ArticlePDF Available

Braodcast Chromatic Numbers of Graphs.


Abstract and Figures

A function � : V ! {1,...,k} is a broadcast coloring of order k if �(u) = �(v) implies that the distance between u and v is more than �(u). The minimum order of a broadcast coloring is called the broadcast chromatic number of G, and is denotedb(G). In this pa- per we introduce this coloring and study its properties. In particular, we explore the relationship with the vertex cover and chromatic num- bers. While there is a polynomial-time algorithm to determine whether �b(G) � 3, we show that it is NP-hard to determine ifb(G) � 4. We also determine the maximum broadcast chromatic number of a tree, and show that the broadcast chromatic number of the infinite grid is finite.
Content may be subject to copyright.
Broadcast Chromatic Numbers of Graphs
Wayne Goddard, Sandra M. Hedetniemi, Stephen T. Hedetniemi
Clemson University
John M. Harris, Douglas F. Rall
Furman University
A function π:V→ {1,...,k}is a broadcast coloring of order k
if π(u) = π(v) implies that the distance between uand vis more
than π(u). The minimum order of a broadcast coloring is called the
broadcast chromatic number of G, and is denoted χb(G). In this pa-
per we introduce this coloring and study its properties. In particular,
we explore the relationship with the vertex cover and chromatic num-
bers. While there is a polynomial-time algorithm to determine whether
χb(G)3, we show that it is NP-hard to determine if χb(G)4. We
also determine the maximum broadcast chromatic number of a tree,
and show that the broadcast chromatic number of the infinite grid is
1 Introduction
The United States Federal Communications Commission has established
numerous rules and regulations concerning the assignment of broadcast fre-
quencies to radio stations. In particular, two radio stations which are as-
signed the same broadcast frequency must be located sufficiently far apart
so that neither broadcast interferes with the reception of the other. The
0Correspondence to: W. Goddard, Dept of Computer Science, Clemson University,
Clemson SC 29634-0974, USA
geographical distance between two stations which are assigned the same fre-
quency is, therefore, directly related to the power of their broadcast signals.
These frequency, or channel, assignment regulations have inspired a vari-
ety of graphical coloring problems. One of these is the well-studied L(2,1)-
coloring problem [2]. Let d(u, v) denote the distance between vertices u
and v, and let e(u) denote the eccentricity of u. Given a graph G= (V, E),
an L(2,1)-coloring is a function c:V→ {0,1,...}such that (i) d(u, v) = 1
implies |c(u)c(v)| ≥ 2, and (ii) d(u, v) = 2 implies |c(u)c(v)| ≥ 1. For
a survey of frequency assignment problems, see [4].
In a similar way, Dunbar et al. [1] define a function b:V→ {0,1,...}to
be a dominating broadcast if for every uV(i) b(u)e(u), and (ii) b(u) = 0
implies there exists a vertex vVwith b(v)>0 and d(u, v)b(v). A
broadcast is called independent if b(u) = b(v) implies that d(u, v)> b(u);
that is, broadcast stations of the same power must be sufficiently far apart
so that neither can hear each other’s broadcast.
In this paper we introduce a new type of graph coloring. A function
π:V→ {1,...,k}is called a broadcast coloring of order kif π(u) = π(v)
implies that d(u, v)> π(u). The minimum order of a broadcast coloring of a
graph Gis called the broadcast chromatic number, and is denoted by χb(G).
Equivalently, a broadcast coloring is a partition Pπ={V1, V2,...,Vk}of V
such that each color class Viis an i-packing (pairwise distance more than i
apart). Note that in particular, every broadcast coloring is a proper coloring.
Also, if His a subgraph of G, then χb(H)χb(G).
Throughout this article, we assume that graphs are simple: no loops or
multiple edges. For terms and concepts not defined here, see [3]. In particu-
lar, we shall use the following notation: α0(G) for the vertex cover number,
β0(G) for the independence number, χ(G) for the chromatic number, ω(G)
for the clique number, and ρr(G) for the largest cardinality of an r-packing.
2 Basics
Every graph Gof order nhas a broadcast coloring of order n, since one can
assign a distinct integer between 1 and nto each vertex in V. There is a
better natural upper bound.
Proposition 2.1 For every graph G,
χb(G)α0(G) + 1,
with equality if Ghas diameter two.
Proof. For the upper bound, give color 1 to every vertex in a maximum
independent set in G. Then give every other vertex a distinct color. Since
nβ0(G) = α0(G) by Gallai’s theorem, the result follows.
If a graph has diameter two, then no two vertices can receive the same
color i, for any i2. On the other hand, since the vertices which receive the
color 1 form an independent set, there are at most β0(G) such vertices. 2
From this it follows that the complete, the complete multipartite graphs,
and the wheels have broadcast chromatic number one more than their ver-
tex cover number. It also follows that computing the broadcast chromatic
number is NP-hard (since vertex cover number is NP-hard for diameter 2).
If a graph is bipartite and has diameter three, then there is also near
equality in Proposition 2.1.
Proposition 2.2 If Gis a bipartite graph of diameter 3, then α0(G)
χb(G)α0(G) + 1.
Proof. By the diameter constraint, each color at least 3 appears at most
once. Since the graph is bipartite of diameter 3, color 2 cannot be used
twice on the same partite set, and thus can be used only twice overall. 2
For an example of equality in the upper bound, consider the 6-cycle;
for the lower bound, consider the 6-cycle where two antipodal vertices have
been duplicated.
We now present the broadcast chromatic numbers for paths and cycles.
Proposition 2.3 For 2n3,χb(Pn) = 2; and for n4,χb(Pn) = 3.
Proof. The first nentries in the pattern below represent a broadcast col-
oring for Pn.
This is clearly best possible. 2
Proposition 2.4 For n3, if nis 3or a multiple of 4, then χb(Cn) = 3;
otherwise χb(Cn) = 4.
Proof. Since the cycle contains either P4or K3,χb(Cn)3. Let v0, v1,
...,vn1, v0be the vertices of Cnand suppose there is a broadcast coloring
of order 3 with n4.
Then there cannot be two consecutive vertices neither of which has
color 1. For suppose that v2has color 2 and v3has color 3, say. Then
neither v0nor v1can receive color 2 or 3, and only one can receive color 1,
a contradiction. Since vertices with color 1 cannot be consecutive, it follows
that the vertices with color 1 alternate. In particular, nis even.
Say the even-numbered vertices have color 1. But then no two consec-
utive odd-numbered vertices can receive the same color, and so they must
alternate between colors 2 and 3. In particular, nis a multiple of 4. It
follows that if nis not a multiple of 4, then χb(Cn)4.
We consider now optimal broadcast colorings. For cycles with order na
multiple of 4, the pattern
is a broadcast coloring. When nis not a multiple of 4, the pattern consists
of repeated blocks of “1,2,1,3” with an adjustment at the very end:
n= 4r+ 1 : 1,2,1,3,1,2,1,3,...,1,2,1,3,4
n= 4r+ 2 : 1,2,1,3,1,2,1,3,...,1,2,1,3,1,4
n= 4r+ 3 : 1,2,1,3,1,2,1,3,...,1,2,1,3,1,2,4
Hence, if nis a multiple of 4, then χb(Cn)3; otherwise χb(Cn)4. 2
Just as the natural upper bound involves the vertex cover number, the
natural lower bound involves the chromatic number.
Proposition 2.5 For every graph G,
It would be nice to characterize those graphs where the broadcast chro-
matic number is equal to the clique number. It is certainly necessary that
the neighbors of any maximum clique form an independent set, and at least
one vertex of such a clique has no neighbors outside the clique (so it can
receive color 1). If the graph Gis a split graph, then this necessary condition
is sufficient. (Recall that a split graph is a graph whose vertex set can be
partitioned into two sets, Aand B, where Ainduces a complete subgraph
and Bis an independent set.)
On the other hand, a necessary condition for χb(G) = χ(G) is that the
clique number be large.
Proposition 2.6 For every graph G, if χb(G) = χ(G)then ω(G)χ(G)2.
Proof. For, assume χb(G) = χ(G) = m. Consider the broadcast coloring as
a proper coloring. If one can reduce the color of every vertex colored mwhile
still maintaining a proper coloring, then one has a contradiction. So there
exists a vertex vmthat has a neighbor of each smaller color: say v1,...,vm1
with vihaving color i(viis unique for i2). Now, if one can reduce the
color of the vertex vm1, this too will enable one to lower the color of vm.
It follows that vm1has a neighbor of each smaller color. By the properties
of the broadcast coloring, that neighbor must be vifor i3. By repeated
argument, it follows that the vertices v3, v4,...,vmform a clique. 2
In another direction, we note that if one has a broadcast coloring, then
one can choose any one color class Vito be a maximal i-packing. (Recolor
vertices far away from Viwith color iif necessary.) However, one cannot
ensure that all color classes are maximal i-packings.
3 Graphs with Small Broadcast Chromatic Num-
We show here that there is an easy algorithm to decide if a graph has broad-
cast chromatic number at most 3. In contrast, it is NP-hard to determine
if the broadcast chromatic number is at most 4. We start with a character-
ization of graphs with broadcast chromatic number 2.
Proposition 3.1 For any connected graph G,χb(G) = 2 if and only if G
is a star.
Proof. We know that the star K1,m has broadcast chromatic number 2.
So assume χb(G) = 2. Then Gdoes not contain P4and diam(G)2. By
Proposition 2.1, it follows that α0(G) = 1; that is, Gis a star. 2
As regards those graphs with χb(G) = 3, we start with a characterization
of the blocks with this property. If Gis a graph, then we denote by S(G)
the subdivision graph of G, which is obtained from Gby subdividing every
edge once. In S(G) the vertices of Gare called the original vertices; the
other vertices are called subdivision vertices.
Proposition 3.2 Let Gbe a 2-connected graph. Then χb(G) = 3 if and
only if Gis either S(H)for some bipartite multigraph Hor the join of K2
and an independent set.
Proof. Assume that χb(G) = 3. Let π:V(G)→ {1,2,3}be a broadcast
coloring and let Vi=π1(i), for 1 i3. It follows that Viis an i-packing
for each i.
Let vV1. The set V1is an independent set, so N(v)V2V3. But v
has at most one neighbor in V2, since V2is a 2-packing. Similarly, vhas at
most one neighbor in V3. Since vhas at least two neighbors, it follows that
vhas degree 2 and is adjacent to exactly one vertex in each of V2and V3.
It follows from Proposition 2.4 that the length of any cycle in Gis either 3
or a multiple of 4. Assume Gcontains a triangle {x, y, z}. These vertices
receive different colors; say xV2and yV3. Then for every neighbor t
of xapart from y, it follows that tV1and (since it is too close to yto
have another neighbor of color 3) that N(t) = {x, y}. Since ycannot be a
cut-vertex, it follows that this is the whole of G.
So assume that every cycle length is a multiple of 4. By the proof of
Proposition 2.4, in any broadcast coloring of order 3 of a cycle, every alter-
nate vertex receives color 1. It follows that V2V3is an independent set.
(Since Gis a block, every edge lies in a cycle.) In particular, Gis the subdi-
vision of some multigraph Hwhere every subdvision vertex receives color 1.
Furthermore, since V2and V3are 2-packings in G, they are independent sets
in H; that is, (V2, V3) is a bipartition of H.
Conversely, to broadcast color the subdivision of a bipartite multigraph,
take V1as the subdivision vertices, and (V2, V3) as the original bipartition. 2
Now, in order to characterize general graphs with broadcast chromatic
number 3, we define a T-add to a vertex vas introducing a vertex wvand
a set Xvof independent vertices, and adding the edge vwvand some of the
edges between {v, wv}and Xv. By extending the above result one can show:
Proposition 3.3 Let Gbe a graph. Then χb(G) = 3 if and only if Gcan
be formed by taking some bipartite multigraph Hwith bipartition (V2, V3),
subdividing every edge exactly once, adding leaves to some vertices in V2V3,
and then performing a single T-add to some vertices in V3.
Thus there is an algorithm for determining whether a graph has broad-
cast chromatic number at most 3. The key is that the colors 2 and 3 can
seldom be adjacent. In particular, if vertices uand vare adjacent with u
with color 2 and vwith color 3, then any neighbor aof uapart from vhas
N(a)⊆ {u, v}. Apart from that, V2V3must be an independent set, while
every vertex of V1has degree at most 2. In particular, if two vertices with
degree at least 3 are joined by a path of odd length, then at one of the ends
of this path there must be two consecutive V2V3vertices.
So a graph can be tested for having broadcast chromatic number 3 by
identifying the places where V2and V3must be adjacent, coloring and trim-
ming these appropriately, trimming leaves that are in V1, and then seeing
whether what remains with the partial coloring is a subdivision of a bipartite
graph. We omit the details.
4 Intractable Colorings
In contrast to the above, the problem of determining whether a graph has
a broadcast 4-coloring is intractable. We will need the following gener-
alization. For a sequence of positive integers s1s2... sk, an
(s1, s2,...,sk)-coloring is a weak partition π= (V1, V2, . . . , Vk), where Vj
is an sj-packing for 1 jk. Then we define the decision problem:
(s1, s2,...,sk)-Coloring
Instance: Graph G
Question: Does Ghave an (s1, s2,...,sk)-coloring?
For example, a 3-coloring is a (1,1,1)-coloring. The Broadcast 4-
Coloring problem is equivalent to the (1,2,3,4)-Coloring problem. We
will need the intractability of a special 3-coloring problem.
Proposition 4.1 (1,1,2)-Coloring is NP-hard.
Proof. The proof is by reduction from normal 3-coloring. The reduction
is to form G0from Gas follows. Replace each edge uv by the following:
add a pentagon Puv and join u, v to nonadjacent vertices of Puv; add a
pentagon Quv and join u, v to adjacent vertices of Quv and add an edge
joining two degree-two vertices of Quv. (See Figure 1.) The vertices u, v are
called original in G0.
We claim that G0has a (1,1,2)-coloring iff Gis 3-colorable.
Assume that Ghas a 3-coloring with colors red, blue and gold. We will
3-color G0such that the gold vertices form a 2-packing. Start by giving the
original vertices of G0their color in G. For each two adjacent vertices u, v
of G, color gold one vertex from each of Puv and Quv chosen as follows.
u v
Figure 1: Replacing an edge uv
For Quv it is the degree-2 vertex; for Puv it is the degree-2 vertex that is
distance-3 from whichever of uor vis gold if one of them is gold, and it is
the degree-2 vertex at distance 2 from both otherwise. Then it is easy to
color the remaining vertices in Puv and Quv with red and blue.
Conversely, suppose G0has a (1,1,2)-coloring πwhere the gold vertices
form a 2-packing. Then adjacent vertices uand vcannot have the same
color. For, if they are both red or both blue, then there is no possible
coloring of Quv (since one of the vertices in the triangle is gold); if they are
both gold, then there is no possible coloring of Puv. That is, restricted to
V(G), the coloring πis a 3-coloring of G.2
Theorem 4.2 Broadcast 4-Coloring is NP-hard.
Proof. We reduce from (1,1,2)-Coloring as follows. Given a connected
graph H, form graph H0by quadrupling each edge and then subdividing each
edge. Thus, all original vertices have degree at least 4 and all subdivision
vertices have degree 2.
The (1,1,2)-coloring of Hwith red, blue and gold, becomes a broadcast
4-coloring (V1, V2, V3, V4) of H0by making V1all the subdivision vertices, V2
all the red vertices, V3all the blue vertices and V4all the gold vertices. On
the other hand, in a broadcast 4-coloring of H0, none of the original vertices
can receive color 1. If we maximize the number of vertices receiving color 1,
it follows that all subdivision vertices receive color 1, and all original vertices
receive color 2, 3, or 4. So, the vertices colored 2 or 3 form an independent
set in Hand the vertices colored 4 form a 2-packing in H. Thus we have a
(1,1,2)-coloring of H.2
Comment: This proves that Broadcast 4-Coloring is also NP-hard
for planar graphs. It is an open question what the complexity is for cubic
or 4-regular graphs. In another direction, it is easy to determine whether a
graph is (2,2,2)-colorable (as only paths of any length and cycles of length
a multiple of 3 are). But what is the complexity of (1,2,2)-Coloring?
5 Trees
We are interested in trees with large broadcast chromatic numbers. In order
to prove the best possible general result, it is necessary to examine the small
A tree of diameter 2 (that is, a star) has broadcast chromatic number 2.
A tree of diameter 3 has broadcast chromatic number 3. The case of a
tree of diameter 4 is more complicated, but one can still write down an
explicit formula. We say that a vertex is large if it has degree 4 or more,
and small otherwise. The key to the formula is the numbers of large and
small neighbors of the central vertex.
Proposition 5.1 Let Tbe a tree of diameter 4with central vertex v. For
i= 1,2,3, let nidenote the number of neighbors of vof degree i, and let L
denote the number of large neighbors of v. If L= 0 then
χb(T) = (4if n32and n1+n2+n33
and if L > 0then
χb(T) =
L+ 3 if n31and n1+n2+n32
L+ 1 if n1=n2=n3= 0
L+ 2 otherwise.
2 5 6
Figure 2: The tree T5: the unlabeled vertices have color 1
Proof. To show the upper bound we need to exhibit a broadcast coloring.
A simple coloring is to color the center and the leaves not adjacent to
it with color 1, and the remaining vertices with unique colors. This uses
L+n3+n2+n1+ 1 colors. It is optimal when n1=n2= 0 and either
2n33 and L= 0 or 0 n32 and L > 0. (Note that L+n3+n22
by the diameter condition.)
Another good coloring is as follows: put color 1 on the small neighbors
of the center vand on the children of large neighbors; put color 2 on one
large neighbor (if one exists) and on one child of each small neighbor; put
color 3 on the remaining children of small neighbors; and put unique colors
on the remaining vertices. If L= 0, then this uses 4 colors if n3>0 and 3
values otherwise. If L > 0, then this uses L+ 3 colors if n3>0 and L+ 2
colors otherwise. This coloring is illustrated in Figure 2.
The only case not covered by the above two colorings is when L= 0
and n3= 1. In this case 3 colors suffice: use color 3 on the central vertex,
color 2 on its degree-3 neighbor and the children of its degree-2 neighbors,
and color 1 on the remaining vertices.
For a lower bound, proceed as follows. If the center is colored 1, then
the coloring uses L+n3+n2+n1+ 1 colors, which is at least the above
bound. So we may assume that the center is not colored 1.
Figure 3: The smallest trees with χb(T) = 4
If a large neighbor receives any color other than 2, then either it or one
of its children receives a unique color. Thus we may remove it and induct.
So we may assume that every large neighbor of the center receives color 2.
This means there is at most one large neighbor.
At least three colors are always needed. For the case that L= 0, it is
enough to argue that 4 colors are needed if n3= 2, n2= 0 and n1= 1 (as
any other case contains this as a subgraph). (This is tree A8in Figure 3.) If
any degree-3 vertex receives color 1, then three more colors are needed for
its neighbors. On the other hand, the degree-3 vertices induce a P3, and so
require three new colors if 1 is not used.
In fact, this observation also takes care of the case where there is only
one large neighbor. 2
Proposition 5.2 The minimum order of a tree with broadcast chromatic
number 2is 2. For 3it is 4and for 4it is 8. Furthermore, P4is the unique
tree on 4vertices that needs 3colors. The two trees on 8 vertices that need
4colors are (i) the diameter-4tree with n3= 2,n1= 1 and L=n2= 0,
called A8; and (ii) the diameter-6tree where the two central vertices have
degree-3and for each central vertex its three neighbors have degrees 1,2and
3respectively, called B8. (These are depicted in Figure 3.)
Proof. By the above result, A8is the unique smallest tree with diame-
ter 4 that needs four colors. So we need only examine the small trees with
diameter 5 or more, which is easily done. 2
In another direction there is an extension result.
Proposition 5.3 Let Tbe a graph but not P4. Suppose Tcontains a path
t, u, v, w where thas degree 1and uand vhave degree 2. Then χb(T) =
Proof. Take an optimal broadcast coloring of Tt. Since T6=P4,Tt
contains P4and hence uses at least three colors.
If ureceives any color except 1, then one can color twith 1. So assume
ureceives color 1. If vreceives any color except 2, then one can color twith
color 2. So assume vreceives color 2. If wreceives any color except 3, then
one can color twith color 3. So assume wreceives color 3. But then one
can recolor as follows: vgets color 1, ugets color 2, and tgets color 1. 2
For example, this shows that χb(Pn) = 3 for all n4.
We are now ready to determine the maximum broadcast chromatic num-
ber of a tree. An extremal tree Tdfor d2 is constructed as follows: it
has diameter 4; n1=n3= 1, n2= 0, L=d2 and all large vertices have
degree exactly 4. The tree Tdhas 4d3 vertices and χb(Td) = d+ 1. The
tree T5is shown in Figure 2.
Theorem 5.4 For all trees Tof order nit holds that χb(T)(n+ 7)/4,
except when n= 4 or 8, when the bound is 1/4more, and these bounds are
Proof. We have shown sharpness above. The proof of the bound is by
induction on n.
If n8 the result follows from Proposition 5.2. If diam(T)3, then
χb(T)3. If diam(T) = 4, then the bound follows from Proposition 5.1.
So assume n9 and diam(T)5.
Define a penultimate vertex as one with exactly one non-leaf neighbor.
Suppose some penultimate uhas degree 4 or more. Define T0to be the
tree after the removal of uand all its leaf-neighbors. Then take an optimal
broadcast coloring of T0, and extend to a broadcast coloring of Tby giving u
a new unique color and its leaf neighbors color 1. By the inductive hypothesis
it follows that χb(T)χb(T0) + 1 (n+ 3)/4 + 1 = (n+ 7)/4, unless T0is
an exceptional tree.
But the three exceptional trees can be broadcast colored with 4 colors
so that, for any specific vertex v, neither it nor any of its neighbors receives
color 2. So let vbe u’s other neighbor and color T0thus; then color uwith
color 2 and its leaf-neighbors with color 1. So, in this case χb(T)4, which
establishes the bound.
So we may assume that every penultimate vertex has degree 2 or 3.
Now, define a late vertex as one that is not a penultimate, but at most
one of its neighbors is not a penultimate or a leaf. (For example, the third-
to-last vertex on a diametrical path.) For a late vertex v, define Tvas the
subtree consisting of v, all its penultimate neighbors, and any leaf adjacent
to one of these. Since diam(T)5, Tvis not the whole of T; let wbe v’s
other neighbor. Then define T0=TTv.
If |Tv|= 3, then vhas degree 2 and its penultimate neighbor has degree 2.
So by the above result, χb(T) = χb(T0) and we are done. Therefore we may
assume that |Tv| ≥ 4.
Give T0an optimal broadcast coloring. Then, if wis not colored 3, one
can extend this to a broadcast coloring of Tby giving va new unique color,
all its neighbors in Tvcolor 1 and the remaining vertices of Tvcolors 2 or 3.
We are done by induction—one new color for at least four vertices—unless
|Tv|= 4 and T0is an exceptional tree. But in this case one can readily argue
that χb(T)4.
So assume the vertex wreceives color 3 in any coloring of T0. (In partic-
ular, this means that T0is not one of the exceptional trees.) Now, we can
afford to recolor wwith a new color and proceed as above if |Tv| ≥ 8. So
assume that |Tv| ≤ 7.
If vhas only one penultimate neighbor of degree 3, then one can color Tv
with colors 1 and 2 except for v, and so are done. So we may assume that v
has two degree-3 neighbors. But then vhas exactly two neighbors in Tvand
these have degree 3. But then one can color Tvwith colors 1 and 2, except
for one neighbor of v, with vreceiving color 1. And hence we are done by
the inductive hypothesis. 2
6 Grids
We will now investigate broadcast colorings of grids Gr,c with rrows and c
columns. The exact values for r5 are given in the following result
Proposition 6.1 χb(G2,c) = 5 for c6;χb(G3,c) = 7 for c12;χb(G4,c) =
8for c10; and χb(G5,c) = 9 for c10. The values for smaller grids are
as follows:
m\n2 3 4 5 6 7 8 9 10 11 12
2 3 4 4 4 5 . . .
3 4556666667. . .
4 5777778...
5 7 7 7 8 8 9 . . .
Proof. Consider the following coloring pattern. For c2, the first c
columns indicate that the values for χb(G2,c) stated are in fact upper bounds.
Consider the following coloring pattern. For c3, the first ccolumns
indicate that the values for χb(G3,c) stated are in fact upper bounds.
Consider the following coloring pattern. For c4, the first ccolumns
indicate that the values for χb(G4,c) stated are in fact upper bounds.
Consider the following coloring pattern. Let iand jdenote the row and
column of a vertex, with 1 irand 1 jc. Then assign color 1 to
every vertex with i+jodd; assign color 2 to every vertex with iand jodd
and i+jnot a multiple of 4; and assign color 3 to every other vertex with
iand jodd. The picture looks as follows.
2 1 3 1 2 1 3 1 . ..
1 – 1 – 1 – 1 – . ..
3 1 2 1 3 1 2 1 . ..
1 – 1 – 1 – 1 – . ..
2 1 3 1 2 1 3 1 . ..
(The uncolored vertices induce a copy of the grid in the square of the graph.)
For G5,c the uncolored vertices should be colored as follows:
It is to be noted that the patterns for G5,8and G5,9are exceptions.
Lower bounds in general can be verified by computer. Some can be
verified by hand. One useful idea is the following. For the lower bound for
G2,c where c6, note that any copy of G2,3contains a color greater than 3.
If one considers the three columns after a column containing a 4, then that
G2,3has at least one of its vertices colored 5 or greater. 2
The following table provides some more upper bounds. An asterisk in-
dicates an exact value.
m\n6 7 8 9 10 11 12 13 14 15 16
6 89999910 10 10 11 11
7 9910 10 11 11 11 11 12 12
Often, the greedy approach produces a bound close to optimal. By the
time the grids have around 20 rows (together with several hundred columns),
the greedy approach uses more than 25 colors. As the following theorem
implies, these bounds are not best possible.
Theorem 6.2 For any grid Gm,n,χb(Gm,n)23.
Proof. There is a broadcast coloring of the infinite grid that uses 23 colors.
The coloring is illustrated below. This provides a broadcast coloring of any
finite subgrid.
As in the coloring of G5,n above, we start by coloring with 1s, 2s and 3s
such that the uncolored vertices occur in every alternate row and column.
Then the following coloring is used to tile the plane.
4 5 8 4 5 9 4 5 8 4 5 9
10 6 11 7 12 6 10 7 11 6 13 7
5 4 9 5 4 8 5 4 9 5 4 8
14 7 15 6 13 7 16 6 17 7 12 6
4 5 18 4 5 11 4 5 19 4 5 11
20 6 21 7 10 6 14 7 15 6 10 7
5 4 8 5 4 9 5 4 8 5 4 9
13 7 11 6 22 7 12 6 11 7 23 6
4 5 9 4 5 8 4 5 9 4 5 8
12 6 10 7 15 6 13 7 10 6 14 7
5 4 17 5 4 11 5 4 18 5 4 11
16 7 19 6 14 7 20 6 21 7 15 6
The coloring was found by placing the colors 4 through 9 in a specific pattern,
and then using a computer to place the remaining colors. 2
Schwenk [5] has shown that the broadcast chromatic number of the in-
finite grid is at most 22.
7 Other Grid-like Graphs
While the broadcast chromatic number of the family of grids is bounded,
this is not the case for the cubes. We start with a simple result on the
cartesian product 2with K2.
Proposition 7.1 If χb(G)diam(G)+x, then χb(G2K2)diam(G2K2)+
The result is true for x0, since χb(G2K2)χb(G). So assume x1.
Suppose there exists a broadcast coloring πof G2K2that uses at most
diam(G2K2) + 2x2 colors. It follows that the 2x1 biggest colors—call
them Z—are used at most once in G2K2. It follows that one of the copies
of Gis broadcast-colored by the colors up to diam(G2K2)1 = diam(G)
together with at most x1 colors of Z. Thus χb(G)diam(G) + x1, a
contradiction. 2
For example, this shows that the broadcast chromatic number of the
cube is at least a positive fraction of its order. Next is a result regarding
the first few hypercubes Qd.
Proposition 7.2 χb(Q1) = 2,χb(Q2) = 3,χb(Q3) = 5,χb(Q4) = 7, and
χb(Q5) = 15.
Proof. The values for Q1and Q2follow from earlier results. Consider
Q3. The upper bound is from Proposition 2.1. To see that five colors are
required, note that since β0(Q3) = 4, at most four vertices can be colored
1. Further, at most two vertices can be colored 2; but, if four vertices are
colored 1 then no more than one vertex can be colored 2. Therefore, the
number of vertices colored 1 or 2 must be at most five. Since diam(Q3) = 3,
no color greater than 2 can be used more than once. As there are eight
vertices in Q3, this means that at least five colors are required. The lower
bound for Q4follows from the value for Q3by the above proposition.
For a suitable broadcast coloring in each case, use the greedy algorithm
as follows. Place color 1 on a maximum independent set; then color with
color 2 as many as possible, then color 3 and so on. 2
We look next at the asymptotics. Bounds for the packing numbers of the
hypercubes are well explored in coding theory. For our purposes it suffices
to note the bounds:
i=0 k
and ρ2(Qk)2k1/k by, for example, the (computer) Hamming code.
Proposition 7.3 χb(Qk)(1
Proof. For the upper bound, color a maximum independent set with color 1.
Then color as many vertices with color 2 as possible. Clearly one can choose
at least half a maximum 2-packing. And then use unique colors from there
The lower bound is from the packing bounds above, together with the
fact that β0(Qk) = 2k1.2
For the sake of interest, the following table gives some bounds computed
by using these approaches.
n6 7 8 9 10 11
χb(Qn)15 28 63 132 285 610
χb(Qn)25 49 95 219 441 881
We consider next another relative of the grid. Define a path of thick-
ness was the lexicographic product Pn[Kw], that is, every vertex of the path
is replaced by a clique of wvertices. A broadcast coloring of a thick path
is equivalent to assigning wdistinct colors to each vertex of the path and
requiring a broadcast coloring.
Let g(w) denote the broadcast chromatic number of the infinite path of
thickness w. A natural approach is to assign one color to every vertex, then
a second and so on. So the following parameter arises naturally. Let f(m)
denote the broadcast chromatic number of the infinite path given that no
color smaller than mis used.
Proposition 7.4 For all msufficiently large, f(m)3m1. Indeed, for
all m f (m)3m+ 2.
Proof. This result is easy to verify for small m. For the cases up to m= 33
a computer search produced a suitable coloring.
The bound f(m)3m1 is by induction on m. The base case is
m= 34, where it can be checked by computer that the greedy algorithm
eventually settles into a cycle of length 176400 moves and uses no color more
than 101.
In general, take the broadcast coloring for f(m). Then replace the ver-
tices of color mwith the three colors in succession 3m, 3m+ 1,3m+ 2. The
result is still a broadcast coloring. 2
So if we have the thick path, it follows that g(w)(1 + o(1))3w. The
idea is to fill the first level, then the second and so on.
As for lower bounds, it holds for the path that at most 1/(i+ 1) of the
vertices can receive color i. Hence for H(s, t) = Pt
s1/i it follows that we
need tsuch that H(m, t)1. By standard estimates for the harmonic series,
H(1, t)ln(t)γ, where γis the Euler constant. Thus f(m)em o(m).
As for g(w), similar considerations imply that we need H(1, t)wso
that g(w)Ω(ew). It is unclear if this is the correct order of magnitude.
We note that we quickly run into problems with achieving a proportion of
1/(i+ 1) for any beyond the first wcolors.
8 Open Problems
1. Can the broadcast chromatic number of a tree be computed in poly-
nomial time?
2. What is the maximum broadcast chromatic number of a grid (and
when is it first obtained)?
3. What about other “grids” such as the three-dimensional grid or the
hexagonal lattice?
4. What is the maximum broadcast chromatic number of a cubic graph
on nvertices?
9 Acknowledgment
The authors express their appreciation to Charlie Shi of Clemson University,
and to James Knisley of Bob Jones University. Their insightful comments
and efforts helped to improve our results on broadcast colorings of grids.
[1] J. Dunbar, D. Erwin, T. W. Haynes, S. M. Hedetniemi, and S. T. Hedet-
niemi. Submitted.
[2] J. R. Griggs and R. K. Yeh, The L(2,1)-labeling problem on graphs,
SIAM J. Discrete Math.,9(1996) 309–316.
[3] W. Imrich and S. Klavˇzar, Product Graphs, John Wiley & Sons, New
York, 2000.
[4] R.A. Murphey, P.M. Pardalos and M.G.C. Resende, Frequency assign-
ment problems, In: Handbook of combinatorial optimization, Supplement
Vol. A, Kluwer Acad. Publ., 1999, 295–377.
[5] A. Schwenk, personal communication, 2002.
... The distance between broadcasting stations is directly related to the frequency they may receive, since two stations may only be assigned the same frequency if they are located far enough apart for their frequencies not to interfere with each other. This colouring was first introduced by Goddard et al. [15] where it was called broadcast colouring. Brešar et al. [1] were the first to use the term packing colouring. ...
... Goddard et al. [15] investigated amongst others the packing chromatic number of paths, trees and the infinite square lattice, Z 2 . They found that for a tree of diameter two (that is, a star) the packing chromatic number is 2; for a tree of diameter three the packing chromatic number is 3, and for a tree of diameter 4, they gave an explicit formula based on the number of large neighbors (degree 4 or more) and small neighbors (degree 3 or less) of the central vertex. ...
... Determining the packing chromatic number is considered to be difficult. Finding χ ρ for general graphs is NP-hard [15], and in fact, deciding whether χ ρ (G) ≤ 4 is already NP-complete. In [11], Fiala and Golovach showed that the decision whether a tree allows a packing colouring with at most k classes is NP-complete. ...
Full-text available
An $i$-packing in a graph $G$ is a set of vertices that are pairwise distance more than $i$ apart. A \emph{packing colouring} of $G$ is a partition $X=\{X_{1},X_{2},\ldots,X_{k}\}$ of $V(G)$ such that each colour class $X_{i}$ is an $i$-packing. The minimum order $k$ of a packing colouring is called the packing chromatic number of $G$, denoted by $\chi_{\rho}(G)$. In this paper we investigate the existence of trees $T$ for which there is only one packing colouring using $\chi_\rho(T)$ colours. For the case $\chi_\rho(T)=3$, we completely characterise all such trees. As a by-product we obtain sets of uniquely $3$-$\chi_\rho$-packable trees with monotone $\chi_{\rho}$-coloring and non-monotone $\chi_{\rho}$-coloring respectively.
... The packing chromatic number was introduced by Goddard, S. M. Hedetniemi, S. T. Hedetniemi, Harris and Rall [20] in 2008 under the name broadcast chromatic number. The current name was given in the second paper on the topic by Brešar, Klavžar and Rall [10]. ...
... The current name was given in the second paper on the topic by Brešar, Klavžar and Rall [10]. The concept has a very wide spectrum of potential applications, such as frequency assignments [20] or applications in resource placements and biological diversity [10]. ...
... Proof. Let T be the graph in Fig. 2. Since T has at least one edge, is a connected graph and is not a star, χ ρ (T) ≥ 3 [20]. Clearly, there exists a 3-packing coloring c ′ of T defined as follows: Fig. 2 for the notation of the vertices). ...
Full-text available
Given a graph G, a function c : V(G) ?{1,..., k} with the property that for every u?v, c(u) = c(v) = i implies that the distance between u and v is greater than i, is called a k-packing coloring of G. The smallest integer k for which there exists a k-packing coloring of G is called the packing chromatic number of G, and is denoted by ??(G). Packing chromatic vertex-critical graphs are the graphs G for which ??(G ? x) < ??(G) holds for every vertex x of G. A graph G is called a packing chromatic critical graph if for every proper subgraph H of G, ??(H) < ??(G). Both of the mentioned variations of critical graphs with respect to the packing chromatic number have already been studied [6, 23]. All packing chromatic (vertex-)critical graphs G with ??(G) = 3 were characterized, while there were known only partial results for graphs G with ??(G) = 4. In this paper, we provide characterizations of all packing chromatic vertex-critical graphs G with ??(G) = 4 and all packing chromatic critical graphs G with ??(G) = 4.
... Packing coloring was introduced by Goddard, Hedetniemi, Harris and Rall [10] under the name broadcast coloring, though all subsequent papers starting from [6] used the term packing coloring. This type of coloring became quite popular and was recently surveyed in [4], where one can find 68 citations. ...
... Proposition 7.1. [10] If G is a graph, then χ ρ (G) ≤ n(G) − α(G) + 1, with equality if diam(G) = 2. ...
Full-text available
The competition-independence game, as introduced by Phillips and Slater, is played on a graph by two players, Diminisher and Sweller, who are taking turns in choosing a vertex that is not in the closed neighborhood of any of the previously chosen vertices. The goal of Diminisher is to minimize the (maximal independent) set of chosen vertices at the end of the game, while Sweller wants just the opposite. Assuming that both players are playing optimally according to their goals, two graph invariants arise depending on who starts the game. In this paper, we introduce a variation of the game in which players are allowed at any stage in the game to use an alternative move called prevention. That is, a player can decide that in his/her move he/she will mark (not choose!) a previously unplayed vertex x by which x is prevented to be chosen during the rest of the game; in particular, x is not in the final set of chosen vertices. Given a graph G, and assuming that both players play optimally according to their goals, ?Id(G) (resp.?Is(G)) denotes the size of the set of chosen vertices in the competition-independence game with prevention if Diminisher (resp. Sweller) moves first. By using the Partition Strategy we prove that for any positive integer n, ?Id(Pn) = [2n+3/6] and ?Is(Pn) = [2n+4/6]. While it is not hard to establish the general bounds, 1 ??Id(G) ? [n/2] and 1 ??Is(G) ? [n/2], we characterize the classes of (connected) graphs G that attain each of the four bounds. Finally, a close connection of the new game with a version of the coloring game called the packing coloring game is established for graphs with diameter 2, and several open problems are posed.
... Let b ∈ R be fixed. Differentiating (13) with respect to c gives ...
... Now substituting b into (15), we obtain c = θ d θ s + θ d θ s+1 + θ s θ s+1 . Finally, substituting t = s or t = d into (13), and using b and c above, we obtain the desired result in (9). Now the quotient matrix B of p(A) resulting from the proof of Theorem 3 ...
Full-text available
A $k$-independent set in a connected graph is a set of vertices such that any two vertices in the set are at distance greater than $k$ in the graph. The $k$-independence number of a graph, denoted $\alpha_k$, is the size of a largest $k$-independent set in the graph. Recent results have made use of polynomials that depend on the spectrum of the graph to bound the $k$-independence number. They are optimized for the cases $k=1,2$. There are polynomials that give good (and sometimes) optimal results for general $k$, including case $k=3$. In this paper, we provide the best possible bound that can be obtained by choosing a polynomial for case $k=3$.
... The interference is intricately linked to the station's geographical position, the closer the stations are uncommon the greater the interference that may occur. Also the packing coloring problem arise from the restrictions concerning the assignment of broadcast frequencies to radio stations, which is similar to radio labeling [1,2,3]. The division of channels allocated to the neighbouring stations must be wide enough to prevent interference. ...
... The packing chromatic number p  (G) of G is the smallest integer k for which G has packing kcoloring. The concept of packing coloring emerges from the area of frequency assignment in wireless networks and was introduced by Goddard et al. [1,2,3,4] by the name Broadcast coloring. It has several applications such as resource replacement and biological diversity. ...
In this article we discuss packing chromatic number of unicycle hairy graphs and its generalization.
In a proper edge‐coloring the edges of every color form a matching. A matching is induced if the end‐vertices of its edges induce a matching. A strong edge‐coloring is an edge‐coloring in which the edges of every color form an induced matching. We consider intermediate types of edge‐colorings, where edges of some colors are allowed to form matchings, and the remaining form induced matchings. Our research is motivated by the conjecture proposed in a recent paper of Gastineau and Togni on S $S$‐packing edge‐colorings asserting that, for subcubic graphs, by allowing three additional induced matchings, one is able to save one matching color. We prove that every graph with maximum degree 3 can be decomposed into one matching and at most eight induced matchings, and two matchings and at most five induced matchings. We also show that if a graph is in class I, the number of induced matchings can be decreased by one, hence confirming the above‐mentioned conjecture for class I graphs.
By providing a new framework, we extend previous results on locally checkable problems in bounded treewidth graphs. As a consequence, we show how to solve, in polynomial time for bounded treewidth graphs, [k]-Roman domination and Grundy domination, among other problems for which no such algorithm was previously known. Moreover, by proving that fixed powers of bounded degree and bounded treewidth graphs are also bounded degree and bounded treewidth graphs, we can enlarge the family of problems that can be solved in polynomial time for these graph classes, including distance coloring problems and distance domination problems (for bounded distances).
Full-text available
An $L(2,1)$-labeling of a graph $G$ is a function $f$ from the vertex set $V(G)$ to the set of all nonnegative integers such that $|f(x)-f(y)|\ge 2$ if $d(x,y)=1$ and $|f(x)-f(y)|\ge 1$ if $d(x,y)=2$. The $L(2,1)$-labeling number $\lambda(G)$ of $G$ is the smallest number $k$ such that $G$ has an $L(2,1)$-labeling with $\max\{f(v):v\in V(G)\}=k$. In this paper, we give exact formulas of $\lambda(G\cup H)$ and $\lambda(G+H)$. We also prove that $\lambda(G)\le\Delta^2+\Delta$ for any graph $G$ of maximum degree $\Delta$. For odd-sun-free (OSF)-chordal graphs, the upper bound can be reduced to $\lambda(G)\le 2\Delta+1$. For sun-free (SF)-chordal graphs, the upper bound can be reduced to $\lambda(G)\le \Delta+2\chi(G)-2$. Finally, we present a polynomial time algorithm to determine $\lambda(T)$ for a tree $T$.
Full-text available
Given a simple graph G=(V,E) and a positive number d, an L d (2,1)-labelling of G is a function f:V(G)→[0,∞) such that whenever x,y∈V are adjacent, |f(x)-f(y)|≥2d, and whenever the distance between x and y is two, |f(x)-f(y)|≥d. The L d (2,1)-labelling number λ(G,d) is the smallest number m such that G has an L d (2,1)-labelling f with max{f(v):v∈V}=m. It is shown that to determine λ(G,d), it suffices to study the case when d=1 and the labelling is nonnegative integral-valued. Let λ(G)=λ(G,1). The labelling numbers of special classes of graphs, e.g., λ(C)=4 for any cycle C, are described. It is shown that for graphs of maximum degree Δ, λ(G)≤Δ 2 +2Δ. If G has diameter 2, λ(G)≤Δ 2 , a sharp bound for some Δ. Determining λ(G) is shown to be NP-complete by relating it to the problem of finding Hamilton paths.
Full-text available
The term frequency assignment has been used to describe many types of problems which, quite often, have different modeling needs and objectives. These problems include: 1. Planning models for permanent spectrum allocation, licensing, and regulation which maximize utilization of all radio spectra [94]. 2. Planning models for network design within a given allocation to include; aeronautical mobile, land mobile, maritime mobile, broadcast, land fixed (point-to-point) and satellite. 3. On-line algorithms for dynamically assigning frequencies to users within an established network. Of special interest here are land cellular mobile systems, where an enormous amount of research has been done. A paper by Katzela and Naghshineh [55] contains nearly 100 references to works just in cellular dynamic channel assignment.
We say that a function f:V→{0,1,…,diam(G)}f:V→{0,1,…,diam(G)} is a broadcast if for every vertex v∈Vv∈V, f(v)⩽e(v)f(v)⩽e(v), where diam(G)diam(G) denotes the diameter of G and e(v)e(v) denotes the eccentricity of vv. The cost of a broadcast is the value f(V)=∑v∈Vf(v). In this paper we introduce and study the minimum and maximum costs of several types of broadcasts in graphs, including dominating, independent and efficient broadcasts.
Frequency assignment problems, In: Handbook of combinatorial optimization
  • R A Murphey
  • P M Pardalos
  • M G C Resende
R.A. Murphey, P.M. Pardalos and M.G.C. Resende, Frequency assignment problems, In: Handbook of combinatorial optimization, Supplement Vol. A, Kluwer Acad. Publ., 1999, 295–377.
The L(2, 1)-labeling problem on graphs
  • J R Griggs
  • R K Yeh
J. R. Griggs and R. K. Yeh, The L(2, 1)-labeling problem on graphs, SIAM J. Discrete Math., 9 (1996) 309-316.