Topics in Algebraic Combinatorics: Preliminary Version of 19 April 2012

1
TOPICS IN ALGEBRAIC COMBINATORICS

Richard P. Stanley
Preliminary version of 19 April 2012
2
CONTENTS
Notation 3
Chapter 1 Walks in graphs 5
Chapter 2 Cubes and the Radon transform 15
Chapter 3 Random walks 25
Chapter 4 The Sperner property 29
Chapter 5 Group actions on boolean algebras 45
Chapter 6 Young diagrams and q-binomial coecients 61
Chapter 7 Enumeration under group action 81
Chapter 8 A glimpse of Young tableaux 111
Chapter 9 The Matrix-Tree Theorem 149
Chapter 10 Eulerian digraphs and oriented trees 169
Chapter 11 The cycle space and bond space 181
Chapter 12 Miscellaneous gems of algebraic combinatorics 209
References 227
3
Basic Notation
P positive integers
N nonnegative integers
Z integers
Q rational numbers
R real numbers
C complex numbers
[n] the set 1, 2, . . . , n for n N (so [0] = )
Z
n
the group of integers modulo n
F
q
the nite eld with q elements
#S or [S[ cardinality (number of elements) of the nite set S
S T the disjoint union of S and T, i.e., S T, where S T =
2
S
the set of all subsets of set S
_
S
k
_
the set of k-element subsets of S
__
S
k
__
the set of k-element multisets on S
KS the vector space with basis S over the eld K
B
n
the poset of all subsets of [n], ordered by inclusion
[x
n
]F(x) coecient of x
n
in the polynomial or power series F(x)
ij
the Kronecker delta, which equals 1 if i = j and 0 otherwise
[L[ the sum of the parts (entries) of L, if L is any array of nonnegative integers
4
() length (number of parts) of the partition
p(n) number of partitions of the integer n 0
Chapter 1
Walks in graphs.
Given a nite set S and integer k 0, let
_
S
k
_
denote the set of k-element
subsets of S. A multiset may be regarded, somewhat informally, as a set
with repeated elements, such as 1, 1, 3, 4, 4, 4, 6, 6. We say that a multiset
M is on a set S if every element of M belongs to S. Thus the multiset in
the example above is on the set S = 1, 3, 4, 6 and also on any set containg
S. Let
__
S
k
__
denote the set of k-element multisubsets on S. For instance, if
S = 1, 2, 3 then (using abbreviated notation),
_
S
2
_
= 12, 13, 23,
__
S
2
__
= 11, 22, 33, 12, 13, 23.
A (nite) graph G consists of a vertex set V = v
1
, . . . , v
p
and edge set
E = e
1
, . . . , e
q
, together with a function : E
__
V
2
__
. We think that
if (e) = uv (short for u, v), then e connects u and v or equivalently e is
incident to u and v. If there is at least one edge incident to u and v then
we say that the vertices u and v are adjacent. If (e) = vv, then we call e a
loop at v. If several edges e
1
, . . . , e
j
(j > 1) satisfy (e
1
) = = (e
j
) = uv,
then we say that there is a multiple edge between u and v. A graph without
loops or multiple edges is called simple. In this case we can think of E as
just a subset of
_
V
2
_
[why?].
The adjacency matrix of the graph G is the pp matrix A = A(G), over
the eld of complex numbers, whose (i, j)-entry a
ij
is equal to the number of
edges incident to v
i
and v
j
. Thus A is a real symmetric matrix (and hence
has real eigenvalues) whose trace is the number of loops in G. For instance,
if G is the graph
5
6 CHAPTER 1. WALKS IN GRAPHS.
1
3
5
4
2
then
A(G) =
_
_
2 1 0 2 0
1 0 0 0 1
0 0 0 0 0
2 0 0 0 1
0 1 0 1 1
_
_
.
A walk in Gof length from vertex u to vertex v is a sequence v
1
, e
1
, v
2
, e
2
, . . . ,
v
, e
, v
+1
such that:
each v
i
is a vertex of G
each e
j
is an edge of G
the vertices of e
i
are v
i
and v
i+1
, for 1 i
v
1
= u and v
+1
= v.
1.1 Theorem. For any integer 1, the (i, j)-entry of the matrix A(G)
is equal to the number of walks from v

i
to v
j
in G of length .
Proof. This is an immediate consequence of the denition of matrix multi-
plication. Let A = (a
ij
). The (i, j)-entry of A(G)
is given by
(A(G)
)
ij
=
a
ii
1
a
i
1
i
2
a
i
1
j
,
where the sum ranges over all sequences (i
1
, . . . , i
1
) with 1 i
k
p.
But since a
rs
is the number of edges between v
r
and v
s
, it follows that the
summand a
ii
1
a
i
1
i
2
a
i
1
j
in the above sum is just the number (which may
be 0) of walks of length from v
i
to v
j
of the form
v
i
, e
1
, v
i
1
, e
2
, . . . , v
i
1
, e
, v
j
7
(since there are a
ii
1
choices for e
1
, a
i
1
i
2
choices for e
2
, etc.) Hence summing
over all (i
1
, . . . , i
1
) just gives the total number of walks of length from v
i
to v
j
, as desired.
We wish to use Theorem 1.1 to obtain an explicit formula for the number
(A(G)
)
ij
of walks of length in G from v
i
to v
j
. The formula we give will
depend on the eigenvalues of A(G). The eigenvalues of A(G) are also called
simply the eigenvalues of G. Recall that a real symmetric p p matrix M
has p linearly independent real eigenvectors, which can in fact be chosen to
be orthonormal (i.e., orthogonal and of unit length). Let u
1
, . . . , u
p
be real
orthonormal eigenvectors for M, with corresponding eigenvalues
1
, . . . ,
p
.
All vectors u will be regarded as p1 column vectors. We let
t
denote trans-
pose, so u
t
is a 1p row vector. Thus the dot (or scalar or inner) product of
the vectors u and v is given by u
t
v (ordinary matrix multiplication). In par-
ticular, u
t
i
u
j
=
ij
(the Kronecker delta). Let U = (u
ij
) be the matrix whose
columns are u
1
, . . . , u
p
, denoted U = [u
1
, . . . , u
p
]. Thus U is an orthogonal
matrix, so
U
t
= U
1
=
_
_
u
t
1
u
t
p
_
_
,
the matrix whose rows are u
t
1
, . . . , u
t
p
. Recall from linear algebra that the
matrix U diagonalizes M, i.e.,
U
1
MU = diag(
1
, . . . ,
p
),
where diag(
1
, . . . ,
p
) denotes the diagonal matrix with diagonal entries
1
, . . . ,
p
.
1.2 Corollary. Given the graph G as above, x the two vertices v
i
and v
j
.
Let
1
, . . . ,
p
be the eigenvalues of the adjacency matrix A(G). Then there
exist real numbers c
1
, . . . , c
p
such that for all 1, we have
(A(G)
)
ij
= c
1
1
+ +c
p
n
.
In fact, if U = (u
rs
) is a real orthogonal matrix such that U
1
AU = diag(
1
, . . . ,
p
),
then we have
c
k
= u
ik
u
jk
.
Proof. We have [why?]
U
1
A
U = diag(
1
, . . . ,
p
).
Hence
A
= U diag(
1
, . . . ,
p
)U
1
.
Taking the (i, j)-entry of both sides (and using U
1
= U
t
) gives [why?]
(A
)
ij
=
k
u
ik
k
u
jk
,
as desired.
In order for Corollary 1.2 to be of any use we must be able to compute the
eigenvalues
1
, . . . ,
p
as well as the diagonalizing matrix U (or eigenvectors
u
i
). There is one interesting special situation in which it is not necessary to
compute U. A closed walk in G is a walk that ends where it begins. The
number of closed walks in G of length starting at v
i
is therefore given by
(A(G)
)
ii
, so the total number f
G
() of closed walks of length is given by
f
G
() =
p
i=1
(A(G)
)
ii
= tr(A(G)
),
where tr denotes trace (sum of the main diagonal entries). Now recall that
the trace of a square matrix is the sum of its eigenvalues. If the matrix M
has eigenvalues
1
, . . . ,
p
then [why?] M
has eigenvalues
1
, . . . ,
p
. Hence
we have proved the following.
1.3 Corollary. Suppose A(G) has eigenvalues
1
, . . . ,
p
. Then the number
of closed walks in G of length is given by
f
G
() =
1
+ +
p
.
We now are in a position to use various tricks and techniques from linear
algebra to count walks in graphs. Conversely, it is sometimes possible to
count the walks by combinatorial reasoning and use the resulting formula to
determine the eigenvalues of G. As a rst simple example, we consider the
complete graph K
p
with vertex set V = v
1
, . . . , v
p
, and one edge between
any two distinct vertices. Thus K
p
has p vertices and
_
p
2
_
=
1
2
p(p 1) edges.
9
1.4 Lemma. Let J denote the p p matrix of all 1s. Then the eigenvalues
of J are p (with multiplicity one) and 0 (with multiplicity p 1).
Proof. Since all rows are equal and nonzero, we have rank(J) = 1. Since a
pp matrix of rank pm has at least m eigenvalues equal to 0, we conclude
that J has at least p 1 eigenvalues equal to 0. Since tr(J) = p and the
trace is the sum of the eigenvalues, it follows that the remaining eigenvalue
of J is equal to p.
1.5 Proposition. The eigenvalues of the complete graph K
p
are as follows:
an eigenvalue of 1 with multiplicity p 1, and an eigenvalue of p 1 with
multiplicity one.
Proof. We have A(K
p
) = J I, where I denotes the pp identity matrix. If
the eigenvalues of a matrix M are
1
, . . . ,
p
, then the eigenvalues of M +cI
(where c is a scalar) are
1
+ c, . . . ,
p
+ c [why?]. The proof follows from
Lemma 1.4.
1.6 Corollary. The number of closed walks of length in K
p
from some
vertex v
i
to itself is given by
(A(K
p
)
)
ii
=
1
p
((p 1)
+ (p 1)(1)
). (1.1)
(Note that this is also the number of sequences (i
1
, . . . , i
) of numbers 1, 2, . . . , p
such that i
1
= i, no two consecutive terms are equal, and i
,= i
1
[why?].)
Proof. By Corollary 1.3 and Proposition 1.5, the total number of closed walks
in K
p
of length is equal to (p 1)
+ (p 1)(1)
. By the symmetry of
the graph K
p
, the number of closed walks of length from v
i
to itself does
not depend on i. (All vertices look the same.) Hence we can divide the
total number of closed walks by p (the number of vertices) to get the desired
answer.
A combinatorial proof of Corollary 1.6 is quite tricky (Exercise 1.1). Our
algebraic proof gives a rst hint of the power of algebra to solve enumerative
problems.
What about non-closed walks in K
p
? Its not hard to diagonalize ex-
plicitly the matrix A(K
p
) (or equivalently, to compute its eigenvectors), but
there is an even simpler special argument. We have
(J I)
k=0
(1)
k
_
k
_
J
k
, (1.2)
by the binomial theorem.
1
Now for k > 0 we have J
k
= p
k1
J [why?], while
J
0
= I. (It is not clear a priori what is the correct value of J
0
, but in
order for equation (1.2) to be valid we must take J
0
= I.) Hence
(J I)
k=1
(1)
k
_
k
_
p
k1
J + (1)
I.
Again by the binomial theorem we have
(J I)
=
1
p
((p 1)
(1)
)J + (1)
I. (1.3)
Taking the (i, j)-entry of each side when i ,= j yields
(A(K
p
)
)
ij
=
1
p
((p 1)
(1)
). (1.4)
If we take the (i, i)-entry of (1.3) then we recover equation (1.1). Note the
curious fact that if i ,= j then
(A(K
p
)
)
ii
(A(K
p
)
)
ij
= (1)
.
We could also have deduced (1.4) from Corollary 1.6 using
p
i=1
p
j=1
_
A(K
p
)
_
ij
= p(p 1)
,
the total number of walks of length in K
p
. Details are left to the reader.
We now will show how equation (1.1) itself determines the eigenvalues
of A(K
p
). Thus if (1.1) is proved without rst computing the eigenvalues
of A(K
p
) (which in fact is what we did two paragraphs ago), then we have
another means to compute the eigenvalues. The argument we will give can
be applied to any graph G, not just K
p
. We begin with a simple lemma.
1.7 Lemma. Suppose
1
, . . . ,
r
and
1
, . . . ,
s
are nonzero complex numbers
such that for all positive integers , we have
1
+ +
r
=
1
+ +
s
. (1.5)
Then r = s and the s are just a permutation of the s.
1
We can apply the binomial theorem in this situation because I and J commute. If
A and B are p p matrices that dont necessarily commute, then the best we can say is
(A + B)
2
= A
2
+ AB + BA + B
2
, and similarly for higher powers.
11
Proof. We will use the powerful method of generating functions. Let x be a
complex number whose absolute value (or modulus) is close to 0. Multiply
(1.5) by x
and sum on all 1. The geometric series we obtain will

converge, and we get
1
x
1
1
x
+ +

r
x
1
r
x
=

1
x
1
1
x
+ +

s
x
1
s
x
. (1.6)
This is an identity valid for suciently small (in modulus) complex num-
bers. By clearing denominators we obtain a polynomial identity. But if two
polynomials in x agree for innitely many values, then they are the same
polynomial [why?]. Hence equation (1.6) is actually valid for all complex
numbers x (ignoring values of x which give rise to a zero denominator).
Fix a complex number ,= 0. Multiply (1.6) by 1 x and let x 1/.
The left-hand side becomes the number of
i
s which are equal to , while
the right-hand side becomes the number of
j
s which are equal to [why?].
Hence these numbers agree for all , so the lemma is proved.
1.8 Example. Suppose that G is a graph with 12 vertices, and that the
number of closed walks of length in G is equal to 3 5
+ 4
+ 2(2)
+ 4.
Then it follows from Corollary 1.3 and Lemma 1.7 [why?] that the eigenvalues
of A(G) are given by 5, 5, 5, 4, 2, 2, 1, 1, 1, 1, 0, 0.
References for Chapter 1.
The connection between graph eigenvalues and the enumeration of walks
is considered folklore. The subject of spectral graph theory, which is con-
cerned with the spectrum (multiset of eigenvalues) of various matrices asso-
ciated with graphs, began around 1931 in the area of quantum chemistry.
The rst mathematical paper was published by L. Collatz and U. Sinogowitz
in 1957. A good general reference is the book
2
[19] by Cvetkovic, Doob,
and Sachs. Two textbooks on this subject are by Cvetkovic, Rowlinson, and
Simic [20] and by Brouwer and Haemers [11].
2
All citations to the literature refer to the bibliography beginning on page 227.
Chapter 2
Cubes and the Radon
transform.
Let us now consider a more interesting example of a graph G, one whose
eigenvalues have come up in a variety of applications. Let Z
2
denote the
cyclic group of order 2, with elements 0 and 1, and group operation being
addition modulo 2. Thus 0 + 0 = 0, 0 + 1 = 1 + 0 = 1, 1 + 1 = 0. Let Z
n
2
denote the direct product of Z
2
with itself n times, so the elements of Z
n
2
are
n-tuples (a
1
, . . . , a
n
) of 0s and 1s, under the operation of component-wise
addition. Dene a graph C
n
, called the n-cube, as follows: The vertex set of
C
n
is given by V (C
n
) = Z
n
2
, and two vertices u and v are connected by an
edge if they dier in exactly one component. Equivalently, u +v has exactly
one nonzero component. If we regard Z
n
2
as consisting of real vectors, then
these vectors form the set of vertices of an n-dimensional cube. Moreover,
two vertices of the cube lie on an edge (in the usual geometric sense) if and
only if they form an edge of C
n
. This explains why C
n
is called the n-cube.
We also see that walks in C
n
have a nice geometric interpretation they
are simply walks along the edges of an n-dimensional cube.
We want to determine explicitly the eigenvalues and eigenvectors of C
n
.
We will do this by a somewhat indirect but extremely useful and powerful
technique, the nite Radon transform. Let 1 denote the set of all functions
f : Z
n
2
R, where R denotes the eld of real numbers.
1
Note that 1
is a vector space over R of dimension 2
n
[why?]. If u = (u
1
, . . . , u
n
) and
1
For abelian groups other than Z
n
2
it is necessary to use complex numbers rather than
real numbers. We could use complex numbers here, but there is no need to do so.
15
16 CHAPTER 2. CUBES AND THE RADON TRANSFORM.
v = (v
1
, . . . , v
n
) are elements of Z
n
2
, then dene their dot product by
u v = u
1
v
1
+ +u
n
v
n
, (2.1)
where the computation is performed modulo 2. Thus we regard u v as an
element of Z
2
. The expression (1)
uv
is dened to be the real number +1
or 1, depending on whether u v = 0 or 1, respectively. Since for integers
k the value of (1)
k
depends only on k (mod 2), it follows that we can treat
u and v as integer vectors without aecting the value of (1)
uv
. Thus, for
instance, formulas such as
(1)
u(v+w)
= (1)
uv+uw
= (1)
uv
(1)
uw
are well-dened and valid. From a more algebraic viewpoint, the map Z
1, 1 sending n to (1)
n
is a group homomorphism, where of course the
product on 1, 1 is multiplication.
We now dene two important bases of the vector space 1. There will be
one basis element of each basis for each u Z
n
2
. The rst basis, denoted B
1
,
has elements f
u
dened as follows:
f
u
(v) =
uv
, (2.2)
the Kronecker delta. It is easy to see that B
1
is a basis, since any g 1
satises
g =
uZ
n
2
g(u)f
u
(2.3)
[why?]. Hence B
1
spans 1, so since #B
1
= dim1 = 2
n
, it follows that B
1
is
a basis. The second basis, denoted B
2
, has elements
u
dened as follows:
u
(v) = (1)
uv
.
In order to show that B
2
is a basis, we will use an inner product on 1 (denoted
, )) dened by
f, g) =
uZ
n
2
f(u)g(u).
Note that this inner product is just the usual dot product with respect to
the basis B
1
.
2.1 Lemma. The set B
2
=
u
: u Z
n
2
forms a basis for 1.
17
Proof. Since #B
2
= dim1 (= 2
n
), it suces to show that B
2
is linearly
independent. In fact, we will show that the elements of B
2
are orthogonal.
We have
u
,
v
) =
wZ
n
2
u
(w)
v
(w)
=
wZ
n
2
(1)
(u+v)w
.
It is left as an easy exercise to the reader to show that for any y Z
n
2
, we
have
wZ
n
2
(1)
yw
=
_
2
n
, if y = 0
0, otherwise.
where 0 denotes the identity element of Z
n
2
(the vector (0, 0, . . . , 0)). Thus
u
,
v
) = 0 if and only u + v = 0, i.e., u = v, so the elements of B
2
are
orthogonal (and nonzero). Hence they are linearly independent as desired.
We now come to the key denition of the Radon transform.
Given a subset of Z
n
2
and a function f 1, dene a new function
f 1 by
f(v) =
w
f(v +w).
The function
f is called the (discrete or nite) Radon transform of f (on

the group Z
n
2
, with respect to the subset ).
We have dened a map
: 1 1. It is easy to see that
is a linear
transformation; we want to compute its eigenvalues and eigenvectors.
2.2 Theorem. The eigenvectors of
are the functions

u
, where u Z
n
2
.
The eigenvalue
u
corresponding to
u
(i.e.,
u
=
u
u
) is given by
u
=
w
(1)
uw
.
Proof. Let v Z
n
2
. Then
u
(v) =
u
(v +w)
=
w
(1)
u(v+w)
=
_
w
(1)
uw
_
(1)
uv
=
_
w
(1)
uw
_
u
(v).
Hence
u
=
_
w
(1)
uw
_
u
,
as desired.
Note that because the
u
s form a basis for 1 by Lemma 2.1, it follows
that Theorem 2.2 yields a complete set of eigenvalues and eigenvectors for
. Note also that the eigenvectors

u
of
are independent of ; only the

eigenvalues depend on .
Now we come to the payo. Let =
1
, . . . ,
n
, where
i
is the ith
unit coordinate vector (i.e.,
i
has a 1 in position i and 0s elsewhere). Note
that the jth coordinate of
i
is just
ij
(the Kronecker delta), explaining
our notation
i
. Let [
] denote the matrix of the linear transformation
: 1 1 with respect to the basis B

1
of 1 given by (2.2).
2.3 Lemma. We have [
] = A(C
n
), the adjacency matrix of the n-cube.
Proof. Let v Z
n
2
. We have
f
u
(v) =
w
f
u
(v +w)
=
w
f
u+w
(v),
since u = v +w if and only if u +w = v. There follows [why?]
f
u
=
w
f
u+w
. (2.4)
19
Equation (2.4) says that the (u, v)-entry of the matrix
is given by
(
)
uv
=
_
1, if u +v
0, otherwise.
Now u +v if and only if u and v dier in exactly one coordinate. This
is just the condition for uv to be an edge of C
n
, so the proof follows.
2.4 Corollary. The eigenvectors E
u
(u Z
n
2
) of A(C
n
) (regarded as linear
combinations of the vertices of C
n
, i.e., of the elements of Z
n
2
) are given by
E
u
=
vZ
n
2
(1)
uv
v. (2.5)
The eigenvalue
u
corresponding to the eigenvector E
u
is given by
u
= n 2(u), (2.6)
where (u) is the number of 1s in u. (The integer (u) is called the Ham-
ming weight or simply the weight of u.) Hence A(C
n
) has
_
n
i
_
eigenvalues
equal to n 2i, for each 0 i n.
Proof. For any function g 1 we have by (2.3) that
g =
v
g(v)f
v
.
Applying this equation to g =
u
gives
u
=
u
(v)f
v
=
v
(1)
uv
f
v
. (2.7)
Equation (2.7) expresses the eigenvector
u
of
(or even
for any Z
n
2
)
as a linear combination of the functions f
v
. But
has the same matrix

with respect to the basis of the f
v
s as A(C
n
) has with respect to the vertices
v of C
n
. Hence the expansion of the eigenvectors of
in terms of the f
v
s
has the same coecients as the expansion of the eigenvectors of A(C
n
) in
terms of the vs, so equation (2.5) follows.
According to Theorem 2.2 the eigenvalue
u
corresponding to the eigen-
vector
u
of
(or equivalently, the eigenvector E

u
of A(C
n
)) is given by
u
=
w
(1)
uw
. (2.8)
Now =
1
, . . . ,
n
, and
i
u is 1 if u has a one in its ith coordinate and is
0 otherwise. Hence the sum in (2.8) has n(u) terms equal to +1 and (u)
terms equal to 1, so
u
= (n (u)) (u) = n 2(u), as claimed.
We have all the information needed to count walks in C
n
.
2.5 Corollary. Let u, v Z
n
2
, and suppose that (u + v) = k (i.e., u and v
disagree in exactly k coordinates). Then the number of walks of length in
C
n
between u and v is given by
(A
)
uv
=
1
2
n
n
i=0
k
j=0
(1)
j
_
k
j
__
n k
i j
_
(n 2i)
, (2.9)
where we set
_
nk
ij
_
= 0 if j > i. In particular,
(A
)
uu
=
1
2
n
n
i=0
_
n
i
_
(n 2i)
. (2.10)
Proof. Let E
u
and
u
be as in Corollary 2.4. In order to apply Corollary 1.2,
we need the eigenvectors to be of unit length (where we regard the f
v
s as
an orthonormal basis of 1). By equation (2.5), we have
[E
u
[
2
=
vZ
n
2
((1)
uv
)
2
= 2
n
.
Hence we should replace E
u
by E
u
=
1
2
n/2
E
u
to get an orthonormal basis.
According to Corollary 1.2, we thus have
(A
)
uv
=
1
2
n
wZ
n
2
E
uw
E
vw
w
.
Now E
uw
by denition is the coecient of f
w
in the expansion (2.5), i.e.,
E
uw
= (1)
u+w
(and similarly for E
v
), while
w
= n 2(w). Hence
(A
)
uv
=
1
2
n
wZ
n
2
(1)
(u+v)w
(n 2(w))
. (2.11)
The number of vectors w of Hamming weight i which have j 1s in common
with u + v is
_
k
j
__
nk
ij
_
, since we can choose the j 1s in u + v which agree
21
with w in
_
k
j
_
ways, while the remaining i j 1s of w can be inserted in the
n k remaining positions in
_
nk
ij
_
ways. Since (u + v) w j (mod 2), the
sum (2.11) reduces to (2.9) as desired. Clearly setting u = v in (2.9) yields
(2.10), completing the proof.
It is possible to give a direct proof of equation (2.10) avoiding linear al-
gebra, though we do not do so here. Thus by Corollary 1.3 and Lemma 1.7
(exactly as was done for K
n
) we have another determination of the eigen-
values of C
n
. With a little more work one can also obtain a direct proof of
(2.9). Later in Example 9.9.12, however, we will use the eigenvalues of C
n
to
obtain a combinatorial result for which no nonalgebraic proof is known.
2.6 Example. Setting k = 1 in (2.9) yields
(A
)
uv
=
1
2
n
n
i=0
__
n 1
i
_
_
n 1
i 1
__
(n 2i)
=
1
2
n
n1
i=0
_
n 1
i
_
(n 2i)
+1
n i
.
Note (for those familiar with the representation theory of nite groups).
The functions
u
: Z
n
2
R are just the irreducible (complex) characters
of the group Z
n
2
, and the orthogonality of the
u
s shown in the proof of
Lemma 2.1 is the usual orthogonality relation for the irreducible characters
of a nite group. The results of this chapter extend readily to any nite
abelian group. Exercise 5 does the case Z
n
, the cyclic group of order n.
For nonabelian nite groups the situation is much more complicated because
not all irreducible representations have degree one (i.e., are homomorphisms
G C), and there do not exist formulas as explicit as the ones for abelian
groups.
The Radon transform rst arose in a continuous setting in the paper [69]
of J. K. A. Radon and has been applied to such areas as computerized tomog-
raphy. The nite version was rst dened by E. Bolker [8]. For some further
applications to combinatorics see J. Kung [49]. For the Radon transform on
the n-cube Z
n
2
, see P. Diaconis and R. Graham [22]. For the generalization
to Z
n
k
, see M. DeDeo and E. Velasquez [21].
Chapter 3
Random walks.
Let G be a nite graph. We consider a random walk on the vertices of G
of the following type. Start at a vertex u. (The vertex u could be chosen
randomly according to some probability distribution or could be specied
in advance.) Among all the edges incident to u, choose one uniformly at
random (i.e., if there are k edges incident to u, then each of these edges is
chosen with probability 1/k). Travel to the vertex v at the other end of the
chosen edge and continue as before from v. Readers with some familiarity
with probability theory will recognize this random walk as a special case of
a nite state Markov chain. Many interesting questions may be asked about
such walks; the basic one is to determine the probability of being at a given
vertex after a given number of steps.
Suppose vertex u has degree d
u
,i.e., there are d
u
edges incident to u (count-
ing loops at u once only). Let M = M(G) be the matrix whose rows and
columns are indexed by the vertex set v
1
, . . . , v
p
of G, and whose (u, v)-
entry is given by
M
uv
=

uv
d
u
, (3.1)
where
uv
is the number of edges between u and v (which for simple graphs
will be 0 or 1). Thus M
uv
is just the probability that if one starts at u,
then the next step will be to v. An elementary probability theory argument
(equivalent to Theorem 1.1) shows that if is a positive integer, then (M
)
uv
is equal to probability that one ends up at vertex v in steps given that one
has started at u. Suppose now that the starting vertex is not specied, but
rather we are given probabilities
u
summing to 1 and that we start at vertex
u with probability
u
. Let P be the row vector P = [
v
1
, . . . ,
vp
]. Then again
25
26 CHAPTER 3. RANDOM WALKS.
an elementary argument shows that if PM
= [
v
1
, . . . ,
vp
], then
v
is the
probability of ending up at v in steps (with the given starting distribution).
By reasoning as in Section 1, we see that if we know the eigenvalues and
eigenvectors of M, then we can compute the crucial probabilities (M
)
uv
and
u
.
Since the matrix M is not the same as the adjacency matrix A, what
does all this have to do with adjacency matrices? The answer is that in one
important case M is just a scalar multiple of A. We say that the graph G
is regular of degree d if each d
u
= d, i.e., each vertex is incident to d edges.
In this case its easy to see that M(G) =
1
d
A(G). Hence the eigenvectors
E
u
of M(G) and A(G) are the same, and the eigenvalues are related by
u
(M) =
1
d
u
(A). Thus random walks on a regular graph are closely related
to the adjacency matrix of the graph.
3.1 Example. Consider a random walk on the n-cube C
n
which begins at
the origin (the vector (0, . . . , 0)). What is the probability p
that after
steps one is again at the origin? Before applying any formulas, note that
after an even (respectively, odd) number of steps, one must be at a vertex
with an even (respectively, odd) number of 1s. Hence p
= 0 if is odd.
Now note that C
n
is regular of degree n. Thus by (2.6), we have
u
(M(C
n
)) =
1
n
(n 2(u)).
By (2.10) we conclude that
p
=
1
2
n
n
i=0
_
n
i
_
(n 2i)
.
Note that the above expression for p
does indeed reduce to 0 when is odd.

Random walks on graphs is a vast subject, of which we have barely
scratched the surface.
Chapter 4
The Sperner property.
In this section we consider a surprising application of certain adjacency ma-
trices to some problems in extremal set theory. An important role will also
be played by nite groups. In general, extremal set theory is concerned with
nding (or estimating) the most or least number of sets satisfying given set-
theoretic or combinatorial conditions. For example, a typical easy problem
in extremal set theory is the following: What is the most number of subsets
of an n-element set with the property that any two of them intersect? (Can
you solve this problem?) The problems to be considered here are most con-
veniently formulated in terms of partially ordered sets, or posets for short.
Thus we begin with discussing some basic notions concerning posets.
4.1 Denition. A poset (short for partially ordered set) P is a nite set,
also denoted P, together with a binary relation denoted satisfying the
following axioms:
(P1) (reexivity) x x for all x P
(P2) (antisymmetry) If x y and y x, then x = y.
(P3) (transitivity) If x y and y z, then x z.
One easy way to obtain a poset is the following. Let P be any collection
of sets. If x, y P, then dene x y in P if x y as sets. It is easy to see
that this denition of makes P into a poset. If P consists of all subsets
of an n-element set S, then P is called a (nite) boolean algebra of rank n
and is denoted by B
S
. If S = 1, 2, . . . , n, then we denote B
S
simply by B
n
.
Boolean algebras will play an important role throughout this section.
29
30 CHAPTER 4. THE SPERNER PROPERTY.
There is a simple way to represent small posets pictorially. The Hasse
diagram of a poset P is a planar drawing, with elements of P drawn as dots.
If x < y in P (i.e., x y and x ,= y), then y is drawn above x (i.e., with a
larger vertical coordinate). An edge is drawn between x and y if y covers
x, i.e., x < y and no element z is in between, i.e., no z satises x <
z < y. By the transitivity property (P3), all the relations of a nite poset
are determined by the cover relations, so the Hasse diagram determines P.
(This is not true for innite posets; for instance, the real numbers R with
their usual order is a poset with no cover relations.) The Hasse diagram of
the boolean algebra B
3
looks like
23
3
123
13
12
2
1
We say that two posets P and Q are isomorphic
if there is a bijection (one-to-one and onto function) : P Q such
that x y in P if and only if (x) (y) in Q. Thus one can think that
two posets are isomorphic if they dier only in the names of their elements.
This is exactly analogous to the notion of isomorphism of groups, rings,
etc. It is an instructive exercise (see Exercise 4.1 to draw Hasse diagrams
of the one poset of order (number of elements) one (up to isomorphism),
the two posets of order two, the ve posets of order three, and the sixteen
posets of order four. More ambitious readers can try the 63 posets of order
ve, the 318 of order six, the 2045 of order seven, the 16999 of order eight,
the 183231 of order nine, the 2567284 of order ten, the 46749427 of order
eleven, the 1104891746 of order twelve, the 33823827452 of order thirteen,
the 1338193159771 of order fourteen, the 68275077901156 of order fteen,
and the 4483130665195087 of order sixteen. Beyond this the number is not
currently known.
A chain C in a poset is a totally ordered subset of P, i.e., if x, y C then
either x y or y x in P. A nite chain is said to have length n if it has
31
n + 1 elements. Such a chain thus has the form x
0
< x
1
< < x
n
. We say
that a nite poset is graded of rank n if every maximal chain has length n.
(A chain is maximal if its contained in no larger chain.) For instance, the
boolean algebra B
n
is graded of rank n [why?]. A chain y
0
< y
1
< < y
j
is
said to be saturated if each y
i+1
covers y
i
. Such a chain need not be maximal
since there can be elements of P smaller than y
0
or greater than y
j
. If P is
graded of rank n and x P, then we say that x has rank j, denoted (x) = j,
if some (or equivalently, every) saturated chain of P with top element x has
length j. Thus [why?] if we let P
j
= x P : (x) = j, then P is a
disjoint union P = P
0
P
1
P
n
, and every maximal chain of P has the
form x
0
< x
1
< < x
n
where (x
j
) = j. We call P
i
the ith level of P.
We write p
j
= [P
j
[, the number of elements of P of rank j. For example, if
P = B
n
then (x) = [x[ (the cardinality of x as a set) and
p
j
= #x 1, 2, . . . , n : [x[ = j =
_
n
j
_
.
(Note that we use both [S[ and #S for the cardinality of the nite set S.)
We say that a graded poset P of rank n (always assumed to be nite)
is rank-symmetric if p
i
= p
ni
for 0 i n, and rank-unimodal if p
0

p
1
p
j
p
j+1
p
j+2
p
n
for some 0 j n. If P is both
rank-symmetric and rank-unimodal, then we clearly have
p
0
p
1
p
m
p
m+1
p
n
, if n = 2m
p
0
p
1
p
m
= p
m+1
p
m+2
p
n
, if n = 2m+ 1.
We also say that the sequence p
0
, p
1
, . . . , p
n
itself or the polynomial F(q) =
p
0
+ p
1
q + + p
n
q
n
is symmetric or unimodal, as the case may be. For
instance, B
n
is rank-symmetric and rank-unimodal, since it is well-known
(and easy to prove) that the sequence
_
n
0
_
,
_
n
1
_
, . . . ,
_
n
n
_
(the nth row of Pas-
cals triangle) is symmetric and unimodal. Thus the polynomial (1 + q)
n
is
symmetric and unimodal.
A few more denitions, and then nally some results! An antichain in a
poset P is a subset A of P for which no two elements are comparable, i.e.,
we can never have x, y A and x < y. For instance, in a graded poset P the
levels P
j
are antichains [why?]. We will be concerned with the problem of
nding the largest antichain in a poset. Consider for instance the boolean
algebra B
n
. The problem of nding the largest antichain in B
n
is clearly
equivalent to the following problem in extremal set theory: Find the largest
collection of subsets of an n-element set such that no element of the collection
contains another. A good guess would be to take all the subsets of cardinality
n/2 (where x denotes the greatest integer x), giving a total of
_
n
n/2
_
sets in all. But how can we actually prove there is no larger collection? Such
a proof was rst given by Emanuel Sperner in 1927 and is known as Sperners
theorem. We will give three proofs of Sperners theorem in this section: one
proof uses linear algebra and will be applied to certain other situations; the
second proof is an elegant combinatorial argument due to David Lubell in
1966; while the third proof is another combinatorial argument closely related
to the linear algebra proof. We present the last two proofs for their cultural
value. Our extension of Sperners theorem to certain other situations will
involve the following crucial denition.
4.2 Denition. Let P be a graded poset of rank n. We say that P has the
Sperner property or is a Sperner poset if
max#A: A is an antichain of P = max#P
i
: 0 i n.
In other words, no antichain is larger than the largest level P
i
.
Thus Sperners theorem is equivalent to saying that B
n
has the Sperner
property. Note that if P has the Sperner property there may still be an-
tichains of maximum cardinality other than the biggest P
i
; there just cant
be any bigger antichains.
4.3 Example. A simple example of a graded poset that fails to satisfy the
Sperner property is the following:
We now will discuss a simple combinatorial condition which guarantees
that certain graded posets P are Sperner. We dene an order-matching from
P
i
to P
i+1
to be a one-to-one function : P
i
P
i+1
satisfying x < (x)
for all x P
i
. Clearly if such an order-matching exists then p
i
p
i+1
(since is one-to-one). Easy examples show that the converse is false, i.e.,
if p
i
p
i+1
then there need not exist an order-matching from P
i
to P
i+1
.
We similarly dene an order-matching from P
i
to P
i1
to be a one-to-one
function : P
i
P
i1
satisfying (x) < x for all x P
i
.
33
4.4 Proposition. Let P be a graded poset of rank n. Suppose there exists
an integer 0 j n and order-matchings
P
0
P
1
P
2
P
j
P
j+1
P
j+2
P
n
. (4.1)
Then P is rank-unimodal and Sperner.
Proof. Since order-matchings are one-to-one it is clear that
p
0
p
1
p
j
p
j+1
p
j+2
p
n
.
Hence P is rank-unimodal.
Dene a graph G as follows. The vertices of G are the elements of P.
Two vertices x, y are connected by an edge if one of the order-matchings
in the statement of the proposition satises (x) = y. (Thus G is a
subgraph of the Hasse diagram of P.) Drawing a picture will convince you
that G consists of a disjoint union of paths, including single-vertex paths
not involved in any of the order-matchings. The vertices of each of these
paths form a chain in P. Thus we have partitioned the elements of P into
disjoint chains. Since P is rank-unimodal with biggest level P
j
, all of these
chains must pass through P
j
[why?]. Thus the number of chains is exactly
p
j
. Any antichain A can intersect each of these chains at most once, so the
cardinality [A[ of A cannot exceed the number of chains, i.e., [A[ p
j
. Hence
by denition P is Sperner.
It is now nally time to bring some linear algebra into the picture. For
any (nite) set S, we let RS denote the real vector space consisting of all
formal linear combinations (with real coecients) of elements of S. Thus S
is a basis for RS, and in fact we could have simply dened RS to be the
real vector space with basis S. The next lemma relates the combinatorics we
have just discussed to linear algebra and will allow us to prove that certain
posets are Sperner by the use of linear algebra (combined with some nite
group theory).
4.5 Lemma. Suppose there exists a linear transformation U : RP
i
RP
i+1
(U stands for up) satisfying:
U is one-to-one.
For all x P
i
, U(x) is a linear combination of elements y P
i+1
satisfying x < y. (We then call U an order-raising operator.)
Then there exists an order-matching : P
i
P
i+1
.
Similarly, suppose there exists a linear transformation U : RP
i
RP
i+1
satisfying:
U is onto.
U is an order-raising operator.
Then there exists an order-matching : P
i+1
P
i
.
Proof. Suppose U : RP
i
RP
i+1
is a one-to-one order-raising operator. Let
[U] denote the matrix of U with respect to the bases P
i
of RP
i
and P
i+1
of
RP
i+1
. Thus the rows of [U] are indexed by the elements y
1
, . . . , y
p
i+1
of P
i+1
(in some order), and the columns by the elements x
1
, . . . , x
p
i
of P
i
. Since U
is one-to-one, the rank of [U] is equal to p
i
(the number of columns). Since
the row rank of a matrix equals its column rank, [U] must have p
i
linearly
independent rows. Say we have labelled the elements of P
i+1
so that the rst
p
i
rows of [U] are linearly independent.
Let A = (a
ij
) be the p
i
p
i
matrix whose rows are the rst p
i
rows of
[U]. (Thus A is a square submatrix of [U].) Since the rows of A are linearly
independent, we have
det(A) =
a
1(1)
a
p
i
(p
i
)
,= 0,
where the sum is over all permutations of 1, . . . , p
i
. Thus some term
a
1(1)
a
p
i
(p
i
)
of the above sum in nonzero. Since U is order-raising, this
means that [why?] y
k
> x
(k)
for 1 k p
i
. Hence the map : P
i
P
i+1
dened by (x
k
) = y
1
(k)
is an order-matching, as desired.
The case when U is onto rather than one-to-one is proved by a completely
analogous argument.
Note. Although it does not really help in understanding the theory, it
is interesting to regard a one-to-one order-raising operator as a quantum
order-matching. Rather than choosing a single element y = (x) that is
matched with x P
i
, we choose all possible elements y P
i+1
satisfying
y > x at the same time. If U(x) =
y>x
c
y
y (where c
y
R), then we
are choosing y with weight c
y
. As explained in the proof of Lemma 4.5
below, we break the symmetry and obtain a single matched element (x)
by choosing some nonvanishing term in the expansion of a determinant.
35
We now want to apply Proposition 4.4 and Lemma 4.5 to the boolean
algebra B
n
. For each 0 i < n, we need to dene a linear transformation
U
i
: R(B
n
)
i
R(B
n
)
i+1
, and then prove it has the desired properties. We
simply dene U
i
to be the simplest possible order-raising operator, namely,
for x (B
n
)
i
, let
U
i
(x) =
y(Bn)
i+1
y>x
y. (4.2)
Note that since (B
n
)
i
is a basis for R(B
n
)
i
, equation (4.2) does indeed dene
a unique linear transformation U
i
: R(B
n
)
i
R(B
n
)
i+1
. By denition U
i
is
order-raising; we want to show that U
i
is one-to-one for i < n/2 and onto for
i n/2. There are several ways to show this using only elementary linear
algebra; we will give what is perhaps the simplest proof, though it is quite
tricky. The idea is to introduce dual operators D
i
: R(B
n
)
i
R(B
n
)
i1
to
the U
i
s (D stands for down), dened by
D
i
(y) =
x(Bn)
i1
x<y
x, (4.3)
for all y (B
n
)
i
. Let [U
i
] denote the matrix of U
i
with respect to the bases
(B
n
)
i
and (B
n
)
i+1
, and similarly let [D
i
] denote the matrix of D
i
with respect
to the bases (B
n
)
i
and (B
n
)
i1
. A key observation which we will use later is
that
[D
i+1
] = [U
i
]
t
, (4.4)
i.e., the matrix [D
i+1
] is the transpose of the matrix [U
i
] [why?]. Now let
I
i
: R(B
n
)
i
R(B
n
)
i
denote the identity transformation on R(B
n
)
i
, i.e.,
I
i
(u) = u for all u R(B
n
)
i
. The next lemma states (in linear algebraic
terms) the fundamental combinatorial property of B
n
which we need. For
this lemma set U
n
= 0 and D
0
= 0 (the 0 linear transformation between the
appropriate vector spaces).
4.6 Lemma. Let 0 i n. Then
D
i+1
U
i
U
i1
D
i
= (n 2i)I
i
. (4.5)
(Linear transformations are multiplied right-to-left, so AB(u) = A(B(u)).)
Proof. Let x (B
n
)
i
. We need to show that if we apply the left-hand side
of (4.5) to x, then we obtain (n 2i)x. We have
D
i+1
U
i
(x) = D
i+1
_
_
_
|y|=i+1
xy
y
_
_
_
=
|y|=i+1
xy
|z|=i
zy
z.
If x, z (B
n
)
i
satisfy [x z[ < i 1, then there is no y (B
n
)
i+1
such that
x y and z y. Hence the coecient of z in D
i+1
U
i
(x) when it is expanded
in terms of the basis (B
n
)
i
is 0. If [x z[ = i 1, then there is one such y,
namely, y = x z. Finally if x = z then y can be any element of (B
n
)
i+1
containing x, and there are n i such y in all. It follows that
D
i+1
U
i
(x) = (n i)x +
|z|=i
|xz|=i1
z. (4.6)
By exactly analogous reasoning (which the reader should check), we have for
x (B
n
)
i
that
U
i1
D
i
(x) = ix +
|z|=i
|xz|=i1
z. (4.7)
Subtracting (4.7) from (4.6) yields (D
i+1
U
i
U
i1
D
i
)(x) = (n 2i)x, as
desired.
4.7 Theorem. The operator U
i
dened above is one-to-one if i < n/2 and
is onto if i n/2.
Proof. Recall that [D
i
] = [U
i1
]
t
. From linear algebra we know that a (rect-
angular) matrix times its transpose is positive semidenite (or just semide-
nite for short) and hence has nonnegative (real) eigenvalues. By Lemma 4.6
we have
D
i+1
U
i
= U
i1
D
i
+ (n 2i)I
i
.
Thus the eigenvalues of D
i+1
U
i
are obtained from the eigenvalues of U
i1
D
i
by adding n 2i. Since we are assuming that n 2i > 0, it follows that the
eigenvalues of D
i+1
U
i
are strictly positive. Hence D
i+1
U
i
is invertible (since
37
it has no 0 eigenvalues). But this implies that U
i
is one-to-one [why?], as
desired.
The case i n/2 is done by a dual argument (or in fact can be deduced
directly from the i < n/2 case by using the fact that the poset B
n
is self-
dual, though we will not go into this). Namely, from the fact that
U
i
D
i+1
= D
i+2
U
i+1
+ (2i + 2 n)I
i+1
we get that U
i
D
i+1
is invertible, so now U
i
is onto, completing the proof.
Combining Proposition 4.4, Lemma 4.5, and Theorem 4.7, we obtain the
famous theorem of Sperner.
4.8 Corollary. The boolean algebra B
n
has the Sperner property.
It is natural to ask whether there is a less indirect proof of Corollary
4.8. In fact, several nice proofs are known; we give one due to David Lubell,
mentioned before Denition 4.2. Lubells proof of Sperners theorem.
First we count the total number of maximal chains = x
0
< x
1
< <
x
n
= 1, . . . , n in B
n
. There are n choices for x
1
, then n 1 choices for
x
2
, etc., so there are n! maximal chains in all. Next we count the number of
maximal chains x
0
< x
1
< < x
i
= x < < x
n
which contain a given
element x of rank i. There are i choices for x
1
, then i 1 choices for x
2
, up
to one choice for x
i
. Similarly there are n i choices for x
i+1
, then n i +1
choices for x
i+2
, etc., up to one choice for x
n
. Hence the number of maximal
chains containing x is i!(n i)!.
Now let A be an antichain. If x A, then let C
x
be the set of maximal
chains of B
n
which contain x. Since A is an antichain, the sets C
x
, x A
are pairwise disjoint. Hence
[
_
xA
C
x
[ =
xA
[C
x
[
=
xA
((x))!(n (x))!
Since the total number of maximal chains in the C
x
s cannot exceed the total
number n! of maximal chains in B
n
, we have
xA
((x))!(n (x))! n!
Divide both sides by n! to obtain
xA
1
_
n
(x)
_ 1.
Since
_
n
i
_
is maximized when i = n/2, we have
1
_
n
n/2
_
1
_
n
(x)
_,
for all x A (or all x B
n
). Thus
xA
1
_
n
n/2
_ 1,
or equivalently,
[A[
_
n
n/2
_
.
Since
_
n
n/2
_
is the size of the largest level of B
n
, it follows that B
n
is Sperner.
There is another nice way to show directly that B

n
is Sperner, namely,
by constructing an explicit order-matching : (B
n
)
i
(B
n
)
i+1
when i <
n/2. We will dene by giving an example. Let n = 21, i = 9, and
S = 3, 4, 5, 8, 12, 13, 17, 19, 20. We want to dene (S). Let (a
1
, a
2
, . . . , a
21
)
be a sequence of 1s, where a
i
= 1 if i S, and a
i
= 1 if i , S. For the
set S above we get the sequence (writing for 1)
1 1 1 1 1 1 1 1 1 .
Replace any two consecutive terms 1 with 0 0:
1 1 0 0 0 0 1 0 0 0 0 1 0 0.
Ignore the 0s and replace any two consecutive terms 1 with 0 0:
1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0.
Continue:
0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0.
39
At this stage no further replacement is possible. The nonzero terms consist
of a sequence of s followed by a sequence of 1s. There is at least one
since i < n/2. Let k be the position (coordinate) of the last ; here k = 16.
Dene (S) = S k = S 16. The reader can check that this procedure
gives an order-matching. In particular, why is injective (one-to-one), i.e.,
why can we recover S from (S)?
It can be checked that if we glue together the order-matchings (B
n
)
i

(B
n
)
i+1
for i < n/2, along with an obvious dual construction (B
n
)
i
(B
n
)
i1
for i > n/2 then we obtain more than just a partition of B
n
into saturated
chains passing through the middle level (n even) or middle two levels (n odd),
as in the proof of Proposition 4.4. We in fact have the additional property
that these chains are all symmetric, i.e., they begin at some level i n/2
and end at level n i. Such a decomposition of a rank-symmetric, rank-
unimodal graded poset P into saturated chains is called a symmetric chain
decomposition. A symmetric chain decomposition implies that for any j 1,
the largest size of a union of j antichains is equal to the largest size of a
union of j levels of P (Exercise 4.7). (The Sperner property corresponds to
the case j = 1). It can be a challenging problem to decide whether certain
posets have a symmetric chain decomposition (e.g., Exercise 5), though we
will not discuss this topic further here.
In view of the above elegant proof of Lubell and the explicit description of
an order-matching : (B
n
)
i
(B
n
)
i+1
, the reader may be wondering what
was the point of giving a rather complicated and indirect proof using linear
algebra. Admittedly, if all we could obtain from te linear algebra machinery
we have developed was just another proof of Sperners theorem, then it would
have been hardly worth the eort. But in the next section we will show how
Theorem 4.7, when combined with a little nite group theory, can be used to
obtain many interesting combinatorial results for which simple, direct proofs
are not known.
For further information on combinatorial aspects of partially ordered sets
in general, see P. Fishburn [26], R. Stanley [85, Ch. 3], and W. Trotter [89].
Sperners theorem (Corollary 4.8) was rst proved by E. Sperner [78]. The
elegant proof of Lubell appears in [53]. A general reference on the Sperner
property is the book by K. Engel [25]. For more general results on the com-
binatorics of nite sets, see I. Anderson [2]. The linear algebraic approach to
the Sperner property discussed here is due independently to M. Pouzet [65]
(further developed by Pouzet and Rosenberg [66]) and R. Stanley [79][81].
For further information on explicit order matchings, symmetric chain decom-
positions, etc., see the text of Anderson [2] mentioned above.
Chapter 5
Group actions on boolean
algebras.
Let us begin by reviewing some facts from group theory. Suppose that X is
an n-element set and that G is a group. We say that G acts on the set X if
for every element of G we associate a permutation (also denoted ) of X,
such that for all x X and , G we have
((x)) = ()(x).
Thus [why?] an action of G on X is the same as a homomorphism : G
S
X
, where S
X
denotes the symmetric group of all permutations of X. We
sometimes write x instead of (x).
5.1 Example. (a) Let the real number act on the xy-plane by rotation
counterclockwise around the origin by an angle of radians. It is easy to
check that this denes an action of the group R of real numbers (under
addition) on the xy-plane. The kernel of this action, i.e., the kernel of the
homomorphism : R S
R
2, is the cyclic subgroup of R generated by 2.
(b) Now let R act by translation by a distance to the right (i.e.,
adding (, 0)). This yields a completely dierent action of R on the xy-plane.
This time the action is faithful, i.e., the kernel is the trivial subgroup 0.
(c) Let X = a, b, c, d and G = Z
2
Z
2
= (0, 0), (0, 1), (1, 0), (1, 1).
Let G act as follows:
(0, 1) a = b, (0, 1) b = a, (0, 1) c = c, (0, 1) d = d
(1, 0) a = a, (1, 0) b = b, (1, 0) c = d, (1, 0) d = c.
45
46 CHAPTER 5. GROUP ACTIONS ON BOOLEAN ALGEBRAS.
The reader should check that this does indeed dene an action. In particular,
since (1, 0) and (0, 1) generate G, we dont need to dene the action of (0, 0)
and (1, 1) they are uniquely determined.
(d) Let X and G be as in (c), but now dene the action by
(0, 1) a = b, (0, 1) b = a, (0, 1) c = d, (0, 1) d = c
(1, 0) a = c, (1, 0) b = d, (1, 0) c = a, (1, 0) d = b.
Again one can check that we have an action of Z
2
Z
2
on a, b, c, d. The
two actions of G = Z
2
Z
2
that we have just dened are quite dierent;
for instance, in the rst action we have some elements of X xed by some
nonidentity element of G (such as (0, 1) c = c), while the second action
fails to have this property. See also Example 5.2(c,d) below for another
fundamental way in which the two actions dier.
Recall what is meant by an orbit of the action of a group G on a set X.
Namely, we say that two elements x, y of X are G-equivalent if (x) = y
for some G. The relation of G-equivalence is an equivalence relation,
and the equivalence classes are called orbits. Thus x and y are in the same
orbit if (x) = y for some G. The orbits form a partition of X, i.e,
they are pairwise-disjoint, nonempty subsets of X whose union is X. The
orbit containing x is denoted Gx; this is sensible notation since Gx consists
of all elements (x) where G. Thus Gx = Gy if and only if x and y are
G-equivalent (i.e., in the same G-orbit). The set of all G-orbits is denoted
X/G.
5.2 Example. (a) In Example 5.1(a), the orbits are circles with center (0, 0)
(including the degenerate circle whose only point is (0, 0)).
(b) In Example 5.1(b), the orbits are horizontal lines. Note that although
in (a) and (b) the same group G acts on the same set X, the orbits are
dierent.
(c) In Example 5.1(c), the orbits are a, b and c, d.
(d) In Example 5.1(d), there is only one orbit a, b, c, d. Again we have
a situation in which a group G acts on a set X in two dierent ways, with
dierent orbits.
We wish to consider the situation where X = B
n
, the boolean algebra of
rank n (so [B
n
[ = 2
n
). We begin by dening an automorphism of a poset
P to be an isomorphism : P P. (This denition is exactly analogous
47
to the denition of an automorphism of a group, ring, etc.) The set of all
automorphisms of P forms a group, denoted Aut(P) and called the automor-
phism group of P, under the operation of composition of functions (just as
is the case for groups, rings, etc.)
Now consider the case P = B
n
. Any permutation of 1, . . . , n acts on
B
n
as follows: If x = i
1
, i
2
, . . . , i
k
B
n
, then
(x) = (i
1
), (i
2
), . . . , (i
k
). (5.1)
This action of on B
n
is an automorphism [why?]; in particular, if [x[ = i,
then also [(x)[ = i. Equation (5.1) denes an action of the symmetric group
S
n
of all permutations of 1, . . . , n on B
n
[why?]. (In fact, it is not hard
to show that every automorphism of B
n
is of the form (5.1) for S
n
.) In
particular, any subgroup G of S
n
acts on B
n
via (5.1) (where we restrict
to belong to G). In what follows this action is always meant.
5.3 Example. Let n = 3, and let G be the subgroup of S
3
with elements
e and (1, 2). Here e denotes the identity permutation, and (using disjoint
cycle notation) (1, 2) denotes the permutation which interchanges 1 and 2,
and xes 3. There are six orbits of G (acting on B
3
). Writing e.g. 13 as
short for 1, 3, the six orbits are , 1, 2, 3, 12, 13, 23, and 123.
We now dene the class of posets which will be of interest to us here.
Later we will give some special cases of particular interest.
Let G be a subgroup of S
n
. Dene the quotient poset B
n
/G as follows:
The elements of B
n
/G are the orbits of G. If o and o
are two orbits, then

dene o o
in B
n
/G if there exist x o and y o
such that x y in B
n
.
(Its easy to check that this relation is indeed a partial order.)
5.4 Example. (a) Let n = 3 and G be the group of order two generated
by the cycle (1, 2), as in Example 5.3. Then the Hasse diagram of B
3
/G is
shown below, where each element (orbit) is labeled by one of its elements.
q
q q
q q
q
d
d
dd
d
d
dd
d
d
dd
3 1
13 12
123
(b) Let n = 5 and G be the group of order ve generated by the cycle
(1, 2, 3, 4, 5). Then B
5
/G has Hasse diagram
q
q
q q
q q
q
q
d
d
dd
r
r
r
r
r
rr
d
d
dd
1
12 13
123 124
1234
12345
One simple property of a quotient poset B
n
/G is the following.
5.5 Proposition. The quotient poset B
n
/G dened above is graded of rank
n and rank-symmetric.
Proof. We leave as an exercise the easy proof that B
n
/G is graded of rank
n, and that the rank of an element o of B
n
/G is just the rank in B
n
of any
of the elements x of o. Thus the number of elements p
i
(B
n
/G) of rank i is
equal to the number of orbits o (B
n
)
i
/G. If x B
n
, then let x denote the
set-theoretic complement of x, i.e.,
x = 1, . . . , n x = 1 i n: i , x.
Then x
1
, . . . , x
j
is an orbit of i-element subsets of 1, . . . , n if and only if
x
1
, . . . , x
j
is an orbit of (ni)-element subsets [why?]. Hence [(B
n
)
i
/G[ =
[(B
n
)
ni
/G[, so B
n
/G is rank-symmetric.
Let S
n
. We associate with a linear transformation (still denoted
)
: R(B
n
)
i
R(B
n
)
i
by the rule
_
_
x(Bn)
i
c
x
x
_
_
=
x(Bn)
i
c
x
(x),
49
where each c
x
is a real number. (This denes an action of S
n
, or of any
subgroup G of S
n
, on the vector space R(B
n
)
i
.) The matrix of with
respect to the basis (B
n
)
i
is just a permutation matrix, i.e., a matrix with
one 1 in every row and column, and 0s elsewhere. We will be interested in
elements of R(B
n
)
i
which are xed by every element of a subgroup G of S
n
.
The set of all such elements is denoted R(B
n
)
G
i
, so
R(B
n
)
G
i
= v R(B
n
)
i
: (v) = v for all G.
5.6 Lemma. A basis for R(B
n
)
G
i
consists of the elements
v
o
:=
xo
x,
where o (B
n
)
i
/G, the set of G-orbits for the action of G on (B
n
)
i
.
Proof. First note that if o is an orbit and x o, then by denition of orbit we
have (x) o for all G (or all S
n
). Since permutes the elements
of (B
n
)
i
, it follows that permutes the elements of o. Thus (v
o
) = v
o
,
so v
o
R(B
n
)
G
i
. It is clear that the v
o
s are linearly independent since any
x (B
n
)
i
appears with nonzero coecient in exactly one v
o
.
It remains to show that the v
o
s span R(B
n
)
G
i
, i.e., any v =
x(Bn)
i
c
x
x
R(B
n
)
G
i
can be written as a linear combination of v
o
s. Given x (B
n
)
i
, let
G
x
= G: (x) = x, the stabilizer of x. We leave as an exercise the
standard fact that (x) = (x) (where , G) if and only if and belong
to the same left coset of G
x
, i.e., G
x
= G
x
. It follows that in the multiset
of elements (x), where ranges over all elements of G and x is xed, every
element y in the orbit Gx appears #G
x
times, and no other elements appear.
In other words,
G
(x) = [G
x
[ v
Gx
.
(Do not confuse the orbit Gx with the subgroup G
x
!) Now apply to v and
sum on all G. Since (v) = v (because v R(B
n
)
G
i
), we get
[G[ v =
G
(v)
=
G
_
_
x(Bn)
i
c
x
(x)
_
_
=
x(Bn)
i
c
x
_
G
(x)
_
=
x(Bn)
i
c
x
(#G
x
) v
Gx
.
Dividing by [G[ expresses v as a linear combination of the elements v
Gx
(or
v
o
), as desired.
Now let us consider the eect of applying the order-raising operator U
i
to an element v of R(B
n
)
G
i
.
5.7 Lemma. If v R(B
n
)
G
i
, then U
i
(v) R(B
n
)
G
i+1
.
Proof. Note that since G is an automorphism of B
n
, we have x < y
in B
n
if and only if (x) < (y) in B
n
. It follows [why?] that if x (B
n
)
i
then
U
i
((x)) = (U
i
(x)).
Since U
i
and are linear transformations, it follows by linearity that U
i
(u) =
U
i
(u) for all u R(B
n
)
i
. (In other words, U
i
= U
i
.) Then
(U
i
(v)) = U
i
((v))
= U
i
(v),
so U
i
(v) R(B
n
)
G
i+1
, as desired.
We come to the main result of this section, and indeed our main result
on the Sperner property.
5.8 Theorem. Let G be a subgroup of S
n
. Then the quotient poset B
n
/G
is graded of rank n, rank-symmetric, rank-unimodal, and Sperner.
51
Proof. Let P = B
n
/G. We have already seen in Proposition 5.5 that P
is graded of rank n and rank-symmetric. We want to dene order-raising
operators

U
i
: RP
i
RP
i+1
and order-lowering operators

D
i
: RP
i
RP
i1
.
Let us rst consider just

U
i
. The idea is to identify the basis element v
o
of
RB
G
n
with the basis element o of RP, and to let

U
i
: RP
i
RP
i+1
correspond
to the usual order-raising operator U
i
: R(B
n
)
i
R(B
n
)
i+1
. More precisely,
suppose that the order-raising operator U
i
for B
n
given by (4.2) satises
U
i
(v
o
) =
(Bn)
i+1
/G
c
o,o
v
o
, (5.2)
where o (B
n
)
i
/G. (Note that by Lemma 5.7, U
i
(v
o
) does indeed have the
form given by (5.2).) Then dene the linear operator

U
i
: R((B
n
)
i
/G)
R((B
n
)
i
/G) by
U
i
(o) =
(Bn)
i+1
/G
c
o,o
o
.
Note. We can depict the transport of U
i
to

U
i
by a commutative diagram:
(RB
n
)
G
i
U
i
(RB
n
)
G
i+1
=
R(B
n
/G)
i
U
i
R(B
n
/G)
i+1
The arrows pointing down are the linear transformations induced by v
o
o.
The map obtained by applying the top arrow followed by the rightmost down
arrow is the same as applying the leftmost down arrow followed by the bottom
arrow.
We claim that

U
i
is order-raising. We need to show that if c
o,o
,= 0, then
o
> o in B
n
/G. Since v
o
=
, the only way c

o,o
,= 0 in (5.2) is for
some x
to satisfy x
> x for some x o. But this is just what it means

for o
> o, so

U
i
is order-raising.
Now comes the heart of the argument. We want to show that

U
i
is one-
to-one for i < n/2. Now by Theorem 4.7, U
i
is one-to-one for i < n/2. Thus
the restriction of U
i
to the subspace R(B
n
)
G
i
is one-to-one. (The restriction
of a one-to-one function is always one-to-one.) But U
i
and

U
i
are exactly the
same transformation, except for the names of the basis elements on which
they act. Thus

U
i
is also one-to-one for i < n/2.
An exactly analogous argument can be applied to D
i
instead of U
i
. We
obtain one-to-one order-lowering operators

D
i
: R(B
n
)
G
i
R(B
n
)
G
i1
for i >
n/2. It follows from Proposition 4.4, Lemma 4.5, and (4.4) that B
n
/G is
rank-unimodal and Sperner, completing the proof.
We will consider two interesting applications of Theorem 5.8. For our rst
application, we let n =
_
m
2
_
for some m 1, and let M = 1, . . . , m. Let
X =
_
M
2
_
, the set of all two-element subsets of M. Think of the elements of X
as (possible) edges of a graph with vertex set M. If B
X
is the boolean algebra
of all subsets of X (so B
X
and B
n
are isomorphic), then an element x of B
X
is a collection of edges on the vertex set M, in other words, just a simple
graph on M. Dene a subgroup G of S
X
as follows: Informally, G consists
of all permutations of the edges
_
M
2
_
that are induced from permutations
of the vertices M. More precisely, if S
m
, then dene S
X
by
(i, j) = (i), (j). Thus G is isomorphic to S
m
.
When are two graphs x, y B
X
in the same orbit of the action of G on
B
X
? Since the elements of G just permute vertices, we see that x and y are
in the same orbit if we can obtain x from y by permuting vertices. This is
just what it means for two simple graphs x and y to be isomorphic they
are the same graph except for the names of the vertices (thinking of edges
as pairs of vertices). Thus the elements of B
X
/G are isomorphism classes of
simple graphs on the vertex set M). In particular, #(B
X
/G) is the number
of nonisomorphic m-vertex simple graphs, and #((B
X
/G)
i
) is the number of
nonisomorphic such graphs with i edges. We have x y in B
X
/G if there
is some way of labelling the vertices of x and y so that every edge of x is an
edge of y. Equivalently, some spanning subgraph of y (i.e., a subgraph of y
with all the vertices of y) is isomorphic to x, as illustrated in Figure 5.1 for
the case m = 4. Hence by Theorem 5.8 there follows the following result,
which is by no means obvious and has no known non-algebraic proof.
5.9 Theorem. (a) Fix m 1. Let p
i
be the number of nonisomorphic
simple graphs with m vertices and i edges. Then the sequence p
0
, p
1
, . . . , p
(
m
2
)
is symmetric and unimodal.
(b) Let T be a collection of simple graphs with m vertices such that no
element of T is isomorphic to a spanning subgraph of another element of T.
Then #T is maximized by taking T to consist of all nonisomorphic simple
graphs with
1
2
_
m
2
_
edges.
53
Figure 5.1: The poset B
X
/G of nonisomorphic graphs with four vertices
Our second example of the use of Theorem 5.8 is somewhat more subtle
and will be the topic of the next section.
Digression: edge reconstruction. Much work has been done on re-
construction problems, that is, trying to reconstruct a mathematical struc-
ture such as a graph from some of its substructures. The most famous of
such problems is vertex reconstruction: given a simple graph G on p vertices
v
1
, . . . , v
p
, let G
i
be the subgraph obtained by deleting vertex v
i
(and all in-
cident edges). Given the multiset G
1
, . . . , G
p
of vertex-deleted subgraphs
graphs, can G be uniquely reconstructed? It is important to realize that the
vertices are unlabelled, so given G
i
we dont know for any j which vertex is
v
j
. The famous vertex reconstruction conjecture (still open) states that for
p 3 any graph G can be reconstructed from the multiset G
1
, . . . , G
p
.
Here we will be concerned with edge reconstruction, another famous open
problem. Given a simple graph G with edges e
1
, . . . , e
q
, let H
i
= Ge
i
, the
graph obtained from G by removing the edge e
i
.
Edge Reconstruction Conjecture. A simple graph G can be uniquely
reconstructed from its number of vertices and the multiset H
1
, . . . , H
q
of
edge-deleted subgraphs.
Note. As in the case of vertex-reconstruction, the subgraphs H
i
are
unlabelled. The reason for including the number of vertices is that for a
graph with no edges, we have H
1
, . . . , H
q
= , so we need to specify the
number of vertices to obtain G.
Note. It can be shown that if G can be vertex-reconstructed, then
G can be edge reconstructed. Hence the vertex-reconstruction conjecture
implies the edge-reconstruction conjecture.
The techniques developed above to analyze group actions on boolean alge-
bra can be used to prove a special case of the edge-reconstruction conjecture.
Note that a simple graph with p vertices has at most
_
p
2
_
edges.
5.10 Theorem. Let G be a simple graph with p vertices and q >
1
2
_
p
2
_
edges.
Then G is edge-reconstructible.
Proof. Let P
i
be the set of all simple graphs with i edges on the vertex set [p],
so #P
i
=
_
(
p
2
)
i
_
. Let RP
i
denote the real vector space with basis P
i
. Dene a
linear transformation
i
: RP
i
RP
i1
by
i
() =
1
+ +
i
,
where
1
, . . . ,
i
are the (labelled) graphs obtained from by deleting a
single edge. By Theorem 4.7,
i
is injective for i >
1
2
_
p
2
_
. (Think of
i
as
adding edges to the complement of , i.e., the graph with vertex set [p] and
edge set
_
[p]
2
_
E().)
The symmetric group S
p
acts on P
q
by permuting the vertices, and hence
acts on RP
q
, the real vector space with basis P
q
. A basis for the xed space
(RP
q
)
Sp
consists of the distinct sums

=
Sp
(), where P
q
. We
may identify

with the unlabelled graph isomorphic to , since

=

if and
only if and
are isomorphic. Just as in the proof of Theorem 5.8, when

we restrict
q
to (RP
q
)
Sp
for q >
1
2
_
p
2
_
we obtain an injection
q
: (RP
q
)
Sp
(RP
q1
)
Sp
. In particular, for nonisomorphic unlabelled graphs

,

with p
vertices, we have
1
+ +

q
=
q
(
) ,=
q
(
) =

1
+ +

q
.
Hence the unlabelled graphs

1
, . . . ,

q
determine

, as desired.
55
Polynomials with real zeros There are many other techniques than
the linear algebra used to prove Theorem 5.8 for showing that sequences
are unimodal. Here we will discuss a technique based on simple analysis
(calculus) for showing that sequences are unimodal. In fact, we will consider
some stronger properties than unimodality.
A sequence a
0
, a
1
, . . . , a
n
of real numbers is called logarithmically concave,
or log-concave for short, if a
2
i
a
i1
a
i+1
for 1 i n 1. We say that
a
0
, a
1
, . . . , a
n
is strongly log-concave if b
2
i
b
i1
b
i+1
for 1 i n 1, where
b
i
= a
i
/
_
n
i
_
. Strong log-concavity is equivalent to [why?]
a
2
i

_
1 +
1
i
__
1 +
1
n i
_
a
i1
a
i+1
, 1 i n 1,
from which it follows that strong log-concavity implies log-concavity.
Assume now that each a
i
0. Does log-concavity then imply unimodal-
ity? The answer is no, a counterexample being 1, 0, 0, 1. However, only this
type of counterexample can occur, as we now explain. We say that the se-
quence a
0
, a
1
, . . . , a
n
has no internal zeros if whenever we have i < j < k,
a
i
,= 0, and a
k
,= 0, then a
j
,= 0.
5.11 Proposition. Let = (a
0
, a
1
, . . . , a
n
) be a sequence of nonnegative real
numbers with no internal zeros. If is log-concave, then is unimodal.
Proof. Otherwise there would exist 1 i n1 for which a
i1
> a
i
a
i+1
,
so a
2
i
< a
i1
a
i+1
.
Now we come to a fundamental method for proving log-concavity.
5.12 Theorem (I. Newton). Let
P(x) =
n
i=0
b
i
x
i
=
n
i=0
_
n
i
_
a
i
x
i
be a real polynomial all of whose zeros are real numbers. Then the sequence
b
0
, b
1
, . . . , b
n
is strongly log-concave, or equivalently, the sequence a
0
, a
1
, . . . , a
n
is log-concave. Morevover, if each b
i
> 0 (so the zeros of f(x) are negative)
then the sequence b
0
, b
1
, . . . , b
n
has no internal zeros.
Proof. Let deg P(x) = d. By the Fundamental Theorem of Algebra, P(x)
has exactly d real zeros. Suppose that is a zero of multiplicity m > 1,
so P(x) = (x )
m
L(x) for some polynomial L(x) satisfying L() ,= 0. A
simple computation shows that is a zero of P
(x) (the derivative of P(x))

of multiplicty m1. Moreover, if < are both zeros of P(x), then Rolles
theorem shows that P
(x) has a zero satisfying < < . It follows

[why?] that P
(x) has at least d 1 real zeros. Since deg P
(x) = d 1 we
see that P
(x) has exactly d 1 real zeros and no other zeros.

Let Q(x) =
d
i1
dx
i1
P(x). Thus Q(x) is a polynomial of degree at most
n i + 1 with only real zeros. Let R(x) = x
ni+1
Q(1/x), a polynomial of
degree at most n i + 1. The zeros of R(x) are just reciprocals of those
zeros of Q(x) not equal ot 0, with possible new zeros at 0. At any rate, all
zeros of R(x) are real. Now let S(x) =
d
ni1
dx
ni1
R(x), a polynomial of degree
at most two. By Rolles theorem (with a suitable handling of multiple zeros
as above), every zero of R(x) is real. An explicit computation yields
S(x) =
n!
2
(a
i1
x
2
+ 2a
i
x +a
i+1
).
If a
i1
= 0 then trivially a
2
i
a
i1
a
i+1
. Otherwise S(x) is a quadratic
polynomial. Since it has real zeros, its discriminant is nonnegative. But
= (2a
i
)
2
4a
i1
a
i+1
= 4(a
2
i
a
i1
a
i+1
) 0,
so the sequence a
0
, a
1
, . . . , a
n
is log-concave as claimed.
It remains to show that if each a
i
> 0 then the sequence a
0
, a
1
, . . . , a
n
has no internal zeros. Suppose to the contrary that for some i < j < k we
have a
i
> 0, a
j
= 0, a
k
> 0. By arguing as in the previous paragraph we will
obtain a polynomial of the form c+dx
ki
with only real zeros, where c, d > 0.
But since k i 2 we have that every such polynomial has a nonreal zero
[why?], a contradiction which completes the proof.
In order to give combinatorial applications of Theorem 5.12 we need to
nd polynomials with real zeros whose coecients are of combinatorial in-
terest. One such example appears in Exercise 9.6, based on the fact that the
characteristic polynomial of a symmetric matrix has only real zeros.
The techniques developed in this section had their origins in papers of
L. H. Harper [41] and M. Pouzet and I. G. Rosenberg, [66]. The closest
57
treatment to ours appears in a paper of R. P. Stanley [81]. This latter
paper also contains the proof of Theorem 5.10 (edge reconstruction) given
here. This result was rst proved by L. Lovasz [52] by an inclusion-exclusion
argument. The condition q >
1
2
_
p
2
_
in Theorem 5.10 was improved to q >
p(log
2
p 1) by V. M uller [58] (generalizing the argument of Lovasz) and by
I. Krasikov and Y. Roditty [48] (generalizing the argument of Stanley).
For further information on Newtons Theorem 5.12, see e.g. G. H. Hardy,
J. E. Littlewood, and G. Polya [40, p. 52]. For a general survey on uni-
modality, log-concavity, etc., see Stanley [83], with a sequel by F. Brenti
[10].
Chapter 6
Young diagrams and q-binomial
coecients.
A partition of an integer n 0 is a sequence = (
1
,
2
, . . . ) of integers
i
0 satisfying
1

2
and
i1
i
= n. Thus all but nitely
many
i
are equal to 0. Each
i
> 0 is called a part of . We sometimes
suppress 0s from the notation for , e.g., (5, 2, 2, 1), (5, 2, 2, 1, 0, 0, 0), and
(5, 2, 2, 1, 0, 0, . . . ) all represent the same partition (of 10, with four parts).
If is a partition of n, then we denote this by n or [[ = n.
6.1 Example. There are seven partitions of 5, namely (writing e.g. 221 as
short for (2, 2, 1)): 5, 41, 32, 311, 221, 2111, and 11111.
The subject of partitions of integers has been extensively developed, and
we will only be concerned here with a small part related to our previous
discussion. Given positive integers m and n, let L(m, n) denote the set of all
partitions with at most m parts and with largest part at most n. For instance,
L(2, 3) = , 1, 2, 3, 11, 21, 31, 22, 32, 33. (Note that we are denoting by
the unique partition (0, 0, . . . ) with no parts.) If = (
1
,
2
, . . . ) and
= (
1
,
2
, . . . ) are partitions, then dene if
i

i
for all i.
This makes the set of all partitions into a very interesting poset, denoted
Y and called Youngs lattice (named after the British mathematician Alfred
Young, 18731940). (It is called Youngs lattice rather than Youngs
poset because it turns out to have certain properties which dene a lattice.
However, these properties are irrelevant to us here, so we will not bother to
dene the notion of a lattice.) We will be looking at some properties of Y
in Section 8. The partial ordering on Y , when restricted to L(m, n), makes
61
62CHAPTER 6. YOUNGDIAGRAMS AND Q-BINOMIAL COEFFICIENTS.
1
2
3
4
1
2
21
22
11
1
2
3
31
32
33
22
21
11
Figure 6.1: The lattices L(1, 4), L(2, 2), and L(2, 3)
L(m, n) into a poset which also has some fascinating properties. Figure 6.1
below shows L(1, 4), L(2, 2), and L(2, 3), while Figure 6.2 shows L(3, 3).
There is a nice geometric way of viewing partitions and the poset L(m, n).
The Young diagram (sometimes just called the diagram) of a partition is
a left-justied array of squares, with
i
squares in the ith row. For instance,
the Young diagram of (4, 3, 1, 1) looks like:
If dots are used instead of boxes, then the resulting diagram is called a
Ferrers diagram. The advantage of Young diagrams over Ferrers diagrams is
that we can put numbers in the boxes of a Young diagram, which we will do
in Section 7. Observe that L(m, n) is simply the set of Young diagrams D
tting in an mn rectangle (where the upper-left (northwest) corner of D is
the same as the northwest corner of the rectangle), ordered by inclusion. We
will always assume that when a Young diagram D is contained in a rectangle
R, the northwest corners agree. It is also clear from the Young diagram point
of view that L(m, n) and L(n, m) are isomorphic partially ordered sets, the
63
11
1
2
3
31
32
33
21
22
311
321
221
211
111
322 331
333
222
332
Figure 6.2: The lattice L(3, 3)
isomorphism being given by transposing the diagram (i.e., interchanging rows
and columns). If has Young diagram D, then the partition whose diagram
is D
t
(the transpose of D) is called the conjugate of and is denoted
. For
instance, (4, 3, 1, 1)
= (4, 2, 2, 1), with diagram

6.2 Proposition. The poset L(m, n) is graded of rank mn and rank-symmetric.
The rank of a partition is just [[ (the sum of the parts of or the number
of squares in its Young diagram).
Proof. As in the proof of Proposition 5.5, we leave to the reader everything
except rank-symmetry. To show rank-symmetry, consider the complement

of in an m n rectangle R, i.e., all the squares of R except for . (Note

that

depends on m and n, and not just .) For instance, in L(4, 5), the
complement of (4, 3, 1, 1) looks like
If we rotate the diagram of

by 180
then we obtain the diagram of a

partition

L(m, n) satisfying [[+[
[ = mn. This correspondence between

and

shows that L(m, n) is rank-symmetric.
Our main goal in this section is to show that L(m, n) is rank-unimodal
and Sperner. Let us write p
i
(m, n) as short for p
i
(L(m, n)), the number of
elements of L(m, n) of rank i. Equivalently, p
i
(m, n) is the number of par-
titions of i with largest part at most n and with at most m parts, or, in
other words, the number of distinct Young diagrams with i squares which
65
t inside an mn rectangle (with the same northwest corner, as explained
previously). Though not really necessary for this goal, it is nonetheless in-
teresting to obtain some information on these numbers p
i
(m, n). First let us
consider the total number [L(m, n)[ of elements in L(m, n).
6.3 Proposition. We have [L(m, n)[ =
_
m+n
m
_
.
Proof. We will give an elegant combinatorial proof, based on the fact that
_
m+n
m
_
is equal to the number of sequences a
1
, a
2
, . . . , a
m+n
, where each a
j
is either N or E, and there are m Ns (and hence n Es) in all. We will
associate a Young diagram D contained in an m n rectangle R with such
a sequence as follows. Begin at the lower left-hand corner of R, and trace
out the southeast boundary of D, ending at the upper right-hand corner of
R. This is done by taking a sequence of unit steps (where each square of R
is one unit in length), each step either north or east. Record the sequence of
steps, using N for a step to the north and E for a step to the east.
Example. Let m = 5, n = 6, = (4, 3, 1, 1). Then R and D are given by:
The corresponding sequence of Ns and Es is NENNEENENEE.
It is easy to see (left to the reader) that the above correspondence gives
a bijection between Young diagrams D tting in an mn rectangle R, and
sequences of m Ns and n Es. Hence the number of diagrams is equal to
_
m+n
m
_
, the number of sequences.
We now consider how many elements of L(m, n) have rank i. To this end,
let q be an indeterminate; and given j 1 dene [j] = 1 + q + q
2
+ +
q
j1
. Thus [1] = 1, [2] = 1 + q, [3] = 1 + q + q
2
, etc. Note that [j] is a
polynomial in q whose value at q = 1 is just j (denoted [j]
q=1
= j). Next
dene [j]! = [1][2] [j] for j 1, and set [0]! = 1. Thus [1]! = 1, [2]! = 1+q,
[3]! = (1 + q)(1 + q + q
2
) = 1 + 2q + 2q
2
+ q
3
, etc., and [j]!
q=1
= j!. Finally
dene for k j 0,
_
k
j
_
=
[k]!
[j]![k j]!
.
The expression
_
k
j
_
is called a q-binomial coecient (or Gaussian coecient).
Since [r]!
q=1
= r!, it is clear that
_
k
j
_
q=1
=
_
k
j
_
.
One sometimes says that
_
k
j
_
is a q-analogue of the binomial coecient
_
k
j
_
.
6.4 Example. We have
_
k
j
_
=
_
k
kj
_
[why?]. Moreover,
_
k
0
_
=
_
k
k
_
= 1
_
k
1
_
=
_
k
k 1
_
= [k] = 1 +q +q
2
+ +q
k1
_
4
2
_
=
[4][3][2][1]
[2][1][2][1]
= 1 +q + 2q
2
+q
3
+q
4
_
5
2
_
=
_
5
3
_
= 1 +q + 2q
2
+ 2q
3
+ 2q
4
+q
5
+q
6
.
In the above example,
_
k
j
_
was always a polynomial in q (and with non-
negative integer coecients). It is not obvious that this is always the case,
but it will follow easily from the following lemma.
6.5 Lemma. We have
_
k
j
_
=
_
k 1
j
_
+q
kj
_
k 1
j 1
_
, (6.1)
whenever k 1, with the initial conditions
_
0
0
_
= 1,
_
k
j
_
= 0 if j < 0 or
j > k (the same intial conditions satised by the binomial coecients
_
k
j
_
).
67
Proof. This is a straightforward computation. Specically, we have
_
k 1
j
_
+q
kj
_
k 1
j 1
_
=
[k 1]!
[j]![k 1 j]!
+q
kj
[k 1]!
[j 1]![k j]!
=
[k 1]!
[j 1]![k 1 j]!
_
1
[j]
+
q
kj
[k j]
_
=
[k 1]!
[j 1]![k 1 j]!
[k j] +q
kj
[j]
[j][k j]
=
[k 1]!
[j 1]![k 1 j]!
[k]
[j][k j]
=
_
k
j
_
.
Note that if we put q = 1 in (6.1) we obtain the well-known formula
_
k
j
_
=
_
k 1
j
_
+
_
k 1
j 1
_
,
which is just the recurrence dening Pascals triangle. Thus equation (6.1)
may be regarded as a q-analogue of the Pascal triangle recurrence.
We can regard equation (6.1) as a recurrence relation for the q-binomial
coecients. Given the initial conditions of Lemma 6.5, we can use (6.1) in-
ductively to compute
_
k
j
_
for any k and j. From this it is obvious by induction
that the q-binomial coecient
_
k
j
_
is a polynomial in q with nonnegative inte-
ger coecients. The following theorem gives an even stronger result, namely,
an explicit combinatorial interpretation of the coecients.
6.6 Theorem. Let p
i
(m, n) denote the number of elements of L(m, n) of
rank i. Then
i0
p
i
(m, n)q
i
=
_
m+n
m
_
. (6.2)
( Note. The sum on the left-hand side is really a nite sum, since p
i
(m, n) =
0 if i > mn.)
Proof. Let P(m, n) denote the left-hand side of (6.2). We will show that
P(0, 0) = 1, and P(m, n) = 0 if m < 0 or n < 0 (6.3)
P(m, n) = P(m, n 1) +q
n
P(m1, n). (6.4)
Note that equations (6.3) and (6.4) completely determine P(m, n). On
the other hand, substituting k = m + n and j = m in (6.1) shows that
_
m+n
m
_
also satises (6.4). Moreover, the initial conditions of Lemma 6.5
show that
_
m+n
m
_
also satises (6.3). Hence (6.3) and (6.4) imply that
P(m, n) =
_
m+n
m
_
, so to complete the proof we need only establish (6.3)
and (6.4).
Equation (6.3) is clear, since L(0, n) consists of a single point (the empty
partition ), so
i0
p
i
(0, n)q
i
= 1; while L(m, n) is empty (or undened, if
you prefer) if m < 0 or n < 0,
The crux of the proof is to show (6.4). Taking the coecient of q
i
of both
sides of (6.4), we see [why?] that (6.4) is equivalent to
p
i
(m, n) = p
i
(m, n 1) +p
in
(m1, n). (6.5)
Consider a partition i whose Young diagram D ts in an mn rectangle
R. If D does not contain the upper right-hand corner of R, then D ts in
an m (n 1) rectangle, so there are p
i
(m, n 1) such partitions . If on
the other hand D does contain the upper right-hand corner of R, then D
contains the whole rst row of R. When we remove the rst row of R, we
have left a Young diagram of size i n which ts in an (m1) n rectangle.
Hence there are p
in
(m1, n) such , and the proof follows [why?].
Note that if we set q = 1 in (6.2), then the left-hand side becomes
#L(m, n) and the right-hand side
_
m+n
m
_
, agreeing with Proposition 6.3.
Note: There is another well-known interpretation of
_
k
j
_
, this time not
of its coecients (regarded as a polynomial in q), but rather at its values for
certain q. Namely, suppose q is the power of a prime. Recall that there is
a eld F
q
(unique up to isomorphism) with q elements. Then one can show
that
_
k
j
_
is equal to the number of j-dimensional subspaces of a k-dimensional
vector space over the eld F
q
. We will not discuss the proof here since it is
not relevant for our purposes.
As the reader may have guessed by now, the poset L(m, n) is isomorphic
to a quotient poset B
s
/G for a suitable integer s > 0 and nite group G
acting on B
s
. Actually, it is clear that we must have s = mn since L(m, n)
has rank mn and in general B
s
/G has rank s. What is not so clear is the
right choice of G. To this end, let R = R
mn
denote an m n rectangle of
squares. For instance, R
35
is given by the 15 squares of the diagram
69
We now dene the group G = G
mn
as follows. It is a subgroup of the group
S
R
of all permutations of the squares of R. A permutation in G is allowed
to permute the elements in each row of R in any way, and then to permute the
rows themselves of R in any way. The elements of each row can be permuted
in n! ways, so since there are m rows there are a total of n!
m
permutations
preserving the rows. Then the m rows can be permuted in m! ways, so it
follows that the order of G
mn
is given by m!n!
m
. (The group G
mn
is called
the wreath product of S
n
and S
m
, denoted S
n
S
m
or S
n
wr S
m
. However,
we will not discuss the general theory of wreath products here.)
6.7 Example. Suppose m = 4 and n = 5, with the boxes of X labelled as
follows.
6
1 2 3 4 5
7 8 9 10
12 13 11 14 15
16 17 18 19 20
Then a typical permutation in G(4, 5) looks like
19
4 3
12 13
16 20 17 18
1 5 2
15 14 11
9 6 10 7 8
i.e., (16) = 1, (20) = 2, etc.
We have just dened a group G
mn
of permutations of the set R
mn
of
squares of an mn rectangle. Hence G
mn
acts on the boolean algebra B
R
of
all subsets of the set R. The next lemma describes the orbits of this action.
6.8 Lemma. Every orbit o of the action of G
mn
on B
R
contains exactly one
Young diagram D (i.e., exactly one subset D R such that D is left-justied,
and if
i
is the number of elements of D in row i of R, then
1

2

m
).
Proof. Let S be a subset of R, and suppose that S has
i
elements in row i.
If G
mn
and S has
i
elements in row i, then
1
, . . . ,
m
is just some
permutation of
1
, . . . ,
m
[why?]. There is a unique permutation
1
, . . . ,
m
of
1
, . . . ,
m
satisfying
1

m
, so the only possible Young diagram
D in the orbit S is the one of shape = (
1
, . . . ,
m
). Its easy to see
that the Young diagram D
of shape is indeed in the orbit S. For by

permuting the elements in the rows of R we can left-justify the rows of S, and
then by permuting the rows of R themselves we can arrange the row sizes of
S to be in weakly decreasing order. Thus we obtain the Young diagram D
as claimed.
We are now ready for the main result of this section.
6.9 Theorem. The quotient poset B
Rmn
/G
mn
is isomorphic to L(m, n).
Proof. Each element of B
R
/G
mn
contains a unique Young diagram D
by
Lemma 6.8. Moreover, two dierent orbits cannot contain the same Young
diagram D since orbits are disjoint. Thus the map : B
R
/G
mn
L(m, n)
dened by (D
) = is a bijection (one-to-one and onto). We claim that

in fact is an isomorphism of partially ordered sets. We need to show the
following: Let o and o
be orbits of G
mn
(i.e., elements of B
R
/G
mn
). Let D
and D
be the unique Young diagrams in o and o
, respectively. Then there

exist D o and D
satisfying D D
if and only if
in L(m, n).
The if part of the previous sentence is clear, for if
then D
. So assume there exist D o and D
satisfying D D
. The
lengths of the rows of D, written in decreasing order, are
1
, . . . ,
m
, and
similarly for D
. Since each row of D is contained in a row of D
, it follows
that for each 1 j m, D
has at least j rows of size at least

j
. Thus
the length
j
of the jth largest row of D
is at least as large as
j
. In other
words,
j

j
, as was to be proved.
Combining the previous theorem with Theorem 5.8 yields:
6.10 Corollary. The posets L(m, n) are rank-symmetric, rank-unimodal,
and Sperner.
71
Note that the rank-symmetry and rank-unimodality of L(m, n) can be
rephrased as follows: The q-binomial coecient
_
m+n
m
_
has symmetric and
unimodal coecients. While rank-symmetry is easy to prove (see Proposi-
tion 6.2), the unimodality of the coecients of
_
m+n
m
_
is by no means ap-
parent. It was rst proved by J. Sylvester in 1878 by a proof similar to
the one above, though stated in the language of the invariant theory of bi-
nary forms. For a long time it was an open problem to nd a combinatorial
proof that the coecients of
_
m+n
m
_
are unimodal. Such a proof would give
an explicit injection (one-to-one function) : L(m, n)
i
L(m, n)
i+1
for
i <
1
2
mn. (One diculty in nding such maps is to make use of the hy-
pothesis that i <
1
2
mn.) Finally around 1989 such a proof was found by K.
M. OHara. However, OHaras proof has the defect that the maps are not
order-matchings. Thus her proof does not prove that L(m, n) is Sperner, but
only that its rank-unimodal. It is an outstanding open problem in algebraic
combinatorics to nd an explicit order-matching : L(m, n)
i
L(m, n)
i+1
for i <
1
2
mn.
Note that the Sperner property of L(m, n) (together with the fact that the
largest level is in the middle) can be stated in the following simple terms: The
largest possible collection ( of Young diagrams tting in an mn rectangle
such that no diagram in ( is contained in another diagram in ( is obtained
by taking all the diagrams of size
1
2
mn. Although the statement of this fact
requires almost no mathematics to understand, there is no known proof that
doesnt use algebraic machinery. (The several known algebraic proofs are all
closely related, and the one we have given is the simplest.) Corollary 6.10 is
a good example of the ecacy of algebraic combinatorics.
An application to number theory. There is an interesting application
of Corollary 6.10 to a number-theoretic problem. Fix a positive integer k.
For a nite subset S of R
+
= R : > 0, and for a real number > 0,
dene
f
k
(S, ) = #
_
T
_
S
k
_
:
tT
t =
_
In other words, f
k
(S, ) is the number of k-element subsets of S whose el-
ements sum to . For instance, f
3
(1, 3, 4, 6, 7, 11) = 2, since 1 + 3 + 7 =
1 + 4 + 6 = 11.
Given positive integers k < n, our object is to maximize f
k
(S, ) subject
to the condition that #S = n. We are free to choose both S and , but k
and n are xed. Call this maximum value h
k
(n). Thus
h
k
(n) = max
R
+
SR
+
#S=n
f
k
(S, ).
What sort of behavior can we expect of the maximizing set S? If the elements
of S are spread out, say S = 1, 2, 4, 8, . . . , 2
n1
, then all the subset
sums of S are distinct. Hence for any R
+
we have f
k
(S, ) = 0 or 1.
Similarly, if the elements of S are unrelated (e.g., linearly independent over
the rationals, such as S = 1,
2,
3, ,
2
), then again all subset sums are
distinct and f
k
(S, ) = 0 or 1. These considerations make it plausible that
we should take S = [n] = 1, 2, . . . , n and then choose appropriately. In
other words, we are led to the conjecture that for any S
_
R
+
n
_
and R
+
,
we have
f
k
(S, ) f
k
([n], ), (6.6)
for some R
+
to be determined.
First let us evaluate f
k
([n], ) for any . This will enable us to determine
the value of in (6.6). Let S = i
1
, . . . , i
k
[n] with
1 i
1
< i
2
< < i
k
n, i
1
+ +i
k
= . (6.7)
Let j
r
= i
r
r. Then (since 1 + 2 + +k =
_
k+1
2
_
)
n k j
k
j
k1
j
1
0, j
1
+ +j
k
=
_
k + 1
2
_
. (6.8)
Conversely, given j
1
, . . . , j
k
satisfying (6.8) we can recover i
1
, . . . , i
k
satisfying
(6.7). Hence f
k
([n], ) is equal to the number of sequences j
1
, . . . , j
k
satisfying
(6.8). Now let
(S) = (j
k
, j
k1
, . . . , j
1
).
Note that (S) is a partition of the integer
_
k+1
2
_
with at most k parts
and with largest part at most n k. Thus
f
k
([n], ) = p
(
k+1
2
)
(k, n k), (6.9)
or equivalently,
(
k+1
2
)
f
k
([n], )q
(
k+1
2
)
=
_
n
k
_
.
73
By the rank-unimodality (and rank-symmetry) of L(nk, k) (Corollary 6.10),
the largest coecient of
_
n
k
_
is the middle one, that is, the coecient of
k(n k)/2. It follows that for xed k and n, f
k
([n], ) is maximized for
= k(n k)/2 +
_
k+1
2
_
= k(n + 1)/2. Hence the following result is
plausible.
6.11 Theorem. Let S
_
R
+
n
_
, R
+
, and k P. Then
f
k
(S, ) f
k
([n], k(n + 1)/2).
Proof. Let S = a
1
, . . . , a
n
with 0 < a
1
< < a
n
. Let T and U be distinct
k-element subsets of S with the same element sums, say T = a
i
1
, . . . , a
i
k
and U = a
j
1
, . . . , a
j
k
with i
1
< i
2
< < i
k
and j
1
< j
2
< < j
k
.
Dene T
= i
1
, . . . , i
k
and U
= j
1
, . . . , j
k
, so T
, U

_
[n]
k
_
. The crucial
observation is the following:
Claim. The elements (T
) and (U
) are incomparable in L(k, n k),

i.e., neither (T
) (U
) nor (U
) (T
).
Proof of claim. Suppose not, say (T
) (U)
to be denite. Thus
by denition of L(k, n k) we have i
r
r j
r
r for 1 r k. Hence
i
r
j
r
for 1 r k, so also a
ir
a
jr
(since a
1
< < a
n
). But
a
i
1
+ + a
i
k
= a
j
1
+ + a
j
k
by assumption, so a
ir
= a
jr
for all r. This
contradicts the assumption that T and U are distinct and proves the claim.
It is now easy to complete the proof of Theorem 6.11. Suppose that
S
1
, . . . , S
r
are distinct k-element subsets of S with the same element sums.
By the claim, (S
1
), . . . , (S
r
) is an antichain in L(k, n k). Hence r
cannot exceed the size of the largest antichain in L(k, nk). By Theorem 6.6
and Corollary 6.10, the size of the largest antichain in L(k, nk) is given by
p
k(nk)/2
(k, nk). By equation (6.9) this number is equal to f
k
([n], k(n+
1)/2). In other words,
r f
k
([n], k(n + 1)/2),
which is what we wanted to prove.
Note that an equivalent statement of Theorem 6.11 is that h
k
(n) is equal
to the coecient of q
k(nk)/2
in
_
n
k
_
[why?].
Variation on a theme. Suppose that in Theorem 6.11 we do not want
to specify the cardinality of the subsets of S. In other words, for any R
and any nite subset S R
+
, dene
f(S, ) = #T S :
tT
t = .
41
1
2
1
3
31
32
1
21
4
21
3
31
32
42 321
421
431
432
4321
321
21
(1) M
(2) M
(3) M
M(4)
2
43
2
Figure 6.3: The posets M(1), M(2), M(3) and M(4)
How large can f(S, ) be if we require #S = n? Call this maximum value
h(n). Thus
h(n) = max
R
+
SR
+
#S=n
f(S, ). (6.10)
For instance, if S = 1, 2, 3 then f(S, 3) = 2 (coming from the subsets 1, 2
and 3). This is easily seen to be best possible, i.e., h(3) = 2.
We will nd h(n) in a manner analogous to the proof of Theorem 6.11.
The big dierence is that the relevant poset M(n) is not of the form B
n
/G,
so we will have to prove the injectivity of the order-raising operator U
i
from
scratch. Our proofs will be somewhat sketchy; it shouldnt be dicult for
the reader who has come this far to ll in the details.
Let M(n) be the set of all subsets of [n], with the ordering A B
if the elements of A are a
1
> a
2
> > a
j
and the elements of B are
b
1
> b
2
> > b
k
, where j k and a
i
b
i
for 1 i j. (The empty set
is the bottom element of M(n).) Figure 6.3 shows M(1), M(2), M(3), and
M(4).
75
It is easy to see that M(n) is graded of rank
_
n+1
2
_
. The rank of the subset
T = a
1
, . . . , a
k
is
rank(T) = a
1
+ +a
k
. (6.11)
It follows [why?] that the rank-generating function of M(n) is given by
F(M(n), q) =
(
n+1
2
)
i=0
(#M(n)
i
)q
i
= (1 +q)(1 +q
2
) (1 +q
n
).
Dene linear transformations
U
i
: RM(n)
i
RM(n)
i+1
, D
i
: RM(n)
i
RM(n)
i1
by
U
i
(x) =
yM(n)
i+1
x<y
y, x M(n)
i
D
i
(x) =
vM(n)
i1
v<x
c(v, x)v, x M(n)
i
,
where the coecient c(v, x) is dened as follows. Let the elements of v be
a
1
> > a
j
> 0 and the elements of x be b
1
> > b
k
> 0. Since x covers
v, there is a unique r for which a
r
= b
r
1 (and a
k
= b
k
for all other k). In
the case b
r
= 1 we set a
r
= 0. (E.g., if x is given by 5 > 4 > 1 and v by
5 > 4, then r = 3 and a
3
= 0.) Set
c(v, x) =
_ _
n+1
2
_
, if a
r
= 0
(n a
r
)(n +a
r
+ 1), if a
r
> 0.
It is a straightforward computation (proof omitted) to obtain the com-
mutation relation
D
i+1
U
i
U
i1
D
i
=
__
n + 1
2
_
2i
_
I
i
, (6.12)
where I
i
denotes the identity linear transformation on RM(n)
i
. Clearly by
denition U
i
is order-raising. We want to show that U
i
is injective (one-to-
one) for i <
1
2
_
n+1
2
_
. We cant argue as in the proof of Lemma 4.6 that U
i1
D
i
is semidenite since the matrices of U
i1
and D
i
are no longer transposes of
one another. Instead we use the following result from linear algebra. For two
proofs, see pp. 331-333 of Selected Papers on Algebra (S. Montgomery, et al.,
eds.), Mathematical Association of America, 1977.
6.12 Lemma. Let V and W be nite-dimensional vector spaces over a eld.
Let A : V W and B : W V be linear transformations. Then
x
dimV
det(AB xI) = x
dimW
det(BA xI).
In other words, AB and BA have the same nonzero eigenvalues.
We can now prove the key linear algebraic result.
6.13 Lemma. The linear transformation U
i
is injective for i <
1
2
_
n+1
2
_
and
surjective (onto) for i
1
2
_
n+1
2
_
.
Proof. We prove by induction on i that D
i+1
U
i
has positive real eigenvalues
for i <
1
2
_
n+1
2
_
. For i = 0 this is easy to check since dimRM(n)
0
= 1. Assume
for some i <
1
2
_
n+1
2
_
1, i.e., assume that D
i
U
i1
has positive eigenvalues.
By Lemma 6.12, U
i1
D
i
has nonnegative eigenvalues. By (6.12), we have
D
i+1
U
i
= U
i1
D
i
+
__
n + 1
2
_
2i
_
I
i
.
Thus the eigenvalues of D
i+1
U
i
are
_
n+1
2
_
2i more than those of U
i1
D
i
.
Since
_
n+1
2
_
2i > 0, it follows that D
i+1
U
i
has positive eigenvalues. Hence it
is invertible, so U
i
is injective. Similarly (or by symmetry) U
i
is surjective
for i
1
2
_
n+1
2
_
.
The main result on the posets M(n) now follows by a familiar argument.
6.14 Theorem. The poset M(n) is graded of rank
_
n+1
2
_
, rank-symmetric,
rank-unimodal, and Sperner.
Proof. We have already seen that M(n) is graded of rank
_
n+1
2
_
and rank-
symmetric. By the previous lemma, U
i
is injective for i <
1
2
_
n+1
2
_
and
surjective for i
1
2
_
n+1
2
_
. The proof follows from Proposition 4.4 and
Lemma 4.5.
77
Note. As a consequence of Theorem 6.14, the polynomial F(M(n), q) =
(1 +q)(1 +q
2
) (1 +q
n
) has unimodal coecients. No combinatorial proof
of this fact is known, unlike the situation for L(m, n) (where we mentioned
the proof of OHara above).
We can now determine h(n) (as dened by equation (6.10)) by an argu-
ment analogous to the proof of Theorem 6.11.
6.15 Theorem. Let S
_
R
+
n
_
and R
+
. Then
f(S, ) f
_
[n],
_
1
2
_
n + 1
2
___
= h(n).
Proof. Let S = a
1
, . . . , a
n
with 0 < a
1
< < a
n
. Let T and U be
distinct subsets of S with the same element sums, say T = a
r
1
, . . . , a
r
j
and
U = a
s
1
, . . . , a
s
k
with r
1
< r
2
< < r
j
and s
1
< s
2
< < s
k
. Dene
T
= r
1
, . . . , r
j
and U
= s
1
, . . . , s
k
, so T
, U
M(n). The following

fact is proved exactly in the same way as the analogous fact for L(m, n) (the
claim in the proof of Theorem 6.11) and will be omitted here.
Fact. The elements T
and U
are incomparable in M(n), i.e., neither

T
nor U
.
It is now easy to complete the proof of Theorem 6.15. Suppose that
S
1
, . . . , S
t
are distinct subsets of S with the same element sums. By the
above fact, S
1
, . . . , S
t
is an antichain in M(n). Hence t cannot exceed the
size of the largest antichain in M(n). By Theorem 6.14, the size of the largest
antichain in M(n) is the size p
1
2
(
n+1
2
)
of the middle rank. By equation (6.11)
this number is equal to f([n],
1
2
_
n+1
2
_
). In other words,
t f
_
[n],
_
1
2
_
n + 1
2
___
,
which is what we wanted to prove.
Note. Theorem 6.15 is known as the weak Erdos-Moser conjecture. The
original (strong) Erdos-Moser conjecture deals with the case S R rather
than S R
+
. There is a dierence between these two cases; for instance,
h(3) = 2 (corresponding to S = 1, 2, 3 and = 3), while the set 1, 0, 1
has four subsets whose elements sum to 0 (including the empty set). (Can
you see where the proof of Theorem 6.15 breaks down if we allow S R?)
The original Erdos-Moser conjecture asserts that if #S = 2m + 1, then
f(S, ) f(m, m + 1, . . . , m, 0). (6.13)
This result can be proved by a somewhat tricky modication of the proof
given above for the weak case; see Exercise 6.3. No proof of the Erdos-
Moser conjecture (weak or strong) is known other than the one indicated
here (sometimes given in a more sophisticated context, as explained in the
next Note).
Note. The key to the proof of Theorem 6.15 is the denition of U
i
and
D
i
which gives the commutation relation (6.12). The reader may be wonder-
ing how anyone managed to discover these denitions (especially that of D
i
).
In fact, the original proof of Theorem 6.15 was based on the representation
theory of the orthogonal Lie algebra o(2n+1, C). In this context, the deni-
tions of U
i
and D
i
are built into the theory of the principal subalgebras of
o(2n+1, C). R. A. Proctor was the rst to remove the representation theory
from the proof and present it solely in terms of linear algebra.
For a undergraduate level introduction to the theory of partitions, see
Andrews and Eriksson, [4]. A more extensive treatment is given by Andrews
[3], while a brief introduction appears in [85, 1.8].
As already mentioned in the text, the rank-unimodality of L(m, n), that
is, of the coecients of the q-binomial coecient
_
m+n
m
_
, is due to J. J.
Sylvester [88], with a combinatorial proof later given by K. M. OHara [61].
An explication of OHaras work was given by D. Zeilberger [95].
The unimodality of the coecients of the polynomial (1+q)(1+q
2
) (1+
q
n
) is implicit in the work of E. B. Dynkin [23][24, p. 332]. J. W. B. Hughes
was the rst to observe explicitly that this polynomial arises as a special
case of Dynkins work. The Spernicity of L(m, n) and M(n), and a proof
of the Erdos-Moser conjecture were rst given by Stanley [79]. It was men-
tioned in the text above that R. A. Proctor [67] was the rst to remove the
representation theory from the proof and present it solely in terms of linear
algebra.
Chapter 7
Enumeration under group
action.
In Sections 5 and 6 we considered the quotient poset B
n
/G, where G is a
subgroup of the symmetric group S
n
. If p
i
is the number of elements of rank
i of this poset, then the sequence p
0
, p
1
, . . . , p
n
is rank-symmetric and rank-
unimodal. Thus it is natural to ask whether there is some nice formula for the
numbers p
i
. For instance, in Theorem 5.9 p
i
is the number of nonisomorphic
graphs with m vertices (where n =
_
m
2
_
) and i edges; is there some nice
formula for this number? For the group G
mn
= S
n
S
m
of Theorem 6.6 we
obtained a simple generating function for p
i
(i.e., a formula for the polynomial
i
p
i
q
i
), but this was a very special situation. In this section we will present
a general theory for enumerating inequivalent objects subject to a group of
symmetries, which will include a formula for the generating function
i
p
i
q
i
as a special case, where p
i
is the number of elements of rank i of B
n
/G. The
chief architect of this theory is G. Polya (though much of it was anticipated
by J. H. Redeld [70]) and hence is often called Polyas theory of enumeration
or just Polya theory. See the references at the end of this chapter for further
historical information.
Polya theory is most easily understood in terms of colorings of some ge-
ometric or combinatorial object. For instance, consider a row of ve squares:
In how many ways can we color the squares using n colors? Each square can
81
82 CHAPTER 7. ENUMERATION UNDER GROUP ACTION.
be colored any of the n colors, so there are n
5
ways in all. These colorings
can by indicated as
C A B E D
where A, B, C, D, E are the ve colors. Now assume that we are allowed to
rotate the ve squares 180
, and that two colorings are considered the same

if one can be obtained from the other by such a rotation. (We may think that
we have cut the row of ve squares out of paper and colored them on one
side.) We say that two colorings are equivalent if they are the same or can be
transformed into one another by a 180
rotation. The rst naive assumption

is that every coloring is equivalent to exactly one other (besides itself), so
the number of inequivalent colorings is n
5
/2. Clearly this reasoning cannot
be correct since n
5
/2 is not always an integer! The problem, of course, is
that some colorings stay the same when we rotate 180
. In fact, these are

exactly the colorings
A B C B A
where A, B, C are any three colors. There are n
3
such colorings, so the total
number of inequivalent colorings is given by
1
2
(number of colorings which dont equal their 180
rotation)
+(number of colorings which equal their 180
rotation
=
1
2
(n
5
n
3
) +n
3
=
1
2
(n
5
+n
3
).
Polya theory gives a systematic method for obtaining formulas of this sort
for any underlying symmetry group.
The general setup is the following. Let X be a nite set, and G a subgroup
of the symmetric group S
X
. Think of G as a group of symmetries of X. Let
C be another set (which may be innite), which we think of as a set of
colors. A coloring of X is a function f : X C. For instance, X could
be the set of four squares of a 2 2 chessboard, labelled as follows:
83
1 2
3 4
Let C = r, b, y (the colors red, blue, and yellow). A typical coloring of
X would then look like
y r
r b
The above diagram thus indicates the function f : X C given by f(1) =
r, f(2) = b, f(3) = y, f(4) = r.
Note. We could work in the slightly greater generality of a group G
acting on the set X, i.e., we are given a homomorphism : G S
X
that
need not be injective. However, we then have a well-dened induced injective
homomorphism : H S
X
, where H = G/ker. The results obtained
below for H are identical to those we get for G, so nothing is lost by assuming
that is injective. In this case we can identify G with its image (G).
We dene two colorings f and g to be equivalent (or G-equivalent, when
it is necessary to specify the group), denoted f g or f
G
g, if there exists
an element G such that
g((x)) = f(x) for all x X.
We may write this condition more succinctly as g = f, where g denotes
the composition of functions (from right to left). It is easy to check, using
the fact that G is a group, that is an equivalence relation. One should
think that equivalent functions are the same up to symmetry.
7.1 Example. Let X be the 2 2 chessboard and C = r, b, y as above.
There are many possible choices of a symmetry group G, and this will af-
fect when two colorings are equivalent. For instance, consider the following
groups:
G
1
consists of only the identity permutation (1)(2)(3)(4).
G
2
is the group generated by a vertical reection. It consists of the
two elements (1)(2)(3)(4) (the identity element) and (1, 2)(3, 4) (the
vertical reection).
G
3
is the group generated by a reection in the main diagonal. It
consists of the two elements (1)(2)(3)(4) (the identity element) and
(1)(4)(2, 3) (the diagonal reection).
G
4
is the group of all rotations of X. It is a cyclic group of order four
with elements (1)(2)(3)(4), (1, 2, 4, 3), (1, 4)(2, 3), and (1, 3, 4, 2).
G
5
is the dihedral group of all rotations and reections of X. It has
eight elements, namely, the four elements of G
4
and the four reections
(1, 2)(3, 4), (1, 3)(2, 4), (1)(4)(2, 3), and (2)(3)(1, 4).
G
6
is the symmetric group of all 24 permutations of X. Although this
is a perfectly valid group of symmetries, it no longer has any connec-
tion with the geometric representation of X as the squares of a 2 2
chessboard.
Consider the inequivalent colorings of X with two red squares, one blue
square, and one yellow square, in each of the six cases above.
(G
1
) There are twelve colorings in all with two red squares, one blue square,
and one yellow square, and all are inequivalent under the trivial group
(the group with one element). In general, whenever G is the trivial
group then two colorings are equivalent if and only if they are the same
[why?].
(G
2
) There are now six inequivalent colorings, represented by
r r
y
r
r y
r
r
y y
r r
r
y r
r y
r b
b
b
b b
b
Each equivalence class contains two elements.
(G
3
) Now there are seven classes, represented by
r r
y
r r
y
y
r r
y
r r
r
y r
r
r y r
y r
b b b
b b b b
85
The rst ve classes contain two elements each and the last two classes
only one element. Although G
2
and G
3
are isomorphic as abstract
groups, as permutation groups they have a dierent structure. Speci-
cally, the generator (1, 2)(3, 4) of G
2
has two cycles of length two, while
the generator (1)(4)(2, 3) has two cycles of length one and one of length
two. As we will see below, it is the lengths of the cycles of the elements
of G that determine the sizes of the equivalence classes. This explains
why the number of classes for G
2
and G
3
are dierent.
(G
4
) There are three classes, each with four elements. The size of each
class is equal to the order of the group because none of the colorings
have any symmetry with respect to the group, i.e., for any coloring
f, the only group element that xes f (so f = f) is the identity
( = (1)(2)(3)(4)).
r r
y
r r
y
r
y r b b
b
(G
5
) Under the full dihedral group there are now two classes.
r
y
r r
y r b
b
The rst class has eight elements and the second four elements. In
general, the size of a class is the index in G of the subgroup xing
some xed coloring in that class [why?]. For instance, the subgroup
xing the second coloring above is (1)(2)(3)(4), (1, 4)(2)(3), which
has index four in the dihedral group of order eight.
(G
6
) Under the group S
4
of all permutations of the squares there is clearly
only one class, with all twelve colorings. In general, for any set X if the
group is the symmetric group S
X
then two colorings are equivalent if
and only if each color appears the same number of times [why?].
Our object in general is to count the number of equivalence classes of
colorings which use each color a specied number of times. We will put the
information into a generating function a polynomial whose coecients are
the numbers we seek. Consider for example the set X, the group G = G
5
(the
dihedral group), and the set C = r, b, y of colors in Example 7.1 above.
Let (i, j, k) be the number of inequivalent colorings using red i times, blue
j times, and yellow k times. Think of the colors r, b, y as variables, and form
the polynomial
F
G
(r, b, y) =
i+j+k=4
(i, j, k)r
i
b
j
y
k
.
Note that we sum only over i, j, k satisfying i +j +k = 4 since a total of four
colors will be used to color the four-element set X. The reader should check
that
F
G
(r, b, y) = (r
4
+b
4
+y
4
) + (r
3
b +rb
3
+ r
3
y +ry
3
+b
3
y +by
3
)
+2(r
2
b
2
+r
2
y
2
+b
2
y
2
) + 2(r
2
by +rb
2
y +rby
2
).
For instance, the coecient of r
2
by is two because, as we have seen above,
there are two inequivalent colorings using the colors r, r, b, y. Note that
F
G
(r, b, y) is a symmetric function of the variables r, b, y (i.e., it stays the
same if we permute the variables in any way), because insofar as counting
inequivalent colorings goes, it makes no dierence what names we give the
colors. As a special case we may ask for the total number of inequivalent col-
orings with four colors. This obtained by setting r = b = y = 1 in F
G
(r, b, y)
[why?], yielding F
G
(1, 1, 1) = 3 + 6 + 2 3 + 2 3 = 21.
What happens to the generating function F
G
in the above example when
we use the n colors r
1
, r
2
, . . . , r
n
(which can be thought of as dierent shades
of red)? Clearly all that matters are the multiplicities of the colors, without
regard for their order. In other words, there are ve cases: (a) all four colors
the same, (b) one color used three times and another used once, (c) two
colors used twice each, (d) one color used twice and two others once each,
and (e) four colors used once each. These ve cases correspond to the ve
partitions of 4, i.e., the ve ways of writing 4 as a sum of positive integers
without regard to order: 4, 3+1, 2+2, 2+1+1, 1+1+1+1. Our generating
87
function becomes
F
G
(r
1
, r
2
, . . . , r
n
) =
i
r
4
i
+
i=j
r
3
i
r
j
+2
i<j
r
2
i
r
2
j
+ 2
i=j
i=k
j<k
r
2
i
r
j
r
k
+ 3
i<j<k<l
r
i
r
j
r
k
r
l
,
where the indices in each sum lie between 1 and n. If we set all variables
equal to one (obtaining the total number of colorings with n colors), then
simple combinatorial reasoning yields
F
G
(1, 1, . . . , 1) = n +n(n 1) + 2
_
n
2
_
+ 2n
_
n 1
2
_
+ 3
_
n
4
_
=
1
8
(n
4
+ 2n
3
+ 3n
2
+ 2n). (7.1)
Note that the polynomial (7.1) has the following description: The denomina-
tor 8 is the order of the group G
5
, and the coecient of n
i
in the numerator
is just the number of permutations in G
5
with i cycles! For instance, the
coecient of n
2
is 3, and G
5
has the three elements (1, 2)(3, 4), (1, 3)(2, 4),
and (1, 4)(2, 3) with two cycles. We want to prove a general result of this
nature.
The basic tool which we will use is a simple result from the theory of
permutation groups known as Burnsides lemma. It was actually rst proved
by Cauchy when G is transitive (i.e., [Y/G[ = 1 in Lemma 7.2 below) and by
Frobenius in the general case, and is sometimes called the Cauchy-Frobenius
lemma.
7.2 Lemma (Burnsides lemma). Let Y be a nite set and G a subgroup of
S
Y
. For each G, let
Fix() = y Y : (y) = y,
so #Fix() is the number of cycles of length one in the permutation . Let
Y/G be the set of orbits of G. Then
[Y/G[ =
1
#G
G
#Fix().
An equivalent form of Burnsides lemma is the statement that the average
number of elements of Y xed by an element of G is equal to the number of
orbits. Before proceeding to the proof, let us consider an example.
7.3 Example. Let Y = a, b, c, d,
G = (a)(b)(c)(d), (a, b)(c, d), (a, c)(b, d), (a, d)(b, c),
and
G
= (a)(b)(c)(d), (a, b)(c)(d), (a)(b)(c, d), (a, b)(c, d).

Both groups are isomorphic to Z
2
Z
2
(compare Example 5.1(c) and (d)).
By Burnsides lemma the number of orbits of G is
1
4
(4 + 0 + 0 + 0) = 1.
Indeed, given any two elements i, j Y , it is clear by inspection that there
is a G (which happens to be unique) such that (i) = j. On the other
hand, the number of orbits of G
is
1
4
(4 + 2 + 2 + 0) = 2. Indeed, the two
orbits are a, b and c, d.
Proof of Burnsides lemma. For y Y let G
y
= G : y = y
(the set of permutations xing y). Then
1
#G
G
[Fix()[ =
1
#G
yY
y=y
1
=
1
#G
yY
G
y=y
1
=
1
#G
yY
#G
y
.
Now (as in the proof of Lemma 5.6) the multiset of elements y, G,
contains every element in the orbit Gy the same number of times, namely
#G/#Gy times. Thus y occurs #G/[Gy[ times among the y, so
#G
#Gy
= #G
y
.
Thus
1
#G
G
#Fix() =
1
#G
yY
#G
#Gy
=
yY
1
#Gy
.
89
How many times does a term 1/#O appear in the above sum, where O is
a xed orbit? We are asking for the number of y such that Gy = O. But
Gy = O if and only if y O, so 1/#O appears #O times. Thus each orbit
gets counted exactly once, so the above sum is equal to the number of orbits.
7.4 Example. How many inequivalent colorings of the vertices of a regular

hexagon H are there using n colors, under cyclic symmetry? Let (
n
be the
set of all n-colorings of H. Let G be the group of all permutations of (
n
which
permute the colors cyclically, so G

= Z
6
. We are asking for the number of
orbits of G [why?]. We want to apply Burnsides lemma, so for each of the
six elements of G we need to compute the number of colorings xed by
that element. Let be a generator of G.
= 1 (the identity): All n
6
colorings are xed by .
= ,
1
: Only the n colorings with all colors equal are xed.
=
2
,
4
: Any coloring of the form ababab is xed (writing the colors
linearly in the order they appear around the hexagon, starting at any
xed vertex). There are n choices for a and n for b, so n
2
colorings in
all.
=
3
: The xed colorings are of the form abcabc, so n
3
in all.
Hence by Burnsides lemma, we have
number of orbits =
1
6
(n
6
+n
3
+ 2n
2
+ 2n).
The reader who has followed the preceding example will have no trouble
understanding the following result.
7.5 Theorem. Let G be a group of permutations of a nite set X. Then the
number N
G
(n) of inequivalent (with respect to G) n-colorings of X is given
by
N
G
(n) =
1
#G
G
n
c()
, (7.2)
where c() denotes the number of cycles of .
Proof. Let
n
denote the action of G on the set (
n
of n-colorings of
X. We want to determine the set Fix(
n
), so that we can apply Burnsides
lemma. Let C be the set of n colors. If f : X C is a coloring xed by ,
then for all x X we have
f(x) =
n
f(x) = f((x)).
Thus f Fix(
n
) if and only if f(x) = f((x)). Hence f(x) = f(
k
(x)) for
any k 1 [why?]. The elements y of X of the form
k
(x) for k 1 are just
the elements of the cycle of containing x. Thus to obtain f Fix(
n
),
we should take the cycles
1
, . . . ,
c()
of and color each element of
i
the
same color. There are n choices for each
i
, so n
c()
colorings in all xed by
. In other words, #Fix(
n
) = n
c()
, and the proof follows by Burnsides
lemma.
We would now like not just to count the total number of inequivalent
colorings with n-colors, but more strongly to specify the number of occurences
of each color. We will need to use not just the number c() of cycles of each
G, but rather the lengths of each of the cycles of . Thus given a
permutation of an n-element set X, dene the type of to be
type() = (c
1
, c
2
, . . . , c
n
),
where has c
i
i-cycles. For instance, if = 4, 7, 3, 8, 2, 10, 11, 1, 6, 9, 5, then
type() = type (1, 4, 8)(2, 7, 11, 5)(3)(6, 10, 9)
= (1, 0, 2, 1, 0, 0, 0, 0, 0, 0, 0).
Note that we always have
i
ic
i
= n [why?]. Dene the cycle indicator of
to be the monomial
Z
= z
c
1
1
z
c
2
2
z
cn
n
.
(Many other notations are used for the cycle indicator. The use of Z
comes
from the German word Zyklus for cycle. The original paper of Polya was
written in German.) Thus for the example above, we have Z
= z
1
z
2
3
z
4
.
Now given a subgroup G of S
X
, the cycle indicator (or cycle index poly-
nomial ) of G is dened by
Z
G
= Z
G
(z
1
, . . . , z
n
) =
1
#G
G
Z
.
Thus Z
G
(also denoted P
G
, Cyc(G), etc.) is a polynomial in the variables
z
1
, . . . , z
n
.
91
7.6 Example. If X consists of the vertices of a square and G is the group
of rotations of X (a cyclic group of order 4), then
Z
G
=
1
4
(z
4
1
+z
2
2
+ 2z
4
).
If reections are also allowed (so G is the dihedral group of order 8), then
Z
G
=
1
8
(z
4
1
+ 3z
2
2
+ 2z
2
1
z
2
+ 2z
4
).
We are now ready to state the main result of this section.
7.7 Theorem (Polyas theorem, 1937). Let G be a group of permutations of
the n-element set X. Let C = r
1
, r
2
, . . . be a set of colors. Let (i
1
, i
2
, . . . )
be the number of inequivalent (under the action of G) colorings f : X C
such that color r
j
is used i
j
times. Dene
F
G
(r
1
, r
2
, . . . ) =
i
1
,i
2
,...
(i
1
, i
2
, . . . )r
i
1
1
r
i
2
2
.
(Thus F
G
is a polynomial or a power series in the variables r
1
, r
2
, . . . , de-
pending on whether or not C is nite or innite.) Then
F
G
(r
1
, r
2
, . . . ) = Z
G
(r
1
+r
2
+r
3
+ , r
2
1
+r
2
2
+r
2
3
+ , . . . , r
j
1
+r
j
2
+r
j
3
+ ).
(In other words, substitute
i
r
j
i
for z
j
in Z
G
.)
Before giving the proof let us consider an example.
7.8 Example. Suppose that in Example 7.6 our set of colors is C = a, b, c, d,
and that we take G to be the group of cyclic symmetries. Then
F
G
(a, b, c, d) =
1
4
_
(a +b +c +d)
4
+ (a
2
+b
2
+c
2
+d
2
)
2
+ 2(a
4
+b
4
+c
4
+d
4
)
_
= (a
4
+ ) + (a
3
b + ) + 2(a
2
b
2
+ ) + 3(a
2
bc + ) + 6abcd.
An expression such as (a
2
b
2
+ ) stands for the sum of all monomials in the
variables a, b, c, d with exponents 2, 2, 0, 0 (in some order). The coecient of
all such monomials is 2, indicating two inequivalent colorings using one color
twice and another color twice. If instead G were the full dihedral group, we
would get
F
G
(a, b, c, d) =
1
8
_
(a +b +c +d)
4
+ 3(a
2
+b
2
+c
2
+d
2
)
2
+ 2(a +b +c +d)
2
(a
2
+b
2
+c
2
+d
2
) + 2(a
4
+b
4
+c
4
+d
4
)
_
= (a
4
+ ) + (a
3
b + ) + 2(a
2
b
2
+ ) + 2(a
2
bc + ) + 3abcd.
Proof of Polyas theorem. Let #X = t and i
1
+ i
2
+ = t, where
each i
j
0. Let i = (i
1
, i
2
, ...), and let (
i
denote the set of all colorings
of X with color r
j
used i
j
times. The group G acts on (
i
, since if f (
i
and G, then f (
i
. (Rotating a colored object does not change
how many times each color appears.) Let
i
denote the action of on (
i
.
We want to apply Burnsides lemma to compute the number of orbits, so we
need to nd #Fix(
i
).
In order for f Fix(
i
), we must color X so that (a) in any cycle of ,
all the elements get the same color, and (b) the color r
j
appears i
j
times.
Consider the product
H
j
(r
j
1
+r
j
2
+ )
c
j
()
,
where c
j
() is the number of j-cycles (cycles of length j) of . When we
expand this product as a sum of monomials r
j
1
1
r
j
2
2
, we get one of these
monomials by choosing a term r
j
k
from each factor of H
and multiplying
these terms together. Choosing r
j
k
corresponds to coloring all the elements
of some j-cycle with r
k
. Since a factor r
j
1
+ r
j
2
+ occurs precisely c
j
()
times in H
, choosing a term r
j
k
from every factor corresponds to coloring
X so that every cycle is monochromatic (i.e., all the elements of that cycle
get the same color). The product of these terms r
j
k
will be the monomial
r
j
1
1
r
j
2
2
, where we have used color r
k
a total of j
k
times. It follows that the
coecient of r
i
1
i
r
i
2
2
in H
is equal to #Fix(
i
). Thus
H
i
#Fix(
i
)r
i
1
1
r
i
2
2
. (7.3)
Now sum both sides of (7.3) over all G and divide by #G. The left-hand
side becomes
1
#G
j
(r
j
1
+r
j
2
+ )
c
j
()
= Z
G
(r
1
+r
2
+ , r
2
1
+r
2
2
+ , . . . ).
On the other hand, the right-hand side becomes
i
_
1
#G
G
#Fix(
i
)
_
r
i
1
1
r
i
2
2
.
By Burnsides lemma, the expression in brackets is just the number of orbits
of
i
acting on (
i
, i.e., the number of inequivalent colorings using color r
j
a
total of i
j
times, as was to be proved.
93
7.9 Example. (Necklaces) A necklace of length is a circular arrangement
of (colored) beads. Two necklaces are considered the same if they are
cyclic rotations of one another. Let X be a set of (uncolored) beads, say
X = 1, 2, . . . , . Regarding the beads as being placed equidistantly on a
circle in the order 1, 2 . . . , , let G be the cyclic group of rotations of X. Thus
if is the cycle (1, 2, . . . , ), then G = 1, ,
2
, . . . ,
1
. For example, if
= 6 then the elements of G are
0
= (1)(2)(3)(4)(5)(6)
= (1, 2, 3, 4, 5, 6)
2
= (1, 3, 5)(2, 4, 6)
3
= (1, 4)(2, 5)(3, 6)
4
= (1, 5, 3)(2, 6, 4)
5
= (1, 6, 5, 4, 3, 2).
In general, if d is the greatest common divisor of m and (denoted d =
gcd(m, )), then
m
has d cycles of length /d. An integer m satises 1
m and gcd(m, ) = d if and only if 1 m/d /d and gcd(m/d, /d) = 1.
Hence the number of such integers m is given by the Euler phi-function (or
totient function) (/d), which by denition is equal to the number of integers
1 i /d such that gcd(i, /d) = 1. Recall that (k) can be computed by
the formula
(k) = k
p|k
p prime
_
1
1
p
_
. (7.4)
For instance, (1000) = 1000(1
1
2
)(1
1
5
) = 400. Putting all this together
gives the following formula for the cycle enumerator Z
G
(z
1
, . . . , z
):
Z
G
(z
1
, . . . , z
) =
1
d|
(/d)z
d
/d
,
or (substituting /d for d),
Z
G
(z
1
, . . . , z
) =
1
d|
(d)z
/d
d
.
There follows from Polyas theorem the following result (originally proved by
P. A. MacMahon (18541929) before Polya discovered his general result).
7.10 Theorem. (a) The number N
(n) of n-colored necklaces of length

is given by
N
(n) =
1
d|
(/d)n
d
. (7.5)
(b) We have
F
G
(r
1
, r
2
, . . . ) =
1
d|
(d)(r
d
1
+r
d
2
+ )
/d
.
Note: (b) reduces to (a) if r
1
= r
2
= = 1. Moreover, since clearly
N
(1) = 1, putting n = 1 in (7.5) yields the famous identity
d|
(/d) = .
What if we are allowed to ip necklaces over, not just rotate them? Now
the group becomes the dihedral group of order 2, and the corresponding
inequivalent colorings are called dihedral necklaces. We leave to the reader
to work out the cycle enumerators
1
2
_
_
d|
(d)z
/d
d
+mz
2
1
z
m1
2
+mz
m
2
_
_
, if = 2m
1
2
_
_
d|
(d)z
/d
d
+z
1
z
m
2
_
_
, if = 2m + 1.
7.11 Example. n-colored *proper* 4-necklaces ??
7.12 Example. Let G = S
, the group of all permutations of 1, 2, . . . , =

X. Thus for instance
Z
S
3
(z
1
, z
2
, z
3
) =
1
6
(z
3
1
+ 3z
1
z
2
+ 2z
3
)
Z
S
4
(z
1
, z
2
, z
3
, z
4
) =
1
24
(z
4
1
+ 6z
2
1
z
2
+ 3z
2
2
+ 8z
1
z
3
+ 6z
4
).
It is easy to count the number of inequivalent colorings in (
i
. If two colorings
of X use each color the same number of times, then clearly there is some
95
permutation of X which sends one of the colorings to the other. Hence (
i
consists of a single orbit. Thus
F
S
(r
1
, r
2
, . . . ) =
i
1
+i
2
+=
r
i
1
1
r
i
2
2
,
the sum of all monomials of degree .
To count the total number of inequivalent n-colorings, note that
0
F
S
(r
1
, r
2
, . . . )x
=
1
(1 r
1
x)(1 r
2
x)
. (7.6)
since if we expand each factor on the right-hand side into the series
j0
r
j
i
x
j
and multiply, the coecient of x
will just be the sum of all monomials of

degree . For xed n, let f
n
() denote the number of inequivalent n-colorings
of X. Since f
n
() = F
S
(1, 1, . . . , 1) (n 1s in all), there follows from (7.6)

that
0
f
n
()x
=
1
(1 x)
n
.
The right-hand side can be expanded (e.g. by Taylors theorem) as
1
(1 x)
n
=
0
_
n + 1
_
x
.
Hence
f
n
() =
_
n + 1
_
.
It is natural to ask whether there might be a more direct proof of such a
simple result. This is actually a standard result in elementary enumerative
combinatorics. For xed and n we want the number of solutions to i
1
+
i
2
+ +i
n
= in nonnegative integers. Setting k
j
= i
j
+1, this is the same
as the number of solutions to k
1
+k
2
+ +k
n
= +n in positive integers.
Place + n dots in a horizontal line. There are + n 1 spaces between
the dots. Choose n 1 of these spaces and draw a vertical bar in them in
_
n+1
n1
_
=
_
n+1
_
ways. For example, if n = 5 and = 6, then one way of
drawing the bars is
s s s s s s s s s s s
The number of dots in each compartment, read from left to right, gives
the numbers k
1
, . . . , k
n
. For the above example we get 2 +3 +2 +1 +3 = 11,
corresponding to the original solution 1 +2 +1 +0 +2 = 6 (i.e., one element
of X colored r
1
, two elements colored r
2
, one colored r
3
, and two colored r
5
).
Since this correspondence between solutions to i
1
+i
2
+ +i
n
= and sets
of bars is clearly a bijection, we get
_
n+1
_
solutions as claimed.
Recall (Theorem 7.5) that the number of inequivalent n-colorings of X
(with respect to any group G of permutations of X) is given by
1
#G
G
n
c()
,
where c() denotes the number of cycles of . Hence for G = S
we get the
identity
1
!
n
c()
=
_
n + 1
_
=
1
!
n(n + 1)(n + 2) (n + 1).
Multiplying by ! yields
n
c()
= n(n + 1)(n + 2) (n + 1). (7.7)
Equivalently [why?], if we dene c(, k) to be the number of permutations in
S
with k cycles (called a signless Stirling number of the rst kind), then
k=1
c(, k)x
k
= x(x + 1)(x + 2) (x + 1).
For instance, x(x + 1)(x + 2)(x + 3) = x
4
+ 6x
3
+ 11x
2
+ 6x, so (taking
the coecient of x
2
) eleven permutations in S
4
have two cycles, namely,
(123)(4), (132)(4), (124)(3), (142)(3), (134)(2), (143)(2), (234)(1), (243)(1),
(12)(34), (13)(24), (14)(23).
Although it was easy to compute the generating function F
S
(r
1
, r
2
, . . . )
directly without the necessity of computing the cycle indicator Z
S
(z
1
, . . . , z
),
we can still ask whether there is a formula of some kind for this polynomial.
First we determine explicitly its coecients.
97
7.13 Theorem. Let
ic
i
= . The number of permutations S
with c
i
cycles of length i (or equivalently, the coecient of z
c
1
1
z
c
2
2
in
!Z
S
(z
1
, . . . , z
)) is equal to !/1
c
1
c
1
!2
c
2
c
2
! .
Example. The number of permutations in S
15
with three 1-cycles, two
2-cycles, and two 4-cycles is 15!/1
3
3! 2
2
2! 4
2
2! = 851, 350, 500.
Proof of Theorem 7.13. Fix c = (c
1
, c
2
, . . . ) and let X
c
be the set
of all permutations S
with c
i
cycles of length i. Given a permutation
= a
1
a
2
a
in S
, construct a permutation f() X

c
as follows. Let
the 1-cycles of f() be (a
1
), (a
2
), . . . , (a
c
1
). Then let the 2-cycles of f()
be (a
c
1
+1
, a
c
1
+2
), (a
c
1
+3
, a
c
1
+4
), . . . , (a
c
1
+2c
2
1
, a
c
1
+2c
2
). Then let the 3-cycles
of f() be (a
c
1
+2c
2
+1
, a
c
1
+2c
2
+2
, a
c
1
+2c
2
+3
), (a
c
1
+2c
2
+4
, a
c
1
+2c
2
+5
, a
c
1
+2c
2
+6
), . . . ,
(a
c
1
+2c
2
+3c
3
2
, a
c
1
+2c
2
+3c
3
1
, a
c
1
+2c
2
+3c
3
), etc., continuing until we reach a
and
have produced a permutation in X
c
. For instance, if = 11, c
1
= 3, c
2
=
2, c
4
= 1, and = 4, 9, 6, 11, 7, 1, 3, 8, 10, 2, 5, then
f() = (4)(9)(6)(11, 7)(1, 3)(8, 10, 2, 5).
We have dened a function f : S
X
c
. Given X
c
, what is #f
1
(),
the number of permutations sent to by f? A cycle of length i can be written
in i ways, namely,
(b
1
, b
2
, . . . , b
i
) = (b
2
, b
3
, . . . , b
i
, b
1
) = = (b
i
, b
1
, b
2
, . . . , b
i1
).
Moreover, there are c
i
! ways to order the c
i
cycles of length i. Hence
#f
1
() = c
1
!c
2
!c
3
! 1
c
1
2
c
2
3
c
3
,
the same number for any X
c
. It follows that
#X
c
=
#S
c
1
!c
2
! 1
c
1
2
c
2

=
!
c
1
!c
2
! 1
c
1
2
c
2

,
as was to be proved.
As for the polynomial Z
S
itself, we have the following result. Write

exp y = e
y
.
7.14 Theorem. We have
0
Z
S
(z
1
, z
2
, . . . )x
= exp
_
z
1
x +z
2
x
2
2
+z
3
x
3
3
+
_
.
Proof. There are some sophisticated ways to prove this theorem which ex-
plain why the exponential function appears, but we will be content here
with a naive proof. Write
e
z
1
x+z
2
x
2
2
+z
3
x
3
3
+
= e
zx
e
z
2
x
2
2
e
z
3
x
3
3

=
_
n0
z
n
1
x
n
n!
__
n0
z
n
2
x
2n
2
n
n!
__
n0
z
n
3
x
3n
3
n
n!
_
.
When we multiply this product out, the coecient of z
c
1
1
z
c
2
2
x
, where
= c
1
+ 2c
2
+ , is given by
1
1
c
1
c
1
!2
c
2
c
2
!
=
1
!
_
!
1
c
1
c
1
!2
c
2
c
2
!
_
.
By Theorem 7.13 this is just the coecient of z
c
1
1
z
c
2
2
in Z
S
(z
1
, z
2
, . . . ), as
was to be proved.
As a check of Theorem 7.14, set each z
i
= n to obtain
0
Z
S
(n, n, . . . )x
= e
nx+n
x
2
2
+n
x
3
3
+
= e
n(x+
x
2
2
+
x
3
3
+ )
= e
nlog(1x)
1
=
1
(1 x)
n
=
0
_
n
_
(x)
0
_
n + 1
_
x
,
the last step following from the easily checked equality
_
n
_
= (1)
_
n+1
_
.
Equating coecients of x
in the rst and last terms of the above string of

equalities gives
Z
S
(n, n, . . . ) =
_
n + 1
_
=
n(n + 1) (n + 1)
!
,
99
agreeing with Theorem 7.5 and equation (7.7).
Theorem 7.14 has many enumerative applications. We give one such
result here as an example.
7.15 Proposition. Let f(n) be the number of permutations w S
n
of odd
order. Equivalently, w
k
= 1 for some odd k. Then
f(n) =
_
1
2
3
2
5
2
(n 1)
2
, n even
1
2
3
2
5
2
(n 2)
2
n, n odd.
Proof. A permutation has odd order if and only if all its cycle lengths are
odd. Hence [why?]
f(n) = n!Z
Sn
(z
i
= 1, i odd; z
i
= 0, i even).
Making this substitution in Theorem 7.14 gives
n0
f(n)
x
n
n!
= exp
_
x +
x
3
3
+
x
5
5
+
_
.
Since log(1 x) = x +
x
2
2
+
x
3
3
+ , we get [why?]
n0
f(n)
x
n
n!
= exp
_
1
2
(log(1 x) + log(1 +x))
_
= exp
1
2
log
_
1 +x
1 x
_
=
_
1 +x
1 x
.
We therefore need to nd the coecients in the power series expansion of
_
(1 +x)/(1 x) at x = 0. There is a simple trick for doing so:
_
1 +x
1 x
= (1 +x)(1 x
2
)
1/2
= (1 +x)
m0
_
1/2
m
_
(x
2
)
m
=
m0
(1)
m
_
1/2
m
_
(x
2m
+x
2m+1
),
where by denition
_
1/2
m
_
=
1
m!
_
1
2
__
3
2
_

_
2m1
2
_
.
It is now a routine computation to check that the coecient of x
n
/n! in
_
(1 +x)/(1 x) agrees with the desired value of f(n).
Quotients of boolean algebra. We will show how to apply Polya
theory to the problem of counting the number of elements of given rank in a
quotient poset B
X
/G. Here X is a nite set, B
X
is the boolean algebra of all
subsets of X, and G is a group of permutations of X (with an induced action
on B
X
). What do colorings of X have to do with subsets? The answer is
very simple: A 2-coloring f : X 0, 1 corresponds to a subset S
f
of X by
the usual rule
s S
f
f(s) = 1.
Note that two 2-colorings f and g are G-equivalent if and only if S
f
and S
g
are in the same orbit of G (acting on B
X
). Thus the number of inequivalent
2-colorings f of X with i values equal to 1 is just #(B
X
/G)
i
, the number of
elements of B
X
/G of rank i. As an immediate application of Polyas theorem
(Theorem 7.7) we obtain the following result.
7.16 Corollary. We have
i
#(B
X
/G)
i
q
i
= Z
G
(1 +q, 1 +q
2
, 1 +q
3
, . . . ).
Proof. If (i, j) denotes the number of inequivalent 2-colorings of X with
the colors 0 and 1 such that 0 is used j times and 1 is used i times (so
i +j = #X), then by Polyas theorem we have
i,j
(i, j)x
i
y
j
= Z
G
(x +y, x
2
+y
2
, x
3
+y
3
, . . . ).
Setting x = q and y = 1 yields the desired result [why?].
Combining Corollary 7.16 with the rank-unimodality of B
X
/G (Theo-
rem 5.8) yields the following corollary.
7.17 Corollary. For any nite group G of permutations of a nite set X,
the polynomial Z
G
(1 +q, 1 +q
2
, 1 +q
3
, . . . ) has symmetric, unimodal, integer
coecients.
101
7.18 Example. (a) For the poset P of Example 5.4(a) we have G = (1)(2)(3), (1, 2)(3),
so Z
G
(z
1
, z
2
, z
3
) =
1
2
(z
3
1
+z
1
z
2
). Hence
3
i=0
(#P
i
)q
i
=
1
2
_
(1 +q)
3
+ (1 +q)(1 +q
2
)
_
= 1 + 2q + 2q
2
+q
3
.
(b) For the poset P of Example 5.4(b) we have G = (1)(2)(3)(4)(5),
(1, 2, 3, 4, 5), (1, 3, 5, 2, 4), (1, 4, 2, 5, 3), (1, 5, 4, 3, 2), so Z
G
(z
1
, z
2
, z
3
, z
4
, z
5
) =
1
5
(z
5
1
+ 4z
5
). Hence
5
i=0
(#P
i
)q
i
=
1
5
_
(1 +q)
5
+ 4(1 +q
5
)
_
= 1 +q + 2q
2
+ 2q
3
+q
4
+q
5
.
(c) Let X be the squares of a 2 2 chessboard, labelled as follows:
1 2
3 4
Let G be the wreath product S
2
S
2
, as dened in Section 6. Then
G = (1)(2)(3)(4), (1, 2)(3)(4), (1)(2)(3, 4), (1, 2)(3, 4),
(1, 3)(2, 4), (1, 4)(2, 3), (1, 3, 2, 4), (1, 4, 2, 3),
so
Z
G
(z
1
, z
2
, z
3
, z
4
) =
1
8
(z
4
1
+ 2z
2
1
z
2
+ 3z
2
2
+ 2z
4
).
Hence
4
i=0
(#P
i
)q
i
=
1
4
_
(1 +q)
4
+ 2(1 +q)
2
(1 +q
2
) + 3(1 +q
2
)
2
+ 2(1 +q
4
)
_
= 1 +q + 2q
2
+q
3
+q
4
=
_
4
2
_
,
agreeing with Theorem 6.6.
Using more sophisticated methods (such as the representation theory of
the symmetric group), the following generalization of Corollary 7.17 can be
proved: Let P(q) be any polynomial with symmetric, unimodal, nonnegative,
integer coecients, such as 1 + q + 3q
2
+ 3q
3
+ 8q
4
+ 3q
5
+ 3q
6
+ q
7
+ q
8
or
q
5
+q
6
(= 0+0q + +0q
4
+q
5
+q
6
+0q
7
+ +0q
11
). Then the polynomial
Z
G
(P(q), P(q
2
), P(q
3
), . . . ) has symmetric, unimodal, nonnegative, integer
coecients.
Graphs. A standard application of Polya theory is to the enumeration
of nonisomorphic graphs. We saw at the end of Section 5 that if M is an
m-element vertex set, X =
_
M
2
_
, and S
(2)
m
is the group of permutations of
X induced by permutations of M, then an orbit of i-element subsets of X
may be regarded as an isomorphism class of graphs on the vertex set M with
i-edges. Thus #(B
X
/S
(2)
m
)
i
is the number of nonisomorphic graphs (without
loops or multiple edges) on the vertex set M with i edges. It follows from
Corollary 7.16 that if g
i
(m) denotes the number of nonisomorphic graphs
with m vertices and i edges, then
(
m
2
)
i=0
g
i
(m)q
i
= Z
S
(2)
m
(1 +q, 1 +q
2
, 1 +q
3
, . . . ).
Thus we would like to compute the cycle enumerator Z
S
(2)
m
(z
1
, z
2
, . . . ). If two
permutations and of M have the same cycle type (number of cycles of each
length), then their actions on X also have the same cycle type [why?]. Thus
for each possible cycle type of a permutation of M (i.e., for each partition
of m) we need to compute the induced cycle type on X. We also know
from Theorem 7.13 the number of permutations of M of each type. For
small values of m we can pick some permutation of each type and compute
directly its action on X in order to determine the induced cycle type. For
m = 4 we have:
CYCLE INDUCED CYCLE
LENGTHS PERMUTATION LENGTHS
OF NUMBER
OF
1, 1, 1, 1 1 (1)(2)(3)(4) (12)(13)(14)(23)24)(34) 1, 1, 1, 1, 1, 1
2, 1, 1 6 (1, 2)(3)(4) (12)(12, 23)(14, 24)(34) 2, 2, 1, 1
3, 1 8 (1, 2, 3)(4) (12, 23, 13)(14, 24, 34) 3, 3
2, 2 3 (1, 2)(3, 4) (12)(13, 24)(14, 23)(34) 2, 2, 1, 1
4 6 (1, 2, 3, 4) (12, 23, 34, 14)(13, 24) 4, 2
103
It follows that
Z
S
(2)
4
(z
1
, z
2
, z
3
, z
4
, z
5
, z
6
) =
1
24
(z
6
1
+ 9z
2
1
z
2
2
+ 8z
2
3
+ 6z
2
z
4
).
If we set z
i
= 1 +q
i
and simplify, we obtain the polynomial
6
i=0
g
i
(4)q
i
= 1 + q + 2q
2
+ 3q
3
+ 2q
4
+q
5
+q
6
.
Suppose that we instead wanted to count the number h
i
(4) of nonisomorphic
graphs with four vertices and i edges, where now we allow at most two edges
between any two vertices. We can take M, X, and G = S
(2)
4
as before, but
now we have three colors: red for no edges, blue for one edge, and yellow
for two edges. A monomial r
i
b
j
y
k
corresponds to a coloring with i pairs of
vertices having no edges between them, j pairs having one edge, and k pairs
having two edges. The total number e of edges is j + 2k. Hence if we let
r = 1, b = q, y = q
2
, then the monomial r
i
b
j
y
k
becomes q
j+2k
= q
e
. It follows
that
i(i1)
i=0
h
i
(4)q
i
= Z
S
(2)
4
(1 +q +q
2
, 1 +q
2
+q
4
, 1 + q
3
+q
6
, . . . )
=
1
24
_
(1 +q +q
2
)
6
+ 9(1 +q +q
2
)
2
(1 +q
2
+q
4
)
2
+8(1 +q
3
+q
6
)
2
+ 6(1 +q
2
+q
4
)(1 +q
4
+q
8
)
_
= 1 +q + 3q
2
+ 5q
3
+ 8q
4
+ 9q
5
+ 12q
6
+ 9q
7
+ 8q
8
+ 5q
9
+3q
10
+q
11
+q
12
.
The total number of nonisomorphic graphs on four vertices with edge multi-
plicities at most two is
i
h
i
(4) = 66.
It should now be clear that if we restrict the edge multiplicity to be r,
then the corresponding generating function is Z
S
(2)
4
(1+q +q
2
+ +q
r1
, 1+
q
2
+q
4
+ + q
2r2
, . . . ). In particular, to obtain the total number N(r, 4)
of nonisomorphic graphs on four vertices with edge multiplicity at most r,
we simply set each z
i
= r, obtaining
N(r, 4) = Z
S
(2)
4
(r, r, r, r, r, r)
=
1
24
(r
6
+ 9r
4
+ 14r
2
).
This is the same as number of inequivalent r-colorings of the set X =
_
M
2
_
(where #M = 4) [why?].
Of course the same sort of reasoning can be applied to any number of ver-
tices. For ve vertices our table becomes the following (using such notation
as 1
5
to denote a sequence of ve 1s).
CYCLE INDUCED CYCLE
LENGTHS PERMUTATION LENGTHS
OF NO.
OF
1
5
1 (1)(2)(3)(4)(5) (12)(13) (45) 1
10
2, 1
3
10 (1, 2)(3)(4)(5) (12)(13, 23)(14, 25)(15, 25)(34)(35)(45) 2
3
, 1
4
3, 1
2
20 (1, 2, 3)(4)(5) (12, 23, 13)(14, 24, 34)(15, 25, 35)(45) 3
3
, 1
2
2
, 1 15 (1, 2)(3, 4)(5) (12)(13, 24)(14, 23)(15, 25)(34)(35, 45) 2
4
, 1
2
4, 1 30 (1, 2, 3, 4)(5) (12, 23, 34, 14)(13, 24)(15, 25, 35, 45) 4
2
, 2
3, 2 20 (1, 2, 3)(4, 5) (12, 23, 13)(14, 25, 34, 15, 24, 35)(45) 6, 3, 1
5 24 (1, 2, 3, 4, 5) (12, 23, 34, 45, 15)(13, 24, 35, 14, 25) 5
2
Thus
Z
S
(2)
5
(z
1
, . . . , z
10
) =
1
120
(z
10
1
+10z
4
1
z
3
2
+20z
1
z
3
3
+15z
2
1
z
4
2
+30z
2
z
2
4
+20z
1
z
3
z
6
+24z
2
5
),
from which we compute
10
i=0
g
i
(5)q
i
= Z
S
(2)
5
(1 +q, 1 +q
2
, . . . , 1 +q
10
)
= 1 +q + 2q
2
+ 4q
3
+ 6q
4
+ 6q
5
+ 6q
6
+ 4q
7
+ 2q
8
+q
9
+q
10
.
For an arbitrary number m = #M of vertices there exist explicit formulas
for the cycle indicator of the induced action of S
M
on
_
M
2
_
, thereby
obviating the need to compute
explicitly as we did in the above tables,

but the overall expression for Z
S
(2)
m
cannot be simplied signicantly or put
into a simple generating function as we did in Theorem 7.14. For reference
we record
Z
S
(2)
6
=
1
6!
(z
15
1
+ 15z
7
1
z
4
2
+ 40z
3
1
z
4
3
+ 45z
3
1
z
6
2
+ 90z
1
z
2
z
3
4
+ 120z
1
z
2
z
2
3
z
6
+144z
3
5
+ 15z
3
1
z
6
2
+ 90z
1
z
2
z
3
4
+ 40z
5
3
+ 120z
3
z
2
6
)
105
(g
0
(6), g
1
(6), . . . , g
15
(6)) = (1, 1, 2, 5, 9, 15, 21, 24, 24, 21, 15, 9, 5, 2, 1, 1).
Moreover if u(n) denotes the number of nonisomorphic simple graphs with n
vertices, then
(u(0), u(1), . . . , u(11))
= (1, 1, 2, 4, 11, 34, 156, 1044, 12346, 274668, 12005168, 1018997864).
A table of u(n) for n 75 is given at
http://www.research.att.com/njas/sequences/b000088.txt
In particular,
u(75) = 91965776790545918117055311393231179873443957239
0555232344598910500368551136102062542965342147
8723210428876893185920222186100317580740213865
7140377683043095632048495393006440764501648363
4760490012493552274952950606265577383468983364
6883724923654397496226869104105041619919159586
8518775275216748149124234654756641508154401414
8480274454866344981385848105320672784068407907
1134767688676890584660201791139593590722767979
8617445756819562952590259920801220117529208077
0705444809177422214784902579514964768094933848
3173060596932480677345855848701061537676603425
1254842843718829212212327337499413913712750831
0550986833980707875560051306072520155744624852
0263616216031346723897074759199703968653839368
77636080643275926566803872596099072,
a number of 726 digits! Compare
2
(
75
2
)
75!
= .9196577679054591809 10
726
,
which agrees with u(75) to 17 signicant digits [why?].
Burnsides lemma (Lemma 7.2) was actually rst stated and proved by
Frobenius [31, end of 4]. Frobenius in turn credits Cauchy [16, p. 286]
for proving the lemma in the transitive case. Burnside, in the rst edi-
tion of his book [13, 118119], attributes the lemma to Frobenius, but in
the second edition [14] this citation is absent. For more on the history of
Burnsides lemma, see [59] and [93]. Many authors now call this result the
Cauchy-Frobenius lemma. The cycle indicator Z
G
(z
1
, z
2
, . . . ) (where G is a
subgroup of S
n
) was rst considered by J. H. Redeld [70], who called it
the group reduction function, denoted Grf(G). G. Polya [63] independently
dened the cycle indicator, proved the fundamental Theorem 7.7, and gave
numerous applications. For an English translation of Polyas paper, see [64].
Much of Polyas work was anticipated by Redeld. For interesting histori-
cal information about the work of Redeld and its relation to Polya theory,
see [37][39][51][71] (all in the same issue of Journal of Graph Theory). The
Wikipedia article John Howard Redeld also gives information and refer-
ences on the interesting story of the rediscovery and signicance of Redelds
work.
The application of Polyas theorem to the enumeration of nonisomorphic
graphs appears in Polyas original paper [63]. For much additional work on
graphical enumeration, see the text of Harary and Palmer [38].
Subsequent to Polyas work there have been a huge number of expositions,
applications, and generalizations of Polya theory. We mention here only the
nice survey [12] by N. G. de Bruijn.
Theorem 7.14 (the generating function for the cycle indicator Z
S
of the
symmetric group S
) goes back to Frobenius (see [32, bottom of p. 152 of

GA]) and Hurwitz [44, 4]. It is clear that they were aware of Theorem 7.14,
even if they did not state it explicitly.
Chapter 8
A glimpse of Young tableaux.
We dened in Section 6 Youngs lattice Y , the poset of all partitions of all
nonnegative integers, ordered by containment of their Young diagrams.
21
1
2
3
4
5
11
111
1111
11111
2111
221
311
32
41
22 31 211
Here we will be concerned with the counting of certain walks in the Hasse
diagram (considered as a graph) of Y . Note that since Y is innite, we cannot
talk about its eigenvalues and eigenvectors. We need dierent techniques for
counting walks. (It will be convenient to denote the length of a walk by n,
rather than by as in previous sections.)
Note that Y is a graded poset (of innite rank), with Y
i
consisting of all
partitions of i. In other words, we have Y = Y
0
Y
1
(disjoint union),
111
112 CHAPTER 8. A GLIMPSE OF YOUNG TABLEAUX.
where every maximal chain intersects each level Y
i
exactly once. We call Y
i
the ith level of Y .
Since the Hasse diagram of Y is a simple graph (no loops or multiple
edges), a walk of length n is specied by a sequence
0
,
1
, . . . ,
n
of vertices
of Y . We will call a walk in the Hasse diagram of a poset a Hasse walk.
Each
i
is a partition of some integer, and we have either (a)
i
<
i+1
and
[
i
[ = [
i+1
[ 1, or (b)
i
>
i+1
and [
i
[ = [
i+1
[ + 1. A step of type
(a) is denoted by U (for up, since we move up in the Hasse diagram),
while a step of type (b) is denoted by D (for down). If the walk W has
steps of types A
1
, A
2
, . . . , A
n
, respectively, where each A
i
is either U or D,
then we say that W is of type A
n
A
n1
A
2
A
1
. Note that the type of a
walk is written in the opposite order to that of the walk. This is because
we will soon regard U and D as linear transformations, and we multiply
linear transformations right-to-left (opposite to the usual left-to-right reading
order). For instance (abbreviating a partition (
1
, . . . ,
m
) as
1

m
), the
walk , 1, 2, 1, 11, 111, 211, 221, 22, 21, 31, 41 is of type UUDDUUUUDUU =
U
2
D
2
U
4
DU
2
.
There is a nice combinatorial interpretation of walks of type U
n
which
begin at . Such walks are of course just saturated chains =
0
<
1
<
<
n
. In other words, they may be regarded as sequences of Young
diagrams, beginning with the empty diagram and adding one new square at
each step. An example of a walk of type U
5
is given by
.
We can specify this walk by taking the nal diagram and inserting an i into
square s if s was added at the ith step. Thus the above walk is encoded by
the tableau
2 1
3 5
4
.
Such an object is called a standard Young tableaux (or SYT). It consists
113
of the Young diagram D of some partition of an integer n, together with
the numbers 1, 2, . . . , n inserted into the squares of D, so that each number
appears exactly once, and every row and column is increasing. We call the
shape of the SYT , denoted = sh(). For instance, there are ve SYT of
shape (2, 2, 1), given by
1 1 1 1 1 2
3 5
5
2
4 3
4
5
2
3 4
5 4
3
2 4
3
5 2
.
Let f
denote the number of SYT of shape , so for instance f

(2,2,1)
= 5.
The numbers f
have many interesting properties; for instance, there is a

famous explicit formula for them known as the Frame-Robinson-Thrall hook
formula. For the sake of completeness we state this formula without proof,
though it is not needed in what follows.
Let u be a square of the Young diagram of the partition . Dene the
hook H(u) of u (or at u) to be the set of all squares directly to the right of u
or directly below u, including u itself. The size (number of squares) of H(u)
is called the hook length of u (or at u), denoted h(u). In the diagram of the
partition (4, 2, 2) below, we have inserted the hook length h(u) inside each
square u.
5
3
2
6 2 1
2
1
8.1 Theorem (hook length formula). Let n. Then
f
=
n!
u
h(u)
.
Here the notation u means that u ranges over all squares of the Young
diagram of .
For instance, the diagram of the hook lengths of = (4, 2, 2) above gives
f
=
8!
6 5 2 1 3 2 2 1
= 56.
In this section we will be concerned with the connection between SYT
and counting walks in Youngs lattice. If w = A
n
A
n1
A
1
is some word
in U and D and n, then let us write (w, ) for the number of Hasse
walks in Y of type w which start at the empty partition and end at . For
instance, (UDUU, 11) = 2, the corresponding walks being , 1, 2, 1, 11 and
, 1, 11, 1, 11. Thus in particular (U
n
, ) = f
[why?]. In a similar fashion,

since the number of Hasse walks of type D
n
U
n
which begin at , go up to a
partition n, and then back down to is given by (f
)
2
, we have
(D
n
U
n
, ) =
n
(f
)
2
. (8.1)
Our object is to nd an explicit formula for (w, ) of the form f
c
w
,
where c
w
does not depend on . (It is by no means a priori obvious that
such a formula should exist.) In particular, since f
= 1, we will obtain by
setting = a simple formula for the number of (closed) Hasse walks of
type w from to (thus including a simple formula for (8.1)).
There is an easy condition for the existence of any Hasse walks of type
w from to , given by the next lemma.
8.2 Lemma. Suppose w = D
s
k
U
r
k
D
s
2
U
r
2
D
s
1
U
r
1
, where r
i
0 and s
i

0. Let n. Then there exists a Hasse walk of type w from to if and
only if:
k
i=1
(r
i
s
i
) = n
j
i=1
(r
i
s
i
) 0 for 1 j k.
Proof. Since each U moves up one level and each D moves down one level,
we see that
k
i=1
(r
i
s
i
) is the level at which a walk of type w beginning at
ends. Hence
k
i=1
(r
i
s
i
) = [[ = n.
After
j
i=1
(r
i
+s
i
) steps we will be at level
j
i=1
(r
i
s
i
). Since the lowest
level is level 0, we must have
j
i=1
(r
i
s
i
) 0 for 1 j k.
The easy proof that the two conditions of the lemma are sucient for
the existence of a Hasse walk of type w from to is left to the reader.
If w is a word in U and D satisfying the conditions of Lemma 8.2, then
we say that w is a valid -word. (Note that the condition of being a valid
-word depends only on [[.)
115
The proof of our formula for (w, ) will be based on linear transforma-
tions analogous to those dened by (4.2) and (4.3). As in Section 4 let RY
j
be the real vector space with basis Y
j
. Dene two linear transformations
U
i
: RY
i
RY
i+1
and D
i
: RY
i
RY
i1
by
U
i
() =
i+1
<
D
i
() =
i1
<
,
for all i. For instance (using abbreviated notation for partitions)
U
21
(54422211) = 64422211 + 55422211 + 54432211 + 54422221 + 544222111
D
21
(54422211) = 44422211 + 54322211 + 54422111 + 5442221.
It is clear [why?] that if r is the number of distinct (i.e., unequal) parts of ,
then U
i
() is a sum of r + 1 terms and D
i
() is a sum of r terms. The next
lemma is an analogue for Y of the corresponding result for B
n
(Lemma 4.6).
8.3 Lemma. For any i 0 we have
D
i+1
U
i
U
i1
D
i
= I
i
, (8.2)
the identity linear transformation on RY
i
.
Proof. Apply the left-hand side of (8.2) to a partition of i, expand in terms
of the basis Y
i
, and consider the coecient of a partition . If ,= and
can be obtained from by adding one square s to (the Young diagram of)
and then removing a (necessarily dierent) square t, then there is exactly
one choice of s and t. Hence the coecient of in D
i+1
U
i
() is equal to 1.
But then there is exactly one way to remove a square from and then add
a square to get , namely, remove t and add s. Hence the coecient of in
U
i1
D
i
() is also 1, so the coecient of when the left-hand side of (8.2) is
applied to is 0.
If now ,= and we cannot obtain by adding a square and then deleting
a square from (i.e., and dier in more than two rows), then clearly
when we apply the left-hand side of (8.2) to , the coecient of will be 0.
Finally consider the case = . Let r be the number of distinct (unequal)
parts of . Then the coecient of in D
i+1
U
i
() is r+1, while the coecient
of in U
i1
D
i
() is r, since there are r + 1 ways to add a square to and
then remove it, while there are r ways to remove a square and then add it
back in. Hence when we apply the left-hand side of (8.2) to , the coecient
of is equal to 1.
Combining the conclusions of the three cases just considered shows that
the left-hand side of (8.2) is just I
i
, as was to be proved.
We come to one of the main results of this section.
8.4 Theorem. Let be a partition and w = A
n
A
n1
A
1
a valid -word.
Let S
w
= i : A
i
= D. For each i S
w
, let a
i
be the number of Ds in w to
the right of A
i
, and let b
i
be the number of Us in w to the right of A
i
. Thus
a
i
b
i
is the level we occupy in Y before taking the step A
i
= D. Then
(w, ) = f
iSw
(b
i
a
i
).
Before proving Theorem 8.4, let us give an example. Suppose w =
U
3
D
2
U
2
DU
3
= UUUDDUUDUUU and = (2, 2, 1). Then S
w
= 4, 7, 8
and a
4
= 0, b
4
= 3, a
7
= 1, b
7
= 5, a
8
= 2, b
8
= 5. We have also seen earlier
that f
221
= 5. Thus
(w, ) = 5(3 0)(5 1)(5 2) = 180.
Proof. Proof of Theorem 8.4. For notational simplicity we will omit the
subscripts from the linear transformations U
i
and D
i
. This should cause no
confusion since the subscripts will be uniquely determined by the elements
on which U and D act. For instance, the expression UDUU() where i
must mean U
i+1
D
i+2
U
i+1
U
i
(); otherwise it would be undened since U
j
and
D
j
can only act on elements of RY
j
, and moreover U
j
raises the level by one
while D
j
lowers it by one.
By (8.2) we can replace DU in any word y in the letters U and D by
UD + I. This replaces y by a sum of two words, one with one fewer D and
the other with one D moved one space to the right. For instance, replacing
the rst DU in UUDUDDU by UD +I yields UUUDDDU +UUDDU. If
we begin with the word w and iterate this procedure, replacing a DU in any
word with UD+I, eventually there will be no Us to the right of any Ds and
the procedure will come to an end. At this point we will have expressed w
117
as a linear combination (with integer coecients) of words of the form U
i
D
j
.
Since the operation of replacing DU with UD + I preserves the dierence
between the number of Us and Ds in each word, all the words U
i
D
j
which
appear will have i j equal to some constant n (namely, the number of Us
minus the number of Ds in w). Specically, say we have
w =
ij=n
r
ij
(w)U
i
D
j
, (8.3)
where each r
ij
(w) Z. (We also dene r
ij
(w) = 0 if i < 0 or j < 0.) We
claim that the r
ij
(w)s are uniquely determined by w. Equivalently [why?],
if we have
ij=n
d
ij
U
i
D
j
= 0 (8.4)
(as an identity of linear transformations acting on the space RY
k
for any k),
where each d
ij
Z (or d
ij
R, if you prefer), then each d
ij
= 0. Let j
be the least integer for which d

j
+n,j
,= 0. Let j
, and apply both sides

of (8.4) to . The left-hand side has exactly one nonzero term, namely, the
term with j = j
[why?]. The right-hand side, on the other hand

1
, is 0, a
contradiction. Thus the r
ij
(w)s are unique.
Now apply U on the left to (8.3). We get
Uw =
i,j
r
ij
(w)U
i+1
D
j
.
Hence (using uniqueness of the r
ij
s) there follows [why?]
r
ij
(Uw) = r
i1,j
(w). (8.5)
We next want to apply D on the left to (8.3). It is easily proved by
induction on i (left as an exercise) that
DU
i
= U
i
D +iU
i1
. (8.6)
(We interpret U
1
as being 0, so that (8.6) is true for i = 0.) Hence
Dw =
i,j
r
ij
(w)DU
i
D
j
=
i,j
r
ij
(w)(U
i
D +iU
i1
)D
j
,
1
The phrase the right-hand side, on the other hand does not mean the left-hand side!
from which it follows [why?] that
r
ij
(Dw) = r
i,j1
(w) + (i + 1)r
i+1,j
(w). (8.7)
Setting j = 0 in (8.5) and (8.7) yields
r
i0
(Uw) = r
i1,0
(w) (8.8)
r
i0
(Dw) = (i + 1)r
i+1,0
(w). (8.9)
Now let (8.3) operate on . Since D
j
() = 0 for all j > 0, we get w() =
r
n0
(w)U
n
(). Thus the coecient of in w() is given by
(w, ) = r
n0
(w)(U
n
, ) = r
n0
f
,
where as usual n. It is easy to see from (8.8) and (8.9) that
r
n0
(w) =
jSw
(b
j
a
j
),
and the proof follows.
Note. It is possible to give a simpler proof of Theorem 8.4, but the proof
we have given is useful for generalizations not appearing here.
An interesting special case of the previous theorem allows us to evaluate
equation (8.1).
(D
n
U
n
, ) =
n
(f
)
2
= n!.
Proof. When w = D
n
U
n
in Theorem 8.4 we have S
w
= n + 1, n +
2, . . . , 2n, a
i
= n i + 1, and b
i
= n, from which the proof is immediate.
Note (for those familiar with the representation theory of nite groups).
It can be shown that the numbers f
, for n, are the degrees of the

irreducible representations of the symmetric group S
n
. Given this, Corol-
lary 8.5 is a special case of the result that the sum of the squares of the
degrees of the irreducible representations of a nite group G is equal to the
order #G of G. There are many other intimate connections between the rep-
resentation theory of S
n
, on the one hand, and the combinatorics of Youngs
119
lattice and Young tableaux, on the other. There is also an elegant combina-
torial proof of Corollary 8.5, based on the RSK algorithm (after Gilbert de
Beauregard Robinson, Craige Schensted, and Donald Knuth) or Robinson-
Schensted correspondence, with many fascinating properties and with deep
connections with representation theory. In the Appendix to this section we
give the denition of the RSK algorithm.
We now consider a variation of Theorem 8.4 in which we are not concerned
with the type w of a Hasse walk from to w, but only with the number of
steps. For instance, there are three Hasse walks of length three from to the
partition 1, given by , 1, , 1; , 1, 2, 1; and , 1, 11, 1. Let (, ) denote the
number of Hasse walks of length from to . Note the two following easy
facts:
(F1) (, ) = 0 unless [[ (mod2).
(F2) (, ) is the coecient of in the expansion of (D + U)
() as a
linear combination of partitions.
Because of (F2) it is important to write (D+U)
as a linear combination
of terms U
i
D
j
, just as in the proof of Theorem 8.4 we wrote a word w in U
and D in this form. Thus dene integers b
ij
() by
(D +U)
i,j
b
ij
()U
i
D
j
. (8.10)
Just as in the proof of Theorem 8.4, the numbers b
ij
() exist and are well-
dened.
8.6 Lemma. We have b
ij
() = 0 if i j is odd. If i j = 2m then
b
ij
() =
!
2
m
i! j! m!
. (8.11)
Proof. The assertion for i j odd is equivalent to (F1) above, so
assume i j is even. The proof is by induction on . Its easy to check
that (8.11) holds for = 1. Now assume true for some xed 1. Using
(8.10) we obtain
i,j
b
ij
( + 1)U
i
D
j
= (D +U)
+1
= (D +U)
i,j
b
ij
()U
i
D
j
=
i,j
b
ij
()(DU
i
D
j
+U
i+1
D
j
).
In the proof of Theorem 8.4 we saw that DU
i
= U
i
D + iU
i1
(see equation
(8.6)). Hence we get
i,j
b
ij
( + 1)U
i
D
j
=
i,j
b
ij
()(U
i
D
j+1
+iU
i1
D
j
+U
i+1
D
j
). (8.12)
As mentioned after (8.10), the expansion of (D + U)
+1
in terms of U
i
D
j
is
unique. Hence equating coecients of U
i
D
j
on both sides of (8.12) yields
the recurrence
b
ij
( + 1) = b
i,j1
() + (i + 1)b
i+1,j
() +b
i1,j
(). (8.13)
It is a routine matter to check that the function !/2
m
i!j!m! satises the same
recurrence (8.13) as b
ij
(), with the same intial condition b
00
(0) = 1. From
this the proof follows by induction.
From Lemma 8.6 it is easy to prove the following result.
8.7 Theorem. Let n and n, with n even. Then
(, ) =
_
n
_
(1 3 5 ( n 1))f
.
Proof. Apply both sides of (8.10) to . Since U
i
D
j
() = 0 unless j = 0,
we get
(D +U)
() =
i
b
i0
()U
i
()
=
i
b
i0
()
i
f
.
Since by Lemma 8.6 we have b
i0
() =
_
i
_
(1 3 5 ( i 1)) when i is
even, the proof follows from (F2).
Note. The proof of Theorem 8.7 only required knowing the value of
b
i0
(). However, in Lemma 8.6 we computed b
ij
() for all j. We could have
carried out the proof so as only to compute b
i0
(), but the general value of
b
ij
() is so simple that we have included it too.
8.8 Corollary. The total number of Hasse walks in Y of length 2m from
to is given by
(2m, ) = 1 3 5 (2m1).
121
Proof. Simply substitute = (so n = 0) and = 2m in Theorem 8.7.
The fact that we can count various kinds of Hasse walks in Y suggests
that there may be some nite graphs related to Y whose eigenvalues we can
also compute. This is indeed the case, and we will discuss the simplest case
here. Let Y
j1,j
denote the restriction of Youngs lattice Y to ranks j 1 and
j. Identify Y
j1,j
with its Hasse diagram, regarded as a (bipartite) graph.
Let p(i) = #Y
i
, the number of partitions of i. (The function p(i) has been
extensively studied, beginning with Euler, though we will not discuss its
fascinating properties here.)
8.9 Theorem. The eigenvalues of Y
j1,j
are given as follows: 0 is an eigen-
value of multiplicity p(j) p(j 1); and for 1 s j, the numbers
s
are eigenvalues of multiplicity p(j s) p(j s 1).
Proof. Let A denote the adjacency matrix of Y
j1,j
. Since RY
j1,j
=
RY
j1
RY
j
(vector space direct sum), any vector v RY
j1,j
can be written
uniquely as v = v
j1
+v
j
, where v
i
RY
i
. The matrix A acts on the vector
space RY
j1,j
as follows [why?]:
A(v) = D(v
j
) +U(v
j1
). (8.14)
Just as Theorem 4.7 followed from Lemma 4.6, we deduce from Lemma 8.3
that for any i we have that U
i
: RY
i
RY
i+1
is one-to-one and D
i
: RY
i

RY
i1
is onto. It follows in particular that
dim(ker(D
i
)) = dimRY
i
dimRY
i1
= p(i) p(i 1),
where ker denotes kernel.
Case 1. Let v ker(D
j
), so v = v
j
. Then Av = Dv = 0. Thus ker(D
j
) is
an eigenspace of A for the eigenvalue 0, so 0 is an eigenvalue of multiplicity
at least p(j) p(j 1).
Case 2. Let v ker(D
s
) for some 0 s j 1. Let
v
=
_
j sU
j1s
(v) +U
js
(v).
Note that v
RY
j1,j
, with v
j1
=
j sU
j1s
(v) and v
j
= U
js
(v).
Using equation (8.6), we compute
A(v
) = U(v
j1
) +D(v
j
)
=
_
j sU
js
(v) +DU
js
(v)
=
_
j sU
js
(v) +U
js
D(v) + (j s)U
js1
(v)
=
_
j sU
js
(v) + (j s)U
js1
(v)
=
_
_
j s
_
v
. (8.15)
Its easy to verify (using the fact that U is one-to-one) that if v(1), . . . , v(t)
is a basis for ker(D
s
), then v(1)
, . . . , v(t)
are linearly independent. Hence

by (8.15) we have that
j s is an eigenvalue of A of multiplicity at least

t = dim(ker(D
s
)) = p(s) p(s 1).
We have found a total of
p(j) p(j 1) + 2
j1
s=0
(p(s) p(s 1)) = p(j 1) +p(j)
eigenvalues of A. (The factor 2 above arises from the fact that both +
j s
and
j s are eigenvalues.) Since the graph Y

j1,j
has p(j 1) + p(j)
vertices, we have found all its eigenvalues.
An elegant combinatorial consequence of Theorem 8.9 is the following.
8.10 Corollary. Fix j 1. The number of ways to choose a partition of
j, then delete a square from (keeping it a partition), then insert a square,
then delete a square, etc., for a total of m insertions and m deletions, ending
back at , is given by
j
s=1
[p(j s) p(j s 1)]s
m
, m > 0. (8.16)
Proof. Exactly half the closed walks in Y
j1,j
of length 2m begin at
an element of Y
j
[why?]. Hence if Y
j1,j
has eigenvalues
1
, . . . ,
r
, then by
Corollary 1.3 the desired number of walks is given by
1
2
(
2m
1
+ +
2m
r
).
Using the values of
1
, . . . ,
r
given by Theorem 8.9 yields (8.16).
For instance, when j = 7, equation (8.16) becomes 4 + 2 2
m
+ 2 3
m
+
4
m
+5
m
+7
m
. When m = 1 we get 30, the number of edges of the graph Y
6,7
[why?].
123
APPENDIX 1: THE RSK ALGORITHM
We will describe a bijection between permutations S
n
and pairs
(P, Q) of SYT of the same shape n. Dene a near Young tableau (NYT)
to be the same as an SYT, except that the entries can be any distinct integers,
not necessarily the integers 1, 2, . . . , n. Let P
ij
denote the entry in row i and
column j of P. The basic operation of the RSK algorithm consists of the
row insertion P k of a positive integer k into an NYT P = (P
ij
). The
operation P k is dened as follows: Let r be the least integer such that
P
1r
> k. If no such r exists (i.e., all elements of the rst row of P are
less than r), then simply place k at the end of the rst row. The insertion
process stops, and the resulting NYT is P k. If, on the other hand, r does
exist then replace P
1r
by k. The element k then bumps P
1r
:= k
into the
second row, i.e., insert k
into the second row of P by the insertion rule just

described. Either k
is inserted at the end of the second row, or else it bumps

an element k to the third row. Continue until an element is inserted at the
end of a row (possibly as the rst element of a new row). The resulting array
is P k.
8.11 Example. Let
P =
3 7 9 14
6 11 12
10 16
13
15
Then P 8 is shown below, with the elements inserted into each row (either
by bumping or by the nal insertion in the fourth row) in boldface. Thus
the 8 bumps the 9, the 9 bumps the 11, the 11 bumps the 16, and the 16 is
inserted at the end of a row. Hence
(P 8) =
3 7 8 14
6 9 12
10 11
13 16
15
.
We omit the proof, which is fairly straightforward, that if P is an NYT,
then so is P k. We can now describe the RSK algorithm. Let =
a
1
a
2
a
n
S
n
. We will inductively construct a sequence (P
0
, Q
0
), (P
1
, Q
1
),
. . . , (P
n
, Q
n
) of pairs (P
i
, Q
i
) of NYT of the same shape, where P
i
and Q
i
each have i squares. First, dene (P
0
, Q
0
) = (, ). If (P
i1
, Q
i1
) have been
dened, then set P
i
= P
i1
a
i
. In other words, P
i
is obtained from P
i1
by row inserting a
i
. Now dene Q
i
to be the NYT obtained from Q
i1
by
inserting i so that Q
i
and P
i
have the same shape. (The entries of Q
i1
dont change; we are simply placing i into a certain new square and not row-
inserting it into Q
i1
.) Finally let (P, Q) = (P
n
, Q
n
). We write
RSK
(P, Q).
8.12 Example. Let = 4273615 S
7
. The pairs (P
1
, Q
1
), . . . , (P
7
, Q
7
) =
(P, Q) are as follows:
P
i
Q
i
4 1
2 1
4 2
2 7 1 3
4 2
2 3 1 3
4 7 2 4
2 3 6 1 3 5
4 7 2 4
1 3 6 1 3 5
2 7 2 4
4 6
1 3 5 1 3 5
2 6 2 4
4 7 6 7
8.13 Theorem. The RSK algorithm denes a bijection between the symmet-
ric group S
n
and the set of all pairs (P, Q) of SYT of the same shape, where
the shape is a partition of n.
Proof (sketch). The key step is to dene the inverse of RSK. In other
words, if (P, Q), then how can we recover uniquely from (P, Q)? More-
over, we need to nd for any (P, Q). Observe that the position occupied by
125
n in Q is the last position to be occupied in the insertion process. Suppose
that k occupies this position in P. It was bumped into this position by some
element j in the row above k that is currently the largest element of its row
less than k. Hence we can inverse bump k into the position occupied by j,
and now inverse bump j into the row above it by the same procedure. Even-
tually an element will be placed in the rst row, inverse bumping another
element t out of the tableau altogether. Thus t was the last element of
to be inserted, i.e., if = a
1
a
2
a
n
then a
n
= t. Now locate the position
occupied by n1 in Q and repeat the procedure, obtaining a
n1
. Continuing
in this way, we uniquely construct one element at a time from right-to-left,
such that (P, Q).
Thus we have obtained a bijective proof of Corollary 8.5, that is,
n
(f
)
2
= n!.
APPENDIX 2: PLANE PARTITIONS
In this appendix we show how a generalization of the RSK algorithm leads
to an elegant generating function for a two-dimensional generalization of
integer partitions. A plane partition of an integer n 0 is a two-dimensional
array = (
ij
)
,j1
of integers
ij
0 that is weakly decreasing in rows and
columns, i.e.,
ij

i+1,j
,
ij

i,j+1
,
such that
i,j

i,j
= n. It follows that all but nitely many
ij
are 0, and
these 0s are omitted in writing a particular plane partition . Given a plane
partition , we write [[ = n to denote that is a plane partition of n. More
generally, if L is any array of nonnegative integers we write [L[ for the sum
of the parts (entries) of L.
There is one plane partition of 0, namely, all
ij
= 0, denoted . The
plane partitions of the integers 0 n 3 are given by
1 2 11 1 3 21 111 11 2 1
1 1 1 1
1
.
If pp(n) denotes the number of plane partitions of n, then pp(0) = 1, pp(1) =
1, pp(2) = 3, and pp(3) = 6.
Our object is to give a formula for the generating function
F(x) =
n0
pp(n)x
n
= 1 +x + 3x
2
+ 6x
3
+ 13x
4
+ 24x
5
+ .
More generally, we will consider plane partitons with at most r rows and at
most s columns, i.e.,
ij
= 0 for i > r or j > s. As a simple warmup, let us
rst consider the case of ordinary partitions = (
1
,
2
, . . . ) of n.
8.14 Proposition. Let p
s
(n) denote the number of partitions of n with at
most s parts. Equivalently, p
s
(n) is the number of plane partitions of n with
at most one row and at most s columns [why?].Then
n0
p
s
(n)x
n
=
s
k=1
(1 x
k
)
1
.
127
Proof. First note that the partition has at most s parts if and only if the
conjugate partition
dened in Chapter 6 has largest part at most s. Thus

it suces to nd the generating function
n0
p
s
(n)x
n
, where p
s
(n) denotes
the number of partitions of n whose largest part is at most s. Now expanding
each factor (1 x
k
)
1
as a geometric series gives
s
k=1
1
1 x
k
=
s
k=1
_
m
k
1
x
m
k
k
_
.
How do we get a coecient of x
n
? We must choose a term x
m
k
k
from each
factor of the product, 1 k s, so that
n =
s
k=1
m
k
k.
But such a choice is the same as choosing the partition of n such that the
part k occurs m
k
times. For instance, if s = 4 and we choose m
1
= 5, m
2
= 0,
m
3
= 1, m
4
= 2, then we have chosen the partition = (4, 4, 3, 1, 1, 1, 1, 1)
of 16. Hence the coecient of x
n
is the number of partitions of n whose
largest part is at most s, as was to be proved.
Note that Proposition 8.14 is trivial in the sense that it can be seen
by inspection. There is an obvious correspondence between (a) the choice of
terms contributing to the coecient of x
n
and (b) partitions of n with largest
part at most r. Although the generating function we will obtain for plane
partitions is equally simple, it will be far less obvious why it is correct.
Plane partitions have a certain similarity with standard Young tableaux,
so perhaps it is not surprising that a variant of RSK will be applicable. In-
stead of NYT we will be dealing with column-strict plane partitions (CSPP).
These are plane partitions for which the nonzero elements strictly decrease
in each column. An example of a CSPP is given by
7 7 4 3 3 3 1
4 3 3 1
3 2
2 1
1
. (8.17)
We say that this CSPP has shape = (7, 4, 2, 2, 1), the shape of the Young
diagram which the numbers occupy, and that it has ve rows, seven columns,
and 16 parts (so 16).
If P = (P
ij
) is a CSPP and k 1, then we dene the row insertion P k
as follows: Let r be the least integer such that P
1,r
< k. If no such r exists
(i.e., all elements of the rst row of P are greater than r), then simply place
k at the end of the rst row. The insertion process stops, and the resulting
CSPP is P k. If, on the other hand, r does exist, then replace P
1r
by k.
The element k then bumps P
1r
:= k
into the second row, i.e., insert k
into
the second row of P by the insertion rule just described, possibly bumping a
new element k
into the third row. Continue until an element is inserted at

the end of a row (possibly as the rst element of a new row). The resulting
array is P k. Note that this rule is completely analogous to row insertion
for NYT: for NYT an element bumps the leftmost element greater than it,
while for CSPP an element bumps the leftmost element smaller than it.
8.15 Example. Let P be the CSPP of equation (8.17). Let us row insert 6
into P. The set of elements which get bumped are shown in bold:
7 7 4 3 3 3 1
4 3 3 1
3 2
2 1
1
.
The nal 1 that was bumped is inserted at the end of the fth row. Thus we
obtain
(P 5) =
7 7 6 3 3 3 1
4 4 3 1
3 3
2 2
1 1
.
We are now ready to describe the analogue of RSK needed to count plane
partitions. Instead of beginning with a permutation S
n
, we begin with
an r s matrix A = (a
ij
) of nonnegative integers, called for short an r s
N-matrix. We convert A into a two-line array
w
A
=
_
u
1
u
2
u
N
v
1
v
2
v
N
_
,
where
u
1
u
2
u
N
129
If i < j and u
i
= u
j
, then v
i
v
j
.
The number of columns of w
A
equal to
i
j
is a
ij
. (It follows that N =
a
ij
.)
It is easy to see that w
A
is uniquely determined by A, and conversely. As an
example, suppose that
A =
_
_
1 0 0 2
1 1 1 0
2 1 0 0
_
_
. (8.18)
Then
w
A
=
_
3 3 3 2 2 2 1 1 1
2 1 1 3 2 1 4 4 2
_
.
We now insert the number v
1
, v
2
, . . . , v
N
successively into a CSPP. That
is, we start with P
0
= and dene inductively P
i
= P
i1
v
i
. We also start
with Q
0
= , and at the ith step insert u
i
into Q
i1
(without any bumping
or other altering of the elements of Q
i1
) so that P
i
and Q
i
have the same
shape. Finally let (P, Q) = (P
N
, Q
N
) and write A
RSK
(P, Q).
8.16 Example. Let A be given by equation (8.18). The pairs (P
1
, Q
1
), . . . ,
(P
9
, Q
9
) = (P, Q) are as follows:
P
i
Q
i
2 3
2 1 3 3
2 1 1 3 3 3
3 1 1 3 3 3
2 2
3 2 1 3 3 3
2 1 2 2
3 2 1 1 3 3 3 2
2 1 2 2
4 2 1 1 3 3 3 2
3 1 2 2
2 1
4 4 1 1 3 3 3 2
3 2 2 2
2 1 1 1
4 4 2 1 3 3 3 2
3 2 1 2 2 1
2 1 1 1
It is straightforward to show that if A
RSK
(P, Q), then P and Q are

CSPP of the same shape. We omit the proof of the following key lemma,
which is analogous to the proof of Theorem 8.13. Let us just note a crucial
property (which is easy to prove) of the correspondence A
RSK
(P, Q) which
allows us to recover A from (P, Q), namely, equal entries of Q are inserted
from left-to-right. Thus the last number placed into Q is the rightmost
occurrence of the least entry. Hence we can can inverse bump the number in
this position in P to back up one step in the algorithm, just as for the usual
RSK correspondence
RSK
(P, Q).
131
8.17 Lemma. The correspondence A
RSK
(P, Q) is a bijection from the set

of r s matrices of nonnegative integers to the set of pairs (P, Q) of CSPP
of the same shape, such that the largest part of P is at most s and the largest
part of Q is at most r.
The next step is to convert the pair (P, Q) of CSPP of the same shape
into a single plane partition . We do this by merging the ith column
of P with the ith column of Q, producing the ith column of . Thus we
rst describe how to merge two partitions and with distinct parts and
with the same number of parts into a single partition = (, ). Draw the
Ferrers diagram of but with each row indented one space to the right of the
beginning of the previous row. Such a diagram is called the shifted Ferrers
diagram of . For instance, if = (5, 3, 2) then we get the shifted diagram
Do the same for , and then transpose the diagram. For instance, if
= (6, 3, 1) then we get the transposed shifted diagram
Now merge the two diagrams into a single diagram by identifying their main
diagonals. For and as above, we get the diagram (with the main diagonal
drawn for clarity):
Dene (, ) to be the partition for which this merged diagram is the Ferrers
diagram. The above example shows that
(532, 631) = 544211.
The map (, ) (, ) is clearly a bijection between pairs of partitions
(, ) with k distinct parts, and partitions whose main diagonal (of the
Ferrers diagram) has k dots. Equivalently, k is the largest integer j for which
j
j. Note that
[[ = [[ +[[ (). (8.19)
We now extend the above bijection to pairs (P, Q) of reverse SSYT of the
same shape. If
i
denotes the ith column of P and
i
the ith column of Q,
then let (P, Q) be the array whose ith column is (
i
,
i
). For instance, if
P =
4 4 2 1
3 1 1
2
and Q =
5 3 2 2
4 2 1
1
,
then
(P, Q) =
4 4 2 1
4 2 2 1
4 2
2
2
.
It is easy to see that (P, Q) is a plane partition. Replace each row of (P, Q)
by its conjugate to obtain another plane partition
(P, Q). With (P, Q) as

above we obtain
(P, Q) =
4 3 2 2
4 3 1 1
2 2 1 1
1 1
1 1
.
Write [P[ for the sum of the elements of P, and write max(P) for the largest
element of P, and similarly for Q. When we merge P and Q into (P, Q),
max(P) becomes the largest part of (P, Q). Thus when we conjugate each
row, max(P) becomes the number col(
(P, Q)) of columns of
(P, Q) [why?].
Similarly, max(Q) becomes the number row(
(P, Q)) of rows of (P, Q) and

of
(P, Q). In symbols,

max P = col(
(P, Q))
max Q = row(
(P, Q)).
(8.20)
Moreover, it follows from equation (8.19) that
[
(P, Q)[ = [(P, Q)[ = [P[ +[Q[ (P), (8.21)

133
where (P) denotes the number of parts of P (or of Q).
We now have all the ingredients necessary to prove the main result of this
appendix.
8.18 Theorem. Let pp
rs
(n) denote the number of plane partitions of n with
at most r rows and at most s columns. Then
n0
pp
rs
(n)x
n
=
r
i=1
s
j=1
(1 x
i+j1
)
1
.
Proof. Let A = (a
ij
) be an r s N-matrix. We can combine the bijections
discussed above to obtain a plane partition (A) associated with A. Namely,
rst apply RSK to obtain A
RSK
(P, Q), and then apply the merging process

and row conjugation to obtain (A) =
(P, Q). Since a column

i
j
of the
two-line array w
A
occurs a
ij
times and results in an insertion of j into P and
i into Q, it follows that
[P[ =
i,j
ja
ij
[Q[ =
i,j
ia
ij
max(P) = maxj : a
ij
,= 0
max(Q) = maxi : a
ij
,= 0
Hence from equations (8.20) and (8.21), we see that the map A (A) is
a bijection from r s N-matrices A to plane partitions with at most r rows
and at most s columns. Moreover,
[(A)[ = [P[ +[Q[ (P)
=
i,j
(i +j 1)a
ij
.
Thus the enumeration of plane partitions is reduced to the much easier enu-
meration of N-matrices. Specically, we have
n0
pp
rs
(n)x
n
=
row()r
col()s
x
||
=
rs Nmatrices A
x
P
(i+j1)a
ij
=
r
i=1
s
j=1
_
_
a
ij
0
x
P
(i+j1)a
ij
_
_
=
r
i=1
s
j=1
(1 x
i+j1
)
1
.
Write P
r
(n) for the number of plane partitions of n with at most r rows.
Letting c and then r in Theorem 8.18 produces the elegant
generating functions of the next corollary.
n0
pp
r
(n)x
n
=
i1
(1 x
i
)
min(i,r)
(8.22)
n0
pp(n)x
n
=
i1
(1 x
i
)
i
. (8.23)
Note. Once one has seen the generating function
1
(1 x)(1 x
2
)(1 x
3
)
for one-dimensional (ordinary) partitions and the generating function
1
(1 x)(1 x
2
)
2
(1 x
3
)
3
. . .
for two-dimensional (plane) partitions, it is quite natural to ask about higher-
dimensional partitions. In particular, a solid partition of n is a three-dimensional
array = (
ijk
)
i,j,k1
of nonnegative integers, weakly decreasing in each of
135
the three coordinate directions, and with elements summing to n. Let sol(n)
denote the number of solid partitions of n. It is easy to see that for any
integer sequence a
0
= 1, a
1
, a
2
, . . . , there are unique integers b
1
, b
2
, . . . for
which
n0
a
n
x
n
=
i1
(1 x
i
)
b
i
.
For the case a
n
= sol(n), we have
b
1
= 1, b
2
= 3, b
3
= 6, b
4
= 10, b
5
= 15,
which looks quite promising. Alas, the sequence of exponents continues
20, 26, 34, 46, 68, 97, 120, 112, 23, 186, 496, 735, 531, 779, . . . .
The problem of enumerating solid partitions remains open and is considered
most likely to be hopeless.
Standard Young tableaux (SYT) were rst enumerated by P. A. MacMa-
hon [55, p. 175] (see also [56, 103]). MacMahon formulated his result in
terms of generalized ballot sequences or lattice permutations rather than
SYT, but they are easily seen to be equivalent. He stated the result not in
terms of the products of hook lengths as in Theorem 8.1, but as a more com-
plicated product formula. The formulation in terms of hook lengths is due to
J. S. Frame and appears rst in the paper [29, Thm. 1] of Frame, Robinson,
and R. M. Thrall; hence it is sometimes called the Frame-Robinson-Thrall
hook-length formula. (The actual denition of standard Young tableaux is
due to A. Young [94, p. 258].)
Independently of MacMahon, F. G. Frobenius [32, eqn. (6)] obtained the
same formula for the degree of the irreducible character
of S
n
as MacMa-
hon obtained for the number of lattice permutations of type . Frobenius was
apparently unaware of the combinatorial signicance of deg
, but Young
showed in [94, pp. 260261] that deg
was the number of SYT of shape

, thereby giving an independent proof of MacMahons result. (Young also
provided his own proof of MacMahons result in [94, Thm. II].)
A number of other proofs of the hook-length formula were subsequently
found. C. Greene, A. Nijenhuis, and H. S. Wilf [36] gave an elegant proba-
bilistic proof. A proof of A. Hillman and R. Grassl [42] shows very clearly the
role of hook lengths, though the proof is not completely bijective. A bijec-
tive version was later given by C. F. Krattenthaler [46]. Completely bijective
proofs of the hook-length formula were rst given by D. S. Franzblau and D.
Zeilberger [30] and by J. B. Remmel [72]. An exceptionally elegant bijective
proof was later found by J.-C. Novelli, I. M. Pak, and A. V. Stoyanovskii
[60].
The use of the operators U and D to count walks in the Hasse diagram
of Youngs lattice was developed independently, in a more general context,
by S. Fomin [27][28] and R. Stanley [82][84]. See also [85, 3.21] for a short
exposition.
The RSK algorithm (known by a variety of other names, either cor-
respondence or algorithm in connection with some subset of the names
Robinson, Schensted, and Knuth) was rst described, in a rather vague form,
by G. de B. Robinson [73, 5], as a tool in an attempted proof of a result
now known as the Littlewood-Richardson Rule. The RSK algorithm was
later rediscovered by C. E. Schensted (see below), but no one actually ana-
lyzed Robinsons work until this was done by M. van Leeuwen [50, 7]. It
is interesting to note that Robinson says in a footnote on page 754 that I
am indebted for this association I to Mr. D. E. Littlewood. Van Leeuwens
analysis makes it clear that association I gives the recording tableau Q of
the RSK algorithm
RSK
(P, Q). Thus it might be correct to say that if
S
n
and
RSK
(P, Q), then the denition of P is due to Robinson, while
the denition of Q is due to Littlewood.
No further work related to Robinsons construction was done until Schen-
sted published his seminal paper [75] in 1961. (For some information about
the unusual life of Schensted, see [5].) Schensteds purpose was the enumera-
tion of permutations in S
n
according to the length of their longest increasing
and decreasing subsequences. According to Knuth [47, p. 726], the connec-
tion between the work of Robinson and that of Schensted was rst pointed
out by M.-P. Sch utzenberger, though as mentioned above the rst person to
describe this connection precisely was van Leeuwen.
Plane partitions were discovered by MacMahon in a series of papers which
were not appreciated until much later. (See MacMahons book [56, Sections
IX and X] for an exposition of his results.) MacMahons rst paper dealing
with plane partitions was [54]. In Article 43 of this paper he gives the deni-
tion of a plane partition (though not yet with that name). In Article 51 he
137
conjectures that the generating function for plane partitions is the product
(1 x)
1
(1 x
2
)
2
(1 x
3
)
3
(1 x
4
)
4

(our equation (8.23)). In Article 52 he conjectures our equation (8.22) and
Theorem 8.18, nally culminating in a conjectured generating function for
plane partitions of n with at most r rows, at most s columns, and with largest
part at most t. (See Exercise 8.33.) MacMahon goes on in Articles 5662
to prove his conjecture in the case of plane partitions with at most 2 rows
and s columns (the case r = 2 of our Theorem 8.18), mentioning on page
662 that an independent solution was obtained by A. R. Forsyth. (Though a
publication reference is given to Forsyths paper, apparently it never actually
appeared.)
We will not attempt to describe MacMahons subsequent work on plane
partitions, except to say that the culmination of his work appears in [56,
Art. 495], in which he proves his main conjecture from his rst paper [54] on
plane partitions, viz., our Exercise 8.33. MacMahons proof is quite lengthy
and indirect.
In 1972 E. A. Bender and D. E. Knuth [6] showed the connection between
the theory of symmetric functions and the enumeration of plane partitions.
They gave simple proofs based on the RSK algorithm of many results involv-
ing plane partitions, including the rst bijective proof (the same proof that
we give) of our Theorem 8.18.
Chapter 9
The Matrix-Tree Theorem.
The Matrix-Tree Theorem is a formula for the number of spanning trees of
a graph in terms of the determinant of a certain matrix. We begin with the
necessary graph-theoretical background. Let G be a nite graph, allowing
multiple edges but not loops. (Loops could be allowed, but they turn out to
be completely irrelevant.) We say that G is connected if there exists a walk
between any two vertices of G. A cycle is a closed walk with no repeated
vertices or edges, except for the the rst and last vertex. A tree is a connected
graph with no cycles. In particular, a tree cannot have multiple edges, since
a double edge is equivalent to a cycle of length two. The three nonisomorphic
trees with ve vertices are given by:
r r r r r r r r
r
r
r
r
r
r
r
d
d
d
d
d
d
d
d
d
A basic theorem of graph theory (whose easy proof we leave as an exercise)
is the following.
9.1 Proposition. Let G be a graph with p vertices. The following conditions
are equivalent.
(a) G is a tree.
(b) G is connected and has p 1 edges.
(c) G is has no cycles and has p 1 edges.
(d) There is a unique path (= walk with no repeated vertices) between any
two vertices.
149
150 CHAPTER 9. THE MATRIX-TREE THEOREM.
A spanning subgraph of a graph G is a graph H with the same vertex set
as G, and such that every edge of H is an edge of G. If G has q edges, then the
number of spanning subgraphs of G is equal to 2
q
, since we can choose any
subset of the edges of G to be the set of edges of H. (Note that multiple edges
between the same two vertices are regarded as distinguishable, in accordance
with the denition of a graph in Section 1.) A spanning subgraph which is
a tree is called a spanning tree. Clearly G has a spanning tree if and only if
it is connected [why?]. An important invariant of a graph G is its number of
spanning trees, called the complexity of G and denoted (G).
9.2 Example. Let G be the graph illustrated below, with edges a, b, c, d,
e.
r r
r r
d
d
d
d
d
dd
d
a
e
c
b
Then G has eight spanning trees, namely, abc, abd, acd, bcd, abe, ace, bde, and
cde (where, e.g., abc denotes the spanning subgraph with edge set a, b, c).
9.3 Example. Let G = K
5
, the complete graph on ve vertices. A simple
counting argument shows that K
5
has 60 spanning trees isomorphic to the
rst tree in the above illustration of all nonisomorphic trees with ve vertices,
60 isomorphic to the second tree, and 5 isomorphic to the third tree. Hence
(K
5
) = 125. It is even easier to verify that (K
1
) = 1, (K
2
) = 1, (K
3
) =
3, and (K
4
) = 16. Can the reader make a conjecture about the value of
(K
p
) for any p 1?
Our object is to obtain a determinantal formula for (G). For this we
need an important result from matrix theory which is often omitted from a
beginning linear algebra course. (Later (Theorem 10.4) we will prove a more
general determinantal formula without the use of the Binet-Cauchy theorem.
However, the use of the Binet-Cauchy theorem does aord some additional
algebraic insight.) This result, known as the Binet-Cauchy theorem (or some-
times as the Cauchy-Binet theorem), is a generalization of the familiar fact
that if A and B are n n matrices, then det(AB) = det(A) det(B) (where
151
det denotes determinant). We want to extend this formula to the case where
A and B are rectangular matrices whose product is a square matrix (so that
det(AB) is dened). In other words, A will be an m n matrix and B an
n m matrix, for some m, n 1.
We will use the following notation involving submatrices. Suppose A =
(a
ij
) is an m n matrix, with 1 i m, 1 j n, and m n. Given an
m-element subset S of 1, 2, . . . , n, let A[S] denote the mm submatrix of A
obtained by taking the columns indexed by the elements of S. In other words,
if the elements of S are given by j
1
< j
2
< < j
m
, then A[S] = (a
i,j
k
),
where 1 i m and 1 k m. For instance, if
A =
_
_
1 2 3 4 5
6 7 8 9 10
11 12 13 14 15
_
_
and S = 2, 3, 5, then
A[S] =
_
_
2 3 5
7 8 10
12 13 15
_
_
.
Similarly, let B = (b
ij
) be an n m matrix with 1 i n, 1 j m and
m n. Let S be an m-element subset of 1, 2, . . . , n as above. Then B[S]
denotes the m m matrix obtained by taking the rows of B indexed by S.
Note that A
t
[S] = A[S]
t
, where
t
denotes transpose.
9.4 Theorem (the Binet-Cauchy Theorem). Let A = (a
ij
) be an m n
matrix, with 1 i m and 1 j n. Let B = (b
ij
) be an n m matrix
with 1 i n and 1 j m. (Thus AB is an m m matrix.) If m > n,
then det(AB) = 0. If m n, then
det(AB) =
S
(det A[S])(det B[S]),
where S ranges over all m-element subsets of 1, 2, . . . , n.
Before proceeding to the proof, let us give an example. We write [a
ij
[ for
the determinant of the matrix (a
ij
). Suppose
A =
_
a
1
a
2
a
3
b
1
b
2
b
3
_
, B =
_
_
c
1
d
1
c
2
d
2
c
3
d
3
_
_
.
Then
det(AB) =
a
1
a
2
b
1
b
2
c
1
d
1
c
2
d
2
a
1
a
3
b
1
b
3
c
1
d
1
c
3
d
3
a
2
a
3
b
2
b
3
c
2
d
2
c
3
d
3
.
Proof of Theorem 9.4 (sketch). First suppose m > n. Since from
linear algebra we know that rank(AB) rank(A) and that the rank of an
m n matrix cannot exceed n (or m), we have that rank(AB) n < m.
But AB is an mm matrix, so det(AB) = 0, as claimed.
Now assume m n. We use notation such as M
rs
to denote an r s
matrix M. It is an immediate consequence of the denition of matrix multi-
plication (which the reader should check) that
_
R
mm
S
mn
T
nm
U
nn
_ _
V
mn
W
mm
X
nn
Y
nm
_
=
_
RV +SX RW +SY
TV +UX TW +UY
_
. (9.1)
In other words, we can multiply block matrices of suitable dimensions as
if their entries were numbers. Note that the entries of the right-hand side
of (9.1) all have well-dened dimensions (sizes), e.g., RV +SX is an mn
matrix since both RV and SX are mn matrices.
Now in equation (9.1) let R = I
m
(the m m identity matrix), S = A,
T = O
nm
(the n m matrix of 0s), U = I
n
, V = A, W = O
mm
, X = I
n
,
and Y = B. We get
_
I
m
A
O
nm
I
n
_ _
A O
mm
I
n
B
_
=
_
O
mn
AB
I
n
B
_
. (9.2)
Take the determinant of both sides of (9.2). The rst matrix on the left-hand
side is an upper triangular matrix with 1s on the main diagonal. Hence its
determinant is one. Since the determinant of a product of square matrices is
the product of the determinants of the factors, we get
A O
mm
I
n
B
O
mn
AB
I
n
B
. (9.3)
It is easy to see [why?] that the determinant on the right-hand side of
(9.3) is equal to det(AB). So consider the left-hand side. A nonzero term in
the expansion of the determinant on the left-hand side is obtained by taking
the product (with a certain sign) of m + n nonzero entries, no two in the
same row and column (so one in each row and each column). In particular,
153
we must choose m entries from the last m columns. These entries belong to
m of the bottom n rows [why?], say rows m + s
1
, m + s
2
, . . . , m + s
m
. Let
S = s
1
, s
2
, . . . , s
m
1, 2, . . . , n. We must choose n m further entries
from the last n rows, and we have no choice but to choose the 1s in those
rows m+i for which i , S. Thus every term in the expansion of the left-hand
side of (9.3) uses exactly n m of the 1s in the bottom left block I
n
.
What is the contribution to the expansion of the left-hand side of (9.3)
from those terms which use exactly the 1s from rows m + i where i , S?
We obtain this contribution by deleting all rows and columns to which these
1s belong (in other words, delete row m + i and column i whenever i
1, 2, . . . , n S), taking the determinant of the 2m 2m matrix M
S
that
remains, and multiplying by an appropriate sign [why?]. But the matrix M
S
is in block-diagonal form, with the rst block just the matrix A[S] and the
second block just B[S]. Hence det M
S
= (det A[S])(det B[S]) [why?]. Taking
all possible subsets S gives
det AB =
S{1,2,...,n}
|S|=m
(det A[S])(det B[S]).
It is straightforward but somewhat tedious to verify that all the signs are +;
we omit the details. This completes the proof.
In Section 1 we dened the adjacency matrix A(G) of a graph G with
vertex set V = v
1
, . . . , v
p
and edge set E = e
1
, . . . , e
q
. We now dene
two related matrices. Assume for simplicity that G has no loops. (This
assumption is harmless since loops have no eect on (G).)
9.5 Denition. Let G be as above. Give G an orientation o, i.e, for every
edge e with vertices u, v, choose one of the ordered pairs (u, v) or (v, u). (If
we choose (u, v), say, then we think of putting an arrow on e pointing from
u to v; and we say that e is directed from u to v, that u is the initial vertex
and v the nal vertex of e, etc.)
(a) The incidence matrix M(G) of G (with respect to the orientation o)
is the p q matrix whose (i, j)-entry M
ij
is given by
M
ij
=
_
_
_
1, if the edge e
j
has initial vertex v
i
1, if the edge e
j
has nal vertex v
i
0, otherwise.
(b) The laplacian matrix L(G) of G is the pp matrix whose (i, j)-entry
L
ij
is given by
L
ij
=
_
m
ij
, if i ,= j and there are m
ij
edges between v
i
and v
j
deg(v
i
), if i = j,
where deg(v
i
) is the number of edges incident to v
i
. (Thus L(G) is symmetric
and does not depend on the orientation o.)
Note that every column of M(G) contains one 1, one 1, and q 2
0s; and hence the sum of the entries in each column is 0. Thus all the
rows sum to the 0 vector, a linear dependence relation which shows that
rank(M(G)) < p. Two further properties of M(G) and L(G) are given by
the following lemma.
9.6 Lemma. (a) We have MM
t
= L.
(b) If G is regular of degree d, then L(G) = dI A(G), where A(G)
denotes the adjacency matrix of G. Hence if G (or A(G)) has eigenvalues
1
, . . . ,
p
, then L(G) has eigenvalues d
1
, . . . , d
p
.
Proof. (a) This is immediate from the denition of matrix multiplication.
Specically, for v
i
, v
j
V (G) we have
(MM
t
)
ij
=
e
k
E(G)
M
ik
M
jk
.
If i ,= j, then in order for M
ik
M
jk
,= 0, we must have that the edge e
k
connects the vertices v
i
and v
j
. If this is the case, then one of M
ik
and M
jk
will be 1 and the other 1 [why?], so their product is always 1. Hence
(MM
t
)
ij
= m
ij
, as claimed.
There remains the case i = j. Then M
ik
M
ik
will be 1 if e
k
is an edge
with v
i
as one of its vertices and will be 0 otherwise [why?]. So now we get
(MM
t
)
ii
= deg(v
i
), as claimed. This proves (a).
(b) Clear by (a), since the diagonal elements of MM
t
are all equal to
d.
Now assume that G is connected, and let M
0
(G) be M(G) with its last
row removed. Thus M
0
(G) has p 1 rows and q columns. Note that the
number of rows is equal to the number of edges in a spanning tree of G. We
call M
0
(G) the reduced incidence matrix of G. The next result tells us the
determinants (up to sign) of all (p1) (p1) submatrices N of M
0
. Such
155
submatrices are obtained by choosing a set S of p 1 edges of G, and taking
all columns of M
0
indexed by the edges in S. Thus this submatrix is just
M
0
[S].
9.7 Lemma. Let S be a set of p 1 edges of G. If S does not form the set
of edges of a spanning tree, then det M
0
[S] = 0. If, on the other hand, S is
the set of edges of a spanning tree of G, then det M
0
[S] = 1.
Proof. If S is not the set of edges of a spanning tree, then some subset R of
S forms the edges of a cycle C in G. Suppose that the cycle C dened by
R has edges f
1
, . . . , f
s
in that order. Multiply the column of M
0
[S] indexed
by f
j
by 1 if in going around C we traverse f
i
in the direction of its arrow;
otherwise multiply the column by 1. Then add these modied columns.
It is easy to see (check a few small examples to convince yourself) that we
get the 0 column. Hence the columns of M
0
[S] are linearly dependent, so
det M
0
[S] = 0, as claimed.
Now suppose that S is the set of edges of a spanning tree T. Let e be an
edge of T which is connected to v
p
(the vertex which indexed the bottom row
of M, i.e., the row removed to get M
0
). The column of M
0
[S] indexed by e
contains exactly one nonzero entry [why?], which is 1. Remove from M
0
[S]
the row and column containing the nonzero entry of column e, obtaining a
(p2)(p2) matrix M
0
. Note that det(M
0
[S]) = det(M
0
) [why?]. Let
T
be the tree obtained from T by contracting the edge e to a single vertex

(so that v
p
and the remaining vertex of e are merged into a single vertex u).
Then M
0
is just the matrix obtained from the incidence matrix M(T
) by
removing the row indexed by u [why?]. Hence by induction on the number
p of vertices (the case p = 1 being trivial), we have det(M
0
) = 1. Thus
det(M
0
[S]) = 1, and the proof follows.
Note. An alternative way of seeing that det(M
0
S) = 1 when S is the
set of of edges of a spanning tree T is as follows. Let u
1
, u
2
, . . . , u
p1
be an
ordering of the vertices v
1
, . . . , v
p1
such that u
i
is an endpoint of the tree
obtained from T by removing vertices u
1
, . . . , u
i1
. (It is easy to see that
such an ordering is possible.) Permute the rows of M
0
[S] so that the ith row
is indexed by u
i
. Then permute the columns in the order e
1
, . . . , e
p1
so that
e
i
is the unique edge adjacent to u
i
after u
1
, . . . , u
i1
have been removed.
Then we obtain a lower triangular matrix with 1s on the main diagonal,
so the determinant is 1.
We have now assembled all the ingredients for the main result of this
section (due originally to Borchardt). Recall that (G) denotes the number
of spanning trees of G.
9.8 Theorem (the Matrix-Tree Theorem). Let G be a nite connected graph
without loops, with laplacian matrix L = L(G). Let L
0
denote L with the
last row and column removed (or with the ith row and column removed for
any i). Then
det(L
0
) = (G).
Proof. Since L = MM
t
(Lemma 9.6(a)), it follows immediately that L
0
=
M
0
M
t
0
. Hence by the Binet-Cauchy theorem (Theorem 9.4), we have
det(L
0
) =
S
(det M
0
[S])(det M
t
0
[S]), (9.4)
where S ranges over all (p1)-element subsets of 1, 2 . . . , q (or equivalently,
over all (p 1)-element subsets of the set of edges of G). Since in general
A
t
[S] = A[S]
t
, equation (9.4) becomes
det(L
0
) =
S
(det M
0
[S])
2
. (9.5)
According to Lemma 9.7, det(M
0
[S]) is 1 if S forms the set of edges of a
spanning tree of G, and is 0 otherwise. Therefore the term indexed by S in
the sum on the right-hand side of (9.5) is 1 if S forms the set of edges of a
spanning tree of G, and is 0 otherwise. Hence the sum is equal to (G), as
desired.
The operation of removing a row and column from L(G) may seem
somewhat contrived. We would prefer a description of (G) directly in
terms of L(G). Such a description will follow from the next lemma. Re-
call that the characteristic polynomial of a p p matrix A is dened to be
det(A xI). Note. Sometimes the characteristic polynomial is dened to
be det(xI A) = (1)
p
det(AxI). We will use the denition det(AxI).
9.9 Lemma. Let M be a p p matrix (with entries in a eld) such that
the sum of the entries in every row and column is 0. Let M
0
be the matrix
obtained from M by removing the last row and last column (or more generally,
any row and any column). Then the coecient of x in the characteristic
polynomial det(M xI) of M is equal to p det(M
0
). (Moreover, the
constant term of det(M xI) is 0.)
157
Proof. The constant term of det(M xI) is det(M), which is 0 since the
rows of M sum to 0.
For simplicity we prove the rest of the lemma only for removing the last
row and column, though the proof works just as well for any row and column.
Add all the rows of MxI except the last row to the last row. This doesnt
eect the determinant, and will change the entries of the last row all to x
(since the rows of M sum to 0). Factor out x from the last row, yielding a
matrix N(x) satisfying det(MxI) = xdet(N(x)). Hence the coecient of
x in det(M xI) is given by det(N(0)). Now add all the columns of N(0)
except the last column to the last column. This does not eect det(N(0)).
Because the columns of M sum to 0, the last column of N(0) becomes the
column vector [0, 0, . . . , 0, p]
t
. Expanding the determinant by the last column
shows that det(N(0)) = p det(M
0
), and the proof follows.
9.10 Corollary. (a) Let G be a connected (loopless) graph with p vertices.
Suppose that the eigenvalues of L(G) are
1
, . . . ,
p1
,
p
, with
p
= 0. Then
(G) =
1
p
2

p1
.
(b) Suppose that G is also regular of degree d, and that the eigenvalues of
A(G) are
1
, . . . ,
p1
,
p
, with
p
= d. Then
(G) =
1
p
(d
1
)(d
2
) (d
p1
).
Proof. (a) We have
det(LxI) = (
1
x) (
p1
x)(
p
x)
= (
1
x)(
2
x) (
p1
x)x.
Hence the coecient of x is
1
2

p1
. By Lemma 9.9, we get
1
2

p1
=
p det(L
0
). By Theorem 9.8 we have det(L
0
) = (G), and the proof follows.
(b) Immediate from (a) and Lemma 9.6(b).
Let us look at a couple of examples of the use of the Matrix-Tree Theorem.
9.11 Example. Let G = K
p
, the complete graph on p vertices. Now K
p
is
regular of degree d = p 1, and by Proposition 1.5 its eigenvalues are 1
(p 1 times) and p 1 = d. Hence from Corollary 9.10 there follows the
elegant result
(K
p
) =
1
p
((p 1) (1))
p1
= p
p2
.
9.12 Example. Let G = C
n
, the n-cube discussed in Section 2. Now C
n
is regular of degree n, and by Corollary 2.5 its eigenvalues are n 2i with
multiplicity
_
n
i
_
for 0 i n. Hence from Corollary 9.10 there follows the
amazing result
(C
n
) =
1
2
n
n
i=1
(2i)
(
n
i
)
= 2
2
n
n1
n
i=1
i
(
n
i
)
.
A direct combinatorial proof (though not an explicit bijection) was found by
O. Bernardi in 2012.
159
APPENDIX: THREE ELEGANT COMBINATORIAL
PROOFS
In this appendix we give three elegant combinatorial proofs that the num-
ber of spanning trees of the complete graph K
p
is p
p2
(Example 9.11). The
proofs are given in chronological order of their discovery.
First proof (Pr ufer). Given a spanning tree T of K
p
, i.e., a tree on the
vertex set [p], remove the largest endpoint (leaf) v and write down the vertex
a
1
adjacent to v. Continue this procedure until only two vertices remain,
obtaining a sequence (a
1
, . . . , a
p2
) [p]
p2
, called the Pr ufer sequence of T.
For the tree below, we rst remove 11 and then record 8. Next remove 10
and record 1. Then remove 8 and record 4, etc., ending with the sequence
(8, 1, 4, 4, 1, 4, 9, 1, 9) and leaving the two vertices 1 and 9.
6
3
1
2 7
9 4
5 10
8
11
We claim that the map just dened from trees T on [p] to sequences
(a
1
, . . . , a
p2
) [p]
p2
is a bijection, thereby completing the proof since
clearly [p]
p2
has p
p2
elements. The crucial observation is that the rst
vertex to be removed from T is the largest vertex of T missing from the se-
quence [why? this takes a little thought]. This vertex is adjacent to a
1
. For
our example, we get that 11 was the rst vertex removed, and that 11 is ad-
jacent to 8. We can now proceed recursively. If T
1
denotes T with the largest
missing vertex removed, then the Pr ufer sequence of T
1
is (a
2
, . . . , a
p2
). The
rst vertex to be removed from T
1
is the largest vertex of T
2
missing from
(a
2
, . . . , a
p2
). This missing vertex is adjacent to a
2
. For our example, this
missing vertex is 10 (since 11 is not a vertex of T
2
), which is adjacent to 1.
Continuing in this way, we determine one new edge of T at each step. At the
end we have found p 2 edges, and the remaining two unremoved vertices
form the (p 1)st edge.
Second proof (Joyal). A doubly-rooted tree is a tree T with one vertex
u labelled S (for start) and one vertex v (which may equal u) labelled E
(end). Let t(p) be the number of trees T on the vertex set [p], and let d(p)
be the number of doubly-rooted trees on [p]. Thus
d(p) = p
2
t(p), (9.6)
since once we have chosen T there are p choices for u and p choices for v.
Let T be a doubly-rooted tree. There is a unique path from S to E,
say with vertices S = b
1
, b
2
, . . . , b
k
= E (in that order). The diagram below
shows such a doubly-rooted tree.
14
S E
11 10 15 7 5 2 3
6 9
1 16
8 13
4 12 17
Let a
1
< a
2
< < a
k
be the increasing rearrangement of the num-
bers b
1
, b
2
, . . . , b
k
. Let be the permutation of the set a
1
, . . . , a
k
given
by (a
i
) = b
i
. Let D
be the digraph of , that is, the vertex set of D
is
a
1
, . . . , a
k
, with a directed edge a
i
b
i
for 1 i k. Since any permu-
tation of a nite set is a disjoint product of cycles, it follows that D
is
a disjoint union of directed cycles (all edges of each cycle point in the same
direction as we traverse the cycle). For the example above, we have k = 7.
(b
1
, . . . , b
7
) = (11, 10, 15, 7, 5, 2, 3) and (a
1
, . . . , a
7
) = (2, 3, 5, 7, 10, 11, 15).
The digraph D
is shown below.
7
2
11 15 5
3 10
Now attach to each vertex v of D
the same subgraph T

v
that was attached
below v in T, and direct the edges of T
v
toward v, obtaining a digraph D
T
.
For our example we get
161
11
15 5
6
9
14 13 8
16
1
2
7
4 12 17
3 10
The graph D
T
has the crucial property that every vertex has outdegree
one, that is, one arrow pointing out. In other words, D
T
is the graph of a
function f : [p] [p], with vertex set [p] and edges i f(i). Conversely,
given a function f : [p] [p], all the above steps can be reversed to ob-
tain a unique doubly-rooted tree T for which D
T
is the graph of f. We
have therefore found a bijection from doubly-rooted trees on [p] to functions
f : [p] [p]. Since the number of such functions f is p
p
, it follows that
d(p) = p
p
. Then from equation (9.6) we get t(p) = p
p2
.
Third proof (Pitman). A forest is a graph without cycles; thus every
connected component is a tree. A planted forest is a forest F for which every
component T has a distinguished vertex r
T
(called the root of T). Thus if a
component T has k vertices, then there are k ways to choose the root of T.
Let P
p
be the set of all planted forests on [p]. Let uv be an edge of a
forest F P
p
such that u is closer than v to the root r of its component.
Dene F to cover the rooted forest F
if F
is obtained by removing the edge

uv from F, and rooting the new tree containing v at v. This denition of
cover denes the covering relation of a partial order on P
p
. Under this partial
order P
p
is graded of rank p 1. The rank of a forest F in P
p
is its number
of edges. The diagram below shows the poset P
3
.
2
2
1 3 3
1
2
3
2
3
1
3
2
3
1
1
1
3
2
2
1
2
1 2
2
3
1 3
1
3
It is an easy exercise to see that an element F of P
p
of rank i covers i
elements and is covered by (p i 1)p elements. We now count in two ways
the number M
p
of maximal chains of P
p
. On the one hand, we can start at
the top. The number of maximal elements of P
p
is p t(p), where t(p) as
above is the number of trees on the vertex set [p], since there are p ways to
choose the root of such a tree. Once a maximal element F is chosen, then
there are p1 elements F
that it covers, then p2 elements that F
covers,
etc., giving
M
p
= p t(p)(p 1)! = p! t(p). (9.7)
On the other hand, we can start at the bottom. There is a unique element
F of rank one (the planted forest with no edges), then (p 1)p elements F
that cover F, then (p 2)p elements that cover F
, etc., giving
M
p
= p
p1
(p 1)!. (9.8)
Comparing equations (9.7) and (9.8) gives t(p) = p
p2
.
Our third proof isnt an explicit bijection like the rst two proofs. On the
other hand, it has the virtue of not depending on the names of the vertices.
Note that in the rst two proofs it is necessary to know when one vertex is
larger than another.
163
The concept of tree as a formal mathematical object goes back to G.
Kirchho and K. G. C. von Staudt. Trees were rst extensively investigated
by A. Cayley, to whom the term tree is due. In particular, in [18] Cayley
states the formula (K
p
) = p
p2
for the number of spanning trees of K
p
, and
he gives a vague idea of a combinatorial proof. Because of this paper, Cayley
is often credited with the enumeration of labelled trees. Cayley pointed
out, however, that an equivalent result had been proved earlier by C. W.
Borchardt [9]. Moreover, this result appeared even earlier in a paper of J.
J. Sylvester [87]. Undoubtedly Cayley and Sylvester could have furnished
a complete, rigorous proof had they the inclination to do so. The elegant
combinatorial proofs given in the appendix are due to E. P. H. Pr ufer, [68],
A. Joyal, [45, Exam. 12, pp. 1516] and J. W. Pitman [62].
The Matrix-Tree Theorem (Theorem 9.8) was rst proved by C. W. Bor-
chardt [9] in 1860, though a similar result had earlier been published by J.
J. Sylvester [87] in 1857. Cayley [17, p. 279] in fact in 1856 referred to the
not-yet-published work of Sylvester. For further historical information on
the Matrix-Tree theorem, see Moon [57, p. 42].
[more??]
Exercise 9.4 is based on a suggestion of P. Venkataramana.
Chapter 10
Eulerian digraphs and oriented
trees.
A famous problem which goes back to Euler asks for what graphs G is there
a closed walk which uses every edge exactly once. (There is also a version for
non-closed walks.) Such a walk is called an Eulerian tour (also known as an
Eulerian cycle). A graph which has an Eulerian tour is called an Eulerian
graph. Eulers famous theorem (the rst real theorem of graph theory) states
that a graph G without isolated vertices (which clearly would be irrelevant) is
Eulerian if and only if it is connected and every vertex has even degree. Here
we will be concerned with the analogous theorem for directed graphs. We
want to know not just whether an Eulerian tour exists, but how many there
are. We will prove an elegant determinantal formula for this number closely
related to the Matrix-Tree Theorem. For the case of undirected graphs no
analogous formula is known, explaining why we consider only the directed
case.
A (nite) directed graph or digraph
D consists of a vertex set V = v
1
, . . . , v
p
and edge set E = e
1
, . . . , e
q
,
together with a function : E V V (the set of ordered pairs (u, v) of
elements of V ). If (e) = (u, v), then we think of e as an arrow from u to v.
We then call u the initial vertex and v the nal vertex of e. (These concepts
arose in the denition of an orientation in Denition 8.5.) A tour in D is
a sequence e
1
, e
2
, . . . , e
r
of distinct edges such that the nal vertex of e
i
is
the initial vertex of e
i+1
for all 1 i r 1, and the nal vertex of e
r
is
the initial vertex of e
1
. A tour is Eulerian if every edge of D occurs at least
once (and hence exactly once). A digraph which has no isolated vertices and
169
170 CHAPTER 10. EULERIAN DIGRAPHS AND ORIENTED TREES.
contains an Eulerian tour is called an Eulerian digraph. Clearly an Eulerian
digraph is connected. The outdegree of a vertex v, denoted outdeg(v), is
the number of edges of G with initial vertex v. Similarly the indegree of v,
denoted indeg(v), is the number of edges of D with nal vertex v. A loop
(edge of the form (v, v)) contributes one to both the indegree and outdegree.
A digraph is balanced if indeg(v) = outdeg(v) for all vertices v.
10.1 Theorem. A digraph D is Eulerian if and only if it is connected and
balanced.
Proof. Assume D is Eulerian, and let e
1
, . . . , e
q
be an Eulerian tour. As we
move along the tour, whenever we enter a vertex v we must exit it, except
at the very end we enter the nal vertex v of e
q
without exiting it. However,
at the beginning we exited v without having entered it. Hence every vertex
is entered as often as it is exited and so must have the same outdegree as
indegree. Therefore D is balanced, and as noted above D is clearly connected.
Now assume that D is balanced and connected. We may assume that D
has at least one edge. We rst claim that for any edge e of D, D has a tour
for which e = e
1
. If e
1
is a loop we are done. Otherwise we have entered
the vertex n(e
1
) for the rst time, so since D is balanced there is some exit
edge e
2
. Either n(e
2
) = init(e
1
) and we are done, or else we have entered
the vertex n(e
2
) once more than we have exited it. Since D is balanced
there is new edge e
3
with n(e
2
) = init(e
3
). Continuing in this way, either
we complete a tour or else we have entered the current vertex once more than
we have exited it, in which case we can exit along a new edge. Since D has
nitely many edges, eventually we must complete a tour. Thus D does have
a tour which uses e
1
.
Now let e
1
, . . . , e
r
be a tour C of maximum length. We must show that
r = q, the number of edges of D. Assume to the contrary that r < q. Since in
moving along C every vertex is entered as often as it is exited (with init(e
1
)
exited at the beginning and entered at the end), when we remove the edges
of C from D we obtain a digraph H which is still balanced, though it need
not be connected. However, since D is connected, at least one connected
component H
1
of H contains at least one edge and has a vertex v in common
with C [why?]. Since H
1
is balanced, there is an edge e of H
1
with initial
vertex v. The argument of the previous paragraph shows that H
1
has a tour
C
of positive length beginning with the edge e. But then when moving along
C, when we reach v we can take the detour C
before continuing with C.

171
This gives a tour of length longer than r, a contradiction. Hence r = q, and
the theorem is proved.
Our primary goal is to count the number of Eulerian tours of a connected
balanced digraph. A key concept in doing so is that of an oriented tree.
An oriented tree with root v is a (nite) digraph T with v as one of its
vertices, such that there is a unique directed path from any vertex u to v.
In other words, there is a unique sequence of edges e
1
, . . . , e
r
such that (a)
init(e
1
) = u, (b) n(e
r
) = v, and (c) n(e
i
) = init(e
i+1
) for 1 i r 1.
Its easy to see that this means that the underlying undirected graph (i.e.,
erase all the arrows from the edges of T) is a tree, and that all arrows
in T point toward v. There is a surprising connection between Eulerian
tours and oriented trees, given by the next result (due to de Bruijn and van
Aardenne-Ehrenfest). This result is sometimes called the BEST Theorem,
after de Bruijn, van Aardenne-Ehrenfest, Smith, and Tutte. However, Smith
and Tutte were not involved in the original discovery.
10.2 Theorem. Let D be a connected balanced digraph with vertex set V .
Fix an edge e of D, and let v = init(e). Let (D, v) denote the number
of oriented (spanning) subtrees of D with root v, and let (D, e) denote the
number of Eulerian tours of D starting with the edge e. Then
(D, e) = (D, v)
uV
(outdeg(u) 1)!. (10.1)
Proof. Let e = e
1
, e
2
, . . . , e
q
be an Eulerian tour E in D. For each vertex
u ,= v, let e(u) be the last exit from u in the tour, i.e., let e(u) = e
j
where
init(e(u)) = u and init(e
k
) ,= u for any k > j.
Claim #1. The vertices of D, together with the edges e(u) for all vertices
u ,= v, form an oriented subtree of D with root v.
Proof of Claim #1. This is a straightforward verication. Let T be the
spanning subgraph of D with edges e(u), u ,= v. Thus if [V [ = p, then T has
p vertices and p 1 edges [why?]. There are three items to check to insure
that T is an oriented tree with root v:
(a) T does not have two edges f and f
satisfying init(f) = init(f
). This
is clear since both f and f
cant be last exits from the same vertex.

(b) T does not have an edge f with init(f) = v. This is clear since by
denition the edges of T consist only of last exits from vertices other
than v, so no edge of T can exit from v.
(c) T does not have a (directed) cycle C. For suppose C were such a cycle.
Let f be that edge of C which occurs after all the other edges of C in
the Eulerian tour E. Let f
be the edge of C satisfying n(f) = init(f
)
(= u, say). We cant have u = v by (b). Thus when we enter u via
f, we must exit u. We cant exit u via f
since f occurs after f
in E.
Hence f
is not the last exit from u, contradicting the denition of T.

Its easy to see that conditions (a)(c) imply that T is an oriented tree with
root v, proving the claim.
Claim #2. We claim that the following converse to Claim #1 is true.
Given a connected balanced digraph D and a vertex v, let T be an oriented
(spanning) subtree of D with root v. Then we can construct an Eulerian tour
E as follows. Choose an edge e
1
with init(e
1
) = v. Then continue to choose
any edge possible to continue the tour, except we never choose an edge f
of E unless we have to, i.e., unless its the only remaining edge exiting the
vertex at which we stand. Then we never get stuck until all edges are used,
so we have constructed an Eulerian tour E. Moreover, the set of last exits
of E from vertices u ,= v of D coincides with the set of edges of the oriented
tree T.
Proof of Claim #2. Since D is balanced, the only way to get stuck is to
end up at v with no further exits available, but with an edge still unused.
Suppose this is the case. At least one unused edge must be a last exit edge,
i.e., an edge of T [why?]. Let u be a vertex of T closest to v in T such that
the unique edge f of T with init(f) = u is not in the tour. Let y = n(f).
Suppose y ,= v. Since we enter y as often as we leave it, we dont use the
last exit from y. Thus y = v. But then we can leave v, a contradiction. This
proves Claim #2.
We have shown that every Eulerian tour E beginning with the edge e
has associated with it a last exit oriented subtree T = T(E) with root
v = init(e). Conversely, given an oriented subtree T with root v, we can
obtain all Eulerian tours E beginning with e and satisfying T = T(E) by
choosing for each vertex u ,= v the order in which the edges from u, except
the edge of T, appear in E; as well as choosing the order in which all the
edges from v except for e appear in E. Thus for each vertex u we have
(outdeg(u) 1)! choices, so for each T we have
u
(outdeg(u) 1)! choices.
Since there are (G, v) choices for T, the proof is complete.
10.3 Corollary. Let D be a connected balanced digraph, and let v be a vertex
of D. Then the number (D, v) of oriented subtrees with root v is independent
173
of v.
Proof. Let e be an edge with initial vertex v. By equation (10.1), we need
to show that the number (G, e) of Eulerian tours beginning with e is indepen-
dent of e. But e
1
e
2
e
q
is an Eulerian tour if and only if e
i
e
i+1
e
q
e
1
e
2
e
i1
is also an Eulerian tour, and the proof follows [why?].
What we obviously need to do next is nd a formula for (G, v). Such a
formula is due to W. Tutte in 1948. This result is very similar to the Matrix-
Tree Theorem, and indeed we will show (Example 10.6) that the Matrix-Tree
Theorem is a simple corollary to Theorem 10.4.
10.4 Theorem. Let D be a loopless connected digraph with vertex set V =
v
1
, . . . , v
p
. Let L(D) be the p p matrix dened by
L
ij
=
_
_
_
m
ij
, if i ,= j and there are m
ij
edges with
initial vertex v
i
and nal vertex v
j
outdeg(v
i
), if i = j.
(Thus L is the directed analogue of the laplacian matrix of an undirected
graph.) Let L
0
denote L with the last row and column deleted. Then
det L
0
= (D, v
p
). (10.2)
Note. If we remove the ith row and column from L instead of the last row
and column, then equation (10.2) still holds with v
p
replaced with v
i
.
Proof (sketch). Induction on q, the number of edges of D. The fewest
number of edges which D can have is p 1 (since D is connected). Suppose
then that D has p 1 edges, so that as an undirected graph D is a tree. If
D is not an oriented tree with root v
p
, then some vertex v
i
,= v
p
of D has
outdegree 0 [why?]. Then L
0
has a zero row, so det L
0
= 0 = (D, v
p
). If
on the other hand D is an oriented tree with root v
p
, then an argument like
that used to prove Lemma 9.7 (in the case when S is the set of edges of a
spanning tree) shows that det L
0
= 1 = (D, v
p
).
Now assume that D has q > p 1 edges, and assume the theorem for
digraphs with at most q 1 edges. We may assume that no edge f of D
has initial vertex v, since such an edge belongs to no oriented tree with root
v and also makes no contribution to L
0
. It then follows, since D has at
least p edges, that there exists a vertex u ,= v of D of outdegree at least
two. Let e be an edge with init(e) = u. Let D
1
be D with the edge e
removed. Let D
2
be D with all edges e
removed such that init(e) = init(e
)
and e
,= e. (Note that D
2
is strictly smaller than D since outdeg(u) 2.)
By induction, we have det L
0
(D
1
) = (D
1
, v
p
) and det L
0
(D
2
) = (D
2
, v
p
).
Clearly (D, v
p
) = (D
1
, v
p
) + (D
2
, v
p
), since in an oriented tree T with
root v
p
, there is exactly one edge whose initial vertex coincides with that of
e. On the other hand, it follows immediately from the multilinearity of the
determinant [why?] that
det L
0
(D) = det L
0
(D
1
) + det L
0
(D
2
).
From this the proof follows by induction.
10.5 Corollary. Let D be a connected balanced digraph with vertex set V =
v
1
, . . . , v
p
. Let e be an edge of D. Then the number (D, e) of Eulerian
tours of D with rst edge e is given by
(D, e) = (det L
0
(D))
uV
(outdeg(u) 1)!.
Equivalently (since D is balanced, so Lemma 9.9 applies), if L(D) has eigen-
values
1
, . . . ,
p
with
p
= 0, then
(D, e) =
1
p
1

p1
uV
(outdeg(u) 1)!.
Proof. Combine Theorems 10.2 and 10.4.
10.6 Example. (the Matrix-Tree Theorem revisited) Let G be a connected
loopless undirected graph. Let

G be the digraph obtained from G by replac-
ing each edge e = uv of G with a pair of directed edges u v and v u.
Clearly

G is balanced and connected. Choose a vertex v of G. There is
an obvious one-to-one correspondence between spanning trees T of G and
oriented spanning trees

T of

G with root v, namely, direct each edge of T
toward v. Moreover, L(G) = L(

G) [why?]. Hence the Matrix-Tree Theorem
is an immediate consequence of the Theorem 10.4.
10.7 Example. (the ecient mail carrier) A mail carrier has an itinerary of
city blocks to which he (or she) must deliver mail. He wants to accomplish
this by walking along each block twice, once in each direction, thus passing
175
along houses on each side of the street. The blocks form the edges of a
graph G, whose vertices are the intersections. The mail carrier wants simply
to walk along an Eulerian tour in the digraph

G of the previous example.
Making the plausible assumption that the graph is connected, not only does
an Eulerian tour always exist, but we can tell the mail carrier how many
there are. Thus he will know how many dierent routes he can take to avoid
boredom. For instance, suppose G is the 3 3 grid illustrated below.
r
r
r
r
r
r
r
r
r
This graph has 128 spanning trees. Hence the number of mail carrier
routes beginning with a xed edge (in a given direction) is 128 1!
4
2!
4
3! =
12288. The total number of routes is thus 12288 times twice the number of
edges [why?], viz., 1228824 = 294912. Assuming the mail carrier delivered
mail 250 days a year, it would be 1179 years before he would have to repeat
a route!
10.8 Example. (binary de Bruijn sequences) A binary sequence is just a
sequence of 0s and 1s. A binary de Bruijn sequence of degree n is a bi-
nary sequence A = a
1
a
2
a
2
n such that every binary sequence b
1
b
n
of
length n occurs exactly once as a circular factor of A, i.e., as a sequence
a
i
a
i+1
a
i+n1
, where the subscripts are taken modulo n if necessary. For
instance, some circular factors of the sequence abcdefg are a, bcde, fgab, and
defga. Note that there are exactly 2
n
binary sequences of length n, so the
only possible length of a binary de Bruijn sequence of degree n is 2
n
[why?].
Clearly any cyclic shift a
i
a
i+1
a
2
na
1
a
2
a
i1
of a binary de Bruijn se-
quence a
1
a
2
a
2
n is also a binary de Bruijn sequence, and we call two such
sequences equivalent. This relation of equivalence is obviously an equivalence
relation, and every equivalence class contains exactly one sequence beginning
with n 0s [why?]. Up to equivalence, there is one binary de Bruijn sequence
of degree two, namely, 0011. Its easy to check that there are two inequivalent
binary de Bruijn sequences of degree three, namely, 00010111 and 00011101.
However, its not clear at this point whether binary de Bruijn sequences exist
for all n. By a clever application of Theorems 10.2 and 10.4, we will not only
show that such sequences exist for all positive integers n, but we will also
count the number of them. It turns out that there are lots of them. For
instance, the number of inequivalent binary de Bruijn sequences of degree
eight is equal to
1329227995784915872903807060280344576.
The reader with some extra time on his or her hands is invited to write down
these sequences. De Bruijn sequences are named after Nicolaas Govert de
Bruijn, who published his work on this subject in 1946. However, it was
discovered in 1975 that de Bruijn sequences had been earlier created and
enumerated by C. Flye Sainte-Marie in 1894. De Bruijn sequences have a
number of interesting applications to the design of switching networks and
related topics.
Our method of enumerating binary de Bruijn sequences will be to set
up a correspondence between them and Eulerian tours in a certain directed
graph D
n
, the de Bruijn graph of degree n. The graph D
n
has 2
n1
vertices,
which we will take to consist of the 2
n1
binary sequences of length n1. A
pair (a
1
a
2
a
n1
, b
1
b
2
b
n1
) of vertices forms an edge of D
n
if and only if
a
2
a
3
a
n1
= b
1
b
2
b
n2
, i.e., e is an edge if the last n 2 terms of init(e)
agree with the rst n2 terms of n(e). Thus every vertex has indegree two
and outdegree two [why?], so D
n
is balanced. The number of edges of D
n
is
2
n
. Moreover, its easy to see that D
n
is connected (see Lemma 10.9). The
graphs D
3
and D
4
look as follows:
01
111
110
001
00
10
11
100
011
101
000
010
Suppose that E = e
1
e
2
e
2
n is an Eulerian tour in D
n
. If n(e
i
) is the
binary sequence a
i1
a
i2
a
i,n1
, then replace e
i
in E by the last bit a
i,n1
. It
177
is easy to see that the resulting sequence (E) = a
1,n1
a
2,n1
a
2
n
,n1
is a
binary de Bruijn sequence, and conversely every binary de Bruijn sequence
arises in this way. In particular, since D
n
is balanced and connected there
exists at least one binary de Bruijn sequence. In order to count the total
number of such sequences, we need to compute det L(D
n
). One way to
do this is by a clever but messy sequence of elementary row and column
operations which transforms the determinant into triangular form. We will
give instead an elegant computation of the eigenvalues of L(D
n
) based on
the following simple lemma.
10.9 Lemma. Let u and v be any two vertices of D
n
. Then there is a unique
(directed) walk from u to v of length n 1.
Proof. Suppose u = a
1
a
2
a
n1
and v = b
1
b
2
b
n1
. Then the unique
path of length n 1 from u to v has vertices
a
1
a
2
a
n1
, a
2
a
3
a
n1
b
1
, a
3
a
4
a
n1
b
1
b
2
, . . . ,
a
n1
b
1
b
n2
, b
1
b
2
b
n1
.
10.10 Theorem. The eigenvalues of L(D
n
) are 0 (with multiplicity one)
and 2 (with multiplicity 2
n1
1).
Proof. Let A(D
n
) denote the directed adjacency matrix of D
n
, i.e., the rows
and columns are indexed by the vertices, with
A
uv
=
_
1, if (u, v) is an edge
0, otherwise.
Now Lemma 10.9 is equivalent to the assertion that A
n1
= J, the 2
n1
2
n1
matrix of all 1s [why?]. If the eigenvalues of A are
1
, . . .
2
n1, then the
eigenvalues of J = A
n1
are
n1
1
, . . . ,
n1
2
n1
. By Lemma 1.4, the eigenvalues
of J are 2
n1
(once) and 0 (2
n1
1 times). Hence the eigenvalues of A are
2 (once, where is an (n 1)-st root of unity to be determined), and 0
(2
n1
1 times). Since the trace of A is 2, it follows that = 1, and we have
found all the eigenvalues of A.
Now L(D
n
) = 2I A(D
n
) [why?]. Hence the eigenvalues of L are 2
1
, . . . , 2
2
n1, and the proof follows from the above determination of
1
, . . . ,
2
n1.
10.11 Corollary. The number B
0
(n) of binary de Bruijn sequences of degree
n beginning with n 0s is equal to 2
2
n1
n
. The total number B(n) of binary
de Bruijn sequences of degree n is equal to 2
2
n1
.
Proof. By the above discussion, B
0
(n) is the number of Eulerian tours in D
n
whose rst edge the loop at vertex 00 0. Moreover, the outdegree of every
vertex of D
n
is two. Hence by Corollary 10.5 and Theorem 10.10 we have
B
0
(n) =
1
2
n1
2
2
n1
1
= 2
2
n1
n
.
Finally, B(n) is obtained from B
0
(n) by multiplying by the number 2
n
of
edges, and the proof follows.
Note that the total number of binary sequences of length 2
n
is N = 2
2
n
.
By the previous corollary, the number of these which are de Bruijn sequences
is just
N. This suggests the following problem, solved by H. Bidkhori and

S. Kishore. Let /
n
be the set of all binary sequences of length 2
n
. Let B
n
be
the set of binary de Bruijn sequences of degree n. Find an explicit bijection
: B
n
B
n
/
n
, thereby giving a combinatorial proof of Corollary 10.11.
The characterization of Eulerian digraphs given by Theorem 10.1 is a
result of I. J. Good [33], while the fundamental connection between oriented
subtrees and Eulerian tours in a balanced digraph that was used to prove
Theorem 10.2 was shown by T. van Aardenne-Ehrenfest and N. G. de Bruijn
[1, Thm. 5a]. This result is sometimes called the BEST Theorem, after
de Bruijn, van Aardenne-Ehrenfest, Smith, and Tutte. However, Smith
and Tutte were not involved in the original discovery. (In [77] Smith and
Tutte give a determinantal formula for the number of Eulerian tours in a
special class of balanced digraphs. Van Aardenne-Ehrenfest and de Bruijn
refer to the paper of Smith and Tutte in a footnote added in proof.) The
determinantal formula for the number of oriented subtrees of a directed graph
(Theorem 10.4) is due to Tutte [92, Thm. 3.6].
Chapter 11
Cycles, bonds, and electrical
networks.
11.1 The cycle space and bond space.
In this section we will deal with some interesting linear algebra related to
the structure of a directed graph. Let D = (V, E) be a digraph. A function
f : E R is called a circulation if for every vertex v V , we have
eE
init(e)=v
f(e) =
eE
n(e)=v
f(e). (11.1)
Thus if we think of the edges as pipes and f as measuring the ow (quantity
per unit of time) of some commodity (such as oil) through the pipe in the
specied direction (so that a negative value of f(e) means a ow of [f(e)[
in the direction opposite the direction of e), then equation (11.1) simply
says that the amount owing into each vertex equals the amount owing
out. In other words, the ow is conservative. The gure below illustrates a
circulation in a digraph D.
3
5 2
3
4
7
-6
-1
181
182CHAPTER 11. CYCLES, BONDS, AND ELECTRICAL NETWORKS.
Let ( = (
D
denote the set of all circulations on D. Clearly if f, g C
and , R then f + g (. Hence ( is a (real) vector space, called the
cycle space of D. Thus if q = [E[, then (
D
is a subspace of the q-dimensional
vector space R
E
of all functions f : E R.
What do circulations have do with something circulating, and what
does the cycle space have to do with actual cycles? To see this, dene a
circuit or elementary cycle in D to be a set of edges of a closed walk, ignoring
the direction of the arrows, with no repeated vertices except the rst and
last. Suppose that C has been assigned an orientation (direction of travel)
o. (Note that this meaning of orientation is not the same as that appearing
in Denition 9.5.)
Dene a function f
C
: E R (which also depends on the orientation o,
though we suppress it from the notation) by
f
C
(e) =
_
_
_
1, if e C and e agrees with o
1, if e C and e is opposite to o
0, otherwise.
It is easy to see that f
C
is a circulation. Later we will see that the circu-
lations f
C
span the cycle space (, explaining the terminology circulation
and cycle space. The gure below shows a circuit C with an orientation o,
and the corresponding circulation f
C
.
1
1
-1
-1
-1
11.1. THE CYCLE SPACE AND BOND SPACE. 183
5
-3
3
-3
3
3
0
4
-1
2
3
5 0
1
Figure 11.1: A function and its coboundary
Given a function p : V R, dene a new function p : E R, called
the coboundary
1
of p, by
p(e) = p(v) p(u), if u = init(e) and v = n(e).
Figure 11.1 shows a digraph D with the value p(v) of some function p : V R
indicated at each vertex v, and the corresponding values p(e) shown at each
edge e.
One should regard as an operator which takes an element p of the vector
space R
V
of all functions V R and produces an element of the vector space
R
E
of all functions E R. It is immediate from the denition of that is
linear, i.e.,
(p +q) = p + q,
for all p, q R
V
and , R. Thus is simply a linear transformation
: R
V
R
E
between two nite-dimensional vector spaces.
A function g : E R is called a potential dierence on D if g = p for
some p : V R. (Later we will see the connection with electrical networks
that accounts for the terminology potential dierence.) Let B = B
D
be the
set of all potential dierences on D. Thus B is just the image of the linear
transformation and is hence a real vector space, called the bond space of
D.
Let us explain the reason behind the terminology bond space. A bond in
a digraph D is a set B of edges such that (a) removing B from D disconnects
some (undirected) component of D (that is, removing B creates a digraph
which has more connected components, as an undirected graph, than D), and
(b) no proper subset of B has this property. A subset of edges satisfying (a)
1
The term coboundary arises from algebraic topology, but we will not explain the
connection here.
is called a cutset, so a bond is just a minimal cutset. Suppose, for example,
that D is given as follows (with no arrows drawn since they are irrelevant to
the denition of bond):
r r
r r
d
d
d
d
d
dd
a
d c b
e
Then the bonds are the six subsets ab, de, acd, bce, ace, bcd.
Let B be a bond. Suppose B disconnects the component (V
, E
) into two
pieces (a bond always disconnects some component into exactly two pieces
[why?]) with vertex set S in one piece and

S in the other. Thus S

S = V
and S

S = . Dene
[S,

S] = e E : exactly one vertex of e lies in S and one lies in

S.
Clearly B = [S,

S]. It is often convenient to use the notation [S,

S] for a
bond.
Given a bond B = [S,

S] of D, dene a function g
B
: E R by
g
B
(e) =
_
_
_
1, if init(e)

S, n(e) S
1, if init(e) S, n(e)

S
0, otherwise.
Note that g
B
really depends not just on B, but on whether we write B as
[S,

S] or [
S, S]. Writing B in the reverse way simply changes the sign of g

B
.
Whenever we deal with g
B
we will assume that some choice B = [S,

S] has
been made.
Now note that g
B
= p, where
p(v) =
_
1, if v S
0, if v

S.
Hence g
B
B, the bond space of D. We will later see that B is in fact
spanned by the functions g
B
, explaining the termininology bond space.
11.1 Example. In the digraph below, open (white) vertices indicate an
element of S and closed (black) vertices an element of

S for a certain bond
B = [S,

S]. The elements of B are drawn with solid lines. The edges are
11.1. THE CYCLE SPACE AND BOND SPACE. 185
labelled by the values of g
B
, and the vertices by the function p for which
g
B
= p.
0
1
1
1
0
0
1
1
0
0
0
1
1
Recall that in Denition 9.5 we dened the incidence matrix M(G) of a
loopless undirected graph G with respect to an orientation o. We may just
as well think of G together with its orientation o as a directed graph. We
also will allow loops. Thus if D = (V, E) is any (nite) digraph, dene the
incidence matrix M = M(D) to be the pq matrix whose rows are indexed
by V and columns by E, as follows. The entry in row v V and column
e E is denoted m
v
(e) and is given by
2
m
v
(e) =
_
_
_
1, if v = init(e) and e is not a loop
1, if v = n(e) and e is not a loop
0, otherwise.
For instance, if D is given by
1
2
1
4
3
5
3
2
2
Actually, this denition gives the negative of the matrix dened in Denition 9.5,
though it makes no dierence here. We will x this inconsistency of notation in a later
version of these notes.
then
M(D) =
_
_
1 1 1 0 0
1 1 0 1 0
0 0 1 1 0
_
_
.
11.2 Theorem. The row space of M(D) is the bond space B
D
. Equivalently,
the functions m
v
: E R, where v ranges over all vertices of D, span B.
Proof. Let g = p be a potential dierence on D, so
g(e) = p(n(e)) p(init(e))
=
vV
p(v)m
v
(e).
Thus g =
vV
p(v)m
v
, so g belongs to the row space of M.
Conversely, if g =
vV
q(v)m
v
is in the row space of M, where q : V
R, then g = q B.
We now dene a scalar product (or inner product) on the space R
E
by
f, g) =
eE
f(e)g(e),
for any f, g R
E
. If we think of the numbers f(e) and g(e) as the coordinates
of f and g with respect to the basis E, then f, g) is just the usual dot product
of f and g. Because we have a scalar product, we have a notion of what it
means for f and g to be orthogonal, viz., f, g) = 0. If 1 is any subspace of
R
E
, then dene the orthogonal complement 1
of 1 by
1
= f R
E
: f, g) = 0 for all g R
E
.
Recall from linear algebra that
dim1 + dim1
= dimR
E
= #E. (11.2)
Furthermore,
_
1
= mathcalV .
Intuitively there is a kind of duality between elementary cycles and
bonds. Cycles hold vertices together, while bonds tear them apart. The
precise statement of this duality is given by the next result.
11.3 Theorem. The cycle and bond spaces of D are related by ( = B
.
(Equivalently, B = (
.)
11.2. BASES FOR THE CYCLE SPACE AND BOND SPACE. 187
Proof. Let f : E R. Then f is a circulation if and only if
eE
m
v
(e)f(e) = 0
for all v V [why?]. But this is exactly the condition that f B
.
11.2 Bases for the cycle space and bond space.
We want to examine the incidence matrix M(D) in more detail. In particu-
lar, we would like to determine which rows and columns of M(D) are linearly
independent, and which span the row and column spaces. As a corollary, we
will determine the dimension of the spaces B and (. We begin by dening
the support |f| of f : E R to be the set of edges e E for which f(e) ,= 0.
11.4 Lemma. If 0 ,= f (, then |f| contains an undirected circuit.
Proof. If not, then |f| has a vertex of degree one [why?], which is clearly
impossible.
11.5 Lemma. If 0 ,= g B, then |g| contains a bond.
Proof. Let 0 ,= g B, so g = p for some p : V R. Choose a vertex v
which is incident to an edge of |g|, and set
U = u V : p(u) = p(v).
Let

U = V U. Note that

U ,= , since otherwise p is constant so g = 0.
Since g(e) ,= 0 for all e [U,

U] [why?], we have that |g| contains the cutset
[U,

U]. Since a bond is by denition a minimal cutset, it follows that |g|
contains a bond.
A matrix B is called a basis matrix of B if the rows of B form a basis
for B. Similary dene a basis matrix C of (.
Recall the notation of Theorem 9.4: Let A be a matrix with at least as
many columns as rows, whose columns are indexed by the elements of a set
T. If S T, then A[S] denotes the submatrix of A consisting of the columns
indexed by the elements of S. In particular, A[e] (short for A[e]) denotes
the column of A indexed by e. We come to our rst signicant result about
bases for the vector spaces B and (.
11.6 Theorem. Let B be a basis matrix of B, and C a basis matrix of (.
(Thus the columns of B and C are indexed by the edges e E of D.) Let
S E, Then:
(i) The columns of B[S] are linearly independent if and only if S is acyclic
(i.e., contains no circuit as an undirected graph).
(ii) The columns of C[S] are linearly independent if and only if S contains
no bond.
Proof. The columns of B[S] are linearly dependent if and only if there exists
a function f : E R such that
f(e) ,= 0 for some e S
f(e) = 0 for all e , S
eE
f(e)B[e] = 0 the column vector of 0s. (11.3)
The last condition is equivalent to f, m
v
) = 0 for all v V , i.e., f is a
circulation. Thus the columns of B[S] are linearly dependent if and only if
there exists a nonzero circulation f such that |f| S. By Lemma 11.4,
|f| (and therefore S) contains a circuit. Conversely, if S contains a circuit
C then 0 ,= f
C
( and |f
C
| = C S, so f
C
denes a linear dependence
relation (11.3) among the columns. Hence the columns of B[S] are linearly
independent if and only if S is acyclic, proving (i). (Part (i) can also be
deduced from Lemma 9.7.)
The proof of (ii) is similar and is left as an exercise.
11.7 Corollary. Let D = (V, E) be a digraph with p vertices, q edges, and
k connected components (as an undirected graph). Then
dimB = p k
dim( = q p +k.
Proof. For any matrix X, the rank of X is equal to the maximum number of
linearly independent columns. Now let B be a basis matrix of B. By Theo-
rem 11.6(i), the rank of B is then the maximum size (number of elements)
of an acyclic subset of E. In each connected component D
i
of D, the largest
acyclic subsets are the spanning trees, whose number of edges is p(D
i
) 1,
where p(D
i
) is the number of vertices of D
i
. Hence
rank B =
k
i=1
(p(D
i
) 1)
= p k.
Since dimB+dim( = dimR
E
= q by equation (11.2) and Theorem 11.3, we
have
dim( = q (p k) = q p +k.
(It is also possible to determine dim( by a direct argument similar to our
determination of dimB.)
The number q p + k (which should be thought of as the number of
independent cycles in D) is called the cyclomatic number of D (or of its
undirected version G, since the direction of the edges have no eect).
Our next goal is to describe explicit bases of ( and B. We begin by
dening a forest to be an undirected graph without circuits. Thus a forest
is a disjoint union of trees. We extend the denition of forest to directed
graphs by ignoring the arrows, i.e., a directed graph is a forest if it has no
circuits as an undirected graph. Equivalently [why?], dim( = 0.
Pick a maximal forest T of D = (V, E). Thus T restricted to each com-
ponent of D is a spanning tree. If e is an edge of D not in T, then it is easy
to see that T e contains a unique circuit C
e
.
11.8 Theorem. Let T be as above. Then the set S of circulations f
Ce
, as e
ranges over all edges of D not in T, is a basis for the cycle space (.
Proof. The circulations f
Ce
are linearly independent, since for each e
E(D) E(T) only f
Ce
doesnt vanish on e. Moreover,
#S = #E(D) #E(T) = q p +k = dim(,
so S is a basis.
11.9 Example. Let D be the digraph shown below, with the edges a, b, c of
T shown by dotted lines.
a
b
d
c f
e
Orient each circuit C
t
in the direction of the added edge, i.e., f
Ct
(t) = 1.
Then the basis matrix C of ( corresponding to the basis f
C
d
, f
Ce
, f
C
f
is
given by
C =
_
_
0 1 1 1 0 0
1 1 1 0 1 0
0 0 1 0 0 1
_
_
. (11.4)
We next want to nd a basis for the bond space B analogous to that of
Theorem 11.8.
11.10 Lemma. Let T be a maximal forest of D = (V, E). Let T
= DE(T)
(the digraph obtained from D by removing the edges of T), called a cotree if
D is connected. Let e be an edge of T. Then E(T
) e contains a unique
bond.
Proof. Removing E(T
) from D leaves a maximal forest T, so removing one

further edge e disconnects some component of D. Hence E(T
) e contains
a bond B. It remains to show that B is unique. Removing e from T breaks
some component of T into two connected graphs T
1
and T
2
with vertex sets
S and

S. It follows [why?] that we must have B = [S,

S], so B is unique.
Let T be a maximal forest of the digraph D, and let e be an edge of T.
By the previous lemma, E(T
)e contains a unique bond B

e
. Let g
Be
be the
corresponding element of the bond space B, chosen for deniteness so that
g
Be
(e) = 1.
11.11 Theorem. The set of functions g
Be
, as e ranges over all edges of T,
is a basis for the bond space B.
Proof. The functions g
Be
are linearly independent, since only g
Be
is nonzero
on e E(T). Since
#E(T) = p k = dimB,
it follows that the g
Be
s are a basis for B.
11.12 Example. Let D and T be as in the previous diagram. Thus a basis
for B is given by the functions g
Ba
, g
B
b
, g
Bc
. The corresponding basis matrix
is given by
B =
_
_
1 0 0 0 1 0
0 1 0 1 1 0
0 0 1 1 1 1
_
_
.
Note that the rows of B are orthogonal to the rows of the matrix C of
equation (11.4), in accordance with Theorem 11.3. Equivalently, BC
t
= 0,
the 3 3 zero matrix. (In general, BC
t
will have q p +k rows and p k
columns. Here it is just a coincidence that these two numbers are equal.)
The basis matrices C
T
and B
T
of ( and B obtained from a maximal
forest T have an important property. A real matrix m n matrix A with
m n is said to be unimodular if every mm submatrix has determinant
0, 1, or 1. For instance, the adjacency matrix M(D) of a digraph D is
unimodular, as proved in Lemma 9.7 (by showing that the expansion of the
determinant of a full submatrix has at most one nonzero term).
11.13 Theorem. Let T be a maximal forest of D. Then the basis matrices
C
T
of ( and B
T
of B are unimodular.
Proof. First consider the case C
T
. Let P be a full submatrix of C (so
P has q p + k rows and columns). Assume det P ,= 0. We need to
show det P = 1. Since det P ,= 0, it follows from Theorem 11.6(ii) that
P = C
T
[T
1
] for the complement T
1
of some maximal forest T
1
. Note that
the rows of the matrix C
T
[T
1
] are indexed by T
and the columns by T
1
.
Similarly the rows of the basis matrix C
T
1
are indexed by T
1
and the columns
by E (the set of all edges of D). Hence it makes sense to dene the matrix
product
Z = C
T
[T
1
]C
T
1
,
a matrix whose rows are indexed by T
and columns by E.
Note that the matrix Z is a basis matrix for the cycle space ( since its
rows are linear combinations of the rows of the basis matrix C
T
1
, and it has
full rank since the matrix C
T
[T
1
] is invertible. Now C
T
1
[T
1
] = I
T
1
(the
identity matrix indexed by T
1
), so Z[T
1
] = C
T
[T
1
]. Thus Z agrees with the
basis matrix C
T
in columns T
1
. Hence the rows of Z C
T
are circulations
supported on a subset of T
1
. Since T
1
is acyclic, it follows from Lemma 11.4
that the only such circulation is identically 0, so Z = C
T
.
We have just shown that
C
T
[T
1
]C
T
1
= C
T
.
Restricting both sides to T
, we obtain
C
T
[T
1
]C
T
1
[T
] = C
T
[T
] = I
T
.
Taking determinants yields
det(C
T
[T
1
]) det(C
T
1
[T
]) = 1.
Since all the matrices we have been considering have integer entries, the
above determinants are integers. Hence
det C
T
[T
1
] = 1,
as was to be proved. (This proof is due to Tutte in 1965.)
A similar proof works for B
T
.
11.3 Electrical networks.
We will give a brief indication of the connection between the above discus-
sion and the theory of electrical networks. Let D be a digraph, which for
convenience we assume is connected and loopless. Suppose that at each edge
e there is a voltage (potential dierence) V
e
from init(e) to n(e), and a cur-
rent I
e
in the direction of e (so a negative current I
e
indicates a current of
[I
e
[ in the direction opposite to e). Think of V and I as functions on the
edges, i.e., as elements of the vector space R
E
. There are three fundamental
laws relating the quantities V
e
and I
e
.
Kirchhos First Law. I (
D
. In other words, the current owing
into a vertex equals the current owing out. In symbols,
e
init(e)=v
I
e
=
e
n(e)=v
I
e
,
for all vertices v V .
Kirchhos Second Law. V (
D
= B. In other words, the sum of the
voltages around any circuit (called loops by electrical engineers), taking into
account orientations, is 0.
11.3. ELECTRICAL NETWORKS. 193
Ohms Law. If edge e has resistance R
e
> 0, then V
e
= I
e
R
e
.
The central problem of electrical network theory dealing with the above
three laws
3
is the following: Which of the 3q quantities V
e
, I
e
, R
e
need to
be specied to uniquely determine all the others, and how can we nd or
stipulate the solution in a fast and elegant way? We will be concerned here
only with a special case, perhaps the most important special case in practical
applications. Namely, suppose we apply a voltage V
q
at edge e
q
, with resis-
tances R
1
, . . . , R
q1
at the other edges e
1
, . . . , e
q1
. Let V
i
, I
i
be the voltage
and current at edge e
i
. We would like to express each V
i
and I
i
in terms
of V
q
and R
1
, . . . , R
q1
. (By physical intuition there should be a unique
solution, since we can actually build a network meeting the specications of
the problem.) Note that if we have quantities V
i
, I
i
, R
i
satisfying the three
network laws above, then for any scalar the quantities V
i
, I
i
, R
i
are also
a solution. This means that we might as well assume that V
q
= 1, since we
can always multiply all voltages and currents afterwards by whatever value
we want V
q
to be.
As an illustration of a simple method of computing the total resistance of
a network, the following diagram illustrates the notion of a series connection
D
1
+D
2
and a parallel connection D
1
| D
2
of two networks D
1
and D
2
with
a distinguished edge e at which a voltage is applied.
D
2
A
e
B e
B
A
A
B
e e
D
1
D + D
1 2
D || D
1 2
3
Of course the situation becomes much more complicated when one introduces dynamic
network elements like capacitors, alternating current, etc.
If R(D) denotes the total resistance V
e
/I
e
of the network D together
with the distinguished edge e, then it is well-known and easy to deduce from
the three network Laws that
R(D
1
+D
2
) = R(D
1
) +R(D
2
)
1
R(D
1
| D
2
)
=
1
R(D
1
)
+
1
R(D
2
)
.
A network that is built up from a single edge by a sequence of series and
parallel connections is called a series-parallel network. An example is the
following, with the distinguished edge e shown by a broken line from bottom
to top.
e
The simplest network which is not a series-parallel network and has no mul-
tiple edges (as an undirected graph) is called the Wheatstone bridge and is
illustrated below. (The direction of the arrows has been chosen arbitrarily.)
We will use this network as our main example in the discussion that follows.
1
4
3
5
2
6
We now return to an arbitrary connected loopless digraph D, with cur-
rents I
i
, voltages V
i
, and resistances R
i
at the edges e
i
. Recall that we are
xing V
q
= 1 and R
1
, . . . , R
q1
. Let T be a spanning tree of D. Since I is a
current if and only if it is orthogonal to the bond space B (Theorem 11.3 and
Kirchhos First Law), it follows that any basis for B denes a complete and
minimal set of linear relations satised by the I
i
s (namely, the relation that
I is orthogonal to the basis elements). In particular, the basis matrix C
T
denes such a set of relations. For example, if D is the Wheatstone bridge
shown above and if T = e
1
, e
2
, e
5
, then we obtain the following relations
by adding the edges e
1
, e
2
, e
5
of T in turn to T
.
I
1
I
3
I
4
= 0
I
2
+I
3
+I
4
+I
6
= 0 (11.5)
I
4
+I
5
+I
6
= 0
These three (= p 1) equations give all the relations satised by the I
i
s
alone, and the equations are linearly independent.
Similary if V is a voltage then it is orthogonal to the cycle space (. Thus
any basis for ( denes a complete and minimal set of linear relations satised
by the V
i
s (namely, the relation that V is orthogonal to the basis elements).
In particular, the basis matrix C
T
denes such a set of relations. Continuing
our example, we obtain the following relations by adding the edges e
3
, e
4
, e
6
of T
in turn to T.
V
1
V
2
+V
3
= 0
V
1
V
2
+V
4
V
5
= 0 (11.6)
V
2
+V
5
= 1,
These three (= q p +k) equations give all the relations satised by the V
i
s
alone, and the equations are linearly independent.
In addition, Ohms Law gives the q 1 equations V
i
= R
i
I
i
, 1 i q 1.
We have a total of (p k) + (q p +k) + (q 1) = 2q 1 equations in the
2q 1 unknowns I
i
(1 i q) and V
i
(1 i q 1). Moreover, it is easy to
see that these 2q 1 equations are linearly independent, using the fact that
we already know that just the equations involving the I
i
s alone are linearly
independent, and similarly the V
i
s. Hence this system of 2q 1 equations in
2q 1 unknowns has a unique solution. We have now reduced the problem
to straightforward linear algebra. However, it is possible to describe the
solution explicitly. We will be content here with giving a formula just for the
total resistance R(D) = V
q
/I
q
= 1/I
q
.
Write the 2q 1 equations in the form of a (2q 1) 2q matrix K. The
columns of the matrix are indexed by I
1
, I
2
, . . . , I
q
, V
1
, V
2
, . . . , V
q
. The last
column V
q
of the matrix keeps track of the constant terms of the equations.
The rows of K are given rst by the equations among the I
i
s, then the
V
i
s, and nally Ohms Law. For our example of the Wheatstone bridge, we
obtain the matrix
K =
I
1
I
2
I
3
I
4
I
5
I
6
V
1
V
2
V
3
V
4
V
5
V
6
1 0 1 1 0 0 0 0 0 0 0 0
0 1 1 1 0 1 0 0 0 0 0 0
0 0 0 1 1 1 0 0 0 0 0 0
0 0 0 0 0 0 1 1 1 0 0 0
0 0 0 0 0 0 1 1 0 1 1 0
0 0 0 0 0 0 0 1 0 0 1 1
R
1
0 0 0 0 0 1 0 0 0 0 0
0 R
2
0 0 0 0 0 1 0 0 0 0
0 0 R
3
0 0 0 0 0 1 0 0 0
0 0 0 R
4
0 0 0 0 0 1 0 0
0 0 0 0 R
5
0 0 0 0 0 1 0
We want to solve for I
q
by Cramers rule. Call the submatrix consisting of
all but the last column X. Let Y be the result of replacing the I
q
column of
X by the last column of K. Cramers rule then asserts that
I
q
=
det Y
det X
.
We evaluate det X by taking a Laplace expansion along the rst p 1 rows.
In other words,
det X =
S
det(X[[p 1], S]) det(X[[p 1]
c
,

S]), (11.7)
where (a) S indexes all (p1)-element subsets of the columns, (b) X[[p1], S]
denotes the submatrix of X consisting of entries in the rst p 1 rows and
in the columns S, (c) X[[p 1]
c
,

S] denotes the submatrix of X consisting
of entries in the last 2q p rows and in the columns other than S. In
order for det(X[[p 1], S]) ,= 0, we must choose S = I
i
1
, . . . , I
i
p1
, where
e
i
1
, . . . , e
i
p1
is a spanning tree T
1
(by Theorem 11.6(i)). In this case,
det(X[[p 1], S]) = 1 by Theorem 11.13. If I
q
, S, then the I
q
column of
X[[p 1]
c
,

S] will be zero. Hence to get a nonzero term in (11.7), we must
have e
q
S. The matrix X[[p 1]
c
,

S] will have one nonzero entry in each
of the rst q p + 1 columns, namely, the resistances R
j
where e
j
is not an
edge of T
1
. This accounts for q p + 1 entries from the last q 1 rows of
X[[p 1]
c
,

S]. The remaining p 2 of the last q 1 rows have available only
one nonzero entry each, a 1 in the columns indexed by V
j
where e
j
is an
edge of T
1
other than e
q
. Hence we need to choose q p+1 remaining entries
from rows p through q and columns indexed by V
j
for e
j
not edge of T
1
. By
Theorems 11.6(ii) and 11.13, this remaining submatrix has determinant 1.
It follows that
det(X[[p 1], S]) det(X[[p 1]
c
,

S]) =
e
j
E(T
1
)
R
j
.
Hence by (11.7), we get
det X =
T
1
_
_
e
j
E(T
1
)
R
j
_
_
, (11.8)
where T
1
ranges over all spanning trees of D containing e
q
. A careful analysis
of the signs
4
shows that all signs in (11.8) are plus, so we nally arrive at the
remarkable formula
det X =
spanning trees T
1
containing eq
e
j
E(T
1
)
R
j
.
4
To be inserted.
For example, if D is the Wheatstone bridge as above, and if we abbreviate
R
1
= a, R
2
= b, R
3
= c, R
4
= d, R
5
= e, then
det X = abc +abd +abe +ace +ade +bcd +bde +cde.
Now suppose we replace column I
q
in X by column V
q
in the matrix K,
obtaining the matrix Y . There is a unique nonzero entry in the new column,
so it must be chosen in any nonzero term in the expansion of det Y . The
argument now goes just as it did for det X, except we have to choose S to
correspond to a spanning tree T
1
that doesnt contain e
q
. We therefore obtain
det Y =
spanning trees T
1
not containing eq
e
j
E(T
1
)
e
j
=eq
R
j
.
For example, for the Wheatstone bridge we get
det Y = ac +ad +ae +bc +bd +be +cd +ce.
Recall that I
q
= det(Y )/ det(X) and that the total resistance of the net-
work is 1/I
q
. Putting everything together gives our main result on electrical
networks.
11.14 Theorem. In the situation described above, the total resistance of the
network is given by
R(D) =
1
I
q
=
spanning trees T
1
containing eq
e
j
E(T
1
)
R
j
spanning trees T
1
not containing eq
e
j
E(T
1
)
e
j
=eq
R
j
.
11.15 Corollary. If the resistances R
1
, . . . , R
q1
are all equal to one, then
the total resistance of the network is given by
R(D) =
1
I
q
=
number of spanning trees containing e
q
number of spanning trees not containing e
q
.
In particular, if R
1
= = R
q1
= 1, then the total resistance, when
reduced to lowest terms a/b, has the curious property that the number (D)
of spanning trees of D is divisible by a +b.
11.4. PLANAR GRAPHS (SKETCH). 199
11.4 Planar graphs (sketch).
A graph G is planar if it can be drawn in the plane R
2
without crossing
edges. A drawing of G in this way is called a planar embedding.
If the vertices and edges of a planar embedding of G are removed from R
2
,
then we obtain a disjoint union of open sets, called the regions (or faces) of
G. (More precisely, these open sets are the regions of the planar embedding
of G. Often we will not bother to distinguish between a planar graph and a
planar embedding if no confusion should result.) Let R = R(G) be the set
of regions of G, and as usual V (G) and E(G) denote the set of vertices and
edges of G, respectively.
Note. If G is simple (no loops or multiple edges) then it can be shown
that there exists a planar embedding with edges as straight lines and with
regions (regarding as the sequence of vertices and edges obtained by walking
around the boundaries of the regions) preserved.
The dual G
of the planar embedded graph G has vertex set R(G) and

edge set E
(G) = e
: e E(G). If e is an edge of G, then let r and r
be
the regions on its two sides. (Possibly r = r
; there are ve such edges in the

example above.) Then dene e
to connect r and r
. We can always draw

G
to be planar, letting e and e
intersect once. If G is connected then every

region of G
contains exactly one nonisolated vertex of G and G

= G.
11.16 Example. Let G consist of two disjoint edges. Then G
has one vertex

and two loops, while G
is a three-vertex path. The unbounded region of

G
contains two vertices of G, and G
= G.
Orient the edges of the planar graph G in any way to get a digraph D.
Let r be an interior (i.e., bounded) region of D. An outside edge of r is an
edge e such that r lies on one side of the edge, and a dierent region lies on
the other side. The outside edges of any interior region r dene a circulation
(shown as solid edges in the diagram below), and these circulations (as r
ranges over all interior regions of D) form a basis for the cycle space (
G
of
G.
11.4. PLANAR GRAPHS (SKETCH). 201
r
Given the orientation D of G, orient the edges of G
as follows: as we
walk along e in the direction of its orientation, e
points to our right.

11.17 Theorem. Let f : E(G) R. Dene f
: E(G
) R by f
(e
) =
f(e). Then
f B
G
f
(
G
f (
G
f
B
G
.
11.18 Proposition. The set S is the set of edges of a spanning tree T of G
if and only if S
= e
: e S is the set of edges of a cotree T
of G
.
11.19 Corollary. (G) = (G
)
For nonplanar graphs there is still a notion of a dual object, but it is
no longer a graph but rather something called a matroid. Matroid theory is
a ourishing subject which may be regarded as a combinatorial abstraction
of linear algebra.
11.5 Squaring the square.
A squared rectangle is a rectangle partitioned into nitely many (but more
than one) squares. A squared rectangle is perfect if all the squares are of
dierent sizes. The earliest squared rectangle was found in 1936; its size is
33 32 and consists of nine squares:
4
8
9
7
10
14
18
15
1
The question then arose: does there exist a perfect squared square? An
isolated example with 55 squares was found by Sprague in 1939. Then
Brooks, Smith, Stone, and Tutte developed a network theory approach which
we now explain.
The Smith diagram D of a squared rectangle is a directed graph whose
vertices are the horizontal line segments of the squared rectangle and whose
11.5. SQUARING THE SQUARE. 203
squares are the edges, directed from top to bottom. The top vertex (cor-
responding to the top edge of the rectangle being squared) and the bottom
vertex (corresponding to the bottom edge) are called poles. Label each edge
by the side length of the square to which it corresponds. The gure below
shows the Smith diagram of the (perfect) squared rectangle above.
7
18
15
8
1
9
10
14
4
pole
pole
The following result concerning Smith diagrams is straightforward to ver-
ify. 11.20 Theorem. (a) If we set I
e
and V
e
equal to the label of edge e, then
Kirchhos two laws hold (so R
e
= 1) except at the poles.
(b) The Smith diagram is planar and can be drawn without separation of
poles. Joining the poles by an edge from the bottom to the top gives a
3-connected graph, i.e., a connected graph that remains connected when
one or two vertices are removed.
Call the 3-connected graph of Theorem 11.20 the extended Smith diagram
of the squared rectangle. If we label the new edge e
1
between poles by the
horizontal length b of the squared rectangle and set V
e
1
= I
e
1
= b, then
Kirchhos two laws hold at all vertices.
7
18
15
8
1
9
10
14
4
pole
pole
32
We therefore have a recipe for searching for perfect squared rectangles
and squares: start listing all 3-connected planar graphs. Then choose an
edge e
1
to apply a voltage V
1
. Put a resistance R
e
= 1 at the remaining
edges e. Solve for I
e
(= V
e
) to get a squared rectangle, and hope that one of
these will be a square. One example found by Brooks et al. was a 112 75
rectangle with 14 squares. It was given to Brooks mother as a jigsaw puzzle,
and she found a dierent solution ! We therefore have found a squared
square (though not perfect):
11.5. SQUARING THE SQUARE. 205
112 x 112

75 x 75
Building on this idea, Brooks et al. nally found two 422 593 per-
fect rectangles with thirteen squares, all 26 squares being of dierent sizes.
Putting them together as above gives a perfect squared square. This example
has two defects: (a) it contains a smaller perfect squared rectangle (and is
therefore not simple), and (b) it contains a cross (four squares meeting
a point). They eventually found a perfect squared square with 69 squares
without either of these defects. It is now known (thanks to computers) that
the smallest order (number of squares) of a perfect squared square is 21. It
is unique and happens to be simple and crossfree. See the gure below. It is
known that the number (up to symmetry) of simple perfect squared squares
of order n for n 21 is 1, 8, 12, 26, 160, 441, 1152, . . . .
6
2
25
16
9
7
33
29
19
8
27
35
50
37
42
18
11
15
4
24
Chapter 12
Miscellaneous gems of algebraic
combinatorics.
12.1 Oddtown
The village of Oddtown has a population of n people. Inhabitants of Oddtown
like to form clubs. Every club has an odd number of members, and every
pair of clubs share an even number of members.
12.1 Theorem. There are at most n clubs.
Proof. Let k be the number of clubs. Dene a matrix M = (M
ij
) over the
two-element eld F
2
as follows. The rows of M are indexed by the clubs C
i
and the columns by the inhabitants x
j
of Oddtown. Set
M
ij
=
_
1, x
j
C
i
0, otherwise.
The matrix M is called the incidence matrix corresponding to the clubs and
their members.
In general, let S be a subset of [n], and let
S
Z
n
be the characteristic
vector of S, i.e.,
S
= (a
1
, . . . , a
n
) where
a
i
=
_
1, i S
0, i , S.
If T is another subset of [n], then the key observation is that the scalar (dot)
product of
S
and
T
is given by
S

T
= #(S T). Hence if we now work
209
210CHAPTER 12. MISCELLANEOUS GEMS OF ALGEBRAIC COMBINATORICS.
over F
2
, then
S

T
=
_
1, if #(S T) is odd
0, if #(S T) is even.
(12.1)
Let A = MM
t
, a k k matrix. By equation (12.1) and the assumption
that every club has an odd number of members, we see that main diagonal
elements of A are 1. Similarly the o-diagonal elements of A are 0, so
A = I
k
, the k k identity matrix. Hence rank(A) = k.
Recall that if B is a k m matrix and C is an mn matrix (over some
eld), then rank(BC) rank(B) (as well as rank(BC) rank(C)), since for
any matrix D, rank(D) = dimimage(D). Hence, since M has n columns,
n rank(M) rank(MM
t
) = rank(A) = k.
While Theorem 12.1 can be proved without linear algebra, the proof is
not easy.
12.2 Complete bipartite partitions of K
n
Figure 12.1 show the six edges of the complete graph K
4
partitioned (ac-
cording to the edge label) into the edge sets of the three complete bipartite
graphs K
3,1
, K
2,1
, and K
1,1
. Clearly we can extend this construction, achiev-
ing a partition of the edges E(K
n
) of K
n
into the edge sets of n1 complete
bipartite graphs. Specically, Let E
1
be the set of edges incident to a xed
vertex v. Thus E
1
is the edge set of a complete bipartite graph K
n1,1
.
Remove E
1
from E(K
n
) and proceed by induction, obtaining a partition of
E(K
n
) into the edges of K
n1,1
, K
n2,1
, . . . , K
1,1
. The question thus arises
as to whether E(K
n
) can be partitioned into fewer than n 1 edge sets of
complete bipartite graphs.
12.2 Theorem. If E(K
n
) is the disjoint union of the edge sets of m complete
bipartite graphs, then m n 1.
Proof. Let E(K
n
) = E(B
1
) E(B
1
) E(B
m
) (disjoint union), where B
k
is a complete bipartite graph with vertex bipartition (X
k
, Y
k
) (so X
k
Y
k
= ).
For 1 i n, dene an n n matrix A
k
by
(A
k
)
ij
=
_
1, i X
k
, j Y
k
0, otherwise.
12.2. COMPLETE BIPARTITE PARTITIONS OF K
N
211
1
2 1
1 2
3
Figure 12.1: A decomposition of the edges of K
4
into three complete bipartite
graphs
All nonzero rows of A
k
are equal, so rank(A
k
) = 1. Let S =
m
k=1
A
k
. For
i ,= j, exactly one of the 2m numbers (A
k
)
ij
and (A
k
)
ji
, 1 k m, is equal
to 1, since every edge ij of K
n
appears in one E(B
k
) with either i X
k
and
j Y
k
, or else j X
k
and i Y
k
. Hence
S +S
t
= J I,
where as usual J is the nn all 1s matrix, and I is the nn identity matrix.
Claim. If T is any real matrix satisfying T +T
t
= J I, then rank(T)
n 1.
Suppose to the contrary that rank(T) n 2. Then T has (at least)
two linearly independent eigenvectors x, y such that Tx = Ty = 0 [why?].
Since J has rank one, the space x, y) spanned by x and y contains a nonzero
vector z satisfying Jz = 0 [why?]. Then from T + T
t
= J I and Tz = 0
we get z = T
t
z. Take the dot product with z
t
on the left. We get
[z[
2
= z
t
T
t
z
= (z
t
T
t
z)
t
(since a 1 1 matrix is symmetric)
= z
t
Tz (since in general (AB)
t
= B
t
A
t
)
= 0 (since Tz = 0),
contradicting z ,= 0. Hence the claim is proved, so in particular rank(X)
n 1. But is general rank(A + B) rank(A) + rank(B) [why?]. Therefore
from S =
m
k=1
A
k
and rank(A
k
) = 1 we get rank(S) m. It follows that
m n 1, completing the proof.
12.3 The nonuniform Fisher inequality
A balanced incomplete block design (BIBD) with parameters (v, k, , r, b) is
a v-element set X and a collection / of k-element subsets (blocks), with
#/ = b, such that any two points x, y X lie in exactly blocks, and
each point is in exactly r blocks. We also assume that k < v, which is the
reason for the word incomplete. We can draw a BIBD as a bipartite graph
with vertex bipartition (X, /). There is an edge from x X to A / if
x A. Thus the degree of each vertex x X is r, and the degree of each
vertex A / is k. It follows that vr = kb (the total number of edges of the
graph). We can also count the number of two-element sets of edges that are
incident to the same vertex of /. On the one hand, since each vertex in /
has degree k this number is b
_
k
2
_
. On the other hand, each pair of points in
X are mutually adjacent to points in /, so we get
_
v
2
_
= b
_
k
2
_
. A little
manipulation shows that these two equalities are equivalent to
vr = kb, (v 1) = r(k 1),
the usual form in which they are written.
R. A. Fisher showed in 1940 that b v. This inequality was generalized
by R. C. Bose in 1949. The most convenient way to state Boses inequalities,
known as the nonuniform Fisher inequality, is to reverse the roles of points
and blocks. Thus consider the elements x of X to be sets whose elements are
the blocks A / that contain them. In other words, we have a collection
C
1
, . . . , C
v
of r-element sets whose union contains b points x
1
, . . . , x
b
. Each
point is in exactly k of the sets. Finally, #(C
i
C
j
) = for all i ,= j.
12.3 Theorem. Let C
1
, . . . , C
v
be distinct subsets of a b-element set X such
that for all i ,= j we have #(C
i
C
j
) = for some 1 < b (independent
of i and j). Then v b.
Proof. Case 1: some #C
i
= . Then all other C
j
s contain C
i
and are
disjoint otherwise, so
v 1
..
from C
i
+ b
. .
from all C
j
=C
i
b.
12.4. ODD NEIGHBORHOOD COVERS 213
Figure 12.2: The 4 3 grid graph
Case 2: all #C
i
< . Let
i
= #C
i
. Let M be the incidence matrix of
the set system C
1
, . . . , C
v
, i.e., the rows of M correspond to the C
i
s and the
columns to the elements x
1
, . . . , x
b
of X, with
M
ij
=
_
1, x
j
C
i
0, x
j
, C
i
.
Let A = MM
t
. The hypotheses imply that A = J +C, where J as usual
is the all 1s matrix (of size v), and C is the diagonal matrix diag(
1
, . . . ,
v
).
Claim: rank(A) = v (i.e., A is invertible). We would then have
v = rank(A) rank(M) b,
the last inequality because M has b columns.
As in the proof of Theorem 4.7, a real symmetric matrix B is positive
semidenite if it has nonnegative eigenvalues. Equivalently, by basic linear
algebra, uBu
t
0 for all row vectors u of length v. Moreover B is positive
denite (and so has positive eigenvalues) if uBu
t
> 0 for all u ,= 0.
Now we easily compute that
u(J +C)u
t
= (u
1
+ +u
v
)
2
+
1
u
2
1
+ +
v
u
2
v
> 0
for all u ,= 0. Thus A = J + C is positive denite and hence of full rank
v.
12.4 Odd neighborhood covers
Consider an mn grid graph. The case m = 4, n = 3 is shown in Figure 12.2.
At each vertex is a turned on light bulb and also a switch that changes the
state of its bulb and those of its neighbors (adjacent vertices). Can all the
lights be turned o?
This problem was open for many years until in 1989 K. Sutner, then a
graduate student, showed using automata theory that the answer if yes for
any (nite) graph! More explicitly, let G be a nite graph with a turned on
light bulb at each vertex. At each vertex is a switch that changes the state
of that vertex and all its neighbors. Then it is possible to turn o all the
lights. We will give a modication of a simpler proof due to Y. Caro based
on linear algebra.
Without loss of generality we may assume that G is simple. If v V (G),
then the neighborhood N(v) of v is the set consisting of v and all vertices
adjacent to v. A little thought shows that we need to prove the following
result.
12.4 Theorem. There exists a subset S V = V (G) such that #(SN(v))
is odd for all v V . (It follows that switching at the vertices v S turns all
the lights o.)
Proof. Let #V (G) = v
1
, . . . , v
p
. Let A be the adjacency matrix of G
over the eld F
2
, and let y = (1, 1, . . . , 1) F
p
2
. Write row(B) for the row
space of a matrix B. Given S V , let
S
= (a
1
, . . . , a
p
) F
p
2
denote the
characteristic vector of S, i.e.,
a
i
=
_
1, v
i
S
0, v
i
, S.
Note that switching at S turns all the lights o if and only if
S
(A+I) = y.
Hence we need to show that y row(A+I) [why?].
Let us recall from linear algebra some standard facts about orthogonal
subspaces. Let K be a eld, and for u, v K
n
let u v be the usual dot
product (2.1) of u and v, so u v K. If W is a subspace of K
n
, then dene
the orthogonal subspace W
by
W
= u K
n
: u v = 0 for all v W.
Let d = dimW. Since W is the set of solutions to d linearly independent
homogeneous linear equations [why?], we have
dimW + dimW
= n.
As a consequence,
(W
= W. (12.2)
Note. Though irrelevant here, let us point out that if K has characteristic
0 then W W
= 0, but that this fact need not hold in characteristic

p ,= 0.
Now by equation (12.2) the vector y = (1, 1, . . . , 1) (or any vector in F
n
1
)
lies in the row space of A+ I if and only if it is orthogonal to every vector
12.5. CIRCULANT HADAMARD MATRICES 215
in row(A+I)
= ker(A+I). Thus we need to show that if (A+I)v

t
= 0,
then v y = 0. Equivalently, if yv
t
,= 0 then (A+ I)v
t
,= 0. Note that (a)
yv
t
,= 0 means the v has an odd number of 1s, and (b) (A+I)v
t
is the sum
of the rows of A+ I indexed by the positions of the 1s in v. Thus we need
to show that A+I does not have an odd number of rows summing to 0.
Suppose that v
1
, . . . , v
k
are vertices indexing rows of A summing to 0.
Let H be the subgraph induced by v
1
, . . . , v
k
, i.e., H consists of the vertices
v
1
, . . . , v
k
and all edges of G between two of these vertices. Let b
ij
be the
(i, j)-entry of A+ I. Since
k
i=1
b
ij
= 0 for 1 j n, and each b
ii
= 1, it
follows that every vertex of H has odd degree. Since [why?]
vV (H)
deg(v) = 2 #E(H),
we have that k = #V (H) is even, completing the proof.
12.5 Circulant Hadamard matrices
For our next gem of algebraic combinatorics, we will provide some variety
by leaving the realm of linear algebra and looking at some simple algebraic
number theory.
An n n matrix H is a Hadamard matrix if its entries are 1 and its
rows are orthogonal. Equivalently, its entries are 1 and HH
t
= nI. In
particular [why?],
det H = n
n/2
. (12.3)
It is easy to see that if H is an nn Hadamard matrix then n = 1, n = 2, or
n = 4m for some integer m 1. It is conjectured that the converse is true,
i.e., for every such n there exists an n n Hadamard matrix.
An n n matrix A = (b
ij
) is a circulant if it has the form b
ij
= a
ij
for some a
0
, a
1
, . . . , a
n1
, where the subscript i j is taken modulo n. For
instance,
A =
_
_
a b c d
d a b c
c d a b
b c d a
_
_
is a circulant. Let A = (a
ij
) be an n n circulant, and let = e
2i/n
,
a primitive nth root of unity. It is straightforward to compute that for
0 j < n the column vector [1,
j
,
2j
, . . . ,
(n1)j
]
t
is an eigenvector of A
with eigenvalue a
0
+
j
a
1
+
2j
a
2
+ +
(n1)j
a
n1
. Hence
det(A) =
n1
j=0
(a
0
+
j
a
1
+
2j
a
2
+ +
(n1)j
a
n1
). (12.4)
Note that the matrix
_
_
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
_
_
is both a Hadamard matrix and a circulant.
Conjecture Let H be an nn circulant Hadamard matrix. Then n = 1
or n = 4.
The rst signicnat work on this conjecture is due to R. J. Turyn. He
showed that there does not exist a circulant Hadamard matrix of order 8m,
and he also excluded certain other orders of the form 4(2m + 1). Turyns
proofs use the machinery of algebraic number theory. Here we will give a
proof for the special case n = 2
k
, k 3, where the algebraic number theory
can be dumbed down to elementary commutative algebra and eld theory.
(Only in Theorem 12.13 do we use a little Galois theory, which can be avoided
with a bit more work.) It would be interesting to nd similar proofs for other
values of n.
12.5 Theorem. There does not exist a circulant Hadamard matrix H of
order 2
k
, k 3.
Note. It is curious that the numbers 2
k
(k 2) are the easiest multiples
of 4 to show are not the orders of circulant Hadamard matrices, while on
the other hand the numbers 2
k
(k 1) are the easiest numbers to show are
the orders of Hadamard matrices. To see that 2
k
is the order of a Hadamard
matrix H, rst note that the case k = 1 is trivial. It is routine to show that
if H
1
is a Hadamard matrix of order a and H
2
is a Hadamard matrix of order
b, then the tensor (or Kronecker) product A B is a Hadamard matrix of
order ab. It follows that there exists a Hadamard matrix of order 2
k
, k 1.
From now on we assume n = 2
k
and = e
2i/2
k
. Clearly is a zero of
the polynomial p
k
(x) = x
2
k1
+ 1. We will be working in the ring Z[], the
smallest subring of C containing Q and . Write Q() for the quotient eld
of Z[], i.e., the eld obtained by adjoining to Q.
12.6 Lemma. The polynomial p
k
(x) is irreducible over Q.
Proof. If p
k
(x) is reducible then so is p
k
(x + 1). A standard fact about
polynomial factorization is Gauss lemma, namely, an integral polynomial
that factors over Q also factors over Z. If p(x), q(x) Z[x], write p(x)
q(x) (mod2) to mean that the coecients of p(x) q(x) are even. Now
p
k
(x + 1) (x + 1)
2
k1
+ 1 x
2
k1
(mod2).
Hence any factorization of p
k
(x+1) over Z into two factors of degree at least
one has the form p
k
(x +1) = (x
r
+ 2a)(x
s
+ 2b), where r +s = 2
k1
and a, b
are polynomial of degrees less than r and s, respectively. Hence the constant
term of p
k
(x + 1) is divisible by 4, a contradiction.
It follows by elementary eld theory that every element u Z[] can be
uniquely written in the form
u = b
0
+b
1
+b
2
2
+ +b
n/21
n/21
, b
i
Z.
The basis for our proof of Theorem 12.5 is the two dierent ways to
compute det H given by equations (12.3) and (12.4), yielding the formula
n1
j=0
(a
0
+
j
a
1
+
2j
a
2
+ +
(n1)j
a
n1
) = n
n/2
= 2
k2
k1
. (12.5)
Thus we have a factorization in Z[] of 2
k2
k1
. Algebraic number theory is
concerned with factorization of algebraic integers (and ideals) in algebraic
number elds, so we have a vast amount of machinery available to show that
no factorization (12.5) is possible (under the assumption that each a
j
= 1).
Compare Kummers famous approach toward Fermats Last Theorem (which
led to his creation of algebraic number theory), in which he considered the
equation x
n
+y
n
= z
n
as
n
=1
(x +y) = z
n
.
We are continuing to assume that H = (a
ji
) is an n n circulant
Hadamard matrix. We will denote the eigenvalues of H by
j
= a
0
+a
1
j
+a
2
2j
+ +a
n1
(n1)j
.
12.7 Lemma. For 0 j n 1 we have
[
j
[ =
n.
Thus all the factors appearing on the left-hand side of (12.5) have absolute
value

n.
First proof (naive). Let H
i
denote the ith row of H, and let denote the
usual dot product. Then
j

j
= (a
0
+a
1
j
+ +a
n1
(n1)j
)(a
0
+a
1
j
+ +a
n1
(n1)j
)
= H
1
H
1
+ (H
1
H
2
)
j
+ (H
2
H
3
)
2j
+ + (H
1
H
n
)
(n1)j
.
By the Hadamard property we have H
1
H
1
= n, while H
1
H
k
= 0 for
2 k n, and the proof follows.
Second proof (algebraic). The matrix
1
n
H is a real orthogonal matrix.
By linear algebra, all its eigenvalues have absolute value 1. Hence all eigen-
values
j
of H have absolute value

n.
12.8 Lemma. We have
2 = (1 )
n/2
u, (12.6)
where u is a unit in Z[].
Proof. Put x = 1 in
x
n/2
+ 1 =
n1
j=0
j odd
(x
j
)
to get 2 =
j
(1
j
). Since
1
j
= (1 )(1 + + +
j1
),
it suces to show that 1 + + +
j1
is a unit when j is odd. Let
j
j 1 (modn). Then
(1 + + +
j1
)
1
=
1
1
j
=
1 (
j
)
j
1
j
Z[],
as desired.
12.9 Lemma. We have Z[]/(1 )

= F
2
.
Proof. Let R = Z[]/(1). The integer 2 is not a unit in Z[], e.g., because
1/2 is not an algebraic integer. Thus by Lemma 12.8, 1 is also not a unit.
Hence R ,= 0.
For all j we have
j
= 1 in R since = 1 in R. Hence all elements of R
can be written as ordinary integers m. But 0 = 2 in R by Lemma 12.8, so
the only elements of R are 0 and 1.
12.10 Lemma. For all 0 j n 1 there is an integer h
j
0 such that
a
0
+a
1
j
+a
2
2j
+ +a
n1
(n1)j
= v
j
(1 )
h
j
,
where v
j
is a unit in Z[].
Proof. Since 2 is a multiple of 1 by Lemma 12.8, we have by (12.5) that
n1
j=0
(a
0
+a
1
j
+a
2
2j
+ +a
n1
(n1)j
) = 0
in Z[]/(1 ). Since Z[]/(1 ) is an integral domain by Lemma 12.10,
some factor a
0
+ a
1
j
+ + a
n1
(n1)j
is divisible by 1 . Divide this
factor and the right-hand side of (12.5) by 1 , and iterate the procedure.
We continue to divide a factor of the left-hand side and the right-hand side
by 1 until the right-hand side becomes the unit u. Hence each factor of
the original product has the form v(1 )
h
, where v is a unit.
12.11 Corollary. Either
0
/
1
Z[] or
1
/
0
Z[]. (In fact, both
0
/
1

Z[] and
1
/
0
Z[], as will soon become apparent, but we dont need this
fact here.)
Proof. By the previous lemma, each
j
has the form v
j
(1 )
h
j
. If h
0
h
1
then
0
/
1
Z[]; otherwise
1
/
0
Z[].
We now need to appeal to a result of Kronecker on elements of Z[]
of absolute value one. For completeness we include a proof of this result,
beginning with a lemma.
12.12 Lemma. Let be an algebraic integer such that and all its conjugates
have absolute value one. Then is a root of unity.
Proof. Suppose the contrary. Let deg() = d, i.e., [Q() : Q] = d. Now ,
2
,
3
, . . . are all distinct and hence innitely many of them have the property
that no two are conjugate. Each
j
Q[] and so is the root of a monic
integral polynomial of degree at most d. If
1
,
2
, . . . ,
d
are the conjugates
of , then all the conjugates of
j
are among
j
1
,
j
2
, . . . ,
j
d
. Hence each
j
satises the hypothesis that all its conjugates have absolute value 1 (and
j
is an algebraic integer). Thus the rth elementary symmetric function e
r
in
j
and its conjugates has at most
_
d
r
_
terms, each of absolute value 1, so
[e
r
[
_
d
r
_
. Moreover, e
r
Z since
j
is an algebraic integer. It follows that
there are only nitely many possible polynomials that can be the irreducible
monic polynomials with roots one of the
j
s, contradicting the fact that
there are innitely many
j
s for which no two are conjugate.
12.13 Theorem (Kronecker). Let be any root of unity and Z[] with
[[ = 1. Then is a root of unity.
Proof. Since Z[], we see that is an algebraic integer. We use the basic
fact from Galois theory that the Galois group of the extension eld Q()/Q
is abelian. Let be a conjugate of , so = w() for some automorphism
w of Q(). Apply w to the equation = 1. Since complex conjugation is
an automorphism of Q() it commutes with w, so we obtain
= 1. Hence
all the conjugates of have absolute value one, so is a root of unity by the
previous lemma.
We now have all the ingredients to complete the proof of Theorem 12.5.
Note that we have yet to use the hypothesis that a
i
= 1. By Lemma 12.7
we have
[
1
/
0
[ = [
0
/
1
[ = 1.
Hence by Corollary 12.11 and Theorem 12.13 we have
0
=
r
1
for some r.
Expand
0
and
r
1
uniquely as integer linear combinations of 1, ,
2
, . . . ,
n
2
1
:
0
= a
0
+a
1
+ +a
n1
=
1
=
r
((a
0
a
n/2
) + (a
1
a
n/2+1
) + )
= (a
r
a
n/2+r
) + (a
r+1
a
n/2+r+1
) + .
Equating coecients of
0
yields
n = a
r
a
n/2+r
. Since each a
i
= 1,
we must have n 4, completing the proof.
12.6. P-RECURSIVE FUNCTIONS 221
12.6 P-recursive functions
A function f : N R is called polynomially recursive, or P-recursive for
short, if there exist polynomial P
0
(n), . . . , P
d
(n), with P
d
(n) ,= 0, such that
P
d
(n)f(n +d) +P
d1
(n)f(n +d 1) + +P
0
(n)f(n) = 0 (12.7)
for all n 0.
For instance, the Fibonacci sequence F
n
is P-recursive since F
n+2
F
n+1
F
n
= 0 for all n 0. Here d = 2 and P
2
(n) = P
1
(n) = 1, P
0
(n) = 1. This
situation is quite special since the polynomials P
i
(n) are constants. Another
P-recursive function is f(n) = n!, since f(n + 1) (n + 1)f(n) = 0 for all
n 0.
Let T denote the set of all P-recursive functions f : N R. Our goal in
this section is to prove that T is an R-algebra, that is, for any f, g T and
, R, we have
f +g T, fg T.
There is one techical problem that needs to be dealt with before proceeding
to the proof. We would like to conclude from equation (12.7) that
f(n +d) =
1
P
d
(n)
(P
d1
(n) + +P
0
(n)). (12.8)
This formula, however, is problematical when P
d
(n) = 0. This can happen
only for nitely many n, so equation (12.8) is valid for n suciently large.
Thus we want to deal with functions f(n) only for n suciently large. To
this end, dene f g if f(n) = g(n) for all but nitely many n. Clearly
is an equivalence relation; the equivalence classes are called germs at of
functions f : N R. The germ containing f is denoted [f]. Write ( for the
set of all germs.
12.14 Lemma. (a) If f is P-recursive and f g, then g is P-recursive.
In other words, the property of P-recursiveness is compatible with the
equivalence relation .
(b) Write R
N
for the real vector space of all functions f : N R. Let ,
R and f
1
, f
2
, g
1
, g
2
R
N
. If f
1
f
2
and g
1
g
2
, then f
1
+ g
1

f
2
+ g
2
and f
1
g
1
f
2
g
2
. In other words, linear combinations and
multiplication are compatible with the equivalence relation . Thus
the set ( has the structure of an R-algebra, i.e., a real vector space and
a ring (with obvious compatibility properties such as (f)g = f(g) =
(fg)).
Proof. (a) Suppose that f(n) = g(n) for all n > n
0
. Let (12.7) be the
recurrence satised by f. Multiply both sides by
n
0
j=0
(nj). We then
get a recurrence relation satised by g. Hence g is P-recursive.
(b) This is clear.
Let R[n] denote the ring of real polynomials in n. Let R(n) denote the
quotient eld of R[n], i.e., the eld of all rational functions P(n)/Q(n), where
P, Q R[n]. Suppose that f R
N
and R R(n). Then f(n)R(n) is dened
for n suciently large (i.e., when the denominator of R(n) is nonzero). Thus
we can dene the germ [f(n)R(n)] ( to be the germ of any function that
agrees with f(n)R(n) for n suciently large. It is easy to see that this
denition of scalar multiplication makes ( into a vector space over the eld
R(n). We now come to the key characterization of P-recursive functions (or
their germs).
12.15 Lemma. A function f R
N
is P-recursive if and only if the vector
space 1
f
over R(n) spanned by the germs [f(n)], [f(n+1)], [f(n+2)], . . . is
nite-dimensional.
Proof. Suppose that f(n) satises equation (12.7). Let 1
f
be the vector
space over R(n) spanned by [f(n)], [f(n + 1)], [f(n + 2)], . . . , [f(n +d 1)],
so dim
R(n)
1
f
d. Equation (12.8) shows that [f(n + d)] 1
f
. Substitute
n+1 for n in equation (12.8). We get that [f(n+d +1)] is in the span (over
R(n)) of [f(n + 1)], [f(n + 2)], . . . , [f(n + d)]. Since these d germs are all
in 1
f
, we get that [f(n + d + 1)] 1
. Continuing in this way, we get by

induction on k that f(n + d + k) 1
for all k 0, so 1
f
= 1
f
. Thus 1
f
is
nite-dimensional.
Conversely, assume that dim
R(n)
1
f
< . Then for some d, the germs
[f(n)], [f(n+1)], . . . , [f(n+d)] are linearly dependent over R(n). Write down
this linear dependence relation and clear denominators to get a recurrence
(12.7) satised by f. Hence f is P-recursive.
We now have all the ingredients necessary for the main result of this
section.
12.6. P-RECURSIVE FUNCTIONS 223
12.16 Theorem. Let f, g T and , R.
(a) f +g T
(b) fg T
Proof. (a) By Lemma 12.15 it suces to show that dim1
f+g
< . Now
by denition, the direct sum 1
f
1
g
is the vector space consisting of
all linear combinations [u] + [v], where [u] 1
f
and [v] 1
g
and
, R(n). In particular, 1
f
1
g
contains all the germs [f(n+k)] +
[g(n +k] = [f(n +k)] +g(n +k)], k 0. Hence
1
f+g
1
f
1
g
.
Now if V and W are subspaces of some vector space, then V W is
spanned by the union of a basis for V and basis for W. In particular, if
V and W are nite-dimensional, then dim(V W) dimV + dimW.
Hence
dim1
f+g
dim(1
f
1
g
) dim1
f
+ dim1
g
< ,
as was to be proved.
(b) The proof is analogous to (a), except that instead of the direct sum
V
K
W we need the tensor product V W over the eld K. Recall
from linear algebra that V
K
W may be thought of (somewhat naively)
as the vector space spanned by all symbols v w, where v V and
w W, subject to the conditions
(v
1
+v
2
) w = v
1
w +v
2
w
v (w
1
+w
2
) = v w
1
+v w
2
.
v w = v w = (v w),
where is a scalar. A standard consequence is that if V has the basis
v
1
, . . . , v
m
and W has the basis w
1
, . . . , w
n
, then V
K
W has the
basis v
i
w
j
, for 1 i m and 1 j n. In particular,
dimV
K
W = (dimv)(dimW).
Now by the basic universality property of tensor products, there is a
unique linear transformation : 1
f

R(n)
1
g
( satisfying
[f(n +i)] g[(n +j)]
[f(n +i)g(n +j)].

The image of contains all germs [f(n+i)g(n+i)], so 1
fg
image().
Thus
dim1
fg
dim(1
f

R(n)
1
g
) = (dim1
f
)(dim1
g
) < ,
and the proof follows.
The Oddtown theorem is due to E. R. Berlekamp [7]. Theorem 12.2 on
decomposing E(K
n
) is a result of R. Graham and H. Pollak [34][35].
Sutners original proof of the odd neighborhood theorem (Theorem 12.4)
appears in [86]. We have given a simplied version due to Y. Caro.
The circulant Hadamard matrix conjecture was rst mentioned in print
by H. J. Ryser [74, p. 134], though its precise origin is obscure. The work of
Turyn mentioned in the text appears in [90][91]. Some more recent progress
is due to B. Schmidt [76].
While P-recursive functions and their cousins the D-nite series of Exer-
cise 12.15 were known to 19th century analysts, the rst systematic treatment
of them did not appear until the paper of Stanley [80] in 1980, which includes
a statement and proof of Theorem 12.16.
Bibliography
[1] T. van Aardenne-Ehrenfest and N. G. de Bruijn, Circuits and trees in
oriented linear graphs, Simon Stevin (= Bull. Belgian Math. Soc.) 28
(1951), 203217.
[2] I. Anderson, Combinatorics of Finite Sets, Oxford University Press,
Oxford-New York, 1987; corrected republication by Dover, New York,
2002.
[3] G. E. Andrews, The Theory of Partitions, Addison-Wesley, Reading,
Mass., 1976.
[4] G. E. Andrews and K. Eriksson, Integer Partitions, Cambridge Univer-
sity Press, Cambridge, 2004.
[5] E. A. Beem, Craige and Irene Schensted dont have a car in the world,
Maine Times, March 12, 1982, pp. 2021.
[6] E. A. Bender and D. E. Knuth, Enumeration of plane partitions, J.
Combinatorial Theory 13 (1972), 4054.
[7] E. R. Berlekamp, On subsets with intersections of even cardinality,
Canad. Math. Bull. 12 (1969), 363366.
[8] E. D. Bolker, The nite Radon transform, Integral geometry (Brunswick,
Maine, 1984), Contemp. Math. 63 Amer

Math
Soc., Providence, RI, 1987,

pp. 2750.
[9] C. W. Borchardt, Ueber eine der Interpolation entsprechende Darstel-
lung der EliminationsResultante, J. reine angew. Math. 57 (1860), 111
121.
227
228 BIBLIOGRAPHY
[10] F. Brenti, Log-concave and unimodal sequences in algebra, combina-
torics, and geometry: an update, in Jerusalem Combinatorics 93, Con-
temp. Math. 178, Amer. Math. Soc., Providence, RI, 1994.
[11] A. E. Brouwer and W. H. Haemers, Spectra of Graphs, Springer, New
York, 2012.
[12] N. G. de Bruijn, Polyas theory of counting, in Applied Combinatorial
Mathematics (E. F. Beckenbach, ed.), Wiley, New York, 1964; reprinted
by Krieger, Malabar, Florida, 1981.
[13] W. Burnside, Theory of Groups of Finite Order, Cambridge University
Press, 1897.
[14] W. Burnside, Theory of Groups of Finite Order, second edition, Cam-
bridge University Press, 1911; reprinted by Dover, New York, 1955.
[15] Y. Caro, Simple proofs to three parity theorems, Ars Combin. 42 (1996),
175180.
[16] A. L. Cauchy, Memoire sur diverses proprietes remarquables des sub-
stitutions regulaires ou irr`egulaires, et des syst`emes de substitutions
conjuguees (suite), C. R. Acad. Sci. Paris 21 (1845), 972987; Oeuvres,
ser. 1, vol. 9, pp. 371387.
[17] A. Cayley, Note sur une formule pour la reversion des series, J. reine
ang. Math. (= Crelles Journal ) 52 (1856), 276284.
[18] A. Cayley, A theorem on trees, Quart. J. Math. 23 (1889), 376378;
Collected Papers, vol. 13, Cambridge University Press, 1897, pp. 2628.
[19] D. M. Cvetkovic, M. Doob, and H. Sachs, Spectra of Graphs: Theory
and Applications, 3rd ed., Wiley-VCH, 1998.
[20] D. M. Cvetkovic, P. Rowlinson, and S. Simic, An Introduction to the
Theory of Graph Spectra, London Mathematical Society, Student Texts
75, Cambridge University Press, Cambridge, 2010.
[21] M. R. DeDeo and E. Velasquez, The Radon transform on Z
n
k
, SIAM J.
Discrete Math. 18 (2004/05), 472-478 (electronic).
[22] P. Diaconis and R. L. Graham,
BIBLIOGRAPHY 229
[23] E. B. Dynkin, Some properties of the weight system of a linear represen-
tation of a semisimple Lie group (in Russian), Dokl. Akad. Nauk SSSR
(N.S) 71 (1950), 221224.
[24] E. B. Dynkin, The maximal subgroups of the classical groups, in Amer.
Math. Soc. Transl., Ser. 2 6 (1957), 245378. Translated from Trudy
Moskov. Mat. Obsc. 1, 39166.
[25] K. Engel, Sperner Theory, Encyclopedia of Mathematics and Its Appli-
cations, no. 65, Cambridge University Press, Cambridge, 1997.
[26] P. Fishburn, Interval Orders and Interval Graphs: Study of Partially
Ordered Sets, John Wiley & Sons, ??, 1985. ??
[27] S. Fomin, Duality of graded graphs, J. Algebraic Combinatorics 3
(1994), 357404.
[28] S. Fomin, Schensted algorithms for dual graded graphs, J. Algebraic
Combinatorics 4 (1995), 545.
[29] J. S. Frame, G. de B. Robinson, and R. M. Thrall, The hook graphs of
S
n
, Canad. J. Math. 6 (1954), 316324.
[30] D. S. Franzblau and D. Zeilberger, A bijective proof of the hook-length
formula, J. Algorithms 3 (1982), 317343.
[31] F. G. Frobenius,

Uber die Congruenz nach einem aus zwei endlichen
Gruppen gebildeten Doppelmodul, J. Reine Ang. Math. (= Crelles J.)
101 (1887), 273299; reprinted in Gesammelte Abhandlungen, vol. 2,
Springer-Verlag, Heidelberg, 1988, pp. 304330.
[32] F. G. Frobenius,

Uber die Charaktere der symmetrischen Gruppe,
Sitzungsber. Kon. Preuss. Akad. Wissen. Berlin (1900), 516534;
Gesammelte Abh. III (J.-P. Serre, ed.), Springer-Verlag, Berlin (1968),
pp. 148166.
[33] I. J. Good, Normally recurring decimals, J. London Math. Soc. 21
(1947), 167169.
[34] R. L. Graham and H. O. Pollak, On the addressing problem for loop
switching, Bell System Tech. J. 50 (1971), 24952519.
230 BIBLIOGRAPHY
[35] R. L. Graham and H. O. Pollak, On embedding graphs in squashed
cubes, in Lecture Notes in Mathematics 303, Springer Verlag, New York-
Berlin-Heidelberg, 1973, pp. 99-110.
[36] C. Greene, A. Nijenhuis, and H. S. Wilf, A probabilistic proof of a
formula for the number of Young tableaux of a given shape, Advances
in Math. 31 (1979), 104109.
[37] J. I. Hall, E. M. Palmer, and R. W. Robinson, Redelds lost paper in
a modern context, J. Graph Theory 8 (1984), 225240.
[38] F. Harary and E. M. Palmer, Graphical Enumeration, ??
[39] F. Harary and R. W. Robinson, The rediscovery of Redelds papers, J.
Graph Theory 8 (1984), 191192.
[40] G. H. Hardy, J. E. Littlewood, and G. Polya, Inequalities, second ed.,
Cambridge University Press, Cambridge, England, 1952.o
[41] L. H. Harper, Morphisms for the strong Sperner property of Stanley and
Griggs, Linear and Multilinear Algebra 16 (1984), 323337.
[42] A. P. Hillman and R. M. Grassl, Reverse plane partitions and tableaux
hook numbers, J. Combinatorial Theory (A) 21 (1976), 216221.
[43] J. W. B. Hughes, Lie algebraic proofs of some theorems on partitions,
in Number Theory and Algebra (H. Zassenhaus, ed.), Academic Press,
New York, 1977, pp. 135155.
[44] A. Hurwitz,

Uber die Anzahl der Riemannschen Flachen mit gegebenen
Verweigungspunkten, Math. Ann. 55 (1902), 5366.
[45] A. Joyal, Une theorie combinatoire des series formelles, Advances in
Math. 42 (1981), 182.
[46] C. Krattenthaler, Bijective proofs of the hook formulas for the number
of standard Young tableaux, ordinary and shifted, Electronic J. Combi-
natorics 2 (1995), R13, 9pp.
[47] D. E. Knuth, Permutations, matrices, and generalized Young tableaux,
Pacic J. Math. 34 (1970), 709727.
BIBLIOGRAPHY 231
[48] I. Krasikov and Y. Roditty, Balance equations for reconstruction prob-
lems, Arch. Math. (Basel) 48 (1987), 458464.
[49] J. P. S. Kung, Radon transforms in combinatorics and lattice theory,
Combinatorics and ordered sets (Arcata, Calif., 1985), Contemp. Math.
57, Amer

Math. Soc., Providence, RI, 1986, pp. 3374.
[50] M. A. A. van Leeuwen, The Robinson-Schensted and Sch utzenberger
algorithms, Part 1: new combinatorial proofs, Preprint no. AM-R9208
1992, Centrum voor Wiskunde en Informatica, 1992.
[51] E. K. Lloyd, J. Howard Redeld: 18791944, J. Graph Theory 8 (1984),
195203.
[52] L. Lovasz, A note on the line reconstruction problem, J. Combin. Theory
Ser. B 13 (1972), 309310.
[53] D. Lubell, A short proof of Sperners lemma, J. Combinatorial Theory
1 (1966), 299.
[54] P. A. MacMahon, Memoir on the theory of the partitions of numbers
Part I, Phil. Trans. Royal Soc. London (A) 187 (1897), 619673;
Collected Works, vol. 1 (G. E. Andrews, ed.), M.I.T. Press, Cambridge,
Massachusetts, 1978, pp. 10261080.
[55] P. A. MacMahon, Memoir on the theory of the partitions of numbers
Part IV, Phil. Trans. Royal Soc. London (A) 209 (1909), 153175;
Collected Works, vol. 1 (G. E. Andrews, ed.), M.I.T. Press, Cambridge,
Massachusetts, 1978, pp. 12921314.
[56] P. A. MacMahon, Combinatory Analysis, vols. 1 and 2, Cambridge
University Press, Cambridge, 1915, 1916; reprinted in one volume by
Chelsea, New York, 1960.
[57] J. W. Moon, Counting Labelled Trees, Canadian Math. Monographs,
No. 1, Canadian Mathematical Congress, 1970.
[58] V. M uller, The edge reconstruction hypothesis is true for graphs with
more than nlog
n
n edges, J. Combin. Theory Ser. B 22 (1977), 281283.
[59] P. M. Neumann, A lemma that is not Burnsides, Math. Sci. 4 (1979),
133141.
232 BIBLIOGRAPHY
[60] J.-C. Novelli, I. Pak, and A. V. Stoyanovskii, A new proof of the hook-
length formula, Discrete Math. and Theoretical Computer Sci. 1 (1997),
053067.
[61] K. M. OHara, Unimodality of Gaussian coecients: a constructive
proof, J. Combin. Theory Ser. A 53 (1990), 2952.
[62] J. Pitman, Coalescent random forests, J. Combinatorial Theory Ser. A
85 (1999), 165193,
[63] G. Polya, Kombinatorische Anzahlbestimmungen f ur Gruppen,
Graphen, und chemische Verbindungen, Acta Math. 68 (1937), 145254.
[64] G. Polya and R. C. Read, Combinatorial Enumeration of Groups,
Graphs, and Chemical Compounds, Springer-Verlag, New York/Berlin,
1987.
[65] M. Pouzet, Application dune propriete combinatoire des parties dun
ensemble aux groupes et aux relations, Math. Zeit. 150 (1976), 117134.
[66] M. Pouzet and I. G. Rosenberg, Sperner properties for groups and rela-
tions, European J. Combin. 7 (1986), 349370.
[67] R. A. Proctor, ??, Amer. Math. Monthly 89 (1982), 721634.
[68] H. Pr ufer, Neuer Beweis eines Satzes uber Permutationen, Arch. Math.
Phys. 27 (1918), 742744.
[69] J. K. A. Radon,
[70] J. H. Redeld, The theory of group reduced distributions, American J.
Math. 49 (1927), 433455.
[71] J. H. Redeld, Enumeration by frame group and range groups, J. Graph
Theory 8 (1984), 205223.
[72] J. B. Remmel, Bijective proofs of formulae for the number of standard
Young tableaux, Linear and Multilinear Algebra 11 (1982), 45100.
[73] G. de B. Robinson, On the representations of S
n
, Amer. J. Math. 60
(1938), 745760.
BIBLIOGRAPHY 233
[74] H. J. Ryser,
[75] C. E. Schensted, Longest increasing and decreasing subsequences, Cana-
dian J. Math. 13 (1961), 179191.
[76] B. Schmidt, Cyclotomic integers and nite geometry, Journal Amer.
Math. Soc. 12 (1999), 929952.
[77] C. A. B. Smith and W. T. Tutte, On unicursal paths in a network of
degree 4, Amer. Math. Monthly 48 (1941), 233237.
[78] E. Sperner, Ein Satz uber Untermengen einer endlichen Menge, Math.
Zeit. 27(1) (1928), 544548.
[79] R. Stanley, Weyl groups, the hard Lefschetz theorem, and the Sperner
property, SIAM J. Algebraic and Discrete Methods 1 (1980), 168184.
[80] R. Stanley, Dierentiably nite power series, European J. Combinatorics
1 (1980), 175188.
[81] R. Stanley, Quotients of Peck posets, Order 1 (1984), 2934.
[82] R. Stanley, Dierential posets, J. Amer. Math. Soc. 1 (1988), 919961.
[83] R. Stanley, Unimodal and log-concave sequences in algebra, combina-
torics, and geometry, in Graph Theory and Its Applications: East and
West, Ann. New York Acad. Sci., vol. 576, 1989, pp. 500535.
[84] R. Stanley, Variations on dierential posets, in Invariant Theory and
Tableaux (D. Stanton, ed.), The IMA Volumes in Mathematics and Its
Applications, vol. 19, Springer-Verlag, New York, 1990, pp. 145165.
[85] R. Stanley, Enumerative Combinatorics, vol. 1, 2nd ed., Cambridge Uni-
versity Press, Cambridge, 2012.
[86] K. Sutner, Linear cellular automata and the Garden-of-Eden, Math.
Intelligencer 11 (1989), 4953.
[87] J. J. Sylvester, On the change of systems of independent variables,
Quart. J. Math. 1 (1857), 4256. Collected Mathematical Papers, vol.
2, Cambridge, 1908, pp. 6585.
234 BIBLIOGRAPHY
[88] J. J. Sylvester,
[89] W. T. Trotter, Combinatorics and Partially Ordered Sets: Dimension
Theory, Johns Hopkins Studies in the Mathematical Sciences 6, Johns
Hopkins University Press, Baltimore, 1992.
[90] R. J. Turyn, Character sums and dierence sets, Pacic J. Math. 15
(1965), 319346.
[91] R. J. Turyn, Sequences with small correlation, in Error Correcting Codes
(H. B. Mann, ed.), Wiley, New York, 1969, pp. 195228.
[92] W. T. Tutte, The dissection of equilateral triangles into equilateral tri-
angles, Proc. Cambridge Phil. Soc. 44 (1948), 463482.
[93] E. M. Wright, Burnsides lemma: a historical note, J. Combinatorial
Theory (B) 30 (1981), 8990.
[94] A. Young, Qualitative substitutional analysis (third paper), Proc. Lon-
don Math. Soc. (2) 28 (1927), 255292.
[95] D. Zeilberger, Kathy OHaras constructive proof of the unimodality of
the Gaussian polynomials, Amer. Math. Monthly 96 (1989), 590602
Index
4483130665195087, 30
van Aardenne-Ehrenfest, T., 178
acts on (by a group), 45
adjacency matrix, 5
adjacent, 5
Anderson, Ian, 40
Andrews, George W. Eyre, 78
antichain, 31
antisymmetry, 29
automorphism
of a poset, 46
automorphism group
of a poset, 47
balanced digraph, 170
balanced incomplete block design, 212
Bender, Edward Anton, 137
Berlekamp, Elwyn Ralph, 224
Bernardi, Olivier, 158
BEST theorem, 178
BIBD, 212
Binet-Cauchy theorem, 151
block (of a block design), 212
block design, 212
Bolker, Ethan David, 21
boolean algebra, 29
Borchardt, Carl Wilhelm, 163
Bose, ??, 212
Brenti, Francesco, 57
de Bruijn, Nicolaas Govert, 106, 178
bump
in RSK algorithm for CSPP, 128
in RSK algorithm for SYT, 123
Burnsides lemma, 87, 106
Burnside, William, 106
Caro, Yair, 224
Cauchy, Augustin Louis, 106
Cauchy-Binet theorem, 151
Cauchy-Frobenius lemma, 87, 106
Cayley, Arthur, 163
chain, 30
characteristic polynomial, 156
characteristic vector (of a set), 209
circulant (matrix), 215
closed walk, 8
Collatz, Lothar, 11
coloring, 82
column-strict plane partition, 127
commutative diagram, 51
complete graph, 8
complexity (of a graph), 150
conjugate (partition), 64
connected graph, 149
covers (in a poset), 30
CSPP, 127
cube (graph), 15
Cvetkovic, Dragos M., 11
cycle index polynomial, 90
cycle indicator, 106
of a group of permutations, 90
235
236 INDEX
of a permutation, 90
DeDeo, Michelle Rose, 21
degree (of a vertex), 25
Diaconis, Persi, 21
diagram (of a partition), 62
digraph, 169
dihedral necklace, 94
direct sum (of vector spaces), 223
directed graph, 169
Doob, Michael, 11
down (linear transformation), 35
Dynkin, Eugene (Evgenii) Borisovitsh,
78
edge reconstruction conjecture, 53
edge set, 5
eigenvalues of a graph, 7
Engel, Konrad, 39
equivalent
colorings, 82, 83
Erdos-Moser conjecture, 77
weak, 77
Eriksson, Kimmo, 78
Eulerian cycle (in a graph), 169
Eulerian digraph, 170
Eulerian graph, 169
Eulerian tour
in a digraph, 169
in a grpah, 169
faithful action, 45
Ferrers diagram, 62
nal vertex (of an edge)
in a digraph, 169
in an orientation, 153
Fishburn, Peter, 39
Fisher, ??, 212
Fomin, Sergey Vladimirovich, 136
forest, 161
Forsyth, Andrew Russell, 137
Frame, James Sutherland, 135
Frame-Robinson-Thrall, 113
Franzblau, Deborah Sharon, 136
Frobenius, Ferdinand Georg, 106, 135
Fundamental Theorem of Algebra, 55
Gauss lemma, 217
Gaussian coecient, 66
generalized ballot sequence, 135
generating function, 11, 86
G-equivalent, 46
colorings, 83
germ, 221
Good, Irving John, 178
graded poset, 31
Graham, Ronald Lewis, 21, 224
graph, 5
Grassl, Richard, 135
Greene, Curtis, 135
group reduction function, 106
Hadamard matrix, 215
Hamming weight, 19
Harary, Frank, 106
Hardy, Godfrey Harold, 57
Harper, Lawrence Hueston, 56
Hasse diagram, 30
Hasse walk, 112
Hillman, Abraham, 135
hook length formula, 113, 135
Hughes, J. W. B., 78
Hurwitz, Adolf, 106
incidence matrix, 153
Oddtown, 209
incident, 5
INDEX 237
indegree (of a vertex), 170
induced subgraph, 215
initial vertex (of an edge)
in a digraph, 169
in an orientation, 153
internal zero, 55
inverse bump (in RSK algorithm), 125
isomorphic
graphs, 52
posets, 30
isomorphism class (of simple graphs),
52
Joyal, Andre, 159, 163
Kirchho, Gustav, 163
Knuth, Donald Ervin, 119, 136, 137
Krasikov, Ilia, 57
Krattenthaler, Christian Friedrich, 136
Kronecker, ??, 219
Kung, Joseph PeeSin, 21
laplacian matrix, 153
lattice, 61
lattice permutation, 135
van Leeuwen, Marc A. A., 136
length
of a chain, 30
of a necklace, 93
of a walk, 6
level (of a ranked poset), 31
Littlewood, Dudley Ernest, 136
Littlewood, John Edensor, 57
Littlewood-Richardson rule, 136
log-concave, 55
logarithmically concave, 55
loop, 5
Lovasz, Laszlo, 57
Lubell, David, 32, 37, 39
MacMahon, Percy Alexander, 135137
Markov chain, 25
Matrix-Tree Theorem, 156, 163
maximal chain, 31
Moon, John W., 163
M uller, Vladimr, 57
multiple edge, 5
multiset, 5
necklace, 93
neighborhood (of a vertex), 214
Newton, Isaac, 55, 57
Nijenhuis, Albert, 135
N-matrix, 128
no internal zero, 55
nonuniform Fisher inequality, 212
Novelli, Jean-Christophe, 136
Oddtown, 209
OHara, Kathleen Marie, 71, 78
orbit, 46
order-matching, 32
explict for B
n
, 38
order-raising operator, 33
orientation (of a graph), 153
orthogonal Lie algebra, 78
orthogonal subspace, 214
outdegree (of a vertex), 170
Pak, Igor M., 136
Palmer, Edgar Milan, 106
part (of a partition of n), 61
partially ordered set, 29
partition
of a set X, 46
of an integer n, 61
Pascals triangle, 67
q-analogue, 67
Pitman, James William, 161, 163
238 INDEX
plane partition, 126
history of, 136
planted forest, 161
Pollak, Henry Otto, 224
Polya, George (Gyorgy), 57, 81, 106
Polya theory, 81
polynomially recursive function, 221
poset, 29
positive denite, 213
positive semidenite, 36, 213
Pouzet, Maurice Andre, 40, 56
P-recursive function, 221
Proctor, Robert Alan, 78
Pr ufer sequence, 159
Pr ufer, Ernst Paul Heinz, 159, 163
q-binomial coecient, 66
quantum order-matching, 34
quotient poset, 47
Radon transform, 17
Radon, Johann Karl August, 21
rank
of a boolean algebra, 29
of a graded poset, 31
of a poset element, 31
rank-symmetric, 31
rank-unimodal, 31
Redeld, John Howard, 81, 106
reduced incidence matrix, 154
reexivity, 29
regular graph, 26
Remmel, Jerey Brian, 136
Robinson, Gilbert de Beauregard, 119,
135, 136
Robinson-Schensted correspondence,
119
Roditty, Yehuda, 57
Rolles theorem, 56
root (of a tree), 161
Rosenberg, Ivo G., 40, 56
row insertion, 123, 128
Rowlinson, Peter, 11
RSK algorithm, 119, 136
Ryser, Herbert John, 224
Sachs, Horst, 11
Schensted, Craige Eugene, 119, 136
Schmidt, Bernard, 224
Sch utzenberger, Marcel-Paul, 136
semidenite, 36, 213
shape (of a CSPP), 127
shifted Ferrers diagram, 131
Simic, Slobodan, 11
simple graph, 5
Sinogowitz, Ulrich, 11
Smith, Cedric Austen Bardell, 178
solid partition, 134
spanning subgraph, 52, 150
spectral graph theory, 11
Sperner poset, 32
Sperner property, 32
Sperners theorem, 32
Sperner, Emanuel, 39
stabilizer, 49
standard Young tableau, 112
Stanley, Richard Peter, 39, 40, 57, 78,
136, 224
von Staudt, Karl Georg Christian, 163
Stirling number, signless of the rst
kind, 96
Stoyanovskii, Alexander V., 136
strongly log-concave, 55
Sutner, Klaus, 213, 224
Sylvester, James Joseph, 71, 78, 163
symmetric chain decomposition, 39
INDEX 239
symmetric function, 86
symmetric sequence, 31
SYT, 112
tensor product (of vector spaces), 223
Thrall, Robert McDowell, 135
tour (in a digraph), 169
transitivity, 29
tree, 149
Trotter, William Thomas, Jr., 39
Turyn, Richard Joseph, 216, 224
Tutte, William Thomas, 178
two-line array, 128
type
of a Hasse walk, 112
unimodal sequence, 31
up (linear transformation), 33
valid -word, 114
Velasquez, Elinor Laura, 21
venkataramana, Praveen, 163
vertex reconstruction conjecture, 53
vertex set, 5
walk, 6
weight (of a binary vector), 19
Wilf, Herbert Saul, 135
wreath product, 69
Young diagram, 62
Young, Alfred, 61, 135
Youngs lattice, 61
Zeilberger, Doron, 78, 136
Zyklus, 90

Topics in Algebraic Combinatorics: Preliminary Version of 19 April 2012

Uploaded by

Copyright:

Available Formats

Topics in Algebraic Combinatorics: Preliminary Version of 19 April 2012

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Topics in Algebraic Combinatorics: Preliminary Version of 19 April 2012

Uploaded by

Copyright:

Available Formats

1

TOPICS IN ALGEBRAIC COMBINATORICS

is equal to the number of walks from v

and sum on all 1. The geometric series we obtain will

f is called the (discrete or nite) Radon transform of f (on

: 1 1. It is easy to see that

are the functions

. Note also that the eigenvectors

are independent of ; only the

] denote the matrix of the linear transformation

: 1 1 with respect to the basis B

has the same matrix

(or equivalently, the eigenvector E

does indeed reduce to 0 when is odd.

There is another nice way to show directly that B

are two orbits, then

, the only way c

> x for some x o. But this is just what it means

are isomorphic. Just as in the proof of Theorem 5.8, when

(x) (the derivative of P(x))

(x) has a zero satisfying < < . It follows

(x) has at least d 1 real zeros. Since deg P

(x) has exactly d 1 real zeros and no other zeros.

= (4, 2, 2, 1), with diagram

of in an m n rectangle R, i.e., all the squares of R except for . (Note

then we obtain the diagram of a

[ = mn. This correspondence between

of shape is indeed in the orbit S. For by

) = is a bijection (one-to-one and onto). We claim that

be the unique Young diagrams in o and o

, respectively. Then there

. So assume there exist D o and D

. Since each row of D is contained in a row of D

has at least j rows of size at least

) are incomparable in L(k, n k),

M(n). The following

are incomparable in M(n), i.e., neither

, and that two colorings are considered the same

rotation. The rst naive assumption

. In fact, these are

= (a)(b)(c)(d), (a, b)(c)(d), (a)(b)(c, d), (a, b)(c, d).

7.4 Example. How many inequivalent colorings of the vertices of a regular

(n) of n-colored necklaces of length

(1) = 1, putting n = 1 in (7.5) yields the famous identity

, the group of all permutations of 1, 2, . . . , =

will just be the sum of all monomials of

(1, 1, . . . , 1) (n 1s in all), there follows from (7.6)

, construct a permutation f() X

itself, we have the following result. Write

in the rst and last terms of the above string of

explicitly as we did in the above tables,

) goes back to Frobenius (see [32, bottom of p. 152 of

denote the number of SYT of shape , so for instance f

have many interesting properties; for instance, there is a

[why?]. In a similar fashion,

be the least integer for which d

, and apply both sides

[why?]. The right-hand side, on the other hand

, for n, are the degrees of the