Exercise set 2 – Solutions

1. Let H and H 0 be the two components of T − e and let F ⊆ E(T 0 ) consist of the edges of T 0 with one
endpoint in V (H), the other in V (H 0 ). Since T 0 is connected, F 6= ∅. Furthermore, since T has the unique
edge e joining H and H 0 , F ⊆ E(T 0 ) \ E(T ). T 0 + e contains a unique cycle C of which e is an edge. C
leaves H and enters H 0 via e. In order to complete the cycle, one must use one edge e0 of E(T 0 ) to come
back from H 0 to H. But then e0 ∈ F . It is now clear that for this e0 both T − e + e0 and T 0 + e − e0 are
spanning trees of G. (Note that the cycle C, after coming back to H, may again enter H 0 and subsequently
return back to H. Every time it does so, it has to use two new edges from F . That is, the choice of e 0 is not
always unique.)

2. (a) [if] e is not a cut-edge of G. Then G \ e is connected and hence has a spanning tree T . But then T is a
spanning tree of G too and e ∈ / E(T ).
[only if] Let e be a cut-edge of G with endpoints u and v. The only u, v-path in G is the edge e (since
another u, v-path in G produces a cycle in conjunction with e). Let T be a spanning tree in G. Since T is
connected, T contains a u, v-path which has to be the edge e. Thus e ∈ E(T ).

(b) [if] Let e be a non-loop edge of G and let u be an endpoint of G. We can grow a BFS (or DFS) spanning
tree of G rooted at u and containing the edge e.
[only if] A tree is a simple graph and hence does not contain a loop.

3. Let S be the set of all spanning trees of Kn . By Cayley’s formula |S| = nn−2 . In order to get τ (Kn \ e) for
a given e ∈ E(Kn ) we have to subtract from nn−2 the number k of spanning trees of Kn containing
the particular edge e. Because of symmetry k is independent of the choice of e. Look at the sum
P
σ := T ∈S e(T ). Since each tree T in the sum has n − 1 edges, we have σ = (n − 1)nn−2 . On the
other hand, each edge of Kn is counted k times in the above sum, so that σ = k × e(Kn ) = kn(n − 1)/2.
Equating the two expressions for σ gives k = 2nn−3 . Thus τ (Kn \ e) = nn−2 − k = (n − 2)nn−3 .

4. Let the partite sets of Ks,t be X and Y with X = {x1 , . . . , xs } and Y = {y1 , . . . , yt }. The Q-matrix under
the vertex ordering x1 , . . . , xs , y1 , . . . , yt is then

t 0 ··· 0 −1 −1 ··· −1 
 0 t ··· 0 −1 −1 ··· −1 
 .. .. .. .. .. .. 
 .. .. 
 . . . . . . . . 
 
 0 0 ··· t −1 −1 ··· −1 
Q=
 −1
.
 −1 · · · −1 s 0 ··· 0  

 −1 −1 · · · −1 0 s ··· 0  
 . .. .. .. .. .. .. .. 
 .. . . . . . . . 
−1 −1 · · · −1 0 0 ··· s

Let us choose to delete the first row and the first column of Q to get

t 0 ··· 0 −1 −1 ··· −1
0 t ··· 0 −1 −1 ··· −1
.. .. .. .. .. .. .. ..
. . . . . . . .
0 0 ··· t −1 −1 ··· −1
τ (Ks,t ) = .
−1 −1 · · · −1 s 0 ··· 0
−1 −1 · · · −1 0 s ··· 0
.. .. .. .. .. .. .. ..
. . . . . . . .
−1 −1 · · · −1 0 0 ··· s
 t 0 ··· 0 −t 0 0 ··· 0
0 t ··· 0 −t 0 0 ··· 0
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
0 0 ··· t −t 0 0 ··· 0
τ (Ks,t ) = 0 0 ··· 0 t −1 −1 · · · −1 .
−1 −1 · · · −1 −1 s 0 ··· 0
−1 −1 · · · −1 −1 0 s ··· 0
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
−1 −1 · · · −1 −1 0 0 ··· s

Adding 1/t times each of the first s − 2 rows to each of the last t rows gives:

t 0 ··· 0 −t 0 0 ··· 0
0 t ··· 0 −t 0 0 ··· 0
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
0 0 ··· t −t 0 0 ··· 0
τ (Ks,t ) = 0 0 ··· 0 t −1 −1 · · · −1 .
0 0 ··· 0 −(s − 1) s 0 ··· 0
0 0 ··· 0 −(s − 1) 0 s ··· 0
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
0 0 ··· 0 −(s − 1) 0 0 ··· s

Adding 1/s times each of the last t rows to the (s − 1)-st row then yields

t 0 ··· 0 −t 0 ··· 0
0
0 t ··· 0 −t 0 ··· 0
0
.. .. .. .. .. .. .. .. .
. . . . . . . ..
.
0 0 ··· t −t 0 0 ··· 0
τ (Ks,t ) = 0 0 ··· 0 t/s 0 0 ··· 0 .
0 0 ··· 0 −(s − 1) s 0 · · · 0
0 0 ··· 0 −(s − 1) 0 s · · · 0
.. .. .. .. .. .. .. . . .
. . . . . . . . ..
0 0 ··· 0 −(s − 1) 0 0 · · · s

Finally we add s times the (s − 1)-st row to each of the first s − 2 rows to get

t 0 ··· 0 0 0 0 ··· 0
0 t ··· 0 0 0 0 ··· 0
.. .. .. .. .. .. .. .. .
. . . . . . . . ..
0 0 ··· t 0 0 0 ··· 0
τ (Ks,t ) = 0 0 ··· 0 t/s 0 0 ··· 0 .
0 0 ··· 0 −(s − 1) s 0 ··· 0
0 0 ··· 0 −(s − 1) 0 s ··· 0
.. .. .. .. .. .. .. .. .
. . . . . . . . ..
0 0 ··· 0 −(s − 1) 0 0 ··· s

Now τ (Ks,t ) is the determinant of a lower triangular matrix and hence is equal to the product of the entries
in the main diagonal of this matrix, i.e., τ (Ks,t ) = ts−2 (t/s)st = st−1 ts−1 .

5. We will prove that G is the cycle Cn . Since G \ u is a tree for every u ∈ V (G) and n(G) > 3, G does
not contain multiple edges and loops, i.e., G is simple. If G has more than two components, then deleting
a vertex u in one component leaves other components unaffected, i.e., leaves the graph disconnected, a
 least 2 vertices. Again deleting a vertex u ∈ V (H) leaves a disconnected graph that cannot be a tree. Thus
G has only one component, i.e., G is connected.
By hypothesis G has no cut vertices and hence no cut edges (Exercise 14), i.e., every edge of G lies on
a cycle. If G contains more than one cycle, we can choose two cycles C1 and C2 of G and a vertex
u ∈ V (C1 ) \ V (C2 ). But then G \ u contains the cycle C2 and hence is not a tree. Therefore, G contains
exactly one cycle.

6. (a) If any two of u, v, w lie in different components of G, then the triangle inequality clearly holds. So
assume that all of these three points lie in the same component of G. Let P be a shortest u, v-path and
Q a shortest v, w-path. The u, w-walk consisting of P followed by Q contains a u, w-path R. Hence
d(u, w) 6 |R| 6 |P1 | + |P2 | = d(u, v) + d(v, w).

(b) Let rad G = (u) and diam G = (u0 ) = d(u0 , v 0 ). But then by the triangle inequality diam G 6
d(u0 , u) + d(u, v 0 ) 6 (u) + (u) = 2(u) = 2 rad G.

(c) If r = d = 1, consider P2 . If r = 1 and d = 2, consider P3 . So assume that r > 1. Take G to be

the connected graph that decomposes into a cycle C ∼ = C2r and a path P ∼ = Pd−r+1 such that C and P
share exactly one vertex u which is also an endpoint of P . Let v be the other endpoint of P . (If d = r, then
v = u.) One can readily check that rad G = (u) = r and diam G = (v) = (d − r) + r = d.

7. (a) G = Kn . (Any two vertices of G do not form an independent set.)

(b) The only non-trivial component of G is either a star or a 3-cycle. (Let M := {uv} be a maximal
matching of G. Let n1 (resp. n2 ) be the number of neighbors of u (resp. v) other than v (resp. u), which are
not neighbors of v (resp. u). Also let n3 be the number of common neighbors of u and v. One can easily
check that if at least two of n1 , n2 , n3 are non-zero, then M is not maximal. Also if n1 = n2 = 0, we must
have n3 = 1.)

(c) The only non-trivial component of G is a star. (Every edge of G is incident to the vertex in a minimum
vertex cover.)

(d) G is an edge. (Every vertex of G is incident to the edge in a minimum edge cover.)

8. (a) Let G be an X, Y -bigraph.

[if] By hypothesis |N (S)| > |S| for any S ⊆ X; so G has a matching that saturates X, i.e., |Y | > |X|. By
reversing the roles of X and Y one can similarly prove that |X| > |Y |.
[only if] A perfect matching of G saturates both X and Y and hence by Hall’s theorem |N (U )| > |U | for
every U ⊆ X and also |N (V )| > |V | for every V ⊆ Y . Given S ⊆ V (G) one can write S as the disjoint
union U ∪ V , where U = S ∩ X and V = S ∩ Y . The neighborhoods of U and V are also disjoint and so
|N (S)| = |N (U )| + |N (V )| > |U | + |V | = |S|.

(b) The complete graph K2n+1 , n ∈ N, satisfies


0 if S = ∅,
|N (S)| = 2n if |S| = 1,

2n + 1 if |S| > 2,

that is, |N (S)| > |S| for any S ⊆ V (K2n+1 ). Since K2n+1 contains an odd number of vertices, it cannot
have a perfect matching.

9. Let S be a maximum independent set in G (so that |S| = α(G)) and let T := V (G) \ S. Since S is
P P
an independent set, the sum v∈T d(v) counts each edge of G at least once, i.e., e(G) 6 v∈T d(v) 6
(n(G) − α(G))∆(G). Rearranging gives α(G) 6 n(G) − e(G)/∆(G).
If G is regular, by the degree sum formula we get 2e(G) = n(G)∆(G), i.e., e(G)/∆(G) = n(G)/2, so that
α(G) 6 n(G) − e(G)/∆(G) = n(G)/2.
 in M (i.e. the set of vertices saturated by M ). Q covers all the edges of G, for, if not, then G has an edge
with unsaturated endpoints u and v. But then adding this edge to M will give a matching bigger than M , a
contradiction.
For every k ∈ N we have α0 (kP3 ) = k and β(kP3 ) = 2k.

11. The statement is true. The proof follows from the fact that a tree is acyclic and from the following lemma.
Lemma: Let G be a graph with two distinct perfect matchings M and M 0 . Then every component of the
symmetric difference M M M 0 is an (even) cycle.
Proof Every component of M M M 0 is either an even cycle or a path, the edges in which alternate between
M \ M 0 and M 0 \ M . Let u be an endpoint of such a path P . In view of symmetry we can assume that the
edge e of P incident on u belongs to M \ M 0 . The only edge incident to a vertex of degree 1 must belong
to every perfect matching. Thus dT (u) > 2. Since M 0 saturates u anyway, we can choose a neighbor v of u
not in P such that uv ∈ M 0 . Since e ∈ M , uv ∈
/ M . Thus uv ∈ M 0 \ M . This implies that P cannot be a
component of M M M . 0 •

12. [if] Define a function f : V (T ) → V (T ) as follows. Choose v ∈ V (T ). By hypothesis T \ v has only one
odd component, call it H. If v has two distinct neighbors u1 , u2 in H, then a u1 , u2 -path in H forms a cycle
with the edges vu1 and vu2 . Thus there exists a unique neighbor u of v in H. Define f (v) := u.
First we claim that f is injective. Assume not, i.e., f (v1 ) = f (v2 ) = u for some v1 6= v2 . Think of T
as a tree rooted at u. Let the children of u be v1 , v2 , . . . , vk . Call Ti the subtree of T rooted at vi . Since
f (v1 ) = u, every subtree rooted at a child of v1 is of even order. Therefore, n(T1 ) is odd. Similarly, n(T2 )
is also odd. But then T \ u has at least two odd components (T1 and T2 ), a contradiction to the hypothesis.
Next we claim that f 2 is the identity map on V (T ). Choose v ∈ V (T ). Since V (T ) is finite, the elements
v, f (v), f 2 (v), f 3 (v), . . . cannot be all distinct. Choose 0 6 k < l such that f k (v), . . . , f l−1 (v) are pairwise
distinct, but f k (v) = f l (v) for some l > k. Since T does not contain loops, l > k + 1. If l > k + 2, then
(f k (v), f k+1 (v), . . . , f l−1 (v)) is a cycle in T , a contradiction. So l = k + 2, i.e., f k+2 (v) = f k (v). Since
f is a map from a finite set to itself, its injectivity implies its bijectivity. Applying the function f −k in the
last equation gives f 2 (v) = v, as claimed.
Thus f produces the desired pairing of vertices for a perfect matching.
[only if] Let T have a perfect matching. By Tutte’s 1-factor theorem o(T \ v) 6 1. If o(T \ v) = 0, then T
contains an odd number of vertices. No graph with an odd number of vertices can have a perfect matching.

13. Choose (nonempty) S ⊆ V (G) and count the number k of edges from S to the odd components of G \ S (as
in the case of a corollary proved in the class). G being 3-regular, counting such edges using their endpoints
in S gives k 6 3|S|.
Let H1 , . . . , Hl be all the odd components of G \ S (where l = o(G \ S)) and let li be the number of edges
between Hi and S. The degree sum formula for Hi and the 3-regularity of G yield 2e(Hi ) = 3n(Hi ) − mi .
Since n(Hi ) is odd, we then have mi . Finally since G has at most two cut-edges mi = 1 for at most two i
and mi > 3 otherwise. Therefore, k > 3o(G \ S) − 4.
Combining the two inequalities involving k yields o(G\S) 6 |S|+4/3. Since o(G\S) and |S| are integers,
the last inequality implies o(G \ S) 6 |S| + 1. Assume that o(G \ S) = |S| + 1. It is a straightforward
check that in this case n(G) is odd. But the degree sum formula for G gives 2e(G) = 3n(G), i.e., n(G) is
even, a contradiction. Thus o(G \ S) 6 |S|. Now apply Tutte’s theorem.

14. P2 (an edge) provides a counterexample to the given statement. The corrected assertion is: Let e be a cut-
edge in G. If the component of G containing e has more than two vertices, then at least one endpoint of G
is a cut vertex.
For the proof of the corrected assertion, let H be the component of G containing e and let u and v be the
endpoints of e. Further let H1 and H2 be the two components of H \ e with u ∈ V (H1 ) and v ∈ V (H2 ).
Since n(H) > 3, (at least) one of H1 and H2 has (at least) two vertices. Because of symmetry we can
 separates w from v.

15. Choose two non-adjacent vertices in the Petersen graph G. These two vertices have the label ab and ac
for some permutation a, b, c, d, e of 1, 2, 3, 4, 5. G contains three pairwise internally disjoint ab, ac-paths:
i) ab, de, ac, ii) ab, cd, be, ac, and iii) ab, ce, bd, ac. By Menger’s theorem G is 3-connected, i.e, κ(G) > 3.
Also κ(G) 6 δ(G) = 3. Thus κ(G) = 3. But then κ0 (G) = 3 (since κ(G) 6 κ0 (G) 6 δ(G), or since a
3-regular graph H has κ(H) = κ0 (H)).

16. Let F = [S, S] and F 0 := [S 0 , S 0 ] be two different edge cuts of G. Then S M S 0 is a non-empty proper subset
of V (G). One can readily verify that F M F 0 = [T, T ].

17. The statement is false. A counterexample is provided by the 2-connected

u v
graph shown in the adjacent figure. Consider the u, v-path P shown by
bold edges. The graph contains no u, v-path internally disjoint from P .

18. [if] Clearly G is connected (By hypothesis G satisfies a condition stronger than mere connectivity). Choose
x ∈ V (G). We will show that G \ x is connected. Choose y, z ∈ V (G \ x). By hypothesis there exists an
x, y-path P in G through z. Deletion of x retains the part of P from z to y.
[only if] Suppose that G is 2-connected and choose three pairwise Q

distinct vertices x, y, z ∈ V (G). By Menger’s theorem G contains

two internally disjoint x, z-paths P1 and P2 . Furthermore, G \ x
contains a y, z-path Q. Let w be the vertex on Q, that is nearest to
y and that is also in V (P1 ) ∪ V (P2 ). By symmetry let us assume that
w ∈ V (P2 ). (Note that we can have w = z, in which case w belongs P1
to V (P1 ) as well. Also w 6= x.) x z
P1 , the z, w-subpath of P2 and the w, y-subpath of Q give us an x, y- w
path through z. P2

19. Consider the graph G shown in the adjacent figure.

v1 e1
The vertices v1 , . . . , vκ each has pairwise distinct
δ
sets of δ neighbors in Kκδ+1 . v1 , . . . , vκ−1 has one v2 e2
neighbor each in Kδ+1 , whereas vκ has κ0 − κ + 1 δ
neighbors in Kδ+1 . All these neighbors in Kδ+1 are v3 e3
pairwise distinct. δ

One can now verify that δ(G) = δ (look at the vertex Kκδ+1 Kδ+1
in Kδ+1 having no neighbor in {v1 , . . . , vκ }).
vκ−1 eκ−1
Consider a pair (x, y) of non-adjacent vertices in this δ
vκ eκ
graph. If x = vi and y = vj for some i 6= j, then
δ
there exist δ > κ internally disjoint vi , vj -paths via eκ’
Kκδ+1 . If x = u ∈ V (Kκδ+1 ) and y = vi , we again
get δ > κ internally disjoint u, vi -paths via Kκδ+1 . If x = vi and u = w ∈ V (Kδ+1 , we get κ internally
disjoint vi , w-paths, one via ei , the others via Kκδ+1 , vj and ej for each j 6= i. Finally, if x = u ∈ V (Kκδ+1 )
and y = w ∈ V (Kδ+1 ), then every u, w-path must consist of one of the vertices v1 , . . . , vκ . Also we can
arrange exactly κ u, w-paths one through each vi . By Menger’s theorem κ(G) = κ and {v1 , . . . , vκ } is a
minimum vertex-cut of G.
A similar study shows that κ0 (G) = κ0 and {e1 , . . . , eκ0 } = [S, S], S = V (Kδ+1 ), is a minimum edge-cut
of G. (Use the Elias-Feinstein-Shannon-Ford-Fulkerson theorem.)
 vertices of G. Let us represent the corresponding vertices in B(G) by b 1 , . . . , bs and u1 , . . . , ut respectively.
If t = 0, then B(G) is the single-vertex tree and has no leaf. So let us concentrate on the case that t > 0, so
that s > 1.
Let B be a block of G. Since G is connected and B does not contain all the vertices of G, there exists
x ∈ V (B) such that x has a neighbor y outside B. The edge xy lies in a separate block B 0 of G. But then
B and B 0 share x and so x is a cut vertex of G. Thus every block contains at least one cut vertex.
By definition B(G) is simple. In order to prove that B(G) is connected, it suffices to produce a b i , uj -path
for every i, j. Bi contains a cut vertex of G, say, vk . Since G is connected, there is a vk , vj -path P in G. Let
us follow the path P starting from vk . We simultaneously generate a bi , uj -walk Q. Initially we are at vk in
P and bi in Q. Assume that at some stage we are at x in P and bl in Q. If x is not a cut-vertex of G, we
remain at bl in Q and if P is not yet finished, we proceed to the next vertex on P . If x is a cut-vertex, say
vm , we move from bl to um in Q, and if P is not finished and the next edge on P belongs to Bn , we move
from um to bn and proceed to the next vertex on P . The resulting bi , uj -walk Q contains a bi , uj -path.
Now suppose that (bi1 , ui1 , . . . , bir , uir ) is a cycle in B(G). Since B(G) is simple, r > 3. I will show that
S
B := rk=1 Bik is 2-connected, contradicting the maximality of each block Bik . Choose x ∈ V (B) and
y, z ∈ V (B) \ {x}. By symmetry we can assume that x ∈ V (Bi1 ). Since Bi1 is a block and has more than
one vertex (G contains no isolated vertex), Bi1 \ x is connected. Choose w ∈ V (Bi1 ) \ x. Then there is a
y, w-path P1 and a w, z-path P2 in B \ x. P1 and P2 produce a y, z-walk which contains a y, z-path. (See
the following figure explaining two possibilities for x. P1 (resp. P2 ) may lie entirely in Bi1 , if y (resp. z) is
in V (Bi1 ) \ {x}. If both y and z are in V (Bi1 ) \ {x}, then one directly gets an y, z-path in the connected
graph Bi1 \ x.)

w w
x
x
P1
P2
P1
P2
z z

y
y

Thus B(G) is a tree. What remains is to show that a cut vertex of G must belong to at least two blocks of
G. Assume the contrary, that is, a cut vertex x belongs to only one block B of G. If x has a neighbor y
outside B, then the edge xy belongs to a different block B 0 so that x belongs to B 0 as well. So x does not
have a neighbor outside B. Now choose any y, z ∈ V (G \ x). Since G is connected, there is a y, z-path P
in G. If P does not involve B, it remains in G \ {x}. Otherwise let y 0 and z 0 be the first and last vertices of
P which are in V (B). (We might have y 0 = y and/or z 0 = z.) Since B is 2-connected, there is a y 0 , z 0 -path
Q in B \ x. The y, y 0 -subpath of P , Q and the z 0 , z-subpath of P give a y, z-path in G \ {x}. Thus G \ {x}
is connected, i.e., x is not a cut vertex of G, a contradiction.

21. We start by proving that for any plane graph H the dual H ∗ is connected. For the proof let Xu be the
unbounded face of H and X any other face of H. If we draw a (semi-infinite) ray from the point x ∈ V (H ∗ )
representing X, the ray will eventually go the interior of Xu (because all faces of H other than Xu are
bounded). From a point on the ray in the interior of Xu there is a polygonal curve ending in xu (the vertex
representing Xu ) and lying entirely inside Xu . We can distort the ray, if necessary, to get a simple polygonal
curve C starting from x and ending in xu such that C does not go through any vertex of H. C crosses a
finite number of edges of H (at their internal points). Every time it does so, it may or may not move from
one face of H to another. In any case, C traverses a finite sequence of faces of H, in which two consecutive
faces share an edge. This gives us an x, xu -path in H ∗ . But then for any x, x0 ∈ V (H ∗ ) an x, xu - and an
x0 , xu -path yield an x, x0 -walk and hence an x, x0 -path in H ∗ . Thus H ∗ is connected, as claimed.
First assume that G is disconnected. Taking H = G∗ in the last paragraph shows that H ∗ = (G∗ )∗ is
connected and hence cannot be the same as G.
For the converse, let G be connected. Let us draw G∗ in such a way that each edge of G∗ crosses only the
corresponding edge of G (only once) and no other edge of G or G ∗ . Now let X ∗ be a face of G∗ . Any edge
 the face X ∗ and must terminate before leaving X ∗ . That is, X ∗ contains at least one vertex of G. Assume
that some face of G∗ contains more than one vertex of G, i.e., n > f ∗ . But f = n∗ and e = e∗ . Since G∗
is connected, Euler’s formula gives n∗ − e∗ + f ∗ = 2. Combining all these findings gives n − e + f > 2,
i.e., G is not connected, a contradiction. Thus every face of G∗ contains exactly one vertex of G. But then
we can use each e as (e∗ )∗ and obtain G as the dual of (G∗ )∗ .

22. It is sufficient to prove the assertion for a connected graph, since otherwise we could add an appropriate
number of edges and prove the assertion for the resulting connected outerplane graph. The length of the
outer face of G is n, whereas the length of every other face is at least 3 (since G is simple). Therefore,
the degree sum formula for G∗ gives 2e = 2e∗ > n + 3(f − 1). Since G is plane also, n − e + f = 2.
Eliminating f gives e 6 2n − 3.

23. Consider the following two cases:

Case 1: G is not planar.
Then G is clearly non-outerplanar too. Also G contains a subdivision of K 5 or K3,3 and hence a subdivision
of K4 or K2,3 .
Case 2: G is planar.
Consider the graph H obtained by adding a new vertex to G and joining this new vertex to every vertex in
G. It follows that G is outerplanar ⇐⇒ H is planar ⇐⇒ H contains no subdivision of K 5 or K3,3 ⇐⇒
G contains no subdivision of K4 or K2,3 . This is exemplified in the following figure:

G is outerplanar

G is not outerplanar

24. (a) It is sufficient to prove the assertion for connected graphs. By Euler’s formula n − e + f = 2. Since G
has girth k, every face of G has length at least k, i.e., 2e > kf . Eliminating f gives e 6 (n − 2) k−2
k
.

(b) The Petersen graph has girth 5. By Part (a) every simple planar graph with 10 vertices and of girth 5
contains at most 8 × 5/3 = 13.333 . . . , i.e., at most 13 edges. But the Petersen graph has 15 edges.

25. (a) Let G ∼= G∗ . Since G∗ is connected (Exercise 21),

G is also connected. By Euler’s formula n − e + f = 2.
But the isomorphism G ∼ = G∗ implies n = n∗ = f , i.e.,
n − e + n − 2, i.e., e = 2n − 2.
(b) Let G be the cycle Cn−1 plus another vertex joined to
each vertex of Cn−1 . Then G ∼ = G∗ . The adjacent figure
explains the construction for n = 7.