Training camp: Combinatorics

Nanyang Technological University

9 Jun 2016

1 Linear algebra methods

1.1 Introduction
The small town Eventown has precisely 2016 inhabitants, half of which are male. They are quite a
social bunch and like to form clubs, however, they have to obey the rules for club forming set by the
Major of Eventown, which are

1 A club must have an even number of members.

2 Any two distinct clubs must share an even number of members.

Under these rules, many clubs can be formed. Any idea how many?
The towners felt great social pressure to form all possible clubs, and to be active in them. After
many socially very taxing years, the inhabitants (still the same number, since they had been too
busy to make children) revolted. They wanted FEWER clubs! After a short trial, the major was
executed, and the town was renamed Oddtown. Subsequently, the club-forming rules were slightly
amended, in particular, rule 1 was replaced by the rule

10 A club must have an odd number of members.

This new rules made the social lives of the remaining 2015 inhabitants of the town now named
Oddtown much less stressful. Do you know how less stressful exactly, that is, can you say how many
clubs could now be formed?
In due time, the new club rules were boring everybody to dead, and so new rules were sought to
form clubs while keeping social life on an acceptable level. Another drawback of the old club rules
was that different clubs often had no idea of each others activities. So after some experimentation,
the following rule was adopted.

(a) Any two distinct clubs must share precisely two members;

Can you now guarantee the inhabitants that under this new club-forming rule, their lives will not
get more involved as before?
In the next sections, we will develop some linear algebraic tools to solve combinatorial problems
like those above.

1
Basics on linear algebra
Let F be a field, for example Q, R, C, or a finite field such as Fp , the field of integers molulo a
prime p. We write Fn to denote the vector space of dimension n over F; we denote the set of n × m
matrices with entries from F by MF (n, m).
For x, y ∈ Fn , their inner product (x, y) is defined as
n
X
>
(x, y) = x y = xi y i .
i=1

If V is a subspace of Fn , then the orthogonal complement V ⊥ is defined as

V ⊥ = {x ∈ Fn | (x, v) = 0 for all v ∈ V }.
If V is a subspace of a vector space U , then
V ⊥U = {x ∈ U | (x, v) = 0 for all v ∈ V }.
If V, W are two subspaces of Fn , then
V + W = {v + w | v ∈ V, w ∈ W }.
We will write jn to denote the all-one vector (1, 1, . . . , 1)> of size n; similarly, we will write Jn or
Jn,m to denote the all-one matrix of size n × n or size n × m; if the dimensions are clear from the
context, then we simply write j and J. The all-zero vector (0, . . . , 0)> will be denoted by 0; similarly,
an n × n or n × m all-zero matrix will be denoted by On or On,m , respectively, or simply by O if the
dimensions are clear from the context.
In what follows, we list some useful properties of matrices and vector spaces that are valid for all
fields.
• The dimension dim(V ) of a vector space V is the largest number n for which there is a collection
of n (linearly) independent vectors in V . A set of dim(V ) independent vectors in V is called a
basis of V . Any set of n + 1 vectors v0 , . . . , vn in an n-dimensonal vector space over a field F
is dependent, that is, there are λ0 , . . . , λn in F, not all zero, such that λ0 v0 + · · · + λn vn = 0. If
there are k independent vectors in an n-dimensional vector space, then k ≤ n.
Note that the dimension cannot go down by enlarging the field (see the later item on determi-
nants).
• If M is an n × m matrix over some field F, then the rowspace rowspace(M ) of M , the subspace
of Fm spanned by the rows of M , and the columnspace columnspace(M ) of M , the subspace
of Fn spanned by the columns of F, have the same dimension. This common dimension is
called the rank of M , and is denoted by rank(M ). In particular, rank(M > ) = rank(M ) and
rank(M ) ≤ min(m, n).

2
• The null space nullspace(M ) of M is the collection of all x ∈ Fm such that M x = 0. Obviously,
nullspace(M ) = rowspace(M )⊥ .

• If V is a subspace of a vector space U , then dim(V ) + dim(V ⊥ ) = dim(U ). In particular, if

V ⊆ Fn , then dim(V ) + dim(V ⊥ ) = n. But watch out: in fields like Q, R, and C, we know
that since x> x = 0 implies x = 0, we have V ∩ V ⊥ = {0}. But this need not be true for
other fields; in particular, in finite fields this is false. For example, over a field of characteristic
p such as Fp = Zp , we have for v = j = (1, 1, . . . , 1)> that v > v = (v, v) = p = 0. For
V = {0, (1, 1)> } ⊆ F22 , we have V > = V , both of dimension 1.

• Since nullspace(AB) ⊇ nullspace(B), we have that rank(AB) ≤ min(rank(A), rank(B)).

• It is easily seen that columnspace(A + B) ⊆ columnspace(A) + columnspace(B); consequently,

rank(A + B) ≤ rank(A) + rank(B).

• An m × n matrix M is invertible if and only if m = n = rank(M ). A square matrix M over a

field F is invertible if and only if M x = 0 implies that x = 0.

• Let Sn denote the collection of all permutations of {1, . . . , n}. The sign of a permutation π is
1 or -1 depending on whether π can be written as a product of an even or an odd number of
transpositions. An n × n matrix M over a field F is invertible if and only if the determinant
X n
Y
det(M ) = sign(π) Mi,π(i)
π∈Sn i=1

of A is nonzero. (The permanent perm(M ) is the expression obtained by leaving out the signs;
so over F2 , the determinant and the permanent are equal.)
The determinant det(v1 , . . . , vn ) of a sequence of n vectors in Fn is the determinant of the matrix
V =P [v1 · · · vn ]. Note that the determinant is linear in each of its arguments. In particular, if
vi = j λi,j vi,j for i = 1, . . . , n, then
X Y 
det(v1 , . . . , vn ) = λk,jk det(v1,j1 , . . . , vn,jn ).
j1 ,...,jn k

• If M is a real m × n matrix , then det(M > M ) ≥ 0, with equality iff rank(M ) < n. A matrix
of the form M > M is called a Gram (or Gramian) matrix.

• Sometimes it can be profitable to consider polynomials or functions as vectors in a suitable

vector space. For example, we may consider the collection of all polynomials of degree at

3
 most n with coefficients in some field F as an (n + 1)-dimensional vector space over F with
basis 1, x, . . . , xn . For another example, suppose that we have a set of maps fi : X 7→ F for
1 ≤ i ≤ n and points x1 , . . . , xn in X. If fi (xi ) 6= 0 and fi (xj ) = 0 for i 6= j, then f1 , . . . , fn are
independent.
• We say that a set of vectors V ⊆ Fn is in general position if any n of the vectors of V are
independent. (In that case, the null space of the matrix with as columns the vectors from V is
said to be an MDS code.) If |F| ≥ n, then the |F| vectors

v(a) = (1, a, a2 , . . . , an−1 )>

are in general position. This follows from the fact that the Vandermonde determinant
Y
det(v(a1 ), . . . , v(an )) = (aj − ai ) 6= 0.
i<j

Incidence and adjacency matrices

Let P be a finite set ( the elements of which will be referred to as points), and let B be a collection of
subsets of P, referred to as blocks. For each B ∈ B, the points in B are the points incident to B. We
say that two points are adjacent if they are contained together in some block. In such a situation,
we can form two matrices describing the relations between points and blocks. The incidence matrix
is the |P| × |L| matrix M defined by

1, if P ∈ B;
M (P, B) =
0, otherwise.
In other words, the columns of the incidence matrix M are the characteristic vectors or incidence
vectors of the blocks.
The adjacency matrix is the |P| × |P| matrix A defined by letting A(P, Q) to be the number of
blocks B ∈ B for which P, Q ∈ B. We note that

A = M M >,

hence the determinant of the adjacency matrix (considered as a real matrix) is nonnegative. Another
interesting matrix to consder is the block intersection matrix N ; here for any two blocks B, B 0 , the
corresponding entry is N (B, B 0 ) = |B ∩ B 0 |, the size of the intersection of the blocks B and B 0 .
(In fact, the block intersection matrix is the adjacency matrix of the dual configuration obtained by
interchanging the roles of “points” and “blocks”.)
Sometimes, the notion of a “block” is not present in the problem statement and has to be devised.
For example, if the problem mentions a “friendship” relation (or any other binary symmetric relation),

4
then as blocks we can take pairs of friends. Also, if the problem is modelled as a graph, then the
blocks are the edges of the graph.
Many combinatorial problems can be solved by describing the problem data in terms of suitable
matrices and vectors, and then considering these matrices and vectors over a suitable field .

The town problems revisited

Let us consider the incidence matrix for the clubs in Even/Odd-town, considering the inhabitants
as “points” and the clubs as “blocks”. This incidence matrix is an m × n matrix M , where m is
the number of inhabitants (so m = 2016 here), and n is the number of clubs. It is easily seen that
the club-forming rules of Eventown state that the m × m block-intersection matrix M > M has even
entries. In other words, if we consider the matrix M as a matrix over F2 (that is, modulo 2), then
the condition is that over the field F2 , we have that M > M = O, the all-zero matrix. Let V the
subspace of Fm 2 spanned by the columns of M , that is, by the characteristic vectors of the clubs. The
condition M > M implies that V ⊆ V ⊥ , so we conclude that 2 dim(V ) ≤ m, so that dim(V ) ≤ m/2.
The available columns for M are the vectors from V , of which there are 2bm/2c . So we conclude that
the number n of different possible clubs satisfies n ≤ 2bm/2c . Hence with the given numbers, at most
21006 clubs can be formed under rules 1 and 2.
After the revolution, the Oddtown club-forming rules state that M > M = In ; so M has rank n
and the number n of clubs satisfies n ≤ m. Hence, given that m = 2015, at most 2015 clubs can be
formed.
Finally, suppose that any two clubs share exactly λ members (under rule (a), we have λ = 2). In
terms of the incidence matrix, now considered as real matrix, we now have M > M = λ(J − I) + K,
where K = diag(k1 , . . . , km ) with ki = |Bi |, the size of the i-th block. Hence for any nonzero vector
x, we have
X 2 X
> > > >
kM xk = (M x, M x) = x M M x = λx (J − I)x + x Kx = λ xi + (ki − λ)x2i .
i i
P
Now ki ≥ λ and since all clubs are distinct, at most one club can have size λ; also, if i xi = 0 but
x 6= 0, then at least two xi are nonzero. We conclude that kM xk > 0, hence M x 6= 0, if x 6= 0; so,
nullspace(M ) = {0} and therefore, n ≤ m. Hence, given that m = 2015, again at most 2015 clubs
can be formed under rule (a).

5
Problem Set
[n]


Question 1 Let F be a collection of n + 1 distinct triples from {1, . . . , n}, that is, F ⊆ 3
. Show
that there are two triples in F that intersect in precisely one element.
Question 2 (Putnam 2003-B1) Do there exists polynomials a(x), b(x), c(x), d(x) such that

1 + xy + x2 y 2 = a(x)b(y) + c(x)d(y)

holds identical?
Question 3 Students in a school go for ice cream in groups of at least two. After k > 1 groups
have gone, every two students have gone together exactly once. Prove that the number of students
in the school is at most k.
Question 4 (IMC-2009-1-3) In a town every two residents who are not friends have a friend in
common, and no one is a friend of everyone else. Let us number
Pn the residents from 1 to n and let ai
be the number of friends of the i-th resident. Suppose that i=1 ai = n2 − n. Let k be the smallest
2

number of residents (at least three) who can be seated at a round table in such a way that any two
neighbors are friends. Determine all possible values of k.
Question 5 (IMC-2001-2-4) Let A = (ak,` )k,`=1,...,n be an n × n complex matrix such that for each
m ∈ {1, . . . , n} and 1 ≤ j1 < j2 · · · < jm ≤ n the determinant of the matrix (ajk ,j` )k,`=1,...,m is zero.
Prove that An = 0 and that there exists a permutation σ ∈ Sn such that the matrix (aσ(k),σ(`) )k,`=1,...,n
has all of its nonzero elements above the diagonal.
Question 6 In a party with n people, it is known that for every nonempty subset S of people, there
is at least one person, inside or outside S, such that this person has an odd number of friends in S.
Prove that n is even.
Question 7 The complete graph Kn cannot be decomposed into fewer than n − 1 disjoint complete
bipartite subgraphs.
Question 8∗ Let A1 , . . . , Am be sets of size r and let B1 , . . . , Bm be sets of size s such that

(i) Ai and Bi are disjoint for i = 1, . . . , m;

(ii) Ai and Bj intersect whenever i 6= j.

Then  
r+s
m≤ .
r

6
Question 9 Let A be an m × m symmetric matrix over the field F2 with diag(A) = 0. Show that
for every n ≥ 1, each column of An contains a 0.
Question 10 (IMC-1999-1-5) Suppose that 2n points of an n × n grid are marked. Show that for
some k > 1 one can select 2k distinct marked points, say a1 , . . . , a2k , such that a1 and a2 are in the
same row, a2 and a3 are in the same column, . . . , a2k−1 and a2k are in the same row, and a2k and a1
are in the same column.
Question 11 (IMC-2002-2-2) Two hundred students participated in a mathematical contest. They
had 6 problems to solve. It is known that each problem was correctly solved by at least 120 partici-
pants. Prove that there must be two participants such that every problem was solved by at least one
of these two students.
Question 12 (Putnam 1985-B3) Let f : N2 7→ N. Suppose that each integer in N is image of
exactly 2012 pairs in N2 . Show that there is a pair (n, m) ∈ N2 such that f (n, m) > mn.
Question 13 (IMC-2011-2-2) An alien race has three genders: male, female, and emale. A married
triple consists of three persons, one from each gender, who all like each other. Any person is allowed
to belong to at most one married triple. A special feature of this race is that feelings are always
mutual — if x likes y, then y likes x. The race is sending an expedition to colonize a planet. The
expedition has n males, n females, and n emales. It is known that every expedition member likes at
least k persons of each of the two other genders. The problem is to create as many married triples
as possible to produce healthy offspring so the colony could grow and prosper.
a) Show that if n is even and k = n/2, then it might be impossible to create even one married triple.
b) Show that if k ≥ 3n/4 , then it is always possible to create n disjoint married triples, thus marrying
all of the expedition members.
Question 14 (Putnam 2005-B6) Let Sn denote the set of all permutations of the numbers
1, 2, . . . , n. For π ∈ Sn , let σ(π) = 1 if π is an even permutation and σ(π) = −1 if π is an odd
permutation. Also, let ν(π) denote the number of fixed points of π. Show that
X σ(π) n
= (−1)n+1 .
π∈Sn
ν(π) + 1 n+1

Question 15 (IMC-2013-1-3) There are 2n students in a school (n ∈ N, n ≥ 2). Each week n

students go on a trip. After several trips the following condition was fulfilled: every two students
were together on at least one trip. What is the minimum number of trips needed for this to happen?