Advances in Applied Mathematics 50 (2013) 6–15

Contents lists available at SciVerse ScienceDirect

Advances in Applied Mathematics

www.elsevier.com/locate/yaama

Basis-exchange properties of sparse paving matroids

Joseph E. Bonin
Department of Mathematics, The George Washington University, Washington, DC 20052, United States

a r t i c l e i n f o a b s t r a c t

Article history: It has been conjectured that, asymptotically, almost all matroids
Available online 17 September 2012 are sparse paving matroids. We provide evidence for five long-
Dedicated to Geoff Whittle on his 60th
standing, basis-exchange conjectures by proving them for this large
birthday class of matroids.
© 2012 Elsevier Inc. All rights reserved.
MSC:
05B35

Keywords:
Matroid
Sparse paving matroid
Basis-exchange property

1. Introduction

A matroid is paving if the closure of each nonspanning circuit is a hyperplane; it is sparse paving
if each nonspanning circuit is a hyperplane. Thus, a matroid M of rank r is sparse paving if and
only if each r-subset of E ( M ) is either a basis or a circuit-hyperplane. It follows that the class of
sparse paving matroids is dual-closed. It is easy to show that this class is also minor-closed. Sparse
paving matroids can also be characterized as the matroids M for which both M and its dual, M ∗ , are
paving.
While paving and sparse paving matroids have received increasing attention recently (see, e.g.,
[6,9,13–15]), they have long played important roles in matroid theory. For instance, D. Knuth [12]
constructed at least
n
2(n/2)
/2n

n!

E-mail address: [email protected].

0196-8858/$ – see front matter © 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.aam.2011.05.006
 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15 7

nonisomorphic sparse paving matroids of rank n/2 on n elements; with the upper bound by M. Piff
[18], it follows that the number gn of nonisomorphic simple matroids on n elements satisfies

3
n− log2 n + O (log2 log2 n)  log2 log2 gn  n − log2 n + O (log2 log2 n), (1.1)
2

with sparse paving matroids accounting for the lower bound. Taking this further, in [13], D. Mayhew,
M. Newman, D. Welsh, and G. Whittle have conjectured that, asymptotically, almost all matroids are
sparse paving.
The five basis-exchange conjectures treated in this paper, all of which have been open for decades
and have been proven for only a few classes of matroids, are part of the circle of ideas that revolve
around the well-known symmetric basis-exchange property: for any bases B 1 , B 2 of a matroid M, if
b1 ∈ B 1 − B 2 , then, for some b2 ∈ B 2 − B 1 , both ( B 1 − b1 ) ∪ b2 and ( B 2 − b2 ) ∪ b1 are also bases of M.
The first conjecture concerns the basis pair graph, G ( M ), of a matroid M, which is defined as
follows. The vertices of G ( M ) are the ordered triples ( A 1 , A 2 , A 3 ) of subsets of E ( M ) where A 1 and
A 2 are disjoint bases of M and A 3 is E ( M ) − ( A 1 ∪ A 2 ). (Thus, the inequality | E ( M )|  2r ( M ) must
hold in order for G ( M ) to have any vertices.) Two vertices, say A = ( A 1 , A 2 , A 3 ) and B = ( B 1 , B 2 , B 3 ),
of G ( M ) are adjacent if B can be obtained from A by switching some pair of elements in two different
sets in A, that is, if

| A 1 − B 1 | + | A 2 − B 2 | + | A 3 − B 3 | = 2.

If E ( M ) is the disjoint union of two bases of M, then G ( M ) is isomorphic to the basis–cobasis graph
studied by R. Cordovil and M. Moreira [2]. The following conjecture was posed by M. Farber [3], who
proved it for transversal matroids. (In [4], M. Farber, B. Richter, and H. Shank proved it for graphic
and cographic matroids.)

Conjecture 1.1. The basis pair graph of any matroid is connected.

The second conjecture involves a family of graphs that we can associate with a matroid. Fix an
integer k  2. Let M be a matroid of rank r and let S be a multiset of size kr with elements in E ( M ).
Define the graph G M ( S ) as follows: the vertices of G M ( S ) are all multisets of k bases of M whose
multiset union is S; two vertices are adjacent if one can be obtained from the other by one symmetric
exchange among one pair of bases in one of the vertices. Thus, vertices A = { A 1 , A 2 , . . . , A k } and B =
{ B 1 , B 2 , . . . , B k } are adjacent if, for some bases B i , B j ∈ B and elements bi ∈ B i − B j and b j ∈ B j − B i ,
we obtain A from B by replacing B i by ( B i − b i ) ∪ b j and replacing B j by ( B j − b j ) ∪ b i . (This graph
may be empty.) The conjecture below is due to N. White [20, Conjecture 12].

Conjecture 1.2. For any matroid M and multiset S of size kr ( M ) with elements in E ( M ) and with k  2, the
graph G M ( S ) is connected.

Conjecture 1.2 is sometimes cast in terms of toric ideals. A routine argument shows that the con-
jecture holds for M if and only if it holds for M ∗ . It has been shown for graphic (and so for cographic)
matroids by J. Blasiak [1] and for matroids of rank at most three (and so for matroids of nullity at
most three) by K. Kashiwabara [11]. J. Herzog and T. Hibi [7] have shown that Conjecture 1.2 is equiv-
alent to its counterpart for discrete polymatroids. J. Schweig [19] has proven the counterpart of the
conjecture for certain discrete polymatroids.
While Conjecture 1.2 is the most well-known of the three parts of [20, Conjecture 12], the next
conjecture is the strongest of the three. Consider the graph G M ( S ) in which k-tuples of bases replace
multisets of bases. Thus, its vertices are all k-tuples of bases of M whose multiset union is S; ver-
tices A = ( A 1 , A 2 , . . . , A k ) and B = ( B 1 , B 2 , . . . , B k ) are adjacent if, for some distinct integers i and
j in {1, 2, . . . , k} and some b i ∈ B i − B j and b j ∈ B j − B i , we obtain A from B by replacing B i by
( B i − bi ) ∪ b j and replacing B j by ( B j − b j ) ∪ bi .
8 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15

Conjecture 1.3. For any matroid M and multiset S of size kr ( M ) with elements in E ( M ) and with k  2, the
graph G M ( S ) is connected.

In Theorem 2.5, we show that Conjecture 1.3 holds for a matroid M if Conjecture 1.2 holds for M
and Conjecture 1.1 holds for all of its minors. Since Conjectures 1.1 and 1.2 have been proven for
graphic and cographic matroids, we therefore get the (apparently new) result that Conjecture 1.3 also
holds for such matroids. Likewise, since we prove that Conjectures 1.1 and 1.2 hold for sparse paving
matroids, Conjecture 1.3 therefore also holds for these matroids. It also holds for matroids of rank or
corank at most three (Corollary 2.6). As with Conjecture 1.2, Conjecture 1.3 holds for M if and only if
it holds for M ∗ .
The fourth conjecture was made by Y. Kajitani, S. Ueno, and H. Miyano [10]. A matroid M is
cyclically orderable if there is a cyclic permutation (a1 , a2 , . . . , an ) of E ( M ) in which each set of r ( M )
cyclically-consecutive elements is a basis of M.

Conjecture 1.4. A matroid M is cyclically orderable if and only if, for all nonempty subsets A of E ( M ),
 
r ( M )| A |  r ( A ) E ( M ). (1.2)

A counting argument shows that inequality (1.2) holds if M is cyclically orderable. J. van den Heuvel
and S. Thomassé [8] proved Conjecture 1.4 when r ( M ) and | E ( M )| are relatively prime.
The fifth conjecture was raised as a problem by H. Gabow [5] and stated as a conjecture by R. Cor-
dovil and M. Moreira [2]. To match our work below, we state it in the case of disjoint bases; it is easy
to show that this implies its counterpart for arbitrary bases.

Conjecture 1.5. If B 1 and B 2 are disjoint bases of a rank-r matroid M, then some cycle (b1 , b2 , . . . , br , br +1 ,
. . . , b2r ) has B 1 = {b1 , b2 , . . . , br } and B 2 = {br +1 , br +2 , . . . , b2r }, and has each set of r cyclically-consecutive
elements being a basis of M.

It is not hard to show that if this conjecture holds for M, then it holds for M ∗ and for all mi-
nors of M. H. Gabow [5] noted that the conjecture holds for transversal matroids. It has also been
proven for graphic matroids [2,10]. A. de Mier [16] observed that this conjecture holds for strongly
base-orderable matroids. Recall that M is strongly base-orderable if for each pair of bases B 1 and B 2
of M, there is a bijection φ : B 1 → B 2 such that for every subset X ⊆ B 1 , both ( B 1 − X ) ∪ φ( X ) and
( B 2 − φ( X )) ∪ X are bases. If M is strongly base-orderable, then listing the elements of B 1 in any or-
der followed by their images under φ , in the corresponding order, gives the required cycle. The class
of strongly base-orderable matroids is both minor-closed and dual-closed, and it strictly contains the
class of all gammoids (which include transversal matroids).
As we noted above, Conjecture 1.3 is stronger than Conjecture 1.2. The only other known im-
plications among these conjectures appear to be those mentioned above (namely, Theorem 2.5 and
Corollary 2.6).
In this paper we prove Conjectures 1.1–1.5 in the special case of sparse paving matroids.
Our notation follows J. Oxley [17]. The symmetric difference, ( X − Y ) ∪ (Y − X ), of two sets X and
Y is denoted by X Y . We let [n] denote the set {1, 2, . . . , n}.

2. Proofs of Conjectures 1.1–1.5 in the case of sparse paving matroids

We will use the lemmas below. The first follows easily from the definition of sparse paving.

Lemma 2.1. Let M be a sparse paving matroid of rank r. Let H and B be two r-subsets of E ( M ) with
| H  B | = 2. If H is a circuit-hyperplane of M, then B is a basis.

Although we will not use it, we note that the following strengthening of Lemma 2.1 is easy to
prove: a matroid M of rank r is sparse paving if and only if whenever H and B are two r-subsets
 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15 9

of E ( M ) with | H  B | = 2 and H is not a basis, then B is a basis. (We remark that the analogous
condition on discrete polymatroids winds up being too restrictive to be of interest.)

Lemma 2.2. Let B and B  be distinct bases of a sparse paving matroid M. For a ∈ B − B  and X ⊆ B  − B, there
are at least | X | − 2 elements x ∈ X for which both ( B − a) ∪ x and ( B  − x) ∪ a are bases of M.

Proof. The lemma follows since, by Lemma 2.1, at most one set ( B − a) ∪ x with x ∈ X , and at most
one set ( B  − x ) ∪ a with x ∈ X , is a circuit-hyperplane. 2

We now turn to Conjecture 1.1.

Theorem 2.3. Conjecture 1.1 holds for sparse paving matroids.

Proof. We first prove the result when E ( M ) is the disjoint union of two bases; we will then reduce
the general case to this one. In this case, vertices have the form ( B 1 , B 2 , ∅), which we simplify to
( B 1 , B 2 ) in the next two paragraphs. We must show that for each pair ( A 1 , A 2 ) and ( B 1 , B 2 ) of ver-
tices in G ( M ) with | A 1  B 1 |  4, there is a path between them. For this, it suffices to show that there
is a path from ( B 1 , B 2 ) to a vertex ( B 1 , B 2 ) with | A 1  B 1 | < | A 1  B 1 |.
If | B 1 − A 1 |  3, then fix x ∈ B 1 − A 1 and set X = A 1 − B 1 . We have | X |  3 and X ⊆ B 2 , so,
by Lemma 2.2, the pair (( B 1 − x) ∪ y , ( B 2 − y ) ∪ x) is a vertex of G ( M ) for some y ∈ X . Also,
| A 1 (( B 1 − x) ∪ y )| < | A 1  B 1 |, as needed.
In the remaining case, | B 1 − A 1 | = 2, let B 1 − A 1 = {b1 , b2 } and A 1 − B 1 = {a1 , a2 }. Thus, a1 , a2 ∈ B 2 .
If any of the following four symmetric exchanges yields only bases, it would provide the desired vertex
( B 1 , B 2 ) adjacent to ( B 1 , B 2 ):

(a) ( B 1 − b 1 ) ∪ a1 and ( B 2 − a1 ) ∪ b 1 ,
(b) ( B 1 − b 1 ) ∪ a2 and ( B 2 − a2 ) ∪ b 1 ,
(c) ( B 1 − b 2 ) ∪ a1 and ( B 2 − a1 ) ∪ b 2 ,
(d) ( B 1 − b 2 ) ∪ a2 and ( B 2 − a2 ) ∪ b 2 .

Thus, we may assume that each pair contains a circuit-hyperplane. By symmetry, we may assume that
( B 1 − b1 ) ∪ a1 is a circuit-hyperplane; then ( B 1 − b1 ) ∪ a2 and ( B 1 − b2 ) ∪ a1 are bases by Lemma 2.1, so
( B 2 − a2 ) ∪ b1 and ( B 2 − a1 ) ∪ b2 are circuit-hyperplanes; thus, ( B 2 − a2 ) ∪ b2 is a basis by Lemma 2.1,
so ( B 1 − b2 ) ∪ a2 is a circuit-hyperplane. For all four sets just identified to be circuit-hyperplanes,
we must have r ( M )  3, so there is an element x in B 1 ∩ A 1 . By comparison with the four known
circuit-hyperplanes, it follows that each set in the following symmetric exchanges is a basis:

(e) ( B 1 − x) ∪ a1 and ( B 2 − a1 ) ∪ x,
(f) B 1 = ( B 1 − {x, b2 }) ∪ {a1 , a2 } and B 2 = ( B 2 − {a1 , a2 }) ∪ {x, b2 }.

Since ( B 1 , B 2 ) is adjacent to ( A 1 , A 2 ), the needed path from ( B 1 , B 2 ) to ( A 1 , A 2 ) exists.

In the general case, for two vertices A = ( A 1 , A 2 , A 3 ) and ( B 1 , B 2 , B 3 ) of G ( M ), we will show that
there is a path in G ( M ) from A to a vertex of the form (C 1 , C 2 , B 3 ); the theorem then follows by
applying the case just treated to the basis pair graph of M \ B 3 . (Recall that the third set in these
triples need not be a basis.)
Assume | A 3  B 3 |  4. By symmetry, we may assume | A 1 ∩ B 3 |  1; fix some a1 ∈ A 1 ∩ B 3 .
Since M is sparse paving, the hyperplane cl( A 1 − a1 ) contains at most one element in A 3 − B 3 , so
A 1 = ( A 1 − a1 )∪ a3 is a basis for some a3 ∈ A 3 − B 3 . The vertex ( A 1 , A 2 , A 3 ), where A 3 = ( A 3 − a3 )∪ a1 ,
is adjacent to A and has | A 3  B 3 | < | A 3  B 3 |.
By iterating the argument above, it now suffices to treat the case | A 3  B 3 | = 2. Let A 3 − B 3 = {a3 }
and B 3 − A 3 = {b3 }. We may assume b3 ∈ A 1 . If ( A 1 − b3 ) ∪ a3 is a basis of M, then the claim holds,
so assume instead that this set is a circuit-hyperplane. By symmetrically exchanging any element
a1 ∈ A 1 − b3 with some element a2 ∈ A 2 , we get a vertex (( A 1 − a1 ) ∪ a2 , ( A 2 − a2 ) ∪ a1 , A 3 ) that is
10 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15

adjacent to A and in which, by Lemma 2.1, we can exchange b3 in ( A 1 − a1 ) ∪ a2 with a3 in A 3 , which

completes the proof of the claim and so of the theorem. 2

We now turn to Conjecture 1.2.

Theorem 2.4. Conjecture 1.2 holds for sparse paving matroids.

Proof. Let M be a sparse paving matroid. We prove that G M ( S ) is connected by induction on k, where
| S | = kr ( M ). The base case k = 1 is trivial: G M ( S ) is connected since it has at most one vertex. For
k  2, we claim that for any two vertices

A = { A 1 , A 2 , . . . , A k } and B = { B 1 , B 2 , . . . , B k }

of G M ( S ), there are (possibly trivial) paths from A to some vertex { A 1 , A 2 , . . . , A k } and from B
to some vertex { B 1 , B 2 , . . . , B k } with A 1 = B 1 . Proving this claim gives the result by induction
since having a path from A to B in G M ( S ) follows from having a path from { A 2 , A 3 , . . . , A k } to
{ B 2 , B 3 , . . . , B k } in G M ( S − A 1 ), where S − A 1 is the multiset difference. List the sets in A and B so
that | A 1  B 1 |  | A h  B j | for all h, j ∈ [k]. Set | A 1  B 1 | = 2i. To prove the claim, it suffices to show
that if i > 0, then

(∗) there is a path from B to a vertex { B 1 , B 2 , . . . , B k } with | A 1  B 1 | < 2i.

Set A 1 − B 1 = {a1 , a2 , . . . , ai } and B 1 − A 1 = {b1 , b2 , . . . , b i }. By symmetry, we may assume that

the sum of the multiplicities of the elements in A 1 − B 1 in S is at least as large as the corresponding
sum for B 1 − A 1 . It follows that some basis in B , say B 2 , has more elements from A 1 − B 1 than from
B 1 − A 1 . We consider several options for B 2 .
For the case i  3, first assume B 2 ∩ ( B 1 − A 1 ) = ∅. We may assume a1 ∈ B 2 . Apply Lemma 2.2 with
x = a1 and X = B 1 − A 1 (so | X |  3): for some bh ∈ B 1 − A 1 , both ( B 1 − bh ) ∪ a1 and ( B 2 − a1 ) ∪ bh
are bases, so statement (∗) follows.
Now, along with i  3, assume | B 2 ∩ ( A 1 − B 1 )|  3. Let X = B 2 ∩ ( A 1 − B 1 ). Since B 2 has more
elements from A 1 − B 1 than from B 1 − A 1 , some element in B 1 − A 1 , say b1 , is not in B 2 . Apply
Lemma 2.2 to B 1 and B 2 with x = b1 and X : for some ah ∈ X , both ( B 1 − b1 ) ∪ ah and ( B 2 − ah ) ∪ b1
are bases. Statement (∗) now follows.
We now address the case with B 2 ∩ ( A 1  B 1 ) = {a1 , a2 , b3 }, thereby completing the argument for
i  3. If we can symmetrically exchange one of a1 , a2 in B 2 for one of b1 , b2 in B 1 to get bases,
then statement (∗) holds. Assume that none of these four symmetric exchanges yields only bases. An
argument like that in the third paragraph of the proof of Theorem 2.3 shows that we may assume that

( B 1 − b 1 ) ∪ a1 , ( B 2 − a2 ) ∪ b 1 , ( B 2 − a1 ) ∪ b 2 , and ( B 1 − b 2 ) ∪ a2

are circuit-hyperplanes. In order to have | A 1  B 1 |  | A 1  B 2 | given that B 2 ∩ ( A 1  B 1 ) is {a1 , a2 , b3 },

there must be an element, say y, in B 2 − ( A 1 ∪ B 1 ). From Lemma 2.1 and the circuit-hyperplanes
above, we have that ( B 1 − b1 ) ∪ y and ( B 2 − y ) ∪ b1 are bases, as are ( B 1 − {b1 , b2 }) ∪ { y , a1 } and
( B 2 − { y , a1 }) ∪ {b1 , b2 }. Statement (∗) now follows, which completes the argument for i  3.
Now assume i = 2. By symmetry, there are two cases: B 2 ∩ {b1 , b2 } is either ∅ or {b1 }. First as-
sume B 2 ∩ {b1 , b2 } = ∅. We may assume a1 ∈ B 2 . If a1 in B 2 can be symmetrically exchanged with
either b1 or b2 in B 1 to yield two bases, then statement (∗) holds, so assume this fails. By sym-
metry, H 1 = ( B 1 − b1 ) ∪ a1 and H 2 = ( B 2 − a1 ) ∪ b2 can be assumed to be circuit-hyperplanes. Since
| A 1  B 1 |  | A 1  B 2 |, there are at least two elements, say z2 and z3 , in B 2 − A 1 . By Lemma 2.1, either
( B 2 − z2 ) ∪ b1 or ( B 2 − z3 ) ∪ b1 is a basis; assume the former is. Comparison with H 1 shows that
( B 1 − b1 ) ∪ z2 and ( B 1 − {b1 , b2 }) ∪ { z2 , a1 } are bases; similarly, ( B 2 − { z2 , a1 }) ∪ {b1 , b2 } is a basis by
comparison with H 2 . Statement (∗) now follows.
We now address the case with B 2 ∩ {b1 , b2 } = {b1 }, thus completing the argument for i = 2. Note
that B 2 must also contain a1 and a2 . Statement (∗) holds if b2 in B 1 can be symmetrically exchanged
 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15 11

with either a1 or a2 in B 2 to yield two bases. If neither exchange yields only bases, then, by symmetry,
we may assume that H 1 = ( B 1 − b2 ) ∪ a1 and H 2 = ( B 2 − a2 ) ∪ b2 are circuit-hyperplanes. At least
two elements in A 1 ∩ B 1 , say x3 and x4 , are not in B 2 since | A 1  B 1 |  | A 1  B 2 |. At least one of
( B 2 − a1 ) ∪ x3 and ( B 2 − a1 ) ∪ x4 is a basis by Lemma 2.1; assume the first is. Now ( B 1 − x3 ) ∪ a1 is a
basis by comparison with H 1 . The sets ( B 1 − {x3 , b2 }) ∪ {a1 , a2 } and ( B 2 − {a1 , a2 }) ∪ {x3 , b2 } are also
bases by comparison with H 1 and H 2 , respectively. It follows that statement (∗) holds. This completes
the argument for i = 2.
Finally, assume i = 1, so A 1 − B 1 = {a1 } and B 1 − A 1 = {b1 }. Thus, B 2 contains a1 and not b1 . Let
X = B 2 − a1 . If X ∪ b1 is a basis (as it must be if k is 2), then exchanging a1 and b1 in B 2 and B 1
shows that statement (∗) holds. Thus, assume k  3 and

(A) X ∪ b1 is a circuit-hyperplane.

If 3  h  k and b1 ∈ / B h , and if there is an element y ∈ X − B h , then there is a z ∈ B h − B 2 for which

both ( B h − z) ∪ y and ( B 2 − y ) ∪ z are bases; from Lemma 2.1 and statement (A), it follows that we can
symmetrically exchange a1 in ( B 2 − y ) ∪ z with b1 in B 1 to get two bases, which yields statement (∗).
Thus, we may assume

(B) each basis B h contains either b1 or all of X .

If B h ∩ {a1 , b1 } = {b1 } for some h with 3  h  k, then the assumption about the multiplicities of a1
and b1 implies that B h ∩ {a1 , b1 } = {a1 } for some h with 3  h  k. Symmetrically exchange a1 in
B h − B h for some z ∈ B h − B h to get bases; since B h − a1 is X by statement (B), statement (A) gives
z = b1 . Thus, we may assume

(C) for 3  h  k, if b1 ∈ B h , then a1 ∈ B h .

Assume 3  h  k and a1 , b1 ∈ B h . If | B 2  B h |  4, then for x ∈ ( B h − b1 ) − B 2 , we can symmetrically

exchange x ∈ B h with some y ∈ B 2 (which cannot be a1 ) to yield two bases; with statement (A), this
allows us to exchange b1 in B 1 with a1 in ( B 2 − y ) ∪ x to yield statement (∗). Thus, we may assume

(D) if a1 , b1 ∈ B h , then | B 2  B h | = 2.

The proof is completed by showing that statements (A)–(D) yield a contradiction. Consider the  multi-
sets  A = {{a1 }, A 2 , A 3 , . . . , A k } and B = {{b1 }, B 2 , B 3 , . . . , B k } of sets. Their multiset unions, A ∈A A
and B ∈B B, are equal. Let b1 have multiplicity t + 1 in these unions. Statements (B)–(D) imply that
the sum of the multiplicities of the elements in X in the sets in B is | X |(k − t − 1) + (| X | − 1)t, that
is, | X |(k − 1) − t. By statement (A), X ∪ b1 is not in A , so the sum of the multiplicities of the elements
in X in the sets in A is at most  | X |(k − t − 
2) + (| X | − 1)(t + 1), that is, | X |(k − 1) − t − 1, which, as
desired, contradicts the equality A ∈A A = B ∈B B. 2

We now prove a general connection between Conjectures 1.1, 1.2, and 1.3.

Theorem 2.5. Let M be a matroid for which the basis pair graph of each of its minors is connected. For k  2,
let S be a multiset of size kr ( M ) with elements in E ( M ). If G M ( S ) is connected, then so is G M ( S ).

Proof. Since G M ( S ) is connected, to show that G M ( S ) is connected it suffices to show that for each
vertex A = ( A 1 , A 2 , . . . , A k ) of G M ( S ) and each permutation σ of [k], there is a path in G M ( S ) from A
to Aσ = ( A σ (1) , A σ (2) , . . . , A σ (k) ). Since every permutation is a composition of transpositions, we focus
on a transposition σ , say permuting i and j with i < j. The desired result follows if we show that
there is a path from A to Aσ in which all bases but the i-th and j-th are fixed. This follows by
noting that the sequence of symmetric exchanges that gives a path from ( A i − A j , A j − A i , ∅) to
( A j − A i , A i − A j , ∅) in the basis pair graph of the minor M |( A i ∪ A j )/( A i ∩ A j ) also gives the desired
path from A to Aσ in G M ( S ). 2
12 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15

Corollary 2.6. For any minor-closed class of matroids for which Conjecture 1.1 holds, Conjectures 1.2 and 1.3
are equivalent. In particular, Conjecture 1.3 holds for sparse paving, graphic, and cographic matroids, and for
matroids of rank or corank at most three.

Proof. The assertions about the first three classes of matroids follow from our results on sparse
paving matroids and from the results on graphic and cographic matroids that we mentioned in the
Introduction. For the remaining two assertions, by duality it suffices to treat matroids of corank at
most three and to show that Conjecture 1.1 holds for them. If M has corank at most three, then G ( M )
can be nonempty (and so potentially disconnected) only when r ( M )  3; thus, | E ( M )|  6. It is rou-
tine to show that all matroids that satisfy these conditions are either graphic or transversal; since
Conjecture 1.1 is known for those classes of matroids, the result follows. 2

For Conjecture 1.4, we start with a definition and a lemma. A k-interval in a cycle σ is a set of k
cyclically-consecutive elements, that is, {x, σ (x), σ 2 (x), . . . , σ k−1 (x)} for some x.

Lemma 2.7. Let M be a rank-r sparse paving matroid on n elements. If 2r  n, then, over all cycles on E ( M ),
the average number of r-intervals that are circuit-hyperplanes of M is less than two.

Proof. Let b( M ) and ch( M ) be, respectively, the numbers of bases and circuit-hyperplanes of M. By
focusing on circuit-hyperplanes, it follows that the average of interest is

ch( M )r !(n − r )!
.
(n − 1)!

The desired result follows easily from this expression, the assumed inequality, 2r  n, and the in-
equality
 
1 n
ch( M )  , (2.1)
n−r +1 r

which is a consequence of Theorem 4.8 in [15]. (Alternatively, to get inequality (2.1), consider the
pairs ( H , B ) consisting of a circuit-hyperplane H and basis B of M with | H  B | = 2; the inequality
follows by noting that each circuit-hyperplane
 is in r (n − r ) such pairs, each basis is in at most r such
n
pairs, and b( M ) + ch( M ) = r .) 2

Theorem 2.8. Conjecture 1.4 holds for sparse paving matroids.

Proof. As noted after Conjecture 1.4, inequality (1.2) holds in every cyclically orderable matroid. The
conjecture is easy to verify for all sparse paving matroids that have rank or nullity at most two
(this includes all disconnected sparse paving matroids, i.e., U 0,n , U n,n , U n−1,n ⊕ U 1,1 , U 1,n ⊕ U 0,1 , and
U 1,2 ⊕ U 1,2 ; this also includes all cases in which inequality (1.2) fails), so below we assume that M
has rank and nullity at least three.
We may assume E ( M ) = [n]. For a cycle σ on E ( M ), all r ( M )-intervals in σ are bases of M if
and only if their complements, all r ( M ∗ )-intervals in σ , are bases of M ∗ , so, by replacing M by
M ∗ if needed, we may assume that 2r  n where r = r ( M ). By Lemma 2.7, for some cycle, say
σ1 = (1, 2, . . . , n), on E ( M ), at most one of its r-intervals is a circuit-hyperplane. We may assume
there is such an interval, say

H 1 = {4, 5, . . . , r + 3},

otherwise the desired conclusion holds.

 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15 13

Consider σ2 = (1, 2, 4, 3, 5, . . . , n). (To aid the reader, we underline the entries that differ from σ1 .)
Only two of its r-intervals differ from their counterparts in σ1 , namely, {3, 5, 6, . . . , r + 3}, which is a
basis (use Lemma 2.1 with H 1 ), and

H 2 = {n − r + 4, . . . , n, 1, 2, 4}.

If H 2 is a basis, then σ2 is the cycle we want. Thus, assume that H 2 is a circuit-hyperplane.

We repeatedly apply this type of argument below. For brevity, for each cycle we list its r-intervals
that differ from their counterparts in σ1 and, when possible, the circuit-hyperplanes that, with
Lemma 2.1, show that these intervals are bases. For brevity, we omit the r-interval {i , 5, 6, . . . , r + 3},
with i = 4, which is a basis (compare it to H 1 ). Since the permutations σi below differ from σ1 in at
most four consecutive places, the assumption that the nullity of M is at least three implies that an
r-interval in σi cannot differ from its counterpart in σ1 at both ends.
Consider σ3 = (1, 3, 4, 2, 5, . . . , n). The relevant intervals are

 {4, 2, 5, 6, . . . , r + 2} (compare to H 1 ),
 {n − r + 4, . . . , n, 1, 3, 4} (compare to H 2 ), and

H 3 = {n − r + 3, . . . , n, 1, 3}.

Thus, σ3 has the desired properties unless H 3 is a circuit-hyperplane, so we assume it is.

Consider σ4 = (1, 4, 3, 2, 5, . . . , n). The relevant intervals are

 {n − r + 4, . . . , n, 1, 4, 3} and {n − r + 3, . . . , n, 1, 4} (compare to H 3 ), and

H 4 = {3, 2, 5, 6, . . . , r + 2}.

Thus, σ4 has the desired properties unless H 4 is a circuit-hyperplane, so we assume it is.

Consider σ5 = (3, 4, 1, 2, 5, . . . , n). The relevant intervals are

 {1, 2, 5, 6, . . . , r + 2} (compare to H 4 ),
 {n − r + 4, . . . , n, 3, 4, 1} (compare to H 2 ),
 {n − r + 3, . . . , n, 3, 4} and {n − r + 2, . . . , n, 3} (compare to H 3 ), and

H 5 = {4, 1, 2, 5, 6, . . . , r + 1}.

Thus, σ5 has the desired properties unless H 5 is a circuit-hyperplane, so we assume it is.

Consider σ6 = (4, 3, 1, 2, 5, . . . , n). The relevant intervals are

 {1, 2, 5, 6, . . . , r + 2} (compare to H 4 ),
 {3, 1, 2, 5, 6, . . . , r + 1} (compare to H 5 ),
 {n − r + 4, . . . , n, 4, 3, 1} (compare to H 2 ),
 {n − r + 3, . . . , n, 4, 3} (compare to H 3 ), and

H 6 = {n − r + 2, . . . , n, 4}.

Thus, σ6 has the desired properties unless H 6 is a circuit-hyperplane, so we assume it is.

Finally, consider σ = (2, 3, 4, 1, 5, . . . , n). The relevant intervals are

 {4, 1, 5, 6, . . . , r + 2} (compare to H 1 ),
 {3, 4, 1, 5, 6, . . . , r + 1} (compare to H 5 ),
 {n − r + 4, . . . , n, 2, 3, 4} (compare to H 2 ),
14 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15

 {n − r + 3, . . . , n, 2, 3} (compare to H 3 ), and
 {n − r + 2, . . . , n, 2} (compare to H 6 ).

Thus, σ has the desired properties, which completes the proof. 2

We now turn to Conjecture 1.5.

Theorem 2.9. Conjecture 1.5 holds for sparse paving matroids.

Proof. Consider disjoint bases B = {b1 , b2 , . . . , br } and C = {c 1 , c 2 , . . . , cr } of a sparse paving ma-

troid M. By the basis-exchange property, we may assume that in the cycle

σ = (b1 , b2 , . . . , br , c1 , c2 , . . . , cr ),

every r-interval of the form {b i , b i +1 , . . . , br , c 1 , . . . , c i −1 } is a basis; such cycles are said to start prop-
erly. We say that a problem occurs at c i if {c i , c i +1 , . . . , cr , b1 , . . . , b i −1 } is not a basis; clearly, i > 1. We
will show how, if a problem occurs at c i , then we can switch a few elements so that the number of
problems decreases and the cycle starts properly; iterating this procedure produces the desired cycle.
First assume 1 < i < r. We will show that one of the following cycles starts properly and has fewer
problems (we underline the few elements that are permuted):

σ1 = (b1 , b2 , . . . , br , c1 , c2 , . . . , c i , c i−1 , . . . , cr ),
σ2 = (b1 , b2 , . . . , bi , bi−1 , . . . , br , c1 , c2 , . . . , cr ),
σ3 = (b1 , b2 , . . . , br , c1 , c2 , . . . , c i+1 , c i−1 , c i , . . . , cr ).

Since S 0 = {c i , c i +1 , . . . , cr , b1 , . . . , b i −1 } is a circuit-hyperplane, Lemma 2.1 implies that

{c i −1 , c i +1 , . . . , cr , b1 , . . . , bi −1 } is a basis. Only one other r-interval in σ1 differs from its counter-
part in σ , namely, S 1 = {b i , . . . , br , c 1 , . . . , c i −2 , c i }, so it follows that σ1 starts properly and has fewer
problems than σ unless S 1 is a circuit-hyperplane. Assume S 1 is a circuit-hyperplane. Only two r-
intervals in σ2 differ from their counterparts in σ ; of these, the set {c i , c i +1 , . . . , cr , b1 , . . . , b i −2 , b i } is
a basis by Lemma 2.1 (compare it to S 0 ); if its complement, S 2 = {b i −1 , b i +1 , . . . , br , c 1 , . . . , c i −1 }, is
a basis, then σ2 starts properly and has fewer problems than σ , so we may assume that S 2 is also a
circuit-hyperplane. Four r-intervals in σ3 differ from their counterparts in σ , namely,

T 1 = {c i −1 , c i , c i +2 , . . . , cr , b1 , . . . , b i −1 }, T 2 = {c i , c i +2 , . . . , cr , b1 , . . . , b i },

and their complements. Each of these sets is a basis by Lemma 2.1 since each symmetric difference
T 1  S 0 , T 2  S 0 , ( E ( M ) − T 1 ) S 1 , and ( E ( M ) − T 2 ) S 2 has two elements, so σ3 starts properly and
has fewer problems than σ .
Now assume i = r, so S 0 = {cr , b1 , . . . , br −1 } is a circuit-hyperplane. Consider

σ1 = (b1 , b2 , . . . , br , c1 , c2 , . . . , cr , cr −1 ),
σ2 = (b1 , b2 , . . . , br , br −1 , c1 , c2 , . . . , cr ),
σ3 = (b1 , b2 , . . . , br , c1 , c2 , . . . , cr −1 , cr , cr −2 ).

An argument similar to that above shows that σ1 starts properly and has fewer problems than σ
unless S 1 = {br , c 1 , c 2 , . . . , cr −2 , cr } is a circuit-hyperplane; likewise, σ2 starts properly and has fewer
problems than σ unless S 2 = {br −1 , c 1 , c 2 , . . . , cr −1 } is a circuit-hyperplane. Assume both S 1 and S 2
are circuit-hyperplanes. Only four r-intervals in σ3 differ from their counterparts in σ , namely:

T 1 = {cr , cr −2 , b1 , . . . , br −2 }, T 2 = {cr −2 , b1 , . . . , br −1 },

and their complements. These sets are bases since T 1  S 0 , T 2  S 0 , ( E ( M ) − T 1 ) S 2 , and

( E ( M ) − T 2 ) S 1 each have two elements, so σ3 is the desired cycle on B ∪ C . 2
 J.E. Bonin / Advances in Applied Mathematics 50 (2013) 6–15 15

Acknowledgments

I thank Anna de Mier and Jay Schweig for discussions about Conjectures 1.1–1.5 and for helpful
comments about this paper. I thank the referee for useful suggestions about the introductory material.

References

[1] J. Blasiak, The toric ideal of a graphic matroid is generated by quadrics, Combinatorica 28 (2008) 283–297.
[2] R. Cordovil, M.L. Moreira, Bases-cobases graphs and polytopes of matroids, Combinatorica 13 (1993) 157–165.
[3] M. Farber, Basis pair graphs of transversal matroids are connected, Discrete Math. 73 (1989) 245–248.
[4] M. Farber, B. Richter, H. Shank, Edge-disjoint spanning trees: a connectedness theorem, J. Graph Theory 9 (1985) 319–324.
[5] H. Gabow, Decomposing symmetric exchanges in matroid bases, Math. Program. 10 (1976) 271–276.
[6] J. Geelen, P. Humphries, Rota’s basis conjecture for paving matroids, SIAM J. Discrete Math. 20 (2006) 1042–1045.
[7] J. Herzog, T. Hibi, Discrete polymatroids, J. Algebraic Combin. 16 (2002) 239–268.
[8] J. van den Heuvel, S. Thomassé, Cyclic orderings and cyclic arboricity of matroids, J. Combin. Theory Ser. B 102 (2012)
638–646.
[9] M. Jerrum, Two remarks concerning balanced matroids, Combinatorica 26 (2006) 733–742.
[10] Y. Kajitani, S. Ueno, H. Miyano, Ordering of the elements of a matroid such that its consecutive w elements are indepen-
dent, Discrete Math. 72 (1988) 187–194.
[11] K. Kashiwabara, The toric ideal of a matroid of rank 3 is generated by quadrics, Electron. J. Combin. 17 (2010). Research
Paper 28, 12 pp.
[12] D. Knuth, The asymptotic number of geometries, J. Combin. Theory Ser. A 16 (1974) 398–400.
[13] D. Mayhew, M. Newman, D. Welsh, G. Whittle, On the asymptotic proportion of connected matroids, European J. Com-
bin. 32 (2011) 882–890.
[14] D. Mayhew, G. Royle, Matroids with nine elements, J. Combin. Theory Ser. B 98 (2008) 415–431.
[15] C. Merino, S. Noble, M. Ramírez-Ibañez, R. Villarroel, On the structure of the h-vector of a paving matroid, European J.
Combin. 33 (2012) 1787–1799.
[16] A. de Mier, personal communications, 2010.
[17] J.G. Oxley, Matroid Theory, Oxford University Press, Oxford, 1992.
[18] M. Piff, An upper bound for the number of matroids, J. Combin. Theory Ser. B 14 (1973) 241–245.
[19] J. Schweig, Toric ideals of lattice path matroids and polymatroids, J. Pure Appl. Algebra 215 (2011) 2660–2665.
[20] N. White, A unique exchange property for bases, Linear Algebra Appl. 31 (1980) 81–91.