L(3, 2, 1)-Labeling of Simple Graphs

Jean Clipperton
Simpson College, IA
[email protected]
Jessica Gehrtz
Concordia College-Moorhead, MN
[email protected]
Zsuzsanna Szaniszlo ∗
Valparaiso University, IN
[email protected]
Desmond Torkornoo
University of Richmond, VA
[email protected]
July 11, 2005

Abstract
An L(3, 2, 1)-labeling is a simplified model for the channel assignment
problem. It is a natural generalization of the widely studied L(2, 1)-
labeling.
An L(3, 2, 1)-labeling of a graph G is a function f from the vertex set
V (G) to the set of positive integers such that for any two vertices x, y, if
d(x, y) = 1, then |f (x) − f (y)| ≥ 3; if d(x, y) = 2, then |f (x) − f (y)| ≥
2; and if d(x, y) = 3, then |f (x) − f (y)| ≥ 1. The L(3, 2, 1)-labeling
number k(G) of G is the smallest positive integer k such that G has
an L(3, 2, 1)-labeling with k as the maximum label. In this paper we
determine the L(3, 2, 1)-labeling number for paths, cycles, caterpillars, n-
ary trees, complete graphs and complete bipartite graphs. We also present
an upper bound for k(G) in terms of the maximum degree of G.

1 Introduction
The assignment of FM frequencies to stations became a problem as technology
advanced in the early 20th centuy. As more and more stations requested fre-
quencies, it became difficult to assingn frequencies without having new stations
∗ NSF Grant.

1
interfere with the broadcast of other stations nearby. The channel assignment
problem is an engineering problem in which the task is to assign a channel (non-
negative integer) to each FM radio station in a set of given stations such that
there is no interference between stations and the span of the assigned chan-
nels is minimized. The level of interference between any two FM radio stations
correlates with the geographic locations of the stations. Closer stations have a
stronger interference, and thus there must be a greater difference between their
assigned channels.
In 1980, Hale introduced a graph theory model of the channel assignment
problem where it was represented as a vertex coloring problem [3]. Vertices on
the graph correspond to the radio stations and the edges show the proximity of
the stations.
In 1991, Roberts proposed a variation of the channel assignment problem
in which the FM radio stations were considered either ”close” or ”very close.”
”Close” stations were vertices of distance two apart on the graph and were as-
signed channels that differed by two; stations that were considered ”very close”
were adjacent vertices on the graph and were assigned distinct channels [6].
More precisely, Griggs and Yeh defined the L(2, 1)-labeling of a graph G =
(V, E) as a function f which assigns every x, y in V a label from the set of
positive integers such that |f (x) − f (y)| ≥ 2 if d(x, y) = 1 and |f (x) − f (y)| ≥ 1
if d(x, y) = 2 [2]. L(2, 1)-labeling has been widely studied in recent years.
In 2001, Chartrand et al. introduced the radio-labeling of graphs; this was
motivated by the regulations for the channel assignments in the channel assign-
ment problem [1]. Radio-labeling takes into consideration the diameter of the
graph, and as a result, every vertex is related.
Practically, interference among channels may go beyond two levels. L(3, 2, 1)-
labeling naturally extends from L(2, 1)-labeling, taking into consideration ver-
tices which are within a distance of three apart; however, it remains less difficult
than radio-labeling. An L(3, 2, 1)-labeling of a graph G = (V, E) is a function
f which assigns every x, y in V a label from the set of positive integers such
that |f (x) − f (y)| ≥ 3 if d(x, y) = 1, |f (x) − f (y)| ≥ 2 if d(x, y) = 2, and
|f (x) − f (y)| ≥ 1 if d(x, y) = 3. The L(3, 2, 1)-labeling number, k(G), of G
is the smallest number k such that G has an L(3, 2, 1)-labeling with k as the
maximum label [4].
In this paper we determine the L(3, 2, 1)-labeling number for paths, cycles,
caterpillars, n-ary trees, complete graphs and complete bipartite graphs. We
also present an upper bound for k(G) in terms of the maximum degree of G.
The method for calculating the upper bound was introduced by Griggs and Yeh
[2] and referenced by Jonas [5].

2
2 Definitions and Notation
Definition 1. Let G = (V, E) be a graph and f be a mapping f : V −→ N. f
is an L(3, 2, 1)-labeling of G if, for all x, y ∈ V ,

3, if d(x, y) = 1;

|f (x) − f (y)| ≥ 2, if d(x, y) = 2;

1, if d(x, y) = 3.


Definition 2. The L(3, 2, 1)-number, k(G), of a graph G is the smallest natural

number k such that G has an L(3, 2, 1)-labeling with k as the maximum label.
An L(3, 2, 1)-labeling of a graph G is called a minimal L(3, 2, 1)-labeling of G
if, under the labeling, the highest label of any vertex is k(G).
Note. If 1 is not used as a vertex label in an L(3, 2, 1)-labeling of a graph, then
every vertex label can be decreased by one to obtain another L(3, 2, 1)-labeling
of the graph. Therefore in a minimal L(3, 2, 1)-labeling 1 will necessarily appear
as a vertex label.
Definition 3. Let G = (V, E) be a graph.
G is called a complete graph on n vertices, Kn , if for all vertices x, y ∈ V ,
(x, y) ∈ E.
G is called a complete bipartite graph, Km,n , if:
1. The set of vertices, V , can be partitioned into two disjoint sets of vertices,
A and B, such that |A| = m, |B| = n, and |V | = m + n
2. For all ai , aj ∈ A, (ai , aj ) ∈
/ E and for all bi , bj ∈ B, (bi , bj ) ∈
/E
3. For all ai ∈ A and bj ∈ B, (ai , bj ) ∈ E.
A star, Sn , is a K1,n complete bipartite graph.
Definition 4. Let G = (V, E) be a graph.
G is called a path, Pn , if V = {v1 , v2 , . . . , vn } such that for 1 ≤ i < n, (vi , vi+1 ) ∈
E.
G is called a cycle, Cn , if V = {v1 , v2 , . . . , vn } such that for 1 ≤ i < n (vi , vi+1 ) ∈
E and (v1 , vn ) ∈ E.
Definition 5. Let G = (V, E) be a graph.
G is called a tree if G is connected and has no cycles.
G is called a caterpillar if G is a tree such that the removal of the degree one
vertices produces a path called the spine of the caterpillar. A uniform caterpillar
is a caterpillar with only degree one and degree ∆ vertices. We denote a uniform
caterpillar with n vertices on the spine by Catn .
G is called an n-ary tree if G is a rooted tree such that the root has degree n
and all the other vertices have degree n + 1.

3
3 Theorems
3.1 Complete Graphs
Theorem 1. For any complete graph on n vertices, k(Kn ) = 3n − 2.
Proof. Let G = (V, E) be a complete graph with V = {v1 , v2 , . . . , vn } and let
f be a minimal L(3, 2, 1)-labeling of G with f (vi ) < f (vj ) for all i < j. Then,
for all vi , vj ∈ V and i 6= j, (vi , vj ) ∈ E, implying that d(vi , vj ) = 1. Thus,
|f (vi ) − f (vj )| ≥ 3 for all vi , vj ∈ V and i 6= j. Since there exists vi in V such
that f (vi ) = 1, we have f (v1 ) = 1. Also, since f is a minimal L(3, 2, 1)-labeling
of G, we have that for all vi with 1 < i ≤ n, f (vi ) ≥ f (vi−1 ) + 3. Recursively,
we have
f (vn ) ≥ f (vn−1 ) + 3
≥ f (vn−2 ) + 3(2)
≥ f (vn−3 ) + 3(3)
..
.
≥ f (vn−i ) + 3(n − i)
..
.
≥ f (v1 ) + 3(n − 1) = 1 + 3(n − 1) = 3n − 2.
Therefore k(Kn ) = 3n − 2.

3.2 Complete Bipartite Graphs and Stars

Theorem 2. For any complete bipartite graph, k(Km,n ) = 2(m + n).
Proof. Let G = (V, E) be a complete bipartite graph, Km,n , and let A =
{a1 , a2 , . . . , am } and B = {b1 , b2 , . . . , bn } be the two sets of vertices that par-
tition V as per definition of a complete bipartite graph. Let f be a minimal
L(3, 2, 1)-labeling of G with f (a1 ) ≤ f (a2 ) ≤ · · · ≤ f (am ) and f (b1 ) ≤ f (b2 ) ≤
· · · ≤ f (bn ). Note that d(ai , aj ) = 2 for all ai , aj ∈ A with i 6= j; the same is
true for pairs of vertices in B. Let f (a1 ) = 1, then we need each f (ai ) with i 6= 1
to be odd since f is minimal. Thus, f (A) = {1, 3, 5, . . . , 1 + 2(m − 1)}. Since
each ai ∈ A is adjacent to every bi ∈ B, we need |f (ai ) − f (bi )| ≥ 3. Note that
we need f (b1 ) ≥ f (am ) + 3. Then, since f (am ) = 2m − 1, f (b1 ) ≥ 3 + (2m − 1).
Since f is minimal we have f (b1 ) = 3 + (2m − 1) = 2m + 2. The label-
ing of vertices in B follows an argument similar to the labeling of vertices
in A: since f (b1 ) = 2m − 2 we need each f (bi ) to be even. Thus, f (B) =
{2m + 2, 2m + 4, . . . , 2m + 2 + 2(n − 1)}. Then f (bn ) = 2m + 2 + 2(n − 1) =
2(m + n). Therefore k(Km,n ) = 2(m + n).
Corollary 3. For a star, Sn , k(Sn ) = 2n + 2.
Proof. By definition, a star, Sn , is K1,n . Therefore k(Sn ) = 2n + 2.

4
3.3 Paths
Lemma 4. For a path on n vertices, Pn , with n ≥ 8, k(Pn ) ≥ 8.
Proof. Let f be a minimal L(3, 2, 1)-labeling for a path on n vertices, Pn , and
suppose k(Pn ) < 8 for n ≥ 8. Let v1 be a vertex with label 1. There is an induced
subpath of at least 5 vertices with v1 as an end vertex. Let V = {v1 , v2 , v3 , v4 , v5 }
be this path. We will continue by considering the possibilities for f (v2 ).
Case I: f (v2 ) = 4:
Then we need f (v3 ) = 7, f (v4 ) = 2, f (v5 ) = 5, and f (v6 ) ≥ 8, which
contradicts our assumed k(Pn ).
Case II: f (v2 ) = 5:
Then we need f (v3 ) ≥ 8, a contradiction to our assumed k(Pn ).
Case III: f (v2 ) = 6:
Then we need f (v3 ) = 3, forcing f (v4 ) ≥ 8, a contradiction.
Case IV: f (v2 ) = 7:
Then f (v3 ) = 3 or 4. Either possibility for f (v2 ) forces f (v4 ) ≥ 9, which is
a contradiction.
Therefore we can conclude that k(Pn ) ≥ 8, when n ≥ 8.
Theorem 5. For any path, Pn ,



1, if n = 1;
4, if n = 2;



k(Pn ) = 6, if n = 3, 4;

7,

 if n = 5, 6, 7;


8, if n ≥ 8.

Proof. Let V = {v1 , v2 , . . . , vn } be the set of vertices of Pn such that vi is

adjacent to vi+1 for 1 ≤ i < n. Define f such that f ({v1 , v2 , . . . , v8 }) =
{1, 4, 7, 2, 5, 8, 3, 6} and f (vi ) = f (vj ) if i ≡ j (mod 8). By definition of f
we can conclude that k(Pn ) ≤ 8 for n ≥ 8. We combine this result with that of
Lemma 4 to get k(Pn ) = 8 for n ≥ 8.
For each Pn with n < 8, we proceed by cases using the labeling pattern
defined by f and the observation that each P1 , P2 , . . . , P7 is an induced subpath
of Pn with n ≥ 8:
Case I: n = 1.
This is trivially true.
Case II: n = 2.
The labeling pattern {1, 4} shows that k(P2 ) = 4 because we cannot do any
better.

5
Case III: n = 3, 4.
The labeling pattern {3, 6, 1, 4} shows that k(Pn ) ≤ 6 for n = 3, 4. Suppose
k(Pn ) < 6. There is a vertex vi ∈ V such that f (vi ) = 1. If vi has degree
2, then vertices vi−1 and vi+1 exist such that f (vi−1 ) ≥ 4 and f (vi+1 ) ≥ 6, a
contradiction. If vi has degree 1, let vi = v1 , then the possibilities for f (v2 ) are
4 and 5. In either case, we would need f (v3 ) > 6, a contradiction.
Case IV: n = 5, 6, 7.
The labeling pattern {3, 6, 1, 4, 7, 2, 5} shows that k(Pn ) ≤ 7 for n = 5, 6, 7.
Suppose k < 7. There is a vertex vi ∈ V such that f (vi ) = 1 and vertices vi+1
and vi+2 exist or vi−1 and vi−2 exist. Without loss of generality, suppose vi+1
and vi+2 exist. The possibilities for f (vi+1 ) are 4, 5, and 6. If f (vi+1 ) = 4, 5
we need f (vi+2 ) ≥ 7, a contradiction. If f (vi+1 ) = 6 then f (vi+2 ) = 3. Now
if vi+2 has degree 2, then vi+3 exists and f (vi+3 ) > 7. If vi+2 has degree 1
then vertices vi−1 and vi−2 exist. The only possibility for f (vi−1 ) is 4. But this
forces f (vi−1 ) ≥ 7, a contradiction.

3.4 Cycles
Lemma 6. Let n be an odd integer, n > 3, then k(Cn ) 6= 8.
Proof. Let f be a minimal L(3, 2, 1)-labeling of Cn where n is odd and n > 3.
Suppose k(Cn ) = 8. Then the only possible labelings with 8 as the maximum in-
teger are as follows: f1 (V ) = {8, 5, 2, 7, 4, 1, 6, 3, 8 . . . }, f2 (V ) = {8, 5, 2, 7, 4, 1, 8 . . . },
f3 (V ) = {8, 5, 1, 7, 4, X}, f4 (V ) = {8, 5, 1, 7, 3, X}, f5 (V ) = {8, 4, 1, 6, 3, 8, . . . },
f6 (V ) = {8, 4, 1, 7, 3, X}, f7 (V ) = {8, 3, 6, 1, 8, . . . }, f8 (V ) = {8, 2, 5, X}, f9 (V ) =
{8, 2, 6, X}, and f10 (V ) = {8, 1, 5, X}.
Observe that f1 , f2 , and f7 are labelings for even cycles. Of the remaining
labelings f3 , f4 , f6 , f8 , f9 , and f10 all fail as potential labelings for a cycle
because the missing X labels are forced to be greater than the assumed k(Cn ).
The only labeling left is f5 , which also fails to label a cycle because it does not
provide a repeatable pattern. Therefore, no minimal L(3, 2, 1)-labeling exists
for Cn with odd n and n > 3 such that k(Cn ) = 8.

Lemma 7. k(C4 ) = 8.
Proof. The labeling pattern {1, 6, 3, 8} shows that k(C4 ) ≤ 8. Now, let f be min-
imal L(3, 2, 1)-labeling of C4 and suppose k(C4 ) < 8. Let f (v1 ) = 1 and v2 and
v4 be the two vertices adjacent to v1 . Then the possibilities for {f (v2 ), f (v4 )}
are {4, 6}, {4, 7}, and {5, 7}.
Case I: {f (v2 ), f (v4 )} = {4, 6}:
Then we need f (v3 ) = 9, a contradiction to our assumption that k(C4 ) < 8.
Case II: {f (v2 ), f (v4 )} = {4, 7} or {5, 7}:
Then we need f (v3 ) = 10, a contradiction to the assumed k(C4 ).
Therefore k(C4 ) = 8.

6
Lemma 8. k(C5 ) = 9.
Proof. By Theorem 5 we know that k(C5 ) ≥ 7 since k(P5 ) = 7. Let f be
minimal L(3, 2, 1)-labeling of C5 and suppose k(C5 ) = 7. Let f (v1 ) = 1 and let
v2 and v5 be the two vertices adjacent to v1 . The possibilities for {f (v2 ), f (v5 )}
are {4, 6}, {4, 7}, and {5, 7}.
Case I: {f (v2 ), f (v5 )} = {4, 6}:
Then we need f (v3 ) = 8, a contradiction to our assumption that k(C5 ) < 8.
Case II: {f (v2 ), f (v5 )} = {4, 7} or {5, 7}:
Then we need f (v3 ) = 9, a contradiction to the assumed k(C5 ).
Since k(C5 ) = 7 is impossible, we have k(C5 ) ≥ 8. By Lemma 6 for odd
cycles, we know that k(C5 ) 6= 8. So we have k(C5 ) ≥ 9. The labeling pattern
{5, 1, 7, 3, 9} shows that k(C5 ) ≤ 9.
Therefore k(C5 ) = 9.
Lemma 9. k(C6 ) = 8.
Proof. The labeling pattern {1, 4, 7, 2, 5, 8} shows that k(C6 ) ≤ 8. Now, let f be
minimal L(3, 2, 1)-labeling of C6 and suppose k(C4 ) < 8. Let f (v1 ) = 1 and let
v2 and v6 be the two vertices adjacent to v1 . The possibilities for {f (v2 ), f (v6 )}
are {4, 6}, {4, 7}, and {5, 7}.
Case I: {f (v2 ), f (v6 )} = {4, 6}:
Then we need f (v3 ) = 7 and f (v4 ) = 2. For v5 , we need f (v5 ) = 9, a
contradiction to k(C6 ) < 8.
Case II: {f (v2 ), f (v6 )} = {4, 7} or {5, 7}:
Then we need f (v3 ) = 8, a contradiction.
Therefore k(C6 ) = 8.

Lemma 10. k(C7 ) = 10.

Proof. Let V = {v1 , v2 , . . . , v7 } be the set the vertices of C7 . Suppose k(C7 ) <
10 and let f be a minimal L(3, 2, 1)-labeling of C7 , then the possible values
for f (V ) are in S = {1, 2, . . . , 9}. Since the greatest distance between any
two vertices in V is 3, none of the possible values for f (V ) can be repeated.
Also, only up to two consecutive labels may be used. Using three consecutive
labels is not possible due to the distance constraints of labeling V . Then at
most 6 labels can be used from S, forcing the unlabeled vertex to be labeled
with a value greater than 10, a contradiction to the assumed k(C7 ). Thus,
k(C7 ) ≥ 10. The labeling pattern {1, 5, 8, 2, 10, 7, 4} shows that k(C7 ) ≤ 10.
Therefore, k(C7 ) = 10.
Lemma 11. Let n be an even integer. If n ≥ 4, then n = 4a + 6b; where a and
b are non-negative integers.

7
Proof. Let S = {4a + 6b | a, b ∈ Z; a, b ≥ 0} and let n be an even integer. Then
n ≡ 0 (mod 4) or n ≡ 2 (mod 4). Suppose that n ≥ 4.
Case I: n ≡ 0 (mod 4):
Then there exists an integer q such that n = 4q. Since n ≥ 4, q ≥ 1. This
implies n = 4a + 6b where a = q and b = 0. Then, n ∈ S.
Case II: n ≡ 2 (mod 4):
Then there exists an integer q such that n = 4q + 2. This implies that
n = 4(q − 1) + 6. Since n ≥ 4 and n ≡ 2 (mod 4), we get n ≥ 6. Then
(q − 1) ≥ 0. From this, it follows that n = 4a + 6b where a = (q − 1) and b = 1.
Then, n ∈ S.
Therefore, if n is an even integer and n ≥ 4 then n = 4a + 6b, where a and
b are non-negative integers.

Lemma 12. Let n be an odd integer. If n ≥ 9 and n 6= 11, then n = 4a + 5b;

where a and b are non-negative integers.
Proof. Let S = {4a + 5b | a, b ∈ Z; a, b ≥ 0} and let n be an odd integer. Then
n ≡ 1 (mod 4) or n ≡ 3 (mod 4). Suppose n ≥ 9 and n 6= 11.
Case I: n ≡ 1 (mod 4):
Then there exists an integer q such that n = 4q + 1. This implies that
n = 4(q − 1) + 5. Since n ≥ 9, we need (q − 1) ≥ 1. This implies that
n = 4a + 5b where a = (q − 1) and b = 1. Then, n ∈ S.
Case II: n ≡ 3 (mod 4):
Then there exists an integer q such that n = 4q + 3. This implies that
n = 4(q − 3) + 5(3). Since n ≥ 9, n 6= 11, and n ≡ 3 (mod 4), we get n ≥ 15.
Then we need (q − 3) ≥ 0. Since n = 4a + 5b where a = (q − 3) and b = 3, we
get n ∈ S.
Therefore, if n is an odd integer such that n ≥ 9 and n 6= 11 then n = 4a+5b
for non-negative integers a and b.
Theorem 13. For any cycle, Cn with n ≥ 3,


 7, if n = 3;

8, if n is even;
k(Cn ) =


 9, if n is odd and n 6= 3, 7;
10, if n = 7.


Proof. Let n ≥ 3 and V = {v1 , v2 , . . . , vn } be the vertices of Cn such that for

1 ≤ i < n, (vi , vi+1 ) and (v1 , vn ) are edges in Cn . The following cases describe
all the possibilities for k(Cn ).

8
Case I: n = 3:
Observe that C3 is a complete graph. Therefore, by Theorem 1 for complete
graphs, k(C3 ) = 7.
Case II: n is even:
We have k(C4 ) = 8 by Lemma 7 and we have k(C6 ) = 8 by Lemma 9. We
also know from Theorem 5 for paths that k(Pn ) = 8 for n ≥ 8. Then for all
even cycles, k(Cn ) ≥ 8. Recall the labelings used for C4 and C6 in Lemmas
7 and 9, respectively: for C4 we used f (V ) = {1, 6, 3, 8} and for C6 we used
f (V ) = {1, 4, 7, 2, 5, 8}. Observe that the labeling used for C4 can be repeated
infinitely for all Cn with n a multiple of 4. Likewise, the labeling used for C6
can be repeated infinitely for all Cn with n a multiple of 6. Moreover, the
labelings of C4 and C6 can be joined together to label C10 as such: f (V ) =
{1, 6, 3, 8, 1, 4, 7, 2, 5, 8}. From Lemma 11, we know that every even integer
greater than or equal to 4 can be expressed as some combination of non-negative
multiples of 4 and 6. From this it becomes obvious that the labeling of every
even Cn can be composed of combinations of non-negative multiples of the C4
and C6 labeling patterns. Therefore k(Cn ) = 8 for all even n.
Case III: n is odd and n 6= 3 or 7:
By Lemma 8 we have k(C5 ) = 9, by Theorem 5 for paths we know that
k(Cn ) ≥ 8 for n ≥ 8 since k(Pn ) = 8, and by Lemma 6 we know that k(Cn ) 6= 8
for odd cycles. This implies that for all odd cycles k(Cn ) ≥ 9. In Lemma 8 we
labeled C5 with {5, 1, 7, 3, 9}. Note that labeling of C5 can be repeated infinitely
for all Cn with n a multiple of 5. Also, we can combine the labeling for C5 with
the labeling for C4 used in Lemma 7 to label C9 as such: {5, 1, 7, 3, 9, 1, 6, 3, 8}.
By Lemma 8 we have that every odd integer greater than or equal to 9, with
the exception of 11, can be expressed as some combination of non-negative
multiples of 4 and 5. This implies that every Cn , with n ≥ 9 and n 6= 11, is
composed of combinations of non-negative multiples of C4 and C5 . Therefore,
since k(C5 ) = 9, we have k(Cn ) = 9 for all Cn , where n ≥ 9 and n 6= 11. For
C11 we know that k(C11 ) ≥ 9 (from Lemma 2 and Theorem 3). Define f for C11
such that f (V ) = {1, 6, 3, 8, 5, 1, 9, 6, 2, 8, 4}. Since max(f (V )) = 9, k(C11 ) = 9.
Case IV: n = 7:
By Lemma 10, we have k(C7 ) = 10.

3.5 Caterpillars
Lemma 14. For a uniform caterpillar, Catn with n > 2, k(Catn ) > 2∆ + 2.
Proof. Suppose k(Catn ) ≤ 2∆ + 2. Begin with any vertex, v on the spine,
label it i where 1 ≤ i ≤ 2∆ + 2. Let m be the number of available labels for
the vertices adjacent to v. We will show that m ≤ ∆ if i = 1 or 2∆ + 2 and
m ≤ ∆ − 1 otherwise. To determine m, we count the possible labels, j, for
vertices adjacent to v such that |i − j| ≥ 3. As 1 and 2∆ + 2 are the respective
minimum and maximum possibilities for any label, we know that 1 ≤ j ≤ i − 3

9
or i + 3 ≤ j ≤ 2∆ + 2. Since any pair of vertices adjacent to v has a distance of
2, no two consecutive numbers may be used as labels. Then, we can compute
m by:    
(i − 3) − 1 (2∆ + 2) − (i + 3)
m≤ +1 + +1 (1)
2 2

Case I: i = 1:
As 1 is the lowest possible label for v, all available labels would be greater
than 1. As such, equation 1 becomes:
   
(2∆ + 2) − (1 + 3) 2∆ − 2
m≤ +1 ≤ +1 ≤∆ (2)
2 2
Thus, when v is labeled with with 1, there are ∆ labels available for the ∆
adjacent vertices.
Case II: i = 2∆ + 2:
Then, since 2∆ + 2 is the highest label possible for v, equation 1 becomes:
   
((2∆ + 2) − 3) − 1 2∆ − 2
m≤ +1 ≤ +1 ≤∆ (3)
2 2
Thus, when v is labeled with 2∆ + 2, there are ∆ labels available for the ∆
adjacent vertices.
Case III: 1 < i < 2∆ + 2:
As n > 2, at least one label must be used on the spine aside from 1 and
2∆ + 2. For this spinal vertex, equation 1 is used and yields:
(i − 3) − 1 (2∆ + 2) − (i + 3)
m≤ + +2
2 2
i − 3 − 1 + 2∆ + 2 − i − 3
≤ +2 (4)
2
2∆ − 5 1
≤ +2≤∆− .
2 2
As m must be an integer (there can be no partial labels), for any v labeled
with 1 < i < 2∆ + 2, there are at most ∆ − 1 labels available for ∆ vertices.
Therefore, by the Pigeon-Hole Principle, any vertex labeled 1 < i < 2∆ + 2 on
Catn cannot have k(Catn ) ≤ 2∆ + 2. Thus, k(Catn ) > 2∆ + 2.
Theorem 15. For a uniform caterpillar, Catn , with ∆ > 2 and n > 2,
k(Catn ) = 2∆ + 3.
Proof. The labeling pattern below illustrates the pattern of an L(3, 2, 1)-labeling
with k(Catn ) = 2∆ + 3.

To label the spine, we use the pattern 1, 2∆, 2∆ + 3, 2∆ − 2 repeatedly. To

label the legs adjacent to a vertex labeled i on the spine use positive values from
the set {i ± (3 + 2p) : 0 ≤ p ≤ 2∆}. As k(Catn ) > 2∆ + 2, k(Catn ) = 2∆ + 3
is the lowest possible k.

10
Corollary 16. For any caterpillar Catn with ∆ > 2 and n vertices on the spine,
either k(Catn ) = 2∆ + 2 or k(Catn ) = 2∆ + 3. When there are no more than
two vertices of degree ∆ within any subgraph of four spinal vertices on Catn ,
k(Catn ) = 2∆ + 2.
Proof. This follows from the proof of the uniform caterpillar: only vertices
labeled 1 and 2∆ + 2 had ∆ label options for ∆ vertices. All other spinal labels
yielded at most ∆ − 1 possible labels. Thus, all spinal vertices labeled with
1 < i < 2∆ + 2 must have a degree of at most ∆ − 1, in order to keep the
largest label to 2∆ + 2 or less (case III, Theorem 14). Also, cases I and II of
Theorem 14 show that a non-uniform caterpillar with k(Catn ) = 2∆ + 2 may
only have a degree of ∆ at vertices labeled with 1 and 2∆ + 2. Notice that
the labels 1 and 2∆ + 2 can be repeated every four labels and that using this
pattern maximizes the total number of vertices on the caterpillar. The pattern
below illustrates one labeling of the vertices that fulfills this requirement and
maximizes the quantity of vertices.

3.6 N-ary Tree

Lemma 17. For any n-ary tree, G, k(G) ≤ 2n + 6.
Proof. This proof is by construction. Let G = (V, E) be an n-ary tree. First,
observe that the induced subgraph of each vertex and all of the ∆ vertices
adjacent to it form a star. With this observation, we propose the following
L(3, 2, 1)-labeling of each star: let f be a minimal L(3, 2, 1)-labeling of a star
with central vertex x and define f such that the adjacent vertices of x are labeled
with the lowest positive integers from the set {f (x) ± (3 + 2i) : i ∈ Z+ ∪ {0}}.
In other words, adjacent vertices of x are first labeled with as many positive
elements as possible from the set {f (x) − 3, f (x) − 5, f (x) − 7 . . . } and the re-
maining adjacent vertices are labeled with the lowest positive elements from the
set {f (x) + 3, f (x) + 5, f (x) + 7, . . . }.
We will show that these stars can be joined via overlapping edges to create
an n-ary tree such that f is an L(3, 2, 1)-labeling. It follows from construction
that for any two adjacent vertices x and y, |f (x) − f (y)| ≥ 3. It also follows
from construction that for any two vertices x and y that are mutually adjacent
to a vertex w, |f (x) − f (y)| ≥ 2. As for vertices x, y with d(x, y) = 3, we observe
that for each star (induced subgraph of G), with central vertex x: if f (x) is even
then for all vertices y adjacent to x, f (y) is odd; and if f (x) is odd then for all
vertices y adjacent to x, f (y) is even. Hence for vertices x, y with d(x, y) = 3, x
and y have different parity. Thus, d(x, y) = 3 implies |f (x) − f (y)| ≥ 1. Thus,
f is an L(3, 2, 1)-labeling for any n-ary tree.
Claim: The highest value of f used in the proposed L(3, 2, 1)-labeling is 2∆+
4. Note that the worst case for the proposed L(3, 2, 1)-labeling is when the cen-
tral vertex, x, of a star has f (x) = 3. This is because none of the vertices adja-
cent to x can be labeled using elements of the set {f (x) − 3, f (x) − 5, f (x) − 7, . . . }.

11
Hence, since all the labels for the vertices adjacent to x must come from the set
{f (x) + 3, f (x) + 5, f (x) + 7, . . . }, the highest label needed is f (x) + 3 + 2(∆ −
1) = 2∆ + 4.
Therefore, since ∆ = n + 1, k(G) ≤ 2n + 6.

Lemma 18. Let G = (V, E) be an n-ary tree. Suppose k(G) < 2n + 6. Let f
be a minimal L(3, 2, 1)-labeling of G, and let x be a vertex with f (x) = 1. Let
A = {a1 , a2 , . . . , a∆ }, be the set of vertices adjacent to x. Neither 6 nor 7 can
appear as a label for the elements of A.
Proof. We consider the cases when 6 or 7 is used as a label for elements of A.
Case I: Suppose there is a vertex a1 in A, such that f (a1 ) = 6.
Let B = {b1 , b2 , . . . , b∆−1 } be the set of vertices adjacent to a1 , not including
x. (Recall that f (x) = 1). By the definition of an L(3, 2, 1)-labeling, f (bi ) differs
by at least 3 from f (a1 ) = 6 for all i. This implies that the minimum label used
in B is 3 or 9. If the minimum label is 3, then there is a vertex, b1 in B such
that f (b1 ) = 3. Let C = {c1 , c2 , . . . , c∆ } be the set of vertices adjacent to b1 .
By the definition of an L(3, 2, 1)-labeling, f (ci ) ≥ f (b1 ) + 3 = 6 for all i. Then
there exists ci in C such that f (ci ) ≥ 6 + 2(∆ − 1) = 2∆ + 4 = 2n + 6, a
contradiction to our assumption that k(G) < 2n + 6. If the minimum label used
in B is 9, then for all bi in B, f (bi ) ≥ 9. This implies that there is bj in B such
that f (bj ) ≥ 9 + 2(∆ − 2) = 2∆ + 5 = 2n + 7. This contradicts our assumption
that k(G) < 2n + 6. Therefore, 6 cannot appear as a vertex label in A.
Case II: Suppose there is a vertex in A labeled 7, say f (a1 ) = 7:
Let B = {b1 , b2 , . . . , b∆−1 } be the set of vertices adjacent to a1 , not including
x. The elements of B are distance 2 away from x and they are distance 3 away
from any non-adjacent element of A. By the definition of an L(3, 2, 1)-labeling
the smallest vertex label we could use for elements in B is 3. However, using 3
as a vertex label in B leads to a contradiction as we have shown in case I. The
same argument also shows that the smallest label in B cannot be more than 3.
Therefore, 7 cannot appear as a vertex label in A.

Lemma 19. Let G = (V, E) be an n-ary tree. Suppose k(G) < 2n + 6. Let f
be a minimal L(3, 2, 1)-labeling of G, and let x be a vertex with f (x) = 1. Let
A = {a1 , a2 , . . . , a∆ } be the set of vertices adjacent to x. The only possible sets
of labels for the vertices of A are {4, 6, 8, . . . , 2∆ + 2}, {5, 7, 9, . . . , 2∆ + 3} and
a set of the form {4, . . . , 2∆ + 3}. The last set contains even numbers up to a
certain point then odd numbers afterwards.
Proof. Let x and A be defined as above. Then by definition of L(3, 2, 1)-labeling,
the smallest label of any vertex in A is 4. Also, by our assumption on k(G) no
label of any vertex of A can be bigger than 2∆ + 3. Since any two elements of A
are distance 2 apart in G, their labels must be at least 2 apart. In the first set
there are ∆ even labels that we can use for labeling the vertices in A. Likewise,

12
there are enough labels in the sets {4, . . . , 2∆ + 3} and {5, 7, 9, , 2∆ + 3} for
labeling each element of A, so we can use these sets in an L(3, 2, 1)-labeling.
Now we show that no other sets are possible. Suppose the minimum label
used for elements of A was at least 6. Then in order to have enough labels, a
label at least 6 + 2(∆ − 1) = 2∆ + 4 = 2n + 6 would need to be used, which
contradicts the assumption k(G) < 2n + 6. Therefore, the smallest label must
be less than 6. If the smallest label is 5, or if we use only even labels, the labels
are {5, 7, 9, . . . , 2∆ + 3}, and {4, 6, 8, . . . , 2∆ + 2}, respectively. Otherwise the
labels must come from the set {4, . . . , 2∆ + 3}.

Theorem 20. For any n-ary tree, G, k(G) = 2n + 6.

Proof. Let G be an n-ary tree. By Lemma 17 we know that k(G) ≤ 2n + 6. If
k(G) < 2n + 6 then, by Lemma 18, we know that neither 6 nor 7 can appear as
a label in A. By Lemma 19 we know that the labels of A must come from one of
the following sets: {4, 6, 8, . . . , 2∆ + 2}, {5, 7, 9, . . . , 2∆ + 3} or one of the sets
{4, . . . , 2∆ + 3}. Notice that 6 is an element of the set {4, 6, 8, . . . , 2∆ + 2}, 7
is an element of the set {5, 7, 9, . . . , 2∆ + 3}, and either 6 or 7 is an element of
any set of the form {4, . . . , 2∆ + 3}. Therefore none of these sets can provide
the labels for A.
Therefore, k(G) = 2n + 6.

3.7 The Upper Bound in Terms of Maximum Degree, ∆

Theorem 21. If G is a graph with maximum degree ∆, k(G) ≤ ∆3 + ∆2 + 3∆.
Proof. Let Ni (v) be the set of vertices of distance i from a vertex v; the members
of the sets of 1 ≤ i ≤ 3 shall be referred to as neighbors of v. When constructing
a greedy L(3, 2, 1)-labeling f on a graph G, each vertex surrounding v is given
the lowest available label such that no two labels are repeated within distance
3 of v. Furthermore, each neighbor of v blocks a certain number of labels from
use: N1 (v) blocks up to 5 labels, N2 (v) up to 3, and N3 (v) prohibits the use of
the label given to itself. The maximum label needed for v is calculated through a
sum of the quantity of vertices within distance three of v while weighting each set
of neighbors according to the quantity of labels each blocks. As the maximum
degree of G is at most ∆, |N1 | ≤ ∆, |N2 | ≤ ∆(∆ − 1), and |N3 | ≤ ∆(∆ − 1)2 .
Thus, k(G) ≤ 5∆ + 3∆(∆ − 1) + ∆(∆ − 1)2 , or k(G) ≤ ∆3 + ∆2 + 3∆.

13
References
[1] G. Chartrand, D. Erwin, F. Harary, and P. Zhang, Radio labeling of graphs,
Bull. Inst. Combin. Appl., 33 (2001), 77-85.
[2] J. R. Griggs and R. K. Yeh, Labeling graphs with a condition at distance
two, SIAM J. Discrete Math., 5 (1992) 586-5995.
[3] W. K. Hale, Frequency assignment: theory and application, Proc. IEEE, 68
(1980), 1497-1514.
[4] L. Jia-zhuang and S. Zhen-dong, The L(3, 2, 1)-labeling problem on graphs,
Mathematica Applicata, 17 (4) (2004), 596-602.
[5] K. Jonas, Graph coloring analogues with a condition at distance two:
L(2,1)-labelling and list λ-labellings, Ph.D. thesis, University of South Car-
olina (1993), 8-9.
[6] F. S. Roberts, T-colorings of graphs: recent results and open problems,
Discrete Math., 93 (1991), 229-245.

14