Discrete Mathematics 310 (2010) 1429–1438

Contents lists available at ScienceDirect

Discrete Mathematics
journal homepage: www.elsevier.com/locate/disc

Some results on Reed’s Conjecture about ω, ∆, and χ with respect to α

Anja Kohl a,∗ , Ingo Schiermeyer b
a
Institute of Applied Analysis, TU Bergakademie Freiberg, 09596 Freiberg, Germany
b
Institute of Discrete Mathematics and Algebra, TU Bergakademie Freiberg, 09596 Freiberg, Germany

article info abstract

Article history: Let G be a graph of order n with chromatic number χ , maximum degree ∆, clique number
Received 3 November 2008 ω and independence number α . In 1998 Reed conjectured that χ is bounded from above by
Received in revised form 18 May 2009 d ∆+ω+
2
1
e. We will present some partial solutions for this conjecture with respect to α . For
Accepted 20 May 2009
instance, we will verify Reed’s Conjecture for graphs with independence number α = 2,
Available online 12 June 2009
for graphs with maximum degree ∆ ≥ n − α − 4, and for triangle-free graphs having
8(n−α)+118
maximum degree ∆ ≥ 21
. In addition, we will prove the general upper bound
Keywords:
∆+ω+1
Chromatic number χ ≤ 13 (n − α + ω + 2 + 2
).
Coloring © 2009 Elsevier B.V. All rights reserved.
Reed’s Conjecture

1. Introduction

We refer to [4] for terminology and notation not defined here and consider in this paper only finite, simple, connected
and undirected graphs.
The chromatic number of a graph G, denoted by χ (G), is the minimum number of colors required to color the vertices of
G such that adjacent vertices receive different colors. Since the computation of χ (G) is NP-hard we are interested in lower
and upper bounds.
Bounding χ (G) in terms of other graph invariants such as the clique number ω(G) and the maximum degree ∆(G) has
a long tradition. A well-known proposition is that any graph G satisfies ω(G) ≤ χ (G) ≤ ∆(G) + 1. A first improvement of
the upper bound was done by Brooks [2] in 1941. His famous theorem states that if a connected graph G is neither an odd
cycle nor a complete graph, then χ (G) ≤ ∆(G). With respect to the clique number this theorem says that a graph G with
∆(G) ≥ 3 and ω(G) ≤ ∆(G) fulfills χ (G) ≤ ∆(G). Borodin and Kostochka, Catlin, and Lawrence (cf. [6]) independently
proved χ (G) ≤ r +r 1 (∆(G) + 2) for all graphs G with ω(G) ≤ r and ∆(G) ≥ r ≥ 3. Kostochka (cf. [6]) supplemented
χ(G) ≤ 23 ∆(G) + 2 for graphs G with ω(G) = 2.
In 1998 Reed [12] conjectured that χ (G) is bounded from above by the average of the lower bound ω(G) and the upper
bound ∆(G) + 1:
∆(G)+ω(G)+1
Conjecture 1 ([12]). For every graph G, χ (G) ≤ d 2
e.
This conjecture was answered in the affirmative for special graphs such as line graphs of multigraphs [7] and quasi-line
graphs (see [8] for definition and proof). Moreover, there are the following results concerning graphs with sufficiently large
maximum degree:
∆(G)+ω(G)+1
Theorem 1 ([12]). There is a constant ∆0 , such that χ (G) ≤ 2
for all graphs G with ∆(G) ≥ ∆0 and ω(G) ≥
b(1 − 1
70000000
)∆(G)c.

∗ Corresponding author.
E-mail addresses: [email protected] (A. Kohl), [email protected] (I. Schiermeyer).

0012-365X/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.disc.2009.05.025
1430 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438

∆(G)+ω(G)+1
Theorem 2 ([11]). For every k ≥ 3 there is a constant ck such that χ(G) ≤ 2
for all graphs G with ∆(G) ≥
2|V (G)|
k
+ ck · (ω(G))k−1 .
Reed [12] also proved that Conjecture 1 holds for any graph G of maximum degree |V (G)|−1. Randerath and Schiermeyer [11]
improved this statement and verified the conjecture for all graphs G satisfying ∆(G) ≥ |V (G)| − 4.
Rabern [9] presented two results with respect to the order of the graph:
∆(G)+ω(G)+1
Theorem 3 ([9]). A graph G of order n satisfies χ (G) ≤ d 2
e if
n−α(G)+3
(a) G has chromatic number χ (G) > 2
or
(b) the complement G of G has matching number ν(G) < b 2n c.
In this paper we present results on Reed’s Conjecture with respect to the independence number. Most of these results
are based on a partition algorithm and structural results that we establish in Section 2. In Section 3 we present our main
results. First we verify Conjecture 1 for graphs with independence number 2 and more generally for almost-split graphs.
By an almost-split graph we mean a graph that can be partitioned into a maximum independent set and a graph having
independence number at most 2. After this we prove the general upper bound χ (G) ≤ 31 (|V (G)| − α(G) + ω(G) +
∆(G)+ω(G)+1
2+ 2
) and compare it with other bounds. Moreover we verify Conjecture 1 for graphs with maximum degree
∆(G) ≥ |V (G)| − α(G) − 4. This implies that Reed’s Conjecture is valid if ∆(G) ≥ |V (G)| − 7. At last we focus on triangle-
8(|V (G)|−α(G))+118
free graphs and verify Conjecture 1 for such graphs G having ∆(G) ≥ 21
.

2. Preliminaries and notation

In the following let G = (V , E ) be a graph of order n, with maximum degree ∆, minimum degree δ , clique number ω,
chromatic number χ and independence number α .
Since Reed’s Conjecture is obviously fulfilled if ω = 1 or α = 1, and also if ∆ ≤ ω + 2 (Brooks’ Theorem), we consider
in this paper only graphs G satisfying α, ω ≥ 2 and ∆ ≥ ω + 3 ≥ 5.
For a vertex v ∈ V , N (v) is the open neighborhood of v , i.e. the set of vertices which are adjacent to v . For the closed
neighborhood N (v) ∪ {v} we write N [v]. The degree of v is the number |N (v)| of adjacent vertices and is denoted by d(v)
or dG (v). For a subset V 0 ⊆ V we denote the subgraph induced by V 0 with G[V 0 ], and the subgraph induced by V \ V 0 with
G[V \ V 0 ] or G − V 0 . A k-coloring is a vertex coloring with exactly k colors. Moreover, we use the abbreviations k-IS and
k-clique for an independent set and a clique of cardinality k, respectively.
We partition the vertex set V as follows:

Algorithm 1. • Determine a maximum independent set I in G.

If α = |I | ≥ 3 set G1 := G − I, else G1 := G.
• Determine a maximum set of vertex-disjoint triangles in G1 .
Denote the corresponding 3-IS’s in G1 by P1 , . . . , Pp .
Sp
Set G2 := G1 − i=1 Pi .
• Determine a maximum matching in G2 .
Denote the corresponding 2-IS’s in G2 by Q1 , . . . , Qq .
Sq
Set G3 := G2 − i=1 Qi .
Let Vi := V (Gi ) for i = 1, 2, 3. Obviously, G3 is a clique. We denote |V3 | by r and the r corresponding 1-IS’s by R1 , . . . , Rr .
If α ≥ 3, then the cardinality of V satisfies
n = α + 3p + 2q + r . (1)
Moreover, with the given partition of the vertex set into 1 + p + q + r independent subsets we can bound χ from above by
χ ≤ 1 + p + q + r. (2)
Fig. 1 shows the partition of the vertex set for the case α ≥ 3.
Sq
Some structural results about G2 . Obviously, G2 has independence number at most 2. Let M := i=1 {{v, v 0 } | v, v 0 ∈ Qi , v 6=
v 0 }. In the following we say that the two vertices v, v 0 ∈ Qi are matching partners. Since M corresponds to a maximum
matching in G2 , the chromatic number of G2 is χ (G2 ) = q + r. Hence, Q1 , . . . , Qq , R1 , . . . , Rr are the color classes of a
χ(G2 )-coloring of G2 and the cardinalities of V2 and V3 are |V2 | = χ (G2 ) + q and |V3 | = χ (G2 ) − q, respectively.

Observation 1. For every vertex v ∈ V2 the subgraph induced by V2 \ N [v] is a clique.

Proof. Suppose there exist x, y ∈ V2 \ N [v] such that {x, y} 6∈ E. Then {v, x, y} is a 3-IS, a contradiction. #
We call v ∈ V \ V3 a big vertex, if v is adjacent to all vertices in V3 .
 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438 1431

Fig. 1. The partition of the vertex set provided by Algorithm 1.

Lemma 1. For i ∈ {1, . . . , q} the set Qi contains a big vertex.

Proof. Suppose there exists an integer i ∈ {1, . . . , q} such that Qi := {x, x0 } contains no big vertex. Hence, ∃v ∈ V3 : {x, v} 6∈
E and ∃w ∈ V3 : {x0 , w} 6∈ E. If v = w , the vertices v, x, x0 form a 3-IS, a contradiction. If v 6= w , then (v, x, x0 , w) is an
M-augmenting path in G2 . This is a contradiction, since M is a maximum matching of G2 .
It follows that x or x0 has to be adjacent to all vertices in V3 . 
Let B be a set of big vertices from V2 such that |B ∩ Qi | = 1 for all i ∈ {1, . . . , q} (B ⊆ V2 \ V3 ). B exists, according to
Lemma 1. We denote by C the set of big vertices from V2 that are not chosen for B (C ⊆ V2 \ (V3 ∪ B)). Moreover, for u ∈ V3
let Du be the set of vertices v ∈ V2 \ V3 for which V3 \ N (v) = {u}.

Lemma 2. Let e = {x, y} ∈ E (G) such that x, y ∈ B and let x0 and y0 be the matching partner of x and y, respectively. Then one
of the following two statements holds:
(i) x0 ∈ C or y0 ∈ C ,
(ii) x0 , y0 ∈ Du for a vertex u ∈ V3 .
Proof. Let Qi = {x, x0 } and Qj = {y, y0 } such that x, y ∈ B and {x, y} 6∈ E. Assume (i) is not true, i.e. x0 , y0 6∈ C . Therefore,
|V3 \ N (x0 )| ≥ 1 and |V3 \ N (y0 )| ≥ 1.
If ∃v ∈ V3 \ N (x0 ) and ∃w ∈ V3 \ N (y0 ) such that v 6= w , then (v, x0 , x, y, y0 , w) is an M-augmenting path in G2 , which is
a contradiction since M is a maximum matching of G2 . Hence, V3 \ N (x0 ) = V3 \ N (y0 ) = {u} for a vertex u ∈ V3 . 

For x, y ∈ B we call an edge e = {x, y} ∈ E (G) fulfilling condition (i) of Lemma 2 a C -edge, otherwise a D-edge, particularly
a Du -edge if the corresponding vertex u ∈ V3 is of interest. Moreover, we denote by Mu the set of matching partners of the
vertices in Du (|Mu | = |Du |, Mu ⊆ B).

Remark 1. If u 6= v , then a Du -edge and a Dv -edge are not adjacent in G2 .

Let VC denote the subset of B consisting of the matching partners of all vertices in C and let Nu denote the subset of
B consisting of all vertices which are incidentSto a Du -edge, respectively.
S Moreover, let VDu be a vertex set of minimum
cardinality that covers all Du -edges. Set N := u∈V3 Nu and VD := u∈V3 VDu . Then we can deduce from Lemma 2:

Proposition 1. VC covers all C -edges and |VC | = |C |.

Proposition 2. VD covers all D-edges. Moreover, if Nu 6= ∅ for a vertex u ∈ V3 , then VDu ⊂ Nu ⊆ Mu and |VDu | < |Nu | ≤
|Mu | = |Du |.

Proposition 3. VC ∪ VD is a vertex cover of G[B]. Therefore, the set B \ (VC ∪ VD ) is an independent set in G[B], i.e. it induces a
clique in G[B].
By Proposition 3, we know that also the set Bu := B \ (VC ∪ VD ∪ Mu ) induces a clique in G. Let Eu denote the set of
matching partners of all vertices in Nu (|Nu | = |Eu |, Eu ⊆ Du ).
Fig. 2 visualizes an extract of the graph G2 together with the main notations. The pictured edges all exist in G2 .

Lemma 3. Let u ∈ V3 . Then {y0 , z } ∈ E holds for all y0 ∈ Eu and z ∈ Bu .

Proof. Suppose there exist vertices y0 ∈ Eu and z ∈ Bu such that {y0 , z } 6∈ E. Since y0 ∈ Eu , vertex y0 has a matching partner
y ∈ Nu . Hence, there exists a second pair of matching partners x, x0 (x, x0 6= z) such that x ∈ Nu , x0 ∈ Eu , and {x, y} is a Du -
edge. Let z 0 be the matching partner of z. Because z is neither in VC nor in Mu , vertex z 0 is neither in C nor in Du . Hence, there
exists a vertex v ∈ V3 \ {u}, such that {v, z 0 } 6∈ E. Then (u, x0 , x, y, y0 , z , z 0 , v) is an M-augmenting path in G2 contradicting
the maximality of M. 
1432 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438

Fig. 2. The main notations for the graph G2 .

Proposition 4. Since Bu induces a clique in G and, by Proposition 1, Eu induces a clique as well it follows from Lemma 3 that
Bu ∪ Eu is a clique for every vertex u ∈ V3 . Moreover, because all vertices in Bu ∪ Eu are pairwise adjacent to all vertices in V3 \ {u},
the vertex set (V3 \ {u}) ∪ Bu ∪ Eu induces a clique in G.

From the definition of Du it follows:

S
Proposition 5. A vertex u ∈ V3 is adjacent to all other vertices in V3 , to all vertices in B ∪ C , and to all vertices in v∈V3 \{u} Dv .

3. Main results

3.1. Graphs with independence number two

∆+ω+1
Theorem 4. Let G be a graph with independence number α = 2. If G has matching number ν(G) < b 2n c then χ (G) ≤ 2
,
 ∆+ω+1 
else χ(G) ≤ 2
.

Proof. We apply Algorithm 1. Obviously, G2 = G1 = G, ν(G) = q, and q ≤ b 2n c. Moreover, χ = q + r since q is the

cardinality of a maximum matching of G.
First assume q = b 2n c, i.e. n ≥ 2χ − 1.
∆−δ+2χ−1
By Proposition 1 we immediately obtain ω ≥ n − (δ + 1). Hence, d ∆+ω+
2
1
e ≥ d ∆−δ+
2
n
e≥d 2
e = χ + d ∆−δ−
2
1
e≥
χ +d −1
2
e = χ.
We now assume q < b 2n c, i.e. n ≤ 2χ − 2 and r = |V3 | ≥ 2.
Choose a vertex s ∈ V3 so that |Ds \ Es | is minimal among all vertices in V3 . According to Proposition 5 (recall that |V3 | = r
and |B| = q),
X
∆ ≥ d(s) ≥ |V3 | − 1 + |B| + |C | + |Dv |
v∈V3 \{s}
X
= χ − 1 + |C | + |Dv |. (3)
v∈V3 \{s}

By Proposition 4, (V3 \ {s}) ∪ Bs ∪ Es induces a clique.

S Hence, ω ≥ |V3 | − 1 + |Bs | + |Es | = |V3 | − 1 + |Es | + |B| − |VC ∪ VD ∪ Ms |.
Since VC and VD ∪ Ms are disjoint and VD ∪ Ms = v∈V3 \{s} VDv ∪ Ms , we obtain:
X
ω ≥ |V3 | − 1 + |B| + |Es | − |VC | − |VDv | − |Ms |
v∈V3 \{s}
X
= χ − 1 − | VC | − |VDv | + |Es | − |Ms |
v∈V3 \{s}
X
= χ − 1 − |C | − |VDv | + |Es | − |Ds |. (4)
v∈V3 \{s}
 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438 1433

By inequalities (3) and (4),

X
∆ + ω + 1 ≥ 2χ − 1 + (|Dv | − |VDv |) − (|Ds | − |Es |)
(a) v∈V3 \{s}
X
≥ 2χ − 1 + (|Dv | − |Ev |) − (|Ds | − |Es |)
(b) v∈V3 \{s}
X
= 2χ − 1 + |Dv \ Ev | − |Ds \ Es |.
v∈V3 \{s}

Ev | ≥ |Ds \ Es | for all v ∈ V3 \ {s} it follows ∆ + ω + 1 ≥ 2χ − 1. Moreover, inequality

Since |V3 | ≥ 2 and because of |Dv \P
(a) is strict if ∆ > χ − 1 + |C | + v∈V3 \{s} |Dv |, and inequality (b) is strict if there exists a vertex u ∈ V3 \ {s} such that
∆+ω+1
|VDu | < |Eu |. In both cases χ ≤ 2
.
Hence, it remains to consider the case where ∆ = χ − 1 + |C | + v∈V3 \{s} |Dv |, and |VDv | = |Ev | for all v ∈ V3 \ {s}. The
P
first condition yields d(v) ≤ d(s) and therefore |Ds | ≤ |Dv | for all v ∈ V3 \ {s}. Since |Ev | = |Nv |, the second condition can
only be fulfilled if Nv = ∅ for all v ∈ V3 \ {s}, i.e. there are no Dv -edges (compare Proposition 2). Then B \ (VC ∪ VDs ) ∪ V3 is
a clique and we can bound ω from below by

ω ≥ |B| − |C | − |VDs | + |V3 | = χ − |C | − |VDs |. (5)

By inequality (5) and the condition for ∆,

∆ + ω + 1 ≥ 2χ + |Dv | − |VDs | ≥ 2χ +
P P
|Ds | − |VDs |.
v∈V3 \{s} v∈V3 \{s}

∆+ω+1
Because of |V3 | ≥ 2 and |Ds | = |Ms | ≥ |VDs | it follows ∆ + ω + 1 ≥ 2χ and therefore χ ≤ 2
. 

Obviously, every graph with independence number 2 is the complement of a triangle-free graph. Gravier and Maffray [3]
proved that the chromatic number χ (G) and the choice number χ` (G) of a graph G are equal if G is the complement of a
triangle-free graph. Therefore:

Corollary 1. If G is a graph with independence number α = 2, then χ` (G) ≤ d ∆+ω+

2
1
e.

Note that Theorem 4 has been independently proved by Rabern [9].

3.2. Almost-split graphs

We call G = (V , E ) an almost-split graph if V can be partitioned into two subsets U1 and U2 such that U1 is an independent
set with α(G) vertices and U2 induces a graph G[U2 ] with independence number α(G[U2 ]) ≤ 2.

 ∆+ω+1 
Theorem 5. Let G be an almost-split graph. Then χ (G) ≤ 2
.

Proof. We apply Algorithm 1 and assume that U1 = I and U2 =V2 , i.e. G

2 = G1 and p = 0.
Assume α(G2 ) = 1. Then G is a split graph and χ (G) = ω ≤ ∆+ω+
2
1
.
∆(G2 )+ω(G2 )+1
Now assume α(G2 ) = 2. Since χ (G) ≤ χ (G2 ) + 1 ≤ d 2
e + 1 (by Theorem 4) and ∆(G2 ) ≤ ∆ − 1, it suffices
∆(G2 )+ω(G2 )+4
to investigate the case χ (G) = χ (G2 ) + 1 = 2
, ∆ = ∆(G2 ) + 1, and ω = ω(G2 ). Because of χ (G2 ) = q + r,

|V2 | = 2χ (G2 ) − r = ∆(G2 ) + ω(G2 ) + 2 − r . (6)

∆(G )+ω(G )+1
By Theorem 4 we know that χ (G2 ) ≤ if q < b 22 c. Since this is not the case, it follows that q = b 22 c and
2 2 |V | |V |
2
r = |V3 | ≤ 1.
By Proposition 1, ω(G2 ) ≥ |V2 | − δ(G2 ) − 1. So we obtain by (6), ∆(G2 ) ≥ δ(G2 ) ≥ ∆(G2 ) + 1 − r. Hence, r = 1 and G2
is regular. Because of ∆ = ∆(G2 ) + 1 and since I is a maximum independent set of G, every vertex v ∈ V2 has exactly one
neighbor in I. Let V3 = {vq+1 } and u ∈ I be the neighbor of vq+1 . Then all vertices in I \ {u} can receive the same color as vq+1 .
Since χ (G) = χ (G2 ) + 1, vertex u has to have a neighbor vi ∈ Qi for every i = 1, . . . , q (otherwise we could immediately
find a χ (G2 )-coloring of G).
Suppose there exist i, j ∈ {1, . . . , q + 1} such that i 6= j and vi and vj are not adjacent. Then (I \ {u}) ∪ {vi , vj } is
an independent set with α + 1 vertices, a contradiction. Hence, the vertices v1 , . . . , vq+1 induce a clique C in G2 . Since
ω(G2 ) ≤ q + 1, C is a maximum clique of G2 , and ω(G2 ) = q + 1. Because u is adjacent to all vertices in C the graph G has a
(q + 2)-clique. This is a contradiction to ω = ω(G2 ). 
1434 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438

3.3. General bounds

In 2006 Schiermeyer [13] proved the following upper bound for the chromatic number that confirms Reed’s Conjecture
for graphs satisfying ∆ ≥ n − α :

Theorem 6 ([13]). If G is a graph of order n with clique number ω and independence number α , then χ (G) ≤ (n − α + ω + 1). 1
2
Moreover, χ (G) ≤ 1
2
(n − α + ω), if either n − α = ω − 1 and G is not a split graph or n − α = ω + 1 and G contains no induced
Kω+3 − C5 .
In case of n − α > ∆ + 2 this bound can be improved as follows:

Theorem 7. If G is a graph of order
n with maximum degree ∆, clique number ω and independence number α ≥ 3, then
χ(G) ≤ 13 n − α + ω + 2 + ∆+ω+ 2
1
.
Proof. We apply Algorithm 1. By (1) and (2),

n − α − 2q − r n − α + q + 2r + 3
χ ≤1+ +q+r = . (7)
3 3
Since G2 is a graph with independence number at most 2 we can use results from Section 3.1.
In particular, we conclude that ∆(G2 ) + ω(G2 ) + 1 ≥ 2χ (G2 ) − 1 = 2(q + r ) − 1. Since any vertex in V2 has to have at
least one neighbor in I, we obtain ∆ ≥ ∆(G2 ) + 1. Hence,
∆ + ω ≥ ∆(G2 ) + 1 + ω(G2 ) ≥ 2(q + r ) − 1. (8)
Using (7) and (8) we can further estimate:

n−α+r +3 q+r n−α+r +3 ∆+ω+1

χ≤ + ≤ + . (9)
3 3 3 6
If r < ω, we obtain χ ≤ n−α+ω+3
2
+ ∆+ω+
6
1
.
If r = ω, then for any vertex v ∈ I there exists a vertex u ∈ V3 such that {u, v} 6∈ E. Therefore, v can receive the same
color as u. Hence, we need for I no extra color, but color all vertices from I using only colors assigned to vertices in V3 . Then,
χ ≤ n−α+ω3
+ ∆+ω+6
1
.
∆+ω+1
In both cases we have χ ≤ 1
n−α+ω+2+


3 2
. 

If ∆ is large compared to n − α and ω, then Theorem 7 improves the upper bound of Brooks:
2(n−α)+3ω
Corollary 2. If G is a graph with maximum degree ∆ > 5
+ 1, then χ (G) < ∆.
Recently, Rabern [9] established another upper bound on χ (G) which is:

n+∆+1
Theorem 8 ([9]). If G is a graph of order n with clique number ω and maximum degree ∆, then χ (G) ≤ 1
ω+


2 2
.
This result can be improved in case of α ≥ 3 as follows:

Proposition 6. If G is a graph of order n with maximum degree ∆, clique number ω and independence number α ≥ 3, then
χ(G) ≤ 12 ω + n−α+2 ∆+4 .


 
n−α+∆(G1 )+1
Proof. By applying Algorithm 1 and Theorem 8 we obtain χ (G) ≤ 1 + χ (G1 ) ≤ 1 + 1
2
ω(G1 ) + 2
≤
 
n−α+∆(G1 )+1
1 + 12 ω + 2
. Since any vertex in V1 is adjacent to at least one vertex in I, ∆(G1 ) + 1 ≤ ∆ holds. This completes
the proof. 

Comparing all four upper bounds we notice that each of the results is of the form χ (G) ≤ ω2 + u
2
where u is a trivial
ω ∆ +2
upper bound on χ (G). This is promising in view of Reed’s conjectured upper bound χ (G) ≤ 2 + 2 that is also a convex
combination of a trivial lower and upper bound.
The decision which of the four results yields the best bound on χ (G) depends on the relation of n, ∆ and α where
Theorem 6 and Proposition 6 outperform the other:

Corollary 3. If G is a graph of order n with maximum degree ∆, clique number ω and independence number α ≥ 3, then
1 1

ω  (n − α + 1) + (∆ + 3),
 if ∆ ≤ n − α − 2
χ (G) − ≤ 4 4
2 1
(n − α + 1), if ∆ ≥ n − α − 2.


2
 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438 1435

3.4. Graphs with large maximum degree

As already mentioned in the introduction there were several efforts to prove Reed’s Conjecture for graphs with
sufficiently large maximum degree. A good progress in that direction was made by Rabern [9] who proved:

Theorem 9 ([9]). Let G be a graph of order n with clique number ω, independence number α and maximum degree ∆ = n − k
for some k ≥ 1. Then χ (G) ≤ ∆+ω+ 1
 
2
if
(a) k ≤ α + ω − √
1 or
(b) k ≤ α − 2 + n − α + 5.
The following theorem supplements the previous statements in case of ω ≤ 4 and n − α ≤ 30.

Theorem 10. Let G be a graph of order  n with1 independence number α ≥ 3 and maximum degree ∆ = n − k for some k ≥ 1.
If G satisfies k ≤ α + 4, then χ (G) ≤ ∆+ω+
2
.
Proof. Since the theorem can be easily verified by Brooks’ Theorem in case of ∆ − 2 ≤ ω ≤ ∆ + 1 it suffices to consider
ω ≤ ∆ − 3.
We apply Algorithm 1. All resulting independent sets are subsumed under the set F , i.e. F := {I , P1 , . . . , Pp , Q1 , . . . , Qq ,
R1 , . . . , Rr }.
If k ≤ α + p + ω − r, then ∆ + ω + 1 ≥ (n − k) + (k − α − p + r ) + 1 = 2(p + q + r + 1) − 1 ≥ 2χ − 1, i.e. χ ≤ ∆+ω+ 1
 
2
.
(1) (2)
If p = 0, then χ ≤ ∆+ω+ 1
 
2
by Theorem 5.
Now assume α + 4 ≥ k ≥ α + p + ω − r + 1 and p ≥ 1. This assumption implies r ∈ {ω, ω − 1, ω − 2}.
Case 1. r = ω (i.e. k ≥ α + p + 1 and p ≤ 3).
If q > 0, then every 2-IS Qi , i ∈ {1, . . . , q}, has to have a big vertex, i.e. a vertex that is adjacent to all vertices in V3
(cf. Lemma 1). This would imply a (ω + 1)-clique, a contradiction. Hence, q = 0 and therefore α(G2 ) = 1. As in the proof of
Theorem 7, the vertices of I can be colored using only colors assigned to vertices from V3 . Hence, χ ≤ p + q + r = p + ω.
If k ≥ α + p + 3, then p = 1, and we would obtain ∆ = n − k ≤ n − (α + p + 3) = 2p + ω − 3 = ω − 1. This is not
possible because of ω ≤ ∆ − 3.
If k ≤ α + p + 2, then ∆ + ω + 1 = n − k + ω + 1 = (α + 3p + ω) − k + ω + 1 = 2(p + ω) − 1 + (α + p + 2 − k) ≥
2(p + ω) − 1 ≥ 2χ − 1.
It follows for r = ω and k ≤ α + 4 that χ ≤ ∆+ω+ 1
 
2
.
Case 2. r = ω − 1 (i.e. k ≥ α + p + 2 and p ≤ 2).
The set of big vertices, denoted by J, is independent (otherwise there would exist a (ω + 1)-clique). Since there are at
least q big vertices in V2 , we conclude q ≤ α(G2 ) ≤ 2.
Subcase 2.1. ∃ S ∈ F ∀ v ∈ S ∃ Sv ∈ F \ S : |N (v) ∩ Sv | = ∅.
Then we need for the vertices in S no extra color but color each vertex v ∈ S with the color given to the independent
set Sv . Hence, χ ≤ p + q + r. This implies ∆ + ω + 1 = n − k + r + 2 = 2(p + q + r ) − 1 + (α + p + 3 − k) ≥
(1)
2χ − 1 + (α + p + 3 − (α + 4)) = 2χ − 1 + (p − 1) ≥ 2χ − 1.
Subcase 2.2. ∀ S ∈ F ∃ v ∈ S ∀ S 0 ∈ F \ S : |N (v) ∩ S 0 | 6= ∅.
This means that every set S ∈ F contains a vertex that has a neighbor in every set S 0 ∈ F \ {S }. We call vertices from
V \ V3 with this property b-vertices. Note that every b-vertex is also a big vertex, and any two b-vertices are nonadjacent.
If p + q ≤ 2, then ∆ = n − k = α + 3p + 2q + ω − 1 − k ≤ α + 3p + 2q + ω − 1 − (α + p + 2) = 2(p + q) + ω − 3 ≤ ω + 1,
a contradiction to ω ≤ ∆ − 3.
If p + q ≥ 3, then q ≥ 1. Because of q ≤ 2, the remaining cases are q = 2, p ∈ {1, 2}, and q = 1, p = 2.
Assume q = 2.
Let Q1 = {x, x0 }, Q2 = {y, y0 } where x, y are b-vertices. Moreover, choose a b-vertex z ∈ I. Since x, y, z are pairwise
nonadjacent, z has to be adjacent to x0 and y0 . Hence, neither x0 nor y0 is a big vertex, such that {x, y} ∈ E (G) has to be a
Du -edge for a vertex u ∈ V3 by Lemma 2. If x0 and y0 are not adjacent, then {x0 , y0 , u} is a 3-IS, a contradiction to α(G2 ) = 2.
If x0 and y0 are adjacent, then {x0 , y0 , z } ∪ (V3 \ {u}) is a (ω + 1)-clique, a contradiction as well. Hence, the case q = 2 is not
possible.
Assume q = 1, p = 2.
It follows from α + 4 ≥ k ≥ α + p + 2 that k = α + 4 and ∆ = ω + 3.
D0u shall denote the set of vertices v ∈ I for which V3 \ N (v) = {u}.
Moreover, let P1 = {v, v 0 , v 00 }, P2 = {w, w 0 , w 00 }, Q1 = {x, x0 } and z ∈ I such that v, w, x, z are b-vertices. Obviously,
v, w, z are adjacent to x0 . Set l := ω − 1 − |N (x0 ) ∩ V3 |. We notice that l ≥ 1 and |N (x0 ) ∩ (V \ V3 )| ≤ ∆ − |N (x0 ) ∩ V3 | =
(ω + 3) − (ω − 1 − l) = 4 + l. So, apart from v, w, z, vertex x0 has at most 1 + l neighbors in V \ V3 . We consider a coloring
f : V3 → {1, . . . , ω − 1} and intend to extend this coloring to a (ω + 2)-coloring of G.
Consider |N (x0 ) ∩ (I \ {z })| < l. Since there are l vertices from V3 which are not adjacent to x0 , there exists a vertex
u ∈ V3 \ N (x0 ) such that x0 has no neighbor in D0u . Hence, we can assign color f (u) to the vertices from D0u ∪ {x0 }. Then we
1436 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438

color the remaining vertices from I \ J by suitable colors assigned to vertices from V3 . We further color all vertices from J
with color ω and the vertices from Pi \ J with ω + i (i = 1, 2). This gives a proper (ω + 2)-coloring of G.
Consider |N (x0 ) ∩ (I \ {z })| ≥ l. Apart from v and w , vertex x0 has at most one neighbor in P1 ∪ P2 . W.l.o.g. let
N (x0 ) ∩ P2 = {w}. We color all vertices from J with ω, the vertices from I \ J with ω + 1, the vertices from P1 \ J with
ω + 2, and the vertices from {x0 } ∪ (P2 \ J ) with suitable colors assigned to vertices in V3 . This gives together with f a proper
(ω + 2)-coloring of G. Hence, χ ≤ ω + 2 ≤ ω2 + ∆− 3
2
+ 2 =∆+ω+ 2
1
.
∆+ω+1
It follows for r = ω − 1 and k ≤ α + 4 that χ ≤ 2
.
Case 3. r = ω − 2 (i.e. k ≥ α + p + 3 and p = 1).
From k ≤ α+ 4 we deduce k = α+ 4. The maximum degree in terms of ω is ∆ = n −(α+ 4) = α+ 3p + 2q + r −(α+ 4) =
2q + ω − 3. If χ (G1 ) ≤ q + r then χ ≤ q + r + 1 = ∆+ω+ 2
1
. Hence, we may assume χ (G1 ) = q + r + 1 = q + ω − 1.
Moreover, we obtain q ≥ 3 by the assumption ω ≤ ∆ − 3.
By the use of Ramsey numbers (cf. [10]) we can further show that ω ≥ 3, since |V2 | ≥ 2q ≥ 6 = r (K3 , K3 ).
Subcase 3.1. ω = 3.
If q ≥ 4, then |V2 | = 2q + 1 ≥ 9 = r (K4 , K3 ), a contradiction. If q = 3, then |V1 | = 10 and χ (G1 ) = q + ω − 1 = 5.
By a result of Jensen and Royle [5] we know that the smallest K4 -free graph with chromatic number 5 has 11 vertices, a
contradiction as well. Hence, the case ω = 3 is not possible.
Subcase 3.2. ω ≥ 4.
The (at least) q big vertices from V2 neither induce a 3-clique nor a 3-IS. Hence, q < r (K3 , K3 ) = 6. Summarizing the
previous considerations we conclude q ∈ {3, 4, 5}.
Obviously, ∆(G1 ) ≤ ∆ − 1 = 2q + ω − 4 and |V1 | = 2q + ω + 1. The graph G2 has |V2 | = 2q + ω − 2 vertices and
maximum degree ∆(G2 ) ≤ ∆(G1 ). If a vertex v ∈ V3 has ∆(G1 ) neighbors in G2 , then P1 ∪ {v} is independent. This would
imply a partition of V1 into q + ω − 2 independent sets, contradicting the assumption χ (G1 ) = q + ω − 1. Hence, we may
assume dG2 (v) ≤ ∆(G1 ) − 1 ≤ 2q + ω − 5 for v ∈ V3 . Moreover, since not all (at least q, q ≥ 3) big vertices are independent,
it is clear that ω(G2 ) = ω.
In the following let u, v ∈ V3 , and Qi := {xi , x0i } for i = 1, . . . , q such that xi is a big vertex.
Suppose q = 5.
Since the number of big vertices in G2 has to be smaller than r (K3 , K3 ) = 6, we notice that there are exactly q = 5 big
vertices in G2 , namely xi , i = 1, . . . , 5. These vertices neither induce a 3-clique nor a 3-IS. Hence, they induce a cycle of
length 5. Without loss of generality let (x1 , x2 , x3 , x4 , x5 , x1 ) be this cycle. Since there is no big vertex in {x0i | i = 1, . . . , 5},
the edges {x1 , x3 }, {x1 , x4 }, {x2 , x4 }, {x2 , x5 }, {x3 , x5 } ∈ E (G2 ) have to be Du -edges for the same vertex u ∈ V3 (compare
Lemma 2 and Remark 1). Hence, x0i ∈ Du for i = 1, . . . , 5 and dG2 (v) ≥ 10 + (ω − 3) = ω + 7 > 2q + ω − 5 = ω + 5. This
is a contradiction.
Suppose q = 3.
Since the big vertices x1 , x2 , x3 do not induce a 3-clique, at least two of them are nonadjacent. Without loss of generality
assume {x1 , x2 } 6∈ E. According to Lemma 2, {x1 , x2 } ∈ E (G2 ) is either a C -edge or a Du -edge for a vertex u ∈ V3 . In the latter
case, x01 , x02 ∈ Du , i.e. x01 and x02 are adjacent to v . This implies dG2 (v) ≥ q + 2 + (ω − 3) = ω + 2 > 2q + ω − 5 = ω + 1,
a contradiction. Hence, {x1 , x2 } is a C -edge, and without loss of generality x01 is a big vertex. It follows from this and from
dG2 (u), dG2 (v) ≤ 2q + ω − 5 = ω + 1, that x02 , x03 are not adjacent to u and v . So, {x2 , x3 } ∈ E (otherwise (u, x02 , x2 , x3 , x03 , v)
would be an M-augmenting path in G2 ). Furthermore, we conclude {x01 , x2 } ∈ E (otherwise {x1 , x01 , x2 } would be a 3-IS)
and {x01 , x3 } 6∈ E (otherwise the big vertices x01 , x2 , x3 would induce a 3-clique). Then, (u, x03 , x3 , x01 , x1 , x2 , x02 , v) is an
M-augmenting path in G2 contradicting the maximality of M.
Suppose q = 4.
At first we notice that we can always partition the set {x1 , x2 , x3 , x4 } into two vertex-disjoint 2-IS. So without loss of
generality assume {x1 , x3 }, {x2 , x4 } 6∈ E. Since the number of big vertices in G2 is smaller than r (K3 , K3 ) = 6 there is at most
one further big vertex in {x0i | i = 1, . . . , 4}. Therefore, at least one of the edges {x1 , x3 }, {x2 , x4 } ∈ E (G2 ), say {x1 , x3 }, has to
be a Du -edge for a vertex u ∈ V3 . Hence, x01 , x03 ∈ Du and they are adjacent to v . By dG2 (v) ≤ ω + 3, vertex v is not adjacent
to x02 and x04 and it follows that {x2 , x4 } has to be a Dv -edge, and x02 , x04 ∈ Dv .
Since there is no 3-IS in G2 we can deduce the existence of the edges {x01 , x3 }, {x1 , x03 }, {x02 , x4 }, {x2 , x04 }, {x01 , x03 }, {x02 , x04 } ∈
E. By arguing with M-augmenting paths in G2 we can prove step-by-step the existence of further edges in G2 . Exemplarily,
we show {x4 , x03 } ∈ E. Assume to the contrary, {x4 , x03 } 6∈ E. Then (v, x04 , x4 , x03 , x3 , x1 , x01 , u) is an M-augmenting path in G2 ,
a contradiction. Hence, {x4 , x03 } ∈ E.
In this manner, we are able to prove that G2 − V3 is a complete graph, apart from the missing edges {x1 , x3 }, {x2 , x4 }, and
{xi , x0i } (i = 1, . . . , 4).
If ω ≥ 5, then |V3 \ {u, v}| ≥ 1, and every vertex w ∈ V3 \ {u, v} satisfies dG2 (w) ≥ 2q + (ω − 3) = ω + 5 > ∆(G1 ), a
contradiction. Hence, ω = 4, and G2 is 7-regular.
Suppose, ∀w ∈ P1 ∃ Sw ∈ F \ {I , P1 } : |N (w) ∩ Sw | = ∅.
Then we could easily color G1 with 6 = q + ω − 2 colors, a contradiction to χ (G1 ) = q + ω − 1.
Suppose, ∃w ∈ P1 ∀ S 0 ∈ F \ {I , P1 } : |N (w) ∩ S 0 | 6= ∅.
Hence, |N (w) ∩ V2 | ≥ 6. Let w 0 , w 00 be the two elements of P1 \ {w} and let yi , i = 1, . . . , 6, be six neighbors of w in G2 .
Obviously, there exist two vertex-disjoint 2-IS in G2 [{y1 , . . . , y6 }]. Without loss of generality assume {y1 , y2 }, {y3 , y4 } 6∈ E.
 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438 1437

Because of ∆(G1 ) ≤ 8 and ∆(G2 ) = δ(G2 ) = 7, we deduce |N (yi ) ∩ P1 | = {w} for i = 1, . . . , 6. Hence, {w 0 , y1 , y2 } and
{w00 , y3 , y4 } are two vertex-disjoint 3-IS in G1 , contradicting the definition of p = 1 as the maximum number of vertex-
disjoint 3-IS in G1 .
Altogether, the case ω ≥ 4 is not possible.
So, we investigated all possible cases for r = ω − 2 and k ≤ α + 4 and ascertained χ ≤ ∆+ω+ 1
 
2
. This completes the
proof. 

 G be1 a graph with independence number α ≥ 3 and maximum degree ∆ = n − k for some k with 1 ≤ k ≤ 7.
Corollary 4. Let
Then χ(G) ≤ ∆+ω+ 2
.

3.5. Graphs with clique number two

At first we mention a result that holds not only for graphs with clique number two (the so-called triangle-free graphs) but
more generally for complements of claw-free graphs. By using Algorithm 1 and another approach we obtain:

Theorem 11. If G is a graph with maximum degree ∆ ≥ 5

6
(n − α) and G is the complement of a claw-free graph, then
 ∆+ω+1 
χ(G) ≤ 2
.
Proof. From Theorems 4 and 10 it follows that χ (G) ≤ d ∆+ω+ 2
1
e if α = 2, or if α ≥ 3 and ∆ ≥ n −α − 4. Hence, it suffices to
assume α ≥ 3 and ∆ ≤ n −α− 5. Since ∆(G1 ) ≤ ∆ − 1, we deduce ∆(G1 ) ≤ n −α− 6. Moreover, since the case ∆(G1 ) = 0 is
trivial, we may assume ∆(G1 ) ≥ 1. It follows that δ(G1 ) = |V1 |− 1 − ∆(G1 ) = n −α− 1 − ∆(G1 ) ≥ n −α− 1 −(n −α− 6) = 5
and n − α ≥ ∆(G1 ) + 6 ≥ 7.
By a result of Wang [14] we know for any integer k ≥ 2, that every claw-free graph of order greater than 6(k − 1) with
minimum degree at least 3 contains k vertex-disjoint triangles.
Hence, G1 has at least n−α
6
vertex-disjoint triangles, i.e. p ≥ n−α
6
. Since G3 is a clique, r is at most ω. Using these estimations
for p, r, and the premise ∆ ≥ 5
6
(n − α), we obtain
5
∆+ω+1 ≥ (n − α) + ω + 1 ≥ n − α − p + ω + 1 ≥ n − α − p + r + 1
6
= 2(1 + p + q + r ) − 1 ≥ 2χ − 1.
(1) (2)

∆+ω+1
Therefore, d 2
e ≥ χ. 
Now we focus on triangle-free graphs. We already know that Reed’s Conjecture is true for triangle-free graphs with
maximum degree ∆ ≤ 4 (Brooks’ Theorem). Let n(k) denote the smallest possible number of vertices in a triangle-free
k-chromatic graph. For k ≤ 5 the values for n(k) are already known: n(2) = 2, n(3) = 5, n(4) = 11 and n(5) = 22
(compare [6]). This already confirms Conjecture 1 for triangle-free graphs of order n ≤ 21. Brandt and Thomassé [1] proved
that every triangle-free graph of order n with minimum degree δ > 3n is 4-colorable. So, by the previous considerations,
such graphs also satisfy Conjecture 1.

Theorem 12. If G is a graph with clique number ω = 2 and k ≥ 2, then χ (G) ≤ 1 + (k − 1) n−α
 
n(k)−1
.
Proof. Let k ≥ 2. We partition the vertex set as follows: Choose a maximum independent set I in G and set G1 := G − I.
|V (G1 )|
After this partition the vertex set of G1 into l := d n(k)− 1
e subsets V1 , . . . , Vl such that |Vi | ≤ n(k) − 1 for i = 1, . . . , l.
Obviously, χ (G) ≤ 1+ i=1 χ (G[Vi ]). Since |Vi | ≤ n(k)−1 the graph G[Vi ] is (k−1)-colorable. Therefore, χ (G[Vi ]) ≤ k−1
Pl
for i = 1, . . . , l. So we obtain 1 + (k − 1)l = 1 + (k − 1)d n(nk−α
)−1
e as an upper bound for χ (G). 

Proposition 7. If G is a graph  clique number ω = 2, independence number α < n − 10 and maximum degree
 ∆with
∆ ≥ 8(n−α)+
21
118
, then χ ( G) ≤ +3
2
.
8(n−α)+118
Proof. If ∆ ≥ 21
, then

8(n − α) + 181 n − α + 20 n−α

 
∆+3≥ =8· +1≥8 + 1.
21 21 21

Therefore, d ∆+
2
3 −α
e ≥ 4d n21 e + 1 ≥ χ , by Theorem 12 and setting k = 5. #
In the last proposition we considered only the case α < n − 10. The remaining case α ≥ n − 10 is easy as the next
proposition shows:
 ∆ +3 
Proposition 8. If G is a graph with clique number ω = 2 and independence number α ≥ n − 10, then χ (G) ≤ 2
.
Proof. If we choose a maximum independent set I, then the remaining graph G − I has at most 10 < n(4) vertices. Hence,
G − I is 3-colorable and from this it follows that G is 4-colorable. Because d ∆+
2
3
e = ∆ ≥ χ (G) for ∆ = 3, 4 by the Theorem
of Brooks and d ∆+
 ∆ +3 
2
3
e ≥ 4 for ∆ ≥ 5 we conclude χ (G) ≤ 2
. #
1438 A. Kohl, I. Schiermeyer / Discrete Mathematics 310 (2010) 1429–1438

Acknowledgements

The authors would like to thank the referees for helpful comments and suggestions.

References

[1] S. Brandt, S. Thomassé, Dense triangle-free graphs are four colorable: A solution to the Erdős–Simonovits problem, J. Combin. Theory Ser. B. (to appear).
[2] R.L. Brooks, On colouring the nodes of a network, Proc. Cambridge Phil. Soc. 37 (1941) 194–197.
[3] S. Gravier, F. Maffray, Graphs whose choice number is equal to their chromatic number, J. Graph Theory 27 (1998) 87–97.
[4] J.L. Gross, J. Yellen, Handbook of Graph Theory, CRC Press, 2004.
[5] T.R. Jensen, G.F. Royle, Small graphs with chromatic number 5: A computer search, J. Graph Theory 19 (1995) 107–116.
[6] T.R. Jensen, B. Toft, Graph Coloring Problems, Wiley-Interscience, 1995.
[7] A.D. King, B.A. Reed, A. Vetta, An upper bound for the chromatic number of line graphs, in: DMTCS Proc. AE, 2005, pp. 151–156.
[8] A.D. King, B.A. Reed, Bounding χ in terms of ω and ∆ for quasi-line graphs, J. Graph Theory 59 (3) (2008) 215–228.
[9] L. Rabern, A note on Reed’s conjecture, SIAM J. Discrete Math. 22 (2) (2008) 820–827.
[10] S. Radziszowski, Small Ramsey numbers, Electron. J. Combin. Dynamic Survey DS1 (2006).
[11] B. Randerath, I. Schiermeyer, On Reed’s conjecture about ω, ∆ and χ , in: Graph Theory in Paris (Proceedings of a Conference in Memory of Claude
Berge), in: Trends in Mathematics, 2006, pp. 339–346.
[12] B.A. Reed, ω, ∆, and χ , J. Graph Theory 27 (1998) 177–212.
[13] I. Schiermeyer, A new upper bound for the chromatic number of a graph, Discuss. Math. Graph Theory 27 (1) (2007) 137–142.
[14] H. Wang, Vertex-disjoint triangles in claw-free graphs with minimum degree at least three, Combinatorica 18 (3) (1998) 441–447.