequence?
Certainly it is possible to record a sequence of pairs (deg+ (v), deg− (v)) for every vertex v, but this
is a cumbersome and confusing object. There’s no perfect way to sort it from largest to smallest,
for example; how do you compare pairs like (1, 5) and (3, 2)? We could peel apart the indegree and
outdegree sequences, but then we lose the information of which goes with which.
With tournaments, the situation is much nicer. If the tournament has n vertices, then an outdegree
of deg+ (v) always goes with an indegree of deg− (v) = n − 1 − deg+ (v), because the total number
of arcs that start or end at v is n − 1. So we capture all the information there is by writing down
the sequence of outdegrees.
This is also called the score sequence of the tournament. That’s because, in the round-robin
competition analogy, your outdegree is exactly your score: the number of times you’ve won against
another participant.
Claim 1.2. Up to isomorphism, there is only one tournament with score sequence 0, 1, 2, . . . , n − 1:
the transitive tournament.

Proof. Induct on n. When n = 1, the tournament with score sequence 0 is definitely a transitive
tournament because there’s nothing for it to be. Now assume this holds for n − 1 and let’s try to
prove it for n.
The vertex with outdegree n − 1 beats everyone else. Delete that vertex, and we get a tournament
with score sequence 0, 1, 2, . . . , n − 2. (Nobody else’s score decreases, because we deleted a vertex
that nobody else beat.) By induction, we’re left with a transitive tournament on n − 1 vertices.
When we add the vertex with outdegree n − 1 back in, we can put it first in the topological
ordering; all arcs out of that vertex will go forward, so the topological ordering still works, and our
tournament is still acyclic. Therefore it’s transitive.

2
The score sequence
of an n-vertex tournament has values between 0 and n − 1, and the sum of all n
scores should be n2 = n(n−1)
2 : the total number of arcs in the tournament. But not all sequences
with this property are score sequences. For example, consider the sequence
1, 1, 1, 1, 5, 6, 6
which does satisfy 1 + 1 + 1 + 1 + 5 + 6 + 6 = 72 . What goes wrong here? Well, if we just look at

the four vertices that have

score 1, we conclude that there are at most 4 arcs between them. But
there’s supposed to be 42 = 6 arcs!
In general, we get the following necessary conditions for an ascending sequence p1 ≤ p2 ≤ · · · ≤ pn
to be a score sequence of a tournament:
 
k
p1 + p2 + · · · + pk ≥ k = 1, . . . , n − 1
2
 
n
p1 + p2 + · · · + pn = .
2

Theorem 1.3 (Landau). These conditions are also sufficient.

We won’t go into the details of the proof here. But the intuition for how to construct a tournament
with this score sequence is to assume that the vertex with score pn beats the vertices with scores
pn−1 , pn−2 , . . . , pn−pn , and repeat with the n − 1 other vertices. This is very similar to how we
proved the Havel–Hakimi theorem.

2 Hamiltonian paths and cycles

2.1 Hamiltonian paths
We didn’t really talk about Hamiltonian paths earlier in this class, because we were pressed for
time and these are objectively less interesting than Hamiltonian cycles. But a Hamiltonian path is
just what you’d expect by analogy: it is a path that visits every vertex exactly once.
Why are we talking about them now? Because of this theorem:
Theorem 2.1. Every tournament has a Hamiltonian path.

Proof. Our tournament probably has cycles, so it probably doesn’t have a topological ordering, but
we can still do our best. Let v1 , v2 , . . . , vn be an ordering of the tournament so that as many as
possible of the arcs (vi , vj ) satisfy i < j. Informally, the ordering has as many forward arcs as
possible.2
In particular, for every i = 1, 2, . . . , n − 1, the arc between vi and vi+1 has to be oriented (vi , vi+1 )
in this ordering. If we had the arc (vi+1 , vi ) instead, we could switch the order of vi and vi+1 and
increase the number of forward arcs: the arc (vi+1 , vi ) would become a forward arc, and nothing
else would change. But we assumed we had as many forward arcs as possible, so this can’t happen.
Therefore (v1 , v2 , . . . , vn ) is a Hamiltonian path.
2
This is called a median order of the tournament. It’s very exciting for people who study tournament theory,
but we will only get to see a glimpse of why.

3
Now we can go back and prove that the two notions of “transitive tournament” are equivalent. This
is not a one-line proof in the notes, but that’s only because I’m being nice to you and explaining
the details; it is a one-line proof in spirit.
Corollary 2.2. If a tournament is transitive (defined by the property that whenever (u, v) and
(v, w) are arcs, so is (u, w)) then it is also acyclic.

Proof. Let (v1 , v2 , . . . , vn ) be a Hamiltonian path in a transitive tournament.

We know that it has all the arcs (vi , vi+1 ) for 1 ≤ i ≤ n − 2. By transitivity applied to arcs (vi , vi+1 )
and (vi+1 , vi+2 ), it also has all arcs (vi , vi+2 ) for 1 ≤ i ≤ n − 2. By transitivity again, we get arcs
(vi , vi+3 ) for 1 ≤ i ≤ n − 3. Repeat this (technically we’re giving a proof by induction) and we get
arcs (vi , vj ) for all 1 ≤ i < j ≤ n.
Therefore v1 , v2 , . . . , vn is a topological ordering, and the tournament is acyclic.

2.2 Hamiltonian cycles

Hamiltonian cycles are a harder sell. Some tournaments don’t have them; in particular, the tran-
sitive tournament doesn’t have any cycles.
However, the exciting thing about tournaments is that here, we can write down a simple necessary
and sufficient condition for a Hamiltonian cycle. (By “simple” I mean “can be checked much more
easily than by a brute-force search for Hamiltonian cycles”.)
Theorem 2.3 (Camion). A tournament on n ≥ 2 vertices has a Hamiltonian cycle if and only if
it is strongly connected.

Proof. Why is strongly connected a necessary condition? Well, if you have a Hamiltonian cycle in
a tournament, you can just follow it around to get from any vertex to any other vertex you like.
To prove that it’s sufficient, we’ll start small. We know that the transitive tournament is not
strongly connected: it has a vertex of outdegree 0 that definitely can’t reach any other vertex. So
our strongly connected tournament is not transitive. Therefore it has three vertices u, v, w where
transitivity fails: (u, v) and (v, w) are arcs, but we have the arc (w, u) rather than the arc (u, w).
This gives us a cycle (u, v, w, u) of length 3. Now we will simply work on making this cycle longer
and longer until it’s Hamiltonian.
Let’s suppose our cycle at some step is C = (v1 , v2 , . . . , vk , v1 ). If we are not yet done, then there
are some vertices of our tournament that are not in C.

4
Case 1. Suppose there is a vertex w outside C which has arcs both to and from vertices in C, as
in the diagram on the left.

v1 v2 v3 v4 v5 v1 v2 v3 v4 v5

w w

Then there must be a place along the cycle where the direction of the arcs switches: vi has an arc
to w, but vi+1 has an arc from w. (Maybe the switch happens from vk to v1 instead, but this is
not substantially different.)
Then we can insert w into the cycle between vi and vi+1 : (v1 , v2 , . . . , vi , w, vi+1 , . . . , vk , v1 ) is a
bigger cycle. (See the second diagram above.)
Case 2. If case 1 doesn’t happen, then the vertices outside C come in two types. There is a set
A of vertices such that all the arcs between A and C point from C to A. Also, there is a set B of
vertices such that all the arcs between A and C point from B to C. (See the first diagram below.)

v1 v2 v3 v4 v5 v1 v2 v3 v4 v5 v1 v3 v4 v5

A B a b a b

Because the tournament is strongly connected, there must be a walk leaving cycle C to go to a
vertex outside C, but then coming back to C. That walk must go from C to A (it can’t go to
B) but it can only come back to C from B, so eventually it must go from A to B somehow. In
particular, there must be an arc (a, b) with a ∈ A and b ∈ B. (See the second diagram above.)
It is now easy to extend cycle C pretty much any way we like. One way to get a cycle of length
k + 1 (one longer than C) is (v1 , a, b, v3 , . . . , vk , v1 ). (See the third diagram above.)
In both cases, we can extend the cycle to make it longer. Eventually, we’ll get a Hamiltonian
cycle.

In the proof above, we went a bit out of our way to go from a cycle of length k to a cycle of length
k + 1. (We could have gone from length k to length k + 2 in case 2; do you see how?) But our
reward for that is the following:
Corollary 2.4 (Moon). Every strongly connected n-vertex tournament has a cycle of every length
between 3 and n.

Proof. In the proof, we start with a cycle of length 3 and end with a cycle of length n, so we
encounter a cycle of every intermediate length along the way.

5
3 Practice problems
1. Find all the possible tournaments on four vertices, up to isomorphism!
Okay, not so fast.
(a) First, convince yourself that the only possible score sequences for a 4-vertex tournament
are (0, 1, 2, 3), (0, 2, 2, 2), (1, 1, 1, 3), and (1, 1, 2, 2).
(b) For each of these score sequences, draw a tournament with that score sequence—and
convince yourself that there’s only one way you could have done it.
(c) As a bonus, which of these tournaments are strongly connected?
2. Count the following kinds of graphs—not up to isomorphism, but just counting all the different
possibilities.
(a) Tournaments with vertex set v1 , v2 , . . . , vn .
(b) Directed graphs with vertex set v1 , v2 , . . . , vn . (Loops are allowed, but multiple copies
of the same edge are not.)
(c) Oriented graphs: directed graphs with vertex set v1 , v2 , . . . , vn that have no loops,
pairs of parallel edges, or pairs of “anti-parallel edges”: both (vi , vj ) and (vj , vi ) cannot
be present.
3. Find a tournament on 6 vertices which is not a transitive tournament (so it has cycles) but
in which there is no cycle longer than 3 vertices.
(There are multiple solutions.)
4. Prove that if a tournament (not necessarily strongly connected) has a cycle of length k, then
it also has cycles of all lengths between 3 and k.
5. Ramsey’s theorem for tournaments says that for every k, a tournament with 2k vertices
contains a transitive subtournament with k + 1 vertices.
(a) What does this theorem say when k = 1? Prove it when k = 1.
(b) Use your answer to question 1 to verify this theorem for k = 2.
(c) Prove that in every tournament with 2k vertices, there is a vertex with outdegree at
least 2k−1 .
(d) Use part (c) to prove this theorem by induction on k.
6. (a) Let u and v be two vertices of a tournament such that there is no u − v path of length
1 (that is, no arc (u, v)) or of length 2 (no path of the form (u, w, v) for any w).
Prove that deg+ (u) < deg+ (v).
(b) A king in a tournament is a vertex x such that for all other vertices y, there is an x − y
path of length either 1 or 2.
Use part (a) to prove that every tournament has a king and explain how to find one
quickly.

6
 (c) Give an example of a 7-vertex tournament in which every vertex is a king.
7. In this problem, T is a fixed n-vertex tournament, but looking at an arc to see which way
it points takes time. (Maybe you have to solve a difficult computational problem; or maybe
you are actually scheduling games of some kind between the vertices to see who wins.)
(a) Suppose that we know there is a vertex v with deg− (v) = 0: all arcs point out of v,
rather than into v. But we don’t know which vertex v is. (You have names like 1, 2, . . . , n
for the vertices, and one of them is secretly vertex v.)
Explain how to find v by checking as few arcs as possible.
(b) Suppose that we know there is a vertex v with deg− (v) ≤ 1: there is at most one arc
that points into v.
Explain how to find v by checking as few arcs as possible. (There might be multiple
such vertices; in that case, it’s okay if you only find one.)