An algorithm for finding the Independence Number of a graph

O.Kettani
email :[email protected]

Abstract

In this paper, we prove that for every connected graph G, there exists a split graph G’ with the
same independence number and the same order . Then we propose a first algorithm for finding
this graph, given the degree sequence of the input graph G. Further, we propose a second
algorithm for finding the independence number of G, given the adjacency matrix of G.

Introduction

Let G=(V,E) be a connected graph on n=│V│ vertices and m=│E│ edges.

A subset X of V is called independent if its vertices are mutually non-adjacent. The
independence number α(G) is the largest cardinality among all independent sets of G.
A split graph is a graph in which the vertices set can be partitioned into a clique and an
independent set.
The problem of finding the independence number α of a graph G is know to be NP- hard[1].
In this paper, we prove that for every connected graph G, there exists a split graph G’ with the
same independence number and the same order . Then we propose an algorithm for finding this
graph, given the degree sequence of the input graph G. Further, we propose a second algorithm
for finding the independence number of G, given the adjacency matrix of G.
First we will prove the following proposition :

Proposition 1 :
For every connected graph G, there exists a split graph G’ with the same independence
number α and the same order n. The degree sequence of G’ contains (n- α) times the degree
(n-1) and α times the degree (n-α).

1
Proof :
Let G=(V,E) be a connected graph on n=│V│ vertices and independence number α.
Then ∃I ⊂V, I independent set, with α =│I│and ∃VC ⊂V (VC a vertex cover) such
V=I∪VC and I∩VC= ∅.
Considere the graph G’ constructed from G by the following rules:
1) For every vertex v in I, and for every vertex w in VC non-adjacent to v create an edge
between v and w.
2) If │VC│≥2 then for every non-adjacent vertices v and w in VC, create an edge
between v and w.

Then clearly the resulting graph G’ is such that V=V’ and I =I’ and VC=VC’
and I’∩VC’= ∅.
Since the subgraph induced by VC’ is a clique then G’ is a split graph and ,
│VC│=n- α and v∈VC ⇒ d(v)= α+(n- α)-1=n-1
│I│= α and v∈I ⇒ d(v)= n- α
Thus, the degree sequence of G’ contains (n- α) times the degree (n-1) and α times the
degree (n-α).
Let G α,n- α denote the split graph G’

Example : n=6 and G is the following graph :

the resulting graph G’ is G 3,3:

2
For n=3, G 2,1 is the the following graph :

And is G 1,2 is the triangle K3

For n=4, G 3,1 is the the following graph :

G 2,2

G 1,4

For n=5, G 4,1 is the the following graph :

3
G 3,2

G 2,3

G 1,4 is K5

For n=6, G 5,1 is

G 4,2 is

4
G 3,3 is

G 2,4 is

And G 1,5 is K6
and so on…

Definition 1 :
Given two degree sequence S = (d1 ≥d2≥... ≥dn ) and S’ = (d’1 ≥d’2≥... ≥d’n)
We denote S’≥S iff d’i ≥di for all i.
And if S’≥S, we define S’-S= (d’1 –d1,... ,d’n –dn )
Definition 2 :
A degree sequence S is said realizable (or graphic) iff it represents a simple graph.
Definition 3 :
Let G=(V,E) be a connected graph and G’=(V,E’) be an other connected graph such
E⊂E’.
Then we define the difference graph G’’, by G’’=(V,E’-E).
Then S(G’’)= S(G’)- S(G).

5
Before establishing the main result of the present paper, we will use the following
deterministic procedure HH (S) due to Havel [3] and Hakimi [4], which can be used to decide
whether a degree sequence is or is not realizable (i.e.,represents a simple graph or not). It is
based on the Erd¨os-Gallai theorem [2] which states that given a degree sequence
S = (d1 ≥d2≥... ≥dn), S is realizable if and only if the following equation holds :
k n
∑ di ≤ k(k − 1) + ∑ min(k, di)
i=1 i=k+1
The Havel-Hakimi procedure tests the degree sequence S as follows.
procedure HH(S)
Input S =( d1 ≥d2≥... ≥dn),
1 If there exists an integer d in S such that d > n−1 then halt and output false. That is, we
cannot have a vertex that is adjacent to more than n − 1 other vertices.
2 If there are an odd number of odd numbers in S halt and and output false. That is, there must
be an even number of vertices of odd degree.
3 If the sequence contains a negative number then halt and and output false.
4 If the sequence is all zeros then halt and output True
5 Reorder S such that it is nonincreasing.
6 Delete the first term d1 from S and subtract one from the next d1 terms to form a new
sequence.
Go to step 3 .

Proposition 2 :
Let G=(V,E) be a connected graph, then there exists a unique split graph
G α,n- α such that S(G α,n- α)≥ S(G) and the degree sequence S(G α,n- α)- S(G) is realizable.

Proof :
The existence of the split graph G α,n- α is proved by proposition 1.
Since E(G)⊂ E(G α,n- α) then S(G α,n- α)≥ S(G) and S(G α,n- α)-S(G) is realizable because it
represents the difference graph G’’=(V, E(G α,n- α)- E(G )).
Suppose on the contrary that there exists another split graph Gα’,n- α’ such that
S(Gα’,n- α’)≥ S(G) and the degree sequence S(Gα’,n- α’)- S(G) is realizable.

6
 Then by lemma 1, α = α(G)=α(Gα’,n- α’)= α’ :a contradiction. Consequently, there exists a
unique split graph G α,n- α such that S(G α,n- α)≥ S(G) and the degree sequence
S(G α,n- α)- S(G) is realizable.

Now, we will prove the first main result of this paper :

Proposition 3 :
Given a connected graph G=(V,E) on n=│V│ vertices and m edges with degree sequence
S(G) = (d1 ≥d2≥... ≥dn ), then its independence number α could be computed by the following
algorithm :
Input : S(G) = (d1 ≥d2≥... ≥dn ).
1α←1
2 Repeat
if S((G α,n- α))≥S(G) and HH(S(G α,n- α)- S(G)) then output(α) halt
Else α←α+1 goto step 2
End.

Proof :
By using proposition 1 and 2, the algorithm finds the unique « closest » split graph G α,n- α
and then its independence number α which is also the independence number of G .

Example : n=6 and G is the following graph :

S(G) = (2,2,2,2,1,1 ).
S((G 1,5)) = (5,5,5,5,5,5)
S((G 1,5)) ≥ S(G) but HH(S(G α,n- α)- S(G))=HH((3,3,3,3,4,4) is false.
S((G 2,4)) = (5,5,5,5,4,4) but HH(S(G α,n- α)- S(G))=HH((3,3,3,3,3,3) is false.
S((G 3,3)) = (5,5,5,3,3,3) and HH(S(G α,n- α)- S(G))=HH((3,3,3,1,2,2) is true.
Then the algorithm outputs α =3

7
Let n be an integer ≥3 . Let θ be the nxn null matrix.
For k such 1≤k≤n-1, we define the triangular matrices Tk by :
(Tk+1)i,j=1 if 1≤i≤k and 2≤j≤k+1 and i<j
(Tk+1)i,j=0 otherwise.

8
Proposition 4 :
For k such1≤k≤n-1, (Tk+1)k+1=θ
And (Tk+1)h ≠ θ if h is such 1≤h≤k.

Proof :

For k such 1≤k≤n-1, let Ck+1 denote the matrice defined by:
(Ck+1)i,j=1 if j=k+1 and 1≤i≤k
(Ck+1)i,j=0 otherwise.

For k such 1≤k≤n-1, we define the set of nxn matrices M(k) by :

M∈ M(k) iff:
(M)i,j integer>0 for j=k+1 and 1≤i≤k
(M)i,j=0 otherwise..

Then :
For k such 2≤k≤n-1, we have:
Tk+1= Tk +Ck+1

(Tk+1)2=(Tk)2+ TkCk+1 +Ck+1Tk +(Ck+1)2

9
since (Ck+1)2=θ and Ck+1Tk=θ and TkCk+1∈ M(k-1)

then
(Tk+1)2=(Tk)2+ TkCk+1 and TkCk+1∈ M(k-1)

(Tk+1)3=(Tk)3+ Tk(TkCk+1) since TkCk+1∈ M(k-1)⇒ Tk(TkCk+1) ∈ M(k-2)

then :
(Tk+1)3=(Tk)3+ (Tk)2Ck+1 and (Tk)2Ck+1∈ M(k-2)
Suppose by induction on p<k-1 that:
(Tk+1)p=(Tk)p+ (Tk)p-1Ck+1 and (Tk)p-1Ck+1∈ M(k-p+1)
then
(Tk+1)p+1=(Tk)p+1+ Tk(Tk)p-1 Ck+1 and Tk(Tk)p-1Ck+1∈ M(k-p)
Since the induction is true for p=1
Then for p such k>p≥1 we have :
(Tk+1)p+1=(Tk)p+1+(Tk)pCk+1 and (Tk)pCk+1∈ M(k-p)

Then for p=k-1 we have :

(Tk+1)k=(Tk)k+(Tk)k-1Ck+1 and (Tk)k-1Ck+1∈ M(1)

since (Tk)k-1Ck+1∈ M(1)⇒ Tk(Tk)k-1Ck+1=θ

Then
(Tk+1)k+1=(Tk)k+1+(Tk)kCk+1
Then
(Tk+1)k+1=(Tk)k+1 (1)
(T2)2=θ

10
Suppose by induction on k that: (Tk)k=θ then by equation (1) we have:
(Tk+1)k+1=θ
On the other hand, for p such 1≤p≤k:
(Tk+1)p=(Tk)p+(Tk)p-1Ck+1 and (Tk)p-1Ck+1∈ M(k-p+1)
since 1≤p≤k ⇒ k-p+1≥1 ⇒(Tk)p-1Ck+1≠θ and (Tk+1)p≠θ for p such 1≤p≤k.
Then the induction is proved.

For k such 1≤k≤n-2, we define the matrices B’k+1 by :

(B’k+1)i,j=1 if 1≤ i≤k and k+2≤j≤n
(B’k+1)i,j=0 otherwise.

Let Sk+1 be the matrix :

Sk+1= Tk+1+ Bk+1

Proposition 5 :
For k such 1≤k≤n-1, we have (Sk+1)k+1=θ
And (Sk+1)h ≠ θ if h is such 1≤h≤k.
Proof :
(Sk+1)2= (Tk+1)2+ (Bk+1)2+ Bk+1Tk+1 +Tk+1Bk+1
since (Bk+1)2=θ and Bk+1Tk+1=θ and Tk+1Bk+1=θ then (Sk+1)2= (Tk+1)2

11
Suppose by induction on p that :
(Sk+1)p= (Tk+1)p
then (Sk+1)p+1=(Tk+1)p(Tk+1+ Bk+1)= (Tk+1)p+1 + (Tk+1)p Bk+1
Tk+1Bk+1=θ⇒(Tk+1)p Bk+1=θ
then
(Sk+1)p+1=(Tk+1)p+1
By proposition 4, it implies for k such 1≤k≤n-1, that :
(Sk+1)k+1=θ and
(Sk+1)h ≠ θ if h is such 1≤h≤k.

Similarly, for k such 1≤k≤n-2, we define the matrices B’k+1 by :

(B’k+1)i,j ∈{0,1} if 1≤ i≤k and k+2≤j≤n
(B’k+1)i,j=0 otherwise.

And we define the matrices S’k+1 by:

Definition 4 :
S’k+1= T’k+1+ B’k+1

Proposition 6:
For k such 1≤k≤n-1, we have (S’k+1)k+1=θ
and (S’k+1)h ≠ θ if h is such1≤h≤k.

Proof :
Analogue to proposition 5’proof.

Proposition 7:
Let G be a graph of order n , and UG denote the strictly upper triangular matrix extracted from
the adjacency matrix of G, then
α(G)= n-min{h such (UG)h+1=θ}
0≤h≤n

12
Proof :
By proposition 1, there exists a split graph S such V(G)=V(S) and E(G)⊂E(S) and
α(G)= α(S)=n-k.
Then UG is congruent to a matrix S’k+1 as defined in definition 4.
Then there exists a permutation matrix P and PT its transpose matrix such :
UG = PTS’k+1P
then
(UG)k+1= PT(S’k+1)k+1P
Then :
(UG)k+1= PTθP
And
(UG)h= PT(S’k+1)h P≠θ for h such 1≤h≤k by proposition 6.
Then k= min{h such (UG)h+1=θ}
0≤h≤n
And
α(G)= α(S)=n-k=n- min{h such (UG)h+1=θ}
0≤h≤n

Proposition 8 :
Given a graph G=(V,E) on n=│V│ vertices and let UG denote the strictly upper triangular
matrix extracted from the adjacency matrix of G and I the nxn identity matrix, then the
independence number α(G) could be computed by the following algorithm :
Input : UG
1 k←0 and T←I
2 T←TUG
3 if T=θ then output(n-k) else k←k+1 goto step 2.

Example 1 :

n=5 and G is the complete graph K5

13
(UG)5=θ then α(G)=5-4=1.

Example 2 :

n=5 and G is the following graph :

Then

(UG)3=θ then then α(G)=5-2=3.

Example 3 :
n=5 and G is the graph 5 K1 (five isolated vertices)
UG=θ Then then α(G)=5-0=5.

14
References :

[1]. M. R. Garey, D. S. Johnson, "Computers and intractability. A guide to the theory of NP-
completness".1979
[2] P. Erdos and T. Gallai. Graphs with Prescribed Degrees of Vertices. Mat. Lapok, 11:264–
274, 1960.
[3] S. L. Hakimi. On the realizability of a set of integers as degrees of the vertices of a linear
graph.
SIAM Journal, 10(3):496–506, 1962.
[4] V. Havel. A remark on the existence of finite graphs. Caposis Pest. Mat., 80:496–50

15