CSA E0 221: Discrete Structures (November 18, 2014)
Chalk & Talk Session
Instructor: Arpita Patra Submitted by: (Abhijat Sharma and Mayank Tiwari)
Turan’s Theorem
1 Introduction
Extremal graph theory is the branch of graph theory that studies extremal (maximal
or minimal) graphs which satisfy a certain property. Extremality can be taken with respect
to different graph invariants, such as order, size or girth. For example, a simple extremal
graph theory question is ”which acyclic graphs on n vertices have the maximum number of
edges?” The extremal graphs for this question are trees on n vertices, which have n − 1
edges.[3]
Paul Turan, a Hungarian mathematician, worked extensively in the fields of Number
Theory and Graph Theory. He had a long collaboration with fellow Hungarian mathemati-
cian Paul Erdos, lasting 46 years and resulting in 28 joint papers. Erdos wrote of Turan,
”In 1940-1941 he created the area of extremal problems in graph theory which is now one
of the fastest-growing subjects in combinatorics.”. The field is known more briefly today
as extremal graph theory, which is said to have been founded as a result of the following
theorem by Turan.[4]
2 Turan’s Theorem (1941) - Statement
Theorem 1 Among n-vextex simple graphs with no Kr+1 , Tn,r has the maximum number
of edges. Here, Kr+1 refers to the (r + 1)-clique and Tn,r refers to the Turan Graph on n
vertices having r partitions.
This theorem generalises a previous result by Mantel (1907), which states that ”the
maximum number of egdes in an n-vertex triangle-free simple graph is bn2 /4c.” Observe
that Mantel’s theorem is a special case of the Turan’s theorem with r = 2.
3 Motivation: k-Chromatic Graphs
It might be interesting to know which are the smallest and largest k-chromatic graphs with
n vertices.
1-1
What is the minimum size among k-chromatic graphs with n vertices?
Proposition 1 Every k-chromatic graph with n vertices has at-least (k2 ) edges.
Proof A k-chromatic graph has a k-vertex coloring, which can be viewed as a k-partition
of the vertex set, where each partition is an independent set. Suppose we have a proper
k-coloring of a k-chromatic graph. For any pair of colors in the graph, say i and j, there
exists at-least one edge with end points of colors i and j. If such an edge does not exist, then
the vertices of colors i and j could be combined into a single color. As this new coloring
would use fewer colors, this would contradict our assumption, that the graph is k-chromatic
(cannot be colored in fewer than k colors). Since there are (k2 ) distinct pairs of colors, there
must be at-least (k2 ) edges. Note that the equality clearly holds for a complete graph on
k-vertices plus n − k isolated verices.
What is the maximum size among k-chromatic graphs with n vertices?
Suppose we have a proper k-coloring. As long as we can find pairs of non-adjeacent ver-
tices having different colors, we can continue to add edges without increasing the chromatic
number. Thus, to find the maximum possible edges in k-chromatic graphs, we will only
consider graphs without such vertex pairs.
Definition 1 A complete multipartite graph is a simple graph G whose vertices can
be partitioned into sets such that two vertices are adjacent if and only if they are not in the
same partite sets. Equivalently, every component of G is a complete graph. For k ≥ 2, the
complete k-partite graph with partite sets of sizes n1 , n2 , ..., nk is written as Kn1 ,n2 ,...,nk . ♦
Figure 1: [2]
4 Turan Graph
The Turan Graph, denoted Tn,r is the complete r-partite graph with n vertices, whose
partite sets differ in size by at-most 1. By the pigeon-hole principle, every partite set has
size either dn/re or bn/rc.
Lemma 2 Among simple r-partite graphs (that is, r-colorable) with n vertices, the Turan
graph Tn,r is the unique graph with the most edges.
1-2
Figure 2: [3]
Proof As discussed above, we can add edges without increasing the chromatic number
until it becomes a complete multipartite graph. Now, given a complete r-partite graph with
partite sets differing by more than 1 in size, we can move a vertex v from the largest partite
set (size i) to the smallest partite set (size j). The edges not involving v remain the same
as before, but v gains i − 1 neighbours in its old partite set, and loses j neighbours in its
new partite set. Since i − 1 > j, the number of edges increases due to this switch. Hence,
we maximize the number of edges only by equalizing the sizes of all partite sets, as in Tn,r .
What happens if we wish to add more edges than in Tn,r ? Does it force the chromatic
number to be r + 1? We have seen (Mantel 1907) that there are graphs with chromatic
number 2, that have no triangles. But if we have edges more than bn2 /4c on an n-vertex
graph, then we are forced not only to use 3 colors, but also to have K3 (triangle) as a sub-
graph. Turan generalised this as follows: For an r-colorable graph with n vertices, if we go
beyond the maximum no. of edges, then we are forced not only to use r + 1 colors, but also
to have Kr+1 (i.e r + 1-clique) as a subgraph.
5 Turan’s Theorem (proof )
Theorem 3 Among the n-vertex simple graphs with no r +1-clique, Tn,r , has the maximum
number of edges.
Proof Every r-colorable(or r-partite) graph, including Turan graph Tn,r , has no r + 1-
clique, since each partite set contributes at-most one vertex to each clique. If we can prove
that the maximum edges is achieved by an r-partite graph, then Lemma 2 implies that
the required graph is Tn.r . Thus, it suffices to prove that for every graph G that has no
r + 1-clique, there is an r-partite graph H with the same vertex set as G i.e V (H) = V (G),
and at-least as many edges i.e e(H) ≥ e(G).
1-3
We prove this by induction on r.
For the base case r = 1, any simple graph with no 2-clique is a null-graph (graph with no
edges), and is trivially 1-partite. Thus, in this case, H = G.
For the induction step, G is an n-vertex simple graph with no r + 1-clique, where r > 1. Let
x ∈ V (G) be a vertex of degree k = 4(G). Let the sub-graph G0 be the induced sub-graph
of G by the set N (x), where N (x) is the set of neighbours of x.
Claim 4 If G has no r + 1-clique, then G0 has no r-clique.
As x is adjacent to every vertex in G0 , if G0 had an r-clique then G would have an r+1-clique,
which would be a contradiction.
Thus, we can apply the induction hypothesis to G0 . Thus, there exists an r − 1-partite
graph H 0 with V (H 0 ) = V (G0 ) = N (x) and e(H 0 ) ≥ e(G0 ). Note that V (H 0 ) = N (x) = k.
Let H be the graph formed from H 0 by joining all of N (x) to all of S = V (G) − N (x). Since
S is an independent set of n − k vertices and H 0 is r − 1-partite, thus H would be r-partite.
Claim 5 e(H) ≥ e(G)
By construction, e(H) = e(H 0 ) + k(n − k). We also have e(G) ≤ e(G0 ) + v∈S dG (v) as the
P
difference of edges between G and G0 would only consist of those edges that have at-least
one end-point in the set S = V (G) − V (G0 ). Note that the edges with both end-points
P twice. Since 4(G) = k, we have dG (v) ≤ k for each v ∈ S. As
in the set S are counted
| S |= n − k, we have v∈S dG (v) ≤ k(n − k). Therefore, we have
X
e(G) ≤ e(G0 ) + dG (v) ≤ e(G0 ) + k(n − k) ≤ e(H 0 ) + k(n − k) = e(H).
v∈S
Figure 3: [1]
Example: Distant pairs of points[1]
In a circular city of diameter 1, we might want to locate n police cars to maximize
√ the
number of pairs that are far apart, say separated by distance more than d = 1/ 2. If six
cars occupy equally spaced points on the circle, then the only pairs not more than d apart
are the consecutive pairs around the outside: there are nine good pairs. Instead, putting
1-4
√
two cars each near the vertices of an equilateral triangle with side-length 3/2 yields three
bad pairs and twelve good pairs. (This may not be the socially best criterion!) In general,
with dn/3e or bn/3c cars near each vertex of this triangle, the good pairs correspond to
edges of the tripartite Turan graph.
Figure 4: [1]
References
[1] Douglas B. West, 2002. Introduction to Graph Theory - Second Edition, Pearson Edu-
cation, Singapore.
[2] Chad M. Griep - Extremal Graph Theory. [Link] eaton/[Link]
[3] Extremal Graph Theory - Wikipedia. [Link] graph theory
[4] Paul Turan - Wikipedia. [Link] Turn
1-5
