Appendix B

Hints to Selected Exercises

Section 1.1
1.1.2 (b) Observe that a simple bipartite graph with the maximum number of
edges is necessarily complete bipartite and that Kr,s has rs edges. Now show that
if s − r ≥ 2, then Kr+1,s−1 has more edges than Kr,s . Alternatively, show that
n2 /4 − r(n − r) is a perfect square, and thus nonnegative.
1.1.7 (b) To determine e(Qn ) apply Theorem 1.1.
1.1.9 (a) Counting in two ways (see inset).
1.1.10 Find upper bounds on the degrees and apply Theorem 1.1.
1.1.11 This is a generalization of Exercise 1.1.2. Use a similar proof technique.
1.1.12 If G is not connected, there is a partition (X, Y ) of V such that no edge of
G joins a vertex in X to a vertex in Y . What is the largest number of edges that
G can have if |X| = r and |Y | = n − r?
1.1.13 Observe that if a graph G is disconnected, then there exists a subset X
of V , with |X| ≤ ⌊n/2⌋, such that no edge of G joins a vertex in X to a vertex in
V \ X.
1.1.15 Observe that a set of vectors over GF (2) is linearly dependent if and only
if it contains a subset whose sum is the zero vector.
1.1.16
(a) Necessity: apply Theorem 1.1.
Sufficiency: how many di ’s are odd?
(b) Necessity: apply Theorem P 1.1.
n
Sufficiency: induction on ( i=2 di ) − d1 .
1.1.18 (b) Counting in two ways. Let G = (V, E), where V = {v1 , v2 , . . . , vn } and
d(vi ) = di , 1 ≤ i ≤ n. Set X := {v1 , v2 , . . . , vk }. For vi ∈ X, find a lower bound on
the number of edges joining vi to vertices in V \ X. For vi ∈ V \ X, find an upper
bound on the number of edges joining vi to vertices in X. Deduce lower and upper
bounds on the number of edges with one end in X and the other in V \ X.
1.1.19 (a) To prove the necessity, first show that if G is a simple graph with
u1 v1 , u2 v2 ∈ E and u1 v2 , u2 v1 6∈ E, then (G \ {u1 v1 , u2 v2 }) + {u1 v2 , u2 v1 } has
the same degree sequence as G. Deduce that if d is graphic, then there is a simple
graph G with vertex set {v1 , v2 , . . . , vn } such that d(vi ) = di , 1 ≤ i ≤ n, and
N (v1 ) = {v2 , v3 , . . . , vd1 +1 }.
1.1.20 Define a graph on S in which xi and xj are adjacent if and only if they
are at distance one. How large can the degree of a vertex be in this graph?
1.1.21
(a) This is a classical result of linear algebra, and is true for any real symmetric
matrix.
(b) This is true for any integer square matrix A. Suppose that λ := p/q is an
eigenvalue of A, where p and q are coprime integers. Assume that q > 1. Using
the fact that a scalar multiple of an eigenvector is also an eigenvector, show
that there exists an integral eigenvector corresponding to λ not all of whose
coefficients are divisible by q. Derive a contradiction.
1.1.24 (a) Consider the case where G = (V, E) is simple. Let V = {v1 , v2 , . . . , vn }.
An eigenvector of G corresponding to an eigenvalue λ may be viewed as a
weight function on V such that the sum of the weights on the neighbours of
a vertex is equal to λ times the weight of the vertex. Choose an eigenvector
x = (x1 , x2 , . . . , xn ) of λ such that |xj | := max{|xi | : 1 ≤ i ≤ n} = 1, and
consider the vertex vj .

Section 1.2
1.2.14 (a) Use Exercise 1.2.11 to reduce the number of cases to consider.
1.2.16
(c) How many edges has a self-complementary graph?
(d) An isomorphism mapping G to G induces a pairing of the vertices of G. What
can you say about their degrees?
1.2.17 Since G is not vertex-transitive, there exist vertices x and y such that
no automorphism of G maps x to y. Let X and Y denote the orbits of x and y,
respectively. Consider a neighbour z of x. Using the fact that G is edge-transitive
and that y is not isolated, show that z ∈ Y \ X. Now show that (X, Y ) is a
bipartition of G.
1.2.18 (b) To show that the Folkman graph is not vertex-transitive, observe that
it is bipartite. Find a property of the vertices in one part that is not shared by the
vertices in the other part.
1.2.20 By Exercise 1.2.8, an automorphism of G can be regarded as a permutation
matrix P such that PAPt = A. Using the fact that Pt = P−1 (why?), show that
if x is an eigenvector of G corresponding to an eigenvalue λ, then so is Px. Now
use the fact that G has n distinct eigenvalues, that eigenvectors corresponding to
distinct eigenvalues are orthogonal (why?), and that P is a permutation matrix,
to deduce that Px = ±x. Conclude.

Section 1.3
1.3.7 (b) For each of the four triangles in the graph of Figure 1.20, apply (a) to
the set of three intervals represented by its vertices. Deduce that one of the six
intervals must meet all five of the other intervals.
1.3.15 (a) Assuming that |X| ≥ |Y |, show that
1 1
≥
|X|(|Y | − d(x)) |Y |(|X| − d(y))
for all xy ∈
/ E. Now apply the proof technique of counting in two ways to deduce
that |X| = |Y |.
1.3.17 (a)
(i) The complement of L(K4 ) is a 1-regular graph on six vertices. Use this fact to
show that Aut(L(K4 )) and Aut(K4 ) have different orders.
(ii) Suppose that n ≥ 3, n 6= 4. Show that an automorphism of Kn induces
an automorphism of L(Kn ). Deduce that the symmetric group Sn is a subgroup
of Aut(L(Kn )). Noting that an automorphism of L(Kn ) maps adjacent edges of
Kn to adjacent edges, show that every automorphism of L(Kn ) is induced by an
automorphism of Kn .
1.3.18 (c) Up to isomorphism, there are just two groups of order ten, namely Z10
and Z2 × Z5 . Proceed by considering various cases (see Holton, D.A. and Sheehan,
J. (1993). The Petersen Graph. Cambridge University Press, pp. 292–293).

Section 1.5
1.5.7 (a) Let B be an r × r submatrix of M. If r = 1, det B = 0, ±1. Suppose that
r ≥ 2. If B has a column of 0s, or if each column of B has two nonzero entries,
then det B = 0 (why?). If not, proceed by induction.
1.5.11 (b) Let xy be a fixed edge of G. First, show that there is no automorphism
of G which maps the ordered pair (x, y) to its converse (y, x). Then show that, for
every edge uv of G, the orbit of the ordered pair (x, y) under the automorphism
group of G includes exactly one of the two ordered pairs (u, v) and (v, u) but not
→
both. Now obtain a directed graph G by orienting an edge uv from u to v if (u, v)
is in the orbit of (x, y) and from v to u, otherwise. Show that every automorphism
→ →
of G is also an automorphism of G. Conclude that G is vertex-transitive and hence
diregular. Deduce that G is regular of even degree.
1.5.12 Show that B is skew-symmetric, and hence that det B = 0 when n is odd.
When n is even, use the fact that the number d(n) of derangements (permutations
in Sn with no fixed point) is odd (in fact, d(n) = nd(n − 1) + (−1)n for n ≥ 2) to
deduce that det B 6= 0.
Section 2.1
2.1.2 (a) Consider the ends of a nontrivial maximal path.
2.1.3 (a) If G has a vertex of degree less than two, delete it and apply induction.
2.1.4 (a) Consider a maximal path and the neighbours of one end of the path.
2.1.5 (a) Consider a maximal path and the neighbours of one end of the path.
(b) Apply induction.
2.1.6 Consider a maximal path and the neighbours of one end of the path.
2.1.8 (a) Consider the neighbours of two adjacent vertices.
2.1.9 Consider a vertex, its neighbours, and the neighbours of its neighbours.
2.1.11 (a) Consider the ends of a maximal directed path.
2.1.12 Consider the arcs in a longest directed path.
2.1.14 (a) Show that G is regular. Now consider the degrees in a vertex-deleted
subgraph.
2.1.17 (a) Consider a shortest odd cycle.
2.1.18 Let X be a set of p vertices which induce a subgraph G[X] whose minimum
degree is as large as possible. If there is a vertex u of G[X] whose degree is less
than q, and also a vertex v of G − X whose degree is less than (k − 1)q, consider
the set (X \ {u}) ∪ {v}, and apply induction on k.
2.1.19 Let C be a shortest odd cycle in KGm,n . Show that the union of the
(n − 2m)-sets S \ {x ∪ y}, xy ∈ E(C), is the entire n-set S.
2.1.22 Consider the set of minimal elements of P . Show that they form an an-
tichain, delete them, and apply induction on |X|.
2.1.23 (a) Show that any vertex of degree three or more in G has a neighbour of
degree one, and apply induction on n.

Section 2.2
2.2.11 Consider a shortest path between two nonadjacent vertices.
2.2.18 (a) By contradiction. Let G be a smallest counterexample. Show that the
girth of G is at least five, and that δ ≥ 3. Deduce that n ≤ 8. Conclude that no
such graph exists.
2.2.19 One may assume that G is connected (why?). Consider the cases δ ≥ 3
and δ ≤ 2, and apply induction.
2.2.20 (a) Show that either G has a vertex of odd degree such that G − x is
connected, or G has an edge xy such that G − {x, y} is connected, and apply
induction.
2.2.21 (b) Does the ‘Proof’ take into account all the graphs satisfying the hy-
potheses of the ‘Theorem’ ?
2.2.25 Form a simple graph with vertex set S, two points being adjacent if and
only if their distance is exactly one. Show that any vertex of degree three or more
in this graph has a neighbour of degree one, and apply induction.
2.2.27 Suppose that all induced subgraphs of G on k vertices have l edges. Show
that, for any two vertices vi and vj ,

n−3

e(G) − d(vi ) = e(G − vi ) = l n−1
k / k−2
n−4

e(G) − d(vi ) − d(vj ) + aij = e(G − vi − vj ) = l n−2
k / k−2

where aij = 1 or 0 according as vi and vj are adjacent or not. Deduce that aij is
independent of i and j.

Section 2.4
2.4.6 (a) Label the vertices 0, 1, 2, ..., 2n. Arrange the vertices 1, 2, ..., 2n in a circle
with 0 at the centre. Consider the cycle (0, 1, 2, 2n, 3, 2n − 1, 4, 2n − 2, ..., n + 2, n +
1, 0) and its rotations.
2.4.7 Denote by G the graph in Figure 2.7, and by P1 and P2 its inner and outer
pentagons. Consider a hypothetical cycle decomposition C of G, no member of
which is a 2-cycle. Show that |E(C) ∩ E(Pi )| = 2, i = 1, 2. Derive a contradiction
based on parity.
2.4.9 Consider an arbitrary linear ordering of the vertices of the digraph.
2.4.10 (a) Set P := {v1 , . . . , vn } and L := {L
P1 , . . . , Lm }, Denote by mj P
the jth
column of M, and consider the vectors xi := {mj : vi ∈ Lj } and yi := {mj :
vi ∈
/ Lj }, 1 ≤ i ≤ n. Show that xi has ith coordinate di := d(vi ) and all other
coordinates 1, and that yi has ith coordinate 0 and kth coordinate dk − 1, k 6= i.
Express each unit vector of Rn as a linear combination of the vectors xi and yi ,
1 ≤ i ≤ n.

Section 2.5
2.5.1 (a) Counting in two ways. Consider the submatrix of the incidence matrix
M consisting of the rows corresponding to the vertices in X.

Section 2.6
2.6.6 If G is even, set X := V . If not, proceed by induction on n. Let v be a
vertex of odd degree in G. Consider the graph obtained from G − v by replacing
the subgraph G[N (v)] by its complement.
Section 2.7
2.7.1 Use Kelly’s Lemma (2.20) to obtain constraints on the numbers of subgraphs
of various types in the missing vertex-deleted subgraph.
2.7.3 (b) A tree is a connected acyclic graph. Look for a tree on eleven vertices
containing a path of length eight.
Section 3.1
3.1.1 Consider a shortest xy-walk.
3.1.2 Use induction on k.
3.1.8 Let x be a vertex of degree n − 2, y the vertex nonadjacent to x, and X
and Y the neighbour sets of x and y, respectively. Consider the set X \ Y .

Section 3.2
3.2.3 (a) Review the theorems in Section 2.5.
3.2.4 Proceed by contradiction. Assume that there is a cut edge, delete it, and
consider the degrees of vertices in the resulting subgraph.

Section 3.3
3.3.5 Add a new vertex with neighbour set X.
3.3.7 If G − v contains a cycle C, consider a suitable Euler tour of the component
of G \ E(C) containing v. If G − v is acyclic, let Q be a v-trail of G which is not
an Euler tour. Show that G \ E(Q) has exactly one nontrivial component.
3.3.9 Let W := v0 e1 v1 e2 v2 . . . em vm be an Euler tour of a graph G, where vm = v0 .
←−
Suppose that vi = v0 , where 0 < i < m. Show that v0 W vi em vm−1 W vi is also an
Euler tour of G.

Section 3.4
3.4.7 Apply Rédei’s Theorem (2.3) to an appropriate digraph defined on the same
vertex set.
3.4.11 (b) Suppose that D is strong and contains an odd cycle v1 v2 . . . v2k+1 v1 . If
(vi , vi+1 ) ∈ A, set Pi := (vi , vi+1 ); if (vi , vi+1 ) 6∈ A, let Pi be a directed (vi , vi+1 )-
path. If all the paths Pi are of odd length, apply (a).

Section 3.5
3.5.3 Use Exercise 2.2.6.
3.5.4 (b) Consider all possible symmetric differences of the cycles C1 , C2 , and C3 .
3.5.7 Induction on m. If G has an edge cut of size two, contract one of these
edges. If not, apply the induction hypothesis to each edge-deleted subgraph of G,
and combine appropriate numbers of copies of the m uniform cycle covers thereby
obtained.
3.5.8 (a)(i) To prove sufficiency, reduce to the case in which G has no edge cut
of size one or two, and apply Exercise 3.5.7 to each edge-deleted subgraph of G.
Obtain a positive integer r such that each of the vectors ue , e ∈ E, defined by
 
0 if f = e
ue (f ) :=
r if f ∈ E \ {e}

is a linear combination of vectors in FC .

Section 4.1
4.1.4 Consider each component of F separately.
4.1.7 To prove sufficiency, consider a graph G with degree sequence d =
(d1 , d2 , . . . , dn ) and as few components as possible. If G is not connected, show
by a suitable exchange of edges (as in the hint to Exercise 1.1.19) that there is a
graph with degree sequence d and fewer components than G.
4.1.12 Show that G has a spanning forest with the same set of labels as G. Apply
Exercise 4.1.4 to deduce that some element of S is not a label of G. Conclude.
4.1.14 (b) Form a bipartite graph B(X, Y ), where X := {x1 , x2 , . . . , xm } is the
set of pizzas, Y := {y1 , y2 , . . . , yn } is the set of students, and xi is joined to yj if
a portion of pizza xi is allotted to student yj . Show that B has no more than d
components.
4.1.15 Delete the root and use induction on n.
4.1.19 (a) Let T be a tree. Show by induction that there is a Hamilton cycle in
T 3 that uses any prescribed edge e = xy of T . Apply the induction hypothesis to
the two components of T \ e, choosing an appropriate edge in each component.

Section 4.2
4.2.5
a) Denote by Fn′ the graph obtained from Fn := Pn−1 ∨ K1 by doubling the edge
from the vertex of K1 to one end of Pn−1 . Set fn := t(Fn ) and fn′ := t(Fn′ ).
Apply Proposition 4.9 to derive two recurrence relations involving these two
functions. Deduce that fn −3fn−1 +fn−2 = 0, and solve this recurrence relation.
b) Denote by Wn′ the graph obtained from Wn by deleting a spoke. Set wn :=
t(Wn ) and wn′ := t(Wn′ ). Observe that each spoke of Wn is in the same number
an of spanning trees. Likewise, each rim edge of Wn is in the same number
bn of spanning trees. Find three relations linking an , bn , wn , wn′ , and fn ,
and apply Proposition 4.9 to an appropriate edge e of Wn′ to obtain another
relation linking these quantities. Deduce from the resulting four relations that
wn − wn−1 = fn + fn−1 , and apply (a) to obtain a recurrence relation for wn .
Solve this recurrence relation.
4.2.11 To each labelled tree T with vertex set {1, 2, . . . , n}, associate the sequence
(j1 , j2 , . . . , jn−2 ), where j1 is the neighbour in T1 := T of the smallest leaf i1 of
T1 , j2 is the neighbour in T2 := T1 − i1 of the smallest leaf i2 of T2 , and so on.
4.2.12
a) The multinomial coefficients satisfy the following recursion:
  k 
X 
n n−1
=
d1 , d2 , . . . , dk i=1
d1 , . . . , di − 1, . . . , dk
 Using the fact that every tree on two or more vertices has a vertex of degree one,
show that t(n; d1 , d2 , . . . , dn ) satisfies a similar recursion, and apply induction
on n.
b) The Multinomial Theorem states that
X n

(x1 + x2 + · · · + xk ) = n
xd1 xd2 . . . xdkk
d1 , d2 , . . . , dk 1 2

where the sum is taken over all partitions Pnof n into k nonnegative integers
d1 , d2 , . . . , dk . Consider the expansion of ( i=1 1)n−2 .
4.2.13 Call the vertices in the m-set of Km,n red and the remaining n vertices
blue. Observe that each spanning tree of Km,n rooted at a red vertex is the union
of a spanning forest F of stars rooted at red vertices and a set S of m − 1 blue–red
edges. How many choices are there for F and for S?

Section 4.3
4.3.9 Let C be a Hamilton cycle of G and e an edge of C. Apply Corollary 4.12
to T := C \ e.
4.3.10 Apply Corollary 4.12.
Section 5.1

5.1.1 Consider the ends of a nontrivial maximal path.

5.1.4 Add a new vertex x adjacent to each vertex of X, and a new vertex y
adjacent to each vertex of Y . Apply Exercise 5.1.1 and Theorem 5.1.
5.1.5 Let C1 and C2 be two longest cycles. Apply Exercise 5.1.4 with X := V (C1 )
and Y := V (C2 ).

Section 5.2
C
5.2.3 (a). To show that ∼ is transitive, consider three edges e, f , and g of G. Let
C1 be a cycle containing e and f , and C2 a cycle containing f and g. Show that
C1 ∪ C2 is nonseparable and apply Theorem 5.2.
5.2.5 Show that the edges of a cycle lie in a single block of the graph. Now use
Theorem 2.7 for (a) and Theorem 4.7 for (b).
5.2.7 (a) If G is not bipartite it contains an odd cycle C. Apply Exercise 5.1.4
with X := {x, y} and Y := V (C).
5.2.8 (b) Use induction on the number of blocks of G. If G is separable, consider
a separation {G1 , G2 } of G. Apply the induction hypothesis to G1 and G2 .
5.2.10 There is such a graph on four vertices.
5.2.11 How many endblocks are there in G \ e?
5.2.12 (a) Let G be an even graph. Proceed by induction on n. One may assume
that G has no loops (why?). Consider a vertex v. Each partition into pairs of the
edge cut ∂(v) gives rise to an even graph on n − 1 vertices and m − 12 d(v) edges
by splitting off these pairs.

Section 5.3
5.3.2 Apply Exercise 5.2.11.
5.3.3 Show that no odd cycle has an ear.
5.3.4 Observe that the statement is true if G is a cycle. Use induction on the
number of ears in an ear decomposition of G.
5.3.7 Consider the final ear P in an ear decomposition of G and proceed by
induction on the number of ears. There are various cases to examine, depending
on the number of internal vertices of P in X ∪ Y .

Section 5.4
5.4.4 Proceed along the lines of Exercise 5.3.8, using Theorem 5.13.
Section 6.1
6.1.1 Use induction on ℓ(v).
6.1.7 If T is not a DFS-tree (rooted at x), there is a cross edge in G (an edge
joining two unrelated vertices of T ). Using this edge, find a spanning x-tree which
contradicts the choice of T .
P
6.1.9 (a) Set ρ(Tv ) := u∈V \{v} d(u, v). Show that σ(Tv ) ≤ (n − 1)ρ(Tv ) and that
P
v∈V ρ(Tv ) = 2σ(G).

6.1.11 Let P be a Hamilton path of G. Show that P extends to a Hamilton cycle

C of G and that if xy is a chord of C, then x+ y + and x− y − are also chords of C.
If the length of xCy is at least four, show further that x++ y and x+ y − are chords
of C as well. Now consider whether or not C has a chord xy with xCy of length
two.

Section 6.2
6.2.1 Suppose that T1 and T2 are two optimal trees in G. Apply Exercise 4.3.2,
choosing the edge e appropriately.
6.2.5 Use the fact that log αβ = log α + log β.
6.2.6 (a) Let T2 be a spanning tree whose largest edge-weight is smaller than the
largest edge-weight in T . Apply Exercise 4.3.2 with T1 = T , choosing the edge e
appropriately.

Section 6.3
6.3.7 Return the vertices in an appropriate order depending on the function l.
6.3.8 Return the vertices in an appropriate order depending on the function f .
6.3.13 (b) Consider a DFS-tree in G which is rooted at one end of the path P
and contains P . Orient G accordingly.
Section 7.1
7.1.1 (a) Observe that both {∂ + (v) : v ∈ V } and {∂ − (v) : v ∈ V } are partitions
of the arc set A.
7.1.2 (a) Arcs with exactly one end in X appear in precisely one of the two sets
∪{∂ + (v) : v ∈ X} and ∪{∂ − (v) : v ∈ X}, whereas those with their two ends in X
appear in both of these sets.
7.1.4 Use Theorem 3.6.

Section 7.2
7.2.3 By appropriate scaling, all capacities may be taken to be integers.
7.2.4 If f is a nonzero flow, consider the set of f -positive arcs.

Section 7.3
7.3.1 (a) Use Exercise 7.1.2.
Section 8.1
8.1.3 See Exercise 5.2.7.
8.1.4 Start with a mapping between the centres of the two trees (see Exer-
cise 4.1.8b). Now use a recursive algorithm for deciding whether two rooted trees
are isomorphic.

Section 8.2
8.2.1 Use the vertex-splitting operation illustrated in Figure 8.2 for reducing
Problem 8.5 to Problem 7.10.

Section 8.3
8.3.5 (a) Use a vertex-splitting operation similar to the one illustrated in Fig-
ure 8.2, inserting a path of length two rather than an arc.
8.3.9 There are three literals in each clause of f , and thus eight triples of possible
values, seven of which satisfy the clause. Let each part Vi of G correspond to a
clause of f , and the seven vertices in Vi correspond to these seven triples of values.
8.3.10 (a) One may assume that there is no clause of the form x ∨ x. Form a graph
G = (V, E) whose vertex set V is the set of literals of f , there being an edge xy
for each clause x ∨ y of f , and one for each pair {x, x} of literals contained in V .
Call the latter set of edges M .
Starting with an arbitrary vertex as root, grow a tree T which includes all
vertices of G reachable from the root by paths whose edges are alternately in
E \ M and M . Colour the vertices of T alternately red and blue, the root being
red. If V (T ) 6= V (G), grow another such tree in G−V (T ), and continue in this way
until a spanning forest of G has been constructed. Consider the truth assignment
in which each red vertex is assigned the value 0 and each blue vertex the value 1.

Section 8.4
8.4.3 Find a polynomial reduction from Hamilton Cycle.

Section 8.6
8.6.7 To obtain D(x, y) from G[X, Y ], adjoin two new vertices x and y, join
x to all vertices in X and y to all vertices in Y , and orient the resulting graph
appropriately.
Section 9.1
9.1.2 Observe that any vertex cut of G ∨ H must include either V (G) or V (H).
9.1.3 (a) If G is not complete, consider a vertex cut S of G and a smallest com-
ponent of G − S. How large can the degree of a vertex in such a component be?
9.1.4 If G is not complete, consider a vertex cut S of G and a smallest component
of G − S. How large can the degree of a vertex in such a component be?
9.1.6 Consider the edges of a longest cycle of G.
9.1.7 Observe that every loop is deletable, and that each member of a set of
multiple edges is deletable. Now use Exercise 5.3.2. Note that by a contractible
edge is meant an edge e such that G / e is 2-connected, not one such that κ(G / e) =
κ(G).
9.1.8 Let S be a minimal vertex cut of G. Show that every vertex of S is adjacent
to at least one vertex of each component of G − S. Deduce that there are two
vertices in distinct components of G − S whose distance in G is two.
9.1.10 (a)
(i) Use induction on m. Consider two cases, according to whether or not G / e is
minimally 2-connected, where e ∈ E. In the latter case, by using the fact that
(G / e) \ f = (G \ f ) / e for any edge f of G / e, show that e is incident to a
vertex of degree two.
(ii) Proceed as in (i). If G / e is not minimally 2-connected for any e ∈ E, show
that every edge of G is incident to a vertex of degree two, and apply a counting
argument.
9.1.13 Use induction on the length of P .

Section 9.2
9.2.2 Consider a shortest odd cycle C together with a 3-fan from a vertex not on
C to V (C).
9.2.4 Find a 5-connected plane graph G whose outer face is of degree at least four,
and place the four vertices x1 , y1 , x2 , y2 in an appropriate order on the boundary
of this face.

Section 9.3
9.3.3 Let ∂(X) be a minimum edge cut of G. Use the fact that G is of diameter
two to deduce that either d(X) ≥ |X| or d(X) ≥ |V \X|. Without loss of generality
(why?) assume the former. Now apply Theorem 2.9 and Exercise 9.3.2a to deduce
that d(X) = δ.
9.3.4 Let ∂(X) be a minimum edge cut. Without loss of generality (why?) it
may be assumed that |X| ≤ n/2. Show that |X| ≤ δ. Now use Theorem 2.9 and
Exercise 9.3.2a to deduce that d(X) = δ.
9.3.5 Consider the cases κ = 0, 1, 2, 3 separately.
9.3.12 Let G := G(x, y) be a graph with two specified vertices x and y. Obtain
G′ from G by adding two new vertices x′ and y ′ and joining x′ to x and y ′ to
y. Now let H := H(u, v) denote the line graph of G′ , where u and v correspond
to the edges xx′ and yy ′ of G′ , respectively. Show that the maximum number of
edge-disjoint xy-paths in G is equal to the maximum number of internally-disjoint
uv-paths in H.

Section 9.4
9.4.7 If G / e is not 3-connected, denote by H the graph of minimum degree at
least three of which G \ e is a subdivision. Show that any two distinct vertices of
H are connected by three internally disjoint paths.
9.4.8 There is an error in the statement of this exercise: ‘any edge’ should be
replaced by ‘some edge’. C. Thomassen showed how this theorem (Tutte’s Wheel
Theorem) can be derived from his Theorem 9.10. Assume that G is minimally 3-
connected: κ(G \ e) = 2 for all e ∈ E. By Theorem 9.10, G contains an edge e such
that G / e is 3-connected. Suppose that G / e is not simple, so that e is contained
in a triangle. Show first of all that at least two vertices of that triangle have degree
three. Thus G contains a path of length one whose vertices (i) are of degree three
and (ii) have a common neighbour. Now consider a longest path in G with these
two properties. Show that the ends of this path are adjacent, and that G has no
other vertices. Conclude that G is a wheel.
9.4.9 (a) Show that no graph in G has two disjoint cycles, but that any simple
graph obtained from a member of G by adding an edge or splitting a vertex of
degree greater than three is either a 3-connected graph with two disjoint cycles or
another member of G. Prove the required assertion by induction on the number of
edges, using Exercise 9.4.8.

Section 9.5
9.5.3 For k = 8, an example can be found amongst the figures in Chapter 14.
Generalize this example to obtain one for all k ≥ 5.
9.5.6 Let x and y be the two vertices of odd degree in G. Consider a balanced
orientation of G + xy.
9.5.8 (b) If G is not minimally 2k-edge-connected, an edge may be deleted from
G to obtain a 2k-edge-connected graph of smaller size. On the other hand, if G
is minimally 2k-edge-connected, it has a vertex of degree 2k, by Exercise 9.3.15.
Now use Theorem 9.16.
Section 9.6
9.6.3
(a) (i) Show that d(X) is odd or even according to whether X contains an odd or
an even number of vertices of odd degree. In particular, an odd edge cut
separates some pair of vertices of odd degree.
(ii) Suppose that d(X) is even. Using parity arguments and submodularity,
show that there is a smallest odd edge cut ∂(Y ) which does not cross
∂(X).
(b) By (a), a smallest odd edge cut among the n − 1 cuts determined by the
Gomory-Hu algorithm is a smallest odd edge cut of G.

Section 9.7
9.7.2 Use induction on the length of the sequence of cliques in the decomposition.
9.7.3 Use induction on the order of the graph.
9.7.4 (a) Consider two disjoint subtrees Ti and Tj of a tree T , each of which
intersects a third subtree Tk . Show that Tk contains the path in T connecting Ti
and Tj .
Section 10.1
10.1.5 First, show that G can be assumed 3-connected. Next, apply Theorem 9.6
to obtain a nonspanning cycle C of length at least three. Finally, apply the Fan
Lemma (9.5).
10.1.6 Count the degrees and the number of vertices. Do you recognize this graph?
10.1.7 Place the vertices along the spine.
10.1.8 (c) To show that the Petersen graph has crossing number at least two, use
the fact that the graph is edge-transitive and appeal to Exercise 10.1.3(a).
e of G := Kn+1 , one with exactly cr(Kn+1 )
10.1.9 Consider an optimal drawing G
crossings. Compute the average number of crossings induced in a Kn -subgraph of
e
G.
10.1.10 Consider a pair of crossing edges in an optimal drawing of the graph.
10.1.12 Consider the line segment joining two distinct points (p, p2 , p3 ) and
(q, q 2 , q 3 ) of the given curve. Let (x, y, z) be a generic third point on this line
segment. Show that x2 − y 6= 0. Deduce that no three points of the curve are
collinear. Now compute (xz − y 2 )/(x2 − y). Deduce that the line segments joining
(p, p2 , p3 ) to (q, q 2 , q 3 ) and (r, r2 , r3 ) to (s, s2 , s3 ), where p, q, r, s are distinct, can
meet only if pq = rs. Choose the points at which to place the vertices of the graph
accordingly.

Section 10.2
10.2.4 Use Exercise 10.2.2.
10.2.11 Use Exercise 10.2.12a.
10.2.15 Use induction on v(H), the initial case being when T is a star and H a
wheel. If T is not a star, consider a vertex of T all of whose neighbours but one
are leaves.

Section 10.3
10.3.1 How many edges must be deleted from G in order to obtain a planar
graph?
10.3.2 Look at the proof of Corollary 10.21.
10.3.3 How many edges are there in a spanning tree?
10.3.4
(a) How many edges has K11 ?
(b) There is a self-complementary plane graph on eight vertices which is also self-
dual.
10.3.5 (a) Use Euler’s Formula.
10.3.8 (a)(i) Use Euler’s Formula.
10.3.9 (a) Form an appropriate planar graph.
10.3.10 Form an appropriate planar graph.

Section 10.4
10.4.1 Use Proposition 10.5.
10.4.4 Apply Exercise 9.4.6 and Theorem 10.28.

Section 10.5
10.5.3 (b) Use induction on the number of contractions needed to obtain K5 as a
minor.
10.5.5 Refine the proof of Theorem 10.35.
10.5.6 Place the vertices along the positive and negative x- and y-axes.
Section 11.1
11.1.2 Use planar duality.
11.1.3 Take Z2 × Z2 as the set of colours.
11.1.4 Take Z2 × Z2 as the set of colours.
11.1.7 (a) Apply Exercise 11.1.6d.
11.1.8
(a) Let xy be an edge of a triangulation G = (V, E). Show that there is a unique
3-vertex colouring c : V → Z3 with c(x) = 0 and c(y) = 1. (See also Theo-
rem 21.5.)
(b) Show that every plane graph G has a supergraph H which is an eulerian
plane triangulation. (Given a 3-vertex-colouring c of G, one can in fact find an
eulerian plane triangulation H with a 3-vertex-colouring c′ such that G ⊆ H
and c is the restriction of c′ to V (G).)
Section 12.2
12.2.4 See the remark following the proof of Turán’s Theorem (12.7).
12.2.5 See the proof of Turán’s Theorem (12.7).
12.2.7 (a) Assume that G contains no triangle. Choose a shortest odd cycle C in G.
Show that each vertex in V (G)\V (C) is joined to at most two vertices of C. Apply
Mantel’s Theorem (Exercise 2.1.16) to G − V (C), and obtain a contradiction.
12.2.8 (b) Apply the Cauchy-Schwartz Inequality.
12.2.9 (a) Appealing to Exercise 12.2.7, recursively delete edges in triangles until
the resulting graph either (i) is bipartite , or (ii) has at most ⌊ 14 (n − 1)2 ⌋ + 1 edges.
In case (i), conclude by showing that if G is a simple bipartite graph, then any
two vertices in the same part have at least m − 14 (n − 1)2 common neighbours.
Alternatively, if each edge of G lies in a triangle, show that t(G) > n2 /12 and
hence that t(G) ≥ ⌊n/2⌋. If some edge e = xy lies in no triangle, apply induction
to G − {x, y} using (a); observe, also, that G − {x, y} is not bipartite, and hence
that either x or y belongs to a triangle.
12.2.10 (a) A graph contains K2,k if and only if it has k vertices with a pair of
common neighbours. Apply the Pigeonhole Principle.

Section 12.3
12.3.7
(b) Show that any simple graph G with δ ≥ m − 1 contains every tree on m
vertices.
(c) Use induction on n and the fact that any simple graph with δ ≥ m− 1 contains
every tree on m vertices.
12.3.8 (a)(ii) Let (A1 , A2 , . . . , An ) be a partition of [1, sn − 1]. Set Bi := {a +
2sn − 1 : a ∈ Ai } and Ci := Ai ∪ Bi , 1 ≤ i ≤ n, and Cn+1 := [sn , 2sn − 1]. Show
that (C1 , C2 , . . . , Cn ) is a partition of [1, 3sn − 2] with no solution.
12.3.11 A 2-edge colouring of the countably infinite complete graph on a set
U = {u1 , u2 , ...} is said to be canonical if, for any fixed i ≥ 1, all edges ui uj ,
where j > i, have the same colour. Firstly, show that given any 2-edge colouring
c of a countably infinite complete graph K with V (K) = {v1 , v2 , ...}, there exists
a countably infinite subset U = {u1 , u2 , ...} of V such that the restriction of c
to the subgraph K[U ] of K is canonical. Then show that every canonically 2-
edge coloured countably infinite complete graph contains a monochromatic infinite
complete subgraph.
12.3.12 Let z be a vertex on the boundary of the convex hull of V . Each line
L which passes through z but through no other vertex of G partitions V \ {z}
into two subsets, XL and YL . Consider all such lines L for which XL and YL are
nonempty, and apply induction to G[XL ∪ {z}], G[YL ∪ {z}] and G[V \ {z}].
Section 13.1
13.1.1 Correction: A(S) and A(T ) should be replaced by AS and AT .
13.1.3
(a) Apply the Inclusion–Exclusion Formula (2.3), noting that P (Ai ) = 1 − P (Ai ).
(b) Use (a).
13.1.4 Use induction on |I \ S|.

Section 13.2
13.2.2 (a) Use induction on |I|.
13.2.4 Complete graphs have many crossings.
13.2.8 (a) What is the probability that a given linear ordering of the vertices is a
directed Hamilton path?
13.2.11 (a)(i) Use the inequalities of Exercise 13.2.1.
13.2.18 (b) Let G and H be two such graphs. It may be assumed that V (G) =
V (H) = N. Construct an isomorphism θ : G → H inductively, one vertex at a
time. Initially, define θ(1) = 1. Suppose that θ has been defined on a finite subset
S of N. Let i be the least integer not in S. Using the given property, show that
there is a vertex j ∈ N \ T , where T := θ(S), such that θ may be extended to
S ∪ {i}. Show, likewise, that θ−1 may be extended from T ∪ {j} to T ∪ {j, k},
where k is the least integer not in T ∪ {j}.

Section 13.3
13.3.5 Consider a random 2-colouring of V .
13.3.6 Consider a random tournament.

Section 13.4
13.4.2
(b) Apply Markov’s Inequality.
(c) Apply Chebyshev’s Inequality.
13.4.3
(a) Apply Cayley’s Formula (Theorem 4.8).
(b) Use Exercise 13.4.2.
(c) Apply Chebyshev’s Inequality.
Section 14.1
14.1.1 What is the stability number of this graph? Use symmetry to determine
all the maximum stable sets. Can the vertex set of the graph be expressed as the
union of three of them?
Alternatively, in any 3-colouring of a 5-cycle, one colour is assigned to just one
vertex, and the remaining two colours are assigned to two vertices each. Choose
a 5-cycle in the graph and show that no 3-colouring of it can be extended to a
3-colouring of the entire graph. Use symmetry to limit the number of cases to be
checked.
14.1.2 If G is not 2-connected, consider a cutvertex of G and apply induction on
the number of blocks.
14.1.3 (a) If some colour includes no such vertex, show that this colour is not
needed.
14.1.4 If Si is the set of colours used for Gi , i = 1, 2, find a proper colouring of
G by the set S1 × S2 .
14.1.5 (c) Use induction on n.
14.1.6 For the lower bound, apply one of the inequalities in Section 14.1. For the
upper bound, observe that the m-sets containing a fixed element of the n-set S
form a stable set in KGm,n , as do the m-sets contained in a (2m − 1)-subset of S.
14.1.7 Apply Exercise 1.1.11.
14.1.8 This should be classified as a hard exercise.
(a) Consider an odd cycle C of G. How many colours are needed for C and how
many for G − C?
(b) Use induction on n. Show first that G contains a triangle. Consider a vertex
v. If N (v) is a stable set, contract ∂(v) and apply induction. Now consider the
vertices v1 , v2 , v3 of a triangle T and their neighbour sets V1 , V2 , V3 in G − T .
Can there be edges in both G[Vi \ Vj ] and G[Vj \ Vi ]? If there is no edge in
G[Vi ∩ V2 ∩ V3 ], find a 4-colouring of G.
14.1.9 Order the vertices of G according to their colour in a proper χ-colouring
of G.
14.1.10
(a) Order the vertices of G according to their degree.
(b) Let k := max{min {di + 1, i} : 1 ≤ i ≤ n}. Show that dk ≥ k − 1, and apply
Theorem 1.1.
14.1.11 (a) Use two of the inequalities in Section 14.1.
14.1.12 One may assume that G is connected (why?). Consider a DFS-tree T in
G. Recall that the vertices on any single level form a stable set.
14.1.13 (b) Obtain a lower bound on χ(C5 [K3 ]) by considering α(C5 [K3 ]).
14.1.14 (a) Obtain a lower bound on χ(C5 [Kn ]) by considering α(C5 [Kn ]).
14.1.15 (b) Consider a kernel S of D, colour the vertices of S, delete S, and use
induction.
14.1.16 (a) Let D1 and D2 be spanning subdigraphs of D, where the arcs of D1
are the arcs (u, v) of D for which f (u) ≤ f (v), and the arcs of D2 are the arcs
(u, v) for which f (u) > f (v). Show that either χ(D1 ) ≥ m + 1 or χ(D2 ) ≥ l + 1,
and apply the Gallai-Roy Theorem (14.5).
14.1.17 Consider a proper χ-colouring of G, and orient G accordingly.
14.1.23 Consider a proper χ-colouring of G, and a random partition of its colour-
classes into two equal, or almost equal, subsets.
Alternatively, consider all such partitions and apply an averaging argument.
14.1.26 Observe that the neighbours of any vertex can be properly coloured in two
colours. Iteratively select a vertex of sufficiently high degree k (to be determined)
in the subgraph of G induced by the as yet uncoloured vertices, and assign such
a colouring to its neighbours using two new colours each time. Now colour the
remaining subgraph (whose maximum degree is less than k) by the greedy heuristic,
again using fresh colours.
14.1.28
P
(a) Show that a spanning branching forest F in D which maximises i ini , where
ni is the number of vertices at level i in F , has the desired property.
(c) For 0 ≤ j ≤ l − 1, set Xj := ∪r≥0 Sk+j+rl . Show that some Xj is not a stable
set, and consider the branching or branchings of F containing x and y, where
(x, y) ∈ D[Xj ].
14.1.29 For G countable, let V = {v0 , v1 , . . .} and set Gi := G[{v0 , v1 , . . . , vi }],
i ∈ N. Denote by Ci the set of proper colourings of Gi in the colours 1, 2, . . . , k.
Define a graph with vertex set ∪{Ci : i ∈ N}, where ci ∈ Ci and cj ∈ Cj are
adjacent if j = i + 1 and ci is the restriction of cj to {v0 , v1 , . . . , vi }. Now apply
König’s Lemma (Exercise 4.1.21). (For G uncountable, see Diestel (2005), p. 201.)

Section 14.2
14.2.1 Consider a critical subgraph.
14.2.3 This exercise is incorrect. The Chvtal graph is not 4-critical. A 4-critical
subgraph of the graph may be obtained by deleting two appropriately chosen edges.
14.2.10 Use Exercise 14.2.9.
14.2.12 (a) Use Theorems 14.10 and 14.7.
14.2.16 Let C := (V1 , V2 , . . . , Vk ) be a k-colouring of G with as few singleton colour
classes as possible, and let C ′ := (V1′ , V2′ , . . . , Vl′ ) be a colouring of G in which each
colour class contains at least two vertices. Form a bipartite graph F := F [X, Y ],
where X := {V1 , V2 , . . . , Vk } and Y := {V1′ , V2′ , . . . , Vl′ }, with Vi adjacent to Vj′ if
and only if Vi ∩ Vj′ 6= ∅. Assume that |V1 | = 1. Show that V1 is connected in F to
some Vi with |Vi | ≥ 3, and modify C accordingly.
14.2.18 Apply Theorems 14.7 and 2.4.

Section 14.3
14.3.1 The fact that Gk is k-chromatic is established by Theorem 14.12. There
are three kinds of edges in Gk . For those in Gk−1 , use induction.
14.3.2 Colour the set of all vertices {i, j} of SGn such that 1 ≤ i ≤ n/2 and
n/2 < j ≤ n. Now apply induction to colour the remaining vertices.
14.3.3 If H is k-colourable, there is an n-subset of S all of whose elements receive
the same colour. Consider the corresponding copy of G to obtain a contradiction.

Section 14.4
14.4.2 To show that C2k+1 is imperfect, use an inequality from Section 14.1.
14.4.3 (a) Use Theorem 9.22 and induction on n.
14.4.4 Use induction on the order of the graph.
14.4.7 This should not be classified as a hard exercise. Simply apply Theo-
rem 14.14.
14.4.8 If G is not perfect, it has a minimally imperfect induced subgraph. Give
a succinct certificate to recognize such a subgraph.

Section 14.5
14.5.3 This should be classified as a hard exercise.
It may be assumed that not all the lists are identical (why?). If G is not regular, find
an appropriate orientation of G and apply Theorem 14.20. If G is regular, consider
an endblock B of G. If B contains adjacent vertices with distinct lists, one of them
is not a cut vertex (why?). Assign it a colour not in the other’s list, delete this
colour (if present) from the lists of its neighbours, delete the vertex from G, find
an appropriate orientation of the resulting graph, and apply Theorem 14.20. If all
the vertices of B have the same list, obtain a ∆-list-colouring of G by colouring
separately the block B and the union of the remaining blocks of G.
14.5.5 Denote the bipartition of Kn,nn by (X, Y ), where |X| = n and |Y | = nn . To
show that this graph is not n-list-colourable, assign disjoint n-sets to the vertices
of X and appropriate lists to the vertices of Y .
14.5.7 Consider any choice of colours from the lists in one part of the bipartition
of Kn,n . Let T be the set of colours chosen. Show that |T | ≥ k.
14.5.8
(a) Use induction on k. Consider two cases, according to whether or not there
exist two consecutive vertices on the k-path whose lists are distinct.
(b) Use induction on n. To show that a (bipartite) theta graph with at least two of
its three paths of length three or more is not 2-list colourable, assign the same
list to all the vertices on the third path and appropriate lists to the other two
paths. To show that a 2-list-colourable graph with no vertices of degree less
than two and at least two vertices of degree at least three is a theta graph, find
a list assignment for an even cycle C such that a given vertex of C is assigned
the same colour in every compatible list colouring. Deduce that any two cycles
in a 2-list-colourable graph have at least two vertices in common.
14.5.9 For v ∈ V , let c(v) be a colour chosen at random from L(v). For e ∈ E,
consider the ‘bad’ event Ae that the ends of e receive the same colour.
14.5.10 (a) Show that an interval graph G may be regarded as the intersection
graph of n subpaths P1 , P2 , . . . , Pn of a path P , where Pn is simply the terminal
vertex of P . Apply induction on n to obtain the desired orientation of G.

Section 14.6
14.6.2 (b) Observe that the subgraph of D induced by A(D) \ A(D′ ) is an even
digraph, and apply Exercise 2.4.2.
14.6.5 (a) The converse of an eulerian orientation of G is eulerian. Show that the
two have the same sign.
14.6.6 (a) Apply Corollary 14.21, and appeal to Exercise 21.4.5.

Section 14.7
14.7.3 (a) Apply induction, using the recursion P (G, x) = P (G\e, x)−P (G / e, x).
14.7.4 (b) Compute the coefficient of xn−1 in P (G, x), and apply Exercise 14.7.3(a).
14.7.5 Use the recursion P (G, x) = P (G \ e, x) − P (G / e, x) and Exercise 14.7.4a.
14.7.9 Prove that (−1)n−1 P (G, x) > 0 when G is connected and 0 < x < 1. Show
this for a tree, then use the recursion P (G, x) = P (G \ e, x) − P (G / e, x).
14.7.10 Consider the expansion of P (G, x) in terms of chromatic polynomials of
complete graphs.
14.7.11 Denote by f (G) the number of acyclic orientations of G. Show that
f (G) = f (G \ e) + f (G / e). Now proceed by induction on m.
14.7.12 Consider the set of all k n functions from V (G) into {1, 2, . . . , k}. For
e ∈ E, denote by Ae the set of such functions which assign the same value to both
ends of e. For S ⊆ E, how many functions assign the same value to both ends of e
for all e ∈ S? Express the set of all proper vertex colourings of G in terms of the
sets Ae , and apply the Inclusion-Exclusion Formula (2.3).
Section 15.1
15.1.1 Look for an embedding in which all faces have degree five or six.

Section 15.3
15.3.1 (a) Check the theorems in Chapter 14.

Section 15.4
15.4.2 (a) If δ ≥ 3, use Exercise 10.1.5. If there is a vertex of degree less than
three, delete it and use induction.
15.4.3 See Exercise 14.1.13.
15.4.6 Construct a maximal stable set S in G by means of a greedy algorithm,
adding at each stage a vertex at distance two from the current stable set. Show
that there is a tree T in G containing S with at most 2|S| − 1 vertices.
15.4.7 One implication is easy. For the other, consider a largest induced subgraph
whose vertex set can be partitioned into stable sets of cardinality two, and make
use of the ceiling function ⌈n/2⌉.
Section 16.1
16.1.3 Show that G has no stable sets of size greater than two. Use this observation
to establish a bijection between matchings in G and colourings of G.
16.1.7
(a) Apply Theorem 2.9 to the spanning 1-regular subgraph G[M ] of G.
(b) Let S denote the vertex set of a 5-cycle of the Petersen graph. Show that
|M ∩ ∂(S)| =
6 3. Use part (a) and this property of the Petersen graph.
16.1.8
(a) Apply Exercise 16.1.7a.
(b) Look at the three cut edges.
(c) Construct a (2k + 1)-regular simple graph in which there is a (stable) set of
2k − 1 vertices whose deletion results in a graph with 2k + 1 components, each
of which is odd. (One may construct such a graph whose connectivity is 2k−1.)
16.1.9 Consider the subgraph G[M △ M ∗ ].
16.1.10 Consider the sum of the lengths of the edges of the matching.
16.1.11 If G has a perfect matching M , show that the second player has a
response to every move of the first player. If G has no perfect matching, consider
a maximum matching M of G. By appealing to Berge’s Theorem (16.3), devise
a winning strategy for the first player in which he starts by picking a vertex not
covered by M .
16.1.12 Let M be a maximum matching of G. If |M | < δ, show that any two
vertices not covered by M are connected by an M -augmenting path of length three,
and derive a contradiction.
16.1.13
(a) Correction. This part should read:
Show that G has no M -alternating cycle, and that the first and last
edges of every maximal M -alternating path belong to M .
(b) Show that the two ends of any maximal M -alternating path have degree one
in G.
(c) There is an example on six vertices.
16.1.14
(a) Use Berge’s Theorem (16.3).
(b) The hypothesis implies that every vertex is incident with a link. Use part (a).
(c) Let MA be a maximum matching of G which (i) covers every vertex in A and
(ii) subject to (i), covers as many vertices of B as possible. Define MB similarly.
Assume that there is a vertex u in A which is not covered by MB . Derive a
contradiction by considering the component of G[MA △ MB ] containing the
vertex u.
16.1.15 (b) Let G[X, Y ] be a bipartite graph and let xy be an edge of G with
x ∈ X and y ∈ Y . Suppose that neither x nor y is essential. Then there exist
maximum matchings Mx and My of G such that Mx covers x but not y, and My
covers y but not x. Show that no component of G[Mx △My ] contains both x and y.
Let Px and Py be the components of G[Mx △ My ] containing x and y, respectively.
Obtain a contradiction by considering the subgraph (Px ∪ Py ) + xy. (Thus, in a
bipartite graph, at least one end of each edge is essential.)
16.1.16 Let G denote the complement of the comparability graph associated with
the partially ordered set ({1, 2, . . . , n}, ≺). (Thus ij is an edge of G if and only if
neither i ≺ j nor j ≺ i.) By induction on n, show that (i) the minimum of d1 +d2 is
n−α′ and (ii) there exists a maximum matching M of G and a feasible schedule for
processing all the jobs in which the jobs corresponding to the vertices not covered
by M are processed on the first machine and the pairs of jobs corresponding to
the edges of M are processed on the second machine.

Section 16.2
16.2.4 Consider G − v, where v is an essential vertex.
16.2.7 (a) Consider the set S := X \ K, where K is a minimal covering of G.
16.2.10 How many edges can be incident to k − 1 vertices of G?
16.2.11 (a) Consider random proper k-colourings of G[X] and G[Y ]. Call an edge
e ∈ [X, Y ] bad if the two ends of e receive the same colour. What is the probability
of this event? Deduce that the expected number of bad edges is less than one, and
apply Markov’s Inequality (13.4) with t = 1.
16.2.12 Let M be a maximum matching. For each vertex v not covered by M ,
pick an edge incident to v. and let F be the set of these edges. By considering the
set M ∪ F , deduce that α′ + β ′ ≤ n. To show that α′ + β ′ ≥ n, consider a minimum
edge covering of G, and a maximum matching in the subgraph of G induced by
that edge covering.
16.2.13 For S ⊆ X, consider the bipartite adjacency matrix of G[S ∪N (S)]. Show
that the rows of this matrix are linearly independent over GF (2) by considering
the inner product of any sum of its rows with any single row of that sum. Deduce
that this matrix has at least as many columns as rows, and apply Hall’s Theorem
(16.4).
16.2.14 Use Proposition 1.3.
16.2.15 Apply Hall’s Theorem to a suitable bipartite graph constructed from G.
16.2.17 For k < n, one may assume that the union of any k of the cycles has
at least k + 1 vertices (why?). Deduce that (V (Ci ) : 1 ≤ i ≤ n) has an SDR
(vi : 1 ≤ i ≤ n), and consider suitable arcs ai ∈ A(Ci ), 1 ≤ i ≤ n.
16.2.19 Construct a bipartite graph G[X, Y ] in which X and Y are the sets of
rows and columns, respectively, of Q, and row i is joined to column j if and only if
the entry qij is positive. Show that G has a perfect matching. Now use induction
on the number of nonzero entries of Q.
16.2.24
(a) Suppose that there is a proper subset of X which is tight. Using the fact that
G is connected, find an edge which is not contained in a perfect matching and
conclude that G is not matching-covered. Conversely, if G is not matching-
covered, there exists an edge xy with x ∈ X and y ∈ Y such that G − {x, y}
does not have a perfect matching. Using Hall’s Theorem (16.4), show that
there is a proper subset of X which is tight.
(b) Similar to the second part of (a).
16.2.25 Use Exercise 16.2.24 and induction on the number of vertices.
16.2.27 First show that part (b) follows from part (a).
Prove part (a) by induction on n by proceeding as follows. Let x be the vertex in
X of degree two, and let xy1 and xy2 be the two edges incident with x. Consider
two cases depending on whether or not y1 = y2 .
If y1 = y2 , the two edges xy1 and xy2 are multiple edges. Apply the induction
hypothesis to H := G − {x, y1 }.
If y1 6= y2 , consider the graph H := (G − x)/{y1 , y2 }. Denote the vertex of H
resulting from identifying y1 and y2 by y. If one of y1 and y2 has degree two in
G, the graph H is a cubic bipartite graph on 2n − 2 vertices. Otherwise, vertex y
has degree four in H. Let e1 , e2 , e3 , and e4 denote the four edges incident with y
in H. Apply the induction hypothesis to H \ ei , 1 ≤ i ≤ 4.

Section 16.3
16.3.5 Let M be a maximum matching of G and let U denote the set of vertices
not covered by M . Since v is essential, α′ (G−v) = α′ (G)−1, and hence v is covered
by M . Let uv be the edge of M incident with v. Then observe that M ′ := M \{uv}
is a maximum matching of G−v and U ′ := U ∪{u} is the set of vertices not covered
by M ′ in G − v. Use the definition of a barrier (as applied to G − v) and the fact
that o(G − (B ∪ {v})) = o(G − v − B) to deduce that B ∪ {v} is a barrier of G.
16.3.6 If G is a connected graph without essential vertices, then it follows from
Lemma 16.10 that the empty set is a barrier of G. Thus, we only need to consider
disconnected graphs. Observe that if G1 , G2 , . . . , Gk are the components of G, then
every maximum matching of G is the union of maximum matchings of the Gi .

Section 16.4
16.4.1 If o(G − v) = 1 for all v ∈ V , consider the neighbour y of a leaf x and
apply induction to G − {x, y}.
16.4.4 See the hint to Exercise 16.4.5.
16.4.5 More generally, let us consider the problem of finding necessary and suf-
ficient conditions under which a graph G has a matching which covers all the
vertices of a specified subset X of V . Obtain a graph H from G as follows:
⊲ if n is odd, adjoin an isolated vertex to G;
⊲ for each pair of nonadjacent vertices, both not in X, add an edge joining them.
1. Show that there is a matching in G covering all vertices in X if and only if H
has a perfect matching.
2. By applying Tutte’s Theorem (16.13), deduce that G has a matching covering
all vertices in X if and only

o(G[X \ S]) ≤ |S| for all S ⊆ V

3. Now suppose that G[X] is bipartite. Strengthen the statement in (b) to show
that there is a matching covering all vertices in X if and only if the number of
isolated vertices of G[X \ S] is at most |S| for all S ⊆ V .
The above results may be adapted to solve this exercise, as well the previous
exercise.
16.4.6 Let H be a maximal spanning supergraph of G whose matching number
is the same as that of G. Use the proof technique of Theorem 16.13 to show that
if U is the set of vertices of degree n − 1 in H, then H − U is a disjoint union of
complete graphs.
16.4.7
(a) By hypothesis, G has a perfect matching. Therefore, for any barrier B of G,
we have o(G − B) = |B|. Firstly, suppose that vertices x and y belong to
some barrier B. To show that G − {x, y} has no perfect matching, observe that
(G − {x, y}) − (B \ {x, y}) = G − B. Conversely, suppose that x and y are two
vertices such that G − {x, y} has no perfect matching. Show that if B ′ is any
barrier of G − {x, y}, then B := B ′ ∪ {x, y} is a barrier of G.
(b) Observe that an edge xy of G is contained in some perfect matching of G if
and only if G − {x, y} has a perfect matching.
16.4.8 By contradiction. Suppose that an edge xy is not in any perfect matching.
By Exercise 16.4.7(b), there exists a barrier of G which contains both x and y.
Let B be such a barrier and let G1 , G2 , . . . , G|B| be the odd components of G − B.
From the hypothesis that G is 3-regular and the fact that x and y are adjacent,
deduce that |∂(B)| ≤ 3|B| − 2. Now, derive a contradiction using the hypothesis
that G has no cut edges.
16.4.10
(a) Use Exercise 16.4.7.
(b) Use counting arguments similar to the ones employed in the proof of Petersen’s
Theorem (16.14) and Exercise 16.4.8.
16.4.11
(a) Consider a spanning tree T of G with the least number of leaves. If n ≥ 4,
show that T has a leaf x adjacent to a vertex y of degree two. Apply induction
to G − {x, y}.
Alternatively, let G be a connected claw-free graph on an even number of
vertices, and let S ⊆ V . Form a simple bipartite graph H := H[S, T ], where
T is the set of odd components of G − S, there being an edge of H joining a
vertex of S and an odd component of G − S if and only there is such an edge
in G. Show that dH (v) ≤ 2 for all v ∈ S. Using Exercise 16.3.2, deduce that
|T | ≤ |S|, and apply Tutte’s Theorem.
16.4.13 Let G be a 2-connected graph with a perfect matching M , and let
X := {x1 , x2 , . . . , xk } be a maximal barrier of G. Then G − X has k components,
G1 , G2 , . . . , Gk , each of which is hypomatchable. For 1 ≤ i ≤ k, shrink V (Gi ) to
a single vertex yi and set Y := {y1 , y2 , . . . , yk }. Now delete the edges of G with
both ends in X to obtain the bipartite graph H[X, Y ]. Since, by Exercise 16.4.7,
no edge with both ends in X can be in M , it follows that the restriction of M to
E(H) is a perfect matching of H. Using the fact that each Gi is hypomatchable,
show that any perfect matching of H may be extended to a perfect matching of
G. Now use Exercise 16.1.13.
16.4.14
(a) By Exercise 16.4.13, the graph either is disconnected or has a cut vertex.
Consider its blocks, and apply induction.
(b) Construct such a graph recursively, adding two vertices at a time.
Alternatively, consider the (unique) connected simple graph on 2n vertices
with 2n − 1 distinct degrees.
16.4.16 (b) Let G be a 2k-regular graph with V = {v1 , v2 , . . . , vn }; without loss
of generality, assume that G is connected. Let W be an Euler tour in G. Form
a bipartite graph H[X, Y ], where X = {x1 , x2 , . . . , xn } and Y = {y1 , y2 , . . . , yn },
by joining xi to yj whenever vi immediately precedes vj on W . Show that H is
1-factorable. Deduce that G is 2-factorable.
16.4.17 Apply Exercise 16.4.16b.
16.4.18 Let G be a triangulation. Apply Petersen’s Theorem (16.14) to G∗ .
16.4.19 Let G be a 2-connected 3-regular graph on four or more vertices, and
let M be a perfect matching of G. Then F := G \ M is a 2-factor of G. Orient
each cycle of F as a directed cycle, and use this orientation to associate a path of
length three with each edge of M .
16.4.20 (a) If d(v) ∈ L(v) for all v ∈ V , let f (v) = d(v) for all v ∈ V . If
/ L(y) for some vertex y, let (x, y) ∈ A. Set G′ := G \ xy, D′ := D \ (x, y),
d(y) ∈
L (x) := L(x) \ {d(x)} and L′ (v) := L(v), v ∈ V \ {x}, and apply induction.
′
Section 16.5
16.5.3 Show that any M -alternating path which enters T from G − V (T ) must
enter via a non-matching edge incident with a blue vertex of T . Consider the
successive vertices of this path.
16.5.6 Denote the k parts of G by {x1 , y1 }, {x2 , y2 }, . . . , {xk , yk }. For v ∈ V , let
L(v) be a list of k colours assigned to v. The assertion is trivial for k = 1, so
assume that k ≥ 2. If L(xi ) ∩ L(yi ) = ∅, 1 ≤ i ≤ k, apply Hall’s Theorem (16.4)
to an appropriately defined bipartite graph H[V, C], where C := ∪{L(v) : v ∈ V }.
If not, proceed by induction on k.
Section 17.1
17.1.2 Find a decomposition of Kn,n into perfect matchings.
17.1.3 How many edges can have the same colour?
17.1.5
(a) Gi is clearly connected and cubic. Can it have a cut edge?
(b) In a 3-edge-colouring of G, can e1 and e2 receive different colours? See Exer-
cise 16.1.7a.
17.1.6 (a) In a 3-edge-colouring of G, can two edges of a 3-edge-cut receive the
same colour? See Exercise 16.1.7a.
17.1.8 How many edges are needed to colour the edges of a Hamilton cycle?
17.1.10
(a) Can two classes be taught in the same period by the same teacher?
(b) Consider two periods in which the numbers of classes differ by at least two.
Can the schedule be rearranged?
17.1.11 A regular bipartite graph of positive degree has the same number of
vertices in each part (see Exercise 1.1.9).
17.1.14 Show that χ′ (K8 ) = 7.
17.1.15 What is a vertex colouring of a line graph? What is a clique in a line
graph? Express these concepts in terms of the root graph.
17.1.16 If intersecting triples have distinct colours, how many colours are needed
to colour all triples?
17.1.17
(a) Place all the vertices but one in a circle, and the remaining one at the centre
(as in the hint to Exercise 2.4.6a).
(b) See Exercise 17.1.3a.
17.1.18 Use the proof technique of Theorem 17.2.
17.1.19 Use induction on the number of vertices. Let G be a 3-edge-coloured
complete graph with the given property, and v be a vertex of G. If some colour
class of G − v does not induce a connected spanning subgraph, look at the colours
of the edges connecting two components in the subgraph induced by that colour.
Can both of the other colours be present?
Section 17.2
17.2.1 (a) How many edges can there be of the same colour? (Compare with
Exercise 17.1.3.)
17.2.2
(a) How many edges of the same colour can there be in H?
(b) Look at the constructions described in Exercise 17.2.1b.
17.2.3 This should be classified as a hard exercise.
17.2.4 (a) Consider a (∆ + 1)-edge-colouring of G.
17.2.5 Find a suitable representation of the vertices of V (P  K3 ) by pairs of
elements from {1, 2, 3, 4, 5}, and use it to determine a 5-edge-colouring of P  K3
in the colours 1, 2, 3, 4, 5.
17.2.8 (a) Adapt the proof technique of Lemma 17.3.
17.2.9 A forest has a vertex of degree zero or one. Bearing this in mind, apply
Lemma 17.3.
17.2.10 (a) Can G[Mi ∪ Mj ] be disconnected?
17.2.11 How are 3-edge-colourings and Hamilton cycles related in cubic graphs?
17.2.12 Show, by contradiction, that no regular self-complementary graph belongs
to Class 1. The converse of this assertion is an unsolved problem. (See A.P. Wojda.
A note on the colour class of a self complementary graph. Discrete Math. 213
(2000), 333–336.)
17.2.13 Use the proof technique of Lemma 17.3.
17.2.14 Use the proof technique of Lemma 17.3.

Section 17.3
17.3.3 (a) Use the result in Exercise 16.1.7a. (It implies that if (M1 , M2 , M3 ) is a
3-edge-colouring of a cubic graph G, then, for any subset X of V , the parities of
|∂(X) ∩ Mi |, 1 ≤ i ≤ 3, are all equal to the parity of |∂(X)|.)
17.3.4 (a) Adapt the ideas in the proof of Tait’s Theorem (11.4).
17.3.5 (a) Use the fact that if X is the vertex set of a 5-cycle of Gk , and if M ′ is
any perfect matching, then |M ′ ∩ ∂(X)| = 1 or 5 (see the hint to Exercise 16.1.7b).

Section 17.4
17.4.2 The graph H has a perfect matching M (why?). If H is triangle-free, show
that the graph obtained by duplicating the edges of H \ M is 5-edge-colourable
(consider its line graph). If H has a triangle, contract it to a single vertex and use
induction.
Section 18.1
18.1.3 Could a hypothetical Hamilton cycle in the Meredith graph include both
of the edges joining two K3,4 -subgraphs?
18.1.4 (b) A nontrivial path has just two ends.
18.1.5 Why is the Herschel graph (Figure 18.1b) nonhamiltonian? Modify this
graph.
18.1.7 (b) A Hamilton-connected graph is hamiltonian. The Petersen graph is
nonhamiltonian.
18.1.8 Orient the two triangles of a triangular prism in the same sense.
18.1.10 In a path partition of H, where are the ends of the constituent paths to
be found?
18.1.11 (b) Is a graph in which every vertex-deleted subgraph is traceable path-
tough?
18.1.12 Consider an edge uv and an automorphism α of G such that α(u) = v.
18.1.14 Use Exercises 18.1.2 and 18.1.13.
18.1.15 (b) Recall that the Herschel graph is bipartite.
18.1.16 Use Exercise 18.1.14.
18.1.17 Use the fact that the Petersen graph is hypohamiltonian (Exercise 18.1.16a).
18.1.18
(a) If G has no k-cycle, show that if vx ∈ E then vy ∈
/ E for some y ∈ V . Deduce
that d(v) ≤ (n − 1)/2.
(b) Let C be a Hamilton cycle of G with a specified sense of traversal. Show
that there are consecutive vertices x, y of C such that d(x) + d(y) ≥ n + 1.
Deduce that G − v is hamiltonian for some v ∈ V , and examine the two cases
d(v) ≤ (n − 1)/2 and d(v) ≥ n/2.
18.1.19 (b)
(i) Consider the cycle space of G.
(ii) Bound the number of cycles that include at
most two chords of the Hamilton
cycle. Obtain the sharper result: if n ≤ r+2
2 + 1, then m ≤ n + r − 1.
18.1.20 If D′ := D − {x, y} is strong, apply Camion’s Theorem and look at
Exercise 18.1.18a. If not, y dominates a vertex y ′ in the initial component of D′
and x is dominated by a vertex x′ in the terminal component of D′ . Show that
there exist directed (y ′ , x′ )-paths in D′ of all lengths l′ , 1 ≤ l′ ≤ n − 3.

Section 18.2
18.2.2 Start with a simple nonhamiltonian 2-connected cubic planar graph.
18.2.3 (b) Show that V (Gi ) \ V (Gi−1 ) is a stable set of Gi . Deduce that the
circumference of Gi is at most twice the circumference of Gi−1 , i ≥ 1.

Section 18.3
18.3.2 If there are two different closures, consider the first edge added in the
formation of one of them that is not an edge of the other. Derive a contradiction.
18.3.3 (a) Use Exercise 18.1.6a and Theorem 18.9.
18.3.5 (a) Let P be a longest path in G. If P has length l < 2δ, show, using the
proof technique of Theorem 18.9, that G has a cycle of length l + 1. Now use the
fact that G is connected to obtain a contradiction.
18.3.6 Construct a certain graph H from G, with α(H) = α(G) and κ(H) =
κ(G) + 1. Apply Theorem 18.10.
18.3.7 Assume false. Consider a cycle C through as many vertices of X as possible,
and a path P from X \ C to C. Find two vertices (one on P and one on C) whose
degree sum is less than n.
18.3.8
(a) Apply Theorem 18.1.
(b) Apply Theorem 18.9.
18.3.9 Form a new graph H from G by adding edges between all pairs of vertices
of X. Show that H is hamiltonian if and only if G is hamiltonian.
18.3.10
(a) Apply Corollary

18.8.
(b) If m = n−1
2 + 1, when are G and c(G) identical?
18.3.11
(a) Bound the minimum degree δ.
(b) A ladder is a (2 × n)-grid. Modify such a graph.
18.3.13 Use induction on the length of Q.
18.3.15
(a) Set P := v0 v1 v2 . . . vℓ , where v0 := x, vk := y1 and vℓ := y, and define

X := {vi : xvi+1 ∈ E} and Y := {vi : vi y ∈ E}

Consider the cases X ∩ Y 6= ∅ and X ∩ Y = ∅. In the latter case, show that C

contains every vertex of (X ∪ Y ∪ {y}) \ {vk−1 }.
(b) (ii) Apply (b)(i) to the graph obtained by adding a set S of k − 1 vertices
and joining them to one another and to both x and y. Consider two cases,
according to whether or not the resulting cycle meets S.
Section 18.4
18.4.1
(a) (i) Apply Smith’s Theorem (18.13).
(ii) See Exercises 17.2.10 and 17.2.11.
(b) See Exercise 18.1.2.
18.4.5 Apply the Lollipop Lemma (18.11).
18.4.6
(a) This is incorrect. The theorem published by A. Kotzig (Aus der Theorie der
endlichen regulären Graphen dritten und vierten Grades. Časopis Pěst. Mat.
82 (1957), 76–92) is somewhat weaker than stated, and Exercise 18.4.5a should
read as follows:
Let G be a simple cubic graph. Show that the line graph of G admits a
Hamilton decomposition if and only if G is hamiltonian.
A Hamilton cycle C in a graph G is a dominating eulerian subgraph of G, and
thus gives rise to a Hamilton cycle in L(G) (see Exercise 3.3.8). Show how to
obtain from C two edge-disjoint Hamilton cycles in L(G). (Orienting C as a
directed cycle and assigning arbitrary orientations to the remaining edges of
G might help.)
Conversely, consider a decomposition of L(G) into two Hamilton cycles. Each
vertex of L(G) corresponds to an edge of G, and there are two distinct ways in
which a cycle can traverse a vertex of L(G). Show that the edges of G which
correspond to the vertices of L(G) traversed by the two Hamilton cycles in one
of these ways induce a Hamilton cycle in G.
(b) This part remains valid.
18.4.7 (b) The medial graph of a plane even graph is 2-face-colourable. Consider
the two colour-classes of faces.
18.4.8 (a)(i) Use induction on the distance between e and f .
Alternatively, prove the result for graphs G of even order only by constructing an
appropriate bipartite graph from G and arguing as in Exercise 18.4.2.
18.4.9 Consider first the case k = 4. Reduce the problem to 4-regular graphs.
18.4.10 Subdivide each edge of C − S once.
18.4.11 (a) Use the proof technique of the Lollipop Lemma (18.11).
18.4.12 (a) Use the proof technique of the Lollipop Lemma (18.11).

Section 18.5
18.5.1 Consider a longest path P in G, and define X as in Pósa’s Lemma (The-
orem 18.19).
Section 19.1

Section 19.2

Section 19.3

Section 19.4
19.3.1 Let D := D(x, y) be a digraph. Obtain the digraph D′ := D′ (x, y) from D
by adding k new arcs from y to v for each vertex v ∈ V \ {x, y}. Show that D has k
arc-disjoint directed (x, y)-paths if and only if D′ has k arc-disjoint x-branchings.
19.3.4 Adjoin a new vertex x to D and join it to each v ∈ V by k − d− (v) arcs.
Denote the resulting digraph by D′ . Show that D′ has k arc-disjoint spanning
x-branchings.

Section 19.5
19.4.9
1. See Figure B.1.

Fig. B.1. Two planar digraphs with ν = 1 and τ = 2

2. Clearly, we may assume that D is strong. Suppose the result false, and choose
a counterexample D which is maximal, that is, such that the addition of any
arc joining two nonconsecutive vertices on a face results in a digraph with
two disjoint directed cycles. Let C be a facial directed cycle of D (such a
cycle exists by Exercise 19.4.8). Show that, for any vertex v of C, there exists
a facial directed cycle Cv of D which meets C only at v. Since ν = 1, for
u, v ∈ V (C), u 6= v, the cycles Cu and Cv must have a vertex in common.
Taking the planarity of D into consideration, analyse the possible intersection
pattens amongst the cycles Cv , v ∈ V (C), to obtain a contradiction.
19.4.11 Let C be a maximum laminar family of arc-disjoint directed cycles. Obtain
a graph G whose vertex set is C by joining two elements C and C ′ by an edge if they
have a vertex of D in common. Using the fact that D is 2-diregular and planar,
deduce that G is planar. By the Four-Colour Theorem (11.2), G is 4-colourable.
It follows that D has at least 14 |C| vertex-disjoint directed cycles. Thus |C| ≤ 4k.
By the Lucchesi-Younger Theorem (19.9), D has a set of |C| arcs whose deletion
destroys all directed cycles. By selecting one end of each of these arcs, we obtain
a set of at most 4k vertices whose deletion destroys all directed cycles.
20.4.1 (a) Let Q be a square submatrix of K. Use induction on the order of Q. If
each column of Q contains two nonzero entries, then det Q = 0. Otherwise, expand
det Q about a column with exactly one nonzero entry, and apply the induction
hypothesis.
20.4.3 Here is a proof of unimodularity.
Proof Let Q be a full square submatrix of B (one of order n − 1). Suppose that
Q = B|T1 . We may assume that T1 is a spanning tree of D since, otherwise,
detQ = 0 by Theorem 20.6. Let B1 denote the basis matrix of B corresponding to
T1 . Then (Exercise 20.2.1(b))

(B|T1 )B1 = B.
Restricting both sides to T , we obtain

(B|T1 )(B1 |T) = B|T.

Noting that B|T is an identity matrix, and taking determinants, we get

det(B|T1 ) det(B1 |T) = 1. (B.1)

Both determinants in (B.1), being determinants of integer matrices, are them-
selves integers. It follows that det(B|T1 ) = ±1. 

20.5.1 Let W be the minimum power of an (x, y)-flow in D, and let f be a flow
of power W . Being a flow, f satisfies Kirchhoff’s Law (K1). To show that f is
a current flow in D, we need to show that f also satisfies (K2); in other words,
we need to show that f is a tension. Suppose that this is not the case. Then f
is not orthogonal to the signed incidence vector fC of some cycle C (because the
tension space B is the orthogonal complement of the circulation space C, which is
generated by the vectors {fC }). Thus there is a cycle C, and a sense of traversal
of C, such that f + (C) − f − (C) < 0. But then the flow f + ǫfC , for ǫ sufficiently
small, has power less than W , a contradiction.
To establish uniqueness, let f and g be two electrical currents of value I from
x to y. Then f − g is an electrical current of value zero, so is both a circulation in
D and a tension in D. By the orthogonality of C and B, it follows that f − g = 0,
that is, f = g.
20.6.1 See Figure B.2.

20.6.4 Show, first, that in any perfect rectangle the smallest constituent square
is not on the boundary of the rectangle. Now suppose that there is a perfect cube,
and consider the perfect square induced on the base of this cube by the constituent
cubes.
 16
25 28
9 7 19 22
24
5
5
11
6
36 33
23 25
17

(a) (b)

Fig. B.2. Solution to Exercise 20.6.1

20.6.5
(a) A Fibonacci tiling is shown in Figure B.3.

5
8
3 2
21

13

Fig. B.3. Fibonacci tiling of the plane

(b) The Fibonacci tiling is not perfect because there are two squares of size one.
To remedy this situation, first obtain, by blowing up the Fibonacci tiling 112
times, a dissection in which there is one square of size 112Fi , for i = 1, 2, 3, . . ..
In this dissection there are two squares of size 112, but the sizes of all others
squares are greater than 112 and distinct. Now replace one of the squares of
size 112 by the perfect square shown in Figure 20.10.
21.4.7 By Theorem 21.17, there are two edge-disjoint spanning trees, T and T ′ .
Let X be the set of odd vertices of T . By Exercise 3.3.5b, there are |X|/2 edge-
disjoint paths in T ′ connecting the elements of X in pairs. The union of T and
these paths is a connected even spanning subgraph.
21.2.9 (a) Necessity follows on taking X as the set of vertices with indegree k and
Y as the set of vertices with indegree l. To prove sufficiency, construct a network
N by forming the associated digraph of G, assigning unit capacity to each arc,
and regarding the elements of X as sources and the elements of Y as sinks. By
Theorem ???, there is a flow f in N (which can be assumed integral) in which the
supply at each source and the demand at each sink is |k − l|. The f -saturated arcs
induce a (k, l)-orientation on a subgraph H of G. A (k, l)-orientation of G can now
be obtained by giving the remaining edges an eulerian orientation.
21.2.13 Let S be any subset of A. By Exercise 21.2.12, the number of circulations
in the spanning subdigraph with arc set S is k m−n+c(S) . Thus the number of circu-
lations whose supports are included in S is k m−n+c(S) . Equivalently, the number
of circulations which take value zero on all the arcs of A(D) \ S is k m−n+c(S) .
Since nowhere-zero circulations are those circulations which take the value zero on
no arc, we obtain the desired result by applying the Inclusion-Exclusion Formula
(2.3).
21.3.2
(a) The proof is straightforward if the degree of v is two. So, suppose that d(v) = 4
or d(v) ≥ 6. The required assertion is equivalent to showing that there exists
a pair of edges incident with v such that there is no 5-edge cut of G which
contains both of them. Towards this end, let F denote the set of all subsets
of ∂(v) such that every member of F is contained in a 5-edge cut of G. Let F
be a maximal member of F and let X be a subset of V (G), with v ∈ X and
|∂(X)| = 5. We may assume that F 6= ∅; otherwise, any two edges incident
with v would serve the purpose. First show that F is a proper subset of ∂(v).
Then let f be an edge in F and let e is an edge in ∂(v) \ F and consider any
edge cut ∂(Y ) of G containing {e, f }. Show that d(Y ) 6= 5. (L.M. de
Almeida e Silva)
(b) Let G be a graph with no 1- or 3-edge cuts such that (i) G has no 3-flow,
and (ii) subject to (i), v(G) + e(G) is as small as possible. Show that G is a
4-edge-connected 5-regular graph.
21.3.3
21.3.4
(a) A 2-edge-connected graph with a non-trivial 2-edge cut has at least two edges.
The only such graph with exactly two edges has two vertices with two edges
joining them. Verify the statement in this case. To prove the statement in
general, apply induction by contracting and deleting edges not belonging to
∂(X). (Note that, in order to apply induction using deletions and contractions,
it is necessary to permit loops and multiple edges.)
(b) The proof is similar to that of (a).
21.5.2 First apply Exercise 21.3.1 to show that there exists a 2-edge-connected
cubic graph G′ with v(G′ ) + e(G′ ) ≤ v(G) + e(G), such that G′ does not have a
k-flow. Then apply Exercise 21.3.4.
21.4.5 Consider an arbitrary orientation D of G. Suppose that there exists a
vertex x whose outdegree in D is greater than k. Let X be the set of all vertices
reachable from x in D. Show that X includes a vertex y whose outdegree is smaller
than k. Now reverse
P all arcs on some directed (x, y)-path in D[X] and observe that
the quantity v∈V max {0, (d+ (v) − k)} is thereby reduced.