Cliques and Independent Sets

prepared and instructed by

Shmuel Wimer
Eng. Faculty, Bar-Ilan University

March 2020 Cliques and Independent Sets 1

 Turán’s Theorem
 
Seldom two enemies have another common enemy.
Soon two will ally to beat the third.
Given entities, what is the maximum number of pairs of
enemies sharing no common enemy?
 
Let be a graph, where an edge exists iff its two end
vertices are enemies.
The problem turns to finding the maximum number of
edges in such that it is -free (triangle free).

March 2020 Cliques and Independent Sets 2

 
Proposition: (Mantel 1907) The maximum number of
edges in an -vertex -free graph is .
 
Proof: Let be -free, and , (max degree).
Since has no triangles, there are no edges in
(neighborhood). The sum of the degrees of and all
counts every at least once.
Hence .

 
The complete bipartite graph is -free and has edges,
thus achieving the bound.
March 2020 Cliques and Independent Sets 3
 
Definition: A complete multipartite graph is a graph
whose vertices can be partitioned into sets so that is an
edge iff and belong to different sets.
Equivalently, ’s (complementary) components are
complete graphs.
 
We write where ,…, are the partite sizes and ’s
components are .

𝑛  1 𝑛1 ×𝑛2
  𝑛  2
𝑛1 ×𝑛3
  𝑛2 ×𝑛3
 
  𝐾 𝑛 , 𝑛 ,𝑛
𝑛  3
1 2 3

March 2020 Cliques and Independent Sets 4

 
Let . When ,…, are such that , , we call Turán graph .
Note that is free.

 
Theorem: (Turán 1941) Let be an -vertex simple -free
graph. Then with equality iff (isomorphic).

 
Proof: (Zykov 1949) Use induction on . When there are
no edges. Let and be -vertex and -free.

 
Choose such that . Set and .

March 2020 Cliques and Independent Sets 5

 𝑥
 
 
𝑘 𝑋
 
𝑌
 
𝐺 
 
There is .
 
Since is -free (induced subgraph) is -free, as otherwise .

 
Therefore, by induction with equality iff .

March 2020 Cliques and Independent Sets 6

 
excludes s vertices and the maximum degree of a ’s
vertex is .
Also, a ’s incident edge belongs either to or (linking
edges), hence .
Equality exists iff is independent set with all edge
degrees . (A ’s internal edge cancels two of ).

 
Define by connecting all the vertices of to all the
vertices of an independent set.

March 2020 Cliques and Independent Sets 7

 independent   replace

𝑛 −𝑘  𝑘 × ( 𝑛 − 𝑘 )
  𝑇 𝑟 −1 ,𝑘
 

𝐻
 

 
is -free ⇒ is an -vertex complete -partite graph -free.
⇒

 
We show that among all -vertex complete -partite
graphs has the maximum number of edges.

March 2020 Cliques and Independent Sets 8

 
Otherwise, there exists , , and some , .
Move a vertex from to . The number of edges in is
increased by .
Hence with equality iff .

March 2020 Cliques and Independent Sets 9

 Application to Combinatorial Geometry
 
Given a town of diameter 1, we would like to place 6
police cars so as to maximize the number of car pairs
that are far apart, .

9 good pair, 6 bad pairs. 12 good pair, 3 bad pairs.

March 2020 Cliques and Independent Sets 10

 
Theorem: Let points be placed in a -diameter circle. The
maximum number of pairs such that is .

 
Proof: Create an -vertex graph in the -diameter circle
where an edge exists iff .
The former construction is , which by Turán’s theorem is
the largest excluding . Let us show that is -free.
Suppose contains comprising the points , constituting a
quadrilateral.

March 2020 Cliques and Independent Sets 11

 
For convex quadrilateral, at
𝑤 
least one angle 90○, say . Since
𝐾4
 

and , there is , impossible. 𝑥

  𝑧
 

𝑦
 

𝑤 
𝐾4
 

𝑥
 
 
Otherwise, let 180○. 𝑦
 

𝑧
 

 
The length of the edges is and one of the angles around
120 ○, yielding an opposite edge length , impossible.

March 2020 Cliques and Independent Sets 12

 Ramsey's Theorem
 
Clique in independent set in .
If has no large cliques, it is reasonable that has.
 
Example: Among persons there are always mutual
acquaintances or mutual non-acquaintances.
Proof: The degrees sum of a vertex in and is , hence
either or .
By symmetry let . Let . If any of edges exists in , it
contains a triangle. Else, a triangle exists in .

March 2020 Cliques and Independent Sets 13

 
Ramsey (1930): For any positive integers and there
exists a smallest number such that every -vertex graph
contains either or contains .
are called Ramsey numbers.
 
Ramsey numbers are very difficult to determine.
, and .
The former example showed that . Since (cycle graph)
neither contains nor , there is .
Consequently .

March 2020 Cliques and Independent Sets 14

 
Theorem: For any two integers and
,
with strict inequality if and are even.

 
Proof: Let be an -vertex graph and let .
Let (adjacent to ) and (nonadjacent to ) .
Since , one and only one of the following holds:
1. 2. .

March 2020 Cliques and Independent Sets 15

 
1. : contains either or . Therefore contains either or .

 
2. : contains either or . Therefore contains either or .

 
Because one of 1 and 2 must hold, contains either or ,
⇒.
 
For the second part, let and be even, and have
vertices.

March 2020 Cliques and Independent Sets 16

 
Because is odd, and the total sum of vertex degrees in a
graph is even (why?), there is a with even degree.
In particular .

 
Consequently one and only one of the following holds:
1. 2. .
As in the former proof ⇒ contains either or ⇒ .

The determination of Ramsey number in general is a

very difficult unsolved problem.

March 2020 Cliques and Independent Sets 17

 
Example: Determination of revisited.
.
.
neither contains nor , hence . Consequently .

 
Example: Showing .
 
The graph neither contains nor ,
hence .
 
and are even, hence .

March 2020 Cliques and Independent Sets 18

 Bounds on Ramsey Numbers
 
Theorem: For all positive integers and
.

 
Proof: By induction on . It holds for .
. .
Assume by induction it holds for all .

March 2020 Cliques and Independent Sets 19

 
Corollary: For all positive integers and , , with equality
iff .
 
Proof: is the number of -subsets of a -set, whereas is
the total number of subsets.

 
is called diagonal Ramsey number. The above shows
that it grows at most exponentially. It is also bounded
below exponentially.
 
Theorem: (Erdős 1947) For all positive integer , .

March 2020 Cliques and Independent Sets 20

 
Proof: Because and , we may assume that .
Denote by the set of simple graphs with vertex set .
Because there is a total of pairs, .
Let be these having .
Fixing a particular subgraph excludes edges, hence the
number of graphs in having a particular is .

March 2020 Cliques and Independent Sets 21

 
There are distinct s, hence .
Inequality stems from ’s graphs with more than one ,
counted more than once in .

 
There is therefore .
Suppose .
Then .

March 2020 Cliques and Independent Sets 22

 
In other words, if , fewer than half of contain .
By complementarity, also fewer than half of contain .
Consequently, has graphs containing neither nor .
This holds for any , .

March 2020 Cliques and Independent Sets 23

 Shannon Capacity
 
A message (sequence of words of chars) consisting of
signals belonging to a certain finite alphabet is
transmitted over a noisy channel.

Some pairs of chars are so similar that they can be

confounded by the receiver due to noise.

What is the largest number of distinct words that can be

used in messages without a confusion at the receiver?

March 2020 Cliques and Independent Sets 24

 
Example: and .
If the noise results errors of type and (mod ), the
message can be safely transmitted, because distinct
words are guaranteed under all errors, and erroneous
words are distinctive.
 
Definition. The strong product of two graphs and is
defined by .
Two vertices and are adjacent (in ) iff and , or and , or
and .

March 2020 Cliques and Independent Sets 25

 0
𝐶5
  𝐶5
 
4
0 0
3
4 1 4 1
⊗  2
3 2 3 2 1

0
0 1 2 3 4 0

The strong product is embedded on a torus.

All 0 vertices are identical, mapped to unique point on
the torus.
March 2020 Cliques and Independent Sets 26
 
Let be the graph with vertex set . An edge is defined if
and represent signals that might be confused with each
other.
 
is a strong product of copies of , representing words of
length over .
 
What are the edges of ?
Two words and are connected with an edge iff or for
all .
Edges of correspond to words that might be confused
with each other.

March 2020 Cliques and Independent Sets 27

 
The largest number of distinct words equals the size
of maximum independent set .

  3
In this example and .
2

0
0 1 2 3 4 0

March 2020 Cliques and Independent Sets 28

 
Shannon proposed in information theory (1956) the
parameter
,
known as Shannon capacity of , as a measure of the
capacity for error-free transmission over a noise channel
whose associated graph is .

March 2020 Cliques and Independent Sets 29