On removable cycles through every edge
-
Upload
manoel-lemos -
Category
Documents
-
view
214 -
download
0
Transcript of On removable cycles through every edge
On Removable CyclesThrough Every Edge
Manoel Lemos1 and James Oxley2*1DEPARTAMENTO DE MATEMATICA
UNIVERSIDADE FEDERAL DE PERNAMBUCO, RECIFE
PERNAMBUCO 50740-540, BRAZIL
E-mail: [email protected]
2DEPARTMENT OF MATHEMATICS
LOUISIANA STATE UNIVERSITY
BATON ROUGE, LOUISIANA 70803-4918
E-mail: [email protected]
Received March 6, 2001; Revised August 21, 2002
DOI 10.1002/jgt.10082
Abstract: Mader and Jackson independently proved that every 2-connectedsimple graph G with minimum degree at least four has a removable cycle,that is, a cycle C such that G\E(C) is 2-connected. This paper considers theproblem of determining when every edge of a 2-connected graph G, simpleor not, can be guaranteed to lie in some removable cycle. The main resultestablishes that if every deletion of two edges from G remains 2-connected,then, not only is every edge in a removable cycle but, for every two edges,there are edge-disjoint removable cycles such that each contains one of thedistinguished edges. � 2002 Wiley Periodicals, Inc. J Graph Theory 42: 155–164, 2003
——————————————————
Contract grant sponsor: CNPq; Contract grant number: 522910/96-3; Contractgrant sponsor: ProNEx/CNPq; Contract grant number: 664107/97-4; Contractgrant sponsor: CAPES; Contract grant number: AEX1307/00-2; Contract grantsponsor: National Security Agency; Contract grant numbers: MDA904-99-1-0030and MDA904-01-1-0026.*Correspondence to: James Oxley, Department of Mathematics, Louisiana StateUniversity, Baton Rouge, Louisiana 70803-4918. E-mail: [email protected]
� 2002 Wiley Periodicals, Inc.
Keywords: removable cycle; 2-connected graph; matroid
1. INTRODUCTION
If G is a 2-connected simple graph in which the minimum degree � (G) is at least
four, then, as a special case of a result for k-connected graphs, Mader [7] showed
that G has a cycle C such that GnEðCÞ, the graph obtained by deleting the edges
of C, is 2-connected. Independently, Jackson [5] proved that C can be chosen
to avoid a nominated edge e and to have at least � (G)�1 edges. We call a cycle D
in a 2-connected graph H removable if HnEðDÞ is 2-connected. Lemos and
Oxley [6] extended Jackson’s theorem and obtained, as a corollary, that every
simple 2-connected graph with � ðGÞ � 5 has two removable edge-disjoint cycles
each with at least � ðGÞ þ 1 edges.
Although every simple 2-connected graph G with minimum degree at least four
has a removable cycle, the example shown in Figure 1(a), which was obtained
independently by Robertson (in [5]) and Jackson [5], shows that the requirement
that G be simple cannot be dropped. However, Fleischner and Jackson [2] proved
that this requirement could be replaced by the condition that G is planar. Their
result was extended by Goddyn, van den Heuvel, and McGuinness [3] who
proved the following strengthening of a conjecture of Jackson [5].
Theorem 1.1. Let G be a 2-connected graph with minimum degree at least four.
If G has no minor isomorphic to the Petersen graph, then G has two edge-disjoint
removable cycles.
In this paper, we consider when we can guarantee that every edge of a 2-
connected graph G is in some removable cycle. When G is simple, the re-
quirement that � ðGÞ � 4 is not sufficient to ensure the existence of such a family
of removable cycles for if G has an edge cut of size less than four, none of the
FIGURE 1
156 JOURNAL OF GRAPH THEORY
edges in this cut will be in a removable cycle. But even if we strengthen the
minimum-degree requirement to insist that every edge-cut has size at least four,
we shall not be guaranteed that every edge is in a removable cycle. For example,
every edge cut in the graph G in Figure 1(b) has size at least four, but the only
edges that are in removable cycles are those edges that are in 2-cycles. Although
G is not simple, a straightforward modification produces a simple graph G0 that
has no removable cycle meeting either the outer pentagon or the inner pentagram.
Specifically, we replace each pair fx; yg of parallel edges of G by a copy of K5
whose vertex set meets VðGÞ in the endpoints of x.
In the following theorem, the main result of the paper, the hypothesis is strong
enough to ensure not only that every edge is in a removable cycle but also that,
for every two edges, there is a pair of edge-disjoint removable cycles one con-
taining each of the distinguished edges. We observe here that, for the graph G
in Figure 2, every edge is in a removable cycle, but the only removable cycle
containing e also contains f . This example can be modified as above to produce a
simple graph satisfying the same assertions.
Theorem 1.2. Let G be a 2-connected graph. Suppose that GnX is a 2-
connected graph for every 2-subset X of EðGÞ, and let x and y be distinct edges
of G. Then G has edge-disjoint cycles Cx and Cy containing x and y, respectively,
such that both GnEðCxÞ and GnEðCyÞ are 2-connected.
For comparison with Theorem 1.1, we note the following immediate con-
sequence of Theorem 1.2.
Corollary 1.1. Let G be a 2-connected graph such that GnX is 2-connected for
every 2-subset X of EðGÞ. Then G has two edge-disjoint removable cycles.
The proof of Theorem 1.2 will be given in the next section. Section 3 describes
a counterexample to a natural extension of Theorem 1.2 to cographic matroids.
The graph and matroid terminology that is used here will follow Bondy and
Murty [1] and Oxley [8], respectively. A block is a connected graph with no cut-
vertices. A graph with at least three vertices is a block if and only if it is 2-
connected and loopless. For a path � that is not a cycle, if u and v are vertices of
�, then � ½u; v� denotes the subpath of � having u and v as its ends.
FIGURE 2
REMOVABLE CYCLES 157
2. PROOF OF THE MAIN RESULT
The following result is the main tool in the proof of Theorem 1.2. Once it is
proved, it will not be difficult to complete the proof of the theorem.
Proposition 2.1. Suppose that G and H are blocks such that
(i) H is a subgraph of G; and
(ii) GnX is a block for every X � EðGÞ � EðHÞ such that jXj � 2.
Then, for each e in EðGÞ � EðHÞ, either(iii) there is no cycle of G that contains e and is edge-disjoint from H; or
(iv) there is a cycle C of G that contains e such that C is edge-disjoint from
H and GnEðCÞ is a block.
Proof. Suppose that the proposition fails and choose a counterexample
ðG;H; eÞ for which ðjEðGÞj;�jEðHÞjÞ is lexicographically minimal. Then there is
a cycle C of G that contains e and is edge-disjoint from H. Moreover, GnEðCÞ is
not a block. Let B be the block of GnEðCÞ that contains EðHÞ. By the choice of
ðG;H; eÞ, we have that B ¼ H. Choose C such that the connected component of
GnEðCÞ that contains H as a block has the maximum number of edges.
In the rest of the proof, when � is a subpath of C, we define its complement �c
to be the other subpath of C having the same terminal vertices as �. In particular,
�ci and �0c will denote the complements of �i and �0, respectively.
Lemma 2.1. GnEðCÞ is connected.
Proof. Let H0;H1;H2; . . . ;Hn be the connected components of GnEðCÞ and
let H be a block of H0. We want to show that n ¼ 0. Assume that n � 1. For each
i in f1; 2; . . . ; ng, let �i be the shortest subpath of C containing e such that its
terminal vertices are in VðHiÞ. If �i¼C, then �i and Hi have a unique common
vertex, say vi. Since G is a block, vi is not a cut-vertex of G so vi is an isolated
vertex of GnEðCÞ. But then GnX is not a block when X consists of the two edges
of G incident with vi. This contradiction to (ii) implies that �i 6¼ C, so �i has
different terminal vertices. Let �0i be a path in Hi joining the terminal vertices of
�i. Note that �i [ �0i is a cycle C0 of G that contains e and is edge-disjoint from H.
Therefore,
Vð�ci Þ \ VðH0Þ ¼ ;; ð1Þ
otherwise the connected component of GnEðC0Þ that contains H contains EðH0Þ [Eð�c
i Þ, which is contrary to the choice of C.
Now, we shall define an auxiliary graph K such that VðKÞ ¼ f1; 2; . . . ; ng and
ij is an edge of K if and only if i 6¼ j and E(�ci Þ \ E(�c
j Þ 6¼ ;. Let K 0 be a
connected component of K. Let � be the union of all the paths �ci for which i in
VðK 0Þ. Note that � is a subpath of C that does not contain e. We set
V 0 ¼[
i2VðK 0ÞVðHiÞ:
158 JOURNAL OF GRAPH THEORY
Note that V 0 6¼ VðGÞ because V 0 \ VðH0Þ ¼ ; by the definition of K 0. Let X be
the set of edges of G that join a vertex of V 0 to a vertex of VðGÞ � V 0. By our
construction, X � EðCÞ and, since G is a block, jXj � 2. We shall prove that X is
the set of terminal edges of �c.
First we show that X \ Eð�Þ ¼ ;. Suppose that f 2 X \ Eð�Þ. By definition
of �, the edge f 2 Eð�ci Þ for some i 2 VðK 0Þ. Let f ¼ xy where x 2 V 0 and
y 2 VðGÞ � V 0. By (1), y 62 VðH0Þ. Thus y 2 VðHjÞ for some j in f1; 2; . . . ; ng.
Hence j 62 VðK 0Þ. Let g be the edge of C � ffg incident to y. By definition of �j,
we have that Eð�cj Þ \ ff ; gg 6¼ ;. The definition of K now implies that Eð�c
i Þ \Eð�c
j Þ ¼ ;. Hence g 2 Eð�cj Þ � Eð�c
i Þ and f 2 Eð�ci Þ � Eð�c
j Þ. Thus f is a termi-
nal edge and y a terminal vertex of �ci and hence of �i. Therefore y 2 VðHiÞ.
But y 2 VðHjÞ so j ¼ i. This contradiction implies that X \ Eð�Þ ¼ ;.
We now know that X � Eð�cÞ. Let f 2 X. Then f is incident with some vertex
x of V 0 and x 2 VðKiÞ, say. Since f 2 Eð�cÞ, we deduce that f 2 Eð�iÞ. Now
no internal vertex of �i is in Hi, so x is a terminal vertex of �i. Thus the edge g of
C � f that is incident with x is in Eð�ci Þ and so is in Eð�Þ. Hence f must be a
terminal edge of �c. Thus jXj � 2. But jXj � 2, so jXj ¼ 2 and we have a con-
tradiction since GnX is disconnected. &
Let w1;w2; . . . ;wn be the cut vertices of GnEðCÞ belonging to H. For each k in
f1; 2; . . . ; ng, let Gk be the connected component of GnðEðHÞ [ EðCÞÞ that
contains wk. We set Vk ¼ VðGkÞ.Lemma 2.2. Suppose that fu; vg � Vk \ VðCÞ and that � is a uv-path of Gk.
Let K be the connected component of GknEð�Þ such that wk 2 VðKÞ. If � is the
path in C that joins u and v and does not contain e, then Vð�Þ � Vk or
Vð�Þ \ VðKÞ ¼ ;.Proof. Observe that the union D of �c and � is an Eulerian graph since �c and
� are edge-disjoint paths that have the same terminal vertices. Thus D is a union
of cycles of G. One of these cycles, say C00, contains e. When Vð�Þ \ VðKÞ ¼ ;,
the lemma holds. Therefore, we may suppose that there is a vertex x in Vð�Þ \VðKÞ. Let � be a wkx-path in K. We choose x so that � has minimum length.
Hence Vð�Þ \ Vð�Þ ¼ fxg. In particular, the union of � with either of the paths
�½u; x� or �½x; v� is also a path.
Now suppose that Vð�Þ 6� Vk and choose a vertex w of Vð�Þ � Vk such that its
distance to x in the path � is a minimum. Then, w is a vertex of �½u; x� or �½x; v�.Without loss of generality, we may assume the latter. It follows, by the choice of
w, that Vð�½x;w�Þ � Vk ¼ fwg. Consider the wkw-path � that is the union of � and
�½x;w�.Suppose w 2 VðHÞ. As � is edge-disjoint from H and C00, it follows that
EðHÞ [ Eð�Þ is contained in some block H0 of GnEðC00Þ. This contradicts the
choice of ðG;H; eÞ since H0 has more edges than H. Therefore, w 62 VðHÞ.We may now suppose that w 2 Vl for some l 6¼ k. Let �l be a wwl-path in Gl.
Observe that Vð�lÞ \ Vð�Þ ¼ fwg because Vð�Þ � VðKÞ � Vk and Vð�½x;w�Þ�
REMOVABLE CYCLES 159
Vk ¼ fwg. Thus the union of � and �l is a wkwl-path " that is edge-disjoint from
H and C00. Therefore, EðHÞ [ Eð"Þ is contained in a block H00 of GnEðC00Þ. This
contradicts the choice of ðG;H; eÞ since H00 has more edges than H. We deduce
that w does not exist, so Vð�Þ � Vk. &
Let �k be the longest path in C that has both ends in Vk and does not contain e.
Next we show that
2.3. Vð�kÞ � Vk 6¼ ;:Let X be the set of edges of G that join a vertex of Vk � fwkg to a vertex of
VðGÞ � Vk. Suppose that Vð�kÞ � Vk. Then X � Eð�ckÞ. But the choice of �k
means that only the terminal vertices of �ck are in Vk. Hence each edge of X is
terminal in �ck. Thus jXj � 2. Since GnX has wk as a cut-vertex, it is not difficult
to obtain a contradiction. We conclude that (2.3) holds.
Let ak and bk be the ends of �k. Traverse �k beginning at ak and let vk be the
first vertex that is not in Vk. Let �0k be the shortest subpath of �k that contains vk
and has both ends in Vk. Suppose that �0k joins the vertices a0k and b0k and that ak;
a0k; vk; b0k; bk appear in this order in �k. Note that a0k; vk, and b0k are distinct, but ak
may equal a0k, and bk may equal b0k. We define
Ak ¼ Vð�k½ak; a0k�Þ \ Vk and Bk ¼ Vð�k½b0k; bk�Þ \ Vk:
By the choice of vk, we have that Ak ¼ Vð�k½ak; a0k�Þ.We now define a simple auxiliary graph M with vertex set Ak. If a and a0 are
distinct members of Ak, then aa0 2 EðMÞ if and only if there is no b in Bk such
that Gk has two edge-disjoint paths joining fa; a0g and fb;wkg.
When f is an isthmus of Gk, we denote by Gkð f Þ the connected component of
Gkn f whose vertex set does not contain wk.
Lemma 2.3. Let fa; a0g be a subset of Ak such that if a 6¼ a0, then aa0 2 EðMÞ.Then
(i) Gk has an isthmus that separates fa; a0g from wk.
(ii) If f is an isthmus of Gk that separates fa; a0g from wk such that jVðGkð f ÞÞjis a minimum, then
(a) VðGkð f ÞÞ \ Bk ¼ ;; and
(b) if N is the connected component of M such that fa; a0g � VðNÞ, thenVðGkð f ÞÞ \ Ak � VðNÞ.
Proof. We shall prove (i) and (ii)(a) simultaneously. To do this, we take J ¼Gk and w ¼ wk when there is no isthmus separating fa; a0g from wk; and, when
the hypothesis of (ii)(a) holds, we take J to be Gkð f Þ and w to be the unique
vertex in VðGkð f ÞÞ \ Vðf fgÞ.Next we observe that there is no isthmus in J that separates fa; a0g from w.
This is true by definition when J ¼ Gk. Moreover, if J ¼ Gkð f Þ and g is such an
160 JOURNAL OF GRAPH THEORY
isthmus, then g is also an isthmus in Gk that separates fa; a0g from wk. Since
GkðgÞ is clearly a subgraph of Gkð f Þ � w, the choice of f is contradicted. Hence
g does not exist.
By the result of the last paragraph, Menger’s Theorem implies that J has an
aw-path � and an a0w-path �0 such that Eð�Þ \ Eð�0Þ ¼ ;.
Next we show that
2.3.1. VðJÞ \ Bk ¼ ;:
If b 2 VðJÞ \ Bk, then choose a path � of minimum length joining b to some
vertex of � or �0, say �0. Let b0 be the common vertex of �0 and � and let � be the
union of �0½a0; b0� and �. Then � and � are edge-disjoint paths of J joining a to w
and a0 to b, respectively. The path � can be completed to an awk-path " of Gk that
is edge-disjoint from � by taking its union with a wwk-path in GknEðJÞ. Hence
aa0 62 EðMÞ and so a ¼ a0. Now apply Lemma 2.2 taking ðu; v; �Þ to be ða0; b; �Þ.The path � implies that a0 2 Vð�k½a0; b�Þ \ VðKÞ, where K is the connected com-
ponent of GknEð�Þ that contains wk. Thus Vð�k½a0; b�Þ \ VðKÞ 6¼ ;. Moreover,
the definitions of Ak and Bk imply that Vð�k½a0; b�Þ 6� Vk. Thus Lemma 2.2 is
contradicted. Therefore b does not exist and so (2.3.1) holds.
As ; 6¼ Bk � VðGkÞ, it follows that J 6¼ Gk. Thus (i) holds. Moreover, J ¼Gkð f Þ and (ii)(a) holds by (2.3.1). Observe that (ii)(b) follows because any path of
Gk joining a vertex of VðGkð f ÞÞ to a vertex of fwkg [ Bk must contain f as an
edge by (ii)(a). It follows by the definition of M that any two distinct elements of
Ak \ VðGkð f ÞÞ are adjacent in M. &
Lemma 2.4. If N is a connected component of M, then there is an isthmus f of
Gk such that
VðGkð f ÞÞ \ Ak ¼ VðNÞ and VðGkð f ÞÞ \ Bk ¼ ;:
Proof. We shall prove the following assertion by induction on jV j.
2.4.1. If V � VðNÞ and N½V� is connected, then there is an isthmus fV of Gk such
V � VðGkð fVÞÞ \ Ak � VðNÞ and VðGkð fVÞÞ \ Bk ¼ ;:
The lemma follows by taking V ¼ VðNÞ in (2.4.1).
Suppose that jVj ¼ 1, say V ¼ fag. Now taking a0 ¼ a in Lemma 2.3(i), we get
that there is an isthmus f of Gk that separates fa; a0g from wk. Choose f so that
jVðGkð f ÞÞj is a minimum. Then (2.4.1) follows from Lemma 2.3(ii) by taking
ffag ¼ f . Now assume that (2.4.1) holds for jV j < n and let jVj ¼ n � 2. Since
N½V� is connected, there is an a in V such that N½V � fag� is connected. By in-
duction, the isthmus fV�fag exists. If a 2 VðGkð fV�fagÞÞ, then we take fV ¼ fV�fagand the result follows. We may now suppose that a 62 VðGkð fV�fagÞÞ. Choose a0
in V � fag such that aa0 2 EðNÞ. By Lemma 2.3(i), there is an isthmus f of Gk
REMOVABLE CYCLES 161
that separates fa; a0g from wk because aa0 2 EðMÞ. Choose such an f for which
jVðGkð f ÞÞj is a minimum.
Note that, since a0 2 V � fag � VðGkð fV�fagÞÞ, any a0wk-path in Gk contains
fV�fag and f as edges. Moreover, since fV�fag separates a from a0 in Gk, it follows
that, in each a0wk-path in Gk, the edge fV�fag is closer to a0 than f is. Thus Gkð f Þhas Gkð fV�fagÞ as a subgraph. Hence V � VðGkð f ÞÞ. The remaining parts of
(2.4.1) follow from Lemma 2.3(ii) by taking fV ¼ f . &
Now let N be a connected component of M. By the previous lemma, there is an
isthmus f of Gk such that
VðGkð f ÞÞ \ Ak ¼ VðNÞ and VðGkð f ÞÞ \ Bk ¼ ;: ð2Þ
Let fx; yg be the subset of VðGkð f ÞÞ \ Ak such that ak; x; y; a0k appear in this
order in �k and all the other vertices of VðGkð f ÞÞ \ Ak lie between x and y on �k.
Next we prove the following:
2.4.2. Every vertex of �k½x; y� is in VðGkð f ÞÞ \ Ak.
Suppose that there is a vertex z of G such that z 2 Vð�k½x; y�Þ but
z 62 VðGkð f ÞÞ. Since Vð�k½ak; a0k�Þ ¼ Ak and fx; yg � Vð�k½ak; a0k�Þ, it follows
that z 2 Ak. By the definition of M, the vertices x and z are in different
components of M. Thus there are edge-disjoint paths, � and �, joining fx; zg to
fwk; bg for some b 2 Bk. First suppose that � and � join x to b and z to wk,
respectively. We know that vk is in the path in C that joins x and b and does not
contain e. Since vk 62 Vk, Lemma 2.2 implies that z and wk are in different
connected components of GknEð�Þ. This contradiction to the fact that � joins z to
wk implies that we may assume that � joins z to b and � joins x to wk.
Let VðffgÞ \ VðGkð f ÞÞ ¼ fwg. Since x 2 VðGkð f ÞÞ, the isthmus f of Gk
separates wk from x. Thus f is an edge of �. Hence w is a vertex of �, and f is not
an edge of z. As y 2 VðGkð f ÞÞ, there is a yw-path g of Gkð f Þ. The union " of �and �½w;wk� is a ywk-path of Gk. Moreover, e is edge-disjoint from z because zand � are edge-disjoint, and Vð�Þ � VðGkð f ÞÞ while Vð�Þ avoids VðGkð f ÞÞ since
f is not an edge of z and z 62 VðGkð f ÞÞ. Now apply Lemma 2.2, letting
ðu; v; �Þ ¼ ðz; b; �Þ. Since �k½z; b� contains vk, it is clear that Vð�k½z; b�Þ 6� Vk.
Therefore, Vð�k½z; b�Þ \ VðKÞ ¼ ;, where K is the connected component of
GknEð�Þ that contains wk. But y 2 Vð�k½z; b�Þ, and the path " joins y to wk so
y 2 VðKÞ. Therefore y 2 Vð�k½z; b�Þ \ VðKÞ. This contradiction implies that
(2.4.2) holds.
We now complete the proof of Proposition 2.1. From (2.4.2), (2), and the
choice of x and y, we have that VðGkð f ÞÞ \ VðCÞ ¼ Vð�k½x; y�Þ. Thus there
are just two edges of C that join a vertex of Gkð f Þ to a vertex outside Gkð f Þ.Delete these edges. The resulting graph has f as an isthmus and has at least three
vertices. Therefore, it is not a block. This contradiction completes the proof of
Proposition 2.1. &
162 JOURNAL OF GRAPH THEORY
Proof of Theorem 1.2. The graph obtained from G by deleting all the loops
is a block. If we can prove the theorem in the case that G is a block, then it will
follow in general because each loop is itself a removable cycle. Thus assume that
G is a block. First, observe that Gne is a block for every edge e of G. Let x and
y be distinct edges of G. Since Gny is a block, G has a cycle C that contains x
but not y. Now applying Proposition 2.1 taking H to be the block with edge set
fyg, we obtain a cycle Cx of G such that GnEðCxÞ is a block, x 2 EðCxÞ, and
y 62 EðCxÞ. Since GnEðCxÞ is a block, there is a cycle C0 of this graph such that
y 2 EðC0Þ. Applying Proposition 2.1 again, this time taking H to be the cycle Cx,
we obtain a cycle Cy containing y such that EðCxÞ \ EðCyÞ ¼ ; and GnEðCyÞ is
a block. &
3. MATROID (NON)-EXTENSIONS
Mader’s theorem [7], with which we began this paper, implies that every 2-con-
nected simple graph in which every bond has at least four edges has a removable
cycle. In particular:
3.1. Every 2-connected simple graphic matroid M in which every cocircuit has
at least four elements has a removable circuit, that is, a circuit C such that M \C is
2-connected.
The matroid U3;7 shows that one cannot delete the word ‘‘graphic’’ from the
last result. Oxley [8, Problem 14.4.8] asked whether one could replace ‘‘graphic’’
with ‘‘binary’’ in (3.1) and, specializing this, Goddyn, van den Heuvel, and
McGuinness (in an earlier version of [3]) conjectured that (3.1) holds when we
replace ‘‘graphic’’ with ‘‘cographic’’. Lemos and Oxley [6] gave a counterexample
to this conjecture.
We describe this example next for it provides a counterexample to a natural
extension of Theorem 1.2 to cographic matroids. We begin with a copy of K5;5
having vertex classes V1 and V2. Then, for every 3-element set Z that is a subset
of V1 or of V2, adjoin two new degree-3 vertices vZ and wZ making each of them
adjacent to all the members of Z. The resulting graph G has 50 vertices and
145 edges. Evidently G is bipartite and has girth four. Moreover, G is 3-connected
and 3-edge-connected. It is shown in [6, Proposition 2.1] that G has no bond C�
for which the contraction G=C� is 2-connected. Thus M�ðGÞ has no remo-
vable circuit. However, M�ðGÞnX is 2-connected for all 2-subsets X of EðGÞbecause every single-edge contraction of G is 3-connected. Thus, by contrast with
the graphic case dealt with in Theorem 1.2, in the cographic case, the condition
that MnX is 2-connected for all 2-subsets X of EðMÞ is not strong enough to
guarantee the existence of even one removable circuit.
Although Mader’s theorem does not extend to binary or even cographic
matroids, by strengthening the hypothesis slightly, Goddyn and Jackson [4] were
REMOVABLE CYCLES 163
able to extend the theorem to a class of binary matroids that includes cographic
matroids. In particular, they proved that if e is an element of a connected binary
matroid M such that M does not have both the Fano matroid and its dual as
minors and jXj � 5 for all cocircuits X of M not containing e, then M has a
removable circuit C that avoids e such that rðMnCÞ ¼ rðMÞ.
ACKNOWLEDGMENTS
Manoel Lemos was partially supported by CNPq (Grant No. 522910/96-3),
ProNEx/CNPq (Grant No. 664107/97-4), and CAPES (Grant No. AEX1307/00-2).
James Oxley was partially supported by the National Security Agency (Grant
Nos. MDA904-99-1-0030 and MDA904-01-1-0026).
REFERENCES
[1] J. A. Bondy and U. S. R. Murty, Graph Theory with Applications, North-
Holland, New York, 1976.
[2] H. Fleischner and B. Jackson, Removable cycles in planar graphs, J London
Math Soc (2) 31 (1985), 193–199.
[3] L. A. Goddyn, J. van den Heuvel, and S. McGuinness, Removable circuits in
multigraphs, J Combin Theory Ser B 71 (1997), 130–143.
[4] L. A. Goddyn and B. Jackson, Removable circuits in binary matroids, Combin
Probab Comput 8 (1999), 539–545.
[5] B. Jackson, Removable cycles in 2-connected graphs of minimum degree at
least four, J London Math Soc (2), 21 (1980), 385–392.
[6] M. Lemos and J. Oxley, On removable circuits in graphs and matroids, J Graph
Theory 30 (1999), 51–66.
[7] W. Mader, Kreuzungfreie a; b-Wege in endlichen Graphe, Abh Math Sem
Univ Hamburg 42 (1974), 187–204.
[8] J. G. Oxley, Matroid Theory, Oxford University Press, New York, 1992.
164 JOURNAL OF GRAPH THEORY