Tomescu Problems in Combinatorics
Tomescu Problems in Combinatorics
,'
·S··.·'·I··
;'1 "
.:. ~
~~:~Jt~:anfome,scu
. . .... ·Author
. .:
'e'rfAJ~,M:elter
;
.,'
WILEY-INTERSCIENCE
SERIES IN DISCRETE MATHEMATICS
ADVISORY EDITORS
Ronald L. Graham
AT&T Bell Laboratories, Murray Hill, New Jersey
Tucker
APPLIED COMBI;>;ATORICS
Pless
INTRODUCTION TO THE THEORY OF ERROR-CORRECTI;>;G CODES
Fishburn
INTERVAL ORDERS AND INTERVAL GRAPHS: A STUDY OF PARTIALLY
ORDERED SETS
Tomescu
PROBLEMS IN COMBINATORICS AND GRAPH THEORY
(Translated by Robert A, Melter)
Palmer
GRAPHICAL EVOLUTION: AN INTRODUCTIO]\; TO THE THEORY OF
RANDOM GRAPHS
PROBLEMS IN
COMBINATORICS AND
GRAPH THEORY
IOA~ TOMESClJ
Faculty of Mathematics
University of Bucharest
Bucharest, Romania
A \Viley-Intersdence Publication
JOHN WILEY & SO~S
10 87654 21
Preface
This book is a translation of Proh/eme de ComhinulOricil ~i Teol'ia Gl'Uj/<rilol'.
which was published in Bucharest, Romania in 1981. In Romania, graph theory
is taught in the faculty of mathematics and in particular in what is known as the
chair of informatics (=computer Students in preparatory schools
which specialize in mathematics and also receive instruction in this
Thus the selection of problems presented includes some which are
and self-contained and others which were acces-
in research journals. The author has used the text to prepare
Romanian candidates for in International Mathematical
Olympiads.
Each problem is accompanied by a complete and detailed solution together
with appropriate referenees to the mathematical literature. This should enable
mature students to use the book independently. Teachers of courses in com-
binatorics and graph theory will also find the text useful as a supplement.
since important concepts are in the problems themselves. Even in
the more the reader will learn the important concepts
lOAS Tm1Escu
BudWI'I':,'{, ROIJli/lliu
F""fllllrl' 1985
Contents
1. COMBINATORIAL IDENTITIES 3
5. PARTITIONS OF INTEGERS 30
6. TREES 33
7. PARITY 39
8. CONNECTEDNESS 41
12. PERMUTATIONS 58
I'jj
14. PROBLEVlS OF RA:vISEY TYPE 65
PART n. SOLUTIONS 69
BIBLIOGRAPHY 335
Glossary of Terms Used
cp(n)=n(l-
\
graph whose vertices can be colored with k colors so that each two adjacent
vertices have different colors.
Graph, complete, on n vertices: A denoted Kn , in which every two vertices
are adjacent. It has G) The complete graph on a denumerably
infinite set of vertices is denoted .A is complete if each two
distinct vertices x and yare with respect to either the arc (x,
the arc (y, x), or both. In the complete digraph, denoted K!, each two
distinct vertices x and.r are joined by both the arcs (x, y) and It has
n(n I) arcs.
Graph. connected: A G with the property that every two vertices are the
endpoints of a walk in G. If G is not connected, then it has at least two
connected components (maximal connected subgraphs, which are pairwise
disjoint with to vertices). A connected with at least k + I
vertices is k-connected if the obtained every set Y of
vertices of IY I ~ k 1 is connected.
Graph, strongly connected: A digraph G with the property that for every two
vertices x and y there is a D!:::::: (x • ... , y) and a path , ... , x)
in G.
Graph. Hamiltonian: A graph which contains a Hamiltonian or a directed
graph which has a Hamiltonian circuit.
Graph, multipartite: A G U) whose vertex set can be partitioned as
X = A I U' .. U A k , so that each edge has its in two distinct sets
of the partition. A multipartite is if each of vertices
located in different partition sets is adjacent.
Graph, planar: A G whose vertices can be represented as in the
plane; the become arcs of a Jordan curve which points corre-
sponding to vertices. Two such arcs have in common at most
one endpoint.
Graph, k-regular: A in which each vertex x has d(x)=k, or a
digraph with the that =k for every vertex x.
Graphs, isomorphic: The G (X, U) and H=(Y, V) are
if there exists a bijection f:X - Y such that y] E U if and only if
[f(x), f(y)] E V.
cycle which contains all
1 2 n )
( p(1) p(2) pen)
Stirling number of the second kind, S(n. k). The number of partitions of a set
with n elements into k classes.
Support or transversal set of the of a graph: A set S of vertices with the
property that every has at least one endpoint in S. The smallest
cardinal number of a support for a G is denoted r(G).
Surjections. number of: The number of surjections f: X ...... Y where = m and
IYI == n is denoted sm."'
Symmetric difference: Let A and B be sets. Their symmetric difference is defined
asA t:.B=(A"-B)v(B'A).
System of distinct representatives (SDR) of a family of sets: A family of subsets
M(S)={S!, S2,"" of a set S has an SDR if there is an injective
function f: ...... S such that f(Sd E Si for every i, 1 ~i~m.
Tournament: A complete, antisymmetric, directed graph. Thus between each
two vertices there is one and only one arc (x, y) or (y, x).
Transposition: A permutation g E Sn which has n - 2 fixed points and is therefore
of type I" - 2 2 i •
Tree: A connected
Tree. spanning, of a connected G: A spanning of G which is a tree.
Triangulation: A planar representation of a planar graph in which each face
is a triangle (cycle with three u""nr"'c
Triangulation of an elementary cycle with n vertices: The which consists
of the and the n - 3 which do not intersect in the interior
of the cycle.
Turan's theorem and Turan's number M(n, k): See Problem 9.9.
Vandermonde's formula: See Problem 3.2(a).
Van der Waerden's number W(k, I): The smallest natural number n with the
property that if the set {l, ... , n} is into k then there
exists a class of the partition which contains an arithmetic progression
with t+ 1 terms.
XVII
(a)
[n/21
I
k= 0
-C:1)f = n:1 Cnn).
where is the greatest integer ~ x;
(b) n (n + k) 1 2".
\ n /
1.2 Prove the equalities listed below:
1+ (n + 1)'.
Jo (~)(m~ =(p:q);
(b) n (~)G) (:) 2"-m;
(d)
(f)
= I0 if p is not a multiple of 3.
'1(_l)kG) ifp=3k;
and
al+a3+aS+" -1);
(f) aO+a3+ a6+ a9+'" a 1+a4+ a7+ alO+'"
=a2+ aS+ ag+ all +.,. =3 n - 1.
(g) With to the summations
ao + a4 + as + ... , al + as + ag + ... ,
a2+a6+aIO+ ... ,a3+ a7+ a!l+ .. .
show that three are equal and that the fourth differs from their value
by one.
Verify the inequalities
l=aO<al<'" <an and an >a"+l>'" >U2n
for every n ~ 2.
1.10 Suppose that
(l+x+x 2 +, .. +xm)"=ao+a1x+a2x2+ ... +amnx mn ,
and set
such that
n (2n k-l)
n-l
(x k+ /) (~)"-k
,x+y
6 Problem~ in Combinatorics and Graph Theory
Cl' n2~'''' nJ
n!
I
(s, ..... 5.1
( It
Sl.· .. 'Sk
)=(m-1),
It-l
where the second summation is taken over all choices of the numbers
SI,' • , ,0 which the relations
51 +S2+ ... +Sk=lt;
51 + + ... + =m.
1.18 A function f : {l, ... , n} ...... {1, ... , r} is said to be
for every i,j, 1 ~ i <j ~ n. Show that the number of
on the set {1, ' ... n} with values in the set {1 •... , r} is equal to
where two sums are considered to be different even if they differ only in the order
of their terms? What is the result if Uj > 0 for all i?
1.20 Determine the number of monomials in the expansion of the poly-
nomial
Combinalorialldentilles 7
." +r)J:=n
o
1.23 Prove the
n
for every positive integer n. where the sum is taken over all partitions of n of
the formj! + + ... +njn=n andji~O for 1 ~i~n.
1.25 Show that for h ~ 2 the following relations hold:
(a) max
(n, ..... no)
min
{nt.·,·! "xl
G: =~)(~:- n~(::Y).
1.28 F or every p ~ k justify the
1.29 Prove Abel's identities:
(a) M G) +
U(2m)
Combinatorial Identities 9
1.34 How many words of length n can be formed with letters of the alphabet
A = {a, b, c, d} so that the letters a and b are not adjacent?
1.35 Let
Show that
°
by tp the polynomial obtained by replacing k of the variables x!, ... ,x. in p
with in all possible ways and then summing the ~) polynomials thus obtained.
Show that
p- +z 2 p-'" == {o if m <n;
.
c·x! "·X. Ifm=n,
where c is the coefficient of the monomial XI'" X. in the expansion
of the polynomial p. taking p(x!, ... , x.) + .. + xJ and setting
Xl = ... =x. 1, deduce Euler's
(~) - +"'+(-1)"
n!
x x+n
2
The Principle of Inclusion
and Exclusion;
Inversion Formulas
2.1 In a Romanian high-school class there are 40 students. Among them
14 like mathematics, 16 like physics, and 11 like chemistry. It is also known that
7 like mathematics and physics, 8 like and and 5 like mathe-
matics and All three are favored by 4 students. How many
students like neither nor?
2.2 Justify the following formula, known as the principle of inclusion and
exclusion:
D(n)=n! (1. II
10
The Principle of Inclusion and Exclusion: Inversion Formulas II
D(n+ l)==n{D(n)+D(n-l)j.
2.10 Show that the number Sn.m of surjective functions f: X ..... Y with
IX 1== nand \Y\ == m is given by the expression
Sn.m =mn -(7) (m -l)" +(;) (m- +". +( -11m.
Deduce from this that if E(n, m) denotes the side of this then
E(n, n)nl and E(n, m)=O [or n <m (Euler's identities).
2.11 Denote by Sn.m,. the number of functions f: X ...... Y which have the
that f(X)=:J where =n, !YI=m, and Zc::. Y, =r. Verify the
formula
if, by definition. ao 1.
2.14 A set X is said to be a collection of objects of type 1)" 2)'2 ... n)" jf
there exists a partition of the set X which contains ;.j classes with j elements,
for j 1, ... ,n. Objects which belong to the same class of the partition are
identified. An arrangement of the objects in cells is a function f: where
A is the set of cells. If f(x) = a;, we shall say that the object x E A is in
cell al' definition, two arrangements are equivalent if one can be obtained
from the other by a permutation of the in the same classes of the partition
of X. Classes of this equivalence relation are called arrangement schemes of
objects in cells.
Denote by A 0 (1 A, ••• n).'; 1m) the number of arrangement schemes
of a collection of of type I).' ... n).' in m distinct cells. and by
A(1)" 2.1.2... 1m) the number of arrangement schemes which leave no cell
empty. Show that
also considered to be distinct if differ only in the order of their terms. Use
this to obtain the identity
"-I
=
(_l)l(~)(m-~-I)=(m-11
I 11 /-1 11-1
1)
for any m;;': 11+ I;;.: 2.
2.19 Let V = {x 1, ••• , be a finite set on which is defined a partial order
~ which, definition, satisfies the following three properties:
(aj Xi for every i 1, ... ,11
(reflexivity);
(b) Xi~Xj and Xj~Xi imply XI Xj land thus i=j) for 1 ~i. (anti· 11
symmetry);
(c) Xi and Xk xi~xkforl~i,j,k~l1(transitivity).
A square matrix of order n whose elements are real numbers (aij)i,J=l, .... n will
be said to be compatible with the partial order defined on V, or simply com-
patible, if 0 implies Xi for every i, j= l, ... , n. Show that the sum and
the product of two compatible matrices is a compatible and show that
if a compatible matrix is then its inverse is compatible.
2.20 If V is the ordered set of the preceding problem, show that
there exists a function )J. defined on V x V with the following properties:
(a) j1(x, y) =0 if x is not less than or equal to y;
(b) j1(x, x) == 1 for every X E V;
z j1(x, y)=O for every x < z if x, Z E V.
Show that
f(x}= L x}g(z).
if, by definition. ao 1.
2.14 A set X is said to be a collection of objects of type 1)" 2)'2 ... n)" jf
there exists a partition of the set X which contains ;.j classes with j elements,
for j I, ... ,n. Objects which belong to the same class of the partition are
identified. An arrangement of the objects in cells is a function f: where
A is the set of cells. If f(x) = a;, we shall say that the object x E A is in
cell al' definition, two arrangements are equivalent if one can be obtained
from the other by a permutation of the in the same classes of the partition
of X. Classes of this equivalence relation are called arrangement schemes of
objects in cells.
Denote by A 0 (1 A, ••• n).'; 1m) the number of arrangement schemes
of a collection of of type 1).' ... n).' in m distinct cells. and by
A(1)" 2.1.2... 1m) the number of arrangement schemes which leave no cell
empty. Show that
3.1 For every reat number x and every natural number n let
-1)" (x-n+ 1),
=x(x + 1)" (x + n 1),
Show that
ktl ~: G=~)
= [x] •.
15
16 Problems III Comblnatorics and Graph Theory
3.4 Show that the numbers of the second kind can be asa
function of the number of functions by the relation
S(n. m) m s•. m·
Show that the Stirling numbers also satisfy the recurrence relation
S(n+l,m)=S(n,m l)+mS(n,m),
where S(n, 1) = SCn, n) = 1.
3.5 Justify the following recurrence relations for the numbers of
the second kind, Sen, m), and for the Bell numbers
Part(1 k,
n!
... nkn; n)= (1 [t'k ! (2!t 2 k 2 ! ... (n!)knk.! .
l
the number of permutations p e of type 1k, 2kl ... nk, which contain
k j cycles withj elements for j= 1,2, ... , n is equal to
Perm(1k,
j
) Sen, i = n G) S(k, i)S(n k.
k=O
Show that:
(aj Sa(n, k)=sa(n-l, k-lj-a n_lsa(n-l,k);
(b) Sa(n, k) = Sa(n -1, k -1) + akS.(n - 1, k);
n
(c) sa(n, k)= I s.(n+l. r+l)a~-k;
j"=k
(Kronecker symbol).
3.11 Show that the generating function for the Stirling numbers of the
second kind associated with the sequence (ao, ai' 02 • ... ) can be expressed as
x ~
') Sa(n, k)t" = .
n';:k (1-001)(1-a\t)··· (I-ak l )
3.12 Let S;(n. k) denote the number of partitions of a set X with n elements
into k classes. each one of which contains at least i elements. Show that:
(aJ Si(n,k)=kS;(n-l.k)+.n-l)
( I-I S;(n-i,k-l);
Problems In Combinatorics amI Graph Theory
'"
where the sum is taken over all integral solutions of the equation
+ ".
jl =n which ifor s= 1,,,., k.
3.13 Show that the Stirling numbers of the second kind satisfy the following
relations:
(a) S(n, 2) =2"-1 - 1;
(b) S(n,n-l)=(;);
L
I't"=exp(exp(tl-l).
"=0 n.
3.17 Show that the Bell numbers Bn
1 k"
B"=-
e
Also show that the difference between the number of partitions with an even
number of classes and the number of with an odd number of classes
of a set with n elements is equal to
l)kk"
e
k!
3.18 Let k) denote the number of subsets of the set X == {l ..... n}
which contain k no two of which are consecutive Show that
n-k+ 1)
fin, k)= ( k .
Stirling. Bell. Fibonacci, and Catalan Numbers 19
j*(n,k) n: Cl~k).
1fL.::::: of*(n,k)forn~l,thenLl 1, =3, and I for
every n ~ The numbers L. are called Lucas numbers.
3.20 Show that the Fibonacci numbers the identity
1- =( 1)n+l.
3.21 In how many ways Un can one mount a staircase with n if every
movement involves one or two Show that the generating function is
uo= 1.
3.22 Show that every natural number n~ 1 can be written as a sum of
wise distinct Fibonacci numbers which are not consecutive numbers and
1 of the Fibonacci sequence.
3.23 Show that the generating function of the Catalan numbers satisfies
the equation
1-
+ ". +Cnx"+'" =--,..---
1 (2n-2).
n n-l
20 Problems in Combinalorlcs and Graph Theory
k
is to
3.29 Let go(n+ 1) be the number of sequences (ai' a2,"" an + d of non-
negative integers such that a 1 = 0 and
-ai+ll~l for i=1,2, ... ,n.
Show that
go(n + 1) = c(n, n) + n+l},
where
1
n+
(2n).
,n
3.31 Show that the number of sequences
(x l, ... ,
3.32 Let Sn be the number of functions f: {l, ... , n} .,., n} with the
property that if f takes on the value i, then f takes on the value j for 1 ~ i.
Show that
Stirling, Bell, Fibonacci, and Catalan Numbers 21
(a) lim
q-l
[knJ q
3.36 For n~2 let f(n, k) denote the number of sequences of k integers
1 :(al <az < ... <ak n which
az al=a3- aZ="'=ak-l- ak-Z=1(mod2)
and
Show that:
{n-l)!n!
2n - 1
4.9 Let F = {El' ... , E.} be a family of r-element subsets of a set X. If the
intersection of each r + 1 subsets of F is show that the
of all the subsets of F is also nonempty.
4.10 Let S={X,h";;i";;, be a family of distinct subsets of X with the
property that Xj r-. X d 0 for every i, j = 1,. ., r. If the set X has n elements.
show that max r=2 n - 1 •
4.11 Let X be a nonempty set and let F be a of rn distinct subsets of
X where rn~ 2. Show that the collection of subsets of the form At; B r<:vmrr'lMr,
difference of A and B) where A, B e F contains at least rn distinct sets.
4.12 A covering of a set S is a family of
of S whose union is equal to S. Show that the of an
n-element set is given by the formula
A(n)= n (-1)J C)
4.13 A covering A of a set S by k nonempty subsets is said to be irreducible
if the union of every k - 1 subsets of A is a proper subset of S. If [tn, k) denotes
the number of irreducible by k subsets of an n-element set, show that
k)=
where SU, k) is the number of the second kind. In particular verify that
I(n,n-l) n 1)
and
[(n.2) + 1,
4.14 Let AI,''', be a collection of n distinct sets, and
YrODlCms In LomlJloalorics and Graph Theory
AI" ...• AI, a subfamily of maximal cardinality with the property that it does
not contain the union of the sets, that is.
A,vAdA.
for each three distinct indices k e {i 1 , •• • ,Let min r, where
the minimum is taken over all families of n distinct sets. Show that
1~f(n)~ 1.
4.1 5 Let A 1, A 2 •.•. , An be finite sets such that
IA11=
and let U~= 1 Ai = S. Suppose that for fixed k (1 ~ k ~ n) the union of every k sets
of this family is equal to S, and the union of at most k -1 sets of the family is a
proper subset of S. Show that ISl ~ (k~ 1)' When equality holds, it follows that
IAd= i)foreveryi=l, ... ,n.
4.16 Let (Xdl<;/a be a family of k-element subsets of a set X. Show that
min i d
1 X is to the smallest m such that k ~ (~).
where the sum is taken over all choices of subsets A 1, ...• Ak of an n-element
set X.
IIA1v'" vAkl
where the sum is taken over all choices of subsets A 1,' .. , of an n-element
set X.
n-l G)
4.20 Let (Adl<;/<;m and (Bdl i<;m be two families of sets with the property
that lAd = ... ::::: IAml =p.IBd = ... =IB",I =q and A1(lB j jf and only if
Show that
max p ([n~2J)'
This result is called Sperner's theorem.
4.22 Let X be an n-element set, and F = {A 1, ... , a of subsets
of X which the following conditions:
(1) l==r~nI2foreveryi=1,. ,p;
(")Aj:#:0 for every j= 1, ... , p.
Show that maxp=C=i).
This result is known as the theorem.
4.23 Let X be a finite set, and , ... , Em a family of subsets of X with the
property that the intersection of two distinct sets E; and never has cardinality
exactly equal to one. Further assume that ;,: 2 for i = ... , m.
Show that under these circumstances one can color the elements of X with
two colors so that no subset E j has all its elements colored with the same color.
4.24 Let F , ... , En} be a of r-element subsets of a set X, where
n ~ 2,-1. Show that it is possible to color the elements of X with two colors so
that no subset in the family F has its elements colored with the same color.
4.30 How many triangles can be formed [rom the n vertices of a convex
polygon if no side of a triangle can be a side of the polygon?
4.31 Consider a convex polygon with n vertices. There are n(n-
of the polygon with the property that no two are and no
three are concurrent other than at vertices of the Show that the number
of points of intersection located outside of the polygon is equal to
n(n - 3)(n - 4)(n - 5)
12
4.32 Suppose there are n points on a circle with the property that no three
of the n(n -1)/2 chords determine are concurrent in the interior of the circle.
Show that in this case these chords delimit
4.33 A set of closed curves is drawn in the plane. They do not inter-
sect but each two curves intersect in at least two points. Let np be
the number of points in which exactly p of the curves intersect.
Show that the number of closed of the plane which are bounded
arcs of these curves and which do not contain such an arc in their interior is
equal to
1+nz+2n3+ ... +tp-l)n p + ....
4.34 Let S be a set, and denote by M(S) (S 1. S 2, •.. , an ordered
family of its subset. A system of distinct representatives (SDR) for M(S) is an
rn-tuple ,az, ... ,am ) such that aleS, for l~i~rn and al'i-=aj if i for
1 ~i. rn.
Show that M(S) has an SDR if and only if u u'" uSi,l~k for all
choices of distinct numbers
M(n. k, M(n-l,k,h);
(b) M(n, k, h)~M(n -1, k-l, h -1)+ M(n -1, k, h);
(c) (~)/(~)~M(n, k,
Problems in Combinatorial Set Theory 27
The incidence matrix A == (aU)1 <> 1.)<> v of this configuration is a square matrix
defined as follows: au = 1 if the element i of X belongs to the set X J. and
otherwise. Show that
AT A =(k-i.)/ +Al,
if and only if A is the incidence matrix of a k, A}-configuration where AT is
the of and J is a square matrix of order v all of whose elements
are 1. The matrix / is the identity matrix of order v.
4.43 Consider the functions f: X -> X such that fU(x») =a for every x E X,
where a is a fixed element of X. If Ixj =n~ 2, prove that the set of all such
functions has
p= C~ 1) pn- p
-l.
4.44 Consider the r-element subsets of the set {1, ... , n}. Select the mini-
mum element of each subset. Show that the arithmetical mean of the numbers
obtained in this way is equal to (n+ 1)/(r+ 1).
1,2•... ,2048}. Show that for any subset XcM,jXI=
there are two subsets BcXsuchthat
i=
4.48 Let X be a finite set containing at least four elements, and let
AI, ... , A 100 be subsets of X which are not necessarily distinct. and are such that
lAd >ilxl for any i == 1, ... ,100. Show that there exists Y c X, IYI ~ 4, with
Y for every i= 1, ... ,100.
4.49 The digital plane D is the set of all points (digital points) in the
Euclidean which have integral coordinates. For any two points ,Yd
and P 2(X2, Y2) from D the city-block distance is defined by
which yields a metric for D. For any FeD a subset Be F is said to be a metric
basis for F if for any x, Y E F, x +
there exists b E B such that d4 (x, b) d4 ( Y, b).
Problems in Combinatorial Set Theory 29
Prove that:
(a) D has no finite metric
(b) for any natural number n ~ 3, there exists En C D such that the
minimum number of elements in a metric basis for is equal to n.
4.50 A finite is a BIBD !l with
tv. v,k, k, A) where v;;;: 4 and }. == 1. It is traditional in this context to substitute
the terms point for object and line for block. From Problem 437 one can
deduce that
V=k2_k+1.
The number n:= k -1 is called the order of A Thus for a finite "r,,,,,,-tn,,,'
of order n it can be seen that
r=k n+L
It is convenient to say that a set of points is collinear if it is contained in some line.
Show that a set system !l:= (V, E) where E is a family of subsets of V is a finite
projective if and only if the following three conditions hold:
(1) Every pair of points is contained in exactly one line.
(2) Every pair of lines intersects in exactly one point.
(3) There exists a 4-subset of V no 3-subset of which is collinear.
5
Partitions of Integers
5.1 In how many ways can three numbers be selected from the set
(1, 2, ... , 3n} so that their sum is divisible by 3?
5.2 Show that the number Pin, m) of partitions of an integer n into m parts
satisfies the recurrence relation
P(n+k, k)=P(n, l)+P(n, 2)+ ... +P(n, k),
with P(n, l)=P(n, n)= 1.
5.3 Show that the number of partitions of an integer n into distinct
parts is to the number of of n into odd parts.
5.4 Verify that the number of partitions of a positive n into m
distinct parts is to
1+ ljJ(n)x"
=l-x- +X5 + +
where ljJ(n) Ok if n = (3k 2
± k)f2 and 1jJ( n) 0 if n cannot be represented in
the form n ± (k is an
5.7 Justify the following for functions:
(a) the function for the number P(n) of all of an
30
Partitions of Integers 31
=1;
function of the number Pin, m) of of an
is
(c) the generating function for the number of partitions of n into odd
parts is
==
;.
(- +p(n--F)},
for every n;;:: 3.
5.14 Denote by B(n) the number of the partitions of n into parts which are
powers of two. For B(6) == 6, and the corresponding partitions are the
following:
1+1+1+1+1+1=2+1+1+1+1=2+2+2
=4+2==4+1+1.
Prove that:
+ 1)=B(2n);
(b) B(2n)=B(2n-l)+B(n);
(c) B(n) is even for any n ~ 2.
5.15 Show that P(n}~ for every n ~ 2.
6
Trees
6.1 Let A U) be a tree and , Ud,.· , =(X p • Up) a set of
subtrees of A. If B= nf=1 =1=O, show that B is the set vertices of a subtree
of A.
6.2 Let G1 , ..• , be a collection of subtrees of a tree G with the property
that each two subtrees have at least one vertex in common. Show that the entire
collection has at least one vertex in common.
6.3 Let d 1 , ••• , do be such that
o ~ ... ~dn'
Show that there exists a tree with n vertices of if
d1 + .,. + dn = 2n - 2.
6.4 Let • A2 be two trees of a connected graph G. Show that
there exists a sequence of trees
Al = B 1, ... , Br =
such that B 1... 1 is obtained from by an u and
another edge v between two nonadjacent vertices of B i , for i 1, ... , r-1.
6.5 For a connected graph G let d(x, y) denote the distance between vertices
x and y, that is. the number of edges contained in the shortest walk which joins
x and y. Further define the eccentricity of a vertex x by e(x) = max y d(x, y). The
center of a graph G consists of those vertices Xo with the property that e(xo)
minx e(x) = p{G). [p(G) is called the radius of The diameter of a connected
G, denoted is defined d(G)=max x e(x).
(a) Show that the center of a tree consists of a vertex or two
vertices.
(b) If G is a tree show that is a convex function in the sense that if
y. z are to x, then
2e(x)~ + e(z).
Show that for every connected graph G,
2p{G).
33
34 Problems in Combinalorics and Graph Theory
6.6 Show that every tree with n vertices and with diameter greater than or
to 2k - 3 contains at least
n - k walks oflength equal to k.
6.7 that G is a tree with vertex set X. For x E X let
s(x)= I d(x, y).
Y'X
(a) Show that the function s(x) is strictly convex in the sense that if
y and z are two vertices adjacent to x, then
<s(y)+
(b) Prove that the function s(x) attains its minimum for a single vertex
or two adjacent vertices of the tree G.
6.8 Determine the trees G on n vertices for which d(x. y) is minimal
(maximal).
6.9 Let x I •.•. , x, be terminal vertices of a tree A, and set
Show that:
for every three indices i, j, k the following relations hold:
and
diJ+d jk - =0 tmod 2).
(b) for every four indices i, j. k, I. two of the numbers +dk1 , +dj /!
+ are equal. and the third is less than or to the two
numbers.
6.10 Let A and B be two trees whose terminal vertices are labeled with
numbers from the set {1, 2, ... , If the distances between these terminal
vertices are the same for A and B. that is,
dAti, j) = d8 (i, j)
for every 1 ~ i r, show that the trees A and B are isomorphic.
6.12 Let A be a tree with vertex set X such that IXI =2n + 1. An auto-
morphism of A is a bijection f:X -X which preserves the adjacency of vertices.
that is, [x, yJ is an edge of the tree A if and only if [f(x), fry)] is an of A.
Show that f has at least one fixed point.
6.13 Let A 1 =(X, VI) and A 2 =(X. V 2) be two trees which have the same
vertex set A. that for every vertex x E X the subgraph obtained from
A 1 by the vertex x and the incident with x is isomorphic to
Trees 35
the subgraph obtained from by the same operation. Show that the trees A
and A2 have the same diameter.
6.14 Let G be a tree with vertex set X = and set
D=(dijki=1."n,
where dij= Xj) is the distance between Xi and in G. Show that
n
det D (-1)"- !(n-l)2 - 2.
6.15 Let A be a tree with vertices Xl' ... ,X n • Suppress the terminal vertex
(of 1) which has the smallest index, with the edge incident with
it, and let A n - 1 be the tree thus obtained and al the index of the vertex
to the vertex. Repeat this for the tree !, and determine
......
Yl Y2
. .. .. Yn
Fig. 6.1
D is symmetric;
(4) d1j ~ dik + dkj for every i. j, k and
(5) If dij > 1, there is an index ki= i. j such that dij = dkj •
(1) G is a tree;
(2) G is connected, and the deletion of any of G results in a
G 1 which is not connected;
G has no and if x and J' are any two vertices
of then the graph G l obtained from G by inserting the
[x, y] contains
6.22 Show that for n';il: 3 there are nn- 3 different trees with n unlabeled
vertices and n 1 labeled 1, 2, ... , n 1.
6.23 Let G denote a with n';il: 2 labeled vertices denoted 1,2, ...• nand
m Label the of G with the numbers 1, 2, ...• m, and each
an arbitrary direction. The incidence matrix of G is the n-by-m ma tfix A = (aij),
where 1 ~ i ~ nand 1 m, in which au equals + 1 or -1 if the is directed
away from or towards the vertex i, and zero otherwise.
Prove that if the graph G has n vertices and is connected. then the rank of its
incidence matrix A is to n 1.
6.24 Show that if B is any nonsingular square submatrix of A, then the
determinant of B is + 1 or - 1.
n 2 2n
2
. .lJ
Fig. 6.2
n
6.29 Let T be a tree having vertex set {1, .... n} and set denoted by
E(T). If X, prove that
v ... vA"i~
is defined as follows: The arc (x, y) EO if and only if (x, y) ¢: U for every x, y E X
with X=f y.
Let h(G) denote the number of Hamiltonian paths of the graph G. Show that
h(G) (mod
This property remains true in the case of a nondirected with n;;;: 4 vertices.
7.8 Show that each tournament contains an odd number of Hamiltonian
paths.
7.9 Suppose that the graph G has all of its vertices or odd Show
that each edge of G belongs to an even number of Hamiltonian
7.10 Let G U) be a connected graph with m edges and n vertices.
Show that the number of spanning graphs of G such that every vertex has even
is equal to 2"'-"+ 1.
7.11 The set X of vertices of any graph can be partitioned into two classes
X 1 and X 2 (one of which may be empty) so that the subgraphs with vertex set
X 1 (X 2) have all their vertices of even degree. Show that this property remains
true if the of the vertices of the subgraph by are even and
the of the vertices ofthe generated by X 2 are odd.
7.12 Let C be a collection of pairwise distinct subsets of a nonempty finite
set X with n;;;: 2 elements. Show that the only collections C with the property
that every proper subset of X intersects an even number of sets from Care
P(X) and P(X) '-{0}, where P(X} is the family of all subsets of X.
8
Connectedness
8.1 Let d 1 ~d2~'" ~dn be the of the vertices of a G, and
suppose that dk ~ k for every k ~ n d. 1. Show that G is connected.
with n vertices and 1 ~ k ~ n. Show that
G contains a connected with k vertices.
8.3 Let G be a graph with n vertices, m and p connected components.
Show that
p+m~n.
X=AuB,
such that each edge of the graph has one endpoint in A and the other in B.
Show that a is bipartite if and only if each elementary cycle in G has an
even number of vertices.
8.6 Does there exist a graph with 10 vertices whose vertices have the
sequence
1.1.1,13,3.4.6,7.9?
8.7 Let d l ' ... , d. be such that
Show that these numbers are the of the vertices of a multigraph with n
vertices if and only if
(1) d 1 + ... +d. is even and
(2) d.~dl + ... + l'
8.9 Consider a graph G with n vertices which does not contain a complete
subgraph with three vertices. further that for every two
vertices x and y there are exactly two vertices which are to both x and y.
Show that there is an p ~ 0 such that n = 1 + 2 Also show that the
G is of p.
8.10 Given natural numbers r~ 2 and 3, show that there exists a graph
G which is of r and with girth g(G)=g.
8.11 Let G be a graph of r with n vertices and g(G)=g.
Show that
3)12
n~1+r+r(r-l)+'" +r(r for g odd
and
n~2{1+tr 1)+'" +(r 1)YI2-1} for g even.
8.13 connected graph G contains at least one vertex x which has the
property that the subgraph G x obtained from G by suppressing the vertex x,
and the edges incident with x, is connected. Does this remain true if instead
of eonneetedness one considers connectedness?
8.14 A directed graph G is strongly connected if and only if for every subset
A of vertices, A there exists at least one arc of G of the form (x, y) where
x E A and y ¢ A. Show that this statement remains true if instead of the arc
(x, y) one takes (y, x) where y ¢ A and x EA.
8.17 Show that every tournament G contains a vertex X such that every
other vertex can be reached from x by a path with at most two arcs.
8.18 G contains a set S of pairwise vertices such
COllllectedlless 43
that every vertex x ~ S can be reached by leaving from a vertex YES and travers-
ing a path of length at most equal to 2.
8.19 A tournament T is said to be transitive if, whenever v) and w)
are arcs of T, then (u, w) is also an arc of T. Show that an increasing sequence
S :S\ ~ •• , ~ Sn of n ~ 1 integers is the sequence of
of a transitive tournament with n vertices if and only if S is the sequence
0,1, ... ,n-1.
8.20 Show that the number C(n) of connected with n labeled vertices
satisfies the recurrence relation
C(n)= C{k)
n
n~2 and C(1)=l.
8.21 Show that almost all graphs with n vertices have diameter equal to
2 for n .... 00. This means that if d2 (n) denotes the number of graphs with n vertices
and with diameter equal to 2, then
lim = 1.
8.22 Define a relation - on the set U of of a graph in the
following way: Let Uj - UJ if i or if the Uj and Uj are found on the same
cycle. Show that - is an equivalence relation on U.
8.23 An articulation point of a connected graph G is a vertex x such that
the subgraph Gx obtained from G by suppressing the vertex x and the
incident with x is not connected. A connected graph G which does not contain
an articulation point is said to be 2-connected.
Show that the properties are equivalent for a G with n ~ 3
vertices:
(1) Gis 2-connected:
every two vertices of G belong to an elementary cycle;
(3) G does not have isolated and each two edges of G lie
on some elementary
8.24 Let G be a 2-connected graph. If G contains two of
maximal length, show that these cycles have at least two vertices in common.
8.25 Consider a G and two of its vertices x and y. Let G x y
denote the subgraph of G obtained suppressing the vertices x and y. Suppose
that G 1 and G 2 are two graphs with the same vertex set X and that IXI~4.
If the G\ x - y is to G2 - X - Y for all choices of vertices
x, y E show that the graphs G 1 and are identical,
8.26 Consider a chessboard with n rows and n columns (n odd). Can a
44 Problems in Combinalorics and Graph Theory
Fig. 8.1
v(G) + p(G) = n.
9.2
of then the number of
to
G)- nk (n-k-l).
9.3 Show that a graph G with n vertices and m edges contains at least
(4m/3n){m-n"/4) triangles.
9.4 Show that a tournament with n vertices contains at most H"; I) circuits
with three vertices. Prove that this limit is attainable for n odd.
9.5 Show that a tournament with n vertices contains at least one
Hamiltonian path and at most n !/2n/2 Hamiltonian paths.
9.6 Show that every with n vertices and m > (n/4)(1 +
contains at least one elementary cycle with four vertices.
9.7 Show that if a graph with n vertices does not contain a complete sub-
graph with k vertices (k ~ 2) then it contains at least m = {n/(k - 1)} vertices of
45
Problems in Combinatorics and Graph Theory
M(
n,
k)=k-2.n
-1
2
r2
+
(r)2
ifn=(k l)t+rand r~k 2.
The graph G for which this maximum number of edges is attained is unique
up to isomorphism. G is made up of k - 1 classes of vertices. There are r classes
which contain t+ 1 the remaining classes each contain t vertices.
Each vertex x is adjacent to all the vertices which do not to the class
which contains x. This result is called Turan's theorem.
9.]0 Suppose that a set M contains 3n points in the plane and that the
maximal distance between the points is to I. Show that at most of
the distances between the points are than
9.11 Given 2n points in the plane with no three collinear, show that the
maximum number of line segments which can be constructed with endpoints
in this set of points and so that no are formed is to n2 .
9.12 Find the maximum number of maximal complete (with
to inclusion) in a with n vertices.
9.13 One wants to construct a telephone network points in n
cities. Let c(u) be the cost of constructing the line for an edge u = [x;. in the
complete graph G thus defined. It is desired to minimize the total cost of con-
the network. One must therefore find a spanning tree A of G such that
the sum of the costs associated with the of A is minimal.
Show that the following algorithm produces a minimal tree of G:
(1) select the of G of minimal cost:
(2) among the unchosen select an which does not form a
cycle with the chosen and which has minimal cost.
Repeat step (2) of the algorithm until a set of edges of cardinality n -I has been
chosen.
9.14 Suppose that all the G) edges of the graph of the preceding problem
have different costs. Show that in this case the spanning tree of minimal cost is
unique.
Extremal Problems for Graphs and Networks 47
9.15 Denote by E the set of vectors X=(X 1, X2, ••• , x N ) E,jJ" such that
Xi~ 0 for i= 1, ... ,11 and Xl + ... +xn= 1. Show that if G={X, U) is a
with 11 vertices then the following equality holds:
XiX}'::::: 2 - ~),
where k is the maximum number of vertices of a ('(\T\"\T'ljptp subgraph of G
9.16 Let G be a strongly connected graph. Associate with each arc u in G
a number Cluj ~ O. If a, b are two distinct vertices of G an (a. b)-CUI
is a set C of arcs with the property that every path from a to b contains at least
one arc from C.
Show that
max min =min max
D u~D e Ute
where D runs through the set of all paths D = (a, . . , b) and C includes the set
of all (a, b)-cuts of G.
9.17 F or a digraph G =(X. U) let a. b be two distinct vertices of G, and c a
function c: U ..... !Ii? such that 0 for every arc u E U. Suppose that 9 is a
function g:X -+:?# which satisfies the following two conditions:
g(a)=O;
(2) - g(xJ ~ y) for every arc Y) E U.
The value of the function c for a path D in Ihe graph G is 10 be the sum
of the values of c for all the arcs of G. that is,
c(D) = I clu),
.<A(D)
Vertex a is called the source, vertex b the sink, and the capacity of the arc u.
A function f: V -+1Jt such that for each arc u is called a flow in the
network G with capacity function c [denoted G=(X, V, if the following two
conditions are satisfied:
(Cl Condition of conservation . For every vertex x b the
sum of the flows of the arcs which enter x is equal to the sum of
the flows of the arcs which leave x, that is,
f(u)::.: flu) forevery x EX"",-{a,
Show that:
(a) f(u) = f(u).
!,lEO)
9.20 Consider the following algorithm for obtaining a maximal flow at the
exit b of a network G = (X, V, Assume that the capacity function c(u) *'" 0 takes
on only integer values:
(1) Define the initial flow as having zero component on each arc of
the network, i.e., flu) =0 for each u E V.
Extremal Problems for Graphs and Net"orks 49
for every n;;J: O. Consider the network with 10 vertices illustrated in 9.1.
It contains the arcs of , Y2) of aI'
A3 = , Y3) of a2, together with the
arcs (Yi' )'j), (Yi, Xj) for i, 1, ... ,4 and i and the arcs (a, and
b) for i == 1, ... ,4. In order not to the two arcs in the
senses (x, y) and (y, x) have been by a single non directed
50 Problems in Combinatorics and Graph Theory
b
a
Fig. 9.1
c= 2.
.=
Apply the algorithm of the preceding problem, using the order of the walks
Xt.Yl.b], X2,Y2,X3,Y3. [a,x2.Y2,Yl.X\,Y3,X3,Y4' and soon-
that is, so that the flow fb always it will be equal to aO+al +a2+
a3 + ... , and strictly less than the maximum flow fb =4c.
9.22 For a graph denote by v(G) the maximum number of in a
matching (a set of which has no endpoints pairwise in common); let t(G)
be the minimum number of vertices of a support S of G (a set of vertices such that
every edge has at least one endpoint in S). If G is bipartite, show that v( G) T( G)
by applying the Ford-Fulkerson theorem to a network constructed in a suitable
way from the graph G.
9.23 Let A = be a
1, .... m matrix with " rows and m
number of elements to 1 which are
found in different rows and columns of the matrix is to the minimum
number of rows and columns which together contain all elements to 1
in the matrix. This result is called theorem.
9.24 For a graph G with" vertices define the rollowing two operations:
delete an edge between two
({3) insert an between two nonadjacent vertices.
Let ()2(G) denote the minimum number of operations (IX) and/or (f3) needed to
transform G into the union of two disjoint cliques K. , u where +"2 " "!
and "1.n2~O (by definition is the empty graph).
Extremal Problems for Graphs and Networks 51
Show that for any graph G with n vertices the following relation holds:
c'i 2 ( G)::;; Wn -1) 2 J,
and this inequality becomes an equality if and only if G is isomorphic to the
complete bipartite graph Kp,q where p, q-;:O and p+q=n,
10
Coloring Problems
X(G)~k+ 1.
10.2 Let G be the complement of a graph G with n vertices. Prove that the
following inequalities are satisfied by the chromatic number:
X(GH x(G)~n+ 1,
n~X(G)X(G)~[(n; lYJ.
10.3 Suppose that a planar G has a Hamiltonian cycle. Show that
the faces of all representations of G in the plane can be colored with four colors
so that each two faces which have a common edge are colored differently.
10.4 Draw an arbitrary number of Hnes in the plane so that no three of them
are concurrent. One can obtain a planar graph G by the points of
intersection of the lines as vertices of a graph and the segments between neigh-
boring intersections as edges of the graph. Show that
X(G)~ 3.
10.5 Show that in a connected planar with n vertices and m
there are m - n + 2 faces the infinite face) in every planar rpT1,rp<pnl
tion formula}.
10.6 Show that every planar graph with n vertices has at most 3n - 6 edges
and every planar graph with n vertices which does not contain triangles has at
most 2n - 4 edges.
10.7 Show that the K 5 and K 3. 3 are not planar.
10.9 A planar graph G with n vertices and m edges has the smallest length
52
Coloring Problems 53
m~ g (n-
g-
10.10 Show that each planar G has chromatic number X(G)~ 5.
Verify that for every 11 ~ 4 there is a planar graph with all faces triangular such
that the becomes an
10.14 Let G be a graph with vertex set X of cardinality 11 and set of U.
A t.-coloring of G is a function
f:X- t ..... A}
where A~ 1 is a natural numbcr such that ir y] E U. then f(x) '#
Show that the number of A-colorings ofthc graph G can be expressed in the
form of a polynomial of degree 11 in A [called the chromatic polynomial of the
graph G and denoted Po().)] in the following manner:
( -1
where "
1 k (-lyC)PG(k- il.
where PGP.) is the chromatic polynomial of G.
10.28 Let M(x\. yd and N(X2' Y2) be two points in the Euclidean plane £2.
It is known that the following definitions yield metrics for the Euclidean
d4 (M, = -x2i+ (city-block
ds{M, N)==max([x l [Yl - Y2i)
Define the infinite graphs G4 and Gs as follows: The vertex set of these
is the set of points of £2, two vertices being adjacent if and only if their city-
block or chessboard distance is equal to 1. Prove that the chromatic number of
these graphs is equal to 4, that is,
X(G 4 )=X(Gs)=4.
10.29 If a G contains no and X(G) 3, show that G
has at least p +2 vertices. For any p~ 3 construct a G with p+ 2 vertices
and without p-cliques such that x( G) == p.
11
Hamiltonian Problems
11.1 Show that the contains
t(n-l)!n!
Hamiltonian
11.2 Prove that the number of Hamiltonian cycles in the complete graph
Kn which use h given edges (which pairwise have no common vertices) is equal to
(n h-l)!
for every 0 ~ h ~ n/2.
11.3 Show that for n odd. n ~ 3, the of the
covered (n - Hamiltonian without common
11.4 Let G be a graph with n vertices Xl' ... ,X n whose the
inequality
d 1 ~d2~ ... ~dn'
m~(n;1)+2
56
Hamiltonian Problems 57
is satisfied, then G contains a cycle and there is a with
m =("~ 1)+ 1 which does not contain a Hamiltonian cycle.
t t.9 Let G be a with n vertices of greater than or equal to k.
Show that:
(a) G contains an cycle of length greater than or equal to
k+ 1;
If G is 2-connected, then it contains either a Hamiltonian
or an cycle of greater than or to 2k.
It.l0 Let G be a with n vertices and more than (n-
k ~ 2. Show that G contains an elementary of length at least
11.11 Let G be a digraph with n vertices such that the indegree r (x) and
the outdegree of every vertex x satisfy the
n n
d-(x)~2 and d+(x)~2'
Pin, (;-1
1'
DH(n, k) ~ ) (n-
1
m
L
PES",
n«P) =(n +m
m
1),
~ L
m.~s.
sgn(p)n«PJ=(n).
m
where sgn(pJ is the of the permutation p.
12.4 Show that
is an equivalence relation.
sand t are conjugate if and only if they have the
same number m of cycles and their cycles have, the
same n1fori t, ... ,m.
that the permutation t has in its representation as a
of disjoint cycles: A) cycles of length 1, .... ;.k cycles of
k f).j + 2,1.2 + ... + k),k = n). Show that the number of per-
mutations conjugate to t, or the number of permutations with the
same structure as t, is equal to
1.
12.8 Choose at random a permutation of the set {1. ... , What is the
probability that the cycle which contains the number 1 has length k? (Suppose
that all permutations of these n numbers have probability. This assump-
tion also holds for the two rWf'.t'>II"mc
12.9 What is the probability that a of the set {I, ... , n} chosen
at random contains the numbers 1 and 2 in the same cycle?
12.10 Select a of the set {I, ... , nJ at random. What is its
average number of
12.11 Denote the number of PESo with the property
that p2 = e. Show that:
(a) = 1 +(n 2, where Po:::: PI:::::: 1;
(b) I.
,
(c) L Po to exp(t+
o~o n
12.12 Show that the minimum number of transpositions necessary for
writing a permutation pESo, p=l=e, as a of is equal to
n - c(p), where c(p) is the number of p (including the
cycles of length 1).
60 Problems in Combinatorirs and Graph Theory
12.13 A set T {ll, ... ,tn-d consists of n-1 transpositions of the set
X={l, ... ,n}. Associate with it a graph (X, with vertices 1, ... ,n whose
edges are transpositions in the set T.
Show that the product 1 t 1 t2 •.. t n - 1 is a circular permutation of the set X
if and only if the (X, T) is a tree. Deduce from this that the number of
ways in which a circular permutation on n elements can be written as a
of n 1 transpositions is equal to n"- 2.
12.14 Denote by pen, k) the number of permutations p E of the set
{1, ... ,n} which have = k inversions i <j for which > pun
Show that:
(b) p(n, k) = p
12.15 Show that the number of p of the set {I, ...• n} which
that there exist k elements for which pU»p(i) for every
i<j is to !s(n, k)!.
12.16 Show that the expression g) = maxi~ 1. ..... !/(0 - gUll. where 1
and g are two of the set {I, ... , n}, defines a distance on the set
If one denotes by r) the number of permutations 1 with the that
d(e,f)~r, where e is the permutation 1/(i)-il~r for l~i~n),
show that
F(n,l) the Fibonacci number.
12.17 Denote by an the number of permutations p of the set {1, ... ,
which satisfy Ip(i) II ~ 2 for every i = 1, ... , n. Show that a. is the element in
first row and first column of the matrix An, where
1 0 0
100 0
A= 0 1 0 1 0
o 0 0 1
000 0
12.18 Find the number A(n, p) of permutations of the set \1,2, ..• n} which
satisfy the inequality
p(k)~k+p 1
for k = 1, ... , n.
Permutations 61
If A" denotes the number of up-down permutations of the set {1, 2, ... , n}.
show that the exponential generating function of this number is
sec x + tan x,
where = 1.
12.20 A permutation p(l)p(2)··· pIn) of the set 1. ... , n} is said to be
2-ordered if p(i)<p(i+2) for every 1 ~ i~n 2, and if p(i) < p(i + 3)
for every 1 ~ i~n- 3. Show that the number of permutations of the set {l, .. " n}
which are both 2-ordered and 3-ordered is equal to the Fibonacci number F"
foreveryn~l F1=land +Fn_!forn~l).
12.21 Let f(n) denote the number of sequences U!, U2, , . , , Un formed
from n numbers in the set {I, ' , . , n} and which satisfy the following
for every i = 1, ... , n 2,
for every i= 1.,." n-3.
Determine f(6).
12.22 A permutation pel) p(2) , .. pen) of the set {I, ' .. , n} is said to have a
fall at p(i) if p(i) > p(i + 1), where 1 ~ i ~ n 1: by definition, every permutation
has a fall at p(n). The Eulerian number A(n, k) is defined as the number of per-
mutations of the set {I, ... , exactly k falls. Show that:
(a) A(n, k) = kA(n-l, k)+ (n - k + l}A(n- L k -1) for n ~ 2, and
A(n,n}= l)=lforanyn~l;
(b) A(n, k)=A(n, n-k+l);
+k-
(c) x"= A(n, k) for 11 ??: 1;
n
k-! (
(d) A(n, k)= (-l)j n+ (k- j)".
\ J
12.23 Show that the number N(n) of permutations p E such that
1 2 ...
( n n-1 ...
N(4mJ=N(4m+l)=(2m)!/ml for m??:l, and N(4m+2) N(4m+3)=O
for m??: O.
13
The Number of Classes
of Configurations
Relative to a Group
of Permutations
13.1A ticket-punching machine of the Bucharest Transit System uses nine
prongs in a array inside a ABeD. What
is the number of ways in which a ticket can be punched using all possible
patterns? The ticket can be put into the slot along AB with either of its
faces showing.
13.2 If G c S. is a perm utation group on the set X = {1, ... , n} and x, y E
let x,... y( G) if there exists a permutation f E G such that y = f(x). The relation
thus defined is an equivalence relation, whose equivalence classes are called
orbits of the group G.
If ;.j(g) is the number of of length one of the permutation g or the
number of elements of X which are invariant under the permutation g, then the
number of orbits of a group G c S. is equal to
1 " .
-I
G
I L.. "'l(g)·
gEe
where the sum is taken over all solutions (d) of the equation
d 1 +2d 2 +"'+ndn n, (1)
and
G&=2
1{
k.tl dhd{(k,1)
n
en N '
d
where 1d=·HL~.r=l(k, I)dkd r 1 dk }, and the sum is taken over all non-
negative integer solutions (d)
that is, to the numerical value of the index polynomial of G for all variables
to m. The index polynomial of the group G is defined the equation
Fig. 13.1
14.1 If the poin ts of the are colored with three show that there
will always exist two points of the same color which are 1 unit apart.
14.2 Show that if the of the plane are colored with two there
will exist an equilateral with all its vertices of the same color.
There is, however, a orthe of the plane with two colors for which
no equilateral triangle of side 1 has all of its vertices of the same color.
14.3 Show that whenever the points of the plane are colored with two
colors, there will exist an triangle of side 1 or which has13
an of its vertices of the same color.
14.4 Let T be a 30-60° right triangle with sides 1,13.
and 2. Show that for
any 2 coloring of the points of the plane there is a triangle congruent to T
o
is a of 10.
14.7 Let aj, a2, .. " ah2.,.1 be a sequence of numbers. Show that it contains a
monotone subsequence with k + 1 terms.
14.8 Let f be an function defined on the set {I, ... , 2" - 1 1
with the property that for i == 1, .. , ,2n - 1 one has 1 ~ ~ i. Show that there
exists a sequence
65
66 Problems in Comblnatorics and Graph Theory
1
for which f(al):;S; ... :;s; f(a n ). However, this is no longer true if 2n - is
by 2"-1_1.
14.9 Show that jf nine in the are selected so that no three are
collinear. then five of the points form the vertices of a convex polygon.
14.10 The G is formed from two odd
setsA I , ... , andB1, ... ,B"" The also
of the form B j] for every 1 :;s; i:;S; nand 1 ~j:;S; m. Assume that the mn + m + n
of G are colored red and blue so that no triangle is monochromatic. Show
that the m + n of the and C" are either all colored red or all
colored blue.
14.11 Show that if the edges of a complete graph with no p q) vertices
are colored either red or blue, then there is either a complete subgraph with
p + 1 vertices all of whose edges are or a subgraph with q + 1
vertices all of whose are blue.
The smallest number n ~ no with this property is called the number
with parameters p + 1 and q + 1 and is denoted R(p + 1, q + 1) for every p,
(p,q
14.12 Show that R(3, 3) = 6.
14.13 Show that the number R(k, k) satisfies the inequalities
~ R(k, k)~ 3
for every n):no(k) there exist two nonvoid disjoint subsets X, Y. such that
X. Y, Xu Y have the same color.
14.19 Denote by K the complete
OC) with a countably infinite number
color its edges with r colors. Show that the
\1"'.·11 .... ,""0· K", contains a
infinite monochromatic sub graph.
14.20 Let (a.l.€N be an infinite sequence of real numbers. Show that it
contains an infinite subsequence which is either
decreasing, or constant.
14.21 Consider an infinite set A of in space. Show that A contains:
(1) an infinite subset A! of collinear points or
an infinite subset of with the property that
no three points are collinear or
(3) an infinite subset of points with the property that no four
points are coplanar.
14.22 Show that for every partition of the set of integers {1, 2•... , 9} into
two classes, at least one of the classes contains an arithmetic progression with
three terms.
14.23 Show that in every partition into two classes of the set M =
[1, 2•... , 256} there is a class a geometric with three
terms.
14.24 Prove or From the interval (3" + 1)/2J one can select a
set of 2" containing no arithmetic triple (three numbers in arithmetic
progression).
14.25 Show that the Ramsey numbers satisfy the inequality
(2+3
R(3, r):S;
n+1J
j [ -2- .If n =t= 2 (mod 4),
(b) fJ(n)=l
l~+l ifn::2(mod4).
68 Problems in Combinatorics and Graph Theory
14.28 The number R(F l' F 2) for two graphs F 1, is the minimum
p such that every of the of Kp contains a green or a red
Prove that the Ramsey numbers for stars are given by the formula
m+11 ifm or 11 is odd,
R(K 1
."" KI .nl= { m +n 1 jf m and 11 are both even.
14.29 Let be a tree with m vertices where m - 1 divides n - 1. Show that
R(T"" .n)=m +n-1.
14.30 Show that the Ramsey number R(K"" K 1.") is the formula
R(K""K1 .• l -1)n+1
for any m, n~ l.
14.31 If m ~ 3 is a fixed natural number. find the smallest natural
number rim) with the property that every partition into two classes of the
set {l, 2, ... , r(m)} contains a class with m numbers (not distinct)
Xl, •••• x'" such that
Part II
S L TI S
69
Solutions
CHAPTER 1
Let
and
[n/2]
B.=2 k~O k
(n)( k-l'
n)
+ It=(x+ 1)2. and taking note of standard
it can be seen that for even n, An is the coef-
of (x + 1)2., and B. is the coefficient of x· - 1 in this
expansion. In for even n one can write
= I (n)2
k=Ok
= (2n).
n
B.-_ n ( n \( n )_(n-l
n-k) k-l -
2n ).
Thus in this case
Since
71
it follows that the value of An - Bn is the same, that is, {l/(n+ 1
(b) Let
!tn)::::: · (n+k) k 1 .
k
n+l
0
k 1 nf (n+k)
f(n+l)=2:: (n+ kL+k) 1 =2:: (n+k) k-l
1.+1
k=O k=0
f(n)+ n
+
1
1) 2"+1 1
+H(n+ 1).
Thus f(n + 1) = for every n ~ 1. which implies that =2". [D.
Fibonacci Quarrerly, 2 (1978),
1.2 ta) Recall that the number of strictly increasing words c! Cz ..
with Cl <C2 < ... <Ck and c, E {I, ... , n} for every 1 ~i~k, is equal to
follows that Ck-2 E {k 2, k-1, ... , Let n-2}.
be the number of
words of this form with Ck _ 2 = q. One can then write
"-2 1.= "-2
I
q 2;
Aq[
Z
+C;)-Cn1 + (2n-o 2) +.
1+ l'
+ 1=
nO +
The proof of (c) is completed equating the coefficients of X"-l on both sides
of this equation and using standard properties of binomial coefficients.
(~)G)
nl n!
m k!
(c) The identity is established by induction on m. For m=O both sides are
equal to 1. that (c) is true for m 1, ... , p - 1, with p~ 1, it follows
that
k=O
f (- l)k (n):=
k '"
1 (_ 1t (") + (_ 1)1' (")
\k/ p/
(d) It will be shown that both sides of the equation represent an enumeration
of the same quantity.
The left side can be interpreted in the following way: Choose k elements from a
set M with IMI = m elements in (~) distinct ways. Consider another set N (with
INI "" 11 elements) disjoint from M. Choose m elements from the union of the set
N and the k elements previously chosen. One can do this in (":k) distinct ways.
It follows that the number of ordered pairs or sets (X, Y) where X eM. YeN v X
with IXi=k, !Yi=m is to (;)(":k). If the result is extended to k which
varies from 0 to m, the number of all pairs (X, Y) will be equal to
'" (m)(11 + k).
c k m /
This can also be solved first choosing the sets Y n M and Y n N.
If IY n then 0 and it follows that IY n MI = m Thus we can
choose Y so that IY n NI and IY n MI =m in Cll",':) different ways. It
remains to choose the set X so as to satisfy
YnMcXcM,
since YcNvX and thus YnMc(NvX)nM=(NnM)v(XnM) X
= X. But the set X can be chosen in 2' ways, because for each of the j elements
of the set M ".( Y n M) there exist two possibilities: it belongs or does not
belong to the set X. It follows that the number of pairs (X, Y) with the stated
property is equal to
The notation (:) =p( p -1).. . - q + 1)/q! will also be used for
values of p.
(0 using (a) it follows that
n )(q)(p+q-j)=
+q-j j q (n-p-q n! _ _
I __
j:>O
== (;)(~)C-p)
=(;) e)(:=:) G)(:)'
again by the use of (a).
(g) Identity (f) is an equality between two polynomials in n, of p + q.
This equality holds for every natural number n, and hence the two polynomials
in n are identical. It follows that the two sides are equal if n is replaced by -1- n.
Since
and
one can deduce (g) from (f). Finally, (h) follows from (g) by setting q=p.
1.6 Let S(I, m, n) be the sum of the left-hand side. By replacing
2n +2)
(n+k+l with (2n +1 ) +(2n + 1),
n+k+l n+k
one finds that
= 2)' ( 1 k ( 21)( 2m + I )( 2n + 1 )
'k ) I+k m+k+1 n+k+l
x(m+k+ 1)(1 k+n+k+ 1)
But
+b)P(l ={x(a- b)+b(1 + x)}p(1
= L (p)
II
1=0!
.
and hence the coefficient of XII is also to
and
:" en 3
Xi +xj2"+·
x = -l) obtain Dixon's
C/)(~) j+i) (1
(1)
But
(2)
(3)
(~)Cn~p)xp-j(1+X)i= (-1)P-
i
e)C n
t+i)O+x i . (4)
Solutions 79
(-1)P I(-
;=0
p
-p+ +
p
But
(3) it is seen that the right-hand side is (2")2, which establishes and
hence Dixon's formula. W. Ljunggren. Mal. 1!dskr., 29 (1947),
I
k
(_1)k ( 2n
n+
But
+l)"+C)(x+
+G)(x+ +...
This that the coefficient of x" is equal to
an=l +
n(n
+ -'---'--....,.,--.-----.:. + ....
The exact form of the last term on the parity of n. This coefficient can
80 Problems in Combinatoric;; and Graph Theory
(2k)!(n)
(k 2k'
(b) First show that ak a2n-k' Then set x= 1.'y and multiply both sides of
the .1'2". It follows that
( y2 + .I' + 1)n =
with the fi r5t ak=a2n-k for k=O, . .. , 2n. Now
substitute x for x and obtain
which is
(c) In order to prove (c) one substitutes x 2 for x in the expansion of the
statement of the problem and obtains
and hence
and
After equating the real and imaginary part of the left-hand side to zero and
using the notation
SI=aO+a3+ a6+ ... ,
S2=al+ a4+ a7+ ... ,
S3 = a2 + as + as + ... ,
it can be shown that
SI- .)2'"!-S3=0
2 '
where
more than the other sums, which have a common value. Thus j == i 1 == -/1
(mod m+ This means that j+n 0 (mod m+2) and Sj= +11"-1;,-
++2) 1. But the other sums are equal to
... + S",-rl =(1 + 1 + ... 71)"
7 I)" 1 + since So + S1
+. It follows that the property is true
+
for n.
If n is odd. then from the induction one can conclude that
the sum 7; is greater by one than the sums with a common value. As in (a), one
can show that the sum Sj' n == 0 (mod m 7 2) is smaller by one than the
sums with a common value.
Thus
1,
SO=(~)7(;)+G)+ ... ,
=G)+(~)+ +.",
+ +(;)+ " ..
just established that for every n. two of these sums are
and the third differs by one. For m 2 one obtains property (g) from the
previous
1.11 Recalling the rules for one sees that the desired
coefficient is to the number of ways in which k can be written as the sum
of two integers:
k=al +a2
where 0::;;; ai' Q2::;;; n-1. Two will also be considered to be
distinct if they differ in the order of their terms. If 0::;;; k::;;; n - I. these k + 1
are 07k, 1 +(k-l), ... , k+O. if n::;;;k::;;;2n-2, the
0+ k. 1 + (k 1), . " ,(k n)+ n and those obtained by
two terms do not the ai' a2::;;; n - L Thus in this case the number
of to
k71-2(k n+ 2n k-l.
84 Problems in Combinlliorics and Graph Theory
Note that in both cases the coefficient of x' can be written in the form
n-In k-ll·
1.12 The values a 1> ••• , ar are obtained in the following manner: Let a l
be the largest x which satisfies the inequality
n ~ ~l)
Finally, Qy will represent the difference
(\atTl)
r
>n an d
....
r' 1) -(art) > Q2)
1 ' that is,
r
+( )+ ... +(b,)~(bl)+ -1)+ ...
b2
r-l 1 r r-l
T(bl-(r-
1
In view of this last inequality implies that
.
T'" +
(hi -(r-
I
Solutions 85
(3)
(br+l)=(bl)-:-(
J
r
bl)=(br )+ -1)+ r-2-1)
r 1
1
r-l
)
-:- (b!
r
+...+(b 1)) +
r-l
1 -(r
I
-(r-l))
o .
which contradicts inequality (3).
Thus the values a l , ... , a, are determined and the result is
1.13 The identity can also be written
-k 1)
n-l
(x k -:- +
The coefficient of the term obtained by expanding lx +
help of Newton's binomial rormula is equal to (:,:). On the
recalling the expansion of (x -:- y)k and the standard binomial formulas, one
fi nds that this coefficient is equal to
o:(n, m)=
" (2n n-I
-k 1)[( k )+ ( k \]
n-m m-n) .
Hn=m,then
2n )
( 2n- m = IUn n
Ak = (2)
k=1 I
where the sets Ak are disjoint.
If cn _ '" + 1 = k -t- 1, then the letters C l' . . . , Cn _ m are chosen smaller than
Cn -", 1, and thus Cl ... Cn -", is a strictly word formed from letters
of the {I, ... , The word Cn -",+ 2 •• C2n-", is a strictly
word of n - formed with the letters of the set {k + 2, ... , 2n} of cardi-
nality 2n-k-1. Thus the word C 1 •.. be chosen in (n~m) ways, and the
word cn- m+ 2' .. CZ n - m can be chosen in n- 1 1) distinct ways, which
that
( 2n-k-
IAkl = \ n-l
n
=1
(2n-k-I\( k
n- 1 ) n
=(\2n2n- p),
or
· (2n-k-l)(1m kn)=(2n).
n-l m
One also sees that et(n, m) = e,;) in this case. Toscano, Boll. Soc. Mach.
Calabrese, 16(1965), 1-8.J
1.14 One can write
.,---:---: {n (n - k
\::.1
L.,
(n . 1) n" -
"(
1 k k +1).=
1 \}
L.,
k=O. k / k=O
=n".
[J. Riordan, Ann. Math. 33(1962),178-1
1.15 The n! in the first box can be chosen in (:,) ways. the n2
in the second box can be chosen from the n nl remaining objects in
ways, etc. The total number of arrangements is to
is an ordered of the set {I •... , n}. which may contain empty classes.
note of the upper indices of each in the last slim indicated,
one sees that it an arrangement of the set of objects {I, ... , n} in p
boxes Yl, Y2, ... ,Y p • slIch that h contains nk for k = 1, ...• p. In fact, irone
takes objects iI, ... , in, in box h, ... , objects k h •.. , k np in box YP' then slIch
an arrangement is also obtained. the rules for removing parentheses,
it turns Ollt that these are un'v",.,,,
n!
nl
arrangements of n in p boxes which contain nl'" " np objects respectively.
at at
Take = = ... = at =aj for every 1 ~ i ~ p; the sum under consideration
becomes
where the upper indices now indicate powers. This follows from the fact that
88 Problems in Combinatorics and Graph Theory
.~., nJ
terms equal to a~2 ...
1.17 to Problem Ll9 there are (~_ i) of m as the
sum of h positive Let .'II denote the number of terms
of this type. It follows that SI + ... +sk=h and =m. where
k the term of the sum.
If the n urn bers SI this property are fixed, then the n urn ber of ways in
which m can be written as a sum of h positive terms such that 51 terms are
to i for i;;: 1 is to the number of arrangements of a set of h objects into k
such that the kth box contains ii. This last number is given by
C!'.~.,
Thus the identity is true, since it has been shown that both sides are
equal to the number of different representations of m as the sum of h
1.18 Let N= ... , n}, R={l •... , r}, and C(N, R) the set of increasing
functions f :N -> R. Identify the increasing function f with the increasing word
b l b z ... bn, where bl = f(i) for i = 1, ... ,n. Since f is increasing, it follows that
l~bl~b2~'" ~bn~rfor l~i~n.
Consider a mapping
F:C(N, R)-Pn(X),
where P n(X) is the of n-element subsets of the set X = {I, 2•... ,r + n l}.
The function F is defined the equation
Because b l ~ ~ . . . ~ b", it follows that b l < b 2 + 1 < b 3 + 2 < ... < bn-t"
(n -1), and hence the image of a function f under F is in fact an n-element subset
of X.
lf b l ... bn::fo C I C2 ... CM are both increasing words, then there is an index
i such that, for b1 =c!, . . , I=Ci-1 and b/<Ci where l~i~n.
This that bf + l)<ci+(i-l) and bi+(i-l)fF(cl'" One can
conclude that F(b l .•• bn)::foF(CI ... and thus the mapping F is
If Y X satisfies! Y! n, take Y = {J'l' . , , , rn} and 1 ~YI <.1'2 < ... <
r+ n 1.
Let bi=Yt-(i-lj for every 1 ~j~n. Then b! ~ ~ ... ~bn and bi E R for
every 1 ~ i ~ n. the definition of the function F, one finds that F(b! ... bn )
= {Yl' ... , Yn}, rrom which it follows that F is
Since F is it can be inferred that the number of increasing functions
f:N ...... R
Solutions 89
is equal to IPn(X)I=(n+:-l)=[r]n/n!.
1.19 Define the sum
-'-(n_+_l..:...)(:.....n_+-'---:-.,:-~-....:.. = (n + p -
p-l
1).
For p=2 there are (.""l)=n+l terms; this can also be shown by
Newton's binomial
Problems in Combinatoric, and Graph Theory
1.21 Let
"=0 m"O It
Expanding (1- 1)- 2k by Newton's binomial formula, one sees that
the coefficient of t" in the expression rk(l- t)- 2k is to
-2k)
(l1-k
Now consider powers smaller than or to n. It turns out that the coef-
ficient of x" in the (-l)P+ + ... T x")P/ p is to
jl!h ...
By su bstituting for p the values p = 1, ... , n, the sum on the left-hand side of the
statement of the problem is obtained. [1. Sheehan, Amer. Math. Monthly, 77
(1970), 168.J
1.25 For natural numbers nj and nj such that nj~ nj+ 2, one can show that
or
which maximize (for fixed k) the indicated in (a) and (b). We show
that these are as equal as that is,
(1)
for every i, I, ... , k. In fact, if the contrary is assumed. then there are two
indices i and j such that ml ~ mj + 2. In this case, one can write
which contradicts the maximality of the product mlm2 .•. mk' In the same way
92 Problems in Combinatorics and Graph Theory
-5)
2 '
which is to m2 -m< 10(m- 5)(m- or 9m 2 -109m+ 300>0. This
trinomial has two real roots in the interval which establishes the
inequality. Thus if there exists an ml~ 8, then the of the binomial coef-
ficients from (b) cannot be a since one can the part to
mj two parts to 5 and mj - 5 respectivel?:, and the product of the bi-
nomial coefficients therefore increases. Since G) > (2)@' (~) > G)(~), and G) > (~)(~),
it follows that for n ~ 7 the desired maximum is equal to G).
If n~8, in the case of the maximum it follows that max(mj, ... , mk)~6,
since in the opposite case there exists an m, = 7. But relation (1) implies that
there must exist at least one number mj which is equal to 6 or to 7. For
=360.
=600,
Solutions 93
216=(~)(~)(~)< =225.
If n ~ 8. neither of the numbers rnl can be equal to 2 or to 3. For example, suppose
that there exists an rnj 3. It follows from that there must also exist a number
rnj E {2, 3. 4} and hence
In conclusion, almost all num bers rnj are equal to 5, but at most two of them
are equal to 4 or to 6, so that relation (1) is satisfied,
Let B(n) denote the maximum in case (b). Then:
Consider
max max
l~k",nnl+'··+nk=n
h Y
-Y) ~x+y'
(2h)
Since n! ~ 3h and n2 ~ 3h, it is sufficient to show that
c:~:)e:~y ~J.
Let x + y= k(constant) where 2h. Then
O~ k~
-X)(3h- Y) -X)(3h- Y)
h x h y 2h 2h
the factors two by two in the order in which they are written
it can be observed that their sum is constant and to 6h - k,
6h - k - 2, ... , 2h - k + 2. Thus the become minimal when the factors
differ to the among themselves.
One can two cases'
(a) k~ h.ln this case the products of the two factors become minimal simul-
taneously for x =0, y == k or x == k, y= 0. Let x = 0, y k. In this case it is sufficient
to show that
3h)(3h k) >-
( h h-k"" k
But
2~k)~C:)·
Recallihe standard formulas for binomial coefficients. It suffices to verify that
G~=~)~Ch2~k).
Solutions 95
If the numbers lXI' ct2"'" ctn';JO are then it is possible to selectj elements
in ctj sets among AI"'" Ak in C)a} ways 1,2, ... , n. The union of these
elements is the set Y. On the other hand, one can select ctl sets from among
AI> ... , Ab ct2 sets from the remainder, and so on, in
k!
k!
-(IXI + ... +
distinct ways.
Thus the number of subsets of X can be written as the slim of the
numbers
over all representations P=ctl + + ... +nctn where the cti are
If ctl =P and ct 2 = .. =ct n 0, then + ... +ct., or k';Jp. For
p =k -lone obtains an identity due to I. M. Voloshin (J. Combinatorial
A12 (1972),
1.29 (a) We prove equality (a) by induction on n. For n = 1 both sides are
equal to x-!- y+ 1. Differentiating with respect to y, one sees that
96 Problems in Comblnalorics and Graph Theory
a
,(x+y+n)" n(x+y+n)M-l
oy
+(y+ l)-t-(n-lj}"-1
-t-l)+(n-l)- Il-k
-1) x"
j
!S(j, n)=O,
since the Stirling number of the second kind S(J, n) = for j ° ... , n-1.
(b) The left-hand side of identity tal can be written as follows:
=kt (:)
+ Jo (~) -k).
Thus
-k 1 -n(x+y+n)"-I
+n- y+
=(x+y)(x+y-t-nln - 1
,
1
+ {(x+y+n)"-1_(x+nr 1 }.
y
taking the limit on both sides as x-+O and y-+O, is obtained.
1.30 Let y =
in both sides.
° in the first Abel identity, and equate the coefficient of x
by its initial values flO) and f(1). It is thus sufficient to show that the system
of equations
Cl + = frO),
c 1'1 -t" == f(1),
has a solution for every choice of f(O) = bo and f(1) = bl' The solution of the
system, when rl #-r2. is by
C 1 = =---"""'::""---
r1 -72
which is an identity.
Thus lAn) == nr'i is a solution of recurrence relation (4). One can conclude by
similar reasoning that f(n}= fl(n)+Cd2(n)=r!(C 1 -t"C 2 n) is a solution of
(4). The constants C! and C 2 can be chosen so that fin) satisfies arbitrary initial
conditions for n = 0 and n = 1. In fact one obtains the system
C 1 = f(O),
rl(Cl -+C 2 l= /0),
f(n+k)=at!(n-+k-l)+'" +ad(n)
Solulions 99
If the roots of this equation are r1 , •• " rs and their multiplicities are rpcnPI'!i\IP
ml , . , . , m", where m 1 + '" +ms==k, one can show similarly that the
solution of(5) has the form
!(n) = - 1 )rl'.
for initial values a 1 = 1, Q2 = 3 (two buns on two consecutive days, an ice cream
or a on the first
In order to solve this recurrence relation one must use the characteristic
equation (see Problem 1.31)
r2 r-2=0,
which has solutions r 1 = 2, r2 = 1. The solution has the form
an C 1 2n + 1)",
where and C 2 are determined by the system
= 1,
=t, and
an 1 +( _l)n}.
1.33 Cover the with dominoes by from the side AD of
length 3, One obtains one of the cases of 1. L In the last two cases, a
A A r---"V7r',.,,.,.,.~
D<---'----'
Fig. 1.1
.vv .Problems in Combinalorlcs and Graph Theory
r-~--------------~B
Fig. 1.2
square with side 1 remains uncovered in the 3 x 3 square with side AD. It can
be covered in a unique way by the striped domino. Thus
U(2n+ 2) = 3U(2n)+ 2V(2n- (1)
where V(2n) denotes the number of ways in which one can cover the
of a ABeD of sides 3 and 2n to which one has added a
rectangle with sides 2 and 1, by its long side to the side AD of the
From 1.2 it follows that the additional domino can be covered in two
ways, and hence
U(2n) + V(2n-
Now in (1) use the expression for V(2n-2) given by (2) to obtain
=3U(2n)+ 2)+2V(2n -4). From (1) it follows that 2V(2n
3U(2n- and thus the relation becomes U(2n+ 3U(2n)+2U(2n-2)+
U(2n) - 3 U(2n - 2). This establishes the linear recurrence relation
U(2n+2) 4U(2n)- (3)
By a simple argument one can deduce that 3 and U( 4) == 11. The charac-
teristic equation of the recurrence (3) is r2 4r + 1 = 0, with roots r 1.2 = J3,
and thus the solution of the recurrence (3) is
U(2n)== +
The constants C! and are determined the values U(2) and
U(4):
where
Each word of length n which satisfies the conditions of the problem can be
obtained from a word of length n 1 these Let x.
denote the number of words of n which start with a or with b, and let J'n
denote the number of words with first letter c or d which satisfy the given
conditions. The following recurrence relations hold:
Yo 1+ l'
C1 3+Jfi C 3 4
2 + 2 2 =,
14,
yields
(n-2)-(k
( k-l
This yields the recurrence relation
(1)
---::---=1.
Therefore
and
1.36 Let S(n, k; Xl' " , X.)=(XI + ... + - L (Xl + ... +Xk- +
L(Xl+'" +Xk-2)"- ... ; then the desired sum is Sn(Xl,"" =aoS(n, n;
XI"'" xn)+aIS(n-l, n: Xl>"" xn)+a2S(n 2, n; XI."', x n}+, .. +
anS(O, n; Xl"'" Xn). Since eX=l+ 1+x2/2!+ ... , it can be seen that the
expression for the exponential generating function is
-1)· .. (e>=kz_l)
= +
IUj
By equating the coefficients of z" on the two sides of this equation for n == I, ... , k,
one can conclude that S(n, k; Xl' ... , for n<k and S(n, n: Xl"'" Xn)
==n!Xl ···X•. Thus Sn(Xl,"" aoS(n,n: Xl •... 'xn)=aOn!xl "'X", [L.
Carlitz, Fibonacci Quarterly, 18(1) (1980), 85.J
1.37 Let
., .• Xj-I> O. Xi+b""
equal to 0 for O,,;;k<n and is equal to c=n! for k=n, from which Euler's
identity follows.
1.38 Consider the left-hand side of the identity as a polynomial in n. The
constant term is then equal to
-(i) (p-IJP+ G) (p
(Use Euler's formula, Problem
it can be shown that the coefficient of nP is equal to
since m<p.
1.39 The identity will be established mathematical induction on n.
For n = 1 it reduces to the identity 1/x l/tx + l/x(x + Suppose that the
proposed identity is true. Replacing x by x + 1 yields
(~) -
x+l n+ 1)'
Subtract this relation from the original equation to obtain
=-------
by virtue of the recurrence relation for binomial that is, the identity
in the statement of the problem for n + 1.
Solutions 105
CHAPTER 2
2.1 Subtract from the total of 40 students the number who mathe-
and
40 - 14 - 16 11.
The students who mathematics and physics are subtracted and
thus they must also be added once.
A similar for the two other pairs of yields the number
40 14-16 11+7+8+5
The students who prefer all three subjects were subtracted three times and after
were added three times. In order to obtain the number of students who did not
any of the three subjects it is necessary to subtract once the four students
who had a preference for all three subjects. The final result is
40 14-16-11+7+8+5-4=15.
2.2 The proof uses induction on q~2. For q=2 the formula becomes
U IA11+IA21 lAin
which can be verified.
.... "'''''""\01"' the formula is true for each union of at most q 1 sets. It follows that
IA1u'" uAql=IA1U'" UAq_ll+!Aq:- U'" uA q tln
Applying the distributive property for intersections of sets, one has
(AI U ... U
n
i~ I
regrouping terms, one obtains the formula of inclusion and exclusion for
unions of q sets. The formula is called the Principle of Inclusion and Exclusion
because of the alternating of the right-hand side.
2.3 Let Pc Q= {I, 2,. ., q} be a fixed set with !pl =p. The number of
elements which to all the sets Ai with i E P and do not to any of
the sets Aj with j E Q coincides with the set of elements which belong to the
I 06 Problem~ in Combinatorics and Graph Theory
for j E Q'P.
Ai \+ K:JP
I
IKI=p-!.J IK' =p+ 2
-IPI
( -I
Sum these numbers with respect to all the subsets PcQ with == P elements:
I
L
IPI=p
I(
PCI.:
IjIKI-IPI I
IPI=p
2.4 Let At denote the set of natural numbers less than or to n which
are multiples of PI' It follows that
n-
q n n n
n- I -+ I
i= J Pi I <;i<j.$fq PiP}
-l)q---
P1PZ ... l'q
Solutions 107
2.5 Let Ai denote the set of the (n-I)! permutations which have i as a fixed
point, and apply the of Inclusion and Exclusion to find the number of
permutations which have at least one fixed point This number is to
ButlAI, nAil n ... n = (n - k)!, since a permutation of the set AI, n' . n
has fixed points in the positions ... , ik , and the other positions contain a
permutation of the n - k remaining elements. The latter set of permutations
has cardinality (n - k)!. BUI k positions ii' ... , ik can be chosen from the set
of n positions in ways. and thus
that
E(n)=in!-I;QI Ail
nt IAil+
I
2.7 Let n P~ ... P~q, where Pb ... , Pq are pairwise distinct primes. The
result can be proved by induction on i l + i z + . . + i q • If i l = 1 and i z = .. ,
= iq then n is a and the sum of the len hand side of the
becomes (J)(l) + (J)(n) = 1 + n - t = n, and the is satisfied if, by definition,
rp(1)= 1. Suppose that the property is true for all numbers for which i l + ...
+ r I, and let n be a natural number such that i l + ... + iq=r. Let
(:: :: ~:),
G3 b) C3
If a3~O, the minimum value for Q3 is r+ 3 2(r+ 1)= l-r, and thus by using
OJ, it follows that the sum of any two of the numbers Cl, b2 , C2 is smaller
than r + 3. It follows that the matrix will be completed in row 3 (except
for the element G3), with elements that are greater than zero. Suppose that a3 ~ O.
In this case one can generate all matrices of the indicated form with row and
column sum equal to r+3, for which the unique nonpositive element is Q3' by
writing the number r + 3 + G3 as a sum of four positive numbers. Again by using
Problem 1.19, it can be seen that for G3 - L ... , r + 1, the number of
solutions of equation (1) is equal to
Solutions 109
1- t
1 + 2! - ' , . ) (1 + 1+t 2 + ' , , ),
These two recurrence relations can be obtained by a straightforward calculation,
2.1 0 Let Y = { Y l' , , , , Jim}, and for every I ~ i ~ m let be the set of functions
from X to Y for which Yi is not the image of any element in X:
Ai= :X -> Y!Yi It fIX)},
It follows that Sn,,,, = mil -IA I u A z U .. , U Ami, since the total number of func-
tions from X to Y is to m". By using the Principle of Inclusion and Exclu-
sion one can verify that
m
s",m=m" I
i= 1
But AI is the set of functions defined on X with values in Y J'i}, and thus
iA11(m -It, and AI (1 AJ is the set of functions defined on X with values in
Y"-{Yi,Yj}.Thus (1A J I=(m- . In general,
IAII (1" (1 (m-k)", where 1~il<i2<"'< m.
Since there are (~) subsets of indices K c( 1. , . , , m: with = k, it follows that
each sum
sn.m=m
n (7) (m l)n+(~) (m 2)"-'" +( _l)m-l (m: 1)-
For m = n. sM,m represents the number of bijections f : X ..... Y with IXI IYi = n.
Hence Sn./! = n!, or
- ' " +(
110 Problems in Combinatorics and Graph Theory
If n < m, there does not exist a surjection from X to Y and thus Sn ... = O.
2.11 Let Z={Yl"" Then, in the notation of the problem,
s..... , = m" -IA I v ... v
For r=m one can conclude that Sn.m.m =Sn ... , since Z = Y.
2.12 The number of words which use all 2n letters of the alphabet A is
to
v ... vAnl=
rlAi,rl . . . Il +, ..
(2)
they can be chosen in (~) ways, and hence the number of words which do not
contain two identical letters can be written 0). and (3) as
2.13 digraph which does not contain a circuit has at least one vertex
x at which no arc d+(x)=O] and at least one vertex y in which no
arc terminates [Le .. d-( y) 0]. Denote A; the set of with n
which have labels from the set p ..... n} which do not contain a circuit, and
which have the property d O. It follows that
Qn=iA 1 VA 2 v'" vAni
.AI-
. ,
n
1 I"n Ai'II
i: 1
But
since the number with n - k vertices labeled from the set {I, ... , n: "
[i 1. · · . , ik } and without circuits is equal to an-k. and since the vertices i l • . . . • ik
have the property rOd rU k ) that is. they can be joined by arcs
having a uniquely determined orientation to the other n - k vertices in
= ways.
Thus each term of the sum is equal to 2k(n-k)a n _b and the sum contains
(~) such terms. This observation yields the recurrence relation for the numbers an'
the given formula one finds that at L Q2 3, a3 = 25. These values
can also be obtained by direct computation. W. Robinson, New Directions
in the of Academic New York-London. 1973,
2.14 Let (n; 1m) denote the number of arrangement schemes orn identical
objects in m distinct cells; it follows that . 1m) = "1). In if
the empty cell is included. this number represents the number in which n
can be written as a sum of m non-negative integers. Problem 1.19 one
can show that this number is also to r"':-
l
).
+
This is equivalent to the expression sought, since n;: 1 S,
A. MacMahon, Combinatory Analysis 1, Cambridge
191
2.15 The number under investigation is of the form A(l)" 2)"··· nAn; 1m).
For example, for the number 150=2 x 3 x 52 one has
2 1. 3 _
A(1 2 • 1 ) -
(3)(1)2(2) (3)(2)2(3) +
1 1 2 2 1 2 =21
(n l2
q
- I -uP - (~) (2 q
- 2
and thus M(p, q) has the Since the K p,q and Kq,p are
isomorphic, it follows that M(p, q) M(q, [1. Tomescu, J. Cambll/morial
B28(2} (1980), 127 - 141.J
2.17 From Newton's binomial formula one obtains
and
----_._._- -.
11'1 Problo:ms in Combinatorks and Graph Theory
of representations of m as a sum of II
k E K, it follows that
I (m-i-l\
In A·=
IcK,I n - i - 1 /I
and the set K may be chosen in ways such that IKI i. One thus finds that
I niCK =0 if K :: {1, ... , because m ~ n + 1, and hence one can assume that
i:r;; n - 1. The last equality becomes
(m-n 1\="-1 (_
n-I ) (n)(m-
i l1-i
i
1
2.19 Let A 1 ..... n and B=(bij)I.}=l .... n be two compatible n-by-n
matrices and suppose aij + bl ) In this case at least one of the terms is
nonzero, say (Jij, and thus Xi:r;; Xj for l:r;; I, j:r;; n. It follows that A + B is a com-
patible matrix.
Let AB = C = 1, . . . . ' and suppose that clj =1= O. Since
and and by
°
there exists an index k such that aik =1= and bkJ'r 0. Thus by definition Xk
. It follows that the matrix C is compatible.
Let A a matrix of order n. By the
elements of V one can assume that Xj implies i Thus A is upper tri-
angular, which implies that del A= 1 ali 1=0. It follows that ail"f0 for
1=1, ... , n.
Let B 1. .... n = A I, and suppose that B is not compatible, that is,
there is an element Xi which is not less than or equal to x) and such that bij 1=0.
Choose an index i which is maximal relative to this property for a fixed index j.
Since XI is not less than or to Xi' we have I> j and thus
n
L
k=l
aikbkj=O.
z .. =51 if Xi:r;;Xj.
I) to otherwise.
since mu=zu 1 and mik:fO implies Xj~Xk' and hence tki 0 for k i. Further-
more, sij= I;= m"Zkj=O for Xi which is not less than or equal to Xj' since
mik:fO and imply that Xi~Xk and Xk~Xj; thus Xj~Xj, which is contrary
to the
For Xj<xJ one has su=Ixl<,,,,.;x;mlk=O by use of the third condition
,rn,,.,,..,,,,,ri upon the Moebius function.
Zij= J for i= 1, .... n. it follows from the solution of the
problem that the matrix Z is nonsingular and similarly that M =Z 1 is a
compatible matrix. Thus the desired function p(x, y) is defined the matrix
Z-1
Since M is compatible with the partial order relation defined in ~: it follows
that y) =0 if x is not less than or equal to y. from the fact that '~ii 1
and 5ij = 0 for every i:f j, one can conclude that the function p defined by the
matrix Z-1 satisfies the last two conditions on the Moebius function.
2.21 Observe that the function j.J. can be obtained inductively as follows:
fila, a) = 1, and if y) is defined for every y such that a ~ y < b, then from the
equation
{.l(a, y) =0
b
Thus p(a, x) can be calculated for every x which satisfies a~:>:~b, the
previously calculated values for elements which lie between a and x.
(a) Now we show that p(X. y)=( - -IXI for every Xc Y S. defini-
tion set Y) =0 if X 1. y, In fact, the function thus defined satisfies
I
A C),CB
IAI (IBI: IAI) ( ijk
\1 forA=B,
=<
to for B::JA and B:i:: A.
(b) We show that
{.l(X, Y)=j.J.
p,(a, Y) L /i p,(dj.
a }' b a I}' I b
the rest of the terms are zero by definition of the function /irk). Thus
p,fd) = (-1)$ = O.
it follows from the definition of the function Il(k) that /lta, b)=O if a does
not divide b.
(c) In this case every interval of the form
[a, = {z i a~z~b}
is a chain (totally ordered Now calculate Il(a!, an) for the chain
a! <a2< ... <an'
F(p) = I(q)·
p
It follows that F(p) represents the number of functions I :B .... C which have as
their kernel a partition q of B such that q,;;; p. Thus q can be obtained by taking
the union of some classes of p, which that
since the desired number is equal to the number of functions on a k-element set
and with values in an x-element set. It is thus the case that
L [x]C(q)=
q<;p
For k == 1 the coefficient of x in the left-hand side is equal to 1-1(0, I), since the only
118 Problems in Combinatorlcs and Graph Theory
partition q with c(q) = I is the partition 0, while the coetftcient of x in the poly-
nomial
[x]"= -l)"'(x-n+l)
is equal to ( Ir -lj!. Since this identity holds for all positive integral
values of x, it follows that the two polynomials in x are equal and hence f'(0, I)
=( l(n-l)!. W. Frucht, Gian-Carlo Rota, Scienria, 122 (1963),
111-1
2.24 Let U=Z J=(Uij)l.j=l ..... n' where
ifxi<Xj,
otherwise,
But V has only n and hence such a walk does not exist. One thus
obtains the matrix
(1+ -u+u 2 - + ... +
and therefore
J-U+
that M is the matrix which defines the Moebius function of the set V. It follows
that
n
1)= L (-lYu~:1>
s= 1
where u~~ 1 is the element of the matrix Us located in the row which corresponds
to 0 and in the column which to 1 of V. Finally one finds that
u~!t = L . ,. Ui,.,! = 1 == Ps .
(i,.,."i,J) O<X'l<'" I'" I
since the remaining terms are zero according to the definition of the matrix U.
2.25 Apply induction on n~2. Un 2 then S!-S2~ UA21~Sl' or
IAI +IA21-IAlnA21~IAIUA21~A11+
The side of this relation is obvious. and the left-hand side is a con-
sequence of the Principle ofInclusion and Exclusion.
Solutions 119
Suppose that Bonferroni inequalities are valid for each /1- I subsets of X.
One can see that
1 Ai!,,;;,h-l(_lt- 1 L Ai!
KL{l. .... n-l:
IKI=k
and
Anl!,,;;,h-2 (-Ilk .. L
Kelt. ",n-l i IE/(
,!n. A/nAn!,
IKI =k
CHAPTER 3
s(n, k)=( -k
P(K). ( 1
P "n- i;
=n-k
(1) one finds that the coefficient of Xk in the expansion of [x]" is equal to
L ),
K .... n-Ij
="-k
that is, it is !s(n, k)l.
3.2 Part is called Vandermonde's formula. 1t can be proved by induction
on n. It is clear that the assertion is true for n = l. If (a) holds for n ~ m - I. where
m;;;:.2, then
+ Y]m= + I(X+ Y m+ 1)
m- 1
I
k=O
_\{x k+y-m+k+l)
=
m-l(m~ 1) [X]k+ \[Y]m .-1 + m 1 (mk 1)
f [(mk - 1I) + (m -k 1
<=0
-k
since [111],,=0 for m+ 1 ~k~n. It must be shown that the polynomial x"
S(I1. is zero.
But this polynomial has at most to n -1, since n)
l) ... (x - n + 1) contains the term x", but the other terms in the sum
do not contain x". The equality, which is valid for m = 1. 2.... , n. shows
that this of at most n-l has at least n distinct roots, and
Solutions 121
The same argument can also be used to prove (b) by considering monotone
functions I :X->R where IXI=n, IRI=r, and the set Ris totally ordered.
monotone function I X -> R is in one-to-one correspondence with an increasing
word of ... I(x n), where X = {Xl' ... , xn) and I(xtl:;;;;
and the number of these increasing words-called com-
of r take n-is given by the formula
+n-l)
(Problem U8).
If the number of distinct letters in the word I(x 1) ... j(xn) is k (1:;;;; k ~ n), then
these k letters can be chosen from the set R in (~) ways. The number of increasing
words of length n with exactly k letters is equal to (:= :l.In fact, if these k letters
are a! <£12 < ... <Uk' then the increasing words have the form Cl C2'" Cn
whereci C2 - =a tt ci 1 =ci 1 +1= ... =cI 1 - 1 =a2, .. ·, ci k_
ak. and 2:;;;;1 1 <i2 < ... <ik - l :;;;;n. Thus the number of
words of a length n with k letters is equal to the number of sequences
2~il<i2<'" <ik t:;;;;n,thatis,itis to _:l.It follows that
or
[:r =Jl =D'
[rJn=JI ~: (:=~)
This equality is valid for every r ~ 1. which implies
3.4 To every surjection I of the set X = onto the set
y = {Yl, .... .I'm} there corresponds a partition of the set X into m
r I(YI)U I l(yZ)U'" U
The order of writing the classes is not taken into and hence there
are m! surjective functions from X to Y which the same of
the set X. In other words, if the elements of Yare permuted in m! different ways,
then one obtains ml different surjections by with the
f. However, all of them will generate the same of X into m classes.
Two surjections which differ by a permutation of the elements of Y cannot
generate different part itions, and furthermore every of X into m classes
can be obtained in this manner. It follows that
Sen, m) Sn m k)".
m '
Consider the set of the m into m -1 classes of a
set with n elements Xl, ..• , X •. One can obtain m 1) partitions into m
classes ofa set with n+ 1 elements X l " " ! X n + 1 to each partition a new
class consisting of the element X.+ • The element X n ... 1 can be added to
122 Problems in Combinalorics and Graph Theory
Sin, m) m=1 2 3 4 5
n==1 0 0 0 0
2 I 0 0 0
3 3 1 0 0
4 7 6 1 0
5 15 25 10 1
in a class and the order of the classes in a partition are not it follows
that the same partition generates t1 (21)k, . .. (n different permutations of
the set X by the elements in each class. For each of these permutations
one also obtains k I! k2! ... kit! by the classes with the
same number of elements among themselves. It is easy to see that in this way
one generates without all the n! permutations of the set X. from which
it follows that
Perm( ! ..•
3.7 Use the identities of Problems 3.2 and 3.3 to express the polynomial
[x + rJ" in two different ways:
k
G)xll "
It
n-k
=ktG) (~)
It
Equating the coefficients of xiyl in these two yields the first identity.
In order to obtain the second recurrence relation one must (X +
<;0
k)
k
G) [XJi[ i'
(~)x"yn
It k n-k
(x + y)" k== 11
(:) S(k. i) [x ]; SIn - k. j )[J'Jj.
and
n k
x"= S(n, k)[X]k L k) L s(k,
k=O
+ 1, k)= k-l),
In the case of one can equate the coefficients of (x Ia)k in the expansions
Solutions 125
n-1
Sin, k)(x I a)k xn = x n - 1 • X == Safn -1, k)(x Ia)k(x - ak + Uk),
(xl =--.(xla)n+l
x-an
(!+a~+a;+
x x
···\)txlaJn+l'
/
The proof of (d) is obtained by iterating recurrence relation (b) for different
values of n.
Recurrence relation (e) follows from repeated application of (a) for decreasing
values ofn.
In order to establish (f) one can write
.=0 m=
n (n
= m= k~O S,(n,
[L. Com let, C. R. Acad. Sci. A275 (1972),747-750.]
n
3.11 Let fk= Sa(n, k)t and observe that
1+
Thus
·x
and fo::: L a~tn
""'0
3.12 Let X = The set of partitions of X into k classes which
contain at least i elements can be written in the form P! v where PI repre-
sents the set of partitions of X into k classes which contain at least i elements
for which the element Xn belongs to a class with more than i and
represents the set of partitions with the same property for which Xn belongs to a
class with exactly i elements.
It follows that S;(n, k) = IP + II since
the partitions in P I are without by from the
of the set {x 1, . . . , X n - d into k classes of cardinality greater than or
equal to i. One then adds, in turn, Xn in k ways to each class. The partitions in
can be obtained by considering all the :) subsets of X of cardinality i
which contain Xn as a class of the partition. To this one adds, in turn, the
Si(n i. k-l) partitions of the 11 i remaining elements in k-l classes of cardi-
nality greater than or equal to 1.
In order to justify (b), consider the set of all functions f : ->
{1, .. ,k} such that IF l(sJI for every 1 ~ s ~ k. It is clear that the number of
such functions is to the number of arrangements of a set of n in
126 Problems in Combinatoric. and Graph Theor)
k boxes such that the sth box contains), objects for s == 1, ... , k, which by Problem
1.15 is equal to n !;jl ! ... A!. By summing these numbers for all solutions of the
)1 + ... +A n with js~ i for s == 1, ... , k, one obtains the number of
f which If-l(S)I~ i for every s= 1, ... , k. This number must
be divided by since the order of classes in a partition of X is immaterial.
3.13 Every into two classes X AvB is determined
the set A, which is taken to be different from X and 0. Thus the number of
choices for A is equal to the number of subsets of the n-element set X minus two.
It follows that S(n. 2) (2" 2)/2 = 2"- I - J. since the order of the classes in a
partition is not taken into consideration.
every partition of an n-element set X into n - 1 classes contains
a class with two elements and n - 2 classes with one element. The two elements
can be chosen in (~) ways.
In order to prove (C), first show that the function of
the Stirling numbers of the second kind is
I! +
... and =k! J
0 (- l)k P e) Px
e ,
k! P~o (-1
in view of the known expression for the Stirling numbers of the second kind
(Problem
Thus k k) is the coefficient of x" in the expansion of (~ 1)k, multiplied
by n!. It follows from this that 1-1 2) + 2!S(n, 3) 3 !S(n, 4) + ... }
is the coefficient of x" in the of the sum of the series
(1)
Since x - + - tx4 + ... = InO + one can conclude that the sum (1)
is equal to In( 1 + eX - I) = x, which establishes the identity for n ~ 2.
3.14 In urder to obtain an for ern, 3) one must determine the
number of ways in which one can express X = A u Bv C where A, B, Care
nonempty and each element of X to two of the sets A, B, C. In fact each
element .\ e X can belong to the sets A, B, or C in = 3 ways. In this way one
obtains 3" for X in the form A v Bu C, three of which do not satisfy
the given conditions, since they to situations in which either A, B
or C is Thus the number ofbicoverings of X with three classes is equal to
(3n - ! 1 -1), since the order of the sets A, B, and C is not taken into
consideration. Comtet, Studia Sci. Math. Hung., 3 137-152.J
3.15 Consider a new element z q: and in each of the set .Y v
suppress the class which contains the element :::. One obtains in this way all the'
partial partitions of X without for the case in which the
partition of X v consists of a
Bn+ I -1.
3.16 Let be a sequence of real numbers, and denote the sum
of the series o un[xJn for those values of.\ for which the series is convergent.
Now define operator L by the relation L(j(x)) = I:~o an, if this series is
where = 1. Thus
'" m
lim I anS(n, S(n, k))
m- <=0
In
lim
m-
I
11=0
a"B n
n=
aMB n •
In order to obtain the desired formula consider the Taylor series expansion
r"x"
exp +
n
Substituting e' = u+ 1 and using Newton's binomial formula
yields
oc
e"=(u+l)"= I un.
"=0 n
It follows that
since SIn, k) = 0 for k> n. In view of the for the Stirling number of the
second kind (Problem 3.4) it follows that
l
~ (-It-j(~\r=
ct.
Bn k
k! \)) j!
The difference between the number of partitions of an n-element set into an
even and an odd number of classes is to
n
l),S(n, kl (-ll G)r
3.19 In order to find the expression for the number k), observe that
the subsets which contain the number n cannot contain either 11 -1 or l. The
number of such subsets is equal to - 3, k 1), and there are -1, k) which
do not contain n. In both cases it is assumed that the sets do not contain two
consecutive integers modulo n + 1.
For example, the subsets which contain the number n are obtained from the
f(n - 3, k -1) subsets of X"-{l. n -1, n} which have k -1 elements and do not
contain two consecutive [n this case the number n is added. In view
of the fact that every k-element subset of X which does not contain two consecu-
tive modulo n+ 1 to the union of these two disjoint sets of
k-element subsets which do not contain two consecutive it follows that
k
= + n (n /).
The Lucas numbers can be expressed as a function of the Fibonacci numbers
as follows:
1+
for n ~ 2. One finds that L 1 = 1 and definition, the recur-
rence relation for the Lucas numbers the fact that L2 = 3.
3.20 First we show induction on n that
c o
1)"+ - (FFM
1_ n+l
is satisfied.
Supposing that (1) holds, one can show that
and thus is true for every n. By equating the determinants of the matrices in
both sides of (1) one obtains the desired recurrence relation.
3.21 It is clear that the number Un is also equal to the number of representa-
tions of n as a sum of the numbers 1 or 2. Two representations are considered
130 Problems in Combinatorlcs and Graph Theory
to be distinct even if they difrer only in the order of their terms. For example:
2 1 + 1,
3=2+1=1+2=1+1+1,
4=2+2=2+1+1=1+2+1=1+1+2=1+1+1+1.
The first term is 1 or 2. In the first case the number is
to u. _ 1, since the remainder of the terms are equal to 1 or 2, and their sum is
equal to n -1. In the second case the number of is equal to
Un - 2' It follows that
(1 )
with initial values Ul == 1, Uz == 2. If one takes Uo == 1, the result is the sequence of
Fibonacci numbers, and hence Un
In order to solve equation one can use the characteristic equation
,2=r+ 1.
which has solutions '1 =(1 +.j5)/2 and '2 .j5)/2, and thus the
solution has the form
where the constants eland are determined from the initial conditions
Uo 1 and UI 1. One thus obtains the system
C\+ 1.
It follows that
xj(x) L'" =
n=1
j(x)-xj(x)-x 2j(x) = + -\
n=
from which it follows that
f '( x)=-l_-x--::'
1
Solutions 131
whereC 1 =1.
As in the statement of the problem, let
+ +
It follows that
,-1
+ + +.,,+ ( L
k= 1
where a>O,
This series converges for every x which satisfies < a. If rx E 1M is a
integer, then only a finite number of terms of the series are different from zero
and the expansion is the binomial formula named for
Newton. (This formula was not actually discovered by Newton, Mathematicians
from Central Asia such as Omar Khayyam knew it much earlier. In western
Europe, Blaise Pascal also used the result before Newton. Newton
proposed the to
In order to (1-4x)1/2 in a series of powers of x, let y= -4x, rx
and the binomial (y + , It turns out that the coefficient of x" is
equal to
2(211-2).
11 n-l
3.24 Determine the number of sequences of letters which contain the letter
a k times and the letter b m times. and which have property (P): For every i,
1 ~ i~m + k the number of letters a among the first i letters of the sequence is
greater than or to the number of letters b. It is clear that the number of
these sequences is nonzero if and only if the condition k;?; m >0 is The
number of sequences of letters which contain the letter a k times and the letter
b m times is equal to Plm, k)=(m+k)!/m!k! If one determines the
number of sequences which do not satisfy (P), the desired number is
obtained by subtracting this number from (,";k). It will be shown that the
number of sequences formed from m letters band k letters a which do not satisfy
(P) is equal to P(m-l, k+ 1) 1)' It is equal to the number of sequences
formed from m 1 letters band k + 1 letters a. The proof of this property follows,
Consider a sequence formed from m letters band k letters a and which does
not (P). There exists a position numbered 2s + L where s;?; O. such that
Solutions 133
(m + k) _(m + k) = k- m+ 1 (m + k\).
\ m m-l k+1 m
For k=m=n-l one has
Cn=~ (2n-2).
n n-I
It is clear that this number represents the solution to the problem, since if 1
is replaced by a and -I by h. then condition (1) expresses the fact that the
number of a's is at least equal to the number of b's in the first k positions for
1:::; k:::; 2n - 2. Condition (2) expresses the fact that the number of a's is equal to
the number of h's and both numbers are equal to n-1.
134 Problems in Combinatorlcs and Graph Theory
3.25 It will be shown that there exists a bijection from the triangulations of
a convex polygon with n+ 1 vertices to the set of of n
factors in the order Xl' X2,"" x.'
Let Al A2 ' , . J be a polygon with n + 1 vertices, and traverse the sides
of the polygon from AI to and so on. until 1 is reached, One obtains a
parenthesized of n factors by using the following rules:
(1) When aside, write a new factor xiin the order Xl'" .,x.,
When one arrives at a vertex which is incident to some
of the triangulation, write a number of
to the number of diagonals which have an
vertex and whose other endpoint has been traversed. Write a
number of opening parentheses equal to the number of diagonals
which are incident to this vertex and whose other endpoint has
not been visited.
In this manner it turns out that for the triangulation of the convex polygon
with vertices illustrated in 3.1. one obtains the following product
of seven factors:
(x
It is clear that this is injective, In order to show that it is
surjective, consider a product of n factors in the order Xl' ' , , , X n •
This product contains n 2 opening parentheses and n 2 closing
Each corresponds to a unique
each of opening and parentheses consider the first letter Xi which
occurs to the of the opening parenthesis and the first letter Xj which occurs
to the left of the closing parenthesis, Draw the diagonal AjA j + \'
Since each pair of parentheses contains two factors in its interior. and since
the are correctly it follows that the n - 2 of the
polygon constitute a triangulation for it. Now apply the indicated corre-
spondence to this triangulation. The result is the product
of n factors. which establishes that the transformation is a bijection.
Thus the number of triangulations is to the Catalan number (Euler),
As
Fig. :1.1
Solutions 135
A(5,5)
1 2 4 5
Fig, 3.2
136 Problems in Combinatorics and Graph Theory
Condition (1) expresses the fact that the path d cannot pass through
points which are located above the diagonal. and condition (2) implies that
there are n horizontal and n vertical segments. The path must therefore ter·
minate in the point A.
It follows from Problem 3.24 that the number of sequences of 1's and - 1's
with the is to
1 (2n)
n+l ,n .
This observation ends the proof.
3.27 For n=4 the number of solutions is equal to 2. Now let n:;:;. 5. A vertex
of the polygon can be chosen in n ways. Join the two vertices which are
1-'1
adjacent to l'\ by a diagonal d\, Consider the which has side d l and a
vertex in common with the which is difrerent from VI' It must have a
side in common with the convex Thus the third side must be one of the
two which join a vertex located on d l with one of the of
another vertex on Thus a second diagonal d2 can be chosen in two ways.
Similarly, if d 1" " , d;(i<n-3)have been chosen, then there are two possible
choices for 1- There are thus n 2"-4 ways of selecting a vertex Xl and a se-
quence d\} 3 of diagonals with the desired property,
Each triangulation obtained in this manner has two which contain
two adjacent sides of the polygon, and hence each is counted twice. rt follows
that the desired number is n 2"- 5 for every /1:;:;' 4.
3.28Let ,fj(n + 1) denote the number of sequences with n + 1 terms and
al It follows that f/n+ 1)= +fj~ I and 10(n+ 1)=
It will be shown by induction on k:;:;. 1 that for every k < n the following rela-
tion holds:
Solutions 137
fo(n) = I -k), (1 )
j<;k,'2
1. +1 2j (n-Ik+ 1)),
where
(1 +x+ 1 =[ I k;;.O
elm, k)x k ] (I +x
yields
2)+ 2}
=4g o(n 3l+5YI(n 3)+3Yz(n-3)+Y3(/J-3j
=9g o(n -4) + 12g!(n - 4)+9g 2(n -4) +4Y3(n - 4) +g4(11- 4)
138 Problems in Combinalorics and Graph Theory
In
(1)
(2)
= +l,j)- +I,j-
j)-e(k,j- {dk,j)- j-
=e(k, k)+e(k, k+ 1)
for every 0, since gk -ll) = 1. [L. Math. Naehr., 49 (1971),125-147.]
3.30It was shown in Problem 3.24 that the number of sequences (Xl' X2,
... , Xln) with terms equal to ± 1 which satisfy the conditions Xl + '" +Xk~O
for every 1 ~k~ 2n and Xj + ... .;..x 2n =O is equal to {1/(n + 1}}e:). Observe
that there exists a bijection from the set of sequences X = X 2' ..• , X 2") to the
set of sequences a al' , . , , (l2n+ d which satisfy the conditions. The
mapping is defined f(x)=a if
3.31 Consider a sequence which satisfies the condition of the problem, and
suppress all terms to n. One obtains a sequence of k terms, where 0 ~ k ~ r,
which satisfies the condition. For i n the condition implies that the
number of such sequences is zero for r = n. Start with a sequence of
the numbers 1. .... n - L of length k, with the property that at most i -1 terms
are less than or equal to i for i = 1, ...• n -1. One can insert r k new terms
equal to n so as to obtain a sequence , ...• x r ) with 1 ~ XI ~ n which satisfies
the same property. However. the r - k terms which are equal to n can be inserted
in (;) ways, which to the (r~h)=(~) possible ways of the
positions occupied by r- terms equal to n in a sequence of r terms.
Now start with all the sequences of length k=O, ... , r composed of the
numbers 1, ... , n -1, which satisfy the property. By inserting r - k terms
equal to n in all possible ways, one obtains, without repetitions, all the sequences
(Xl' ... , with I ~xi~n which satisfy the same property. This implies the
recurrence relation
f(n,r) (n-
If n r it has been seen thatf(n, r) = 0, and this coincides with the value by
Let n>r, and suppose that f(n-l, (n-l _1)k-1 for every
k ... , n -1. In view of (1) it follows that
= r G) (n- kt G) 1)'-1
=nr -r " (r-l)(n_
k-J
= n' - rn'- =(n r)n,'-l for every r = 1, ... , n - I.
However, for r n it has been seen that formula (2) is also true. Thus is true
for every n~ r. [H. E. Daniels, Proc. Roy. Soc., A183 (1945), 405-435.]
Sn=
r JUO'~III' In LomomaWflCS and Graph Theory
or
"
=}: for n';:;:1.
n k 0
x"
= 1 + s(x)eX,
and hence = 1/(2- s(x) in a power series:
1
s(x)
*
one can choose any of the q" - 1 vectors other than 0 in the space V. Each vector
v 0 generates a of dimension one which contains q vectors. There
thus exist q" -q vectors, each of which with v forms a system of two
linearly independent vectors and thus can be chosen as the second component
of the system. Let w be one of them. The pair {v, a two-dimensional
subspace which contains vectors. Thus there exist q" _q2 vectors which are
linearly independent of v and w, and any of them can be chosen as the third
vector of the system. Continuing in this way, one has
Un.k _l)(qn_q}'" _qk-l).
3,36 For k=! one has f(n, I; for k 2, a2=n and a2-al is an even
numoer, and hence 2) = [(n 1)/2J. One can prove (a) by induction on n.
This formula is valid for n 2 and n = 3 for all 1, if by definition (g) = 1.
Suppose (al is true for n ~ m L It is clear that k) = IV tfm, kl+ IV 2fm,
where IV1(m, k) is the number of sequences, the same conditions. for
which al 1 and IV 21m, k) is the n umber of such sequences with a 1 ~ 2, It follows
that k)=f(m-l. since the sequence
=(P~l
+(~)+(P:l)+(P;l)+ ... +1,
Gn + 1 = ~1)+ +(;)
+(P;l)+(P:l)+ .. +1,
2 (~)+ G)+(p; 1)
+(P;1)+(P:2)+ ... +1.
But
Solutions 143
(p~ 1)= +
Gn (p 0 +(n+(i)+ 3
1) + (p+4 + ... + 1.
CHAPTER 4
4.4 Consider the set {k+ 1, k+2, .... 2k+ I}. which contains k+ 1 numbers.
The sum of each of distinct numbers selected from this set lies between
2k + 3 and 4k + 1, and is therefore not divisible by 2k + 1. The difference of each
two different numbers lies between 1 and k and is therefore not a multiple of
2k + 1. Let the number be denoted It follows that n(k):;t k + 2.
Let A be a set which consists of k + 2 integers. If there are two numbers in A
which have the same residue modulo 2k + 1, then their difference is divisible
2k+ 1, and A satisfies the given condition.
In the opposite case, all the k + 2 remainders of the num bers in A modulo
2k + I, are pairwise distinct. Thus the set of remainders is a 2)-element
subset of the set {O. 1, ... , 2k i·. One can also consider the remainders as forming a
(k+ subsetofthesetM {-k,-lk-l), ... ,-l,O,l" .. , since
every p = (2k + l)q m, where q is an and m EM. However. for
every choice of k+2 numbers from M, there will exist two among them whose
sum is zero, since otherwise one would have IMI:;t 2(k + 1) + 1 =2k + 3. Thus A
contains two numbers whose sum is divisible by 2k+ 1. But IAI k+2, from
which we deduce that n(k)~k+2. From the two opposite inequalities one can
conclude that n(k) = k + 2.
4.5 Sum the inequality for j = 1, ... , n to obtain
+
If one further sums over i = L .... n. then
IMI + 2nlCkl:;t n3 •
by summing on k it is seen that nIMI+2nIMI:;tn 4 and thus IMI:;tn 3
The lower bound is attained if n 0 (mod 3).
Suppose now that ;MI = n3 /3. and define a partition of M,
M A;u u"'uA~u U"'U u"'uA'iu u"'uA~,
= =x~
Solutions 145
is
4.11 Fix a subset C E F. The function f : F -> P(X) defined by f(A) A c:, C
is In if At 6 C = c:, C, then from the properties of 6 one can
infer that At t 6 C) 6 C c:,C)6C= and thus A l =A 2 • It follows
that the num ber of sets of the form A 6 C, where A runs the family F,
is at least to = m,
4.12Let S = {I, . , , , n}, and denote Ai the set of families of nonempty,
dist inct subsets of S which do not contain the element i E S, Since the
number of families of nonempty pairwise distinct subsets of S is equal to
1, it follows that
IAtv"'v IAinAjl+-"·
1.::>' It
Solutions 147
Since In lEK '~I, where IKI=k, the formula for follows immedi-
if one takes into account the fact that the set K can be selected in G) ways
as a subset of S. [L. Comtet, c.R. Acad. Sci. A262 (1966), 1091 1094.]
4.13 Let A=(Sjh';j<Ck be an irreducible of the set S. [t follows
from the definition that each subset Sj contains a nonempty subsel T; of S.
consisting of elements which do not belong to the other subsets of the
Let T = I It follows that the subsets ... , Ik form a partition of the
set T. If == one sees that k ~ i ~ n and all irreducible of S can be
obtained without in (he following way every i): For each of the
C) choices of T as a subset of S, consider the SU, k) partitions Tl u .. u Ik
from among SI "T
of T. Each of the II i elements to a family of subsets chosen
1 , ••• , Sk "'-..Ik which contains at least two sets. and thus the
number of possibilities is 2k - (~) - (~) = 2k - k -I. From this observation the
formula for I(n, k) follows.
The expression for /(n, n -1) is obtained by using the fact thai S(n. n - 1) == (;).
In order to obtain the expression for I(n, suppose Ihal y 11: S. Then there
exists a from the set of partitions with three classes of Su { onlo the
set of irreducible 2-se! of S, defined as [ollows: The partition S u
== U S2 uS 3 is associated with the irreducible of S whose sets are
SI U(S3 "{ r}) and S2U(S) "{y}), where S3 is the class of the partition which
contains y. It fonows that I(n. 2)=S(n+ 1. The same conclusion holds if one
applies the recurrence relation (a) of Problem 3.5. [T. Hearne, C. Wagner,
Discrete Mathemalics. 5 (1973),247-251.]
4.14 Let t be the smallest integer such that [t 2 /4J > 11. Consider the follow-
sets of natural numbers:
(
t
J' where 1~j~2
or f(n)~ 1.
will be a minimal set in the family {AI" .. , An} "'-..{ ... , Ai ,\' which is not
the union of two distinct sets of the family (AI" ... , . One can select AI,. I
148 Problems in Combinatorics and Graph Theory
in this way if n-s> since there are (~) unions of two distinct sets in
the family ... , Ai.}. This construction defines a subfamily {Ai,> ... ,
where r+ )f2~n, that r~-f2n-l.
The equation Ajv Ak i, j, k E . . . , i,} are pairwise distinct)
cannot hold, and thus family ... , AI,} is union-free. In fact, if were
selected after and A j • this equality would also contradict the construction,
for otherwise it would follow that Ak was not minimal when it was selected,
which contradicts the definition of the family {AI, •... , AiJ Erdos. S. Shelah,
Theory and Applications, Proceedings of the Conference at Western
Michigan University, 1972, Lecture Notes in Mathematics. Springer-Verlag,
1972.]
4.15 For every XES let M(x) {i Ix E }. It can be shown that for every
subset Tc {1, 2, ... , n} with ITI = n - k + 1, there exists XES such that M(x) = T.
In fact, since U ltT Ai is a union of k-1 subsets of the given family, it follows that
there is an element XES which does not belong to this union. One can show that
M(x)::::: T. By construction M(x) c T. If M(x) is a proper subset of T, it follows
that IM(x)1 ~ n k. Thus there exist k sets which do not contain the element x.
which contradicts the hypothesis that every union of k sets of this family is
equal to S. It follows that M(x) = T, and thus the function which associates with
each element xES the subset M(x) has a restriction defined on U c S,
with values in the set Pn-k+ l({l, ... , It follows that
L
n
i= I
=(n-k+ 1 =(n k+ 1) en)
and it can be seen that
== n nk+ 1 (n
k-
(n 1)
== k-l
for every 1 ~ i ~ n. This deduction uses the fact that in this case = n k+ 1
for every XES.
4.16 Let me be the smallest natural number with the property described in
the statement of the problem, and set U~= 1 XI = Y. Since iXil = h for 1 ~ i ~ k
and Xjc Y, it follows that k~(I~') and thus Iyl, or mo~min In orderto
prove the opposite inequality. let Y c X with IYI =mo. Since h ~ k, one can
choose h-element subsets ... , X. such that U~= 1 Xi == Y. In this property
follows from the inequality mo ~ kh. however. that mo > kh. Since
h !) < k by the definition of mo, and since rna ~ kh + 1, one can conclude that
Solutions 149
4.17 Let YcX with IYI=p. The set Y may be written as the union of k
sets, Y = A I V .. , V Ak , in (2 k -l)P different ways. In each of the p elements
of Y can to 2' -1 nonempty families of subsets A I' . . . . . [t follows
from the fact that Y can be chosen in different ways. that
I)
I)
,
which is the desired result. One should also note that each or the indicated sums
contains 2"" terms and each of the subsets A I ' ... , Ak can be selected from among
the subsets of X in 2" ways. [1. E American Math. Monthly, 86
(1979), 223.]
4.19 Suppose that a filter basis S contains the sets A I' A l , .. , It follows
that there exists Bl E S such that BI CAl n Similarly, there exists ES
such that c Bl n A 3 , and so forth. Finally there is a Bp_ I E S such that
I cB p _ 2 nA p and c A 1+ 1 for i= 1, ... , p-l. It is also the case that
c , By construction,
Bp- 1 cB p - 2 c ... cB 2 cB 1,
and hence there exists a set B ! E S such that B for i = !, . , . , p.
Thus every filter basis S has form: There exists Bc B
such that
S={B,Bv .,Bv (1)
The number offilter bases for an arbitrary n-element set is obtained by summing
these numbers over k=O, ...• n-1.
4.20 Let X be the union of all the elements of the sets and Bj where
1~ m. Suppose that X •... , and let (x P(1)' •.• , be an arbitrary
permutation of X.
There exists at most one index i for which each element of the set AI has an
index smaller than each element of the set B;, In fact. that this is not the
case. Let be two indices such that each element has an index smaller
than each element of B; and each element of has an index smaller than each
element of B j. By the hypothesis, there is an element x Plk) E Bin A 2 and an
element xp(r) E B2 n A l' Since xpir) E A 1 and Xp(k) E B 1. it follows that p(r) < p(k).
But Xpl r ) E B2 and Xp(k) E A21 and thus p(r} > p(k), which establishes a contra-
diction.
Thus there exist at most 11 ! B I ) with the desired property, Let i be a
fixed and consider of the indices of the elements of X such
that each element of Ai has an index which is smaller than the index of every
element of B,. The number of such permutations is equal to
n \ I I( In!
( p+q;I p.q. 11 p-q).=(-)
p+q ,
p
To see this, note that one can choose p+q elements from AlvB; in (p:q) ways.
The first p elements in order of indices are taken from Ai' and the
remaining q elements (with indices than the elements of Ai) are associated
with BI • The indices of the elements from Ai can be permuted in p! ways, the
indices of the elements from Bj can be permuted in q! ways, and the remaining
elements' indices can be in (n - p - q)! ways, Thus in this way one
obtains all permutations of {l, ... , n} with the property that each element of Ai
has an index which is smaller than each element of B i •
Thus each pair (Ai> Bi ) is calculated n P times relative to all the permuta-
tions of indices for which each element of Ai has an index smaller than each
element of B;, Thus the number of pairs (A j , BI ) satisfies the inequality
Solutions 151
since max n, C) = ([n~2J)' It follows that max p ~ ([M~2J)' The opposite inequality
is obtained by considering the family G of subsets of X which contain m = [ni2]
elements.
This proofis due to D. Lubell [J. Comb. Theory, 1 (1966), 299].
4.22 Let n=(x!, ... , xn) be a cyclic permutation of the set X. Consider a
cycle C with n edges, and associate the symbols XI, ... , Xn with the edges of C
(in the usual cyclic order). If an r-element subset A of X contains consecutive
elements in the permutation n. then it corresponds to a subwalk of the cycle C
of length r.
By condition (2) the subwalks which correspond to the sets AI"'" Ap have
at least one edge pairwise in common. Let X be a vertex of C. Associate the sub-
walk having the same symbols with the corresponding r-element subset or X.
It follows that X is a terminal vertex for at most one walk AI' In fact. if Al and Aj
have a terminal vertex in common and i=l=} (and thus Ai and Aj are distinct),
then they must leave X in opposite directions on the cycle C. Since n;:; 2r. it
follows that these walks are disjoint relative to edges, which contradicts the
hypothesis. Now, since each walk A J (j = 2....• p) has at least one edge in common
with A). it follows that one of the endpoints of Aj is an interior point of AI' But
since two distinct walks cannot have a common endpoint, one can conclude
that there exist at most r-l walks Aj with 2~}~p, and hence p-l ~r-l, or
p~r.
Finally, for each cyclic permutation n of X, there exist at most r r-element
subsets of X which consist of consecutive vertices relative to TT. and which
152 Problems In CornbinalOrics and Graph Theory
satisfy (2). Since there exist (n -1)! permutations of X, one can in this
way obtain at most - I)! r-element subsets of X.
In order to determine a permutation for which an r-element set A
consists of consecutive one must first order the sets A and X "A.
Since IAI=r. it follows that each set A is counted r!(n-r)! times. Thus the
number of subsets Aj is p~r(n l)!/r!(n-r)!= :). In order to show that
max C~ ~), consider the r-element subsets of X which contain a fixed element
XI EX. There are :) such subsets.
This proof is due to G. O. H. Katona [J. Comb. Theory, 13 (1972),183-
4.23 Let X Color these elements with one of the two colors
a or b, so no set Ei is monochromatic. The element Xl is assumed to be colored
with a.
that one has colored the elements X 1> ••• ,Xi with 1 < i < n with a or
b so that no subset is and consider the case which occurs
when this process cannot be continued. Thus one cannot color the element
X/+ I with the color a, since there exists a set E c {x b . . . , Xi+ I} with Xi + 1 E E
which has all the elements other than Xi+ 1 colored with a. The element Xi+!
can also not be colored with b, since there is a set Fc{xl!"" xi+d with
XI+ ! E F which has all elements distinct from x/+ 1 colored with b. It follows
that E, F are distinct sets chosen from ... , Em with EnF= +l}' which
contradicts the hypothesis. Thus one can color Xi + 1 with either a or b so that
no monochromatic set is "'r",.rll1ro"ro
It has thus been proved process can continue
until one has colored X with two colors so that no set is monochromatic.
4.24 Color the elements of X with red and blue. Assume that the
"'rlP""t'lrlPt'\T and that each color has probability 1. Let Ai be the
The first inequality is strict, since the events A I' . . . , An are not independent;
they occur simultaneously when all the elements of X have the same color.
Thus the probability of the complementary event is strictly positive, which
shows that there exists a coloring with the desired property.
4.25 Assume that n~ 3 and the family F of three-element subsets of M
has the property that for each two distinct subsets A, B E F one has IA nBI =0
or IA n BI =2. In this case it will be shown that ~ n, which establishes the
desired nr""p·rrv
Thus suppose that F contains only three-element sets which either are
Solutions 153
From this one can easily find the maximal cardinality of the family F as a func-
tion ofn. Let max IFI = f(n). Thus ifn =4k, it follows that f(n) = n, and this maxi-
mum is attained when F is obtained by starting from a partition of Minto n/4
four-element classes and considering all four subsets with three elements of each
154 Problems In Combinatorics and Graph Theory
class. For n=4k+ lone sees that ffn)=n-l, and this maximum is attained by
applying the construction to the set M where x is an
element of M The maximum is attained only in this case. For n =4k + 2 one
has = n - 2. The maximum is attained only in the following cases:
(I) The construction is applied for n=4k to the set M "'Ix,
where x, .v are any two distinct elements of M.
F contains n 2 subsets of type (b):
h. X2}' ...•
For n = 4k + 3 it turns out that fIn) = n - 2 and the maximum occurs only in the
fol1owing two cases:
The construction is applied for n =4k to the set M y,
where x, Y. z are distinct elements of M, and the set {x. y,
is then added to the family thus obtained.
(4) F contains n-2 subsets of type (b), as in case (2).
Then
= +2
j(x)+
d(x)
---< (I)
m n-
since O<IAI<n. Now take the sum or the inequalities of form (1) for every pair
A) with X If: A.
It has been assumed that for every .x E X there is a set A E F such that
x r; A, and thus every element x E X belongs to at least one pair A).
for every A E F there exists an element x E X -4= 0 with the property that
x If: A, and hence each set A E F to at least one pair
F or a fixed element X E X there exist m - d(x) sets A with the property
that x r; A. For a fixed set A there are exactly n -IAI elements x with the property
156 Problems in Combinatorics and Graph Theory
I IAI
(m I
(n- A)n_IAI'
or
It follows from the definition of that the two sides of inequality (2) are
which establishes a contradiction. Thus the assumption that m> n is
false and hence m ~ n. [Po R. P. Algebraic Theory
1973.J
4.28 Let {PI' ... , PM 1 denote a set of n points in the plane. Let at be the
number of points at distance 1 from the Pi for 1 ~ i ~ n. The desired number
of will be equal to
Let C, denote the unit circle with center at Pi' Each two such circles
have at most two points of intersection. Thus for each pair of circles
either 0,1, or 2 points of intersection and the numbers thus obtained, one
finds that the sum is less than or -1).
1t is clear that one needs to consider case in which each number at is
at least to 1. Since each point Pi is an intersection point for exactly (~')
of these one can write
-m.
Ctl aY ~n iii
and thus from the preceding inequality one has
4m2=( i != 1
~ -1)+
or
- 2nm 2n 2 (n - 1 O.
Thus m~(n+ <n.Jn for n';::.1. This is immediate from the
observation that the inequality is equivalent to n> -1. [1978 W. L. Putnam
IJCLlUV.Il. Amer. M mh. Monthly, 86 (1979),
4.30 One can choose a vertex of the triangle in n ways. The two other
vertices must be chosen from among the n 3 vertices different from and not
adjacent to the vertex selected. Thus there are C~3)=(n- -4)/2
f/V"~H~'HH'''~, of which n 4 must be eliminated, since they to the
cases in which the two chosen vertices are a side of the polygon.
There therefore remain (11- 4)(n - ways of the two other vertices.
Since anyone of the three vertices can be selected as the first vertex, it follows
that there are -4)(n 5)/6 ways of a which satisfies the
condition.
4.31 No three are concurrent in an interior point of the polygon.
It follows that each interior point of intersection is determined by the
two diagonals which meet on it and thus by the four vertices of the
which are the of these Thus the number of points of inter-
section of the diagonals inside the polygon is equal to G).
Each AB intersects all diagonals which join pairs of vertices, other
than A and B, in points other than the vertices of the polygon. It follows that the
number of points of intersection other than vertices of the diagonal AB with
other diagonals is to the number of diagonals which do not have an end-
point in common with AB. But this number is equal to
n(n 3)
-3}+ 1 1,
vertices is to
Subtract from this number the number of interior points of intersection. The
number of exterior points of intersection is thus equal to
11(11- -7n+14) -5)
-~~-----~
since the number of interior points of intersection is equal to (:). The formula
is valid for every n;;;?! 1.
4.33 The property is established by induction on the number c of curves.
If c= 1, there are no points of intersection and hence the number of is
to 1. Suppose that the property holds for every family of c curves which
have np points of intersection having multiplicity p, for every p;;;?! 2, where the
multiplicity of a point of intersection is defined as being the number of curves
which intersect in it. By the induction hypothesis these c curves bound
1 + n2 + .. + (p -lJnp +. . closed Consider a new curve, and suppose
that it passes through kp points of intersection of multiplicity p with the first c
curves ~ The total number of points of intersection of the new curve is
Solutions 159
equal to kz +k3 + ' " + kp + . , ., and hence the number of arcs induced onthis
curve is equal to k2 + k3 + ' ... Each portion corresponds to a new with
to the family of c curves. Thus the total number of regions bounded
the family of c + 1 curves is equal to
(1)
1+m2+2rn3+" +(p-l)rn p
+ -l)(np-k p + l + +'"
=l+nz+ + ... +(p-l)n p +'" + +k3+'" +kp+ "'.
This is the number (1) of regions formed by the family of c + 1 curves. The proper-
ty is thus established by induction.
4.34 Suppose that (SI" .. , Sm) has an SDR, denoted (a l ., .. , am!. It follows
that SitU'" uSi,::J{a""", aik}' and hence U'" uSiJ~k for every
k and every choice of pairwise distinct numbers ii' .... ik •
The will be established induction on rn. that the family
.... S'" satisfies the condition for the existence of an SDR. For rn = 1 each
element of S 1 forms an SDR.
further that the is true for every rn' < rn, and let
M(S) (S\)., .• Sm)' Two cases will be studied:
(1) Si,U ... uS;. contains at least 1\+1 elements for l~k~rn 1 and
for each choice of distinct numbers it, ... , ik E {I, ... , rn}. Let at E SI'
(/ \ each time it appears in the other sets of M(S), and denote the sets thus obtained
by S;'. . Thus M'(S) ... , S;") satisfies the necessary and
sufficient conditions for the existence of an SDR, since the set Si, U ... U Sj,
contains at least k + 1 elements and a single element a 1 has been suppressed.
By the induction hypothesis M'(S) has an SDR. The element at together with an
SDR for M'(S) forms an SDR for M(S) in which a 1 is a representative of SI
no appears in the sets of M'(S)].
that l~k~rn 1, and let [il"'" e{t, ..• rn; such that
= k. Renumber the family ... , S'" so that Si, becomes
160 Problems ill Combillatorics alld Graph Theory
becomes ... , Si, becomes Sk' By the induction (S I' S2' ... ,
has an SDR. Denote it by D* (aI' a2' ... ,ak), and suppress the elements of D*
whenever appear in the sets SH 1, SH 2, . . . , Sm. Denote the sets thus
obtained I, St+ 2' . . . , S~. The system of m - k sets M*(S) =(S:+ I> •• "
satisfies necessary and sufficient conditions for the existence of an SDR. In
fact, if u ... u stp ' where i 2 , .•• , ip} C {k 1, ... , contains fewer
than p then S 1 U U··· U S~ u ... u contain fewer
than k+p u ... u =k. By construction SI u ... USk
and hence U' .. U SI, U ... u Sipl < k + p, which contra-
dicts the that M(S) satisfies necessary and sufficient conditions for
the existence of an SDR.
By the induction hypothesis M*(S) has an SDR, which with D'"
forms an SDR for M(S}, since M"'(S) does not contain any element of D"'.
[Phillip J. London Math. 10 (1935),26-30.]
4.35 It will first be shown that if F is a minimal of h-element subsets
of X, then there exists an element x E X such that
h
I{E lEE F, x E E}I~- M(n. k. h). (1)
n
Suppose that the property is that is, for every element x E X the number
of sets in the F which contains it is smaller than k, In this
case, the number of occurrences of the elements of X in the sets of F
[= hM(n, k, is strictly smaller than the of the number n of elements
of X and the number k, h), which is an upper bound for the number
of occurrences of each element in the sets of X. It has thus been shown that
hM(n, k, h) < k, h), and this contradiction establishes (1). It follows that if
G= {E lEE F, x f E}, then
n-h
IG\ M(n, k, h)-I{E lEE F, x EE}I"'- M(n, k, h}.
n
Now it will be shown that M(n -1, k, which will a proof of the
first inequality. For this it is sufficient to show that the family G has the property
that every k-element subset of the set Y == X contains at least one h-element
subset of G.
Let Z c Y with == k. Since Z is a k-element subset of X, it follows that there
exists an h-element set T of F such that T c: Z. From the fact that Z does not
contain the element x, one can conclude that x rt T and hence the set T belongs
to the family G. This observation establishes inequality T.
Nemetz, M. Mat. Lapok, 15 228-238.]
By iterating this and the fact that M(k, k, h) = lone can show
that
M(n, k,
Solutions 161
M(n, k,
(:) G)
(~)=(k\'
k-h h)
The upper bound is by induction on the index n and by using in-
equality (b), the recurrence relation for binomial and the fact that
one obtains an for n =. k (and for k = h or h = 1). In fact
~(n-:~~-I)+(n k:h-l)=(n-:+h).
M(2n, 5, 3) 2 (n).
3/
162 Problems in Combinatorics and Graph Theory
G =({x}ug(X))xeT'
One can at this =k for every x E T and also that G
contains h + 1 subsets. Let U == n - and suppose that there are indices
...• ir (O;:;;;r;:;;;h+ 1) such that U does not contain np elements of the set
for every 1;:;;; p;:;;; r. It follows that one can write
and hence
.
IUnTI~ I np+l~r+l.
[1= 1
Thus U contains at least one element x E T such that x if: ...• xd and
hence U {x}ug(xl E G. It has in fact been shown that U at least one ~
k-element subset of that is, M(n. n - h, k);:;;; h + 1. One can finally conclude that
I
M(n. n-h, k)=h+l for every k'#2 and n~k(h+l). Cahiers du
Centre d' Etudes de Recherche OperGtionnel/e, 15(3) (1973).
I
1
Solutions 163
(
~ ), .. , ~)
A k .. A
.., .. ,
.
). ), k
that is, the elements on the main are equal to k and ali the other
elements are to)" If A is the incidence matrix of a k,
follows that each column of the matrix contains k elements
v - k elements are to zero.
Let B AT A. The element bij of the matrix B is the scalar product of columns
i and j of the matrix A. Thus bif = k, that is. the number of 1's in column i of the
matrix A. If i then columns i and j of A have a 1 in row t if and only if the
element t of X to both of the sets XI and Xj' Thus the off-diagonal
element is equal to !Xjll =}.. if a v-by-i' binary matrix A
satisfies =(k ).)1+}"J O<),<k<v-l,thendefine
= {i! aij= l}
for every 1 v. It follows that (1) and (2) are verified.
4.39 One first evaluates the determinant of the matrix Al'A where A is the
incidence matrix of a k, ).}-configuration. Recall the form of the matrix
AT A obtained in the preceding problem; subtract the first row from rows 2
U, and then add columns 2 through to the first column. In this way one
obtains an upper matrix with elements on the main equal
to k + VI, -;. + k and the V 1 elements to k ;.,
Thus det (k leVA }"+k»O, that is, the matrix A is nonsingular,
since det (AT A)=(det A)2.
In order to prove that the k, i.)·configuration of the matrix A is a BIBD
with parameters v, k, k,;') it is necessary to show that each element belongs to
k blocks and each pair of distinct objects is contained in exactly i.
blocks. These conditions hold if and only if the matrix A satisfies the equation
=(k +i.J (1)
164 Problems in Combinalllrics and Grllpll Theory
which is similar to the equation satisfied the incidence matrix of a (v, k, A)-
configuration. The proof of relation (1) is analogous to that given in the pre-
ceding problem, since if C = AA T , the element cij (of the matrix C) is the scalar
product of rows i and} of the matrix A. Thus c/i=k if the number of I's in row i
of the matrix A is to k, and hence each to k blocks.
If i then rows i and j of the matrix A have a 1 in column t if and only if the
elements i and} both belong to block t. It follows that the number of blocks
which contain the pair {i, is equal to ,1..
One must finally show that if the matrix A satisfies the AT A == (k - ).)1
+ then =(k-).ll +AJ. Since each column of the matrix A contains k
elements to 1 and 1:- k O's, one can show that J A = kJ. A further
tion is that
(JA)A 1=(kJ)A 1.
and thus
(2)
But it follows from (2) that
-;,)1 +AJ)A 1
1
=(k-),)A- +),k-1J.
Thus
A -).jA IJ +Ak- =(k-).)A-1J +).k-1vJ,
since J 2 = vJ and the matrix J has order v.
On the other hand A kJ.
identifying the two
and
A IJ =mJ, or J =mAJ,
where m=(k I.k -,1.). But I'J= =J(mAJ)=m{JA)J=mkJ2=mkvJ,
and hence v=mkt, or mk=1. Thus k-)' k 2 so that k 2 -k=).(v-l); and
1
A -IJ =mJ =k-! J, that is, J =k- AJ, or
AJ=kJ
It has thus been shown that J A = AJ; this property is essential to the verifica-
tion of (1). The proceeds as follows:
AAT =A(AT A)A -1 =A((k-;.)l +).J
4.41 In the solution to Problem 4.39 it was shown that del A)2
ltV). - A+ k). where A is the incidence matrix of a (t, k, ).)-configura-
tion or a BIBD with (r, i" k, k, Since r == k and v b, the relations
which the of a BIBD must imply that k(k 1) -1) and
hence L'X-A +k==k2. Thus (k- lk 2 is the square of an Since 1'-1
is odd, it follows that k-A is a perfect square.
4.42 Let p be the maximum number of subsets with the desired property.
Then by the family of sets
uB AcY,IAI=r;BcX,,",y,IBI=[(n k)/2]},
one can show that
which that
4.47 Every positive integer can be written uniquely in the form where
P is a non-negative integer and q is a positive odd integer, called the odd part of
n. Therefore the odd of an from the set M = {1, 2, . , , , 2n: must be
one of the n 1. 3, 5, ... , 2n -- 1. Given n + 1 in M, at least two
must have the same odd : that is, they must be of the form 2P'q and
where PI =FP2' If, for PI <P2 then 2P1 q divides (W. L Putnam
Math. Competition, 1958.)
4.48 Let X {Xl.".' x",}, and denote by nl> ' , . , n", the numbers of sets
among AI, ' , " A100 which contain Xl' , .. , X", respectively. Then
100
n 1 + - .. + n", = I lAd x 100=75m,
i= 1
which implies that there is an index i such that nj;;::' 76. for
that i = 1 or n;;;::' 76, If x I is deleted from X, one obtains a subset X I with
m -1 elements. Let B I' ... , Bs be the sets among A I' ..• , A 100 not containing
X l' If s =0 let Y = Otherwise. it follows that s:'( 100 -- 76 = 24.
Since I, it can similarly be shown that there exists an element
X2 tomorethanisofthesetsBi.Let. " be the sets which
do not contain x 2 (and hence not Xl' construction), where p < s--
or p:'( 5. If p =0 let Y == {x I, X2}' Otherwise. it follows that there exists X3 which
belongs to more than ~p sets from . ' . , Cp' Because p:'( 5, it follows that
pi4:'(i, and hence at most one set from .... Cp does not inc1udex3' If all the
sets contain X3. the set Y={x i • X2, X3}' Otherwise. let X4 be an
element of the unique set C1 which does not contain X3 (and hence nol XI or
In this case one can choose Y = {x 1, X 2, X), X4}; Y has at least one element
in common with every set for 1:'( i:'( 100. A similar problem was at
the W. L. Putnam Math. Competition in 1980 (Problem B-4).
4.49 Suppose that the plane has a finite metric basis B. Then there
exists a sufficiently large ~~'~"M'- which contains B. Examination of 4.1
shows that the x and y have equal d4 -distances from the
Fig. 4.1
/
168 Problems in Combinatorics and Graph Theory
I
BI
A1V'~ B2
~ ~B3
Ir A2:~
. l~B4
A3f~
~ Bs
A41~
I ~ B6
~VI
As
I
0 A6 x
Fig. 4.2
This contradicts the assertion that B is a metric since any set which
contains a metric basis is also a metric basis.
In order to show (b) consider the in the plane bounded by the lines
x = 0, y x + y == n -1 and x + y = n + 1. We shall choose to be the rectangle
R, composed of all digital points in the region with the exception of the four
points having coordinates (0, n), (0, n+ I), (n, 0), (n+ 1,0) Figure 4.2). It
follows that R contains 3n - I digital points. Denote the (digital) intersection
points of the sides of R with lines of 1
An, Bn It is clear that for l~i~n, d4 (A" M}= ,M) for any
Min R such that MfA;, It follows that any 4-metric basis for R must contain
at least one point from every pair {Aj, Bd· for 1 ~ i~ n. This implies that dim4
(R);::' n, where dim4(R) denotes the number of elements in a minimal metric
basis for R (Le., the metric dimension of
It will in fact be shown that B={C 1 , ••• , i, where =Ai or =B, for
l ~ i~ n, is a 4-metric basis for R.
One observes that the pairs of digital points of R which have equal
distances to and 1 are {B I • Bi+J and the of the form {AI. Btl, ....
l' I}, 2, Bi+ . Bn}· Suppose now that B is not a 4-metric
basis for R. Then there exist two distinct digital points M, N such that M, N f B
and M and N have equal distances to all of B. If B contains and Ai+ 1.
it follows that {M, N}={B;. Bi+d, since M and N do not to B. In a
Solutions 169
similar manner one can show that if Ai, B i+1 EB then {M, N}={B j • Aj + 1
if B i , I EB then {M, N} {A;, Bi+d; and if Bh Bid EB then {M,
{Ai' Since n;;:t3, one can apply this argument to the pairs {C I ,
and { . It turns out that {M, N} = A2, B 2 } ("I A3,
== {A 2, B 2}' This establishes a contradiction, since M and N do not
It follows that dim4(R) == nand R has exactly 2" minimal 4-metric bases. [R. A.
Melter and 1. Computer Graphics and Image Processing,
(1984),1
4,50 Suppose first that L\ is a finite plane. Note that the condition
v;;:t4 implies that the order n of L\ is at least 2 and that the block size k is at least
3. Condition (I} is satisfied, since ;. = 1. It will be shown that every two lines of L\
intersect. Suppose that there exist two lines and such that Ll ("I
If == {Xl"'" xd, then by hypothesis Xi to k lines
Lr for every i = 1, ... , k. Condition im plies t hat for every 1"- i, k,
i and 2 ~ m, p ~ k. In fact, iffor LT = then the pair , Xj} is
contained in two lines and , which contradicts In this case the number
of lines of L\ is at least
CHAPTERS
5.1 Let
A ={mll ~m~ m=O
B={mll~m~ m=::l(mod
C={mll ~m~3n, m 2 (mod 3);.
It follows that IAI IBI=lcl n. The sum x+y+z=::O(mod 3) if and if
x, y, z E A or x, y, z E B or x, y, z E C or x, )" z each to a different set
among A, B, C. Thus the number of solutions is to
n(3n 2 - 3n + 2)
The (2k+ l)p entries in the Ferrers diagram associated with the partition of 11
as follows: Put in different rows + 1)2"
1)2" entries. Maintain the order of the
num ber of entries from to bottom. For exam pie, the 7 + 5 + 5 + 3+
3+ 3+ 1 + 1 + 1 + 1 is associated in this way with the 10+ 7+ 6+4+3
into distinct The thus obtained have distinct parts because
every can be uniquely as a of an odd number and a
power or2.
The injeclivity of the mapping follows from the uniqueness of the representa-
tion (1). In order to show that this is a surjection, consider a
partition P of n into distinct parts. Each can be uniquely written as a
nrr,{111rt of an odd number and a power of 2. the terms of the sum
and combining common factors one obtains only terms of the form (2k+ l}p
where k?J:. O. This will generate p terms equal to 2k + l.
By arranging these terms in decreasing order one obtains a partition Q of n
into odd Apply the previously defined mapping to Q. The result is the
partition P and this establishes the surjectivity of this mapping.
5.4 Let at + ... + am n be a partition of n into m distinct parts.
It follows that al > az > ... > 1, which that
this argument. one finds that the numbers which occur in the partition
of n are
1, kl + 1, (k l + 1)(k 2 + 1), .... (k l + 1)(k2 + 1)'" + 1),
where (k l + 1)' .. (k j + 1) has multiplicity to kj + 1 for i= 1, .' ., m. Since the
sum of these terms is to n, one can write
n=k 1 +k 2(k l +l)+'" +km + l (k 1 +l)'''{k",+1)
andhencen+1 +1)'" 1+1).
If n+ 1 is then m=O and the only partition of n which satisfies the
condition of the problem contains only terms to 1. If n + 1 is not
then there is another solution to the problem. In fact it follows that
n+l=(k 1 +l)(k 2 +1)
with k I' k 2 ~ 1. and the partition
n=l+ '+1;-
has the property that every number between 1 and n can be uniq uely rPTITP',pn
as a partial sum of this partition.
5.6 It follows from the rules for removing that I/I(n)=
Qe(n) - Qo(n) where the number of partitions of n into an even
number of distinct and Qo(n) represents the number of partitions of n
into an odd number of distinct parts.
In order to prove Euler's identity, one defines a transformation of a Ferrers
with an even number of rows into a diagram with the same number of
cells and an odd number of rows and vice versa. Since one considers only
partitions into pairwise distinct parts, the diagram of this kind of partition is
formed of several trapezoids placed next to each other as in Figure 5.1. Let the
number of cells in the last row of the diagram be equal to m, and let the number
of rows in the upper be to k.
}~T
SOUTH
fig. 5.1
Solutions 173
If m:s;:; k, one suppresses the last row of the (labeled SOUTH) and adds
one cell to each of the fm;t m rows of the upper trapezoid (in a line inclined at 45'
to the east in the This transformation does not change the total
number of cells. One obtains in this way a new diagram in which each row has a
different length. The parity of the number of rows is changed with respect to the
initial diagram. The diagram of Figure 5.1 corresponds to the partition
23 = 7 +6+ 5 + 3 +2,
After performing this transformation, one finds the row SOUTH laid
EAST, Since m=2 and k=3, the 23=8+7+5+3 is
obtained.
if the diagram contains at least two trapezoids and if m> k, then take one cell
from EAST in each row of the upper trapezoid, and with these cells make a new
row SOUTH in the new diagram. This construction is possible because In> k
and thus the row SOUTH is shorter than the old row SOUTH in the diagram. The
length of each row in the upper trapezoid has been shortened by one. It foHows
that all the rows of the new are different in length, The new
diagram contains the same number of cells as the old diagram, but the parity
of the number of rows has The new diagram contains one more or
one less row than the
This operation can be carried out when the consists of a single
trapezoid (when the diagonal EAST contains k cells and k is equal to the number
of parts of n) if m d: k and m l' k + 1.
The transformation just described is an involution on the set of partitions of
n into pairwise distinct parts. (This means that if this transformation is applied
one obtains the original diagram.) It follows that the transformation is
Thus the Ferrers diagrams for partitions of /1 which admit this trans-
formation can be divided into an equal number of with an odd and
even number of rows. One can now find the which do not
admit this transformation. They consist of a single trapezoid for which m = k
or m k+ 1 (Figure
In the first case
fig. 5.2
1/4 Problems in Combinatorirs and Graph Theory
(l-al
=(l+alx+
1 +a1x+ +
Observe that the term which appears in the coefficient of x" is
such that Al + -I- ..• + k)'k = n, and thus it determines the following partition
of n:
n=k+k+'" +k+'" +2+2+'" ..... 2+1 1+'" +1.
_ . - - - '----v---'
The rules for removing imply that in this case the exponents of
the symbols which appear in the coefficient of x" generate, without repetitions,
all the =
of n. If one sets a1 a2 = . . . 1, then the coefficient of x"
will be to the number of partitions ofn. Part of the problem
now follows immediately, since }'2 =}'4 = .,. O.
The proof of part (e) is contained in the solution of Problem 5.13; the proof
of part (d) is analogous to that of part (eJ.
The property in Problem 5.3 follows from an algebraic calculation
due to Euler and based on the use of generating functions:
1-
-----,.,.--,---,,,.-- =--.
(1- 1 x
+ x)(1 +
5.8 Euler's identity and the for the generating function
of the num bers P(n). one can write
Solutions 175
After to zero the coefRcient of x" on the left-hand side of this equa-
it turns out that
I P(n- 0,
j"O
or in other words
P(nJ+ I
k" 1
(-1)' {p +p(n- =0,
n
f
(l-
p;t 3)
(1 +x p +x 2p )(l- 1-
In fact
(j
..... H f H ", "'IIVW 11 \lfOm rroOlem ,.1) that the function of the
numbers P(lIl is
(4)
The
then leads to the following relation between the associated formal series:
The proof of part (a) is completed by equating the coefficients of on the two
sides of this identity.
The proof of part (b} starts with the identity
(y+ z) +z)=x+)'+2z>x+ y.
Thus x + y.), + z. x + z are the lengths of the sides of a triangle with perimeter 2n.
It is also the case that no two sides are equal, since
.\:+y>x+z>y+z.
It follows that
(11-1) 112-311+2
\ 2 2·
different ways (Problem 1.19). It follows from this that two terms are equal in
each of the following of 11:
--~rtlM~"V'~---'-''''·'''''· .
• • • • •_ _ _ _ _. .IiIIIIIIIII-IIIIIIIII--Fll!li...
178 Problems in Combinalorics and Graph Theory
There are thus 3(n - representations if n is even and 3(1l- 1li2 representa-
tions if n is odd and n:;;;: 4. But for n = 1, 2, 3 one has Q(n, 3) =0, and hence the
formula is verified.
If n is a of 3, one must also subtract 2 from the numbers obtained,
since there is a of n with three terms which is counted three
times instead of being counted only once. For example. for n=6 one obtains a
unique partition with two terms to 2. 6 = 2 + 2 + 2.
In order to obtain the num ber of representations with three distinct
terms, one must subtract from the number of all representations of n as the
sum of three the number of representations which contain equal terms.
The result is that this number is equal to
n2 - 3n+2
1(n-2)+2 if n
2
nl - 3n -I.. 2 nl- 6n -I.. 5
-I) ----- if n = 6k -1..1,
2
,,2 3n+2 nl -6n+ 8
- 2) if I! =6k+ 2,
,
11- 3n-l..2 n l -6n+9
1)+2= if n=6k+3,
n l -3n+2 nl-6n+8
-2)= if n 6k+4,
2
,
n l - 3n-l.. 2 n- 6n-l.. 5
-1)= if n 6k~5.
+1 for every n.
and thus
-6n..l-5
12
It follows from the inequalities
n2 6n..l-5 Q I) n2-6n..l-12
tn. -
that
Q(n, 3) ==
5.13 One first shows that the number of partitions of n into distinct
odd terms is equal to the number of partitions of n with a symmetric Ferrers
diagram. (The axial symmetry of a Ferrers is with respect to a line
drawn from the top left of the at an of 45° with the horizontaL)
Define a from the set of partitions of n with symmetric Ferrers
onto the set of partitions of n into distinct odd terms as
follows: that a F errers has k cells on the
and let a 1 be the total number of cells which are found in the first row and first
column of the diagram. Since the diagram is symmetric, it follows that a 1 is odd.
Now let Q2 denote the number of cells which are found in the first row and first
column of the diagram obtained by suppressing the a 1 cells. Similarly let a3 be
the number of cells which are found in the first row and first column of the
obtained when the a2 cells are and so on. One thus obtains
k odd numbers a 1>a2> ... >ak which define a of n: n=
a1+a2 + ... +ak'
For example, consider the partition of n=30 with the Ferrers
diagram of Figure 5.3:
30=7+7+5+4+3+2+2
Fig. 5.3
180 Problems in Combinatorics and Graph Theory
In this case al = 13, {/2 = 11, {/3 = 5, (/4 = 1, which to the partition
ono into four odd 30= 13 + 11 + 5 + 1.
It is clear that this correspondence is injective. In order to establish the
surjectivity let n = {l1 + ... + ak be a of n into k distinct odd
parts. A symmetric Ferrers will now be constructed Or cells in
the rth row and column for 1 ~ r~ k. It will turn out that for the described corre-
the of this partition with Ferrers is
nrF·t'i,plv the partition of n into odd parts, and this establishes the
surjectivity and hence the bijectivity of the ,..",·rp,:nr"".i
The number of partitions of n with
coefficient of x" in the product
In fact, in order to obtain a term equal to x" in this expansion one must multiply
terms of the form x D " • • • ,xu,, where 0 1 +02 + ... + Ok = n, al •... ' (lk are
pairwise distinct odd numbers, and we set (11;;:;: (12;;:;: ••• ;;:;: {lk in order not to
count these monomials twice, Since the coefficient of each product of the form
XD' ••. X , is
O
to 1, it turns out that the coefficient of x" is equal to the
number of partitions of n into pairwise distinct odd parts.
To obtain the function for the number of Ferrers
with k cells on the one suppresses the square with k cells
on a side which lies in the upper left-hand portion of the
Now consider the number of cells in row i plus the number of cells in column
i (i;;:;: k + 1) as a new term in a partition of n with a term to k 2 and a sequence
of other terms less than or equal to 2k. For the diagram of Figure 5.3 this new
partition can be written
30 +6+4+4.
One thus has shown that the number of partitions of n with symmetric Ferrers
having k cells on the diagonal is equal to the number of of
n of the form
where k;;:;: III ;;;:;: (/2;;:;: Similarly one finds that the number of these
... ;;;:;: (I,.
partitions of n is to the coefficient of x" in the expansion of the product
In fact the number of partitions of m = n - k2 into even parts which are less
than or equal to 2k is the coefficient of xm in the expansion of the product
(l
+ (I~XR + .. ), .. (I + GkX2k +
.•• (I~' + ' , ')X + " .
m
= 1+ (11
The term (11' which appears in the coefficient of xm satisfies the
relation
+ . , , + 2k}'k m,
and thus it defines a partition of m of the form
2k+ 2k+ ... + 4+'" +4+2+ ." +2.
It follows from the rules for removing that the exponents of the
which appear in the coefficient of xm generate, without all
of m into even Thus if (11 {/2 ' ., =(lk = 1. then the
coefficient of xm will be equal to the number of partitions of m into even
which are less than or equal to 2k. This the which has been
obtained for the function of the Ferrers
with k cells on the diagonal. Thus
represents the generating function for the number of symmetric Ferrers diagrams
with n cells; this the proof of the first identity.
In order to prove Euler's identity, observe that (as in the discussion of the
function of the number of of n into distinct odd the
number of of n into k distinct odd is the coefficient of in
the of the product
(1
On the other hand, it has been seen that there exists a bijection of the set of
partitions of n with Ferrers with k cells on the diagonal
onto the set of partitions of n into k distinct odd parts, Thus the number of
partitions of n into k distinct odd is equal to the coefficient of x" in or
the coefficient of in
CHAPTER 6
6.1 There is a unique walk which joins each pair of vertices of a tree. The
subgraph induced by B does not contain since A itself does not contain
If x, y E B and x i=}', then x, yare vertices of each subtree AI' ....
and thus each of these subtrees contains the unique walk Z l' . . . , Zk' ,l'J
joins x and J in A. Hence Zl"",ZkEXl"'" or Zl,,,,,Zk EB. and thus
the subgraph induced by the set of vertices B is connected and is in fact a tree.
6.2 The proof uses induction on the number of vertices in the tree G.
lf G has two vertices the property is immediate. Suppose that the property
holds for all trees with at most n vertices. and let G be a tree with n + 1 vertices.
G contains a vertex x of degree 1 which is adjacent to a vertex .\'. By the induction
hypothesis the property is valid for the subtree obtained from G sup-
pressing the vertex x and the [x, If no subtree G 1 , ... , Gk is the graph
of only the vertex x, then all the subtrees , ... , Gk obtained from
Solutions 183
6.4 Suppose U I is the set of edges of the tree A l' and U 2 is the set of
of the tree . let u E U I " If the edge u is suppressed from the tree
then a graph G l is obtained which contains two connected components eland
In fact, if the graph obtained had contained at least three connected com·
ponents, then by adding the edge u between two vertices located in different
components the resulting graph would not have been connected and hence
A 1 would not be connected. Thus the definition of a tree would be contradicted.
184 Problems in Combinatorics and Graph Theory
On the other hand, it has been seen that every tree with n vertices has n - I
6.3), and thus by suppressing u the resulting graph IS
no since it has n- 2
There is an v E U 2 which joins two vertices located in components C 1
and . For otherwise it would follow that the of the tree only
of vertices located either in , which would imply that A2 is not
connected. But All a tree, is and hence there exists an
v E U 2 which joins vertices located in and . Since does
not contain a cycle, it follows that the obtained from G 1 by the
edge v also does not contain cycles and is connected. It is thus a tree. which
wi!! be denoted
But is a spanning tree of G which has more edges in common with
than A l ' this transformation, replacing all the edges U E U I ""'U 2
of U 2, to obtain finally a tree Br = where r;:::' 2.
6.5 We show that if all the vertices of 1 of the tree G are sup-
pressed then e(x) is decreased by 1 for every vertex of the
All the vertices at a distance e(x) from x have 1, and thus by
them decreases for all of the vertices. One can also observe that
by this operation e(x) decreases exactly one, since the longest walk which
leaves x ends in a vertex 1 in G, which is then suppressed. The property
is true for a graph with a vertex. Assume therefore that it is true for all
trees with at most n 1 vertices. Let G be a tree with n;:::. 2 vertices. Denote by
C the set of vert ices x in the center of G, that is, those for which e(x) is a minimum.
Suppose that C does not contain a vertex of 1, and suppress all vertices
of l in the G. For all the vertices the value of e(x) is
reduced by one. and hence this operation a new tree G' is obtained with the
same center C. Since G' has at most n - 2 it follows from the induction
hypothesis that C consists of one vertex or two and nr(\T\prtv
(a) is therefore established. If C contains a vertex of 1, then x
will be adjacent to a unique vertex y. It is clear that J' is strictly nearer than x to
every other vertex of G. Thus e(x) can be a minimum only jf e(x) 1 and G is a
tree consisting of x and y joined an In this case C = y) and the
nrr,nprtv is established.
By Problem 8.4, the walks of maximum in a tree have a
intersection. It can be shown that this intersection contains the center of the tree.
(b) Let L be a walk of length which starts at x. If L does not contain
either}' or z, then
e( y) = + 1, e(z)=e(x)+ 1.
and hence 2e(x) < + e(z). The walk L cannot contain both y and z, since
both vertices are adjacent to x. For example, if L contains y, then
e()')~e(x) 1, e(z) +1.
and thus e(y) +
Solulions 185
(c) It is easily shown that the distance thus defined on the set of vertices ofa
connected graph is a metric for this set and thus satisfies the inequality.
Let x, y be vertices such that d(x, y) =d(G), and let z be a vertex of minimal
eccentricity [ It can be shown that
d(G)=: y)~d(x, z)+d(z, y)~p(G)+p(a)=2p(a).
The points in which s(x) attains its minimum form the barycenter of the tree
G. By a walk with an odd even} number of vertices
one observes that the vertex (or of two adjacent
If Xl' ... , xp are the vertices of a walk and Yl , ... ,}'q are other vertices
by an to Xl' then if p is even, it follows that the center of the tree
thus obtained is the vertex . If q is sufficiently large, for q
then the barycenter of the tree Xl' while the distance between the center
the is p/2 - 1. This number can be for values of p
which are sufficiently
6.8 If the tree G has vertex set X of cardinality n, it follows that it also has
n- 1 Then in the sum d(x. ,1') there are exactly 2(n -1) terms equal
to 1, and the other nonzero terms are greater than or equal to 2. If G is a star
(the K 1,n-l), all the terms different from 0 x = .v) and 1 are equal to 2.
If G is not a star, there is at least one term equal to 3, and thus the desired mini-
mum is attained only for the tree 1 In order to find the maximum of this
sum one can show that s(xj is a maximum in the set of terminal vertices of a
tree when the tree is a walk and x is one of its If L is a walk
with n vertices and x is one of its endpoints, then
s(x)=1+2+" +(n-l).
If G is a tree and x is a terminal vertex with e(x) =d. then there will exist at
least one vertex at a distance 1, 2, ' .. , d from x. It follows from the definition
of that the sum which enters into is of the form
s(x)=l +2+ .. , +d+d 1 + ... +
where d 1 , . , . , dn - 1 -d"d. comparing this with the expression for in the
case of the walk L. one finds that the maximum of s(x) is attained only when
d = n -1, that is. when G is a walk and x is one of its One now uses
induction on n to prove that Lx,yeX d(x, y) is a maximum in the set of trees G
with n vertices from the set X when G is a walk. For 1 ~ n" 3 the property is
immediate, since in this case every tree with n vertices is a walk,
Suppose now that the property is true for all trees with at most n -1 vertices,
and let G be a tree with n vertices which consist of the set X. If a is a vertex of
1 in the tree then
d(x, .v) 2s(a) + L d(x, y).
Xd'El'
" P---C--<>--r!v
XI
Xj
Fig. 6.3
There is also a vertex v E such that the walk from I' to Xk has no
vertices in common with other than the 1'. It follows that
Xj)+
which (a).
Consider i,j, k, I E {I, ... , r} which are pairwise difTerent. If, for example, i
the three numbers become dkb dki+dl/, dkl+d ll , and thus two are equal and the
third satisfies the inequality
dkl~dkl+di1'
In fact, the distance defined between the vertices of a satisfies the con-
ditions ror a metric, including the
Let Lij and Lkl be walks which vertices Xi to Xj and Xk to XI> respectively.
These walks can have no, one, or two or more vertices in common.
If the walks have no vertices in common, let u be a vertex of the walk Lij and t'
a vertex of the walk Lk/ such that the walk with endpoints u and r has
in common with the walks Lij and Lk/ only the u and v (Figure 6.3).
In this case, it follows that
dik+dp = +djk=dij+ + vJ,
which establishes (b).
If and have exactly one vertex u in common, then dij + dki = d1k + djl =
+
If the walks and Lkl have at least two vertices u, v in common tFigure 6.3)
then (b) is also true. If u and w denote the endpoints of the subwalk common to
the two walks, then it follows that
dij+dki=di/+djk=dik+djl+2d(u. w).
Let r~ 2 and (diAJ = j , . , be a symmetric matrix with "('\,,.",'0
such that dij =0 if and only if i which satisfies
to show (by induction on r) that there exists a tree A with r terminal vertices
Xl •... ' xr such that d(x" xj)=dijfor i,j= 1, ... , r. A. Uspelli Mat.
Nauk,20 94-96.]
6.10 The proof will proceed by induction on r. For r=2 the property is
since every tree with two terminal vertices is an
walk and the fact that dAO, 2)=dn(l, 2) shows that the two walks have the same
number of vertices and are thus isomorphic.
Let r>2, and suppose that the property is true for all trees with at most r-1
terminal vertices. Let VA denote the vertex of degree greater than or equal to 3
of the tree A, which is closest to the terminal vertex labeled r. Similarly let Vn
denote the vertex of the tree B. It follows that there exists an
elementary walk [VA"'" r] in the tree A, such that all vertices located between
VA and r have 2 in the A (their set may possibly be empty). There
is a similar elementary walk [VB"'" r] in B. the terminal vertices
with label r as well as all internal vertices on the walks ... , rJ and [VB" .. , r]
In this manner one obtains trees AI and B\ which each have r-1
terminal vertices with labels selected from the set {1, ... , r 1}. The distances
between the terminal vertices i and j are the same in A and B for 1 ~ r-1
and thus will remain unchanged for the trees A 1 and B I' By the induction
hypothesis A 1 and are isomorphic trees, and thus there exists a bijection f
from the set of vertices of A I onto the set of vertices of which preserves the
of vertices.
One can assume that i for i=l, ... ,r-l, because it is possible to
relabel the terminal vertices of B J so that this condition is satisfied and the
distances between the terminal vertices i and j are the same in A 1 and B I' It
is now possible to show that f(vA)=vn. that f(VAl7'=Vn. There exists a
unique walk in B I which joins !(VA) and Vn which can be extended in an arbitrary
fashion to a walk which the terminal vertices i and j in B l ' It follows that
dn(i,j) + dBti, r) - dBU, r) = 2d B(i, Since f(VA) is found on the walk which joins
i andj in Bl (which is isomorphic to AI)' it follows that VA is found on the walk
which in AI' one can conclude similarly that
dA(i,j)+dA(i, r)-dAU, r)=2d A(i, VA)'
Since the distances between terminal vertices are the same in the trees A and B,
it also follows that dn(i, VB) = dA(i, VA)' But dB(i, =dBtU, vn) and dA(i, VA)=
dA,(i,vA)' Since and BI are isomorphic under f, it follows that dA,(i, VA)=
dB,(i,J'(VA)) =dB,(i, VB), which contradicts the fact that f(VAl 7'= eB and the vertices
f(VA), Vn, and i are found on the same walk. f(v A) = Vs. Let i, j with
1 ~i r-1 be labels for two terminal vertices such that VA and !(VA)=VB
are found on the walk with endpoints i and j in A! and and
hence in A and B.
One can thus write
Since the left-hand sides are equal by the hypothesis, it follows that dA(r, ==
dll (r, 1:n), and hence the walks [VA, ... , r] and [vn,"" r] have the same length.
",UlUllons 189
subtree A" of A with vertex set X,,, In the solution of Problem 6.5 it was shown
that A and A,. have the same center. It therefore follows from the induction
hypothesis that either the center of A is equal to u, in which case g(u)= I(u)=u,
or the center of A consists of the adjacent vertices u and v, in which case either
g(u)= flu)=u and gft')= !(v)=t'. or g(u)= !(u)=r and g(v)= I(t') =11.
The stated property is thus valid for every m. Since it must be shown that 1
has a fixed point, the only case which must be investigated is that in which A
has center (u, d with j(u) [' and f(r) u. Let and Au denote the subtrees of
A obtained by suppressing the [1/, 1'J and which contain the vertices u and !'
respectively. Let Ii l ' ... , II, denote the vertices in Au which are adjacent to u.
Since 1 preserves adjacency in A. it follows that f(UI),"" flu,) are adjacent
to v in the subtree AI" and so on.
Let and X" denote the vertex sets of the trees and Au It
follows that I(X.) = Consider the restriction of the function 1 to the set
Xu; denote it by h: X .-X v' The function h is a bijection, and y] is an
in the tree A. if and only if hU')] is an in the tree Av. Thus h is an
isomorphism of the trees Ali and Av. I t follows that and have the same
number of which that 1= + and is thus an
even num ber.
h was assumed that 2n+ 1, and thus the case in which f(u)=v and
u does not occur. It foHows that! has at least one fixed point.
6.13 Let AI X and x denote the graphs obtained from AI and
by x and the incident with x. Since a tree with /1
vertices has /1-1 (Problem 6.3), the of the vertex x in the tree Al
is equal to
(2)
The fact that the Al -x and - x are isomorphic implies that
1l(A 1 -x)=m(A 2 -x).
By and (2) one can show that dA,(xl=dA,(x) for every vertex x EX
md thus the trees A! and A2 have the same terminal vertices 1). Let
r be the set of terminal vertices for the trees A 1 and A 2 • If ITI it follows that
41 and A2 are walks oflength IXI- I, and thus Al and have the same diameter.
iuppose that iTI ~ 3, and let L be an elementary walk of maximal length in the
ree A l ' The length of L is by definition to d(A d, the diameter of Ai'
The endpoints of this walk are two terminal vertices in the set T. The set T
dso contains at least one other terminal vertex x which does not to the
valk L. By hypothesis - x is isomorphic to Al - x. Since the walk L is con-
ained in the A I X, it follows that - x contains an walk
Solutions 191
which has the same length as the walk L. Thus the graph
walk of the same as L. It follows that
the roles of A 1 and in the one can
conclude that ~ d(A 2 ) and hence A 1 and have the same diameter.
It can also be shown that under the conditions the trees A 1 and
are [P.1. Kelly, Pacific J. Math., 7 (1957), 961-968.J
6.14 One can obtain an arborescence from the tree G by the
vertex Xl to be the root and directing all the of the tree so that for every
vertex yi=x l there is a unique which in Xl and terminates in y.
In this way a order is defined on the set X: Let Xi ~ Xj if the unique
from Xl to Xj contains the vertex Xc.
Let the matrix Z be defined as follows: zu=l if x!~
oif Xi is not less than or to Xl' In the solution to Problem 2.19
that one can renumber the elements of X so that Z is upper
with Zji= 1 for i= 1, ... , n. It follows that del Z = L
Let
~\
1 I
-2 o
o 2
o o -2)
It will be shown that ZT AZ = D. In fact, the element in row i and column j of
the matrix ZT AZ is
n
I ZkiQklZlj= I I Ukl'
1= 1 Xk4!tXi :q~:t;j
I (- 2)+ I 1+ I 1,
since x 1 ~ Xj for every i = 1, ... , n. Let X,.denote the last common vertex of the
paths from Xl to Xi and from Xl to with XI being considered as the initial
point. It follows that
,x,.)+ 1)+ +1)+(d(x 1 , +1)
6.15 (a) Since every tree has at least two terminal vertices. it follows that
the procedure previously described one obtains a tree only
of the vertex Xn and another vertex adjacent to X n• The Priifer code of A is com-
pleted by a new position (In-I = n.
Let b l , ... ,bn I be the indices of the terminal vertices which are
when the algorithm determining the Priifer code is applied to the tree A. It will
now be shown how to determine the numbers bi from the Priifer code
, ... ,a,,-l)·
It is clear that bi is dilTerent from b l •.. " I and also from (Ii' since Xb,]
is an in the tree A. Since the vertex Xh, has been it cannot be
adjacent to a terminal vertex at a later Thus bti=aj for i.
Conversely, if k t {hi," , ' .. , an - d, then the vertex Xk is a terminal
vertex of the tree A-{Xb ••• , obtained from A by suppressing the ver-
tices with indices ..." , bi - I - otherwise it would have to be to
a vertex which will be suppressed at a later step. It follows that k E {a;, ... , (in- d,
which contradicts the hypothesis. Hence
(1)
Since
(2)
194 Problems in Combinatorics lind Graph Theory
This equation is in fact identity (c) of Problem 1.29 and implies that t n =
Dziobek, Sitzungsber, Berl. Math, G" 17 (1917), 64-67.J
6.17 that the terminal vertices are fixed and in fact that they are
Since d 1 = . , , = dp = 1. it follows from the problem that
number is equal to
In 2)!
(1)
6.18 (al Let f(n) be the desired number of possible ways of selection. It
follows that f(I)= 1 and f(2)=2. This corresponds to the choice of edge
[Xl' Yt] and respectivcly to [Xl' X2J and [.1'1' .l'2J or [Xl' J'IJ and [X2, Y2]. If
the ladder graph has 2n vertices. then one can select [x ,. }'tJ and for the remain-
ing edges there 1) possihle choices. One may also select and
[YI' Y2], leaving further choices. It then
f(n-l)+ f(n-2).
and this relation the initial values f(l) implies
that fin) = Fn' num ber.
(b) Let g(nl number of spanning trees It
follows that g(l) = 1 w hcn thc spanning tree consists of LX l ' .It J. Also g(2) = 4
for the trees which are obtained from a cycle with four vertices if one suppresses.
in turn, an edge of the cycle.
In order to prove the recurrence relation for Yin), consider the graph of
Figure 6.1 with 2n+2 vertices:
xl'yl····· X n+I'Yn+'·
g(n)= +
6.20 It will be shown that is equivalent to (2) and that (2) is equivalent
to (3).
Suppose that G is a tree, that is, G is connected and without cycles. If the
obtained from G by [x, is then there is a walk
between x and y in . It follows that in G I there exists an walk
between x and y, which together with yJ generates a in G, and this is a
contradiction. Hence (1) implies (2). Suppose G is a graph satisfying (2) and
G has a cycle [x, Zl"'" Zb y, xl By deleting one edge [x, y] in this cycle one
obtains a new graph which is connected; this is a contradiction. It follows
that G is connected and without and hence is a tree.
OJ implies If G is a tree, it follows that G does not contain a For
any two vertices x, y of G there is an elementary walk between
x and y. This walk together with the yJ generates a cycle in , and
hence (3) holds.
It remains to show that (l). To show this let G be a
One must prove that G is connected. Suppose that G is not connected. There
are therefore two vertices x and y belonging to difTerent connected components
of G. By the one obtains a graph G 1 which does not contain
a which is a contradiction.
6.21 If is a tree with n labeled vertices, one may choose a root a in n ways.
F or any such selection a unique arborescence is obtained directing each
so that any vertex x =f a to be reached from (I by a unique path. Hence the number
of arborescences with n labeled vertices is to n'nn- 2 = nn-I by Cayley's
formula.
6.22 Choose a vertex of a labeled tree with vertex set {l.,." nJ (say
the vertex and call it the root. There exists a unique walk from any other
vertex i< n to the root. If [i, J1 is the first edge in this walk, let The
function J is called the tree function of the label i to the [i,
where JU), for i = 1, ' .. , n -1. This defines a mapping of the set of nn-
vertex-labeled trees TN onto the set of edge-labeled trees,
When n ~ 3, each tree is the of n vertex-labeled trees,
since the vertex n may be chosen in n ways and the labels of the other vertices
are uniquely determined the labels of the It follows that the number of
trees with n unlabeled vertices and n - 1 labeled to n" - = n" - 3
by Cayley's formula, [E. M. Palmer, J. Combinatorial 6(1969),206-207.J
6.23 Each column of an incidence matrix A contains one +- 1, one 1,
and n - 2 zeros, and hence the sum of all n rows vanishes. The sum of any r
rows of A must contain at least one nonzero entry if r<n, for otherwise G
would not be connected. This implies that no r rows are if
r < n. In if there exist r rows ai" ' . , ,ai, of A (r < n) whose sum equals the
null vector wiLh rn components, then it follows that there is no which is
directed away from or towards the vertex set {i 1 ,., of G; this contradicts
the hypothesis that G is connected,
198 Problems in Comblnalorics and Graph Theory
-I -I
I -I n-I -I -1 n-I -1 0 n 0
=11"- 2,
From the initial conditions one can conclude that C 1 =(3 + 2j3)j6 and C 2 =
-2j3)/6. [H, Prodinger. R, F. Tichy. Fibonacci Quarterlv. 19821,
16·21.]
6.29 The proof is by induction on n. For n=2 this inequality becomes an
equality and coincides with the Principle of Inclusion and Exclusion.
Suppose that the inequality holds for any n - 1 subsets of X and any choice
a Iree on vertices 1, ... n-l Without loss of generality one ean suppose
that n is a terminal vertex of the tree T. Denote by T I th e tree obtained from T
suppressing and its incident edge nJ. It follows from the induction
hypothesis that
i(Alu ... uA._1)uA"!=IA1u·" UAn-ll+IAnl-i(A1U , .. uA._ 1)IlA.1
"
~
i
L
1
IA{i- L [Lj]EE(Ttl
IAIIlAjl-IAkIlA.1
=2:" i= 1
L
[t.j]Et:{T)
iAi Ajl,
6.30 It is clear that (1) is satisfied. In order to show suppose that the
of I induce a spanning graph (X,1) with p components , ... , Cp and that
these components contain respectively n 1 , ••• , np vertices (nl + ... + np= n).
Since (X, 1) does not contain a cycle, it follows that C I, ..• , C p are trees having
respectively n l -1, ... ,np-l edges. hypothesis (X, J) also has no
which implies that J contains at most nj-l edges with both endpoints in the
component Cil for i =: 1, ... , p. Because IJI III + 1, it follows that there is
at least one ee whose lie in different components and
(i:l= j). It follows that J v is an independent set, since (X, I v {e}) also
contains no
From Problem 6.20 one can deduce that the bases of the matroid M(G)
coincide with the edge sets of spanning trees of G. If (X, S) has p components
containing respectively m 1 , ••• , mp vertices, then the fact that independent
sets of edges contain no cycles implies that
p(S)= (m/1)=n-p
CHAPTER 7
!J
SolutIons 201
Thus 13 the number of faces whose vertices are colored with all the
colors a, b, and c, since each edge with colors a, b is counted twice (in both faces
with it). Thus /3 == 0
This is a
which is to an
The result remains valid for an
7.3 The of the condition is since if a graph G without
isolated vertices has an Eulerian then it is connected. In fact each two
x and y, incident to an arc which belongs to an Eulerian
are a walk.
On the other hand, an Eulerian circuit uses all the arcs which at
and terminate in every vertex x. The Eulerian circuit uses a unique arc which
ends at x and a unique arc which starts at x each time the vertex x is traversed,
and hence r(x)=d"'"(x). In order to show the sufficiency,let C be a circuit of
the graph G which contains a maximal number of arcs. The graph G is con-
nected and d+(x) for every vertex x. If C does not contain all the arcs of
then it follows from the fact that G is connected that there is an arc (x, y)
which has a vertex in common with the circuit C Suppose, for that
x e C One can make this because the circuit C uses the same
number of and exit arcs in each vertex, but the of the vertex y
is to the of the vertex y. Therefore if y e C there will exist an
arc of the form (y, t).
Let Gc denote the spanning subgraph of G induced the arcs which do not
belong to the circuit C Since d-(x) for every vertex x. and since the
circuit C uses the same number of entry and exit arcs at every vertex. it follows
that all the vertices of have equal indegrees and outdegrees.
Leave by the arc (x, y) of Gc , and move the arcs of Gc each arc
once. Continue this process as as possible. One cannot end in a
vertex z since the vertices of have and and
each traversal of a vertex z + uses an entry arc and an exit arc. Thus if one has
+
arrived at a vertex z x, one can also leave this vertex on an exit arc. Since the
number of vertices of Gc is finite, one must terminate at the vertex x, and this
produces a circuit C 1 in
The union of the arcs of the circuits C and C 1 is a circuit which is
than C, which contradicts the assumption made. Thus C is an Eulerian circuit.
7.4 One can suppose that the graph G has at least one Eulerian circuit C,
since otherwise the property is evident. The Eulerian Cifcuit passes through
every arc of G and hence passes the vertex x with d+ (xl ~ 3.
Move the circuit C by from x and returning to x. At each
traversal of x one obtains a circuit. Let these circuits be • C 2l ••• , where
m =d+ (xl. permutation of the circuits .... , Cm determines an order of
passing the arcs of the graph G once, and hence an Eulerian circuit.
Two Eulerian circuits obtained in this way are identical if and only if the per-
mutations of the cycles C I, .•• , Cm are identical as cyclic The
202 Problems In Combinatorics and Graph Theory
In the solution of Problem 6.3, it was shown that d(x)~2 for every vertex X
implies the existence of an elementary cycle C I in G. By suppressing the edges
of the cycle Clone obtains a spanning graph of G which may contain some
isolated vertices. After suppressing the isolated vertices one obtains a spanning
subgraph G! of G whose vertices have even degrees and which does not contain
isolated vertices. In fact, by suppressing the edges of the cycle C 1 some vertices
of G have their degree reduced by 2. One can write
G=C 1 uG 1 ,
where G1 has at least three edges less than G. G has all vertex degrees even and
lacks isolated vertices.
By continuing this procedure the edges of Glare eventually exhausted and
one finally obtains a single cycle C k' that is, G = C 1 U ... U C k such that C 1"'" C k
do not have an edge pairwise in common.
Now let G be a graph all of whose vertices have even degree. Excluding
isolated vertices, it has been shown that G can be expressed as the union of
elementary cycles C j , ••• , C k which do not contain an edge pairwise in com·
mono By selecting a sense for traversing each cycle Cp' one directs the edges of
the cycle in the sense of their traversal. Thus the in degree and the outdegree of
every vertex on the cycle Cp increase by one. Finally, after all the edges of the
Solutions 203
cycles C p have been directed for p = 1,. .. k, one obtains a directed which
satisfies the given conditions.
7.6 If the graph G has an Eulerian then it follows that G is connected
and has vertices of even since each traversal of a vertex uses two
if G is connected and has even then by the
can be directed so as to obtain a directed graph which satisfies
for every vertex x. It follows from Problem 7.3 that the
so obtained has an Eulerian circuit, which corresponds to an Eulerian
in the G.
that G is connected and has 2k vertices of odd (k ~ I). Let
G 1 be the graph obtained from G by adding a new vertex which is joined by
to all the 2k vertices of odd in G. It follows that is connected
and has even and thus has an Eulerian After the
additional vertex and the 2k incident with it. the Eulerian decom-
poses into k walks which are disjoint with respect to their do not use the
same twice, and cover all the of G.
It can be seen that the sum of the of the vertices of the graph is even.
and hence the number of vertices of odd is also even.
7.7 Let K: denote the complete directed graph with vertex set X of cardi-
nality n and arc set or the form {(x.y)lx.yeX and X:fy}.lt follows that the
graph K: has n(n 1) arcs and the set of arcs of the G is the complement
of the set of arcs of the G with respect to the set of arcs of The number
of Hamiltonian ofK: is equal to n!, since there is a bijection from the
set of Hamiltonian paths of K onto the set of permutations of X. Let U j • • • . , Urn
denote the arcs of the and let Ai denote the set of Hamiltonian paths
of K: which contain the arc Ui'
One can use the Principle of Inclusion and Exclusion (Problem to obtain
=n!+ iAr,n'" n 0)
Thus if the arcs 1.11" ••• , ul• form elementary paths without common
VI"f'f,f',""o then
IA/,n ... nA,.I=(n-p)!.
Finally, the numbers IA" n'" nAipl are equal 10 I if and if p=n 1
and the arcs UI" ••. , Ul. form a Hamiltonian path in the graph Otherwise
these numbers are even.
By (1) one can conclude that h(G):=h(G) (mod
Let G be a and consider the Kn which has (;) edg~s
and n Hamiltonian walks. Denote U l , ••• , U m the edges of the graph G,
and by the set of Hamiltonian walks of Kn which contain the edge 1.11'
A formula analogous to (1) for h(G) can be obtained immediately by replacing
nl with n which is even for every n;-:4.
One can also show that
n ... n
if the 1.11" ••• , Ut. form q walks without common vertices. The
right-hand side is equal to zero otherwise. In fact, if one considers the n - p
connected components reduced to a single point, then one can form
(n- !
Hamiltonian walks, but each walk L from among the q elementary walks of the
form
L= , ... ,xp ]
can be inserted in each of the (n- Hamiltonian walks in two ways. One
can choose the form , ... , xp] or [X p,"" Xl]. This results in distinct Hamil-
tonian walks which belong to the set AI, n .. n Ai.'
Thus lA/In'" nA/pl is equal to 1 if p=n-l and q=l (in other words
1.1.\, ••• ,1.1/ form a Hamiltonian walk in the G) and is even otherwise.
p -
In view of the fact that n;-: 4, one can conclude that h( G) h(G) (mod
7.8 It can be shown that if the orientation of a unique arc 1.1 = la, b) of the
tournament G is inverted, then a tournament G 1 is obtained such that h(Gtl=
h(G). Let G2 be the graph obtained from G by suppressing the arc u, and let G3
be the obtained from G by adding the arc (b, a) with the same endpoints,
but the opposite orientation to u.
Since G2 is obtained from the direction of all its arcs, it
follows that
h(G z)= h( G3 )
The results of the preceding problem then imply that
h(G 2 )=h{G z) (mod 2).
Solutions 205
Let h 3 (a, b), a), and h3 (O) denote respectively the number of Hamiltonian
paths of G3 which use the arc (a, b), the arc (b, or neither of these two arcs. ThEm
h(G z)=h(G 3 )==h 3 (O)+h 3 (a,b)+ a),
from which one can conclude that h3(a, b)=h3(b, a) modulo 2. Hence
=h 3 {O)+h 3 (a, b)=h 3 (O)+h 3 (b, aj h(GJl.
Now let G be an arbitrary tournament. According to Problem 9.5, G contains
a Hamiltonian path; say (Xl' X2.' .. , B y the direction of some arcs,
It IS to insure that the tournament contains only arcs of the
form (XI. xJ) with i i.e., it becomes a transitive tournament with a unique
path.
It has been shown that if the direction of an arc is the parity of the
number of Hamiltonian paths remains constant. This observation that
h(G)=l(mod AClaLitt. 7(1934),
7.9 Let L be a walk which at a vertex Xo of the
graph G:
L [xo, ... ,xkl
It follows that the vertex Xk is to Xk _ I and to the other d(Xk) -1 vertices
which belong to the walk L, since otherwise L could be extended to a walk.
Suppose that Xk is to a vertex xJ such that 1 ~ k 2. The walk
(A 6, B) 6, B A
for every two sets A and it follows that J is also an involution, that is,
=G 1 ·
It folIows that for every of G" of even degree there exists a
graph G1 = f{G 2 ) of even of G which contains u and is such
f(G d. The mapping f is therefore and hence bijective. Thus
the number of spanning of even ofG is equal to 2P - n + 1 +2 P - nT ! =
1)-"T I. This completes the proof by induction of the property.
IYjn
and in this sum each term is even.
If x E ,it can be shown analogously that if x fE A then dZ2 (x) is even. If
x E A, then
d Z2 (x) -d s(x)+d 6 (x)
This number is always even, since d4 (x) is even by the induction hypothesis and
IY2 n AI is odd. The property is thus found to be true for every n.
It will now be shown that there exists a partition
X= u
such that the degrees of the vertices of the subgraph generated by X 1 are even
and the of the vertices of the subgraph induced by X 2 are odd. To this
end add to the G a new vertex y, which is adjacent to all the vertices of X.
Let Gibe the thus obtained. the previous resul t, there exists a
Y1 U Y2 of the set of vertices X u where and induce
even degree in G I' If, for example, y E Y2, then by denoting Xl = Y! and
X 2 == Y2 "'-{y} one obtains the desired partition of the vertex set of G. [W. K.
Chen, SIAM J. Appl. Math., 20 (1971),526-529.]
7.12 It is clear that if C contains only nonempty then = Cu
satisfies the and hence one can assume that E C.
It follows that C 1= {X}, since otherwise there exists a proper subset Y c:: X
which has elements in common with and hence with an odd number of sub-
sets from C. Thus there exists a subset A in C such that IAI =a is minimum and
a ~ 1. Because a ,,;;; n - 1, it follows that X "'-A is a proper su bset of X, and X "'-A
intersects all sets of C but does not intersect A. By the hypothesis Icl 1 is an
even number, and hence C contains an odd number of subsets of X.
Solulions
CHAPTERS
(1)
r i=1, ... ,p. Equality holds only for components which are trees.
By inequalities (1) for i == 1, ...• p, one finds that m ~ n - p, since no
10 connected components have a common vertex and hence have no common
1ge.
8.4 Suppose that there exist two walks of maximal
d which have no common vertex. Since G is connected, there are two
rtices Xl ELI and X2 E which are joined by a walk Q which has in common
th and L2 only the endpoints Xl and X2' The vertex XI divides the walk
into two subwalks Lil and Li2 (one of which may possibly be empty) for
1, 2.
Let I(L) be the of the walk that is, the number of its One can
;ume that l(Lil)~ I(Ld and I(Ll1)~ I(L 2d. It follows that
I(L 11. ) > I(L 11)+ I(L2 d~ 2/(L2 d ~ I(L 2 d + = I(L 2).
d hence the walk (Lll' Q, L2d is longer than ,which contradicts the
pothesis.
Suppose that G is a tree, and let Lt. be two walks of maximal length in G.
ese walks have at least one vertex in common. In fact the common vertices
1:..1 and form a walk Q lwhich may reduce to a vertex). Thus vertices
u a of the following form: There exist two walks
which originate at one terminal vertex of Q and two walks L'{ and
ich at the other terminal vertex of Q. such that
=(Lj,Q,L'{) and L2 • Q. L:;).
'ollows that I(L'd = I(L 2), since if, for example, I(L'! I(L'1), then the walk
,Q, L~) would be longer than L 2 , which would contradict the
r)
In the same way it is seen that I(L = I(L'2).
;;ince (L'l , is a its is at most to the of a maximal
k, and hence 21(L~)~ I(Ld or I(L'J tl(Ld; analogously, l(L'i) ). Thus
median point or the two median points of the walk belong to
walk Q and hence to Since can be chosen arbitrarily, it follows that
median point(s) of Ll on whether I(Ld is even or
he intersection of all the maximal walks of the tree G.
f the connected graph G contains cycles. this property does not hold.
Solutions 211
where ai E A and bi E B for 1:;;;; i:;;;; k. The length of this cycle is 2k, and thus an
even number.
Now suppose that every ofG has an even number or vertices.
Color the vertices of G with two colors so that each two vertices joined by an
have different colors. The coloring is perrormed as follows: Start with a
vertex Xl which is colored a. The vertices to Xl are colored b. Then the
vertices adjacent to the vertices which are colored b will be colored a, and so
on. In this way no vertex in the connected component which contains XI will be
colored once with a and once with b, which would be a contradiction.
In fact, suppose that the vertex z is colored a in a walk L I == [x l ' ..1'1 , ... , Ykt z]
of even length and is colored b in a walk L2 = [Xl' t I, •.• , [s' z] of odd length.
If Ll and have only endpoints in common. then the union ofthe of
and forms an with an odd number of which contra-
dicts the Otherwise and L2 have a number of vertices in
common, which implies that the union of the of and generates a
0111".or<.nn of G of walks and cycles. It will be shown by
induction on the number of LI that the fact that G does not contain odd
elementary cycles implies that the lengths of Ll and must have the same
parity.
If I(Ld= 1 and I(L 2 ) is even. then the elementary cycle consisting of the edge
of and the edges of is odd, which contradicts the hypothesis. It follows
that I(L 2 ) is odd. Suppose that the property is true for every two walks LI and
orthe indicated form such that /(LI t, and let and L2 be two walks with the
same endpoints such that = t + 1.
II has been assumed that and also have a vertex .r in common other
than Xl and z. Let L~ and be subwalks of and L2 contained
between Xl and Y, Similarly let be sub walks of and which are
located between y and z, Since I(L'd:;;;; I, it follows from the induction
that I(L'd and I(L 2) have the same If Xl is replaced by y, then since
/(L 1):;;;; t, one can conclude similarly that I(L'{) and I(L'2} have the same
Since I(LII==I(L~)+I(L~) and I(L 2)+/(L'2), it follows that I(Ld and I(L 2 )
have the same parity. Thus the vertices of the connected component which
contains XI can be colored in this way by two colors. continuing this process
for all connected components of G one obtains a two colors of the
vertices of G such that each two vertices have different colors. It
rollows that G is bipartite by the set A to be the set of vertices colored a
and the set B to be the set of those vertices colored by b.
It also follows that every graph G without odd which has at least one
edge. is bichromatic, that x( G) = 2,
8.6 The vertex of 9 must be adjacent to all the other vertices of the
212 Problems in Combinalorics and Graph Theory
and thus the endpoints of the diagonals of the polygon are not joined
by an Thus by joining diametrical vertices of the polygon by an each
vertex will have k, and hence a regular of k with n vertices
is obtained.
8.9 Let x be a vertex of the graph G which
Xl' ... ,x'" in G. Since G does not contain a
it follows that the vertices Xl" .. , X", are not pairwise It is
also the case that every two vertices Xh x j with 1:t;;; i, j:t;;; m and i are adjacent
to a vertex z -:/:: x. Hence z and x are not adjacent, since G does not contain a
triangle.
The mapping which associates with every pair {Xil Xj} the vertex z which is
not adjacent to x is injective. In if such a z corresponds to two different
then there are at least three vertices adjacent to both z and x, which con-
tradicts the hypothesis. Since the number of pairs {Xi> is equal to G), it
follows that the number of vertices which are not to x is greater than or
to (;).
But the hypothesis implies that for each vertex z which is not adjacent to x
there are two vertices from Xl' . , , 'X m which are adjacent to z and to x. Call
n '" 12 n '" 12
k '" 6 k 7
Fig. 8.2
214 Problems in Combinalorics and Graph Thoory
these vertices Xr and XS' The mapping which associates to each vertex z which
is nonadjacent to x the pair of vertices {xr' is also injective.
Suppose that the same pair {x,., xs ) corresponds to two vertices Zl -I=Z2< It
follows that there exist at least three vertices x, Z t. Z2 to x, and xs ,
which contradicts the
Thus the number of vertices z which are not adjacent to x is at most equaJ to
(~)< From the two opposite inequalities it follows that the number of vertices
which are not adjacent to x is equal to G), and hence
A necessary condition for the existence of this graph is that lip is an even
number, since it represents twice the number of in the graph.
8.10 Carry out the following construction: Suppose that a regular graph
G(r, g) of r and girth 9 has been constructed. Consider also a graph
G(r', 9 -1), where r' is equal to the number of vertices of the G(r, g).
Replace each vertex of the graph G(r', 9 -1) by r' vertices of degree 1 Figure
8.2).
Now identify these r' vertices with the vertices of a copy of the graph G(r, g}.
In 8.3 the G(r, g) is which is obtained for r=2 and g=5.
Denote by G 1 the graph obtained in this way. The G 1 is regular of
r + 1 by construction. It will now be shown that g( G1) = g. Consider an elemen-
tary of minimal length in a copy of the graph G(r, g). Such a cycle has
Fig. 8.3
Solutions 215
Suppose now that 9 is even, and consider two adjacent vertices x and y.
:t Si be the set of vertices at a distance i from the set {x, y} for i = 1, ... , g/2 -1.
follows that iSII=2(r-lj and ISi+ll=(r-l)ISil for i=1. ... ,g/2-2. From
IS one can conclude that
n~ISol+ISI!+'" +ISg/2-11=2+2(r-l)+ .. · +2(r_l)9/ 2 - 1
•
fig. 8.4
SolutiollS 217
[y, Xl"'"
sr.
8.17 Let x be a vertex of G with maximal outdegree, that is,
d+(x) =max d'"(t), (1)
lEX
where X is the vertex set of the graph G. It will be shown that this vertex satisfies
the conditions of the statement of the problem.
Let y e X and y If G does not contain the arc (x, y), then it must contain
(y, x). This follows from the fact that a tournament is complete. now
that the vertex y cannot be reached by a path of length 2 which at x.
Thus for every arc (x, z) there is an arc (y, z), since otherwise there would exist
an arc (z, y) and hence a path of length 2 from x to z, namely (x, y, z), which is
contrary to the hypothesis.
for every arc which originates at x [of the form z)), there is an
arc z) which originates at y and is also an arc (y, x) which at y
and terminates at x.
It follows that d+( y) > d+(x), and this contradicts (1). The vertex x which was
defined in (1) therefore satisfies the given condition.
8.18 The will be established by induction on the number of vertices
of the graph G. If G consists of two isolated vertices, let S = y}. Otherwise
there is an arc (y, xl and S =
Suppose that the property is true for all graphs with at most n and
take G to be a graph with n + 1 vertices. Let x be a vertex of and denote by A
the set {z\ (x, z) is an arc of the G}. Denote by G 1 the subgraph obtained
from G by suppressing the vertices of the set A v {x}. It follows from the induc-
tion hypothesis that G1 contains a set of pairwise nonadjacent vertices S I with
the following property: vertex z ¢ S 1 can be reached by from a
vertex yeS 1 and traversing a path of G1 of length at most 2. Two cases will
be studied.
(a) The vertices of S I v are not Let S = S 1 v
vertex z in G! such that z ¢ can be reached starting at a vertex y e
and traversing a path of length at most 2. If z e A, then there exists a path (x, z)
of length equal to L
Solutions 219
n = f k (n)k 2("i')C(k).
k= 1
Observe that 2mrepresents the number with vertex set X = {Xl" ", x n }.
In any such it is possible to label an arbitrary vertex in n ways. and hence
n 2(~) is to the number of with vertex set X in which one vertex is
labeled.
It will now be shown that the side of the formula represents the
same quantity. [n order to do this consider a G with vertex set X which
contains a labeled vertex XI> where 1 ~ i~ n. The vertex Xi belongs to a con-
nected component C of G. Let k be the number of vertices in C. It follows that
1 ~ k ~ n and the vertices of C are not joined by any edge to vertices in X
The component C can be chosen in (~) ways, the labeled vertex in C can be
"c.
chosen in k ways, and the number of connected graphs with vertex set C is equal
to C(k}. At the same time the number of graphs with vertex set X"C is equal
to 2(";: ).
Consider in turn k = 1, ... , n. One can see that in this way one generates all
the n 2(;) with vertex set X which have a marked vertex, and this observa-
tion completes the
220 Problems in Comblnatories and Graph Theory
n=l 2 3 4 5 6 7
8.21 Consider the 2/;) graphs with vertex set X == {Xl" .. , x n }. Let GIJ be
the set of graphs among them which have the following properties:
(1) XI and x J are not
(2) Let Xk be another vertex such that Xi and Xk are Then
Xk and xJ are not adjacent.
Thus for k f i, j there are three possible ways of joining the vertices Xi> Xi' and Xk'
It follows that
for every
The set of with vertex set X and diameter at least 3 is Ui<J and
hence
. 8 (n) 3"
!T! () 2 4"=0.
Thus almost all graphs with n vertices have diameter equal to 1 or 2 as n...... 00.
But there exists a unique graph with diameter equal to 1, Thus
almost all graphs with n vertices have diameter equal to 2 as n...... cc. From this
it follows that almost all graphs with n vertices are connected as n...... 00.
It can be shown similarly that almost all directed with n vertices have
the that for each two vertices X and y there is a path of length 1 or 2
from X to y as n ...... oo.
8.22 It follows immediately from the definition that this binary relation is
reflexive and symmetric on the set U. In order to prove the transitivity it will
be shown that if the Ul and U2 are found in the same elementary cycle
and if U2 and U3 are found on the same elementary cycle C 2, then there is an
elementary which contains Ul and Uj.
Traverse in both directions by at the endpoints of the edge U3;
terminate at the first vertex which is found on the . Let X and y be these
vertices on the (see Figure 8.5). It can happen that.\ or yare endpoints
of 1.13' However x since the U2 is found on the part of the cycle
delimited by X and y and which does not contain U!. The cycle (which
indicated by the heavy line in 8.4) is obtained by taking the union of (1)
the elementary walk on the cycle C 1 wh ich joins x and y and which contains the
Solutions 221 '
Flg.8.S
edge u1, and (2) the elementary walk in which connects x and y and contains
the U3'
G which contains the vertices a and b, This contradicts (2) and the
that G is 2-connected,
8.24 Let A= ,al.""a,-I' and B=[b o , b l , . " . 1. bo] be two
elementary cycles of maximal length in the graph G. Suppose that these
have at most one vertex in common, and let ao == bo be that common vertex,
Since G is 2-connected, it follows that there also exists an elementary walk
which does not pass through ao and which has one endpoint in the set of vertices
{al"",a,-d and the other endpoint in the set {bJ.""br-d, Let
[aID Xl' , , , , Xb bq ] be such a walk where k~ 0, p, q and XI does not to
the cycles A or B for 1 :( i:( k. that q. One has therefore obtained an
elementary cycle [ao. ' , . , ap • Xl" , , , Xk, ' •• , br - 1 , ao] which is longer than
A. which contradicts the hypothesis.
Now let A and B be cycles which have no vertex in common. Since G is 2-
connected, there exist two elementary walks without common vertices which
a vertex of A and a vertex of B. There will be an elementary walk of length
than or equal to r/2 which is of the cycle A and has as its endpoints
the endpoints of the two walks. An result can be obtained for the
cycle B. The two parts of cycles A and B, together with the two elementary
walks which join a vertex in A and a vertex in B, form an elementary cycle
which is longer than A, and this again contradicts the hypothesis.
8.25 Let m(G) denote the number of in the graph G. It follows that
since on the left-hand side each edge u contributes 1 for each pair {x, of
vertices which are different from the endpoints of u. The fact that there exist
(n; 2) such pairs of vertices implies (1) for i = 1,2.
By for x the G1 -x - Y is isomorphic to - x - y.
Thus have the same number of and hence (1) that m(Gd:=:
m(G 2 ). Now consider the sum
(2)
where Xo is a fixed vertex. If u is an edge incident with Xo, then its contribution
to the sum (2) is zero. Otherwise its contribution to this sum is equal to n - 3.
This corresponds to the case in which y is different from Xo and from the end-
points of the edge u. Thus it follows that
(n - 3)m(G ,) - m(Gt - Xo - y)=(n -3)d o ,(xo)·
Since the left-hand side of this is independent of i, one can see that
dGj(xo) = dG1(xo) for every vertex Xo'
In the same way one can show that
m(G j ) - m( G1 - x - y) =doj(x) +dGiY) - :x;(x, y). (3)
SolUflons 223
Here (X/(x, y)=O if the vertices x and yare not adjacent in the graph and
(X/(x, y) = 1 if x and yare adjacent in It follows from (3) that for every vertex
one has (X\(x, (X2(X, y), that is, the graphs and coincide.
8.26 LeI G be the graph with n2 vertices which correspond to the squares
of a chessboard. Two vertices are considered to be adjacent if one can be
reached from the other by a Since a knight always moves from a
black square to a white square or vice versa, it follows that this graph is bipartite.
One can conclude from Problem 8.5 that this graph does not contain an elemen-
tary cycle with an odd number of vertices and hence there does not exist an
elementary cycle with n2 vertices. It is therefore impossible for a knight to visit
all the squares of the chessboard in the manner described.
8.27 The set M of perfect of may be written as the union of
two disjoint sets:
M =M)'UvMU5'
where M}'u is the set of the perfect matchings of Gn that contain [y, and Mus
is the set of containing [u, s] (see 8.6). If a perfect
matching of contains then it follows that it also contains s], t],
and [v, and hence 1M},.!
== K(n Thus this is to the number of
perfect matchings of the graph obtained from Gn by deleting both
(X and p. If a matching belongs to MuS' then one can show that it contains the
edge [t, and hence 1M
"5\ = K(n -1). This is the number of perfect matchings
of the graph obtained from Gn by deleting the hexagon p. Since K(l)= 2, K(2)= 3,
and K(n) = K(n -1) + K(n - 2) for n ~ 3, it follows that K(n)... 1 for any n ~ L
[M. Gordon, W. H. T. Davison, J. Chern. 20 428-435.]
fig. 8.6
8.28 As in Problem 6.9, one can show that for any three vertices x, y, z of a
tree d(x, y) + dry, z) + d(x, z) 0 (mod 2) and for any four vertices x, y. z, t
of a tree the numbers d(x, d(z, l), d(x, dry, n. t)+d(y, z) are not all
distinct.
(a) Suppose that G satisfies the hypothesis and contains an odd cycle with
224 Problems in Comblnatorics and Graph Theory
2k + 1 vertices. Let x, y, z be three distinct vertices on this cycle such that d(x, y) ""
d(x, z) and z) 1. In this case
d(x, y)+d(y, z)+d(x, z)=2k+ 1 1 (mod 2),
which is a contradiction.
On the other hand, if G contains an even with 2k vertices (k> 2), let
x, y, z, t be four vertices on the such that
d(x, y) t)=l and d(x,z)=d(y,t)=k-1.
Then
First note that if there exists a subgraph spanned by shortest paths between the
vertices x, y, z which tS a tree, then the equation follows from a previous result.
Otherwise, one can assume that the vertices are distributed as in Figure 8.7 and
d(x, y) u)+d(u, v)+d(v, y),
d(x, z)=d(x, u)+d(u, w)+d(w, z),
dey, z) v)+d(v, w)+d(w, z).
It follows that
d(x, y)+d(y, z)+d(x, z)
u)+2d(y, v)+2d(z, w)+d(u, dew, v)+d(v, u)
:ed(u, w)+d(w. v)+d(v, u) (mod 2),
But this last sum is even, since it is the of a cycle m a bipartite
To complete the proof one must find vertices x', y', z' in S(d(G)-l) such that if
d(x', d)=a, d(y', d) d(z', d)=e. then a+b= y), a+c z), b+e=
dory, z), Let a == a(x, y, z) -do( y, z), b = a(x, y, z) do(x, e = a(x. y, z) y),
where a(x, y, z) = {ddx, y)+ do(x, z) + dory, z)}/2. This is the solution of the
Solutions 225
z
y
Fig. 8.7
and, for example, a= {do (x, y)+ do(x, z)- do( y, z)}/2~ {2d(G)-1}/2,
which implies that a~d(G)- L
(c) Suppose that G is not bipartite. Then G contains an odd with
2k + 1 vertices and, as in the proof of (a), one can choose three vertices x, y, z
such that
d(x, z)+ z):a 1 (mod
which contradicts (b).
One can also show that three vertices x, y. z of a graph G are ,<",mp, .. ,
embeddable in a tree ifand only if y)+d(y, z)+d(x, z):aO(mod 2). [R. A.
Melter and 1. Ars Combinatoria, 12 (1981), 111-1
8.29 There is a round trip by at least one Ai which contains an odd number
of stops. For n::::: 1 the statement is obvious, since one airline serves at least three
cities C l' and hence [C 1, , C 3' C 1] is a cycle with three landings. Use
induction on n, and suppose that n~2. One may assume that all round trips by
A. consist of an even number of stops, since otherwise there is nothing more to
prove. Because the of service by An has no odd it follows from
Problem 8.5 that this is Then one can find a of the
cities into two non empty classes {Q 1, ••. , Qr} and , ... , R.} where r + s == N,
such that each flight by An runs between a Q-city and an R-city. Since r + s =
N~2n+ l,atleast oneofr,sisgreaterthan 2n-l,sayr~2n-l + 1. But {Ql, ... ,Q,}
are only served by AI' ... , An-I, and hence the induction hypothesis at least
one of these flies a round trip with an odd num ber of landings. If there
are N =2" there is a schedule with n airlines which contain no odd round
trip on any of them. Let the cities be k 1, ... ,N-l=2"-1. Write k
as an number in the binary system (possibly starting with one or more
Link C1 and Cj by Al if the first of i andj are distinct, by if the
first digits are the same but the second are ... , under A" if the
first n -1 are the same but the nth are different.
All round under Aj are even, since the ith digit alternates for the vertices
of such a Equivalently, the graph of service by Ai is bipartite for every
1 ~ i~ n. [Problem to the jury of the 24th International Mathematical
Olympiad, 1983.]
Problems in Combinatorics and Graph Theory
APTER 9
G) ~(n;ly=_n(_n-:---_5)
This bound is for n;;;;' 6 and vanishes for n = 5. The with five
vertices is such that neither nor eontains a triangle. It can be shown that
the lower bound is attained if n + 1 where p is a natural number. [A. W.
Goodman, Amer. Math. Momhly, 66 (1959), 778-783.]
9.3 Let G=(X, U) where IXI=n and IUI=m. Let x. y E X, and denote by
A(x) the set of vertices which are adjacent to x. In similar fashion A(y) will
denote the set of vertices which are adjacent to y. ]t follows that
~ L d(y)-n}
3 [x,y]eU
triangles, since each triangle is counted relative to each of its sides, ]n this sum
d(x) occurs exactly d(x) times for every x E X, It follows that the sum is to
n
~ C~x d(X)Y -mn}= 4m (m- n:).
9.4 Consider a set of three vertices {x, y, z} in the graph. Either the is
a circuit or there are two arcs of the form (x, y) and (x, z), In the latter case each
would be the direction of the arc between y and z, the three vertices induce a
transitive subgraph of the tournament. Let Sj denote the number of arcs which
=
originate at the vertex Xi of the graph for i 1, .. , , n. It follows that the number
of transitive is to
228 Problems in Combinatorics and Graph Theory
To see this one uses Jensen's inequality and the fact that
number of circuits with three vertices is bounded above
Let t(n) denote the maximum number of Hamiltonian paths in the class of
tournaments with n vertices. The following values are known: ((3) == 3, t(4) == 5,
c(5) t(6) =45, and t(7) 189.If
_. (t(n))lm
a-lim , ,
tt-CC: n!
then '1 exists and satisfies the inequalities 0.5 == 2 - I ~ 11. ~ 2 - 3,4 <0.6. Szele's
conjecture states that 11. = 0.5, bu t a proof of this has not yet been given. Szele,
Mat. Fiz. 50 (1943), .... .jv.
..l- . .
9.6 Suppose that the graph G has n vertices and m and does not
contain an elementary cycle with four vertices. Count the number of of
vertices {x, in which both members are adjacent to a third vertex z. If the
vertex z is fixed, then there are such pairs. Each pair y} is counted at
most once, since if it was with to z 1 and =2 with z I ::/= Z2, then
[x, Z I, y, Z2' x] would form a cycle with four contrary to hypothesis.
Hence it follows that
where X is the vertex set of the G. Since the function -1 )/2 is convex,
it follows from Jensen's that
(k +1)-(k-3)n=(k-l)p-p+k 2
- (k - 1)p - s + n
=n-p+k-s-2
-p+k-s 2
r
==q+ p
s
p+k-s 2
r+s
=q+ k-s-2
;;:::q+ 1,
tion shows that n - 1 [(n -] )/(k - 1)] = [(k 2)nj(k - 1)]. It follows that a
with n vertices and without complete subgraphs with k vertices and which
contains a maximal number of must have the desired structure. Thus
M(n'k)=(;)-rC~l) (k l-r)(;)
- n - rt( t + 1) (k 1 - r ll(1 -
k-2. n
2_r2 (r)
k 2 + 2 .
9.10 First we show that for every choice of four points A, B, C, D from M,
there exist at least two points which are at a distance less than or equal to
1/J2. If three of the are collinear, then one of the distances is less than
or equal to t< I/J2 and the property is seen to hold. Otherwise it will be shown
that the configuration formed from the four points contains a triangle with an
of at least 900. There are two possible cases:
): b2 + ( 2 ): 2 min (b 2 , el ).
If b):c. it can be seen that ): or C2~ and hence c~l/J2.
which has as its vertices the 3n points; two vertices are
joined by an if the distance between them is greater than 1IJ2. From the
property just one can conclude that this graph does not contain a
subgraph with four vertices. Thus it follows from the
that the number of is at most equal to M(3n, 4)= . The
distances which are than 1/J2 can be chosen in the interval (1-1:,1) for
every e > 0, by grouping each n points sufficiently close to the vertices of an
equilateral triangle of side 1.
Solutions 233
9.11 The proof follows directly from Turan's theorem, since M(2n, 3)=n2.
The graph which realizes this maximum number and does not contain a
triangle is the complete bipartite
9.12 It will be shown that for every n ~ 2 the maximum number f(n) of
maximal (cliques) in the class of with n vertices
takes on the following values: f(n) == 3"13 for n == 0 (mod 3); f(n) 4 X 3("-1)13 1
for n == 1 (mod 3), and f(nJ == 2 X 3(n- 2)13 for n == 2 (mod 3).
For 2~n~4 the expression for f(n) can be obtained by a simple counting
argument. In fact, if n == 2 or n = 3, the maximal number of cliques is obtained for
graphs consisting of isolated vertices. In the case of n=4 one must consider
both a graph which only has isolated vertices and the complete
2'
Let G be a with n ~ 5 vertices which contains f(n) cliques. G contains at
least two nonadjacent vertices x, y, since otherwise G would be the complete
graph on n vertices K n , which contains a unique clique; this would contradict
the maximality of G.
Let V(x) be the set of vertices adjacent to x in G. Denote by y) the graph
obtained from G by all the incident with x and replacing them
with from the vertex x to each vertex in the set V(yJ.
The symbol will the obtained from G by
the vertex x, and a(x) will denote the number of contained
in V(x) and which are maximal with to the subgraph G x . Finally, c(x)
is the number of cliques of G which contain x.
By suppressing the edges incident with x, one causes c(x) a(x) cliques to
But joining x an to all the vertices of V(y) creates c(y)
Thus if c(G) denotes the number of in G, one has
c(G(x, y» = c(G) + c(y) - c(x)+ a(x).
It can be assumed that c(y);;l: because otherwise one only has to consider
the graph G(y, x).
Since dG) is maximal, it follows that c(G(x, y»~c(G) or c(y)=c(x) and
This implies that c(G(x, y» = c(G), and the graph G(x, y) contains the
same maximal number of cliques f(n) as the graph G. It can be similarly
that
c(G(y, x})=c(G)+c(x)-c(y)+a(y).
Thus, in view of the fael that c(x)=c(y), it follows that a(y)=O, and hence
c(G(x, =dG(y, c{G).
Let x be an arbitrary vertex of the and let )'1, ...• Yp be vertices
which are not adjacent to x. Transform the G into the graph G t := )'1' x),
followed by the transformation of G1 into G2 == G 1()' 2, Xl, and so on, until one
finally transforms G p - 1 into Gp=Gp_I(Y P' x), in every case the
number f(n) of cliques. The graph Gp has the property that the vertices
x, Yl"'" yp. are not joined to each other by an edge, and V(x)= V(yd= ...
234 Problems in Combinlltorics and Graph Theory
Suppose now that f(G) is attained for all variables x, > O.1f G is not complete.
then there exist two nonadjacen t say] and 2. Let F (G) = F(x 1, ... , Xn) ==
L[i.n<ux,Xj, and O<C~Xl' Then
where A represents the set of vertices in G which are adjacent to vertex 2, but
to vertex 1. Similarly, B denotes the set of vertices in G which are
to vertex 1, but not to vertex 2. It remains to show that
Xi' For if Lj<AXj> then F(XI-C,x 2+C,X3,""x n»
which is a if the inequality then
+C,X2-C,X), ... ,xnj> for every O<C~X2' which con-
tradicts the maximality of
If C=X 1, then
F(O, Xl +X2, X3,"" F(x),
and hence the maximum is attained for the subgraph G' obtained from G by
the vertex 1. This reduces to the previous case if one of the variables
takes the value zero. It thus follows from the induction
to G' that
=~ (1 _ n xi ) ~ ~ (1 _~).
and hence the inequality is also established in this case. S. Motzkin, E. G.
Canadian J. Mathematics, 17(4) (1965),533-540.]
9.16 Suppose that the arcs U 1, ... ,Urn of G are numbered so that
dull;;<l: C(U2);;<l: •.. ;;<l: c(u m ). Let k be the smallest index with the property that the
set of arcs {Ul,"" uo} contains the arcs of a path Do ... , b). Since
, •.. , Uk -I} does not contain all the arcs of any path from a to b, it follows
Co = {Uk, Uk" I, .•• , Urn} is an (a. with n A(Dol = [Here A{Do)
denotes the set of arcs of the path Do.] It follows that
min c(uJ C(Uk) and max c(u) OJ
.<Do "ECO
g(u)-g(z)~c(z, u).
Ifu, Z E S, the validity of the follows from the induction hypothesis.
If Z E Sand U ==,1', then the method of the arc (x, yJ that
g(y)=g(x)+ +
and (1) is satisfied.
Let UES and z=y.1t will be shown that in this case g(u) and hence
~O~qy. u), and thus is satisfied.
In order to prove this it will be shown by induction that the function (J
increases with an increase in the number of vertices chosen in G. This property
holds for the vertex c selected after a, since gee) g(a) + c(a, c) du, c)~ 0,
g(a).
that this property holds for every set of vertices of cardinality less
than or equal to lSI constructed by the indicated procedure. Suppose further
that there exists a vertex U E S such that g( y) < g(u).
Since g(y) = g(x) + c(x, y) ~ g(x), it follows that g(y) The induc-
tion hypothesis now implies that the vertex x was selected in the set S before
238 Problems in Comblnalorics and Graph Theory
(*)
In view of the condition (C) for the conservation of the flow for every vertex
x =1= a, b, it follows that the term corresponding to x =1= a, b in the sum (*) is zero.
Thus (*) reduces to IUEW-Ib) f(u)- I"Ew+la) f(u), since w-(a)= (b)
But since every arc (x. y) E U belongs to both of the sets wand
it follows that by terms one can express (*) in the form
{J(u)- flu)} =0.
where is the set of arcs u which have both endpoints in the set A. On the
other hand, jfthe arcu Ew-(A), then the flow f(u) appears in (**) with a plus
sign, and if 14 € w + (A) then the flow f(u) appears in (**l with a minus sign,
since there will exist vertices Xl' X 2 € A such that u € w d and u E w + (x 2)'
Comparing the two expressions for the sum (**), one sees that
f(u)- f(u)~ I f(u)~ L c(u)==c(w-(A))
.,w u,,,, -(AI .,w -(AI
This is because condition bounds the flow on each arc of the network and
the flow takes on values.
9.19 Let v be a walk which joins the source a and the sink b. Denote by
the set of arcs of v directed in the same sense as the direction determined in
traversing the walk v from a to b. Similarly v- is the set of arcs oriented in the
opposite direction.
Let
For e>O one can increase the flow fb as follows: [ncrease the flow on each arc
UE by i:, and diminish the flow by <: on each arc u E v -. It follows from the
method of defining e that one obtains a new flow j', which satisfies
O~ j"(u) ~ c(u)
for each arc u € U. The conservation condition is also satisfied at each vertex
x b. The flow at the sink since f{, = fb + <: > fb, and the last arc
of the walk r which terminates in b belongs to the set t.: oj-.
A walk for which I: = 0 is said to be saturated. Thus if a flow f realizes max 1'0.
there will not exist a nonsaturated walk from a to h. Otherwise the flow could be
increased at b.
Let f be a maximal flow in the network and consider all the
walks from a to b. all the arcs u for which there exists an elementary
walk v such that 14 € V and flu) = c(u) or U E v- and Here u is the first
arc encountered with this property in the walk v from a to b.
The spanning graph thus obtained has at least two connected components,
since otherwise there would exist a nonsaturated walk from a to b, which con-
Problems in Combinalori<:5 and Grapb Theory
But it has been seen that for every flow f and every cut induced the set
A c X with a ~ A and b E A one has the inequality
fb:::;; c(w - (A)).
This also follows from the preceding problem.
Thus a maximal flow f and a cut have been found for which the
inequality becomes an Hence the cut (jJ has a minimal capacity
which is equal to the maximal at the sink.
9.20 It is clear that the has a finite number of steps, which
is bounded above, for by In fact, it follows from Problem
9.18 that max f fb:::;; c(w -(A)), where is any cut of the network. If A {b},
then w-(A)=w-(b). From this one can deduce that max! c(w-(A))
L.,"qb) cluj. Start with a zero flow on each arc, so that fb=O. At each step the
flow fb increases e. Recall that the of an arc is a UVI1-UIOl'."'"
Thus the values of the flow and of e itself are and e > 0 that
e ~ 1. Therefore, at each step the flow increases by at least 1, and hence the
number of steps of the algorithm is bounded above by the capacity of a cut.
It will now be shown that when one can no longer mark the sink b, then the
flow obtained has the maximal value at the sink. The set of arcs which join a
marked vertex to an unmarked vertex constitutes a cut with minimal --~'--"J
To show this let A be the set of vertices which cannot be labeled by the
algorithm. It follows that a ~ A and b E since it has been assumed that the
sink of the network cannot be labeled. Thus w - (A) is a cut of the network. One
has flu) == for every arc u E (j) - (A), since if u = (x, y) then x ~ A and yEA.
rr f(u) < c(u). the vertex y could have been labeled, which contradicts the fact
that YEA. Also one has flu) = 0 for every arc u E w + (A), since if u = xl then
yEA and x f. A. If flu) > O. the vertex y could have been labeled starting from
the labeled vertex x, which contradicts the fact that J' EA.
By applying (b) of Problem 9.18 one can write
fb= L f(u)- flu)
UEw-(A)
Soilltion~ 241
But it has been shown that for every flow f and every cut w-(A) one has
fb~ -(A)). For the flow and cut this inequality becomes an equality.
Thus the flow is maximal, and the cut w - (A) has minimal capacity.
This reasoning also provides a new proof of the F ord- Fulkerson theorem of
the problem in the case when the capacities of the arcs are
Observe that this algorithm allows one to find the maximal flow after a finite
number of steps in every network G whose capacities are rational numbers.
Let the arcs of the network be U I, ..• , Urn with c(ull pdq; where Pi' q/ are
integers and 1 for every 1 ~ i ~ m. Multiply the capacity of all
the arcs by the least common multiple , .. " qmJ of the denominators of these
fractions. One th us obtains a transport network G I which has the same
and whose are Let f be a flow in the network G I , and
the flow g by
(1)
for every u e U. The conservation condition at every vertex x =/= a, b and the
boundedness on each arc will be satisfied for the network G.
In from f(u) = +(x) f(u) it follows that
so that A'I has flow reserve 0, A2 has reserve an' A) has reserve an + I, and
has reserve an + I' Notice that for 11 = lone can take Ai = AI for 1~ i ~ 4,
at 11, choose a walk from a to b which includes among the
arcs Ai only the arcs and A); for example.
[a, Y), b],
It follows that 1:1 =min (e an' e-f(Y2. an+l. e-f(Y3. b)). Thus by
the induction hypothesis the amount of flow sent to b up to the present on the
arcs of the network is to ; Qh and thus f(a, : ai' This implies
that c f(a, n ai > an > an+ l ' Similarly one can deduce that
c - fCV2' x)) > an C - f(y3. b) > an and hence £) = Qn+ l ' Increase the flow
on the arcs x~), ,y~), (y2. (X'J' Y3), and ( • b) an + I' The flow at b
then increases by an + I' The arcs A'I' , A~ have reserves of flow equal to
° ,
0, an - an + I = an + 2, and an + 1 respectively.
Now choose a walk v from a to b on which one can increase the flow at b
so that A~ E v+ and v- = {A'!. For example.
v= ~.~.~, ,~, .~,~
I
Solutions 243
max f = 4c, and the maximal flow can therefore be obtained as
follows:
f(a,xd= ,Y2)= f(a, x21= f(xz, l'll
f(Yl. f(a, == f(X3 \ Y4) = f(Y4. b)
=f(a, , b)=c.
and on the other arcs the flow wil1 be equal to zero. [L R. D. R.
Flows in Princeton University 1962.]
Fig. 9.2
Fig. 9.3
244 Problems in Combinatorics and Graph Theory
no flow ~ith integral components can have values other than 0 and 1 on the
arcs of the network. The arcs of the form y) with x E A and y e B which have
flow equal to 1 determine a of the
[x, yJ of the graph C. In fact the arcs with flow equal to 1 cannot have a
common endpoint. since the entry arcs and the exit arcs of the network both
have equal to 1. Thus they can have a maximal flow equal to 1. The
num ber of arcs u y) which have a flow feu) == 1 is thus to the number
of in the associated matching, which is, in turn, equal to the flow fb
at the sink.
The maximal flow in the associated network to a maximal
matching in the graph C obtained by the arcs between
A and B which have a flow equal to 1. It follows that fb' The Ford-
Fulkerson theorem implies that max fb minT c(w-(T)), where T is a set of
vertices such that a f; T and beT. Note that for T = A v B v {b} one has
c(w-(T)) IAI and C IAI + 1. It follows that the minimal capacity of a cut in
the network is attained for a cut which does not contain an arc of the form
(x, y) with x e A and y e B. Thus c(w-(T)) c(w-(To)), where
w ={(a,x)lxeAo}v bllyeB o}, Ao AandBocB.
Now we show that AovBo is a support for the G. To do
consider an arbitrary edge [x, yJ e U to which corresponds the path (a, X, y, b)
in the network associated with C. Since every path from a to b contains at least
one arc of a cut in the network, it follows that either (a, xl e w- (ToJ and hence
x eA o• or b) ew-(To) and hence y e Bo. Thus one has shown that AovBo
is a support ofC of cardinality IAovBol=IAol+IBol=lw-(Toll= ==
vIC), since each arc in w-(Tol has a capacity to 1. It follows that
rIG) ~ IAov Bol =v(C).
On the other if V is a set of which form a and S is a
set of vertices which form a of it follows that IVI ~ lSI. (This is because
the in V do not have an endpoint pairwise in common, but each one has
at least one vertex from S.) This implies that v( C) ~ '1:( C), which, together with
the previously established opposite inequality, establishes the equality of the
two numbers for the
9.23 The matrix A can be considered to be the matrix of a
C Y, U), where X = ... ,xnL y = {YI'" . , and U = {(Xi.
The maximum number of elements to 1 which are found on diITerent
rows and columns corresponds to the of a maximal matching of C,
which contains v( C) If the rows i l , .• ,. i, and the columns j l ' .. ,
together contain all the elements equal to 1 in the matrix A. then , . , .,
Xi" Yit" ..• Yj,} is a support of the graph G.
The minimal number of rows and columns with this property is thus equal
to ,(C). By the preceding problem, v( Cl = r(C), which completes the proof.
9.24 Let E(n) = n:(n _1)2]. The proof is by induction on n. For n ~ 3 the
result follows by all cases. Suppose that the property is
valid for all graphs with at most n and let C be a graph with n + 1
Solutions 245
vertices. Let G., denote the subgraph obtained from G by deleting x and all
incident to it. It follows from the induction hypothesis that D2( G.,)::;; E(n).
Hence by using at most (!Xj and (fJ), G can be transformed into
K n, v wherenj:;;;'O,n2:;;;'O,andnl+n2=n.Ifxis tOPl vertices of
K n, and to P2 vertices of then G can be transformed into Knl v 1 or
.,.jvK 2 either PI operations and n2 P2 operations ({3) or P2
operations (:x) and nl-PI Let zl=Pl+n 2 - and
P2 + nl - Pl' It follows that Zl +Z2 = n, and hence min , Z2)
and min 1) for odd n. one has D2( Gx) +
min . z2)~E(n+ 1), since E(n) =E(n+ 1) for even nand E(n)+t(n-l)=
E(n+ 1) for odd n.
In order to characterize the extremal G with n vertices with this
property, observe that if DleG) = E(n), then for any vertex x of G one has Dl(G x) =
E(n -1). The characterization of all G such that r5 2(G} E(n) can now be
obtained induction on n. For n~3 it can be shown that all extremal
are complete bipartite that this property is true for all
with at most n and let G be a graph with n + 1 vertices such that
D2 ( G) = E(n + 1). If the subgraph with n vertices is composed only of isolated
vertices, then x is also isolated or it is adjacent to all vertices of , since other-
wise there would exist a vertex y such that Gy is not a complete bipartite
graph, and hence D2(G y) <E(n) by the induction hypothesis. But this would imply
that < E(n + which is a contradiction.
Thus G contains only isolated vertices G=K o.n + tl or G= .n'
A similar proof can be used when Kp.q where p, q> 0
and p+q=n. It remains to show that =E(n) when p, q:;;;,O and
p+q=n. For one may obtain a of x vertices from
the set with p and y vertices from the set with q vertices of . the
remaining vertices constitute the second clique. Hence the number opera-
tions (!Xl and ({3) is equal to(~)+ 2 +(i)+(q?)+x(q- Y)+ YlP-X) (x-
(p-q)(x- Y)-tn+t(p2 +qf). This expression has a minimum equal 10
O~x~p, O::;;y~q, p+q=n. and this minimum is reached only if x- r
for even n and x Y=f(p q-l) [or t(p q+ l)J for odd n. For the
Ko,n with n isolated vertices one finds that Gl+ E(n) if x = n, and equality
holds only for x=y for even n and x - I ) , +1) for odd n.
[M. Petersdorf. Wiss. Z. Techn. Hochsch. Ilmenau 12 (1966), and
1. Math. et Sci. Humaines, 42 (1973),
Let maxG where (}.(G) represents the minimum number of opera-
tions (~) andior ({3) which transform G into the union of k disjoint cliques (some
of them may be empty). It is known that ()I =(~) and Dk [;t(n -1)2J for any
2~k~n. [I. Tomescu, Discrete Math., 10 (1974), 173--179.J
CHAPTER 10
Fig. 10.1
coloring the vertices of each part A and B of with the same color one
obtains a coloring of the vertices with two colors. Since all the edges of GT
are of the form [a, bJ with a E A and b E B, it follows that each two adjacent
vertices have difTerent colors.
In the same way, consider the dual graph of the faces in the exterior of C.
It can be seen that one can also color the faces outside of C with two other
colors. One has thus found a with four colors of the faces in a planar
of the graph G, with the desired property.
10.4 Denote (x!. . .. , Yn) the coordinates of the points of inter-
section with to a pair of perpendicular axes. One can assume that the
directions of the axes are chosen so that the abscissas Xl' .••• X are pairwise
M
for Xl <X2 < ... <X n • Color the vertices of the graph in this
order with three colors. If one has colored with three colors the vertices with
abscissas x I •... , Xi _ 1 , then the vertex (x i, JI) has at most two vertices
which have already been colored, since there do not exist three concurrent lines.
Thus there is a third color usable for the vertex Yi) for i 2, ... ,n. The
under consideration is therefore satisfied.
10.5 The property will be established by induction on the number j of
faces of the graph G. If f = 1, it can easily be seen that G is connected and
does not contain cycles; it is thus a tree whose only face is the infinite face. In
this case, the formula is satisfied, since m = n - 1. Suppose now that f> 1 and
that the is true for all planar connected graphs with at most f -1 faces.
248 Problems In Combinatories and Graph Theory
or
m~ 3n-6.
If G does not contain triangles, then one can show similarly that
2m! ~41 =4(m n+
or
m~2n-4.
10.7 Suppose that the complete graph with five vertices is planar. Then
l=m n+2=10-5+2=7.
It follows from the """un,/', problem that
20=2m~31=
v+ 1 m=2, (1)
where [' is the number of f is the number of faces (including the infinite
face), and m is the number of edges of the G. Since each face has three
and each edge to two faces, it follows that
312m. (2)
Suppose now that each vertex has degree d(x)?':: 6. In any graph one has the
relation
L =2m.
since every edge [x, is counted twice, both in d(x) and It is also the case
that
m
2m?':: or }?':: 1:.
Fig. 10.2
and similarly
-2).
<-g-
2g
-4)+- in-2).
g-2 g-2
if C contains edge whose causes C hecome disconnected,
then the inequality is proven and it has been seen that the inequality is in
fact strict.
It remains to consider the case in which C does not contain an edge [a. b]
elimination transforms C into disconnected Let Jibe the number
faces which edges for i let j' be the number of
representation of C. It follows
since in this case each edge [a. h] lies on the boundary between two faces. Thus
2m=
Solutions 251
m,;;. (n 2).
g-
Observe that holds only for 1 = jg+ 2 = . 0, and thus when all
with g vertices. This case can be realized for certain
values of n. for ""<HHI"''', for n = g.
10.10 The property will be proven by induction on the number of vertices
of the G.
If G has at most five vertices, the property is immediate.
that every graph with at most n vertices has chromatic
number less than or to 5. and let G be a with n + 1 vertices.
It follows from Problem 10.8 that there is a vertex x in G with d(x)~5.
Denote by Gx the obtained from G the vertex x
and the incident with x. the induction the vertices of
can be colored by at most colors so that each two adjacent vertices
have different colors. If d(x)';;' 4 then one can
color, difTerent from the colors of the vertices to x. Thus the vertices
of G can be colored with at most five colors, that is, X(G)';;'S.
that d(x) 5 and that the vertices YI' 1'2' Y3' .1'4, Ys to x
are colored with at most four colors in the coloring with at most five colors of
One can find an available color for x, and the property is established. The
only case is that in which d(x) = S and the vertices Yl, . .. , Ys are
colored with exactly five colors, say A. B, C. D. E, in the of Gx
10.3). Consider the connected component M which contains YI of the subgraph
of consisting of vertices colored with either color A or color C. If the vertex
Y3 does not belong to this component, one can interchange the colors A and C
in M to obtain a coloring with five colors of the graph Gx in which YI has the
color C.
Thus the color A has become available, and the vertex x can be colored with
A to obtain a coloring with five colors of the vertices of the graph G. Otherwise,
If .h and Y3 belong to M, it follows that there exists a walk which )'1 and
)'3 and which contains alternatingly the colors A and C.
Consider the connected component N which contains Y2 in the subgraph
consisting of vertices colored B or D. In this case Y4 ¢ N. In fact, there would
otherwise be a walk with endpoints Y2 and Y4 whose vertices are colored alter-
Band D. Since the is it would turn out that this walk
must have a vertex in common with the walk which YI and Y3 (see
lO.3). But this is since these two walks consist of sets of vertices for
which the sets of colors are disjoint. Thus Y4 " N. By the colors
252 Problems in Combinalortcs and Graph Theory
o
o
Fig. to.3
0
Band D in the component one obtains a of the subgraph with
five colors in which the vertex Yz has color D. Thus the color B becomes avail-
able. One can color x with B, and this produces a coloring with five colors of
the vertices of the G.
In 1976 K. Appel and W. Haken proved the Four-Color Theorem. It states
that for every planar graph G one has the inequality X(G)~4. used an
electronic computer to study more than 1900 configurations which occurred
in the proof.
10.11 The graphs with the smallest number of vertices which do not contain
and which have chromatic numbers x(Gd=3 and X(G 2 )=4 respec-
tively are illustrated in 10.4 and 10.5.
Solutions 253
Fig. 10.5
most equal to the number of nonempty subsets of the set of colors, that is,
2'" 1. This is a contradiction, and hence X(G) = IJ". [A. Gyarfas, Discrete Math.,
30(2) (1980), 185.J
10.13 One can assume that all the faces in a planar representation of G
are triangles, that is, G is a triangulation of the plane. In fact, if G is not a triangu-
lation, one can add new to the graph G until one obtains a triangulation
GI If the desired inequality is satisfied by GI, it will be satisfied further for G,
since the degrees of the vertices of G I are larger than the of the vertices
of G. Thus let G be a planar graph with all faces for which the sum of
the squares of the is maximal. Let x be a vertex of minimal degree.
It will be shown that = 3.
In fact, if d(x)~ 2 it follows that d(x) = 2 and the G reduces to
which contradicts the that n~4. that d(x) ~ 4, and let
Xl' X 2' ... ,x, be vertices adjacent to x such that d(x d(x;J for i=2, ... , r,
where r=d(x)~4 (Figure 10.6).
Suppress the edge [x, x 1] and add the edge , x,.] to produce a new planar
graph G l without multiple edges for r~4.
Let S denote the sum of the squares of the of and S 1 the sum of the
squares of the o f . It follows that
d2
= 2d(x,l + 2d(x 2) - 2d(x) - 2d(x 1) + 4> 0,
since ~ d(x d~ d(xtl for i = 2, ... ,r. However, this inequality contradicts
the maximality of the graph G, and thus every graph G for which S is maximal
contains a vertex x of d(x) = 3.
The proof of the property will now be completed induction on n. Let
n =4, and consider a of the K 4' In this case both
sides of the inequality are other with four vertices has a
sum of squares of its of some vertices
the
now that the with at most n
vertices, and let G be a planar with n + 1 vertices for which the sum of the
squares of the is maximal and which has all of its faces triangular. By
fig. 10.6
Solutions 255
Fig. 10.7
256 Problems in Combinalorics and Graph Theory
c
Fig. 10.8
face of the graph G) which is joined by edges to a, b, c as in Figure 10.8. The new
infinite face of the graph is bounded by the cycle [b, c, X n +1, b). Also d(xn+ ==
3, deal 4, deb) = n, and = n. G 1 is a triangulation, and =S +9 +
2(n - 1)2 + 16 - 9 = S + 4n + 14. the induction hypothesis S =
+3)2_ and thus SI= +3)2-62+4n+14=2(n+4)2 62 and the
property is established.
the fact that the function x" I :[0, is convex for IY. ~ 2, it can
be shown analogously that
PG(A)=i."-IA 1 U U'" U
v,
where p{V, ic) represents the number of functions f:X ...... p, ... , A} which take
on the same value for the of every in V. It follows that these
functions have the same value from Lhe set {1, ... , for each connected com-
ponent of the graph (X, V) of G. Thus p(V, = ),r(V). This the
nrpccu,n for the chromatic of the graph G in view of the fact that
Solutions 257
the term ;.n is obtained for V when the graph (X, V) consists
of isolated vertices and thus has 11 connected components.
Observe that a term of degree n is obtained for V =0, ),",
but terms of degree 11 t are obtained for = I. In this case there are (7) =m
terms equal to - )."-1, and hence
PG().)=}."-m)," 1+ "',
where m is the number of edges of the graph G.
If P G().} =).» + 1+ 2 + ... + an, then one can show similarly that
a2 - C3(G), where the number of triangles of the G.
Terms of degree n - 2 are obtained for = n 2, when IVI = 2 and the span-
graph (X. V) of G contains only two or noL This also
occurs if IVI = 3, when the spanning contains three which a
triangle.
Since two can be chosen in (~) ways. one has indeed obtained the
for the coefficient a2'
10.15 If a f of the graph G e has the property that fix) 't' flY),
where e= [x, then f is also a for G and vice versa. It follows that
-e(A)-Pd),) is the number of those f of G - e with the property
that f(x) = f( y).
A A-coloring f of G e of this kind a A-coloring g of G Ie
g(z)=f(x)=f(,l') and g(t)=f{l) for every I+Z. Conversely, every },-coloring g
of the graph GIe induces a }.-coloring f of G e with the fix) =
defining fly) and fit) g(t) for 1,* x, y. Since both corre-
spondences are it follows that
a/~ i- ~) by induction on the number of edges ofG. IfG has the minimal number
of edges, equal to 1'1-1,
then it is a tree, and by Problem 10.16 its chromatic
polynomial is equal to
+( _1),,-1 (1'1-1) x.
\1'1-1
This establishes the inequality G, ~ i_I) for 1 ~ i ~ 1'1 -1.
Suppose that the property is true every connected with n vertices
and m such that
n-l~m -1.
The property will be proved for a connected graph G with n vertices and p ~ (;)
"lr\rt()·(P that G is not a tree. Then there exists an edge e such that G - e
and PI at W2 it turns out that the color (J. becomes available and it is
(J.
from among the colors of the incident with 1), it follows that
e2 = \\'2J can be colored with IX and this yields a coloring of the
the graph G with at most D + 1 colors,
It remains to consider the case in which the vertices v, \1'1' W2 belong to the
same component of the H{IX,
(b) Let P2 of PI be a color which is from the edges incident with "'2'
One can suppose that is the color of an edge incident with r, for otherwise
the proof could be finished coloring the e2 with P2' Let e3 == [v, tV,J be
the edge colored P2 which is incident with /J, Delete the color from the
e3, and color e2 with P2, since P2 does not occur at W2' Following the same
reasoning as in (a), one can consider only the case in which v, 1\'2' W3 belong to
the same component of the spanning graph H(IX, P2)'
(c) The num ber of colors is and the colors themselves PI' P2' P3' ' , .
are distinct. It follows that either one can color all the incident
with (and obtain a of the of G with at most D + 1 colors)
or else one arrives at the following situation: One can no use
of type or (b) to recolor the last WkJ which has color Pk -1' This is
because every color from Wk is a color Pi where i < k 1.
Similarly one can suppose that v, 11'" "',+ 1 belong to the same component C
of the spanning graph H(Ct., Pi)' and thus there is a walk with endpoints Wi and
\\if+ 1 which consists only of colored alternately IX and Pi' The color Ct. does
not occur in the incident with v, and PI does not occur in the incident
with 11','+ l' It follows that C is a walk from v to W i + 1 which passes through Wi
and which contains only edges colored Ct. and Pi alternately (Figure 10.9).
This walk does not contain the vertex Wk, since Pi does not occur among the
colors of the edges which have an endpoint at Wk' If is the of
Hlr1., which contains Irk, then C and Clare disjoint. This roJlows because
Fig, 10,9
262 Problems in Combinatorics and Graph Theory
E~----+-:---\-----~B
4
Fig. 10.10
Solutions 263
E~--------------------~
Fig. 1O.l!
It follows that by taking a new vertex F and joining it to all the vertices of the
pentagon, the coloring of the of is uniqucly extended to a coloring
of the edges of with five colors. For example, the AF is given color 4,
which is available at vertex A, and so on (Figure 10.11). This construction can
be to any two graphs and 1 with 11 odd.
The inequality becomes an equality only when x is an isolated vertex and the
subgraph has a maximum number of minimal This follows from
the fact that the vertex x can be added to a minimal of with k
classes in at most k different ways.
If X(Gx)=k 1. then a minimal coloring of G is given as follows:
, ... , l' where the sets 1 consist of pairwise nonadjacent
vertices and there exist k 1 vertices Xl E , ... , Xk _ 1 E C k _ 1 which are
to x. Otherwise one would have X(G):S;; k-1.
Let X be the vertex set of G. It follows that every of G has a class
which contains the vertex x and a subset of the set X "{ Xl' ••. , Xk - d. Since
x I, ... , Xk _ I} contains n k it follows from the induction
hypothesis that the number of of X which contain in the same class
the vertex x and r vertices of the set X Xl' ••. , is bounded above
r - 1)" -k-, for O~ r:S;; n k. In fact the r vertices can be chosen from the
II - k vertices in ." ) distinct ways. The maximal num ber of (k - 1)-colorings
ofagraphGwithn k-r vertices and X(G) k-lisequalto(k l)n-k-r.Thus
n-k (
Cm(G):S;;r= n~
k) (k
holds only when Xl'" "xk-d induces a subgraph and
n - k vertices are isolated. Thus it has been established in-
duction that ~kn-k for every graph G with n vertices and chromatic
number equal to k. The upper bound is attained only when G consists of a
k-subgraph and n - k isolated vertices. Observe that this is the
unique graph G with X(G) = k and the minimal number of [I.
C. R. Acad. Sci. Paris, A272 (1971), 1301-1303.]
10.23 Let C(n, k) be the desired number of colorings. It is clear that n) =
C(n, 2) = 1 for every n ~ 2, and C( n, for k> n. It will be shown by induction
on n that C(n, k) == S(n - I, k - 1) for every tree with n vertices. For n == 2 the
property is obviously true. that the property is true for every tree with
at most n -1 vertices. It follows that C(n, k) = S(n - 2, k- 2) + (k - l)S(n - 2. k-l).
In fact every tree A with n vertices contains a vertex x of L such that
the Ax obtained from A by the vertex x is a tree with
n - 1 vertices. The set of C(n, k) colorings of A can be as the union
of the set of for which the vertex is alone in a class of the partition
[there are Sen 2, k - 2) such colorings, the number of (k - 1)-colorings of
the tree Ax] and of the set of in which the vertex x occurs in a class
with other vertices of the tree. Since x is to a unique vertex
of the tree, it follows that x can be added to k -J classes of a k-coloring
of a tree with n -1 vertices. by the induction hypothesis this set of color-
has cardinality (k -1)S(n - 2, k I). It follows from the recurrence relation
for numbers of the second kind (Problem 3.4) that C(n, k)=S(n 1, k -1)
for every tree with n vertices.
10.24 The property will be proven by induction on k. For k = 1 the graph G
Solutions 265
does not contain an edge. and it is sufficient to let H = G. In this case dH(x) =
ddx) = 0 for every x EX, and the graph H is monochromatic. since it consists
of isolated vertices.
Suppose that the property is true for all graphs which do not contain a
complete subgraph with k + 1 vertices. and let G be a graph which does not
k + 2 vertices.
degree in G. Denote
adjacent to x, subgraph induced by
Since G1 complete subgraph with
from the induction that there exists a graph
which is k-chromatic
for every Z E :.
Let H be a graph with vertex set X. In H all the vertices of X ",",X 1 are joined
to all the vertices of X l' Also adjacent are all pairs of vertices in X 1 which were
adjacent in H l' The graph H is (k + I)-chromatic by construction, since X(H 1) = k.
If Y f' X 1 then
dH(y) = Ix II =ddx)~dd.\'),
and if Y EX 1, it follows that
-Ix 11 +dH,(Y)
I-IXII +dG,(y)~dd
which establishes
Now let G vertices and without a with
k + 1 vertices, monochromatic subsets
Denote IAI . ,. + Qk = n, and let
the graphs G and H respectively. It follows that
1 1
IUI="2 L
.>:EX
dd x )::;;"2 I
x,X
dH(x)=IVI::;; L
I.; i<j<;k
aiaj,
since there are edges in H only between vertices from distinct sets Ai and Aj'
The last sum is maximal if and only if lai-ak.::;l. Thus aJ = ,., =a,.=
m + 1 and ar + 1 = ' . , = ak = m, where m = [11/kJ and 11 == r (mod k), Observe that
this limit is attained only if the last inequality becomes an equality and hence
in the class of graphs. (See Turim's 9.9.)
[Po Erdos. Mat, 249 - 251.J
10.25 possesses the maximum numher class
of graphs with chromatic number equal k, number
of vertices WIth to 11, for 1::;; i::;; k (111 clear
that any two colors are adjacent.
multipartite nl ~ n2 + 2, then vertex
from the class with nl vertices into the class with n 2 vertices, thus obtaining a
new multipartite complete graph G1 with 11 vertices and mJ edges, It follows that
Problems in Combinatorics and Graph Theory
ml -m=n 1 -1-n2;;;:; I,
which is a contradiction. Hence jn;-nr:(1 for any l:(i.j:(k, and G is iso-
morphic with the Tunin defined in Problem 9.9.
10.26 LetGbea with n vertices. and let =;'"-U n 1+, .. +
( 1)"- la It follows that an _ 1 is the number of of G, and X(G) = r if and
if P G(l)=P G(2) = ... =PG(r 1)=0 and >0. If P G().) = PT(n.kP·), it
follows that ZIG) x(T(n, kJ) = k -\ and that G and k) have the same
number M(n, k) of From Problem 10.25 one can conclude that G is
isomorphic to T(n, k). [c.- Y. Chao, G. A. Novacky, Discrete M a/h., 41 (1982).
139 143.J
10.27 One can show that Pdk) = i!(~)Ci(G) holds, where Ci(G) stands
for the number of of G, are partitions of the vertex set of G.
In fact, i colors can be chosen from the set of k colors in (~) different ways, and a
partition into i classes generates i! into account the order of
the classes. The formula for results from the inverse binomial formulas
(Problem 2.17) if ak = P G(k) and
~ (-ljiG)PG(k-i).
10.28 The set of points having coordinates (0.0), (t, i), H, - and (I. 0) is a
4-c1ique for G4 • and the points (0,0), (0, I), (1, 0), and (1,1) induce a 4-c1ique for
,which that X(G 4 );;;:;4 and X(G 8 );;;:;4.1t remains to define a 4-coloring
for G4 and . In the case of G4 consider all Jines with slope 1 or -I passing
through the of E 2 having coordinates (digital The inter-
section points of these lines are points M( p, q) where p, q E:!l and points
N(r/2, .>/2) where r, S E :!l and r, s 1 Let S denote the set of all such
Color M(p, q) E S with color r:J. if p=q (mod 2) and with color fl if
=
p q + 1 (mod Color =
E S with the color y if r s (mod 4) and with
the color () if r = s + 2 (mod 4). For P( u, vJ E S consider the points Q= 1, v
and R If P is colored with the color a E p, Y, o}, then all interior
points of the segments PQ and PR will also be colored with the color a. In this
way any square ABCD having vertices in S and length of a side equal to Ji/2
will have its four vertices colored with 0:, (3, i', 0, and the colors of the sides will
be a, a, b, c, where u, b, c E fl, y, In this case color all interior points of
ABCD with the color a. Thus all points of will be colored with four colors.
I t is easy to see that if d4 (E, F):.: 1 then E and F have different colors.
A 4-coloring of Gs may be defined in a similar manner. Let S denote the set
of digital points of E2, and color the points of S in the following way: the point
M(p. q) with p, q E :!l will be colored with the color ct. if p =1 (mod 2) and
q=l (mod 2): fJ if (mod 2) and q=O(mod 2); ')' if p=l 2) and
q == 0 (mod /3 if p =0 (mod 2) and q == 1 (mod 2). If Mlp, q) E S is colored with
Solution.~ 267
CHAPTER 11
contains exactly tIn 1) 1n! cycles which pass through each vertex exactly once.
11.2 For h=O there are (n-l) Hamiltonian cycles in the complete
graph and the formula is seen to be satisfied. In fact, every Hamiltonian
determines a permutation of the vertices of Kn; the number of
cyclic on n elements is equal to (n-l)!. It is to obtain
two distinct circular from the same Hamiltonian
the cycle in both directions. Thus the number of Hamiltonian in Kn is
equal to tIn - I)!. This formula can also be established by induction, since for
n=3 the graph Kn is a Hamiltonian Suppose the formula holds for n~m,
and let "'I be the complete graph with m + 1 vertices: Xl, ... , Xm.,.j' Every
Hamiltonian cycle in Km+ 1 can be obtained from a Hamiltonian cycle of
by the vertex X m ... 1 between two adjacent vertices of the Each
Hamiltonian of generates in this way m different of K m +!. and
the induction hypothesis K", has -1) !/2 Hamiltonian It follows
that the number of Hamiltonian in 1 is equal to ((m-l)!/2}m=m!/2,
and hence the property holds for every m~3.
Let h~ 1, and suppose that h are each a new vertex. One
thereby obtains a complete graph with n - h vertices which has (n - h -1}!j2
Hamiltonian cycles. Let z be a vertex which has replaced the edge in one
of the (n - h 1) !/2 Hamiltonian cycles, and let u, L' be to z in this
cycle. The can the vertex z in this Hamiltonian cycle in
exactly two distinct ways' The walk z, t] is Cu. X, y. t] or by
y, x, After out this for all the vertices which represent
theh one obtains {(n-h 1) h-lJ! 1 distinct
Hamiltonian cycles which pass through the h
Observe that every Hamiltonian which passes through the h can
be obtained without repetition in this way. The h have no vertices in
common, and hence it follows that 2h ~ n.
11.3 Let the vertices ofK n be numbered 0. 1.2, ... and consider the fol-
lowing Hamiltonian
C1 [0, 1,2, 2k, 3, 2k- L 4. 2k 2.5 .... , k+3, k, k+2, k+ 1, 0],
k+ 1
Fig. 11.1
which x and y.
Denote by Zlt"'" Zi, the vertices ~~,,~~ ..... with x, where 2 = il < i2 < .. < ik ~ n.
It follows that y is not adjacent to 1 for s = 1, ... ,k, since otherwise G
would contain a Hamiltonian cycle
, . , . , Zit-a, Zn' Zn-l"" ",Zis'
In this case one finds that Xl is adjacent to the vertices XH 1, ... ,X,- l' Xl+ l' ..
Xn. which implies that
270 Problems in Combinatorlcs and Graph Theory
Figure 1 Otherwise there would exist an index i such that 1::;;; i::;;; 2n - 1
and the vertex Xi+ 1 is to Xo, but Xi is not adjacent to Xo. It follows from
an earlier observation that the vertex Xi 1 is adjacent to X2", and one thus
obtains an elementary cycle with 2n vertices in the graph G. namely
X} _ 1 X}
Fig. 11.2
272 Problems in Combinalorics lind Graph Theory
n+ 1,
which contradicts the that Gis n and hence dev!l == n.
(c. St. J. A.
An extension of this result is the following: If G is an n-regular graph of order
2n, n ~ 3, and G =F or G is an n-regular graph of order 2n + 1, n ~ 4, then Gis
Hamiltonian-connected (every two distinct vertices of G are the endpoints of
a Hamiltonian walk in G). [1. Tomescu, J. Graph 7(983),429-436].
It can be shown analogously that there is a non-Hamiltonian
graph of n with 2n + 2 vertices. It consists of two disjoint copies of the
complete
11.7 Let C be an cycle of maximum in the graph G,
and suppose that C is not a Hamiltonian Denote Gl a connected
component of the obtained from G by the vertices of
the cycle C. Let Xl, .•• , be vertices of C which are
No two ofthese are in the cycle C. If, for Xi and Xj are adjacent,
one can the of the cycle C a walk with endpoints Xi and Xj
which passes intermediate vertices of In this way a cycle longer
than C is obtained, which contradicts the hypothesis.
Now traverse the C in either sense. and let h, .... .vs be vertices which
are adjacent to Xl> ..• , Xs. It follows from the observation
that.vi 1£ {Xl' . . .• for every 1 ~ i~s. It can be shown .vl>"" Ys are
wise nonadjacent. if .vi and Yj were adjacent one could suppress the
[XI,.VJ and of the and replace them the Lv" Yj] and
a walk with Xi and Xj which passes through vertices of the graph Gj •
longer than whose existence contradicts
Fig. 11.3
Solutions 27:1
Recall that x I. . , x, are all the vertices of C which are adjacent 10 vertices
of G1 It follows that if Yo is a vertex in G 1> then the set S {)'o, )'1' .. , , is an
independent set of vertices. By from G the vertices x 1> ••. , x, one
obtains at least two connected one of which is G l' Since the
G is it must contain, definition, disconnecting sets with at least k
vertices and hence s:;;;: k. But this implies that
ISI=s+l:;;;:k+l,
which contradicts the hypothesis. Thus G is Hamiltonian. [V. Chvatal, P.
Discrele M Gill .. 2 (1972), 111-11
11.8 The property wi!! be established by induction on n. If n 3 then m:;;;: 3
and hence G is the K 3, which is a Hamiltonian Suppose thai the
with at most n - 1 vertices, and let G be a graph
with n:;;;:4 vertices and m:;;;: 2. I) + 2
It will be shown that G contains a vertex x of degree = n - 1 or = n - 2.
Otherwise one would have d(x) ~ n 3 for every vertex x, and since d(x) 2m.
it would follow that
n(n - 3)
m~ ,
If there exisl two adjacent vertices of the cycle, Xi and Xi+ 1 (where the sum is
taken modulo n 1), which are also adjacent to x, then it is to insert x
between Xi and XI+ 1 so as to obtain a Hamiltonian C in the graph G.
Otherwise, for every i =0, .. "n - 2, if x were to XI it would follow
that x is not to XI';' 1 and hence ~ (n - I )/2. It can thus be seen that
n-2:::;;(n 1)/2, which contradicts the hypothesis that n:;;;:4. One can now con-
clude that G contains a Hamiltonian
(b) There does not exist a vertex x with d(xJ == n - 2. It follows that there are
at least two vertices YI and Y2 with == =n-1. If not, one would have
a unique vertex of n -1 and the other n -1 vertices of less than
or eq ual to n - 3, and hence
274 Problems in Combinatorics and Graph Theory
which is a contradiction.
Now suppress the vertices Yl and Y2 and the 2(n - 1) -1 = 2n - 3 edges
incident with them. The result is a subgraph G 2 with n - 2 vertices and m2
edges such that
2
n -3n+6 (n-3)
m2:;;' 2 -(2n-3)= 2 .
for every n:;;' 6. By adding two edges U l and U2 to the graph G2 between two pairs
of nonadjacent vertices one obtains a graph G 3 with m3=m2+2:;;,(n;3)+2
edges, which by the induction hypothesis must then contain a Hamiltonian
cycle C 3 .
If C 3 does not contain the edges U 1 and U2, then it has been shown that one
can insert vertices Yl and Y2 to obtain a Hamiltonian cycle in the graph G.
Let C 3 =[xo •... , XI- Xi+l •. ,·, X n -3, xoJ, Ul = Xi+l], and suppose that C3
does not contain the edge U 2 . Jt follows that there is a Hamiltonian cycle C for G:
A similar result holds when C 3 contains U2 and does not contain U l' Suppose
that C 3 contains UI and U2. and let U 1 =[Xi-l, xa
and U2=[Xj' Xj+1J with i~j.
In this case let
;"1
Yl Y2 Yl Y2
Fig. 11.4
walk L 1 has an endpoint in Xi, and that L2 does not have a vertex in common
with L!. If L I also has an endpoint in xi' the procedure terminates. Otherwise
consider an walk with endpoints at Xi and Xj' Traverse L3 from Xj
towards Xi' Suppose that one first encounters a vertex of the walk L 2 • In this case
a subwalk of by the subwalk of which terminates at Xj' This pro-
duces the two walks with the desired property. On the other if one first
meets a vertex of the C I' then instead consider a subwalk of L 3 con-
tained between xJ and the first intersection of with the cycle One again
obtains the two walks with the property. Suppose there exists a walk Li
with endpoints Xi and Xj or different walks with these endpoints. In both cases,
one obtains an elementary cycle which contains the vertices Xo and x'" and all
vertices adjacent with them II The C I and C 2 which contain
the vertices adjacent to x 0 and x'" respectively have at most one vertex in com-
mon (when i= j). It follows that the cycle which is formed has a length
than or equal to 2k+ 1. A. Proc. London Math. Soc., 2 (1952),
69-81.]
Xl Xm - 1
xo XO "";:"-_-.1.
Fig. 11.5
One can therefore apply the induction hypothesis to the subgraph G;,
which also contains fewer than n vertices. It follows that and hence the graph
G contains an elementary of length greater than or equal to k + 1. Erdos,
T. Gallai, Acta Math. Acad. Sci. Hung .. 10 (1959j,
11.11 Let C=(x l , •.. , X m , Xl) bc a elementary circuit of the graph
G. The num ber of vertices of C satisfies the m> n/2. I n l e t Q =
bea ofG,andletz", ... , (il<"'<
be the vertices of G such that (Zi s ' is an arc. It follows from maximality
of this path that z", ... , z" belong to the path Q. Since n/2, it follows
that k:;: n/2. and hence the circuit ZI," ., Zj" has length greater
than n/2, which implies that m> n/2.
Suppose that the circuit C is not Hamiltonian, and let D == be a
longest elementary path in the subgraph obtained from G by suppressing the
vertices Xl" .. , Xm of the circuit C. Since it follows that there exist
at least - r arcs of the form (u. where u ¢; D, since there are at most r
arcs of the form (u, with U ED.
All of the arcs of the form (u. Yo) with u ~ D have their initial vertex u in C.
Denote these vertices by Xi, • .••• Xi, with t:;: - rand i l < ... < i,. Similarly
one can find vertices Xh"'" Xl,. S:;: n/2- r such that (Y", Xl.) is an arc of the
graph G for p= 1, .... sand h < ... Suppose that the vertices Xl, and
are different. It follows that the path consisting of arcs of the
C must have (number equal to r+ 2. Otherwise one would
have a circuit of greater than that of C, which is seen to con-
tradict the hypothesis if the path (Xi" ••. , Xjq) is replaced by the path
(Xl., Yo, ... ,y" X jq) of length equal to r + 2.
Now consider an elementary path Dp consisting of arcs of the circuit C
which originates at x,,+ I Xm+ 1 =Xl) and length equal to r for
p = 1. .... t. The following two cases are
(a) Xjq=x,. and thus X j ,,. V(Dp), where V(Dp) denotes the set of
vertices of the path Dp. In fact r ~ n - m -1 ~ nl2 -1, and the length of the
circu it C is m> n12.
xl,f and hence Xjq" V(Dpl. since otherwise it would follow
that the has length equal to r + 1. But it has been shown that
C must have length at least equal to r+ 2.
278 Problems in Comblnatorics and Graph Theory
IpVl V(Dp)i ~ I + r.
since the union of the sets VW p ) contains distinct vertices XI, + 1.' •. !
XI, + l ' Also there exist two vertices. say and such that the
(Xi, + l' ... 'Xi,) formed from arcs of the circuit C a vertex x j .' In this
case it has been seen that the path DI of r does not contain the vertex
Xj" nor therefore the vertex XI,. It follows that the r vertices of the path DI
which are diITerent from XiI + I, .•• , to the union of the sets V(Dp)
for p = I, ... , t. This observation of (2).
Observe that and imply that s ~ m (t + r), so that
n
r+,i -r+r n-r.
Since the path D has r + 1 vertices in a sub graph with n - m vertices, it follows
that
r+l~n-m, or m~n r 1.
But (3) and (4) are contradictory, which (",)lTI1"IIpl,'" the prool that C is a Hamil-
toni an circuit.
It remains an open problem to show that under the condition of the present
problem and for n ~ 5 the graph G contains at least two Hamiltonian circuits
without common arcs. St. J. A. N ash-Williams, The Many Faeets of Graph
Theory. Lecture Notes in Mathematics, 110 (1969),237-243.]
11.12 If the tournament G contains a Hamiltonian then it is strongly
llI<O"L<O'U, since every two vertices of a circuit are by a Now sup-
pose that G is connected. Since for every two vertices x, y of G there
is a path from X to y and from y to x, it follows that G contains a circuit.
Let C == (y 1, ... ,Yk' YI) be an elementary circuit of G with a maximal number
of vertices. that C is not a Hamiltonian and let X be a vertex
which does not to C. Since the graph is
there exists an arc (y 1, If there also exists an arc , x, Yz, . , . ,
Yb yd is a circuit than which contradicts the It follows
that the arc between x and yz has the form (Yz. x). In fact there exist arcs (,Vj, x)
for i = 1, ... , k. Let A be the set of vertices X for which there is an arc (Yl' x).
One can conclude that x) is an arc for x E A and i= 1, ... , k. Since G is
strongly there exists an arc (x, z) with x E A and z (: A. It follows that
z (: C, and since G is and z (: A, one can find an arc of the form Yt),
This yields a circuit (x, z, YI" .. , Yk, x) which contains more vertices than C,
Solutiolls 279
and this contradicts the hypothesis. [Po Camion, C. R. Acad. Sci. Paris, 249
(1959),2151-2152.J
11.13 Let P= ,X2,'" and let S be a selection of k of P
which components. There are four possible cases:
(I) e Sand
(2) eSand j,xnJeS;
(3) !E Sand [x n - I ' xn] ~ S;
(4) !E Sand [xn - j , xn] e S.
It is clear that the number of sets S which satisfy (1) is to the number
of solutions of the system
al+"'+aJ=k,
hj + ". + n-k I,
where ai' hi are and aj, hi): 1; this number is j_1 Problem
In a similar manner, in case (2) one finds a system
+ ... +aj=k,
aj
hI + .,. + I=n-k
with j- 2 :) solutions for aj, 1. In case the system is
al + '" +aj
bj +'" +bj+l=n-k-L
(n-~- 2)(~=:) solutions, and for the last case the system is the same as
for the first case. Hence
as
11.14 Let H [XI' Xl'"'' be a fixed Hamiltonian walk of K n , and
denote its edges by ej Xi + 1] for 1 ~ i ~ n 1. Let Ai be the set of all H ami 1-
tonian walks of Kn which contain the edge ej of H for 1 ~ i ~ n - L Thus H(n, k)
is the number of Hamiltonian walks of which belong to precisely k sets Ai'
By using C Jordan's sieve formula (Problem we can show that
+ (-1)" 1 -k (n k 1),
where Pj(n, i) is Problem 1 since I (\.K
To obtain an for DH(n, k) in the case of a path DH of
K:, denote its arcs by at , .... Gn-l and let Ai be the set of all Hamiltonian
ofK: arc a, of D H for 1 ~ i:( n 1. The rest of the is similar to
that for the numbers H(n, k). since K:- i has (n- il! Hamiltonian and
each such path may be expanded to a Hamiltonian of K:in a unique way.
One also uses the identity
Problem l.S(a)].
11.15 Since G is connected. it has a spanning tree T. It is sufficient to prove
that 1'3 possesses the property of the statement, which says that G 3 is
Hamiltonian-connected. Now prove by induction on n that for any tree T with
n is Hamiltonian-connected. For n~4 it follows that the diameter
of T is at most three, and hence is the graph which is Hamiltonian-
connected. that all trees at most n -1 vertices have as their
cube a Hamiltonian-connected and let T be a tree with n vertices. If x
and yare two distinct vertices of T, one now considers two cases:
(a) x and yare joined in T an edge u == [x. y]. Denote by Tx and the
two subtrees obtained from T by deleting [x. y] such that contains x and
Ty contains y. By the induction hypothesis nand are Hamiltonian-con-
nected. Let x 1 be a vertex of which is adjacent to x, and let Yt be a vertex of
T), which is adjacent to y. If one of the trees or reduces to a single vertex,
then let Xl =x or Yl = Y respectively. The vertices x 1 and Yt are adjacent in T3
because dtxt, yd:(3 in 1'. Let be a Hamiltonian walk with endpoints x and
Solutions 281
CHAPTER 12
12.1 Suppose that the permutation p is a cycle oflenglh k. say p = [i 1." .• ikJ.
It follows that p(i I) i2, p2(i 1) = i3 , •• , • pk- 1 (i 1 J it, and pk(i d = 11' Analogously
one can conclude that pk(j) for every j, and hence k is the smallest number r
such that p' = e; every multiple ks of k has the property that pks = e,
Consider the cycles Pl' ... , PI in the representation of the permutation r
of disjoint P=PI .. ' PI' These cycles commute among
and hence it follows that pr Therefore if pi' = e one can
conclude that r is a common mUltiple of the of the of the permuta-
tion p.
12.2 Let p(n, k) denote the number of permutations of n elements which
have k Denote by e(n, k) the coefficient of Xk in the of [x]".
that is.
\ c(n. (1 )
, and thus it remains to show that cin, k)=p(n, k) for every nand k.
It follows from (1) that ern, 1) =(n l)! and p(n. 1) = (n -1 J!. since the number
of permutations which have only one cycle permutations) is equal to
n :/n. because every can be written in n distinct ways by as the first
Problems in Combinatorics and Graph Theory
clement each of its n elements. It will be shown that p(n, k) and c(n, k) satisfy
the same recurrence relation.
Let Xn={XI""'X n } and Xn+! Xnv d. permutation of X n + 1
with k cycles may contain the element Xn+ I alone in a cycle, with the remaining
n elements a permutation with k -1 Otherwise X n + I is contained
in a cycle together with other elements. The element X n + I can be inserted into a
cycle with p elements in p distinct ways. It follows that
p(n+ 1, k)=p(n, k-1)+npln, k). (2)
all permutations with k of the set X n+ 1 can be obtained, without
in two ways: one can add a new cycle consisting of X n + I to each
permutation of with k 1 cycles, or one can insert the element Xn + 1 in one
of n ways into each permutation of the set X n with k
It follows from (1) that
[X]n+ I = [x]n(x + n),
or
n
=c(n, k)- L IAil+ L IA/nAjl-· .. ·
i= 1 1 (: i<j~11
It follows that two conjugate permutations can be decomposed into the same
number of cycles having respectively the same lengths.
In order to prove the sufficiency of the condition, define the permutation 9
by first considering the decompositions into cycles of the perm utations sand t.
Then let g(t pq ) = Spq. The decompositions of sand t as products of disjoint
cycles both contain each of the numbers 1, ... , n exactly once. It follows that
9 E Sn and, as in the previous calculation, one can show that s == gig -1. This
observation completes the proof of the sufficiency of the condition.
The proof of Cauchy's formula can now be given. Let f be a permutation
which has ;'5 cycles of length S for 1 :(s:( k. Express f as a product of cycles
written in increasing order of length:
285
k
-'--,
. , . [* , , ,
J=
,I
,
distinct ways, On the other hand. each cycle of length i can be written in i distinct
taking as its first element each of the i elements. There are thus a total
., . kA' representations. Starting with all the h()'1 ' , .. ,I.d permutations
which contain ;'$ cycles with s elements, for s"" 1, .. "k one can write each
I, cycle of length s in s different ways. By also permuting the cycles which contain
I the same number of elements in all ways. it is easy to show that one
can obtain without repetitions all the n ~ permutations of the set {1, . , ' , It
G=~)(k-l)!ln-k)! -1)1.
(n-2)
k-l
distinct ways. There are (k -1)! circular of these k elements and
(n-k)! of the n-k
tions in which 1 and 2 to different
.-1 ( ~= 1 2) (k-l)l(n k)!
"-1
-2)1 (n-k)
-2)!---
lim (1
n- 00
+~+
2
... +~ -In n)=I'=0.5772. ..
n
constant).
12.11 Since p2:::::: e, it follows that the permutation p has all its of S
length lor 2. Thus the set of permutations PES. such that == e can be written
in the form Anu where is the set of permutations p E with =e which
satisfy p(n) = n, and Bn consists of those permutations p E which p2 = e and I
in which the element n is contained in a of length 2 in p. It follows that m
+ I +(n 2' In if p(n)=n, the other n 1 elements
ry Solutions 287
ts can turn a cycle of length 2 with each of the other n -1 elements; the remaining
elements form a permutation r E 2 for which r2 == e.
In order to prove (b) let i denote the number of of 1 (fixed
points) of the permutation p, and let j denote the number of cycles of length 2.
It follows that
d Pn =Perm(li2i ; n)=n! '2)i!j
where the summation is taken over all of n in the form i + 2j = n
with i, O. The exponential function of the number Pn is
~)
"" In
L Pn(m) k .
n=O
The num ber 11- c(p) of transpositions can be attained, since each cycle of length
mcan be written as a product of m - 1 transpositions;
.. , [im-!, im].
288 Problems in Combinalorics and Graph Theory
Fig. 12.1
!i associated with its edges. Now permute the transpositions t 1,1 2 "", In-I
in (n -1) ~ distinct ways to obtain distinct products which generate circular
I
i
generate different products of transpositions. Thus (1'1 -1) !AU):::: (n -l)~nn- 2,
: and hence A(j')=nn-2, that is, the number of ways in which a circular permuta-
, tion on n elements can be written as a product of n - 1
to the number of labeled trees with n vertices. [J.
Mat. K Wato Int. Ked, 4 (1959), 63- 71; O.
is equal
Magyar Tud. Akad.
Berl. Mach. G..
17 (191 64-·67.]
12.14 In order to prove (a), observe that if P1 = al G2 ... an is a permutation
"I, in Sn and P2 == anan-I... a!, then
+ 1, k)= (1)
where
Ai = {pip E Sn+ l ' l(p)= k, and p(i)=n+ I},
If the element n + 1 is found at position i, then it is in inversion with all the
n + 1- i elements which are found at i + 1, ' , , , n + 1, Thus if p E Ail
then by the element n + 1 from position i one obtains a permutation
Pi E Sn such that I( Pi) =k - n + i-I, The correspondence which is thereby
defined associates with a permutation pEa permutation Pi E Since it is a
bijection between the set Ai and the set of permutations in with k - n + i-1
inversions, it follows that
iAii=ptn,k n+i-l),
Part (d) now follows by using (1).
in S. has at most (;) inversions. The maximum number
of inversions is attained only for the permutation n, n-l, ... ,LIt follows that
p(n, (~)}= 1 and 0=0 for i> The permutation with no inversions
is the identity permutation 1, 2, ' , ., n, and hence p(n, 0) 1. These initial values
together with recurrence relation (d) a determination of the
matrix of numbers p(n, k).
In order to prove (c), observe that if k<n, then it follows from that
p(n,k)=p(n l,k)+p(n l,k-l)+ ," + -1,0),
p(n,k-l)=p{n-l,k-ll+p(n-l,k-2)+ '" +p(n-l,O),
and hence k)=p(n-l, + k-
For let
(1+x)(1+x+ "'(1+x+ ... +xn-I)
L c(n+ 1, +x)"'(l+x"
k;,O
= (.L
1;;.0
c(n, i)x l ) (1 + x + ... + x n ),
Since the numbers c(n, k) and p(n, k) have the same initial values and satisfy
the same recurrence relation, which uniquely determine them, it follows that
c(n, k) k) for every nand k. This completes the proof of equation (e).
12.15 Write a given permutation pEas a of disjoint
Include cycles of length 1 (fixed points of p), and write the cycles so that their
largest elements are written first. These first elements are to occur in increasmg
order from cycle to cycle. For example, the permutation
1 2 3 4 5 6 7)
p= ( 3 4 5 2 1 7 6 eS 7
will be written in the form
p=[4, 6].
In this way the last cycle will always with the element n. The sequence
of numbers obtained in this way can be considered to be another permutation
f(p). In the previous example one has
f(p} 2 3 4 5 6 7) .
25137 6
The permutation p can be uniquely reconstructed from the permutation
f(p) E . The last of p begins with the number n. The next to last begins
with the number which is not contained in the last cycle, and so on.
If the permutation p contains k then the permutation PI = f(p) will
contain exactly k elementsj for which PIU»PI(i) for every i<j, and these will
be the first elements in these k of the p.
j The permutation p with k cycles can be uniquely reconstructed from the
permutation Pl = f(p) which contains k elements j for which PIUl> pdi) for
every i It follows that! is a between the set of permutations P E Sn
with k cycles and the set of permutations PI E Sn which have the property stated
in the problem.
The number of permutations P E Sn with k is equal to Is(n, k)/ by
Problem 12.2. This observation the [A. Colloquium
Aarhus (1 104-11
12.16 It is clear that d(f, g) == 0 implies !U)=g(i) for every i = 1, .... n, and
hence f == 9 and d(j,g);J;: 0 for every j,g E Sn. Furthermore dig,}'),
and for j; g, hE one has
\f(i) - gU)1 ~ II(O - hU)1 + Ih(i) - g(OI
for every i == 1, ... , n, which implies that
max \f(i) - g(i)\
i
~max {1!(i)-h(OI+!h(i)-g(i)\}
,
If(O- 17(01 +max
I
292 Problems in Combinatorics lind Graph Theory
or
d(J, q)<d(J, h)+d(h, g).
Hence d(j', g) is a metric or distance on Sn' If p E Sn and d(e n , p)< 1, where e.
is the identity permutation in SM' then either p(n) = n or p(n)= n -1 and p(n -1)
== n, since otherwise one would have p);;?; 2.
Let q, r denote the restrictions of p to the first n-J and n- 2 elements of
{1,. , . , n} respectively, In the first case one can conclude that d(e n _! , 1, and
in the second case 2, r)::;;;; 1. Thus F(n, l)==F{n-l, l)+F(n 2,1), which,
together with the values 1)= 1 and F(2, 1)=2, the equation 1)=
for every n;;?; 1.
12.17 Define the permanent of a square matrix A 1 .. ",. by the
000
o 0
o
1 o
o
an=per
o 1 1
o 0 1
o 000
where
Solutions 293
1
o 1
o
o o I 1
o 1
bn=per
o o
By expand ing bn and en on their first column s one can see that
where
o o
I1
1
1
o
.\ dn=per
~
,1,
!
o
1, and all the
) matrice s have order n. Expans ion of dn
; where
1
o 1
o
o 1
en=per
recurrence relations determine an, bn , Cn, dn, en for every n with initial values
forn=2:a2 =C2 =2, e2=1. Let
0
0 0
1}
A= 1 0 1
0 0 0
1 0 0 0
The given recurrence relations can be PYr,rp<:<prt in matrix form as follows:
2= ... =A n IV! =Anv o'
Here
which coincide with the directly obtained values for the case n = 2. Thus an
can be as the scalar of the vector of the first row
of the matrix An and the vector Vo, that is, (A"}ll' [D. H. Lehmer, in: Comb.
Theory Appl., Call. Math. Soc. J. 4, North-Holland, 1970,
12.18 The desired number is the permanent of the matrix
0 0 0 0
1 0 0 0
0
)n- p
.. . I .. I \
1 JP
"---v----'
P n-p
By the permanent on the first row of the matrix one can obtain the
recurrence relation
A(n, p) -1, p).
Solutions 295
2An = (n-1)
0 AoAn- + ("-1)
1 1
+(n-1)
n-1 where Ao=A~ = 1.
Notice that each up-down permutation with the number 1 in position 2i + 1
and the number" in position 2j is counted exactly twice: once among the up-
down permutations with the number 1 in position 2i + 1 and once among the
up-down permutations with the number n in position If ak=Aklk: for 0,
then this recurrence relation can be written
X 11:)
tan ( -2+- :::::
l+sinx
secx+tanx.
4 cos x
[D. Andre, C. R. Acad. Sci. Paris, 88 (1879), 965-967.]
There are many extensions of this classical result [L. Carlitz, Discrete Mathe-
matics, 4 (1973),273-286; L. Carlitz, R. Scoville, Duke Mathemmical J., 39(4)
(1972), . etc.]
296 Problems in Combinatorlcs and Graph Theory
Thus p(1 min(p(3), p(4), p(5), p(6), .. .). One can show analogously that
p(i}< 2 pU).
Let the number be denoted g(n) for every n ~ 1. It can be seen that
g(1)= land = 2, since both of the 1,2 and 2, 1 satisfy the
conditions. The set of permutations of the set which are 2-ordered
and 3-ordered can be written in the form Anv where An is the set of perm uta-
tions of the set (1, ... , n} which are 2-ordered and 3-ordered and for which
p(1) 1. The set Bn consists of those permutations of the set {1, ... , n} which are
2-ordered and 3-ordered and for which 2 and p(2) = 1.
In if the permutation p is 2-ordered and 3-ordered and p(1) = 2, then
p(2l = 1, since otherwise the number 1 would appear in one of the
3,4, .. ,n; this contradicts the inequality
< min{p(3), p(4)•. .. , pen)). (1)
In the same way one can show that p(l)" 2, since if p(l) ~ 3 at least one of the
numbers 1 or 2 appears in one of the positions 3, 4, ... , n, and this again con-
t radicts (1).
It is clear that the sets and are disjoint, and thus g(n}= IAnl + IBnl ==
g(n-l) + g(n - 2), since An has the same cardinality as the set of permutations
of the set ... , n} which are 2-ordered and 3-ordered. The set Bn has the same
cardinality as the set of of the set p, ... , n} which are 2-ordered
and 3-ordered,
Since g(1) = and g(2) = and the numbers g(n) the same recur-
rence relation of the Fibonacci numbers, it follows that g(n) = for every n ~ 1.
[H. B. Mann, Econometrica, 13 (1945),
It follows from the proof of the
12.21 problem that
u;<mink'-'l UIH and hence the sequence can contain at most two consecutive
terms which are equal.
Let q) denote the number of sequences where U 1 , ••• , uq
belong to a set of p distinct numbers which satisfy the
conditions Uj<u1+2(1:r(i:r(n-2) and Uj<ui+3(1"i:r(n 3). It follows that
!(n)=F(n, n).
Suppose that Ul,' .. ,u q E {I, ... , p}. If the sequence Ul, •.• , U q with
I and u2~2, then the number of such sequences is to F(p-l,q-l),
while if U2 = 1, the number of these sequences is to F(p -1, q - since
fJV'''''''''''' 3, 4, ... , q there must be numbers from the set .. , p} which
conditions. If the sequence with Ul =k, then U 1 =k
and u2=1, or ul=k and U2 ... ,or u1=k and u2=k, or u2~k+1. Thus
in this case the number of the sequences (u I, •.. , u q ) is equal to F(p - k, q 1) +
kF(p - k, q one can conclude that the numbers F(p, q) satisfy the
recurrence relation
Solutions 297
for every p, q ~ 3.
It is also true that F(r, 1) rand F(r,2) r2, F(l. r) =0 for r~ 3, F(r,2r) 1
when the unique sequence 112233··· rr is obtained, and 2r + 1) =
F(r, 2r+2)= ... =0. One can also conclude that F(2.3) 2, since in this case
there exist sequences 112 and 122 which satisfy the conditions. These
initial conditions yield the following table of numbers j) in which the first
two rows and the first two columns are determined from the initial conditions:
.. .....o'" ............
2 4 l
.................... 2 0 0
- - .. o
3 9' 3 0
4 16 24 31 22 13 4
5 25 50 85 88 75 42 19
I
numbers A(n, k) row row, taking into account that A(n. k)=O for k>n. For
11 ~
6 the table is given below:
n 2 3 4 5 6
1 0 0 0 0 0
2 1 0 0 0 0
3 1 4 0 0 0
4 11 11 1 0 0
5 26 66 26 0
6 57 302 302 57
k) (m+: (1)
where the sign @ between ap(l) and ap(i + I) is ~ if the permutation p does not
have a fall at p(i) and is < otherwise. But the number of sequences satisfying (2)
is equal to the number of sequences b l b2 ... bn composed of integers
which satisfy the following conditions:
(3)
Let b l = let b2 = ap (2) if p has a fall at pi 1) and b2 Qp(2) + 1 otherwise:
if by =ap(y) +s, let br+ I ap(r+ I) + s if p has a fall at per) and b Y " 1 =ap(r+ 1) + S+ 1
SoIUliol1s 299
= t
k= 1
A(n,n k+l)(m-l+(n-k+l l
n
)= " A(n,k)(m+k-l).
n
which is (l). Since (1) holds for every m~ 1, it follows that polynomial identity
also because the polynomials on both sides take equal values for an
infinite number of values of the variable x.
(d) may be deduced from (cl as follows. For x = lone finds that
A(n, 1) = 1. For x =2 it can be seen that A(n, n-l)= A(n, 2) =2" n -1, which
is in accordance with (d). Suppose that (d) is true for A(n. where 1~ s~ k-1.
It will be shown that this relation is also true for s = k.
Let x = k in (cl to obtain
or
t
5=0
(_1)5+\ (n~l)(n+s)
P s n
0 (4)
for any p ~ 1. By using Newton's generalized binomial formula one can conclude
that
300 Problems in Combinatorics and Graph Theory
1 (1-
X{I+C71)x+(n;2) + .. +(n;s)xs+ .. }
+1)) + 1)=a.
Since 2m + 1 is the unique fixed point of a, it follows that p(2m + 1)=2m + 1
and p has no other fixed points. The restriction of p to the set {1, ... , 4m + I} "
+ 1 ) has properties analogous to those of p, and hence one can conclude that
N(4m + 1)= N(4m)
CHAPTER 13
13.1 Consider the axis of symmetry xx' of the rectangle which is parallel
to the sides AC and BD (Figure 13.2). The desired number of configurations
is equal to
N l +N 2 ,
where N 1 represents the number of configurations which coincide with their
with respect to a reflection in xx'. The number N z represents one-half
the number of configurations which are not self-corresponding under a reflec-
IX
I
I
I
A B
1 12 3
@
~ ®
I
I
I
I
4 15 6
® ®~
I
I
I
I
7 18 9
@ ®@
c ,I D
I
I
I
I
I x'
Fig. 13.2
302 Problems in Combinatorics and Graph Theory
+ 11 +63)=287.
the side AB contains n
so that there are a total of mn perforations. Vrr1,('".,,'l1m
it follows that
1)12J-l_1.
In fact, since is a of G, one can consider the set GIG k whose classes
are sets of the form 9 E Gk }, which form a of the set G and
whose cardinality is equal to =IGI/IGkl. It will be shown that this number
is equal to lOki by a bijection from Ok to Observe that
Solutions 303
1
g(i)=h(i)=k with g, heG implies that h(i)=k, that is, I1g- EG. or
he Gky.
F or every element i e Ok there exists a permutation g/ E G such that g/(I) == k.
since i"" k(G). The bijection J: Ok-+G/Gk is defined as follows: J(i) == Gkg l e ,
where the permutation gi satisfies g,(i) = k. The mapping J is well defined, since
ifthere exist two permutations h, 9 e G such that g(O =h(i) = k, then it has been
shown that hE Gkg, that is. the class of h coincides with the class of g, so Gkh ==
where 0 1 , •••• Or denote the orbits of the group G. It has been seen that if
j, kEO, then IG j =IGkl=IGI/IOd, and hence
IGI
10d 10,(qIGI.
which that q=(l/IGj)
13.3 Denote by R 1 , R 2 , ••• , Rn the clockwise rotations of the poly-
gon about its center through an 2n/n, 4n/n •... ,2n respectively. These
rotations form a group with respect to the composition of rotations; R. is the
identity element of the group. Two polygons PI and P 2 with k vertices are con-
sidered to be identical if there exists a rotation R", with 2nm;n such that
where all the have the same length d=n/(m, n). [t follows that d I k or
n I kim, nl, and the number of cycles in (2) is equal to kid = k(m, n)/n.
Consider the permutation induced by on the set X of vertices of the poly-
gon P and written as a product of It will be shown that every
(m, n) consecutive num bers modulo n of the set {O, I, ... , n - I} have the property
that no of them belong to the same cycle of the induced by
on X. For suppose the contrary. One can then show that there exist two
°
numbers ~ a, b ~ n -1 such that 0< bl ~ (m, n) - 1 and such that they are
found on the same cycle as e.lt follows that c+pm=a(modn) and c+qm=b
(mod n), and hence
a=c+ pm rn and b=c+ qm-sn,
where r, s ~ 0 are Thus one has
la-bl=l(p-q)m+(s- ~(m,n)
because a::f:: b, and Ihis contradicts the inequality la - bl ~ (m, n) - 1.
It follows that for every (m, n) consecutive numbers (modulo n) of the set
{O, ... , n -I} there are kim, n)/n distinct numbers which belong to
different cycles of the permutation (2) induced by R", on X. For otherwise the
number of elements in X would be smaller than {k(m, n)/n} {n/(m. n)} = k, which
is a contradiction. In order to find the number of polygons P with k vertices
which are invariant under Rm, one must find the number of permutations of
form (2) where d == n/(m, n) which contain k(m, n)ln cycles. The permutation
is uniquely determined if one chooses an element in each the other el-
ements of the cycles are obtained repeated addition of m (modulo n) to the
number chosen. Thus if n • k(m, then the number )'1 (Rm) is equal to the
number of ways of k(m, elements from among lm, n)
that is, it is equal to
(m, n) \
fen, k, m) = ( k(: n) ).
If n is not a divisor of k(m, n) then )'1 (RmJ O. In view of (1) the desired number
has the form
n
L fen, k, m). (4)
n m 1
n I k(m •• '
Solutions 305
I
By using the notation d = n!(m, n), the condition n k(m, n) becomes d • k, and
I
since din, it follows that d (n. k),
Now calculate the number of terms in summation (4) which have the same
value. Assume that d is fixed and m takes values between 1 and n such that
k(m, n) is divisible by n. If e = ml(m, n) then e ~ n/(m, n) = d and (e, d) 1. The
I I
condition n k(m, n) is satisfied if d (n, k). Thus for fixed d, fen, k, m) takes the
same value for all numbers e which are prime to and less than d. The number e
uniquely determines m m = e(m, n) = en/d. It follows that the number of
convex polygons with k vertices which cannot be obtained from one another
by a rotation is to
({J(d)
nd
where ({J(d) is Euler's function which the number of positive
smaller than and prime to d (Problem 2.4).
13.4 Proceeding as in the preceding
of colorings with k colors of the vertices of the polygon with n vertices
which cannot be obtained from one another by a rotation is equal to
1 "
- ;'1 (Rm),
n m=
where ;.!(R m ) represents the number of invariant under the rotation
Rm of angle 211m/n. One can show analogously that if a k-coloring is invariant
under a rotation , then this rotation defines a permutation of the n vertices
which can be in the following form as a product of
[11, ... , .... ,in .. · ".. ,in, (1)
where d =n/(m, n) and p=(m, nl. Each cycle in (1) is formed from numbers of the
set ... n - !} to vertices of the regular polygon which are
colored with the same color.
Also, every (m, nl consecutive numbers modulo n in the set {O, ...• n -1 }
belong to different cycles in (1). Thus the number of invariant under
Rm is equal to the number of colorings with k colors of(m, nl consecutive numbers
modulo n from {O, ... , n -1;. that is. to the number of functions
g:{1, ... ,tm.n)} 1, ....
It follows that AdRm)=k(m.n,. Finally the desired number of k-colorings is
equal to
-1 I" ki".m,=
1
I ({J(d)k"'d.
nm=! ndln
13.5 One first obtains a formula for gn' If in the representation of the
permutation p Ii:' Sn as a of disjoint cycles there are dk cycles of
k for k 1, ... , n, then p said to be a permutation of 1dt2d,. . nd n• where
d! + + ... +ndn=n.
306 Problems In Combinatorics and Graph Theory
Suppose that the graph G U) has vertex set X = {1. ... , n}.
permutation pin Sn can be considered to be a permutation of the set of
G defining p(G)=(X,p(U)), where p(U)={[p(i), p(J)]1 E One first
determines the number d(p) of graphs G with n vertices such that p(G)=G,
that is, the number of fixed points of p. Let G be a graph such that p(G)=G,
and suppose that p is of 14 '24 , .•. nd •• The graph G can be partitioned into
subgraphs G z, ... such that two vertices i and j belong to the same sub-
graph if and only if i and j belong to the same cycle in p. Among the
G; there are dk which contain exactly k vertices. for k=l, ... , n.
that 1,2, ... , k} and 1, k+ 2.. ,., k+ I} are sets of vertices for
two of these Gland One can assume that the permutation p
contains the cycles [1,.,., kJ and + 1, ... , k+ IJ. The [k/2] possible
between the vertex 1 and the vertices 2, 3, . , . , [k/2] + 1 in the graph G 1 uniquely
determine the existence of the other of G t in view of the condition p(G) =G.
Similarly the (k, 1) possible between the vertex 1 and the vertices
k + i for i = I, ... , I) uniquely determine the existence or nonexistence of the
other which vertices of and vertices of G 2 in view of the con-
dition p(G)=G. Thus if p(G}=G, one can select in an arbitrary manner the
existence of a number of edges equal to
Thus 2Gd , from which (a) follows, since the number of permutations of
type "'n4 "is ton!/Ndand =n!.
In order to prove (b) one can proceed Two vertices x, y can be
nonadjacent, joined by the arc (x. y) or the arc x), or joined by both the
arc (x, y) and the arc (y, There are thus four possible cases. Also, in the
expression for the summation even !kdk must be replaced by
Lk even t(k - 2)dk • For two vertices u, v of an even cycle of length k, separated by
the maximal distance k12, there are only two either u and v are
or u and v by both arcs (u, v) and (v, u), since the existence
arc would contradict the in variance of the G under the per-
mutation p. Thus by applying Burnside's formula one obtains the numerator of
d., which is expressed by
existence of both arcs (iI' il<+ d and (i k + I' id contradicts the definition of a
tournament, and hence in this case d(p)=O. Otherwise, if p(G) G, one can
choose an direction only for
arcs; the orientation of the remaining arcs is uniquely determined by the orienta-
tion of these arcs, and the numbers d l , d 3, ••• satisfy (3).
The values of the numbers 9n' dn, and In up to n 7 are given in the following
table:
n 9n d. tn
1 1 1 1
2 2 3 1
3 4 16 2
4 11 218 4
5 34 9,608 12
6 156 1,540,944 56
7 1044 882,033,440 456
(1) Reflexivity: f"'- f because f"" fe, where e is the identity permuta-
tion and e E G.
(2) Symmetry: fl"'" implies that f2.... ,since there exists 9 e G
such that and thus f2g- 1 == fl and g-I E G.
(3) Transitivity: fl"" and imply that fl '"
fig f2 and f2h= f3 with g, he G implies (flg)h=
where g17 E G, since G is a group.
Let F denote the set of the m" f:X ..... A. For every permutation
9 of X the == is an injection of F into F. In fact, =1=
the existence of an object ieX such that ft(i)=1= f2(i). Let It
follows that flg(j)= =1= f2(0= f2g(j) and thus flg=l= The mapping
g: F..... F is injective. and since F is finite, this mapping must be surjective and
hence a bijection. In other words g E S, where S denotes the set of permutations
of the set F.
The Ip(g) defines a mapping
Ip:G ..... S.
This mapping is injective, since if gl there exists k e X such that gl (k):i:: 92(k).
It will be shown that the functions iii :F ..... F and ;F ..... F are different,
that is, there exists f E F such that fgl =F f92, in view of the of g.
308 Problems in Comblnatorics and Graph Theory
If m;;r. 2 there is a I which uses different colors for the distinct elements
and since F is the set of all functions I: X ...... A.
Thus =1= or jgI(k)"I" Ig2(k), from which it follows that j9I
and are different and I/> is an injection. If m = 1, then I/> no longer is
However, in this case the number of colorings is equal to L so aJl the
have the same color and P(G;I, ... ,1)=(l/iGI)Lg<c1=1, which completes
the proof of the property.
Let G={9IgeG}. Since I/> is an injection, it follows that there exists a
bijection I/>I :G ..... GcS defined I/>I(g)=I/>(g) for every 9 E and hence
IGi=IGI·
The set Gis a subgroup of the group S of permutations of the set F of colorings,
since iiI' 92 E G implies that 9192 e G. In fact -lg2U)=gl(Jg2)= j(g2gd=
g2gdf), and thus the product of two elements of say 91 and , is an element
of G which to the g2g1 e G. Since S is a finite group, it
of S.
to the given definition, two 11 and are if
there exists 9 e G such that f1 9 or gUt> "" 12' and thus belong 10 the
same orbit of the group This implies that the number of equivalence classes
is to the num ber of orbits of the group By Burnside's theorem (Problem
this number is equal to
a d
r-
I 4
f g
Fig. 13.3
By using theorem from Problem 13.6 one finds that the number of
ways of coloring the faces of a cube with m colors is equal to
'm,.... +3m4 + +8m 2 ).
13.8 In the solution to Problem 13.3 it was shown that the rotation R",
of 2nm/n of a about its center can be written as a
of (n, m) cycles of length d = n/(m, n):
where the numbers it, ... , jl , . . . • . ..• , td are the numbers 0, 1, ... ,
t 1 ••..
n 1 which represent the vertices of the polygon. Thus the cycle index
polynomial is equal to
I 1" x'(n m)
- 1 X~'d 1 1=-
1
rp(d)X~/d,
n ",=1 .'(n.m) --n 11 m"" n
(m,n)=nld
because if e=m/(m, n) then the conditions m~n and (m, n) n/d are satisfied if
and only if e~d and (e, d)= 1. An application of theorem shows that
the number of with k colors of the vertices of a regular polygon with
n vertices which are not obtained from one another a rotation is equal to
(l/n)
In if G is a finite group with p one can consider its repre-
sentation as a group of by for every a EO G
xa
for every x E G. Thus a becomes a permutation of the set G which can be de-
composed into p/k cycles of k, where k is the order of the element a in C.
In fact xa m = x if and only if a'" = 1. as one finds that the
cycle index polynomial of the group G is equal to
1
IGI p
where p=ici and rp(C, d) is the number of elements of order d in the group C.
Observe that in Problem 13.1 the group of rotations about the axis xx' which :
leave invariant the ABeD is formed from permutations
G={e [IJ[2J[3J[4J[5][6][7][8J[9] and ,3][4, [7,9J}
in the notation of Figure 13.1. The cycle index vnr,m I A I is thus
k
n 1
(k.i) JAkIn
[X k;(k. i) •
1
iThe upper limit in this may be taken to be r because r is the least
'common of the cycle of f. H. Redfield, Amer. J. Math.,
:'9 (1927),
13,10 The automorphism group of this is G={[1][2][3J[4],
[2J[4][1. 3J, [1J[3][2, 4], [1, 3][2, 4]}, and hence
13.11 Denote this number by MG(m). It is clear that MGfm)= P(m, 1)+
P(m, 2)+ P(m, since any multigraph with three unlabeled vertices and m
corresponds to a partition of m into at most three parts. One finds that
1, P(m, 2) = [mI2J, and P(m, 3) = 3, 1) + P(m - 3, 2) + P(m - 3,
I1G(m-3) (see Problem 5.2). Thus MG(m)=l +[m/2J+MG(m-3} for any
~~3, where MG(O) = 1.
The formula for MG(m) now follows by induction on m, since MG(O) =
I1G(I)=I, MG(2) 2, MG(3)=3. MGt4) and MG(5)=5.
Suppose that the expression for MGtm) is valid for m<n-l. The fact that it
lIso holds for m = n follows from consideration of the cases n == 0,1, ... , 5 tmod 6).
13.12 Letw(a)=al+a2+ ... +anfor any Boolean vector Then
I~ w(a) ~ n.and if I is symmetric, it follows that I(a) = I(b) for any a, b E Bn
uch that w(a)=w(b). Thus it is sufficient to define any symmetric Boolean
unction of n variables for Vo == to, ... ,0), v! 0, ... ,0), V2 (1, 1,0, ... ,0), ... ,
,=(1,1, .... 1) to either be 0 or I, where w(v[) = i for any 0< i< n. This can be
lone in 2n '" 1 ways.
:HAPTER 14
14,\ Consider the graph of Figure 14.1 in which all edges have length equal
o I. Suppose that the property does not hold, and assume that A is colored a,
! is colored b. D is colored c, and thus that F is colored a. One can conclude
nalogously that G is colored a and hence there are two points F and G at a
istance 1 which have the same color; this contradicts the hypothesis.
.Hl Problems in Combinalorics and Graph Theory
B c
F G
Fig. 14. t
14.2 Suppose that the two colors are red and blue. For every coloring there
will be two M and N with the same color, say red. If the midpoint P of
the segment M N is red, then one has obtained three equidistant collinear points
with the same color. Otherwise, suppose that P is blue. If the point N 1 which is
to N with respect to M is red, then the points N 1, M, and N are red. il
If N 1 is blue, consider the point M 1 which is symmetric to M with respect to N.
If M 1 is red, then the desired points are M. N, M l' Otherwise M 1 is blue and is
thus N 1, P, and M 1 are three equidistant collinear blue points. fe
Thus in any coloring of the of the plane with red and blue there will Cl
exist three collinear points of the same color, say red. Denote these
01
by A, B, C. Construct the equilateral triangle AFC so that E, D are the Sl
of its 14.2). If F is then AFC is the desired
Otherwise F is blue. If D and E are blue, then the equilateral triangle DEF has
blue vertices and the problem is solved. Otherwise at least one of the points D
and E is red. Suppose that D is red and thus the triangle ABD has the desired an
property. In order to show that there exists a 2-coloring of the points of the
in which no triangle of side I is monochromatic, consider
the plane with an x-y coordinate system. The lines y tl
(J3/2)k with k an integer partition the plane into a network of parallel bands. 0
Suppose that the band bounded by the lines (J3/2)k and (J3/2)(k + Ucontains:
all the points of the line (J3/2)k and none of the of the line (.J3/2)(k + 1). ~re
10.
F
lid(
EFI
I
side
:ria:
orig
A B C irial
Fig. 14.2 me
leory j Solutions 313
E
Fig. 14.)
lere Now color the points of the plane red and blue so that all the points of a
) of band have the same color but each two neighboring bands have different colors.
lnts Sihce each band has width J3/2, which is equal to the altitude of an equilateral
his triangie of side 1, it follows that every equilateral triangle of side 1 has vertices
·ed. in two bands and thus does not have all of its vertices the same color.
. N. 14.3 If all the of the plane have the same color, then the property
md is evident. Otherwise there will exist two points A. B at a distance 2, with dif-
ferent colors. In fact every two points of the plane which have different colors
Nill can be joined by a polygonal line which has all of its segments of length 2, and
e~e one of these segments must have of different colors. In 14.3
tl e suppose that
~ e.
1as AC=CB=AD=DC-AE=EC=1
'D
:ed and
~he DE=BD=BE=J3.
jer Let A be colored red and B colored blue. One can assume that C is red, for
)." otherwise one could make the argument below for the which are sym-
?s. metric to D and E with respect to the perpendicular bisector of AB. If D or E is
I~S then one obtains an equilateral monochromatic triangle of side 1. If D and E
). are blue, then the triangle BED has all of its vertices blue and side length equal
IOJ3.
14.4 Consider an equilateral monochromatic red) triangle ABC of
side a. In 14.4, BDE and CH F are equilateral triangles of side b,
EFG is congruent to ABC, and BE and CF are perpendicular to AB.
Problem 14.3 there exists a monochromatic equilateral triangle ABC of
lide a E {I, .j3}. If a= 1, then Jet b=,J3, and if a=J3, then let b= l. The
Iriangles ABE, DBC, GFC, EFH, ACH, and DEG are all congruent to the
T. Now A, B, and C are ali red. If there are no monochromatic
congruent to T, then by considering DBC, and ACH.
me can see that E, D, and H must be blue. DEG forces G to be red.
314 Problems in Combinalorics and Graph Theory
A H
b
b
b
""=:"' _ _-----i F
D
G
Fig. 14.4
Ie
\\
Fig. 14.5 Ie
ry 315
9 6 8 7
Fig. 14.6
14.7 1t will be shown that every sequence of numbers (11, a2' ... ,amn ~ 1 con·
lains either an increasing subsequence with at least m + j terms or a decreasing
with at least n + 1 terms.
Suppose that every sequence has at most m terms and every
sequence has at most n terms.
For each term al define the numbers Ci and d i as the numbers of terms in a
iongest increasing or decreasing sequence which begins with ai' The mapping
which associates with every term al the ordered pair (c" dJ is injective. In fact,
iet and al ~ aj. In this case one has Ci:;;: 1 and hence cd Cj or , dd r
316 Problems in Combinatorics and Graph Theory 50
If a,;;;aj then dj;;;d j + 1, that is, and hence (cl' di):f:(cj, dJ). But
implies that the number of elements of the sequence is less than or 1,
equal to the number of elements in the Cartesian product {L 2.... ,m} x C
{1, 2, ... , n}, that is, mn + 1 ~mn, which is a contradiction. L
C
14.8 Let t(x) denote the maximum length of a sequence x = at < a2 < ...
el
which satisfies f(ad~ f(a2}~ .... Since fO)= 1, it follows that max t(x)==t(l).
It must be shown that 1(1);;;n. Use induction on n. For n=l the is
since t(1) == 1. Suppose that the property holds for every number rr
m~ n 1, and show that /(1);;; n. Under these conditions one can first show that
if ;;;1(1)-k then f(x)~2k for k ... , t(l)-1.
Suppose on the other hand that f(xj > 2k and t(xl;;; t(1) - k. It follows from
the fact that 1 ~f(i)~i that X>2k and k<n 1.
Let x=al < ... <a,ix ! and f(al)~ ... ~f(at(x)' the induction hypothesis
there exists a sequence 1 ~ b l < ... < t ~ 2k with f(b l ) ~ ••• ~ f(b k + I), since
k + 1 ~ n - 1. One can thus conclude that
Since the number of elements in the domain of f is equal to 2"-1, one can con-
clude that
f!l)-I
2"-1 = L It(x) t(1)-k}l~
k=O
---
M
/ Q
/
/
/
/
/
/
/
Fig. 14.7
C
""\
P
""
\
N
M Q
"-
Fig. 14.8
Jill Problrm~ in Combinatoric, and Graph Theory
Sl
pr
C,
ro
B
~----------------------~~
.....
"-
"-
"- .....
......
Fig. 14.9
'Y jSo lutiolls 319
e ~ It has been conjectured that for every choice of 2m 2 + 1 points in the plane
_ ,such that no three are collinear there are m which are the vertices of a convex
. polygon. The has been verified through m = 5. It has also been shown
,that there is a choice of 2m - 2 points in the with no three collinear such
,that no m points form a convex polygon,
14.10 Suppose that there exists a coloring of the of the graph G
.which does not induce a monochromatic and which uses different
colors for the of the Thus there exist two of
,which have different colors, say [AI' A 2 ] red and [A 2 , A3J blue.
Consider an arbitrary vertex B j of the cycle Cm. One can assume, for example,
that the [B), A 2 ] is red. Since there is no monochromatic triangle, it follows
that the [AI, Bj ] is blue.
Let + 1 be a by an edge to B) in the eycle (taking Brn -1 = B 1)'
, If the [B]4- I, is red, then the edge [B)+ I ' Ad is blue, and no
matter whether the edge [B}! B]+ 1] is colored red or blue, one of the triangles
B]B j + I A I or BjB,+ 1 A2 is monochromatic (see 14.10), which contradicts
the hypothesis.
Thus if the [B l , is colored red, then the [B)Tl' is colored
blue. Analogously if the edge [B), is colored blue, then one finds that the
edge l ' A 2 ] must be colored red by A I with A3 in the
argument. Traverse the cycle em, starting from the vertex with the
[B}, colored red and passing through the neigh boring vertices. One will
meet Bi , which implies that A2 J must be colored blue, since m is odd.
This yields a contradiction which establishes that all the of the cycle C.
have the same color. II can be shown since n is also odd, that all
the edges of the cycle must also be colored with the same color.
It remains to show that the two colors of the cycles Cn and Cm are identical.
Suppose that there exists a coloring without monochromatic triangles with the
property that all the of the cycle are blue and all the of the
are red. Suppose, without loss of that the
[otherwise the colors red and blue can be interchanged). (See 14.11.)
It follows from the nonexistence of a monochromatic triangle that [A l ' B 2]
IS blue, , B 2 ] is red, [A2' is blue, and hence [AI' B 3 ] is red. 'Now replace
320 Problems In Combinatorlcs and Graph Theory
BI by and repeat the argument. It turns out that for every two adjacent
and I of the cycle the colors of the edges and
,B J ';'1 are But this conclusion leads to a contradiction, since the
Cm is odd. It follows that the m + n of the and Cn are
colored with the same color.
m == n = 5 this problem was at the 21st International Mathe-
matical Olympiad in London in 1979.)
14.11 The will be established by induction on p + q. For p= 1 or
q = 1 it is immediate. now that p, q> 1. Let x be a vertex of the complete
with no vertices which is colored with two colors. Denote by d,(xJ the
number of red edges which have an endpoint at x, and db(x) the number of
blue which have an endpoint at x.
Since
dr(x)+db(x)=no l=(P -1 p
Suppose, for that the first inequality holds, and let G be the complete
subgraph induced by the vertices which are by red edges to x. (The
second inequality can be treated analogously.) Since G has at least e';~~ I)
vertices and its are colored red and blue, it follows from the induction
hypothesis that G contains either a complete red subgraph with p vertices or a
complete blue subgraph with q + 1 vertices. In the second case, the proof is
finished. But in the first case G contains a complete red subgraph with vertex
set H, where IHI = p. However, in this case H u {x} is a complete red subgraph
with p + 1 and the is finished. It is clear that R(p, q) = p} and
R(p,2)=p. The only known nontrivial values of the numbers q) with
Solutions 321
3 4 5 6 7 9
28
6 9 14 18 23 36
29
25 34
4 9 18
25 38 38
5 14
34 38 102
6 18
36
7 23
28
8
9 36
14.12 Consider the complete graph with five vertices as being repre-
sented a pentagon together with all its diagonals. Color the sides of
the pentagon red and the diagonals blue. There are no monochromatic triangles,
and hence 6.
[n order to prove the opposite it remains to show that every
coloring of the of K 6 with red and blue will yield at least one monochro-
matic triangle.
Let x be a vertex of K 6 . There are five which at x. and hence
at least three of these have the same color, say red. Thus there exist three vertices
ai' a2, G3 which are joined to x by red If one of the determined by
ai' az , a3, say is red. then there is a red triangle x, a I ' a2' Otherwise the
triangle with vertices GI, G2' G3 has all of its edges blue.
Thus every of with two colors contains a monochromatic
triangle.
14.13 The upper bound follows from Problem 14.11. In it has been seen
that
for p=q==k I.
322 Problems in Combinalorics and Graph Theory S
Observe that
2k -12) 2(2k - 3) (2k 4) < 4 (2k 4), (1)
( k k-1 ,k 2 ,k-2
For k=2 one has
v,
2k-2)=2=
( k-l
Suppose that l1
R
2k-2\,;:: 12) tI
( k-l J""
I
c: n'"
(2:)< 4 ~ 4 x
and thus (2) is true for every k';P 2. In order to obtain the lower bound one can
II
o
e
o
F
suppose that the index k ';P 4. In fact for k = 2, 2) = 2 = 2"'2, and for k = 3 g
one finds that R(3, 3)=6>2.ji Problem 14.12.
Let n = that is, the smallest integer which is greater than or equal to
. Let Xbe the set of the complete graph and let E = {E l , ... ,Em }
be the family of subsets of X defined as follows: Ei is the family of edges of a
complete subgraph with k vertices in Kn. Since there are (~) such complete
subgraphs, it follows that m=(~). By a similar one can show that
r= lEd =(~) for every 1", i '" m, since Kk has (~) 1
By the given it is possible to color the elements of X with two colors
so that no subset in the E has all elements of the same color if and
if the complete graph can be colored with two colors so that there is no \'
complete monochromatic subgraph with k vertices. Thus n < R(k, k) if one can e
color the elements of X with two colors so that no subset in the family E is mono-
chromatic. By Problem 4.24 this condition holds for m'"2' - 1, or
(~)"'2(~)-1. (3)
R(al'
+a2
all-
2).
which is the first in the induction on k.
Let k;a: 3, and suppose the numbers I (b! , ... , bk _ d exist for all
values b l , .• " bk - I;a: 1. It will be shown that
-I' ad)·
In order to do this, let G be the complete graph with R. , ... , ak - 2,
R2(ak -1, vertices. Color its arbitrarily with the colors C1 , ••• , ch. For
the time recolor the which have colors Ck _ I and Ck with a new color d,
the induction hypothesis there are two possible cases. Either there exists
an index i, 1<:i<:k-2, such that the G contains a complete subgraph
with aj vertices and with all colored Ci (in which case the proof is
or G contains a subgraph G1 with R 2(ak-!, ak ) vertices and with all
edges colored d. In the G1 the color d with color Ck ! or (;k, which
originally existed for the of G. From the definition of the Ramsey number
1, it follows that (and hence G) contains either a sub-
graph with ak 1 vertices and with all edges colored Ck 1, or a complete subgraph
with ak vertices and with all edges colored Ck-
14.15 Observe that
[k!eJ= ~J==
., k k!.
"
J. J.
The proof will use induction on k.
For k=2 it follows from Problem 14.12 that R 2 (3, 3)=6.
Let x be an arbitrary vertex of a complete graph G with [k! eJ + 1 vertices
) whose are colored with the colors C 1 , ...• Ck' The vertex x is thus an
1 endpoint of [k! eJ in G. Since
it follows that among the [k! e] with an endpoint at x there will exist
at least 1 + [(k -I)! eJ with the same color, say C1' Let X be the set of
r vertices which are joined to x by an of color Cl' If X contains two vertices
joined by an edge of color c1 , then these two together with x, form a
monochromatic of color Cl' and the is proved.
Otherwise X induces a complete with 1 + [(k -1)! e] vertices
of G, whose are colored with k-l colors. the induction
G1 (and hence G) contains a monochromatic triangle whose sides all have one
of the colors C 2, ..• , Ck; th is com pletes the proof of the
Observe that R 3(3, 3, [31 eJ + 1 == 17. In order to prove the
inequality one must find a coloring with three colors of the edges of the com-
graph K 16, which does not contain a monochromatic triangle.
324 Problems in Combinatorlcs and Graph Theory
14.18 Let no(k) = [k! If the nonempty subsets of the set {1, 2, ... , n} are
colored with k then color the [i,n (where 1 ~ n + 1) of the
graph Kn+ 1 with vertices I, 2, ... ,n + 1 with the same color as the
subset {i, ... ,j- to Problem 14.15, 1 contains a mono-
chromatic
Suppose that the vertices of this triangle are p, q, r where I < q < r ~ n + 1.
It follows that X = {p .... , q-l}, Y ={q, ... , r-l}, and Xu Y ={p, ... , r-l}
have the same color.
14.19 The will be established by induction on r. Let r denote
the vertices of by Xl' X2' ... , x., .. , and let r(xl) denote the number of red
incident with Xi; the two colors will be called red and blue.
Two cases will be examined:
(a) There exist a countable infinity of vertices XI" Xi2"'" which satisfy
r(xi) < 00. Define inductively the vertices Yl, .1'2' .. , of the infinite
monochromatic as follows: Let Yt ::= ; Y2 is the first of the vertices
XI) which is not to Yl by a red and Y2 is joined to Yl by a
blue There will exist such a vertex Y2 because there exist only a finite
number of red edges which are incident with YI' The vertex Y3 is the first vertex
among the vertices with indices greater than the index of Y2 which are joined
by a blue to YI and Yl' and so on. The process can be continued
since the set of vertices Xi) with 1 is infinite and each such vertex is incident
with a finite number of red It follows that the infinite subgraph
generated by the set of vertices {YI, Yl""} has all of its edges blue.
(b) Suppose that (a) does not hold, and assume that the graph K does not (f:;
contain a complete infinite subgraph with all of its edges blue. Let Xl =
Ir(x) = It follows that is an infinite set. Choose Yl eX 1, and denote
by X~ c:: X \ the subset of vertices of X I which are joined to Yl by a red
Since (al does not it follows that the vertices of which do not belong
to are finite in number, and thus X; is an infinite set. Denote by X 2 c:: X; the
subset of vertices of X i which are incident with an infinite number of red
with both endpoints in X~. If is finite, it follows that X~ ,",X 2 is an infinite
set of vertices which are incident with a finite number of red which have
both endpoints in The problem is thus reduced to case (a) for the
infinite with vertex set X~. There is a infinite
with all blue which contradicts the Thus
and one can choose Y2 e X 2'
It has been shown that the hypothesis that is an infinite set leads
to a contradiction, and thus X; ,",X 2 must be finite. Since Y2 is incident with an
infinite number of red with endpoints in X~, and since Xl ,",X 2 is finite,
it follows that the subset c:: of vertices in which are to Y2 a
red is infinite. Let X 3 c:: X zbe the subset of vertices of X zwhich are incident
with an infinite number of red with both endpoints in Xl' As before, one
finds that X 3 is an infinite set, so that one may choose Y3 eX 3, and so on.
By induction, a complete infinite graph generated by the set of vertices
326 Problems in Combinalorics and Graph Theory
{J'I' Y2""} is It has all of its red. This establishes the result
for r=2.
Suppose that the property holds for all colorings of with at most r-1
colors, and consider a coloring of with r colors: C1 , C2,' •• , c,., where r?it 3.
Recolor the which are colored C r - ! or c, with a new color Cr + I' the
induction for r 1 there exists a infinite mono-
chromatic If the color of the of this graph is one of the colors
Cl"'" C,-2' then the proof is finished. Otherwise there exists a I'ATnnlPI
infinite subgraph with all colored Cr + l ' recoloring the edges of this
graph which originally were colored Cr - I and Cr with these colors and applying
the property for two colors, one obtains a complete infinite subgraph with all
the same color 1 or cr }. The property is thus demonstrated
for every r.
14.20 Denote the vertices of the graph K y , numbers from the
set {I, 2, .. 'J' The j] is colored red kj and a;<uj. It is colored yellow
if i and aj > aj' and blue if i and ai aj' using the problem
one can show the existence of a complete infinite monochromatic subgraph. If
its color is red, then there will be an infinite strictly increasing subsequence;
for yellow one finds an infinite strictly decreasing subsequence, and for blue an
infinite constant
14.21 First we show that for every infinite set A of points in the plane there
is an infinite subset A 1 of collinear points or an infinite subset A 2 of points such
that no three points are collinear. Consider all the lines determined by pairs of
points from A. If one of them contains an infinite number of points of A, then
the property is demonstrated. Otherwise each line determined by tv;,o points
of A contains a finite number of points of A,
In this case carry out the following construction: Let x 1 and x;; be two
of A, Denote the set obtained from A by eliminating aU the points
of the line x I x 2, including x 1 and X2' 1t follows that B 1 is an infinite set. Let
X3 E B!. Denote by B2 the set obtained from by eliminating all points on the
Jines X3X! and X3X2 which belong to A. It follows that B2 contains an infinite
number of points. If a set of points has been obtained with the
property that no three are and if B'-1 is the infinite set of points which
belong to the set A and are not found on any of the lines determined by of
from , , .. , then let X y + I E B"-I' Denote by B,. the infinite subset
of points of Br _ 1 which are not found on any line x 1 Xr~ ! , ... , XrXy+ I , and so on.
Tt has thus been shown by induction that this construction can be continued
indefinitely, and thus A contains an infinite subset ! ' x 2, , . ,} of points with
whence
y = t(x + z) E [t(3" + 2), tw + 2a.)] = +
Thus this interval is disjoint both from A" and + 3", since + 1)
and t(2 x 3" + 1) < 3" + 1. It follows that y , l ' which is a contradiction. One
can conclude from the construction of the An's that ! has twice as many
elements as An (the sets An and An + 3n being disjoint). Hence the
of An is 2".
For n= 11 this reduces to a problem proposed at the 24th International
Mathematical Olympiad (Paris, 1983): Is it to choose 1983 distinct
positive integers, all less than or equal to , and no three of which are con-
secutive terms of an arithmetic "rrH'n'~OI
14.25 First \Ve establish the following
R(3, t)::;;';R(3, t 1)+ t. (1)
2k+l):S;;R(3, +2k+l,
and hence, by the induction hypothesis,
2k+ I)
or
+3
R(3, 2k+ 1)~ +2k+2
The is concluded showing that the last inequality for R(3, 2k) is strict.
that that in fact R(3, 2k2 + I «4kl + Suppose that there exists a
value of the index k~ 2 for which R(3, 2k)= 2k2 + 2. Thus there exists a graph H
with 2k2 + 1 vertices which contains neither a triangle nor an independent set
with 2k vertices.
If there existed a vertex y with 2k. then no two vertices to r
\vould be selfadjacent. It follows that H contains an independent set with 2k
vertices. which contradicts the hypothesis. Hence for every vertex x of Hone
has d(x):S;; 2k 1. Since H has an odd number of vertices and the sum of the
of its vertices is even, it follows that not all the vertices of H can have
the even degree 2k - 1. Hence H contains a vertex z such that d(z) 2k - 2.
Consider the graph obtained from H by the vertex z and all the
vertices adjacent to z. It can be seen from the induction hypothesis that H 0
has qo where
330 Problems In Combillatorks and Graph Theory
14.26 Suppose that the partite sets of G are A and B and hence = IBI =
2p + 1. Assume that the two colors are a and b, and let U e A. It follows that u is
adjacent to at least p + 1 vertices of B by having the same color, say a.
Let U c B denote the set of vertices of B which are adjacent to u by having
color a, so that I + 1. Let be the connected component composed only
of with color a and containing the vertex Ii eA. Suppose that ICoti Ai =
x ~ 1. It follows that every vertex of A that does not belong to Co is to
all vertices of U by edges of color b only. Denote the number of these vertices
which are found in A ""Co by O. It follows that the x vertices of A ti
together with the vertices of U are included in the connected component Co
having all edges of color a, and the y vertices of A which do not belong to Co
together with the vertices of U are contained in a connected component of G
having all edges of color b. Let 1=r ~ p + 1. One must show that
max(x+r,y+r)~2p+2. thatx+r:;;; +landy+r:;;; +1. In this
case x + y + 2r + 2, but one can write x + y + 2r = + 1 + 2p + 1 ....
2(p + 1} = 4p + 3, which is a contradiction. This completes the proof.
In order to see that the bound 2p + 2 cannot be improved, consider the
partitionsA=Aju andB=B j uB 2 ,whereIA 11 IBt!=p, =p+l.
Now color with a all the between Aj and B and between and
and with b all the remaining of K 2p" U p . j '
14.27 (a) It is necessary to show that if G is not then its com-
plement G is connected. Let x and y be two vertices of G. If x and yare not
adjacent in G, they are adjacent in G, and therefore x and y belong to the same
component C 1 of G. Since G is not connected, there exists a vertex z ¢ C!,
which implies that z is not adjacent to x and to yin G. It follows that z, yJ
is a walk of length 2 in Gbetween x and y, and that Gis connected.
(b) that the of are colored with three colors a, b, c. It will
be shown that there exists a monochromatic connected spanning
of with at least [(n + 1)/2] vertices. If one of the three colors is not
follows from (a) that this property holds. Suppose that Kn contains
colors a, b, and c, and let R denote a connected component of the spanning
subgraph of K. composed of all edges with the color a. If IRI =n, the property is
Solutions 331
n+IRI,
or IWI + IQ;;;: n, which implies that max (I Wi.;;;: Thus it has been
shown that h(n);;;: + 1)/2]. Ifn=.E 2 (mod 4). the inequality also holds.
Let X denote the vertex set of Kn , and consider an equipartition
X=X 1 UX 2 UX 3 UX 4
such that 1 ~ -IXll ~ 1 for every i,j = 1, ... ,4. Color the edges of Kn in the
following way: all between Xl' and between X 3, with the color a:
all between Xl' X 4, and between with the color b; and all
between Xl' X 3 and between X 2' X 4, with the color c. All edges having both
ends in a set Xi where 1 ~ i ~ 4 will be colored arbitrarily with the colors G, b, or c.
If n ¢ 2 (mod 4) it is clear that the maximum number of vertices of a mono-
chromatic connected of is to [(n + 1}j2], and hence
in this case it follows that If n == 2 (mod 4), then for the above-
defined of the set of Kn the maximum number of vertices of a
monochromatic connected spanning subgraph is equal to ni2 + 1, and therefore
in this case 13(n) ~ n!2 + 1. In order to prove the opposite consider
an arbitrary with the colors G, b, and of the for n 4p + 2
(p;;;: 1). It has been shown that there exists a monochromatic connected span-
ning subgraph of K. with at least + 1}/2J = + 1 vertices. Let H be such a
subgraph, and suppose that the edges of H have the color c. If A denotes the
connected component composed of with the color c, which contains the
vertices of and if ;;;:2p+2. it follows that 13(n);;;: + 1. Hence in this
case the 13(n)=nI2+1 holds for n 2(mod4}.
IAI = + 1, and if H denotes the set of the
[HI = + 1. No one extremity in A and the other in H is colored
with the color c, and hence the bipartite graph whose partite sets are
A and H has with the colors a or b only. By Problem 14.26 there exists a
332 Problems in Combinatorics and Graph Theory
For m=2, (1) is immediate, since if I has no red , it follows that it has
only blue and hence any vertex of K ... 1 is a center of a blue star K I •• '
Assume (I) to be true for all m':r.; m-l, and form the tree I removing an
endpoint x of Tm and the edge [x, yJ incident to x in Tm' In a 2-colored K", +._\
one can assume by induction that there is either a blue K l.n or a red l'
that the latter choice obtains. Since there are m + n - 1 - (m 1) = n
vertices Vi of 1 which are not vertices of the red l' and K", +" _ 1 con-
tains no blue Kl.., it follows that some from to some VI must be red. But
this forms a red in I' by induction holds and the proof is
The corresponding results for the case in which m -1 does not divide n-1
are much more complicated, and a solution has not been obtained.
However, in this case (1) still holds and. in fact, for almost all trees
R(T"" K1."l m+n-2 for n sufficiently large. [So A. Burr, Graphs and
Combinatorics, Lecture Notes in Mathematics, 406, Springer-Verlag, Berlin,
1974,
14.30 Form a of Kiln-I)" by taking m - ! copies of K. all having
only green and interconnect them by red This coloring contains no
Solutions 333
red Km and no green K I."! and hence R(K m• K I..) -1)n + 1. The inequality
R(K m , K l,n) ~(m -1)n + 1 can be obtained for m;?: 2 and n;;: 1 as a corollary of
Turan's theorem (Problem 9.9). Indeed, in a graph G with p (m -1)n + 1
vertices and minimum degree (m 2)n + 1 (the complement has maximum
degree n-l), the number q of is bounded below by q;?:(mn-n+ l)(mn-
2n + The number of in Turan's graph with (m -1)n + 1 vertices and
m - 1 ;;: I parts is equal to
m 2 (mn-n+l)2-1
M (mn - n + 1, m) = -m---l •-'------:;:---'---
n(m n+ 2)
The inequality
is equivalent to n + 1 > 0, which holds for n;;: 1. Hence q > M(mn - n + 1, m) and
by Tunin's theorem G must contain For m= 1 the result is obvious.
This formula was to R(K m• d, where I is any tree with
n+ 1 vertices, by V. Chvatal Theory 1(1) (1977), 93].
14.31 It will be shown that r(m) == - m - 1. Let {l, ... ,m 2 m 2} = A u B
be a partition into two classes, where
A {1,2, ... ,m-2,(m-l)2,(m-l)2+1, ... , -m-2}
and
B={m l,m, ... ,(m-I)2-1}.
that this property does not hold. Let 1 EA. It follows that m -1 E B
and (m 1)2 EA. Now consider two cases: (a) mEA and (b) m E B.
(a) If mEA one can see that -2m+l=m+ .'. +m+1, that is,
A is not m-sum-free.
(b) lfmE then since (m l}+m+ .,. +m=m2-m 1 it follows
that m2 -m 1 € A, [n this case one can write
1+ .. , + 1 + m2 - 2m + 1 = - m - 1;
thus A is not m-o.\,4W.-! and this contradicts the
[A. Beute!spacher, W, Brestovansky, Combinatorial Theory,
Lecture Notes Math., Springer-V eriag, Berlin, 969 (1982), 30- 38.J
Bibliography
M. Aigner, Combinatorial Theory, Springer-Verlag, Berlin Heidelberg, 1979.
C. Berge, Principes de Combinaroire, Dunod, Paris, 1968.
C. Berge, Graphes et Hypergraphes, Dunod, Paris, 1970.
M. Behzad, G. Chartrand, L Lesniak-Foster, Graphs and Digraphs, Prindle, Weber and Schmidt,
Boston,I979.
B. Bollobas, Graph Theory. An introductory Course, Springer-Verlag, New York Heidelberg, 1979.
L Comte!, Analyse Combina/oire, 1, n. Presses Cniversitaires de France. Paris. 1970.
L. R. Ford, Jr .. D. R. Fulkerson. Flows in Network.'i. Princeton University Press. 1962.
A. M. Gleason. R. E. Greenwood, L. M. The WL. Putnam Mathematical Campention. Proh-
ferns and SolutIOns: 1938 1964. The Associatton of Amenca. 1980.
1. E. Graver, M, E. Watkins, Combinatorics with Emphasis on the Theory afGraphs, Springer-Verlag,
New York - Heidelberg - Berlin, 1977.
M. Hall, Jr., Combinatorial Theory. Blaisdeli, Waltham, Massachusetts, Toronto- London, 1967.
F, Harary, Graph Theory, Addison-Wesley, Reading, :'vlassachuse!!s, 1969,
D. E. Knuth, The Art oj Compuler Programming. Vol. I,' Fundamenral Algorithms, 2nd cd., Addison-
Wesley, Reading, Massachusetts, 1971
L. Lovasz, Comninatorial Problems and Exercises, Akademiai Kiado, Budapest, 1979.
J. W, Moon, Counring Lanelled Trees, Canadian Mathematical Monographs, I, Canadian Mathe-
matical Congress, 1970.
J. W. Moon, Topics on Tournaments, Holt. Rinehart and Winston. New York, 1968.
H. Ryser, CombinaTorial MathematiCS, Wile), New York. 1963.
I. Tomescu, Introduction to Combinatorics, Collet's, London-Wellingborough. 1975.
N. Vilenkin, Combinatorial Mathcmaticsjor Recrearion, Mir, Moscow, 1972,
R. 1. Wilson. InTroduction to Graph Theory, Longman, London, 1975.