Keywords and Phrases: 3-Edge Coloring, Edge-Critical Graphs, NP-Complete
Keywords and Phrases: 3-Edge Coloring, Edge-Critical Graphs, NP-Complete
Keywords and Phrases: 3-Edge Coloring, Edge-Critical Graphs, NP-Complete
DIAMANTIS KOREAS
ABSTRACT. For a cubic graph G let the following problem: “Among all
proper 4-edge colorings of G find one that minimizes the use of one of the four
colors”. This paper deals with this problem. We consider a 4-edge coloring of G
which minimizes the smallest cardinality m of color classes. Using colors 1, 2, 3
and 4, let this class have the fourth color “4”. We introduce a variable D that
denotes the distance between pairs of edges having color 4. We also introduce
another variable q, which is related to the number of vertices that is necessary
to delete, in order to get a 3-edge critical subgraph of G. Our motivation is to
give inequalities having as parameters m, D and q. Therefore, finding restricted
classes of cubic graphs which guarantee that parameters D or q must have spe-
cific values, then we can get good upper bounds for the cardinality m. Moreover,
we believe that these inequalities are a useful tool in the further investigation of
3-edge critical graphs.
1. Introduction
1.1. Preliminaries. We consider simple and finite graphs in this paper. A
graph G is called ∆-regular if all the vertices have the same degree ∆. A 3-
regular graph is also called cubic graph. An edge-coloring of a simple graph G is
an assignment of colors to the edges of G such that no two incident edges have the
same color. The chromatic index of G, χ0 (G) is the minimum number of colors
in an edge-coloring of G. Vizing proved that χ0 (G)=∆(G) or χ0 (G)=∆(G)+1
for every graph G [6], where ∆ or ∆(G) is the maximum vertex degree of G.
Therefore, each graph belongs to one of the following two categories: either to
class 1, which includes all the graphs G, so that χ0 (G) =∆(G), or to class 2, which
includes all the graphs, so that χ0 (G)=∆(G)+1. The problem of classifying a
graph is NP-complete[4]. This problem remains NP-complete for triangle free
graphs with maximum degree 3 [5]. Planar graphs with ∆<7 have not been
completely classified. So, it is reasonable to search for approximating algorithms
that decide, in polynomial time, if a graph with ∆=k is k-edge colorable for
some restricted classes of graphs. A classical work in the edge coloring of graphs
is in [2].
A different direction to face the k-edge coloring problem is to try to color as
many edges as possible using k colors. This approach is known as max edge
k-coloring: ”Given a graph G and a number k, color as many edges as possible
1
2
using k colors”. In the max edge k-coloring problem we try to find the best
possible ratio of the number of edges that are colored with k-colors to the size
of the given graph. Some well known approximating ratios are: 1 − (1 − k1 )k or
k
k+1 [3],[1].
In this paper we use a variation of the max edge 3-coloring problem: For a
cubic graph G let the following problem: “Among all proper 4-edge colorings
of G find one that minimizes the use of one of the four colors”. Let a 4-edge
coloring of G which minimizes the smallest cardinality m of color classes. Using
colors 1, 2 3 and 4 we assign to this class color the fourth color 4. Max edge
3-coloring problem allows the use of more than 4 colors so, any ratio that we can
achieved in our problem also holds for the original max edge 3-coloring problem.
We introduce a variable D to denote the distance between pairs of edges having
color 4. We also introduce another variable q, which is related to the number
of vertices that is necessary to delete, in order to get a 3-edge critical subgraph
of G. We give inequalities having parameters m, D and q. In other words,
our purpose is to find upper bounds for m using parameters D and q. These
inequalities are useful, since if we find restricted classes of cubic graphs which
guarantee that these parameters must have specific values, then we can get good
upper bounds for the cardinality m.
1.3. Basic lemmas. For our purpose we need one more definition and some
basic results.
Definition 4. We shall call a graph edge-critical if it is connected, it is of class
2 and the removal of any edge transforms it to a graph of class 1. In this paper
when we say “critical graph” we mean “edge-critical graph”. If a critical graph
3
Figure 1. Edges eb and gj have color “4” and their distance is 7. The
distance between edges that get the fourth color can be less than 7 but
then we will use more than twice the color “4”.
has maximum degree ∆, then we shall call it ∆-critical. Some basic results, see
[2], for a 3-edge critical graph G with n vertices are:
From now on, we shall call simply the set of the 2m-2 vertices of degree 2 with
n
M . Since we suppose that H is critical we get 2m − 2≤ and finally
3
n
m≤ + 1 (1)
6
Consider now the case where H is not critical. Graph H can be either con-
nected or disconnected. If H consists of more than one components then only
one of them is not 3-edge colorable, actually the one that has edge v, so all the
3-edge colorable components must be deleted. In both cases the deletion of some
other edges except the edge v will give a subgraph of H that is critical, say H0
of order n0 .
The deletion of other edges except the m-1 ones that we first deleted it is pos-
sible to destroy some vertices of degree 2. During the deletion of those edges a
sequence of graphs H0 , H1 , . . . , Hk is generated.
Obviously in this sequence the first element is H and the last is H0 .
Notice that we do not interested in the procedure that leads into the critical
subgraph H0 .
Let examine the following cases:
Case 1. In graph Hi there are no adjacent vertices of degree 2.
Suppose that the number 2m-2 of vertices of degree 2 was reduced by one due
to the deletion of an edge adjacent to a vertex of degree 2. However, if we delete
one edge adjacent to a vertex of degree 2 then we must also delete the other
one as well. That is because a 3-critical graph cannot have vertices of degree 1.
Therefore, the deletion of the second edge generates a vertex of degree 2 and at
the end of this process two new vertices of degree 2 are generated.
Case 2. In graph Hi there are adjacent vertices of degree 2.
In a 3-critical graph there are not adjacent vertices of degree 2 and these vertices
must be deleted.
Let vertices a and b belong to M . For a subgraph Hi vertices a and b can belong
to a path of adjacent vertices of degree 2 for two reasons:
i. edges ac and bd in which we have assigned color 4 have distance 1, see fig
2, and their deletion makes vertices a and b adjacent.
ii. edges ac and bd, in which we have assigned color 4, have distance greater
than 1, see fig 3. So, between them there are vertices of degree 3. In these in-
termediate vertices we have adjacent edges in which a different color than color
4 is assigned. If all these intermediate edges are deleted then we get a path of
adjacent vertices of degree 2 to which vertices a and b belong.
If one of the previous reasons occurs then the number 2m-2 of vertices of degree
2 can be reduced. At the same time the order n is also reduced. If the number
2m-2 is reduced by q and the order n by Q then we get the inequality:
n 3q − Q
m≤ +1+ (2)
6 6
We conclude that we can have a greater upper bound than this in the inequal-
ity (1) if 3q-Q has a positive value. Trying to examine the possible values of
3q-Q we use the variables Q and Q. Variable Q to count the number of vertices
5
that we have deleted starting from the initial order n. Variable q to count the
difference between the number of vertices of degree 2 from the initial number
2m-2. Sometimes, we use variable Q to denote the number of vertices that we
delete in one step of our procedure and not the number of vertices that we have
deleted. In the same way, we can use variable q to denote the number of vertices
of degree 2 that are deleted only in one step of our procedure. For example,
suppose that in the first step of the procedure one vertex of degree 2 is deleted
and its deletion generates two new vertices of degree 2. Then q=1+(-2)=-1
6
and Q=1, so 3q-Q=-4. Suppose that in the second step of the procedure three
adjacent vertices of degree 2 are deleted and their deletion generates two new
vertices of degree 2. Then q=3+(-2)=1 and Q=3, so 3q-Q=0. Adding the pre-
vious values from step 1 to the variables q and Q, we have their current values,
q=0 and Q=4. In other words the current value of 3q-Q is -4. It is clear that
each step of the procedure contributes an integer value to the expression 3q-Q
(which we shall call simply 3q-Q). We are going now to examine when 3q-Q gets
a maximum, since we want to find an upper bound for m. That means that we
do not interested in cases where a negative or zero value is contributed to it.
Let a path, in which the endpoints are vertices of degree 3 and all the other
vertices are of degree 2.
We will study the following cases:
i. None of the vertices of degree 2 belong to M . Then their deletion contributes
to 3q-Q a negative value.
ii. Between any pair of vertices that belongs to M there are D-1 vertices of
degree 2 that do not belong to M . So, in this path vertices that belong to M
have been generated due to the deletion of edges with the color 4 at distance D.
Let the number of vertices in M in this path be q1 , see fig 4. So, the q+(q-
Figure 4. Edges having the color ”4” in G are deleted and vertices a,
b, c and d have degree 2. Between them there are vertices of degree 3 that
correspond to edges in G to which we have assigned a color different than
color 4. When all these edges are deleted then in the path are 10 adjacent
vertices of degree 2. Vertices a and b, b and c, c and d have distance 3.
Deleting them the number 2m-2 is reduced by 4-2=2, since 2 new vertices
of degree 2 are generated. The order of graph G is reduced by 10.
1)(D-1) vertices of degree 2 must be deleted and this deletion generates two
vertices of degree 2. That means that the number 2m-2 is reduced by q1 -2
and the order by Dq1 -D+1, therefore we contribute to 3q1 -Q the value 3(q1 -2)-
(Dq1 )+D-1=(3-D) q1 +D-7. If some vertices in M are at greater distance than
D then 3(q1 -2)-(Dq1 )+D-1≤(3-D) q1 +D-7.
7
Figure 5. We have the previous example but now different edges are
deleted. Actually, 4 edges are deleted: two edges having endpoints between
vertices a and b and two edges having endpoints between vertices c and
d. We have three consecutively paths. The first has 4 adjacent vertices of
degree 2. This path is followed by a second path with 2 adjacent vertices
of degree 3. This second path is followed by a third path with 4 adjacent
vertices of degree 2. In this way the number 2m-2 is reduced by 4-4=0,
since 4 new vertices of degree 2 are generated. At the same time the order
is reduced by 8.
The value of (3-D)q +k(D-7) takes a maximum when k takes a maximum, so for
q
k= .
2
We avoid details that we have already study in subcase 1. We only note that
inequality (4) does not change if the paths we have described form cycles.
If all adjacent vertices of degree 2 are deleted, then we leave case 2 and we
face again case 1 for some graph Hj , j >i, and so on.
According to the inequalities (3) and (4) and for D≤7 when we get H0 we can
assume that the number of 2m-2 vertices of degree 2 is totally reduced by q and
the size of H0 is totally reduced by at least Dq. Let t be the total number of
vertices of degree 2 in H0 and n0 its order.
n0 (n − Dq) n (3 − D)q
Then we have 2m − 2 − q = t≤ ≤ , so m≤ + + 1. We
3 3 6 6
consider the following cases:
n (3 − D)q
(1) If D < 3 we get the maximum value for + + 1 when q=2m-1
6 6
n
that is + 1.
6
n (3 − D)q n
(2) If D = 3 then the expression + + 1 gets the value + 1.
6 6 6
n (3 − D)q
(3) If D > 3 and D < 7 then the expression + + 1 achieves the
6 6
n 1 3
maximum value when q is zero, so m≤ + + .
2D 2 2D
That means that in the worst case we have the desired inequality:
n 1 3 n
m≤max{ + + , + 1}.
2D 2 2D 6
We complete the first and the second part of the theorem by considering the
q
case where D>7. Using inequality (2) and equation (3) with k = we get
2
n q(D + 1)
m≤ + 1 − .
6 12
n q(D + 1)
The expression + 1 − takes a maximum for q=0. 2
6 12
Corollary 1. Let G be a cubic graph with n vertices. Suppose that G belongs
to class 2 and m is the number of times we have used color 4 in a legal 4-edge
coloring of G. Then the following holds:
If in the assigned 4-edge coloring the distance between any pair of edges, in which
n 1 3 n
color 4 has been assigned, is at least D and m>max{ + + , + 1} then
2D 2 2D 6
this 4-edge coloring is not optimal, i.e. there exists a 4-edge coloring using less
than m times the color 4.
Proof. The proof follows from our theorem.
Corollary 2. Let G be a cubic graph with n vertices. Suppose that G belongs to
class 2 and m is the minimum number of times we have used color 4 (over all
possible 4-edge colorings). Then the followings hold :
9
6(m − 1) − n 6(m − 1) − n
i. for D< 3 we get q> , for D> 3 and D<7 we get q< ,
3−D 3−D
12 + 2n − 12m
and for D > 7 we get: q≤3 −
D+1
6(m − 1) − n
ii. for q> 0 and for D≤7 we get D≤3 − and for D> 7 we get
q
12 + 2n − 12m − q
D≤
q
(7 + D)q (n + 6)
iii. m≥ −
20 10
Proof. The proof for the first and second part follows from the inequalities in
our theorem.
We shall prove the third part using the configuration shown in fig 6. This be-
haves as a single edge in a 3-edge coloring, in the sense that edges ab and cd
must get the same color. In any edge having color 4 are adjacent four edges,
each one having one of the colors 1, 2 or 3. We pick all these four edges and
replace each one by the configuration in fig 6. It is clear that our 4-edge col-
oring remains optimal in the resulting new graph. Also the new graph remains
connected, bridgeless and cubic. The order n and the minimum distance D
have changed. Indeed, the configuration which we use has four vertices and the
sorter path that connects vertices a and d has length 4. Therefore, the order of
the new graph is n+16m and its minimum distance is D + 6. Using the inequal-
ity, for D≥7, we have proved in the second part of our theorem we are finished. 2
2.2. Criteria that guarantee bounds for the distance D. One can ask
“Is it necessary to assign 4-edge colorings to G in order to find the minimum
distance between edges with the fourth color?” In other words “Do we have any
criteria that predetermine the distance between edges that are colored with the
fourth color in an optimal 4-edge coloring?” The answer in the second question
is “Yes”. We will give two simple conditions below:
(1) Let graph G contain a non 3-edge colorable subgraph H, so that in
every 4-edge coloring of H we have to use at least 2 times color 4. Let
10