JOURNAL OF COMBINATORIALTHEORY 4, 52-71 (1968)

An Introduction to Chromatic Polynomials*

RONALD C. READ

Department of Mathematics, University of the West Indies, Kingston, Jamaica

Communicated by Frank Harary

ABSTRACT

This expository paper is a general introduction to the theory of chromatic pol-

ynomials. Chromatic polynomials are defined, their salient properties are derived, and
some practicaI methods for computing them are given. A brief mention is made of
the connection between the theory of chromatic polynomials and map coloring
problems. The paper concludes with some unsolved problems relating to chromatic
polynomials and to applications of the theory to practical problems in operations
research.

1. INTRODUCTION

There are m a n y interesting problems which arise when one considers

the ways of coloring the nodes o f a graph subject to certain restrictions.
T h e object o f this paper is to give a brief account of the fundamentals
o f this branch o f graph theory.
A coloring o f a graph is the result of giving to each node o f the graph
one of a specified set of colors. In more mathematical terms it is a mapping
o f the nodes into (or onto) a specified finite set C (the set o f colors).
We shall leave for the m o m e n t the question o f whether the m a p p i n g is
to be into or onto.
By a proper coloring of a graph will be meant a coloring which satisfies
the restriction that adjacent nodes are not given (i.e., m a p p e d onto) the
same color (element) o f C. A coloring for which this is not true will be
called an improper coloring.
These are the definitions; but it so happens that we shall nearly always
be concerned with proper colorings only, and it will therefore be con-
venient to drop the term " p r o p e r " and agree that by "colorings" of a
g r a p h we m e a n " p r o p e r colorings" unless the contrary is stated.

* This work was supported in part by a National Science Foundation Grant

(GN-2544) from the Chemical Information Program, which is jointly supported by
the Department of Defense, the National Institutes of Health, and the National Science
Foundation.
52
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 53

Of particular interest is the function, associated with a given graph G,

which expresses the number of different ways of coloring G as a function
of the number of specified colors. The number of colors is, for some
obscure reason, usually denoted by )t, and the function will be written as
Ma()0. Before we can define it we must decide, among other things,
whether the mappings are to be into or onto the color set. It turns out to
be more convenient, algebraically, to work with "into" mappings, and
this is what we shall do. Thus Mc0~) is the number of ways of coloring
the Graph G with 2~ colors, with no stipulation that all the )t colors are
in fact used. If we wish to make this stipulation (as we shall later on)
we can describe the colorings as being in "exactly )t colors."
Two other questions must be settled before the meaning of Ma(A) is
dear. First, are we regarding the nodes as fixed, or can we permute them ?
For example, do we regard the colorings

~ _ and ~ 9 black
o white
| red

of the triangle as being the same or different ? Insofar as they differ only
by a cyclic permutation of the nodes (which leaves the graph unchanged)
it might be felt that they should be regarded as equivalent, but we shall
not take this view. Instead we shall regard our graphs as if their nodes
were points fixed in space, so that, for example, a triangle whose apex
(as we look at it) is colored white will be differently colored from one
whose apex has some other colors.
Second, we shall agree that the actual colors used (or rather the distinc-
tions between them) are important. There are two ways of looking at
this. One can think of the allocation of colors to the nodes merely as a
convenient means of partitioning the nodes into a number of disjoint sets.
Thus the colorings of the triangle given above have the effect of dividing
the set of nodes into three subsets, each having one node. If this were the
only purpose that the coloring had to serve, then permuting the colors
would give an equivalent coloring. There would, for example, be no point
in making a distinction between the colorings
54 READ

since they give the same partition of the nodes. Colorings counted in
this way will be called "colorings with color indifference," but we shall
not usually count them in this way. Instead we shall take account of
different colors, and hence regard the two colorings of the graph just
given as being distinct.

2. SOME ILLUSTRATIONS

By way of illustration let us calculate the function Me(A) for some

simple graphs. Take first the graph

as the graph G. We can color the centre node in any of the A colors.
When this has been done, this color is no longer available for coloring the
outer nodes, by the condition for a proper coloring. Hence the outer nodes
can be colored independently each in A -- 1 ways. Thus

Mo(A) = A(A -- 1)2.

Next let us take the triangle

as G. There are A ways of coloring, say, the top node. There are then A -- 1
ways of coloring an adjacent node, and A - - 2 ways of coloring the
remaining node, since no two nodes may be given the same color. Thus

Mc(~) = A(~ - - 1)(~ - - 2).

This can clearly be generalized. Suppose G is the complete graph on n

nodes. We choose a node and color it; this is possible in A ways. Picking
another node we have A -- I colors with which it can be colored, since it is
adjacent to the first node. Pick another node; it is adjacent to both nodes
already colored, and can therefore be colored in A -- 2 ways. We continue
in this way; the last node can be given any of the remaining h -- (n -- 1)
colors. Hence

Mc(A) = A(A -- I ) 0 -- 2)... (a -- n + 1).

We shall use the notation 2~(") for this factorial expression.

 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 55

Finally let G be the empty graph on n nodes, i.e., the graph having no
edges. Its n isolated nodes can be colored independently, each in A ways.
Hence for this graph
m d ~ ) = A-.

It will be seen that, in each of the above examples, Ma(2t) is a polynomial

in t. This is always so, as we shall shortly prove. The function Mc()t) is
called the "chromatic polynomial" of the graph G.

3. A FUNDAMENTALTHEOREM

Let us consider a particular graph G, such as that in figure 1, and

concentrate on a particular pair of non-adjacent nodes, for example those
marked A and B in the figure. Now the colorings of G in A colors are of
two types:
(i) those in which A and B are given different colors, and
(ii) those in which A and B are given the same color.

AQ G,

FIGURE 1 FIGURE 2

A coloring of G of type (i) will be a coloring of the graph G' obtained

from G by adding the edge A B (see Fig. 2), since the addition of this edge
does not infringe the requirements for a proper coloring. Conversely, to
any coloring of G' corresponds a type (i) coloring of G.
Further, a coloring of G of type (ii) will be a coloring of the graph G"
obtained from G by identifying the nodes A and B (see Fig. 3). (Note that
we can replace multiple edges, if any arise, by single edges, since a multiple
edge represents exactly the same restriction on the colors as does a single
edge joining the same nodes). Conversely, any coloring of G" corresponds
to a type (ii) coloring of G.

FIGURE 3
56 READ

F r o m these two results we derive

THEOREM 1. MoO0 = Ma'()t) + Mo"(A)

--a theorem of fundamental importance.
By way of example, if we take G to be

then G' and G" are

A/XB
d~ and I
respectively. We have already seen that
My(A) = )~()~-- 1)~
and
Mo,(A) = A(A -- 1)(A -- 2),
while it is easy to see that
Ma.(A) = a(~ -- 1).
Thus the theorem is verified in this particular case.
By means of Theorem 1 the chromatic polynomial of a graph can be
expressed in terms of the chromatic polynomials of a graph with an extra
edge, and another with one fewer nodes. The theorem can then be applied
again to these graphs, and so on, the process terminating (as it must do)
when none of these graphs has a pair of non-adjacent nodes. The chromatic
polynomial of the given graph will then have been expressed as the sum
of the chromatic polynomials of complete graphs; and these, as we have
seen, are known.
To do this in practice it is convenient to adopt a convention whereby
the actual picture of a graph serves to denote its chromatic polynomial
(with A understood)3 Thus instead of writing Ma(A) = Ma,(~) q- M~-(A),
and having to explain what G, G', and G" stand for, we can simply write

Applying Theorem 1 repeatedly to this graph, and indicating by A and B

the nodes being considered at each stage, we have

1 This useful notational device was introduced by Zykov [9].

 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 57

"- -I-

ll

Hence
Me(l) = 1 ~5~ -k 4t t4~ + 3t 13~.
It can be seen that the chromatic polynomial of any graph can be reduced
to the sum of a number of factorials, and hence is indeed a polynomial.
When a chromatic polynomial is expressed in this way we shall say that
it is in "factorial form." If the factorial form is known the polynomial
itself is readily found. (Tables of Stirling's numbers which, in effect,
give the t ~n> as polynomials in t, come in handy here.) For the above
graph we have
M~(I) = 1~ -- 614 q- 141a -- 152t~ q- 6h.

Theorem 1 can also be used round the other way, id the form

Mo,(~) = M~(a) -- Mo,,(~).

Here the process is that of removing edges, and we end up with our
chromatic polynomial expressed in terms of the chromatic polynomials
of empty graphs.
58 READ

Thus

15I IV
-I I Y A: I
m
D

~ I o --o --2 +

~ 0
-30 +3

--
OOo) (oo
- -3 - +3 - o
0 0 0 0

o o o o

= --4 o + 6 -3 ~
0 0 0 0

so that
MG(~) : A 4 - - 4 h 3 -~- 6,~ 2 - - 3A.

We shall call this process of expressing chromatic polynomials in terms

of chromatic polynomials of complete or empty graphs "chromatic
reduction."

4. SOME SHORT CUTS

Various short cuts are available whereby the calculation of chromatic

polynomials by this sort of method can be facilitated. Some of these are
contained in the following theorems.
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 59

THEOREM 2. If a graph G has connected components Ga, G2 ..... Gk,

~en
Ma(A) = Ma~(A) 9Mc~(A) "" Ma,(A).

PROOF. Since the components are disjoint, the coloring of each is

quite independent of the coloring of the others. Hence the number of
ways of coloring the whole graph is simply the product of the numbers of
colorings of the separate components.

THEOREM 3. If tWO graphs X and Y "overlap" in a complete graph on k

nodes then the chromatic polynomial of the graph formed by X and Y
together is
Mx(A) "Mr(h)
A(k)

(By "overlapping" is meant the following: The node sets of X and Y

are not disjoint, but have k nodes in common, and every pair of these
k nodes is joined both in X and in Y. As an example take the graph

for which

X = ~ and Y =

having the triangle (k = 3) in common:)

PROOF. The number of ways of coloring the common part is A(k). If

we fix the colors of these k nodes there will be Mx(A)/A(k) ways of coloring
the remaining nodes of X, and Mr(A)/A (k) ways of coloring the remaining
nodes of Y. Hence the total number of colorings is

A(k) Mx(A) Mr(h) _ Mx(A)" Mr(h)

A(k) A(k) A(k)

THEOREM 4. The chromatic polynomial of the product of two graphs

X and Y is Mx(A) (2) Mr(A) where 9 denotes a type of multiplication in
which factorials are treated as powers (further explained below).
60 READ

(By the product of two (disjoint) graphs X and Y is meant the graph
obtained by adding to X and Y in all possible ways, edges joining a node
of X to a node of Y. Thus the product of

I and is

PROOF. Let us apply chromatic reduction to X and Y separately. If

they have m and n nodes, respectively, we shall end up with

Mx(it) = Mar,,~(it) -1- aiMKm_~(it) q- a2Mxm_2(it) -t- "'" (1)

when Mx(it) has been expressed in terms of the chromatic polynomials

of the complete graphs K ~ , K~_I, Kin_2 .... on m, m -- 1, m -- 2 .... nodes.
Similarly

Mr(it) = Mr,~(it) -q- bxMK,~_l(it) q- b2MK,~_2(h) --k "". (2)

If we now perform the chromatic reduction of the product of X and Y,

then the X and Y portions of the product graph will be reduced in exactly
the above way. Moreover, at every stage, every node of each graph
obtained in the reduction of X will be joined to every node of each graph
in the reduction of Y. Hence we shall finish by expressing the chromatic
polynomial of the product in terms of all possible products of a complete
graph from (1) and a complete graph from (2). But the product of a
complete graph on (say) p nodes and one on q nodes is itself a complete
graph, o n p q- q nodes. Hence to a term )t(~) in the factorial form of Mx(it),
and a term it(q) in the factorial form of Mr(it), there will be a term in
it(v+~) in the factorial form of the chromatic polynomial of the product
graph.
It follows that this chromatic polynomial can be found in its factorial
form by taking the factorial forms of Mx(it) and Mr(it) and multiplying
them as if the factorials were powers. This is the process that we denoted
symbolically by Mx(it) 9 Mr(it) in the statement of the theorem.
By way of example, if

X =
I and Y : t

then Mx(h) = A(2)and Mr(A) = h (3) -t- A~).

 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 61

Hence
Mx(A) (3 Mr(A) = 1 (2) (3 (1 (a) q- ),(21)
= 1(5~ + am.

5. I L L U S T R A T I O N S

To illustrate the use of these theorems let us calculate a few chromatic

polynomials, beginning with two that have already been found.
O O
First, ~ =Xistheproductof and , and the
O O
o
chromatic polynomial of is A 3 = 1 <2~ § Therefore, by Theo-
o

rem 4,
= (1(2) 2I- 1(1)) @ (I ($) -~- I(1))

= 14 -- 420 + 612 -- 3t.

Second, by Theorem 3, with k = 2,

A(A - - 1)(A -- 2)(A a -- 41 a q- 61 z -- 3t)

= 1(1 -- 1)
=t 5-614§ 141a - 1 5 1 ~ + 6 t .

Finally we have two general results:

THEOREM 5. The chromatic polynomial of any tree having n nodes is

~(a - 1) --1.
PROOF. We can build up any tree by starting with a single edge and
adding edges one by one, each added edge having one node in common
with the tree so far constructed. For example

~/P and I give

\
62 READ

H e n c e b y T h e o r e m 3 w i t h k ---- 1, the c h r o m a t i c p o l y n o m i a l o f the new

tree ( h a v i n g the a d d e d edge) is o b t a i n e d f r o m t h a t o f the o l d tree b y
m u l t i p l y i n g b y the c h r o m a t i c p o l y n o m i a l o f the edge, viz., A(?, - - 1), a n d
d i v i d i n g b y ?,(1), i.e., A. H e n c e the a d d i t i o n o f each edge m u l t i p l i e s the
c h r o m a t i c p o l y n o m i a l b y ) ~ - 1. Since we s t a r t e d with a single edge
( c h r o m a t i c p o l y n o m i a l )~()~ - - 1)) the t h e o r e m follows.

THEOREM 6. The chromatic polynomial of an n-gon is

(A - - 1)" + ( - - 1 ) " 0 t - - 1).

PROOF. L e t Pn(?,) d e n o t e the c h r o m a t i c p o l y n o m i a l o f a n n-gon. Then

by T h e o r e m 1 we have

% # %

n-gon tree (n--, 1)-gon

b y d e l e t i n g a n edge. H e n c e we h a v e

P~(A) = )t()~ - - 1) "-1 - - P,_I(A)

b y T h e o r e m 5. T h i s can be w r i t t e n

P~(A) - - (A - - 1) ~ = (A - - 1) n-x - - Pn-I(A),

f r o m w h i c h it follows t h a t (--1)n{P,,(A) - - (A - - 1) ~} is a c o n s t a n t , which

can be f o u n d b y p u t t i n g n - - 3. F o r

--{Pa(,~) - - (A - - 1) a} = --{A(A - - 1)(A - - 2) - - (A - - 1) 3}

~A--1.
Thus
P,,(A) - - ( A - - I)" = (--1)~(A - - 1),
i.e.,
P~(A) = (A - - 1) ~ + (--1)"(A - - 1).

6. PROPERTIES OF CHROMATIC POLYNOMIALS

W e shall n o w list a n d p r o v e s o m e p r o p e r t i e s o f the c h r o m a t i c p o l y -

n o m i a l M~()0 o f a g r a p h G. W e let n d e n o t e the n u m b e r o f n o d e s of G.
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 63

THEOREM 7. The degree o f Me(A) is n.

PROOF: By the nature of the reduction process there is, at each stage,
exactly one graph having n nodes. This is therefore true at the final stage,
and from this the result follows. F r o m the fact that there is exactly one
such graph we also have

THEOREM 8. The coefficient o f A~ in Ma(A) is 1.

THEOREM 9. MG(A) has no constant term.

PROOF: If not, then M(0) :/: 0, whereas ctearly the number of ways of
coloring a graph in no colors must be zero!

THEOREM 10. The terms in Me(A) alternate in sign.

PROOV. (It is worth remarking here that many properties of chromatic

polynomials can be found by the use of the very elegant theory of MObius
functions developed recently by Gian-Carlo Rota [5]. From the standpoint
of this theory the theorem we are now considering is almost trivially
true; but in the absence of this background we shall work from first
principles. The p r o o f is by two-way mathematical induction.)
It is easily verified that the theorem is true for all graphs having 1, 2,
3 nodes. Let n be any integer such that the theorem is true for all graphs
on n nodes or less. Consider graphs on n + 1 nodes.
The empty graph on n -5 1 nodes certainly satisfies the theorem. Let
k be any integer such that the theorem is true for all graphs on n + 1 nodes
and k or fewer edges. Consider any graph G' with n § 1 nodes and k + 1
edges. By Theorem 1 we have

Ma,(A) = Me(A) -- MG-(A),

where G has n + 1 nodes and k edges, and G" has n nodes. Since the
theorem is therefore true for G and G" we can write

MG(A) = A n + l __ a l An + a2A n-1 _ aaAn-2 q- ...

and
Ma"(A) = A" - - bl An-1 -[- b2A n-2 -~- ...

with every ai, bi positive. Hence

M~'(A) = An + l -- (a 1 -~ 1 ) A "a' -~- ( a 2 Jr- b0 A"+J ....... (3)

in which the coefficients alternate in sign.
64 READ

By mathematical induction on k the theorem is true for all graphs on

n q- 1 nodes; and by mathematical induction on n it is true for all graphs.
By observing that, in this proof, every deletion of an edge added 1 to
the absolute value of the coefficient of the second term of the chromatic
polynomial we derive

THEOREM 11. The absolute value of the second coefficient of Me(A) is

the number o f edges in G.

THEOREM 12. For a connected graph, M~(A) <~ A(A -- 1)'~-1 (A a positive
integer).
PROOF: Consider any spanning tree T of G. Its nodes can be colored in
A(h -- 1)"-1 ways. N o w every coloring of G is a coloring of T, but in
general some colorings of T will not be colorings of G. Hence

Me(A) ~< A(A -- 1)'*-1.

We have already seen that equality holds if G is a tree. The converse

also holds, and we can state

THEOREM 13. A necessary and sufficient condition for a graph G on n

nodes to be a tree is that Me(A) --= A(A -- 1)"-1.
PROOF: The necessary part has already been proved. Let G be a graph
whose chromatic polynomial is A(A -- 1)n-1. Then G is connected, for
otherwise its chromatic polynomial would be the product of those of its
components (Theorem 2), each of which has a factor A (Theorem 9).
Hence M~(A) would have a factor A2 at least, which it has not.
The coefficient ofA n-1 in Mo(A) is --(n -- 1); hence G has n -- 1 edges.
Since G is connected, has n nodes and n -- 1 edges it must be a tree.
The method of considering a spanning tree of a connected graph also
gives us the following theorem:

THEOREM 14. I f G is connected then the absolute value of the coefficient

of A~ in MG(A) is not less than ('~-~).
PROOF. Let T be a spanning tree of G. F r o m equation (3) it follows
that the addition of an edge to a graph cannot decrease the absolute value
of any coefficient. But G can be obtained by adding suitable edges to T.
Hence, since

Mr(A)----= A ( A - 1)n-1 = ~ (--1) n-r-1 ( n - 111A~

rml
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 65

it follows that the absolute value of the coefficient of A~ in Me(A) is not

n--1
less than (~-1). In particular, the coefficient of A for a connected graph
cannot be zero (its absolute value must be ~1). F r o m this result, and
Theorem 2 we deduced the following corollary:

COROLLARY. The smallest number r such that A~has a nonzero coefficient

in Me(A) is the number of components of G.

7. INTERPRETATION OF THE COEFFICIENTS

It is possible to give an interpretation of the coefficients in the chromatic

polynomial, both in its usual form and in its factorial form. We shall
take the factorial form first.
Let Pc(r) denote the number of ways of coloring a graph G with exactly
r colors, with color indifference (as explained in Section I). This is then
the number of ways of partitioning the set of nodes into r disjoint subsets
such that no two nodes of the same subset are joined. To take account of
of the differences between the colors we must allot a color to each subset,
and this is possible in r! ways. Thus the number of colorings of G in
exactly r colors, but recognizing the different colors is r lPe(r).
Let us now reconstruct Me(A). This is the number of ways of coloring
G in ;~ colors, but not necessarily using all )~ of them. Consider those
colorings in which exactly r colors are used. Their number is (~)r !Pc(r),
since there are (~) ways of choosing which r colors are to be used. Summing
over all r, we have

h
~-- ~] A(r) Po(r).
r=l

The right-hand side is now the chromatic polynomial in its factorial

form, and we therefore have

THEOREM 15. The coefficient of A(r) in the factorial form of Me(A) is

the number of ways of coloring G in exactly r colors with color indifference.
The interpretation of the coefficients in the usual form of the chromatic
polynomial is rather less obvious, and requires the use of the principle of
inclusion and exclusion, well known in combinatorial analysis (see, for
example, Riordan [6]). We shall attempt to find the chromatic polynomial

582141x-5
66 READ

of a graph G b y starting with the total number of colorings, both proper

and improper, and subtracting the improper colorings.
The total number of colorings in A colors, including improper colorings,
is clearly/I'~, where n is the number of nodes. Let us consider such a coloring
of G, and let us delete any edge of G which joins nodes of different colors.
We shall get a subgraph of G having the property that adjacent nodes are
always colored alike (what one might call a "highly improper" coloring !).
There is a single color associated with each connected component of this
subgraph, and thus if the subgraph has p components there are A~ of
these highly improper colorings. We observe that to every coloring (proper
or improper) of G there is a highly improper coloring of some subgraph
of G. For proper colorings of G this subgraph is the empty graph. Let
N(p, r) denote the number of subgraphs of G which have p components
and r edges.
We first subtract from the total ~ the number of highly improper
colorings of those subgraphs having just one edge. If we subtract
Z~ N(p, 1)A~ we shall have subtracted these, but much more besides.
For, if AB and CD are edges of G, then the contribution from the subgraph
consisting of the edge AB alone will include colorings in which C and D
are given the same colors, and this will be a highly improper coloring of
the subgraph consisting of AB and CD. Moreover it will have been
subtracted twice, once for AB and once for CD. Similarly colorings for
subgraphs of 3, 4 and more edges will also have been subtracted an
appropriate number of times.
To redress the balance we can add the term ~ N(p, 2)AL This will
compensate for the double subtraction of the colorings of the two-edge
subgraphs, but will now necessitate a compensation for the three-edge
subgraphs, and so on. We obtain, for the number of proper colorings of G,

As -- ~ N(p, 1)A~ + ~] N(p, 2 ) ~ -- ~ N(p, 3) ~~ + ..-.

Since N(n, 0) is dearly = 1 (the empty graph), we can write this expres-
sion as

Mo(A) = ~ ~ (--1) r N(p,r)A"

r ~0=1
(4)

From this we obtain

THEOREM 16. The coefficient of )~ in Mo(A) is ~ =k o ( - - l )N(p,

r r),
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 67

where N(p, r) is the number of subgraphs of G with p components and r edges,

and k is the number of edges in G.
Many results concerning chromatic polynomials, including our
Theorems 7, 8, 9, 11, follow readily from the above form for the
coefficients.
To calculate a chromatic polynomial by means of Theorem 16 would
entail examining all the 2 k subgraphs of G (spanning subgraphs to be
quite unambiguous--they have all n nodes but only some subset of the
edges of G). It was shown by Whitney [7], however, that one need only
consider a comparatively small number of these subgraphs. Let us list
all the circuits in the graph G, and from each circuit remove one edge.
To do this systematically, we can imagine the edges to have been numbered
in some arbitrary fashion; then from each circuit we remove the edge with
highest number. We obtain a set of what are called "broken circuits."

THEOREM 17. Equation (4) still holds if, in finding N(p,r) we consider
only those subgraphs of G which do not contain any broken circuits.
PROOF: The full proof of this is contained in the paper [7] by Whitney,
and we shall not give it here, but merely indicate the lines on which it
proceeds. Briefly, the method is to show that those subgraphs which
contain a broken circuit can be paired off so that the contributions to
Ma(A) from the subgraphs of a pair will cancel.
If a subgraph H of G contains a broken circuit B, but does not contain
the edge (b say) that was deleted to form B, then there is also a subgraph H*
which is H augmented by the edge b. Now H and H* have the same
number of components, since the two nodes of H* joined by b were
already joined in H by the broken circuit. Therefore since H* has one
more edge than H, the contributions of these two subgraphs to Mc(A)
(in equation 4) will cancel.
In general there may be several possible companions for a given
subgraph (if it contains more than one broken circuit) and the part of the
proof that we have omitted is that which ensures that all the subgraphs
which contain some broken circuit can be paired so that their contributions
vanish. It then follows that only the other subgraphs need be considered.

8. MAP COLORING PROBLEMS

Problems concerning the coloring of the regions of a map are special

cases of graph coloring problems. For we can take a point in the interior
of each region, call these points the nodes of a graph G, and join two of
68 READ

these nodes if and only if the regions to which they belong have a boundary
in common. The condition for a map coloring, viz., that two regions
with a common boundary must be given different colors, becomes the
condition that two adjacent nodes must be colored differently, that is,
the condition for a proper coloring of the resultant graph G. If the map is
drawn on a plane, then G will be a planar graph.
Map coloring problems are therefore equivalent to problems in the
coloring of planar graphs. In particular, the famous four-color conjecture,
that any planar map can be colored in four colors, becomes
"Ma(4) 3& 0, if G is planar"
since Ma(4) ----- 0 means that G cannot be colored in four colors.
For a discussion of various results concerning chromatic polynomials
of maps, and other applications to specific map coloring problems see
Birkhoff [2-4] and Whitney [7, 8].

9. UNSOLVED PROBLEMS

There are many unsolved problems in connection with graph colorings;

we shall mention only a few. First and foremost is the question "What
makes a polynomial chromatic?" We have derived various necessary
conditions for a polynomial to be the chromatic polynomial of some
graph (Theorems 7, 8, 9, 10, 12) but none of them is sufficient. For
example, the polynomial
A4 _ 3A3 + 3A~
satisfies these conditions but is not the chromatic polynomial of any graph.
The problem of characterizing chromatic polynomials is unsolved.
Another unsolved problem in a similar vein is that of determining what
numbers can be roots of some chromatic polynomial.
A property that is very noticeable when one has calculated a few
chromatic polynomials is that the coefficients first increase in absolute
magnitude, and then decrease; two successive coefficients may be equal,
but it seems that one never finds a coefficient flanked by larger coefficients,
and it is natural to conjecture that the coefficients always behave in this
way. It is fairly easy to show that the coefficients are bounded in absolute
magnitude by the corresponding coefficients in the chromatic polynomial
of the complete graph on the same number of nodes (the proof of this
will be left as an exercise for the reader); and certainly these upper bounds
first increase and then decrease. But whether this is true for all chromatic
polynomials is, as far as I know, still an open question.
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 69

Again, this increase and decrease in the coefficients suggest that for
large values of n the coefficients in the chromatic polynomials of " m o s t "
graphs on n nodes might approximate to some well-known unimodal
statistical distribution. Several unsolved problems can be formulated
along these lines.
It is clear that distinct graphs may have the same chromatic polynomial.
For example, all trees with n nodes have the same chromatic polynomial.
Less trivially, the following distinct graphs have the same chromatic
polynomial:

This prompts the question " W h a t is a necessary and sufficient condition

for two graphs to have the same chromatic polynomial ?" This question
also is unsolved.

10. T w o APPLICATIONS

This introduction m a y well have left the reader with the impression
that chromatic polynomials are not of any particular practical importance.
To show that this is not necessarily the case, we give two possible applica-
tions:

APPLICATION 1. Allocation o f channels to television stations.

Assume that there are k possible channels (frequencies) available for
use by the n television stations in a certain country. As is well known,
stations that are near to each other cannot use the same channel without
causing interference. Thus, given any two stations, it may or may not be
the case that they can use the same channel. The problem is to allocate
a channel to each station in such a way that any two stations which need
to have different channels get different channels.
Let us construct a graph G whose nodes represent the stations. We join
two nodes by an edge if and only if the corresponding stations cannot use
the same channel. Then any allocation of channels is, effectively, a coloring
of G in k colors, and if it is proper then the condition about nearby stations
being given different channels is satisfied. Thus the problem reduces to
that of coloring a graph, and the chromatic polynomial will give the number
of ways of allocating the k channels.
70 READ

The chromatic number x(G) of a graph G is the smallest number of colors

with which G can be properly colored; that is, it is the smallest integer h
for which Me(A) =~50. In the above example it is the minimum number of
channels that will suffice to satisfy the allocation conditions.

APPLICATION 2. Construction of timetables.

Let us formulate the problem of constructing a timetable, say the

weekly schedule of lectures for courses at a university.
We have a list of lectures to be given in the various subjects. Certain
of these lectures must not be given at the same time as certain others,
however, since there are students who wish to attend both. Therefore we
have also a list of pairs of lectures that must not clash. We now construct
a graph L in the following way. For each lecture that has to be given
there is a node of L; and two nodes of L are joined by an edge if and
only if the lectures they represent m a y not be given at the same time.
There will be a set of periods during the week, probably seven or eight
per day, when lectures may be given. We shall take these as constituting
our set of colors. If we can construct a proper coloring of L with these
"colors" we shall have a timetable; for lectures corresponding to nodes
allocated a certain color will be given during the period represented by that
color; and lectures that may not clash will not be scheduled for the same
period, since the nodes that represent them, being joined, will not have
been given the same color. I f it should happen that the chromatic number
of L is greater than the total number of available periods, then no time-
table is possible with the given restrictions.
For an actual timetable the graph L would have a l a r g e number of
nodes, and it would be impracticable to compute its chromatic polynomial.
(The same remark applies to Application 1.) However, what is needed is
not the number of ways of coloring L but, first of all, the value of x(L)
(in case the construction of a timetable should turn out to be impossible),
and one (at least) actual coloring. This is a practical possibility; in fact the
problem can be stated as a special kind of linear programming problem.
We define a "maximal independent set" of nodes of a graph G as a set
of nodes no two of which are joined by an edge, and which is maximal
for this property. Let the maximal independent sets of G be M1, Mz ..... Ms.
The union of these sets is clearly the set of nodes of G. In general two of
these sets will have non-empty intersection.
To find a coloring of G in k colors it is sufficient to select k of these
maximal independent sets of nodes whose union is the set of all nodes
of G. For then we can allocate one color to the nodes of each set. Nodes
belonging to more than one set may be given the color associated with any
 AN INTRODUCTION TO CHROMATIC POLYNOMIALS 71

o f the sets to which t h e y belong. F u r t h e r , the coloring will be p r o p e r ,

since nodes o f the same set are n o t joined.
Let x~ = 1 if M~ is chosen in the selection o f m a x i m a l i n d e p e n d e n t sets,
a n d xi = 0 if it is not. L e t a~ be 1 if n o d e i belongs to M s , a n d 0 if it
does not. T h e n for a c o l o r i n g o f G we m u s t have

a~x~ > 0 (5)

since n o d e i m u s t b e l o n g to at least one o f the sets, a n d

}-'. xi = n u m b e r o f sets (6)

i

should be a m i n i m u m , for the coloring in the fewest colors.

The m i n i m i z i n g o f (6) subject to the constraints (5) is a p r o b l e m in
integer p r o g r a m m i n g . It is o f a r a t h e r special k i n d in t h a t all coefficients
are O's o r l ' s . I n s o f a r as m e t h o d s are k n o w n [1] for the s o l u t i o n o f such
p r o b l e m s , a n d are well a d a p t e d to c o m p u t a t i o n b y electronic means,
the h a n d l i n g o f p r o b l e m s such as the c h a n n e l a l l o c a t i o n p r o b l e m a n d the
c o n s t r u c t i o n o f timetables s h o u l d be feasible, even w h e n the graphs in
question are quite large.

REFERENCES

1. E. BALAS, An Additive Algorithm for Solving Linear Programs with Zero-One

Variables, Operations Res. 13 (1965), 517-546.
2. G. D. BIRI,ZHOFF,A Determinantal Formula for the Number of Ways of Colouring
a Map, Ann. o f Math. 14 (1912), 42-46.
3. G. D. BmKHOFF,The Reducibility of Maps, Amer. J. Math. 35 (1913), 115-128.
4. G. D. B~RKHOEE,On the Number of Ways of Colouring a Map, Proc. Edinburgh
Math. Soc. 2 (1930), 83-91.
5. G.-C. ROTA, On the Foundations of Combinational theory. I. Theory of M6bius
Functions, Z. Wahrscheinlichkeitstheorie und Verw. Gebiete 2 (1964), 340-368.
6. J. RtORDAN, Introduction to Combinational Analysis, Wiley, New York, 1958.
7. H. WHITNEY, A Logical Expansion in Mathematics, Bull. Amer. Math. Soc. 38
(1932), 572-579.
8. H. Whitney, The Coloring of Graphs, Ann. o f Math. 33 (1932), 688-718.
9. A. A. ZVKOV, On Some Properties of Linear Complexes, Amer. Math. Soc. TransL
No. 79 (1952); translated from Mat. Sb. 24, No. 66 (1949), 163-188.