This action might not be possible to undo. Are you sure you want to continue?
∗
Yongxi Cheng
Institute for Theoretical Computer Science, Tsinghua University, Beijing 100084, China
cyx@mails.tsinghua.edu.cn
Abstract
An antimagic labeling of a ﬁnite undirected simple graph with m edges and n vertices is a
bijection from the set of edges to the integers 1, . . . , m such that all n vertex sums are pairwise
distinct, where a vertex sum is the sum of labels of all edges incident with the same vertex.
A graph is called antimagic if it has an antimagic labeling. In 1990, Hartsﬁeld and Ringel [5]
conjectured that every simple connected graph, but K
2
, is antimagic. In this article, we prove
that a new class of Cartesian product graphs are antimagic. In particular, by combining this
result and the antimagicness result on toroidal grids (Cartesian products of two cycles) in [7],
all Cartesian products of two or more regular graphs of positive degree can be proved to be
antimagic.
Keywords: antimagic; magic; labeling; regular graph; Cartesian product
1 Introduction
All graphs in this paper are ﬁnite, undirected and simple. We follow the notation and terminology
of [5]. In 1990, Hartsﬁeld and Ringel [5] introduced the concept of antimagic graph. An antimagic
labeling of a graph with m edges and n vertices is a bijection from the set of edges to the integers
1, . . . , m such that all n vertex sums are pairwise distinct, where a vertex sum is the sum of labels
of all edges incident with that vertex. A graph is called antimagic if it has an antimagic labeling.
Hartsﬁeld and Ringel showed that paths P
n
(n ≥ 3), cycles, wheels, and complete graphs K
n
(n ≥ 3)
are antimagic. They conjectured that all trees except K
2
are antimagic. Moreover, all connected
graphs except K
2
are antimagic. These two conjectures are unsettled. In [2], Alon et al. showed that
the latter conjecture is true for all graphs with n vertices and minimum degree Ω(log n). They also
proved that complete partite graphs (other than K
2
) and nvertex graphs with maximum degree
at least n − 2 are antimagic. In [6], Hefetz proved several special cases and variants of the latter
conjecture. In particular, he proved that for integers k > 0 a graph with 3
k
vertices is antimagic if
it admits a K
3
factor. The main tool used in his paper is the Combinatorial NullStellenSatz (see
[1]). In [7], Wang showed that the toroidal grids, i.e., Cartesian products of two or more cycles,
are antimagic. In [3], the author proved that Cartesian products of two paths, or of a cycle and a
path, are antimagic.
∗
This work was supported in part by National Natural Science Foundation of China under grant No. 60553001
and National Basic Research Program of China under grant No. 2007CB807900, 2007CB807901.
1
In this paper, we prove that the Cartesian products G
1
G
2
of a regular graph G
1
and a
graph G
2
of bounded degrees are antimagic, provided that the degrees of G
1
and G
2
satisfy some
inequality. By combining this result and the antimagicness result on the Cartesian products of two
cycles [7], all Cartesian products of two or more regular graphs of positive degree (not necessarily
connected) can be proved to be antimagic. First, we introduce another concept about graph labeling
called δapproximately magic.
Deﬁnition 1.1 A δapproximately magic labeling of a graph with m edges is a bijection from the
set of edges to the integers 1, . . . , m such that the diﬀerence between the largest and the smallest
vertex sums is at most δ, where a vertex sum is the sum of labels of all edges incident with that
vertex. A graph is called δapproximately magic if it has a δapproximately magic labeling.
Thus 0approximately magic is the same as magic in [5], or supermagic in some literature. We
ﬁrst prove some approximate magicness results on connected regular graphs, the following is proved
in Section 2.
Theorem 1.1 If G is an nvertex kregular connected graph (k ≥ 1), then G is (
nk
2
−1)approximately
magic in case k is odd and kapproximately magic in case k is even.
Recall that the Cartesian product G
1
G
2
of two graphs G
1
= (V
1
, E
1
) and G
2
= (V
2
, E
2
) is a
graph with vertex set V
1
V
2
, and (u
1
, u
2
) is adjacent to (v
1
, v
2
) in G
1
G
2
if and only if u
1
= v
1
and u
2
v
2
∈ E
2
, or, u
2
= v
2
and u
1
v
1
∈ E
1
.
Using the approximate magicness results in Theorem 1.1, we prove the following theorem in
Section 3.
Theorem 1.2 If G
1
is an n
1
vertex k
1
regular connected graph, and G
2
is a graph (not necessarily
connected) with maximum degree at most k
2
, minimum degree at least one, then G
1
G
2
is an
timagic, provided that k
1
is odd and
k
2
1
−k
1
2
≥ k
2
, or, k
1
is even and
k
2
1
2
≥ k
2
and k
1
, k
2
are not both
equal to 2.
By combining Theorem 1.2 and the antimagicness result on the Cartesian products of two cycles
in [7], the following theorem is obtained in Section 4.
Theorem 1.3 All Cartesian products of two or more regular graphs of positive degree are an
timagic.
Finally, we give a generalization of Theorem 1.1 in which G is not necessarily connected, and a
generalization of Theorem 1.2 in which G
1
is not necessarily connected. The following two theorems
are proved in Section 5.
Theorem 1.4 (generalization of Theorem 1.1) If G is an nvertex kregular graph (k ≥ 1, G is not
necessarily connected), then G is (
nk
2
− 1)approximately magic in case k is odd and (
2n
3
+ k − 1)
approximately magic in case k is even.
Theorem 1.5 (generalization of Theorem 1.2) If G
1
is an n
1
vertex k
1
regular graph, and G
2
is
a graph with maximum degree at most k
2
, minimum degree at least one (G
1
,G
2
are not necessarily
connected), then G
1
G
2
is antimagic, provided that k
1
is odd and
k
2
1
−k
1
2
≥ k
2
, or, k
1
is even and
k
2
1
2
> k
2
.
2
2 1 1 m
e
1
e 3
e
4
e
m
e
2
e
m
1 m
2 p
1 p
2 1 m p
2 1 1 m
e
1
e 3
e
4
e
m
e
2
e
m
1 m
p
1 p
2 m p
Fig. 1. Labeling of the sequence of trails T : t
1
t
2
. . . t
n
2
For more results, open problems and conjectures on magic graphs, antimagic graphs and various
graph labeling problems, please see [4].
Throughout the paper, we denote x (ceiling of x) to be the least integer that is not less than
x, denote x (ﬂoor of x) to be the largest integer that is not greater than x.
2 Proof of Theorem 1.1
We begin with some terms and deﬁnitions (see [5]). A walk in a graph G is an alternating sequence
v
1
e
1
v
2
e
2
e
t−1
v
t
of vertices and edges of G, with the property that every edge e
i
is incident with
v
i
and v
i+1
, for i = 1, . . . , t − 1. Vertices and edges may be repeated in a walk. A trail in a graph
G is a walk in G with the property that no edge is repeated. A circuit is a closed trail, that is a
trail whose endpoints are the same vertex. A cycle is a circuit with the property that no vertex is
repeated. An Eulerian circuit in a graph G is a circuit that contains every edge of G. In order to
prove Theorem 1.1 for the case that k is odd, we need the following theorem ([5], pp. 56),
Theorem 2.1 (part of Listing Theorem). If G is a connected graph with precisely 2h vertices of
odd degree, h = 0, then there exist h trails in G such that each edge of G is in exactly one of these
trails.
If G is a connected nvertex regular graph of odd degree k, by Theorem 2.1, there are n/2 trails
t
1
, t
2
, . . . , t
n
2
in G, such that each edge of G is in exactly one of these trails. Denote [t[ to be the
length (number of edges) of a trail t. Without loss of generality, assume that [t
1
[ ≥ [t
2
[ ≥ . . . ≥ [t
n
2
[.
By concatenating these trails we get a sequence T : t
1
t
2
. . . t
n
2
, which contains all the m (=
nk
2
)
edges of G. Number the edges of G according to their ordering in T, let e
1
, e
2
, . . . , e
m
be the
numbering. Assign the labels 1, 2, . . . ,
m
2
 to the edges of odd indices e
1
, e
3
, . . . etc., and assign
the labels m, m − 1, . . . ,
m
2
 + 1 to the edges of even indices e
2
, e
4
, . . . etc. (see Figure 1). It is
easy to see that for the above labeling, the sum of any two consecutive edges in T is either m + 1
or m + 2. In addition, if e is the ﬁrst or the last edge of a trail, then the largest possible label
received by e is at most m−
k−1
2
(notice that [t
1
[ ≥ k). For each vertex v of G, the k edges incident
with v can be partitioned into
k−1
2
pairs and a singleton, such that each pair is composed of two
consecutive edges within one of the above n/2 trails, and the single edge is the ﬁrst or the last
edge of a trail. Therefore, for the above labeling, the sum received by any vertex of G is at most
(m −
k−1
2
) +
k−1
2
(m + 2) = m +
k−1
2
(m + 1), at least 1 +
k−1
2
(m + 1), implying that this
is an (
nk
2
− 1)approximately magic labeling of G. For the case that the degree k is even, we need
the following lemma.
Lemma 2.2 Every mvertex connected regular graph of degree 2 (i.e., cycle C
m
) is 2approximately
magic, for m ≥ 3.
3
4
5
2
1
3
m
1 m
2 m
3 m
Fig. 2. 2Approximately magic labeling of C
m
Proof: We have the following four cases:
Case 1. m ≡ 1 (mod 4). Let m = 4t + 1, t ≥ 1. Partition the labels 1, 2, . . . , m into 2t + 1 groups
(1), (2, 3), . . . , (2t, 2t + 1), (2t + 2, 2t + 3), . . . , (m − 1, m). First assign label 1 to an arbitrary edge
of C
m
, then assign the labels (m, m − 1), (2, 3), (m − 2, m − 3), (4, 5), . . . , (2t, 2t + 1) in a way that
each pair of labels are assigned to the two edges that have common endpoints with the labeled arc.
Case 2. m ≡ 3 (mod 4). Let m = 4t + 3, t ≥ 0. Partition the labels 1, 2, . . . , m into 2t + 2 groups
(1), (2, 3), . . . , (2t, 2t + 1), (2t + 2, 2t + 3), . . . , (m − 1, m). First assign label 1 to an arbitrary edge
of C
m
, then assign the labels (m, m−1), (2, 3), (m−2, m−3), (4, 5), . . . , (2t +3, 2t +2) in the same
way as in Case 1.
Case 3. m ≡ 0 (mod 4). Let m = 4t + 4, t ≥ 0. Partition the labels 1, 2, . . . , m into 2t + 3 groups
(1), (2, 3), . . . , (2t, 2t +1), (2t +2), (2t +3, 2t +4), . . . , (m−1, m). First assign label 1 to an arbitrary
edge of C
m
, then assign the labels (m, m−1), (2, 3), (m−2, m−3), (4, 5), . . . , (2t +4, 2t +3) in the
way that each pair of labels are assigned to the two edges that have common endpoints with the
labeled arc, ﬁnally assign the label 2t + 2 to the one nonlabeled edge.
Case 4. m ≡ 2 (mod 4). Let m = 4t + 2, t ≥ 1. Partition the labels 1, 2, . . . , m into 2t + 2 groups
(1), (2, 3), . . . , (2t, 2t +1), (2t +2), (2t +3, 2t +4), . . . , (m−1, m). First assign label 1 to an arbitrary
edge of C
m
, then assign the labels (m, m − 1), (2, 3), (m − 2, m − 3), (4, 5), . . . , (2t, 2t + 1), (2t + 2)
in the same way as in Case 3.
It is easy to see that in any of the above cases, the vertex sums of C
m
are all among m, m+1, and
m + 2, implying the assertion of the lemma (see Figure 2).
Recall that a connected graph with all vertices of even degrees has an Eulerian circuit. It follows
that if G is a connected nvertex regular graph of even degree k, G has an Eulerian circuit, without
loss of generality, say e
1
e
2
. . . e
m
, where m =
nk
2
. We label 1, 2, . . . , m to this circuit using the
above 2approximately magic labeling in Lemma 2.2 (here we view this circuit as a cycle). For each
vertex v of G, the k edges incident with v can be partitioned into k/2 pairs such that each pair is
composed of two consecutive edges in the Eulerian circuit e
1
e
2
. . . e
m
, thus the sum of each pair is
among m, m + 1, and m + 2. Therefore, for the above labeling, the sum received by any vertex of
G is at least
k
2
m, at most
k
2
(m+2), implying that the labeling of G is kapproximately magic.
4
3 Proof of Theorem 1.2
Suppose that G
1
is an n
1
vertex k
1
regular connected graph, V (G
1
) = ¦u
1
, u
2
, . . . , u
n
1
¦, and G
2
is
a graph with maximum degree at most k
2
, minimum degree at least one, V (G
2
) = ¦v
1
, v
2
, . . . , v
n
2
¦.
Denote by m
1
(=
k
1
n
1
2
) and m
2
the number of edges of G
1
and G
2
, respectively.
Let f : E(G
1
G
2
) → ¦1, 2, . . . , m
2
n
1
+m
1
n
2
¦ be an edge labeling of G
1
G
2
, and denote the
induced sum at vertex (u, v) by w(u, v) =
f((u, v), (y, z)) , where the sum runs over all vertices
(y, z) adjacent to (u, v) in G
1
G
2
. In the product graph G
1
G
2
, at each vertex (u, v), the
edges incident to this vertex can be partitioned into two parts, one part is contained in a copy of
G
1
component, and the other part is contained in a copy of G
2
component. Denote by w
1
(u, v)
and w
2
(u, v) the sum at vertex (u, v) restricted to G
1
component and G
2
component respectively,
i.e., w
1
(u, v) =
f((u, v), (y, v)), where the sum runs over all vertices y adjacent to u in G
1
, and
w
2
(u, v) =
f((u, v), (u, z)), where the sum runs over all vertices z adjacent to v in G
2
. Therefore,
w(u, v) = w
1
(u, v) + w
2
(u, v).
Given two isomorphic graphs G and G
, and two labelings f and f
of G and G
respectively,
we call f
a δshift of f, if for each edge e ∈ E(G) and its counterpart e
∈ E(G
) under the
isomorphism, we have f
(e
) = f(e) + δ. Now we will present our labeling of G
1
G
2
, which
contains two steps.
Step 1 (renaming vertices): Assign labels 1, 2, . . . , m
1
to the edges of G
1
, such that the labeling is
(
n
1
k
1
2
− 1)approximately magic if k
1
is odd, k
1
approximately magic if k
1
is even. Without loss of
generality, we can rename the vertices of G
1
such that w(u
1
) ≤ w(u
2
) ≤ . . . ≤ w(u
n
1
), denote this
labeling by L
1
. Assign labels 1, n
1
+ 1, 2n
1
+ 1, . . . , (m
2
− 1)n
1
+ 1 arbitrarily to the edges of G
2
.
Similarly, rename the vertices of G
2
such that w(v
1
) ≤ w(v
2
) ≤ . . . ≤ w(v
n
2
), denote this labeling
by L
2
.
Step 2 (labeling G
1
G
2
): Assign labels m
2
n
1
+1, m
2
n
1
+2, . . . , m
2
n
1
+m
1
n
2
to the edges that are
contained in copies of G
1
component. For the ith G
1
component (with vertices (u
1
, v
i
), (u
2
, v
i
),. . . ,
(u
n
1
, v
i
)), label its edges with m
2
n
1
+(i−1)m
1
+1, m
2
n
1
+(i−1)m
1
+2, . . . , m
2
n
1
+(i−1)m
1
+m
1
,
such that the labeling is an [m
2
n
1
+(i −1)m
1
]shift of L
1
, under the natural isomorphism, for i =
1, . . . , n
2
. Since G
1
is regular, we have w
1
(u
1
, v
i
) ≤ w
1
(u
2
, v
i
) ≤ . . . ≤ w
1
(u
n
1
, v
i
), for i = 1, . . . , n
2
.
Assign labels 1, 2, . . . , m
2
n
1
to the edges that are contained in copies of G
2
component. For the
jth G
2
component (with vertices (u
j
, v
1
), (u
j
, v
2
),. . . , (u
j
, v
n
2
)), label its edges with j, n
1
+j, 2n
1
+
j, . . . , (m
2
−1)n
1
+j, such that the labeling is a (j −1)shift of L
2
, under the natural isomorphism,
for j = 1, . . . , n
1
. From the way we name the vertices of G
2
, we have w
2
(u
1
, v
1
) ≤ w
2
(u
1
, v
2
) ≤
. . . ≤ w
2
(u
1
, v
n
2
).
In what follows we will prove that for the above labeling, if k
1
is odd and
k
2
1
−k
1
2
≥ k
2
, or, if k
1
is even and
k
2
1
2
≥ k
2
and k
1
, k
2
are not both equal to 2, then
w(u
1
, v
1
) < w(u
2
, v
1
) < . . . . . . . . . . . . < w(u
n
1
, v
1
) <
w(u
1
, v
2
) < w(u
2
, v
2
) < . . . . . . . . . . . . < w(u
n
1
, v
2
) < (1)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
w(u
1
, v
n
2
) < w(u
2
, v
n
2
) < . . . . . . . . . . . . < w(u
n
1
, v
n
2
),
implying that the above labeling is antimagic.
5
For each i ∈ ¦1, . . . , n
2
¦, we have w
1
(u
1
, v
i
) ≤ w
1
(u
2
, v
i
) ≤ . . . ≤ w
1
(u
n
1
, v
i
), and w
2
(u
1
, v
i
) <
w
2
(u
2
, v
i
) < . . . < w
2
(u
n
1
, v
i
) since w
2
(u
j+1
, v
i
)−w
2
(u
j
, v
i
) = d(v
i
), where d(v
i
) ≥ 1 is the degree of
v
i
in G
2
, j = 1, . . . , n
1
−1. It follows that w(u
1
, v
i
) < w(u
2
, v
i
) < . . . < w(u
n
1
, v
i
), for i = 1, . . . , n
2
.
In order to prove w(u
1
, v
i+1
) > w(u
n
1
, v
i
), for i = 1, . . . , n
2
− 1, we distinguish between two cases.
Case 1. k
1
is odd. For each i ∈ ¦1, . . . , n
2
− 1¦, we have w(u
1
, v
i+1
) ≥ w(u
1
, v
i
) +
n
1
k
2
1
2
since
w
1
(u
1
, v
i+1
) = w
1
(u
1
, v
i
) + m
1
k
1
= w
1
(u
1
, v
i
) +
n
1
k
2
1
2
(notice that the labeling of the (i + 1)th G
1
component is an m
1
shift of the labeling of the ith G
1
component) and w
2
(u
1
, v
i+1
) ≥ w
2
(u
1
, v
i
).
In addition, we have w(u
n
1
, v
i
) ≤ w(u
1
, v
i
) +(
n
1
k
1
2
−1) +k
2
(n
1
−1) since w
1
(u
n
1
, v
i
) ≤ w
1
(u
1
, v
i
) +
(
n
1
k
1
2
− 1) (notice that G
1
is regular and L
1
is (
n
1
k
1
2
− 1)approximately magic when k
1
is odd),
and w
2
(u
n
1
, v
i
) = w
2
(u
1
, v
i
) +d(v
i
)(n
1
−1) ≤ w
2
(u
1
, v
i
) +k
2
(n
1
−1). It follows that w(u
1
, v
i+1
) −
w(u
n
1
, v
i
) ≥ (w(u
1
, v
i
) +
n
1
k
2
1
2
) −(w(u
1
, v
i
) +(
n
1
k
1
2
−1) +k
2
(n
1
−1)) = n
1
(
k
2
1
−k
1
2
−k
2
) +1 +k
2
> 0,
for i = 1, . . . , n
2
− 1.
Case 2. k
1
is even. Similarly, for each i ∈ ¦1, . . . , n
2
− 1¦, we have w(u
1
, v
i+1
) ≥ w(u
1
, v
i
) +
n
1
k
2
1
2
since w
1
(u
1
, v
i+1
) = w
1
(u
1
, v
i
) + m
1
k
1
= w
1
(u
1
, v
i
) +
n
1
k
2
1
2
and w
2
(u
1
, v
i+1
) ≥ w
2
(u
1
, v
i
). In
addition, w(u
n
1
, v
i
) ≤ w(u
1
, v
i
) + k
1
+ k
2
(n
1
− 1) holds since w
1
(u
n
1
, v
i
) ≤ w
1
(u
1
, v
i
) + k
1
(L
1
is
k
1
approximately magic when k
1
is even) and w
2
(u
n
1
, v
i
) = w
2
(u
1
, v
i
) +d(v
i
)(n
1
−1) ≤ w
2
(u
1
, v
i
) +
k
2
(n
1
−1). It follows that w(u
1
, v
i+1
)−w(u
n
1
, v
i
) ≥ (w(u
1
, v
i
)+
n
1
k
2
1
2
)−(w(u
1
, v
i
)+k
1
+k
2
(n
1
−1)) =
n
1
(
k
2
1
2
− k
2
) + k
2
− k
1
.
If
k
2
1
2
> k
2
, since k
1
is even,
k
2
1
2
−k
2
≥ 1, then w(u
1
, v
i+1
) −w(u
n
1
, v
i
) ≥ n
1
(
k
2
1
2
−k
2
) +k
2
−k
1
≥
n
1
+ k
2
− k
1
> 0 (since n
1
> k
1
). If
k
2
1
2
= k
2
, since k
1
, k
2
are not both equal to 2, we have k
1
> 2,
it follows that w(u
1
, v
i+1
) − w(u
n
1
, v
i
) ≥ k
2
− k
1
=
k
2
1
2
− k
1
> 0. Thus, in any case, we have
w(u
1
, v
i+1
) − w(u
n
1
, v
i
) > 0, for i = 1, . . . , n
2
− 1.
Therefore, (1) holds, implying the assertion of Theorem 1.2.
4 Proof of Theorem 1.3
Since the Cartesian product preserves regularity, we only need to prove that all Cartesian products
of two regular graphs are antimagic. We ﬁrst prove Theorem 1.3 for the case that G
1
and G
2
are both connected, then we generalize the proof to the case where G
1
and G
2
are not necessarily
connected.
4.1 Connected Case
Suppose that G
1
is an n
1
vertex k
1
regular connected graph, and G
2
is an n
2
vertex k
2
regular
connected graph. Without loss of generality, assume that k
1
≥ k
2
. Furthermore, we may assume
k
1
≥ 2 since K
2
K
2
can be easily veriﬁed as antimagic. If k
1
= 2 and k
2
= 1, by Theorem
1.2, G
1
G
2
is antimagic. If k
1
= 2 and k
2
= 2, then G
1
G
2
is a toroidal grid graph and its
antimagicness is proved in [7]. For k
1
≥ 3, if k
1
is odd, then
k
2
1
−k
1
2
≥ k
1
≥ k
2
; if k
1
is even, then
k
1
≥ 4,
k
2
1
2
> k
1
≥ k
2
. Thus by Theorem 1.2, G
1
G
2
is antimagic.
6
4.2 Unconnected Case
Denote by c
1
and c
2
the numbers of connected components of G
1
and G
2
, respectively. It is
easy to see that the number of connected components of G
1
G
2
is c = c
1
c
2
, and each of its
connected components is a (k
1
+ k
2
)regular graph (which is product of one k
1
regular connected
graph and one k
2
regular connected graph). Let m
1
, m
2
, . . . , m
c
be the numbers of edges of these
connected components C
1
, C
2
, . . . , C
c
. The labeling of G
1
G
2
goes as follows. Assign 1, 2, . . . , m
1
to the edges of C
1
, assign m
1
+ 1, m
1
+ 2, . . . , m
1
+ m
2
to the edges of C
2
, . . . . . . , and assign
m
1
+. . . +m
c−1
+1, m
1
+. . . +m
c−1
+2, . . . , m
1
+. . . +m
c−1
+m
c
to the edges of C
c
, such that the
labeling of each connected component is antimagic (this can be achieved because of the previous
proof for the case where G
1
and G
2
are both connected and the regularity of each component).
The whole labeling of G
1
G
2
is antimagic, since between any two diﬀerent components, any sum
of k
1
+k
2
labels from a group of larger labels must be greater than any sum of k
1
+k
2
labels from
a group of smaller labels.
5 Generalizations of Theorem 1.1 and 1.2
In this section, we will prove Theorems 1.4, a generalization of Theorem 1.1 in which G is not nec
essarily connected, and Theorem 1.5, a generalization of Theorem 1.2 in which G
1
is not necessarily
connected.
5.1 Proof of Theorem 1.4
For the case k is odd, by Theorem 2.1 (Listing), for each connected component of G (which is a
connected kregular graph), if it has n
i
vertices, we can decompose it into
n
i
2
trails. By running
this decomposition over all connected components of G, we can get a total number of
n
2
trails, such
that each edge of G is in exactly one of these trails. It is easy to see that the largest length of these
trails is at least k. We concatenate these trails into a sequence in the ordering of nonincreasing
lengths, and label the sequence in the same way as in Theorem 1.1, which results in an (
nk
2
− 1)
approximately magic labeling of G. For the case k is even, we ﬁrst prove the following lemma.
Lemma 5.1 If G is an nvertex graph consisting of vertexdisjoint cycles of odd sizes (numbers of
edges), then G is
2n
3
approximately magic, for n ≥ 3.
Proof: Suppose that G is composed of l cycles C
1
,C
2
,. . . ,C
l
(of sizes n
1
, n
2
, . . . , n
l
, where n
1
≥
n
2
≥ . . . ≥ n
l
≥ 3 are odd numbers, and n
1
+ + n
l
= n). Let n = 3t + ε, t ≥ 1, ε ∈ ¦0, 1, 2¦.
We partition the labels 1, . . . , n into three groups 1, 2, . . . , t and t + 1, . . . , 2t + ε and 2t + ε +
1, 2t + ε + 2, . . . , 3t + ε. Let A : a
1
, a
2
, . . . , a
t
denote the sequence 1, 2, . . . , t; let B : b
1
, b
2
, . . . , b
t+ε
denote the sequence 2t + ε, 2t + ε − 1, . . . , t + 1; and let C : c
1
, c
2
, . . . , c
t
denote the sequence
2t +ε+1, 2t +ε+2, . . . , 3t +ε. It is easy to see that 2t +ε+2 ≤ a
i
+c
j
≤ 4t +ε, a
i
+b
i
= 2t +ε+1,
and b
i
+ c
i
= 4t + 2ε + 1, for i, j = 1, 2, . . . , t. In addition, 2t + 3 ≤ b
i
+ b
j
≤ 4t + 2ε − 1, for i = j,
i, j = 1, 2, . . . , t.
Let m
i
=
n
i
−1
2
, i = 1, 2, . . . , l. We will present a labeling of G, which goes as follows. Label the
cycles C
1
,C
2
,. . . ,C
l
one by one. For the ith cycle C
i
, pick the m
i
smallest elements from the current
(remained) Asequence and the m
i
smallest elements from the current (remained) Csequence, if
at this moment there are at least m
i
elements remained in A (also C). Otherwise, pick all the
7
remained elements of the two sequences. Speciﬁcally, we have the following two cases.
Case 1. At the beginning of the labeling of C
i
, there are at least m
i
elements remained in the
current A (also C) sequence. Denote by a
s
i
+1
, a
s
i
+2
, . . . , a
s
i
+m
i
and c
s
i
+1
, c
s
i
+2
, . . . , c
s
i
+m
i
(where
s
1
= 0, and s
i
= m
1
+ + m
i−1
for 1 < i ≤ l) the m
i
smallest elements of the current A (and
C) sequence. Pick b
s
i
+m
i
from the current Bsequence, and label the edges of C
i
sequentially
with b
s
i
+m
i
, c
s
i
+1
, a
s
i
+1
, c
s
i
+2
, a
s
i
+2
,. . . ,c
s
i
+m
i
, a
s
i
+m
i
, then remove these elements from their
sequences. Since 3t +ε +2 ≤ b
s
i
+m
i
+c
s
i
+1
≤ 4t +2ε +1, for the above labeling, each vertex sum
of C
i
is at least 2t + ε + 1, and at most 4t + 2ε + 1.
Case 2. At the beginning of the labeling of C
i
, the number of elements remained in the current
A (also C) sequence is less than m
i
. In this case we must have n
1
≥ 5 (otherwise all cycles are
‘triangles’, i.e. consisting of 3 edges, in our labeling each triangle will be labeled by three elements,
and exactly one element from each sequence, which is a contradiction). We can assume that l ≥ 2,
since for the case l = 1 G has been proved to be 2approximately magic in Lemma 2.2.
If the current A (also C) sequence is empty, then label the remained nonlabeled cycles arbi
trarily using elements remained in Bsequence. Otherwise, pick all the elements a
s
i
+1
, a
s
i
+2
, . . . , a
t
and c
s
i
+1
, c
s
i
+2
, . . . , c
t
from the current A and C sequences. At this moment, besides b
t
(where
t ≥ 2 since l ≥ 2), b
1
is unused (if i = 1, since t ≥ 2, we have b
1
distinct from b
t
and unused;
if i > 1, since n
1
≥ 5, b
1
has not been used for labeling C
1
, thus is unused). Remove b
t
and
b
1
from the current Bsequence, and label the elements b
t
, c
s
i
+1
, a
s
i
+1
, c
s
i
+2
, a
s
i
+2
,. . . ,c
t
, a
t
, b
1
sequentially to an arc of consecutive edges of C
i
. Then, label the remained nonlabeled edges of
C
i
using arbitrary elements remained in Bsequence, and remove these elements from B. Since
3t +ε +2 ≤ b
t
+c
s
i
+1
≤ 4t +2ε +1, and a
t
+b
1
= 3t +ε, we have that for the above labeling, each
vertex sum of C
i
is at least 2t + ε + 1, and at most 4t + 2ε + 1.
Therefore, for the above labeling, the vertex sums of G are at least 2t +ε+1 (which is
2n
3
+1),
at most 4t + 2ε + 1 (which is 2
2n
3
 + 1), implying that the diﬀerences between vertex sums of G
are at most
2n
3
.
Remark 5.2 The result that G is
2n
3
approximately magic in Lemma 5.1 is actually asymptot
ically best possible. Consider the case that G is consisting of
n
3
‘triangles’. Suppose that label 1
is assigned to an edge v
1
v
2
of a triangle v
1
v
2
v
3
, if the edge v
2
v
3
or v
1
v
3
is assigned with a label
l >
2n
3
, then the diﬀerence of the two vertex sums of v
3
and v
1
, or v
3
and v
2
will be at least
2n
3
.
Similarly, suppose that label n is assigned to an edge v
4
v
5
of a triangle v
4
v
5
v
6
, if the edge v
4
v
6
or
v
5
v
6
is assigned with a label l ≤
n
3
, then the diﬀerence of the two vertex sums of v
5
and v
6
, or v
4
and v
6
will be at least
2n
3
. If neither of the above two cases happens, then the vertex sum of v
1
or
v
2
is at most
2n
3
+ 1, and the vertex sum of v
4
is at least
4n
3
+ 1, thus, the diﬀerence of the two
vertex sums of v
4
and v
1
, or v
4
and v
2
is at least
2n
3
.
Now we will prove Theorem 1.4 for the case that k is even. Since k is even, G is an even
graph (a graph with all vertices having even degrees), it follows that G can be decomposed into
edgedisjoint simple cycles. In addition, two cycles having a common vertex can be merged into
one circuit. Therefore, by repeating merging two cycles of odd sizes that having a common vertex
into an even circuit, ﬁnally we will obtain a collection of s (≥ 0) even circuits P
1
, P
2
, . . . , P
s
(of
sizes 2m
1
, 2m
2
, . . . , 2m
s
), together with a collection of t (≥ 0) vertexdisjoint odd cycles Q
1
, Q
2
,
8
. . . , Q
t
(of sizes n
1
, n
2
, . . . , n
t
, and n
1
+ n
2
+ + n
t
≤ n), such that each edge of G is in exactly
one of these circuits or cycles.
Let m =
nk
2
be the number of edges of G. First we label the even circuits P
1
, P
2
, . . . , P
s
. By
viewing these circuits as cycles, using the 2approximately magic labeling in Lemma 2.2, we assign
labels 1, 2, . . . , m
1
and m, m − 1, . . . , m − m
1
+ 1 to P
1
, assign labels m
1
+ 1, m
1
+ 2, . . . , m
1
+ m
2
and m−m
1
, m−m
1
−1, . . . , m−m
1
−m
2
+1 to P
2
, . . . . . . , and assign labels m
1
+. . . +m
s−1
+1,
m
1
+. . . +m
s−1
+2, . . . . . . , m
1
+. . . +m
s−1
+m
s
and m−m
1
−. . . −m
s−1
, m−m
1
−. . . −m
s−1
−1,
. . . . . . , m−m
1
−. . . −m
s−1
−m
s
+1 to P
s
. Thus, the sum of any two consecutive edges of circuit
P
i
(i = 1, . . . , s) is among m, m + 1, and m + 2.
Let m
∗
= m
1
+m
2
+. . . +m
s
, and n
∗
= n
1
+n
2
+. . . +n
t
. If n
∗
= 0 (i.e., there is no odd cycle),
similarly as in Theorem 1.1, the above labeling of G can be proved to be kapproximately magic, by
partitioning the k edges incident with any vertex of G into k/2 pairs such that each pair is composed
of two consecutive edges in some circuit P
i
(i ∈ ¦1, . . . , s¦). Otherwise, we have n
∗
≥ 3. Assign the
remained labels m
∗
+1, m
∗
+2, . . . , m
∗
+n
∗
to the vertexdisjoint odd cycles Q
1
, Q
2
, . . . , Q
t
, using
the
2n
∗
3
approximately magic labeling in Lemma 5.1. Since 2m
∗
+n
∗
= m, and
n
∗
3
+
2n
∗
3
 = n
∗
for all integers n
∗
≥ 1, it follows that the sum of any two consecutive edges of these odd cycles is at
least 2m
∗
+
2n
∗
3
 +1 = m+1 −
n
∗
3
 (≤ m), and at most 2m
∗
+2
2n
∗
3
 +1 = m+1 −
n
∗
3
 +
2n
∗
3

(≥ m + 2). Therefore, for the whole labeling of G, the sum received by any vertex of G is at least
m
k−2
2
+ (m + 1 −
n
∗
3
), at most (m + 2)
k−2
2
+ (m + 1 −
n
∗
3
 +
2n
∗
3
). Since n
∗
≤ n, the
whole labeling of G is (
2n
3
+ k − 1)approximately magic.
5.2 Proof of Theorem 1.5
If k
1
= 2, since
k
2
1
2
> k
2
, k
2
= 1, G
2
is 1regular, by Theorem 1.3, G
1
G
2
is antimagic. In what
follows we assume that k
1
≥ 3.
We do the same labeling on G
1
G
2
as in Theorem 1.2 (when k
1
is even, the labeling L
1
on G
1
here is (
2n
1
3
+ k
1
− 1)approximately magic). We will prove that for this labeling, (1) still holds if
k
1
≥ 3 is odd and
k
2
1
−k
1
2
≥ k
2
, or, if k
1
≥ 4 is even and
k
2
1
2
> k
2
.
w(u
1
, v
i
) < w(u
2
, v
i
) < . . . < w(u
n
1
, v
i
) can be proved by using the same argument in Theorem
1.2, for i = 1, . . . , n
2
. In order to prove w(u
1
, v
i+1
) − w(u
n
1
, v
i
) > 0, for i = 1, . . . , n
2
− 1, there
are two cases.
Case 1. k
1
is odd. Since G
1
is still (
n
1
k
1
2
− 1)approximately magic, by using the same argument
in Theorem 1.2, we can obtain that w(u
1
, v
i+1
) − w(u
n
1
, v
i
) > 0, for i = 1, . . . , n
2
− 1.
Case 2. k
1
is even (thus k
1
≥ 4). G
1
is (
2n
1
3
+ k
1
− 1)approximately magic. For each i ∈
¦1, . . . , n
2
− 1¦, we have w(u
1
, v
i+1
) ≥ w(u
1
, v
i
) +
n
1
k
2
1
2
since w
1
(u
1
, v
i+1
) = w
1
(u
1
, v
i
) +
n
1
k
2
1
2
and w
2
(u
1
, v
i+1
) ≥ w
2
(u
1
, v
i
). In addition, w(u
n
1
, v
i
) ≤ w(u
1
, v
i
) + (
2n
1
3
+ k
1
− 1) + k
2
(n
1
− 1)
since w
1
(u
n
1
, v
i
) ≤ w
1
(u
1
, v
i
) + (
2n
1
3
+ k
1
− 1) and w
2
(u
n
1
, v
i
) = w
2
(u
1
, v
i
) + d(v
i
)(n
1
− 1) ≤
w
2
(u
1
, v
i
) +k
2
(n
1
−1). Therefore, w(u
1
, v
i+1
) −w(u
n
1
, v
i
) ≥ (w(u
1
, v
i
) +
n
1
k
2
1
2
) −(w(u
1
, v
i
) +(
2n
1
3
+
k
1
− 1) + k
2
(n
1
− 1)) = n
1
(
k
2
1
2
−
2
3
− k
2
) + k
2
− k
1
+ 1.
Since k
2
<
k
2
1
2
, there are two cases: k
2
≤
k
2
1
2
− 2 or k
2
=
k
2
1
2
− 1. If k
2
≤
k
2
1
2
− 2, w(u
1
, v
i+1
) −
w(u
n
1
, v
i
) ≥ n
1
(
k
2
1
2
−
2
3
− k
2
) + k
2
− k
1
+ 1 > n
1
+ k
2
− k
1
> 0 (since n
1
> k
1
). If k
2
=
k
2
1
2
− 1,
w(u
1
, v
i+1
) − w(u
n
1
, v
i
) ≥ n
1
(
k
2
1
2
−
2
3
− k
2
) + k
2
− k
1
+ 1 >
k
2
1
2
− k
1
> 0 (since k
1
≥ 4). Thus, in
9
either case, we have w(u
1
, v
i+1
) − w(u
n
1
, v
i
) > 0, for i = 1, . . . , n
2
− 1.
Therefore, (1) holds, the labeling for k
1
≥ 3 is antimagic.
6 Concluding Remarks and Open Problems
Since the Eulerian circuit of an Eulerian graph (consequently the trails in the Listing Theorem)
can be eﬃciently computed, the proofs in this paper provide eﬃcient algorithms for ﬁnding the
antimagic labelings.
It is easy to see that, for cycles, the 2approximate magicness result in Lemma 2.2 is best
possible (i.e., 2 can not be improved to 0 or 1). For nvertex kregular (k > 2) connected graphs,
it may be interesting to prove that they are δapproximately magic for some δ < (
nk
2
− 1) in case
k is odd, or δ < k in case k is even, or, to prove some lower bounds on δ.
Acknowledgments
The author would like to thank Andy Yao for helpful comments, and thank the anonymous reviewer
for helpful suggestions.
References
[1] N. Alon, Combinatorial Nullstellensatz, Combinatorics, Probability and Computing, 8 (1999),
pp. 7–29.
[2] N. Alon, G. Kaplan, A. Lev, Y. Roditty and R. Yuster, Dense graphs are antimagic, Journal
of Graph Theory, 47 (2004), pp. 297309.
[3] Y. Cheng, Lattice grids and prisms are antimagic, Theoretical Computer Science, 374 (2007),
pp. 6673.
[4] J.A. Gallian, A dynamic survey of graph labeling, ninth edition, The Electronic Journal of
Combinatorics, 5 (2005), DS6, pp. 1148.
[5] N. Hartsﬁeld and G. Ringel, Pearls in Graph Theory, Academic Press, INC., Boston, 1990
(Revised version, 1994), pp. 108109.
[6] Dan Hefetz, Antimagic graphs via the Combinatorial NullStellenSatz, Journal of Graph The
ory, 50 (2005), pp. 263272.
[7] TaoMing Wang, Toroidal Grids Are Antimagic, Proc. 11th Annual International Computing
and Combinatorics Conference COCOON’2005, LNCS 3595, Springer, 2005, pp. 671679.
10
1) If G is an nvertex kregular graph (k ≥ 1. E2 ) is a graph with vertex set V1 × V2 . 2 . or supermagic in some literature.2 in which G1 is not necessarily connected. m such that the diﬀerence between the largest and the smallest vertex sums is at most δ. By combining this result and the antimagicness result on the Cartesian products of two cycles [7]. k1 is even and 2 k1 2 > k2 . E1 ) and G2 = (V2 . then G1 × G2 is ank2 −k k2 timagic. where a vertex sum is the sum of labels of all edges incident with that vertex. The following two theorems are proved in Section 5.2 and the antimagicness result on the Cartesian products of two cycles in [7]. provided that k1 is odd and 1 2 1 ≥ k2 .2) If G1 is an n1 vertex k1 regular graph. u2 = v2 and u1 v1 ∈ E1 . we give a generalization of Theorem 1. Thus 0approximately magic is the same as magic in [5]. Theorem 1. Theorem 1. or.5 (generalization of Theorem 1. we introduce another concept about graph labeling called δapproximately magic. then G is ( nk −1)approximately 2 magic in case k is odd and kapproximately magic in case k is even. and G2 is a graph (not necessarily connected) with maximum degree at most k2 . Finally. or. minimum degree at least one (G1 . provided that the degrees of G1 and G2 satisfy some inequality.2 If G1 is an n1 vertex k1 regular connected graph.1. the following theorem is obtained in Section 4.G2 are not necessarily k2 −k connected). minimum degree at least one. First. Using the approximate magicness results in Theorem 1.In this paper. or. then G is ( nk − 1)approximately magic in case k is odd and ( 2n + k − 1)2 3 approximately magic in case k is even. we prove that the Cartesian products G1 × G2 of a regular graph G1 and a graph G2 of bounded degrees are antimagic. We ﬁrst prove some approximate magicness results on connected regular graphs.4 (generalization of Theorem 1.1 A δapproximately magic labeling of a graph with m edges is a bijection from the set of edges to the integers 1. A graph is called δapproximately magic if it has a δapproximately magic labeling. By combining Theorem 1. then G1 × G2 is antimagic. Theorem 1. Theorem 1. and a generalization of Theorem 1. .1 If G is an nvertex kregular connected graph (k ≥ 1). the following is proved in Section 2. all Cartesian products of two or more regular graphs of positive degree (not necessarily connected) can be proved to be antimagic. . we prove the following theorem in Section 3. G is not necessarily connected). and (u1 . v2 ) in G1 × G2 if and only if u1 = v1 and u2 v2 ∈ E2 . k2 are not both equal to 2. Deﬁnition 1. Recall that the Cartesian product G1 × G2 of two graphs G1 = (V1 . .1 in which G is not necessarily connected. provided that k1 is odd and 1 2 1 ≥ k2 . .3 All Cartesian products of two or more regular graphs of positive degree are antimagic. k1 is even and 21 ≥ k2 and k1 . and G2 is a graph with maximum degree at most k2 . Theorem 1. u2 ) is adjacent to (v1 .
t2 . . If G is a connected nvertex regular graph of odd degree k. . A trail in a graph G is a walk in G with the property that no edge is repeated. t n . that is a trail whose endpoints are the same vertex. Labeling of the sequence of trails T : t1 t2 . 1. Number the edges of G according to their ordering in T . please see [4]. . open problems and conjectures on magic graphs. em be the numbering.1 (part of Listing Theorem).1 e1 m e2 2 e3 m 1 e4 p em 1 p 1 em 1 e1 m e2 2 e3 m 1 e4 m 2p 1 p 2 em 1 em p 1 m 2p Fig. e3 . if e is the ﬁrst or the last edge of a trail. the sum of any two consecutive edges in T is either m + 1 or m + 2. . there are n/2 trails t1 . . the sum received by any vertex of G is at most (m − k−1 ) + k−1 × (m + 2) = m + k−1 × (m + 1). h = 0. . A cycle is a circuit with the property that no vertex is repeated. pp. t n in G.1 for the case that k is odd. Lemma 2.2 Every mvertex connected regular graph of degree 2 (i. etc. (see Figure 1). antimagic graphs and various graph labeling problems. . Without loss of generality. . 2.. Theorem 2. An Eulerian circuit in a graph G is a circuit that contains every edge of G. Denote t to be the 2 length (number of edges) of a trail t. . . etc. by Theorem 2. . . and assign 2 the labels m. we need the following theorem ([5]. . It is 2 easy to see that for the above labeling. then there exist h trails in G such that each edge of G is in exactly one of these trails. . . which contains all the m (= nk ) 2 2 edges of G. . Assign the labels 1. .1. for i = 1. A walk in a graph G is an alternating sequence v1 e1 v2 e2 · · · et−1 vt of vertices and edges of G. at least 1 + k−1 × (m + 1). e4 . 56).e. Therefore. let e1 . For the case that the degree k is even. 3 . . . m + 1 to the edges of even indices e2 . A circuit is a closed trail. such that each edge of G is in exactly one of these trails. .1 We begin with some terms and deﬁnitions (see [5]). e2 . m to the edges of odd indices e1 . . with the property that every edge ei is incident with vi and vi+1 . assume that t1  ≥ t2  ≥ . t n 2 For more results. . implying that this 2 2 2 2 is an ( nk − 1)approximately magic labeling of G. . . . we need 2 the following lemma. denote x (ﬂoor of x) to be the largest integer that is not greater than x. for m ≥ 3. Throughout the paper. 2 By concatenating these trails we get a sequence T : t1 t2 . then the largest possible label received by e is at most m − k−1 (notice that t1  ≥ k).. we denote x (ceiling of x) to be the least integer that is not less than x. . for the above labeling. m − 1. t − 1. cycle Cm ) is 2approximately magic. and the single edge is the ﬁrst or the last edge of a trail. . . . . If G is a connected graph with precisely 2h vertices of odd degree. . the k edges incident 2 with v can be partitioned into k−1 pairs and a singleton. For each vertex v of G. ≥ t n . In addition. Vertices and edges may be repeated in a walk. In order to prove Theorem 1. . 2 Proof of Theorem 1. such that each pair is composed of two 2 consecutive edges within one of the above n/2 trails.
(4. Therefore. (2t + 2) in the same way as in Case 3. . (2t + 3. 2t + 1). . t ≥ 0. (m − 1. 2t + 4). thus the sum of each pair is among m. . (m − 2. . m ≡ 3 (mod 4). . (2t + 2). (2. . m ≡ 2 (mod 4). . (2. . . .m 2 1 m 1 3 m 2 4 m 3 5 Fig. (m − 2. For each vertex v of G. m into 2t + 1 groups (1). . . m + 1. (2t + 2. . . (2. 3). m − 1). . without loss of generality. 2. . . . . implying that the labeling of G is kapproximately magic. 2t + 3). m − 1). . . (2t. m + 1. implying the assertion of the lemma (see Figure 2). 2. 5). . . (2t. . 2 2 4 . 5). . . We label 1. . (2. 2t + 3). t ≥ 1. em . (m − 1. 5). . (2t + 4. (2. 3). Partition the labels 1. . m into 2t + 2 groups (1). and m + 2. . .2 (here we view this circuit as a cycle). 2t + 1). . 2Approximately magic labeling of Cm Proof: We have the following four cases: Case 1. (2t. m − 3). . . (m − 2. . m into 2t + 3 groups (1). Case 3. the k edges incident with v can be partitioned into k/2 pairs such that each pair is composed of two consecutive edges in the Eulerian circuit e1 e2 . m). (2t. m). m − 1). . (m − 1. . Partition the labels 1. . . . . (m − 2. 5). and m + 2. . . . . m to this circuit using the 2 above 2approximately magic labeling in Lemma 2. (4. (m − 1. for the above labeling. 2t + 1). (2t + 2. . m). at most k × (m + 2). . m − 3). 2t + 1). (2. First assign label 1 to an arbitrary edge of Cm . . (2. then assign the labels (m. . . m ≡ 0 (mod 4). 3). then assign the labels (m. t ≥ 1. . m − 3). . . . Let m = 4t + 3. 3). 2. . (2t + 3. . 2t + 1) in a way that each pair of labels are assigned to the two edges that have common endpoints with the labeled arc. (4. 2. . then assign the labels (m. 2t + 4). . . Partition the labels 1. First assign label 1 to an arbitrary edge of Cm . . Let m = 4t + 2. m). 2t + 1). 3). (4. Let m = 4t + 4. the sum received by any vertex of G is at least k × m. m − 1). 3). . G has an Eulerian circuit. 2t + 3) in the way that each pair of labels are assigned to the two edges that have common endpoints with the labeled arc. 3). . . . then assign the labels (m. Partition the labels 1. 3). m − 3). where m = nk . . (2t + 3. First assign label 1 to an arbitrary edge of Cm . (2t + 2). (2t. . Recall that a connected graph with all vertices of even degrees has an Eulerian circuit. m ≡ 1 (mod 4). 2. It is easy to see that in any of the above cases. . 2t + 2) in the same way as in Case 1. First assign label 1 to an arbitrary edge of Cm . . Let m = 4t + 1. t ≥ 0. say e1 e2 . em . (2. m into 2t + 2 groups (1). . . Case 4. . It follows that if G is a connected nvertex regular graph of even degree k. 2. (2t. . Case 2. the vertex sums of Cm are all among m. ﬁnally assign the label 2t + 2 to the one nonlabeled edge.
. . . vi ) ≤ . . where the sum runs over all vertices y adjacent to u in G1 . m2 n1 to the edges that are contained in copies of G2 component.. vi ). vn2 ). . (u. Let f : E(G1 × G2 ) → {1. v) = f ((u. . . we call f a δshift of f . 5 . < w(un1 . 2n1 + 1. which contains two steps. V (G2 ) = {v1 . . . . . . .. n1 + 1. m2 n1 + (i − 1)m1 + 2. . .. vn2 )).. V (G1 ) = {u1 . vi ). where the sum runs over all vertices (y..3 Proof of Theorem 1. for i = 1. . Assign labels 1. (u2 . Similarly. v1 ) < w(u1 . . ... such that the labeling is a (j − 1)shift of L2 . . . at each vertex (u. . . such that the labeling is ( n12k1 − 1)approximately magic if k1 is odd. n2 .. v) = f ((u.e. ... In what follows we will prove that for the above labeling. . for j = 1. . .. v). . . ... minimum degree at least one. if k1 ≥ k2 and k1 .. .. . . . . Given two isomorphic graphs G and G .. . vn2 ). < w(un1 . (m2 − 1)n1 + 1 arbitrarily to the edges of G2 .. ≤ w(vn2 ). Since G1 is regular. vi ) ≤ w1 (u2 . z)) . w(u. then w(u1 . we can rename the vertices of G1 such that w(u1 ) ≤ w(u2 ) ≤ .. v). . (1) implying that the above labeling is antimagic. . for i = 1. . . . vn2 ) < ... z)). . respectively. ≤ w2 (u1 . . v2 ) < . Now we will present our labeling of G1 × G2 . . label its edges with m2 n1 + (i − 1)m1 + 1. . For the ith G1 component (with vertices (u1 . . . . . v2 . .. and G2 is a graph with maximum degree at most k2 . z) adjacent to (u. under the natural isomorphism.. .. . . vn2 ) < w(u2 . v) + w2 (u. . under the natural isomorphism. v1 ). v). . such that the labeling is an [m2 n1 + (i − 1)m1 ]shift of L1 . v) by w(u. m1 to the edges of G1 ... . .. 2. . . n2 . w(u1 . denote this labeling by L2 . v1 ) < . v). n1 . . Assign labels 1. . if k1 is odd and is even and 2 k1 2 2 k1 −k1 2 ≥ k2 . . we have w1 (u1 . . Denote by w1 (u. . where the sum runs over all vertices z adjacent to v in G2 .. ≤ w(un1 ). .. v). . .. we have f (e ) = f (e) + δ. m2 n1 + (i − 1)m1 + m1 .. vn2 }. . . Step 2 (labeling G1 × G2 ): Assign labels m2 n1 + 1. . . . (uj .. .. n1 + j. . v1 ) < w(u2 . From the way we name the vertices of G2 . i. 2.. 2n1 + j. v) in G1 × G2 . and two labelings f and f of G and G respectively. . ... 2.2 Suppose that G1 is an n1 vertex k1 regular connected graph. . Without loss of generality. v) = w1 (u. m2 n1 + 2. . v) = f ((u. . ≤ w1 (un1 . and w2 (u. w1 (u.. . . vi ). and denote the induced sum at vertex (u... rename the vertices of G2 such that w(v1 ) ≤ w(v2 ) ≤ . For the jth G2 component (with vertices (uj . v) restricted to G1 component and G2 component respectively. label its edges with j. un1 }. v)). denote this labeling by L1 . . one part is contained in a copy of G1 component.. v) the sum at vertex (u. vi )). v1 ) ≤ w2 (u1 . v2 ) < w(u2 .. Step 1 (renaming vertices): Assign labels 1. . if for each edge e ∈ E(G) and its counterpart e ∈ E(G ) under the isomorphism. .. v2 ). m2 n1 + m1 n2 to the edges that are contained in copies of G1 component. .. In the product graph G1 × G2 . Therefore. .. . (un1 . . . the edges incident to this vertex can be partitioned into two parts. we have w2 (u1 .. or. . v2 ) < . (m2 − 1)n1 + j. . v) and w2 (u. v2 ) ≤ . . (y. . . .. (y. m2 n1 + m1 n2 } be an edge labeling of G1 × G2 .. < w(un1 . n Denote by m1 (= k12 1 ) and m2 the number of edges of G1 and G2 . u2 . and the other part is contained in a copy of G2 component. k1 approximately magic if k1 is even. k2 are not both equal to 2. (uj . ..
vi ). . if k1 is odd. vi ) + d(vi )(n1 − 1) ≤ w2 (u1 . in any case. vi )−w2 (uj . we have w(un1 . for each i ∈ {1. vi+1 ) − k2 −k n k2 w(un1 . . vi ) < w(u2 . 4 Proof of Theorem 1. for i = 1. for i = 1. . n2 − 1}. vi+1 ) > w(un1 . n1 − 1. and w2 (u1 . vi+1 )−w(un1 . . if k1 is even. vi+1 ) − w(un1 . vi ) + k2 (n1 − 1). In addition. .2. we have it follows that w(u1 . vi ). vi ) since w2 (uj+1 . . vi ) + ( n12k1 − 1) + k2 (n1 − 1)) = n1 ( 1 2 1 − k2 ) + 1 + k2 > 0. assume that k1 ≥ k2 . vi+1 ) ≥ w(u1 . n2 }. . For k1 ≥ 3. then G1 × G2 is a toroidal grid graph and its k2 −k antimagicness is proved in [7]. vi+1 ) − w(un1 . . for i = 1. It follows that w(u1 . k2 are not both equal to 2. 6 . . 2 k1 2 > k1 ≥ k2 . . . . vi ) + d(vi )(n1 − 1) ≤ w2 (u1 . k1 is odd. vi ) + 2 n1 k1 2 2 n1 k1 2 since w1 (u1 . vi ) > 0. n2 − 1}. n2 − 1. vi ) + ( n12k1 − 1) (notice that G1 is regular and L1 is ( n12k1 − 1)approximately magic when k1 is odd). vi ) ≤ w(u1 . For each i ∈ {1. vi+1 ) = w1 (u1 . then 1 2 1 ≥ k1 ≥ k2 . We ﬁrst prove Theorem 1. . vi ) + n k2 k2 (n1 −1). w(un1 . < w2 (un1 .3 for the case that G1 and G2 are both connected. vi ) ≤ w(u1 . Case 1. since k1 is even. G1 × G2 is antimagic. we have k1 > 2. . It follows that w(u1 . we distinguish between two cases. vi ) ≤ . we may assume k1 ≥ 2 since K2 × K2 can be easily veriﬁed as antimagic. . 2 k1 2 If > k2 . for i = 1. If k1 = 2 and k2 = 1. It follows that w(u1 . . .1 Connected Case Suppose that G1 is an n1 vertex k1 regular connected graph. In addition. vi )+k1 +k2 (n1 −1)) = n1 ( 2 k1 2 − k2 ) + k2 − k1 . . vi )+ 12 1 )−(w(u1 . vi ) ≤ w1 (u2 . vi ). Therefore. . then w(u1 . vi ) + k1 + k2 (n1 − 1) holds since w1 (un1 . vi+1 ) − w(un1 . by Theorem 1. . n2 − 1. and w2 (un1 . vi ) < . vi ) < w2 (u2 . and G2 is an n2 vertex k2 regular connected graph. . If = k2 . . Thus by Theorem 1. vi ) + k1 (L1 is k1 approximately magic when k1 is even) and w2 (un1 . . vi ) ≥ k2 − k1 = w(u1 . . vi ). vi+1 ) ≥ w(u1 . ≤ w1 (un1 . . . Case 2. G1 × G2 is antimagic. vi ) + 12 1 and w2 (u1 .For each i ∈ {1. vi ) + 12 1 ) − (w(u1 . then we generalize the proof to the case where G1 and G2 are not necessarily connected. If k1 = 2 and k2 = 2. < w(un1 . n2 . vi ) + ( n12k1 − 1) + k2 (n1 − 1) since w1 (un1 . vi ) ≤ w1 (u1 . vi+1 ) ≥ w2 (u1 .3 Since the Cartesian product preserves regularity. Furthermore.2. − k1 > 0. . where d(vi ) ≥ 1 is the degree of vi in G2 . j = 1. we only need to prove that all Cartesian products of two regular graphs are antimagic. vi ) ≥ (w(u1 . since k1 . vi ) = w2 (u1 . In order to prove w(u1 . . . Without loss of generality. . vi ) ≤ w1 (u1 . vi ) ≥ (w(u1 . vi+1 ) = w1 (u1 . 2 k1 2 − k2 ≥ 1. vi ) ≥ n1 ( 2 k1 2 2 k1 2 2 k1 2 − k2 ) + k2 − k1 ≥ n1 + k2 − k1 > 0 (since n1 > k1 ). . 4. . . k1 is even. . implying the assertion of Theorem 1. vi+1 ) ≥ w2 (u1 . . Similarly. Thus. .2. vi ) = d(vi ). vi ) < . vi ) + m1 k1 = w1 (u1 . we have w1 (u1 . vi ) + n k2 2 n1 k1 2 since w1 (u1 . vi ) = w2 (u1 . n2 − 1. vi ). (1) holds. vi ) + (notice that the labeling of the (i + 1)th G1 component is an m1 shift of the labeling of the ith G1 component) and w2 (u1 . then k1 ≥ 4. . we have w(u1 . we have w(u1 . . vi ) + m1 k1 = w1 (u1 .
m1 + . . . and let C : c1 .1 Proof of Theorem 1. We partition the labels 1. j = 1. . . . . . . + mc−1 + mc to the edges of Cc . . any sum of k1 + k2 labels from a group of larger labels must be greater than any sum of k1 + k2 labels from a group of smaller labels. It is easy to see that the largest length of these trails is at least k. 3 Proof: Suppose that G is composed of l cycles C1 . . . Label the cycles C1 . Let n = 3t + ε. . . . . 2t + ε and 2t + ε + 1. m2 .2 Unconnected Case Denote by c1 and c2 the numbers of connected components of G1 and G2 . 3t + ε. nl . ≥ nl ≥ 3 are odd numbers.1 If G is an nvertex graph consisting of vertexdisjoint cycles of odd sizes (numbers of edges). . . . for i.2 In this section. Let A : a1 . . −1 Let mi = ni2 . . . . By running 2 this decomposition over all connected components of G. we will prove Theorems 1. mc be the numbers of edges of these connected components C1 . . 2. .C2 . . b2 . In addition. . l.1 and 1. . . at denote the sequence 1. . . assign m1 + 1.4. . we can get a total number of n trails. ε ∈ {0. 2t + ε + 2. . . . . and label the sequence in the same way as in Theorem 1. . . . . . . and bi + ci = 4t + 2ε + 1. 5 Generalizations of Theorem 1. for each connected component of G (which is a connected kregular graph). The whole labeling of G1 × G2 is antimagic. . n into three groups 1. . . 3t + ε. . . which goes as follows. C2 . . where n1 ≥ n2 ≥ . for i = j. and Theorem 1. 2. . . . Let m1 . bt+ε denote the sequence 2t + ε. . . if at this moment there are at least mi elements remained in A (also C). 2t + ε − 1. . .1 in which G is not necessarily connected. + mc−1 + 2. 2. . . t + 1. i = 1.4 For the case k is odd. . . c2 . We will present a labeling of G. It is easy to see that 2t + ε + 2 ≤ ai + cj ≤ 4t + ε. . . . . For the ith cycle Ci . ai + bi = 2t + ε + 1. 2. . 5. . . let B : b1 . . . . . and assign m1 + . Lemma 5. t.Cl one by one. . . n2 . which results in an ( nk − 1)2 approximately magic labeling of G. . 2. . The labeling of G1 × G2 goes as follows. .2 in which G1 is not necessarily connected. . 2}. Cc . a generalization of Theorem 1.1 (Listing). . . . pick all the 7 . 2. if it has ni vertices. j = 1.4. 2t + 3 ≤ bi + bj ≤ 4t + 2ε − 1. such that the labeling of each connected component is antimagic (this can be achieved because of the previous proof for the case where G1 and G2 are both connected and the regularity of each component). .Cl (of sizes n1 . we can decompose it into ni trails. 2t + ε + 2. For the case k is even. m1 + . . since between any two diﬀerent components. + mc−1 + 1. .1. . and n1 + · · · + nl = n).C2 . .5. m1 + m2 to the edges of C2 . i. t. . . and each of its connected components is a (k1 + k2 )regular graph (which is product of one k1 regular connected graph and one k2 regular connected graph). t and t + 1. then G is 2n approximately magic. . for n ≥ 3. .. . t ≥ 1. . Otherwise.. Assign 1. we ﬁrst prove the following lemma. . . m1 to the edges of C1 . . We concatenate these trails into a sequence in the ordering of nonincreasing lengths. pick the mi smallest elements from the current (remained) Asequence and the mi smallest elements from the current (remained) Csequence. . a2 . t. respectively. . a generalization of Theorem 1. by Theorem 2. . 1. . such 2 that each edge of G is in exactly one of these trails. It is easy to see that the number of connected components of G1 × G2 is c = c1 × c2 . ct denote the sequence 2t + ε + 1. . m1 + 2.
asi +mi and csi +1 . Remove bt and b1 from the current Bsequence. Since 3t + ε + 2 ≤ bsi +mi + csi +1 ≤ 4t + 2ε + 1. . . 3 Remark 5. label the remained nonlabeled edges of Ci using arbitrary elements remained in Bsequence. and the vertex sum of v4 is at least 4n + 1. it follows that G can be decomposed into edgedisjoint simple cycles. Speciﬁcally. . at and csi +1 . we have the following two cases. ct from the current A and C sequences. or v3 and v2 will be at least 2n . . .csi +mi . . then remove these elements from their sequences. the vertex sums of G are at least 2t + ε + 1 (which is 2n + 1). asi +2 . for the above labeling. In this case we must have n1 ≥ 5 (otherwise all cycles are ‘triangles’. . . . .. 3 3 Similarly. csi +1 . together with a collection of t (≥ 0) vertexdisjoint odd cycles Q1 . we have that for the above labeling. 3 at most 4t + 2ε + 1 (which is 2 2n + 1). then the diﬀerence of the two vertex sums of v5 and v6 .remained elements of the two sequences. Since k is even. asi +2 . . Consider the case that G is consisting of n ‘triangles’. then label the remained nonlabeled cycles arbitrarily using elements remained in Bsequence. . since n1 ≥ 5. we have b1 distinct from bt and unused. csi +mi (where s1 = 0. . at . b1 has not been used for labeling C1 . P2 . 2m2 .. there are at least mi elements remained in the current A (also C) sequence.4 for the case that k is even.2. in our labeling each triangle will be labeled by three elements. which is a contradiction). . the diﬀerence of the two 3 3 vertex sums of v4 and v1 . asi +1 . 8 . csi +2 . then the vertex sum of v1 or 3 v2 is at most 2n + 1. . and si = m1 + · · · + mi−1 for 1 < i ≤ l) the mi smallest elements of the current A (and C) sequence. . asi +2 . the number of elements remained in the current A (also C) sequence is less than mi . . ﬁnally we will obtain a collection of s (≥ 0) even circuits P1 .1 is actually asymptot3 ically best possible. if i > 1. asi +mi . At the beginning of the labeling of Ci . and label the elements bt . . by repeating merging two cycles of odd sizes that having a common vertex into an even circuit. We can assume that l ≥ 2. Denote by asi +1 . csi +2 . for the above labeling. Therefore. . csi +1 . In addition. . . since t ≥ 2. besides bt (where t ≥ 2 since l ≥ 2). Ps (of sizes 2m1 .e. .2 The result that G is 2n approximately magic in Lemma 5. and at most 4t + 2ε + 1. . . thus is unused). Otherwise. Case 2. consisting of 3 edges. since for the case l = 1 G has been proved to be 2approximately magic in Lemma 2. . . b1 is unused (if i = 1. and at + b1 = 3t + ε. pick all the elements asi +1 . Since 3t + ε + 2 ≤ bt + csi +1 ≤ 4t + 2ε + 1. b1 sequentially to an arc of consecutive edges of Ci . or v4 and v2 is at least 2n . At the beginning of the labeling of Ci . csi +2 . . Therefore. 3 Now we will prove Theorem 1. suppose that label n is assigned to an edge v4 v5 of a triangle v4 v5 v6 . . and exactly one element from each sequence. or v4 3 and v6 will be at least 2n . and remove these elements from B. each vertex sum of Ci is at least 2t + ε + 1. G is an even graph (a graph with all vertices having even degrees). csi +2 . Case 1. and label the edges of Ci sequentially with bsi +mi . asi +1 . asi +2 . If neither of the above two cases happens. and at most 4t + 2ε + 1. then the diﬀerence of the two vertex sums of v3 and v1 . Then.ct . Q2 . At this moment. implying that the diﬀerences between vertex sums of G 3 are at most 2n . . Suppose that label 1 3 is assigned to an edge v1 v2 of a triangle v1 v2 v3 . if the edge v2 v3 or v1 v3 is assigned with a label l > 2n . thus. i. . two cycles having a common vertex can be merged into one circuit. each vertex sum of Ci is at least 2t + ε + 1. If the current A (also C) sequence is empty. 2ms ). if the edge v4 v6 or v5 v6 is assigned with a label l ≤ n . Pick bsi +mi from the current Bsequence.
. Assign the remained labels m∗ + 1. . s}). m − m1 − m2 + 1 to P2 . the above labeling of G can be proved to be kapproximately magic. . − ms−1 .1. vi ) ≤ w1 (u1 . by Theorem 1. . . . Therefore. Since 2m∗ + n∗ = m. . or. vi+1 ) − k2 k2 2 w(un1 . . vi ). vi+1 ) − w(un1 . we have n∗ ≥ 3. . Otherwise. . Q2 .2 (when k1 is even. . vi ) + ( 2n1 + k1 − 1) and w2 (un1 . . for i = 1. vi+1 ) − w(un1 . vi ) + ( 3 n k2 n k2 + − − k2 ) + k2 − k1 + 1. . m1 and m. . . vi ) + ( 2n1 + k1 − 1) + k2 (n1 − 1) 3 since w1 (un1 . G1 is ( 2n1 + k1 − 1)approximately magic. . . .2. there are two cases: k2 ≤ 21 − 2 or k2 = 21 − 1. Thus. . . . w(u1 . . . there is no odd cycle). . . Let m = nk be the number of edges of G. vi ) + 12 1 and w2 (u1 . . + nt . If k2 ≤ 21 − 2. . Let m∗ = m1 + m2 + . . − ms−1 − ms + 1 to Ps . . since 21 > k2 . . . k2 k2 2 w(u1 . . n2 .2 Proof of Theorem 1. Case 1. we can obtain that w(u1 . w(u1 . for i = 1. . .. assign labels m1 + 1. vi ) + k2 (n1 − 1). by using the same argument in Theorem 1. . and at most 2m∗ + 2 2n + 1 = m + 1 − n + 2n 3 3 3 3 3 (≥ m + 2). . . + ms−1 + 2. . . for the whole labeling of G. vi ) ≥ n1 ( 21 − 3 − k2 ) + k2 − k1 + 1 > 21 − k1 > 0 (since k1 ≥ 4). vi+1 ) − w(un1 . . . If n∗ = 0 (i. similarly as in Theorem 1. m + 1. .5 k2 If k1 = 2. . and n + 2n = n∗ 3 3 3 for all integers n∗ ≥ 1. Qt . . Thus. the 2 3 2 3 3 whole labeling of G is ( 2n + k − 1)approximately magic. vi ) ≥ (w(u1 . vi+1 ) ≥ w(u1 . . For each i ∈ 3 {1. m∗ + n∗ to the vertexdisjoint odd cycles Q1 . Qt (of sizes n1 . n2 − 1}. vi ) < . w(u1 . . m1 + 2. In order to prove w(u1 . . m − m1 − 1. we have w(u1 . (1) still holds if 3 k1 ≥ 3 is odd and 2 k1 −k1 2 ≥ k2 .2.e. Case 2. using the 2approximately magic labeling in Lemma 2. .2. . there are two cases. In addition. . . vi ) + d(vi )(n1 − 1) ≤ 3 w2 (u1 . and n∗ = n1 + n2 + . Since n∗ ≤ n.. . vi ) ≤ w(u1 . . the labeling L1 on G1 here is ( 2n1 + k1 − 1)approximately magic). . . and m + 2. . . . . We will prove that for this labeling. . . + ms . If k2 = 21 − 1. G2 is 1regular. m − m1 − . Ps . m − 1. . . in 9 . . vi+1 ) = w1 (u1 . . vi+1 ) − w(un1 . and assign labels m1 + . . . n2 − 1. such that each edge of G is in exactly one of these circuits or cycles. . . . . . . Since G1 is still ( n12k1 − 1)approximately magic. n2 . the sum received by any vertex of G is at least ∗ ∗ ∗ m × k−2 + (m + 1 − n ). . . . . vi ) > 0.3. . m1 + . < w(un1 . m − m1 + 1 to P1 . m1 + m2 and m − m1 . vi ) ≥ n1 ( 21 − 3 − k2 ) + k2 − k1 + 1 > n1 + k2 − k1 > 0 (since n1 > k1 ). k1 is odd. . 2. vi ) can be proved by using the same argument in Theorem 1. . . . Therefore. First we label the even circuits P1 . s) is among m. we assign labels 1. n2 − 1. . the sum of any two consecutive edges of circuit Pi (i = 1. + ms−1 + 1. G1 × G2 is antimagic. vi ) + 12 1 since w1 (u1 . by partitioning the k edges incident with any vertex of G into k/2 pairs such that each pair is composed of two consecutive edges in some circuit Pi (i ∈ {1. vi ) = w2 (u1 . w(un1 . nt . . P2 . vi ) < w(u2 . + ms−1 + ms and m − m1 − . m∗ + 2. . . at most (m + 2) × k−2 + (m + 1 − n + 2n ). . 2 k1 k2 k2 k2 Since k2 < 2 . vi ) + k1 − 1) + k2 (n1 − 1)) = k2 n1 ( 21 2 3 2 n1 k1 2n1 2 ) − (w(u1 . . . . We do the same labeling on G1 × G2 as in Theorem 1. 3 5. vi+1 ) ≥ w2 (u1 . . k2 = 1. for i = 1. . if k1 ≥ 4 is even and 2 k1 2 > k2 . k1 is even (thus k1 ≥ 4). vi ) > 0. .1. m1 + . By 2 viewing these circuits as cycles. it follows that the sum of any two consecutive edges of these odd cycles is at ∗ ∗ ∗ ∗ ∗ least 2m∗ + 2n + 1 = m + 1 − n (≤ m). − ms−1 − 1. using ∗ ∗ ∗ the 2n approximately magic labeling in Lemma 5. . . . . . m − m1 − . In what follows we assume that k1 ≥ 3. and n1 + n2 + · · · + nt ≤ n).
Toroidal Grids Are Antimagic. Alon. [4] J. 5 (2005). 671679. A. [3] Y. or δ < k in case k is even. 1148. Theoretical Computer Science. Antimagic graphs via the Combinatorial NullStellenSatz. 263272. Combinatorics. to prove some lower bounds on δ. 11th Annual International Computing and Combinatorics Conference COCOON’2005. or. the proofs in this paper provide eﬃcient algorithms for ﬁnding the antimagic labelings. the 2approximate magicness result in Lemma 2. pp. ninth edition. the labeling for k1 ≥ 3 is antimagic. [2] N. Academic Press. for i = 1. we have w(u1 . For nvertex kregular (k > 2) connected graphs. Lattice grids and prisms are antimagic. pp. [5] N. 374 (2007). for cycles. Therefore.. Acknowledgments The author would like to thank Andy Yao for helpful comments. LNCS 3595. Journal of Graph Theory. Probability and Computing. 6673.either case. Dense graphs are antimagic. 1990 (Revised version. Hartsﬁeld and G. 10 . 297309. 2005. A dynamic survey of graph labeling. Proc. [6] Dan Hefetz. INC. DS6. It is easy to see that. . it may be interesting to prove that they are δapproximately magic for some δ < ( nk − 1) in case 2 k is odd. and thank the anonymous reviewer for helpful suggestions. 47 (2004). Y. 8 (1999). vi+1 ) − w(un1 . n2 − 1. 7–29. Lev. pp. 1994). pp. Combinatorial Nullstellensatz. pp. Ringel. The Electronic Journal of Combinatorics. Kaplan. pp. pp. Springer.2 is best possible (i. Journal of Graph Theory.A. . Yuster. Alon. (1) holds. 50 (2005). 108109.e. Boston. G. 6 Concluding Remarks and Open Problems Since the Eulerian circuit of an Eulerian graph (consequently the trails in the Listing Theorem) can be eﬃciently computed. [7] TaoMing Wang. Pearls in Graph Theory. References [1] N. . Cheng. 2 can not be improved to 0 or 1). Roditty and R. vi ) > 0.. . Gallian.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.