Guest post by James Oxley.

1. Introduction

It is well known that a graph is bipartite if and only if every cycle has an even number of edges and that a connected graph is Eulerian if and only if every vertex has even degree. While we cannot characterize arbitrary matroids in which every circuit has even cardinality, we can characterize binary matroids with this property. For reasons that will be clarified later, such a binary matroid is called affine.  The goal of this note is to prove the following two theorems. The first is due to Jim Geelen and Peter Nelson, the second to Nathan Bowler. A circuit $C$ in a matroid is odd if $|C|$ is odd; otherwise $C$ is even.

Theorem 1.1 (Geelen and Nelson).  Let $M$ be a loopless, connected, non-affine, binary matroid. If a largest odd circuit of $M$ has $k$ elements, then $M$ has no circuit of size exceeding $2k-2$.

Theorem 1.2 (Bowler). Let $n$ and $k$ be integers exceeding two where $k$ is odd. Then there is an integer $N(k,n)$ such that every $3$-connected, non-affine, binary matroid whose largest odd circuit has $k$ elements either has at most $N(k,n)$ elements or has a minor isomorphic to $M(K_{3,n})$. 

Both of these theorems emerged two years ago from discussions following the author’s online seminar talk [7] based on the paper by Chun, Oxley, and Wetzler [2]. The proofs of these theorems were emailed to the author by Jim Geelen and Nathan Bowler.  The matroid terminology used here will follow [6].

2. Background

Recall that the rank-$r$ binary projective geometry $PG(r-1,2)$ is the matroid whose elements are the non-zero vectors of $V(r,2)$, the $r$-dimensional vector space over the two-element field, and whose independent sets are the linearly independent subsets of $V(r,2)$. For example, the matroids $PG(0,2)$, $PG(1,2)$, and $PG(2,2)$ are $U_{1,1}$, $U_{2,3}$, and the Fano matroid. Of course, every simple binary matroid of rank $r$ is isomorphic to a restriction of $PG(r-1,2)$.  The rank-$r$ binary affine geometry $AG(r-1,2)$ is the matroid that is obtained from $PG(r-1,2)$ by deleting one of its hyperplanes. Thus $AG(0,2)$, $AG(1,2)$, and $AG(2,2)$ are $U_{1,1}$, $U_{2,2}$, and $U_{3,4}$. Clearly, $AG(r-1,2)$ is the matroid that is obtained from $V(r,2)$ by deleting a $V(r-1,2)$. The symmetry of $V(r,2)$ means that it does not matter which hyperplane we choose to delete. If we delete the hyperplane consisting of the vectors whose first coordinate is zero, then we see that the elements of $AG(r-1,2)$ are the vectors of $V(r,2)$ whose first coordinate is one. It follows that every circuit of $AG(r-1,2)$ is even. To see that a simple binary matroid $M$ having no odd circuits is a restriction of a binary affine geometry, we proceed as follows. Let $A$ be a binary matrix representing $M$. Adjoin a new row to $A$ consisting entirely of ones to get the matrix $A^+$. Then one easily checks that $M[A] = M[A^+]$. Clearly $M[A^+]$ is a restriction of a binary affine geometry.

Adding elements in parallel to existing elements of a simple binary affine matroid yields another binary matroid in which every circuit is even. We deduce that a binary matroid is affine if and only if it has no loops and its simplification is isomorphic to a restriction of a binary affine geometry. The next theorem, which incorporates this observation, is obtained by combining results of Brylawski [1], Heron [4], and Welsh [8]. A proof of the equivalence of i and ii was outlined above. A proof of the equivalence of i and iii may be found in [6, Proposition 9.4.1]. This equivalence, originally shown by Welsh, generalizes to binary matroids the dual relationship between bipartite and Eulerian graphs.

Theorem 2.1.  The following are equivalent for a non-empty binary matroid $M$.

1. Every circuit of $M$ is even.
2. $M$ is loopless and its simplification is isomorphic to a restriction of a binary affine geometry. 
3. $E(M)$ can be partitioned into cocircuits.

3. Proofs and Consequences

This section presents the proofs of the two main theorems. The proof of Theorem 1.1 will use the next two lemmas. A theta-graph is a graph $G$ consisting of two vertices, $u$ and $v$, and three internally disjoint paths joining them. These three $uv$-paths in $G$ are the series classes of the theta-graph.

Lemma 3.1. In a connected binary matroid $M$ with an element $e$ and an odd circuit $C$, either $E(M) = C$, or $M$ has as a restriction the cycle matroid of a theta-graph that contains the element $e$ and has $C$ as the union of two of its series classes. In particular, $M$ has an odd circuit containing $e$.

Proof. We may assume that $E(M) \neq C$. Let $d$ be an element of $E(M) – C$, and let $D$ be a circuit that contains $d$ and meets $C$ such that $D – C$ is minimal. As $M$ is binary, $C \bigtriangleup D$ is a disjoint union of circuits. The minimality of $D- C$ implies that $C \bigtriangleup D$ is a circuit, so $M|(C \cup D)$ is the cycle matroid of a theta-graph in which $C$ is the union of two of the series classes. To ensure that $e \in C \cup D$, we take $e = d$ when $e \notin C$. Because the series classes in $M|(C \cup D)$ do not all have even cardinality, it follows that $e$ is in an odd circuit of $M$.

Lemma 3.2. Let $M$ be a connected, non-affine binary matroid. If $M|X$ is connected and affine having at least two elements, then $M$ has an odd circuit $D$ that meets both $X$ and $E(M) – X$ for which $D-X$ is a circuit of $M/X$. Moreover, if $X$ is a circuit, then $M|(X\cup D)$ is the cycle matroid of a theta-graph in which exactly two of the series classes have the same parity.

Proof. As $M$ is not affine, it has an odd circuit $C$. Thus, by Lemma 3.1, $M$ has an odd circuit containing $e$. Hence we can take an odd circuit $D$ that meets $X$ for which $D – X$ is minimal. Suppose that $D -X$ is not a circuit of $M/X$. Then $M/X$ has a circuit $D_1$ that is properly contained in $D -X$. Thus $M$ has a circuit $D_2$ that meets $D-X$ in $D_1$ and is contained in $D_1 \cup X$. The choice of $D$ implies that $D_2$ is even. Now $D \bigtriangleup D_2$ is a disjoint union of circuits of $M$. As $|D \bigtriangleup D_2|$ is odd, $D \bigtriangleup D_2$ contains an odd circuit. This odd circuit cannot be contained in $X$, nor does it contain $D-X$. Hence it contradicts the choice of $D$. Thus $D -X$ is a circuit of $M/X$.  It follows immediately that if $X$ is a circuit, then $M|(X\cup D)$ is the cycle matroid of a theta-graph. As each of $D$ and $X$ is the union of exactly two of the series classes in this theta-graph, and $|D|$ and $|X|$ have different parities, it follows that exactly two of the series classes in $M|(X\cup D)$ have the same parity.

Proof of Theorem 1.1.  Let $C$ be a largest circuit of $M$. We may assume that $C$ is even. By Lemma 3.2, $M$ has an odd circuit $D$ that meets $C$ such that $M|(C\cup D)$ is the cycle matroid of a theta-graph. Let the series classes of the theta-graph be $S_1$, $S_2$, and $S_3$ where the first two have the same parity. Then $C = S_1 \cup S_2$.
Suppose that each of $|S_1 \cup S_3|$ and $|S_2 \cup S_3|$ is at most $\tfrac{1}{2}|C|$. Then $|S_1| + |S_2| + 2|S_3| \le |S_1| + |S_2|$. Thus $S_3$ is empty, a contradiction. Hence $|S_1 \cup S_3|$ or $|S_2 \cup S_3|$ exceeds $\tfrac{1}{2}|C|$, and the theorem follows.

A theta-graph with two series classes each having $k – 1$ elements and one having exactly one element shows that the bound in Theorem 1.1 is sharp.  Another immediate consequence of Lemma 3.2 is the following.

Corollary 3.3. Every even circuit in a connected, non-affine, binary matroid is the symmetric difference of two odd circuits.

In a connected matroid $M$ with at least two elements, the sizes of a largest circuit and a largest cocircuit are the circumference $c(M)$ and the cocircumference $c^*(M)$, respectively. When $M$ has an odd circuit, we denote the size of a largest odd circuit by $c_o(M)$; when $M^*$ has an odd circuit, we write $c^*_o(M)$ for $c_o(M^*)$.

Corollary 3.4. Let $M$ be a connected binary matroid having an odd circuit and an odd cocircuit. Then
$$|E(M)| \le 2(c_o(M) – 1)(c^*_o(M) – 1).$$

Proof. By a result of Lemos and Oxley [5], $|E(K)| \le \tfrac{1}{2}c(K)c^*(K)$ for all connected matroids $K$. Combining this with Theorem 1.1 gives the result.

In the next proof, $M({\mathcal W}_m)$ denotes the cycle matroid of a rank-$m$ wheel, while a tipless binary spike of rank $m$ is the vector matroid of the binary matrix $[I_m|I_m^c]$ where $I_m^c$ is the $m \times m$ matrix that is obtained from $I_m$ by replacing each entry by the other element in the two-element field.

Proof of Theorem 1.2. Let $m = \max\{n, 2k -1\}$. By a theorem of Ding, Oporowski, Oxley, and Vertigan [3], there is a number $N(m)$ such that every $3$-connected binary matroid having more than $N(m)$ elements has as a minor one of $M({\mathcal W}_m)$, $M(K_{3,m})$, $M^*(K_{3,m})$, or a tipless binary spike of rank $m$. If $M$ is a non-affine such matroid and its largest odd circuit has $k$ elements, then, by Theorem 1.1, its largest circuit has at most $2k-2$ elements. Each of $M({\mathcal W}_m)$, $M^*(K_{3,m})$, and the tipless binary spike of rank $m$ has an $m$-element circuit, so none of these matroids occurs as a minor of $M$. Hence $M$ has a minor isomorphic to $M(K_{3,m})$. Taking $N(k,n)$ to be $N(m)$, we get the theorem.

Corollary 3.5. For every odd integer $k$ exceeding two, there is an integer $N'(k)$ such that every $3$-connected, non-Eulerian, simple graph whose largest odd bond has $k$ edges has at most $N'(k)$ edges.

Bibliography

[1] T.H. Brylawski, A decomposition of combinatorial geometries, Trans. Amer. Math. Soc. 171 (1972), 235-282.

[2] C. Chun, J. Oxley, K. Wetzler, The binary matroids with no odd circuits of size exceeding five, J. Combin. Theory Ser. B 152 (2022), 80-120.

[3] G. Ding, B. Oporowski, J. Oxley, and D. Vertigan, Unavoidable minors of large $3$-connected binary matroids, J. Combin. Theory Ser. B 66 (1996), 334-360.

[4] A.P. Heron, Some topics in matroid theory, D.Phil. thesis, University of Oxford, 1972.

[5] M. Lemos and J. Oxley, A sharp bound on the size of a connected matroid, Trans. Amer. Math. Soc. 353 (2001), 4039-4056.

[6] J. Oxley, Matroid Theory, Second edition, Oxford University Press, New York, 2011.

[7] J. Oxley, The binary matroids with no odd circuits of size exceeding five, Graphs and Matroids Online Seminar, University of Waterloo, April 17, 2020.

[8] D.J.A. Welsh, Euler and bipartite matroids, J. Combin.Theory 6 (1969), 313-316.

Guest post by Attila Joó.

In this post we report an interesting property of infinite matroids that provides a common generalisation of the papers ‘A Cantor-Bernstein theorem for paths in graphs’ by Diestel and Thomassen [1] and ‘A Cantor-Bernstein-type theorem for spanning trees in infinite graphs’ by Erde, Gollin, Joó, Knappe, and Pitz [2]. This generalisation also leads to some new applications.

Let us reformulate the Cantor-Bernstein theorem in the language of graph theory:

Theorem (Cantor-Bernstein [3]): If $ G=(V_0,V_1;E) $ is a bipartite graph and matching $ I_i $ covers $ V_i $ for $ i\in \{ 0,1 \} $, then $ G $ admits a perfect matching.

Ore discovered the following generalisation of the Cantor-Bernstein theorem which is the extension of the Mendelsohn-Dulmage theorem to infinite graphs:

Theorem (Ore [4]): Let $ G=(V_0,V_1;E) $ be a bipartite graph and let $ I_0, I_1\subseteq E $ be matchings in $ G $. Then there exists a matching $ I $ such that $ V(I)\cap V_i\supseteq V(I_i)\cap V_i $ for $ i\in \{ 0,1 \} $.

Diestel and Thomassen [1] examined a more general graph theoretic setting in which disjoint paths are used to connect two vertex sets:

Theorem (Diestel and Thomassen [1]): Assume that $ G=(V,E) $ is a graph, $ V_0, V_1\subseteq V$ and $ \mathcal{P}_i $ is a system of disjoint $V_0V_1$-paths in $ G $ for $ i\in \{ 0,1 \} $. Then there exists a system of disjoint $ V_0V_1 $-paths $ \mathcal{P} $ with $ V(\mathcal{P})\cap V_i \supseteq V(\mathcal{P}_i)\cap V_i $ for $ i\in \{ 0,1 \} $.

In our paper [2] entitled ‘A Cantor-Bernstein-type theorem for spanning trees in infinite graphs’ we investigated if the existence of a $ \kappa $-packing and a $ \kappa $-covering by spanning trees implies the existence of a $ \kappa $-family of spanning trees which is both, i.e. a $ \kappa $-partition:

Theorem (Erde et al. [2]): Let $ G=(V,E) $ be a graph and let $ \kappa $ be a cardinal. If there are $ \kappa $ many pairwise edge-disjoint spanning trees in $ G $ and $ E $ can be covered by $ \kappa $ many spanning trees, then $ E $ can be partitioned into $ \kappa $ many spanning trees.

We believed that the connection between the two previous theorems above is only analogical but it turned out that the connection is actually stronger. Namely, both of them are special cases of a “Cantor-Bernstein theorem” of infinite matroids. To state this matroidal generalisation we fix some notation. We call a matroid (co)finitary if all of its (co)circuits are finite. Note that the cofinitary matroids are exactly the duals of the finitary ones. Let us denote the class of finitary and cofinitary matroids by $ \mathfrak{F} $ and $\mathfrak{F}^{*} $ respectively. Let $ \mathfrak{F}\oplus \mathfrak{F}^{*} $ be the class of matroids having only finitary and cofinitary components, equivalently that are the direct sums of a finitary and a cofinitary matroid (Bowler called them ‘simple matroids’ in the previous post). For a matroid class $ \mathfrak{C} $, let $ \mathfrak{C}(E) $ denote the set of matroids on edge set $ E $ that are in class $\mathfrak{C} $.

Theorem (J.): For $ i\in \{ 0,1 \} $, let $ M_i\in (\mathfrak{F}\oplus \mathfrak{F}^{*})(E) $ and $ F_i\subseteq E $. Then there exists an $ F\subseteq E $ such that $ \mathsf{span}_{M_i}(F)\supseteq F_i $ and $ \mathsf{span}_{M_i^{*}}(E\setminus F)\supseteq E\setminus F_{1-i} $ for $ i\in \{ 0,1 \} $.

The special case where $F_i$ are common independent sets and the matroids are finite is a theorem of Kundu and Lawler [5]. It is worth to mention that $\mathfrak{F}\oplus\mathfrak{F}^{*}$ cannot be replaced by the class of all matroids in the theorem above because the resulting statement fails under the Continuum Hypothesis. We do not know at this point if a counterexample can be found in set theory ZFC alone. It is also unclear if $\mathfrak{F}\oplus\mathfrak{F}^{*}$ can be replaced by some larger “well-behaved” class of matroids like the so called tame matroids. The latter class consists of those matroids in which the intersection of a circuit and a cocircuit is always finite. This class was intensively studied by Bowler, Carmesin and others. It was shown for example by Carmesin that the bases of a fixed tame matroid have the same size. The analogous statement for arbitrary matroids is independent of ZFC (the positive and negative direction of the independence was proved by Higgs and Bowler & Geschke [6] respectively).

In order to derive the mentioned theorem of Diestel and Thomassen [1] from the theorem above, one needs to choose $M_i$ to be the cycle matroid of the graph obtained by contracting $V_{1-i}$ in $G$ and let $F_i:= E(\mathcal{P}_{i})$. One can show that the set of all the $V_0V_1$-paths in $G[F]$ is suitable as $ \mathcal{P}.$ For the spanning tree partitioning result we need a “packing-covering-partitioning family version” of the previous theorem:

Theorem(J.): For $ i\in \Theta$, let $ M_i\in (\mathfrak{F}\oplus \mathfrak{F}^{*})(E) $ and $ P_i,R_i \subseteq E $ such that $ P_i\cap P_j=\varnothing $ for $ i\neq j$ and $ \bigcup_{i\in \Theta}R_i=E $. Then there are $ T_i\subseteq P_i\cup R_i $ for $ i\in \Theta $ forming a partition of $ E $ such that $ \mathsf{span}_{M_i}(T_i)\supseteq P_i $ and $\mathsf{span}_{M_i^{*}}(E\setminus T_i)\supseteq E\setminus R_i $.

This version can be obtained from the previous form in the same way as the Matroid Intersection Conjecture and the Packing-Covering Conjecture can be derived from each other (see the previous post by Bowler). The special case where $P_i$ and $R_i$ are assumed to be bases of $M_i$ is an earlier theorem by Erde et al. [7] and was conjectured by Bowler. By choosing all $M_i$ to be the cycle matroid of $G$, sets $P_i$ to be the (edge sets of) spanning trees forming a packing and the $R_i$ to be spanning trees forming a covering, the sets $T_i$ form the desired partition.

Finally, we mention two applications. According to an unpublished result of Bowler and Carmesin, for every compact graph-like space (introduced by Thomassen and Vella [8]) the edge sets corresponding to pseudo-cycles are the circuits of a cofinitary matroid. This makes it possible to extend the scope of the mentioned theorem of Diestel and Thomassen to compact graph-like spaces (using pseudo-arcs as a replacement of graph theoretic paths).

The other application corresponds to the Matroid Intersection Conjecture of Nash-Williams (discussed in the previous post by Bowler). Let $M_0, M_1 \in (\mathfrak{F}\oplus \mathfrak{F}^{*})(E) $ be fixed. For an $I\in \mathcal{I}_{M_0}\cap \mathcal{I}_{M_1} $ the following statements are known to be equivalent:

1. There is a partition $ E=E_0\sqcup E_1 $ such that $ I_i:=I\cap E_i $ spans $E_i $ in $M_i $ for $ i\in \{ 0,1 \} $.
2. $I$ is strongly maximal, i.e. $ \left|J\setminus I\right|\leq \left|I\setminus J\right| $ for every $ J\in \mathcal{I}_{M_0}\cap \mathcal{I}_{M_1} $.

If the matroids are finite (i.e. $E$ is finite), then ‘strongly maximal’ is equivalent with ‘of maximal size’ but in general strong maximality is a strictly stronger notion. The Matroid Intersection Conjecture claims the existence of a strongly maximal common independent set and is known to be true if $E$ is countable [11].

For $I, J \in \mathcal{I}_{M_0}\cap \mathcal{I}_{M_1} $ let $ J\trianglelefteq I $ iff $ J\subseteq \mathsf{span}_{M_0}(I)\cap \mathsf{span}_{M_1}(I) $. It is not too hard to see that if $E$ is finite, then the strongly maximal (i.e. maximal sized) common independent sets are exactly the $\trianglelefteq $-maximal ones ($I$ is $\trianglelefteq $-maximal iff $ I \trianglelefteq J $ implies $ J \trianglelefteq I $ for every $J\in \mathcal{I}_{M_0}\cap \mathcal{I}_{M_1} $). According to the second application, if $E$ is countable, then the set of the strongly maximal elements of $\mathcal{I}_{M_0}\cap \mathcal{I}_{M_1}$ is $\trianglelefteq $-cofinal in $\mathcal{I}_{M_0}\cap \mathcal{I}_{M_1}$, however, in contrast of the finite case it might be not even upward $\trianglelefteq $-closed in $\mathcal{I}_{M_0}\cap \mathcal{I}_{M_1}$.

[1] R. Diestel and C. Thomassen, A Cantor-Bernstein theorem for paths in graphs, The American Mathematical Monthly 113 (2006), no. 2, 161–166.
[2] J. Erde, J P. Gollin, A. Joó, P. Knappe, and M. Pitz, A Cantor-Bernstein-type theorem for spanning trees in infinite graphs, Journal of Combinatorial Theory, Series B 149 (2021), 16–22.
[3] G. Cantor, Mitteilungen zur lehre vom transfiniten, Zeitschrift für Philosophie und philosophische Kritik 91 (1987), 81–125.
[4] O. Ore, The theory of graphs, American Mathematical Society, 1962.
[5] S. Kundu and E. L Lawler, A matroid generalization of a theorem of Mendelsohn and Dulmage, Discrete Mathematics 4 (1973), no. 2, 159–163.
[6] N. Bowler and S. Geschke, Self-dual uniform matroids on infinite sets, Proceedings of the American Mathematical Society 144 (2016), no. 2, 459–471.
[7] J. Erde, J. P. Gollin, A. Joó, P. Knappe, and M. Pitz, Base partition for mixed families of finitary and cofinitary matroids, Combinatorica 41 (2021), no. 1, 31–52.
[8] C. Thomassen and A. Vella, Graph-like continua, augmenting arcs, and Menger’s theorem, Combinatorica 28 (2008), no. 5, 595–623.
[9] R. Aharoni and R. Ziv, The intersection of two infinite matroids, Journal of the London Mathematical Society 58 (1998), no. 03, 513–525.
[10] A. Joó, Proof of Nash-Williams’ intersection conjecture for countable matroids, Advances in Mathematics 380 (2021), 107608.
[11] A. Joó, On the Packing/Covering Conjecture of infinite matroids (2021). https://arxiv.org/abs/2103.14881
[12] N. Bowler, Infinite matroids, Habilitation Thesis, 2014. https://www.math.uni-hamburg.de/spag/dm/papers/Bowler_Habil.pdf
[13] A. Joó, On a linking property of infinite matroids (2021). https://arxiv.org/abs/2009.08439.

Guest contributor: Jim Geelen

In this post we consider some problems related to the following classical result (posed by Sylvester in 1893 and proved by Gallai in 1944).

Sylvester-Gallai Theorem: Given any finite set of points in the real plane, not all on a line, we can find a line in the plane that contains exactly two of them.

That is, in matroid-thoretic language, every simple, rank-$3$, real-representable matroid contains a $2$-point line. Geometers have considered a number of related problems, but there does not seem to have been much work on this in the matroid theory community. I’ve not thought about these questions much, and have not done a thorough literature review, but I’m just going to put the problems out there for discussion. I’ll start with a brief discussion about what is known for real-representable, complex-representable, and binary matroids.

In geometry, an ordinary line is a line with exactly two points. This has an obvious generalization to higher rank flats, however, the term “ordinary flat” is used to describe a different, less natural, generalization of ordinary lines, which we define later. So, following conventions in geometry, we will call a flat in a matroid elementary if it is an independent set.

Does every simple real-representable matroid with rank at lest $4$ contain an elementary plane? This was answered in the negative by Motzkin in 1951 who presented two counterexamples; the first being the direct sum of two lines. The second example is more subtle; take $5$ planes in $\mathbb R^3$ in general position; any three of the planes intersect in a point, and these 10 points give a counterexample.  These were thought to be the only examples, until 1971, when Bonnice and Kelly presented an infinite family. Later, Hansen added to the list of known counterexamples.

Problem 1: Classify the simple rank-$4$ real-representable matroids that do not have an elementary plane.

Given the list of known examples, that problem is likely to be quite challenging.

Geometers have a non-trivial generalization of the Sylvester-Gallai Theorem, given below. A flat $F$ of a matroid $M$ is ordinary if $M|F$ has a coloop. Thus, in a simple matroid, elementary lines and ordinary lines are the same thing.

Hansen’s Theorem: For each $r\ge 3$, every simple rank-$r$ real-representable matroid contains an ordinary hyperplane.

The definition of ordinary flats seems a bit contrived, but the result does have interesting consequences.

Corollary 2: Every simple rank-5 real-representable matroid has an elementary plane.

Proof: Suppose that $M$ is a simple rank $5$ real-representable matroid. By Hansen’s Theorem there is an ordinary hyperplane $H$ in $M$. By definition, $H=P\cup \{e\}$ for some plane $P$ and element $\{e\}$. By the Sylvester-Gallai Theorem, $P$ contains an elementary line $\{f,g\}$. Now $\{e,f,g\}$ is an elementary plane.$\Box$

Moreover, the same idea easily generalizes to higher-dimensions.

Corollary 3: For each $k\ge 2$ and $r\ge 2k-1$, if $M$ is a simple rank-$r$ real-representable matroid, then $M$ has an elementary rank-$k$ flat.

Since the direct-sum of $k-1$ triangles does not contain an elementary rank=$k$ flat, this bound is best possible.

Complex-representable matroids

It is easy to construct simple rank-$3$ matroids with no ordinary lines; projective planes and affine planes (over fields of size at least $3$) give examples. It is not as easy to come up with examples that are representable over the complex numbers, however, the ternary affine plane is complex representable. There are, in fact, infinitely many examples known in rank-$3$, but the list of known examples is believed to be complete. Increasing the rank to $4$ yields an ordinary line.

Theorem [Kelly, 1986]: Every simple complex-representable matroid with rank at least $4$ contains an ordinary line.

It is natural to wonder whether Corollary 3 extends to complex-representable matroids. I really like this conjecture, but it is hard to imaging that it is new.

Conjecture 4: For each positive integer $k$ there is an integer $R$, such that, if  $M$ is a simple complex-representable matroid with rank at least $R$, then $M$ contains an elementary rank-$k$ flat.

Binary matroids

I was led to these problems by results on binary matroids that I came across in Kazuhiro Nomoto’s Ph.D thesis. In particular, Peter Nelson and Kazuhiro Nomoto characterized the simple binary matroids with no elementary plane. Their theorem statement is not particularly complicated, so I will state it, but it is not directly relavent to any of the conjectures, so I will not go into any detail.

Theorem [Nelson, Nomoto]: A simple binary matroid $M$  has no elementary plane if and only if it can be obtained via lift-joins from:

• the direct sums of two projective geometries,
• even-plane matroids, and
• complements of triangle-free binary matroids,

Here a matroid is even-plane if each of its planes have an even number of points; these matroids have an unexpected description as roots of quadtratic equations. This class is particularly interesting as it has unbounded critical number. You might want to look at the the paper for a definition of a lift-join of simple binary matroids $N_1$ and $N_2$, but here is a cryptic definition; it is the unique maximal simple binary matroid $M$ spanned by the direct sum of $N_1$ and $N_2$ such that $M/E(N_2)$ is loopless and simplifies to $N_1$.

Since this result is already quite complicated, it would be presumably very hard to characterize the simple binary matroids with no elementary flat of rank $4$. For other finite fields $\mathbb F$ and integers $k\ge 2$, one could try to classify the simple $\mathbb F$-representable matroids with no elementary rank-$k$ flat. The only other instance of this problem that looks approachable is:

Exercise 5: Characterize the simple ternary matroids with no elementary line.

Hint: If you consider a simple ternary matroid as a restriction of a ternary projective geometry then the matroid has an elementary line if and only if its complement does. Show that if there is no elementary line then the matroid and its complement cannot both be spanning.

Matroids in general

The class of matroids with no ordinary line looks extremely complicated, but it would be nice to have some general Ramsey-theoretic tools for addressing Sylvester-Gallai-type problems. For example:

Conjecture 6: For any integer $k\ge 1$ there is an integer $R$ such that if $M$ is a simple matroid with rank at least $R$ then there is a rank-$k$ flat in $M$ such that either $M|F$ has no ordinary lines, or every line in $M|F$ is ordinary.

I don’t know whether I particularly believe this conjecture; it may be the case that $R$ may also depend on the maximum line-length.

Claim: Conjecture 6 implies Conjecture 4.

Proof. Consider the smallest integer $k$ such that Conjecture 3 fails. By Hansen’s Theorem, $k\ge 3$. Consider a simple complex-representable matroid $M$ with huge rank but with no elementary rank-$k$ flats. If Conjecture 5 were true we could find a large-rank flat $F$ in which all of the lines are ordinary. Now, for any $e\in F$, by induction $(M|F)/e$ has a rank-$(k-1)$ elementary flat $F’$. Then $F’\cup\{e\}$ is an elementary rank-$k$ flat in $M$. $\Box$

I will say that a class $\mathcal M$ of matroids satisfies the $k$-Sylvester Property if every simple matroid in $\mathcal M$ with sufficiently large rank contains an elementary rank-$k$ flat.  Note that if $\mathcal M$ does not satisfy the $k$-Sylvester Property, then $\mathcal M$ doe not satisfy the $k’$-sylvester Property for any $k’\ge k$. Using the proof idea from the previous claim, Conjecture 6 also implies:

Conjecture 7:  If $\mathcal M$ is a minor-closed class of matroids that satisfies the $2$-Sylvester property, then $\mathcal M$ satisfies the $k’$-Sylvester property for all $k’\ge 2$.

The following weakening of Conjecture 6 would already imply Conjecture 7.

Conjecture 8: For any integer $k\ge 1$ there is an integer $R$ such that if $M$ is a simple matroid with rank at least $R$ then there is a rank-$k$ flat in $M$ such that either there is a point in $M|F$ that is only on ordinary lines, or no line in $M|F$ is ordinary.

Elementary planes in triangle-free matroids

If $M$ is a triangle-free matroid with no elementary plane, then, for each element $e$ of $M$, the matroid $M/ e$ is simple and has no ordinary line. So results about ordinary lines can be lifted to give results about triangle-free matroids with no ordinary plane. For example, the Sylvester-Gallai Theorem implies that:

Theorem 9: If $M$ is a simple, triangle-free real-representable matroid with rank at least $4$, then $M$ has an elementary plane.

This result is implied by Hansen’s Theorem, but the proof of Hansen’s Theorem is considerably harder.

The next result follows from Kelly’s Theorem.

Theorem10: If $M$ is a simple, triangle-free complex-representable matroid with rank at least $5$, then $M$ has an elementary plane.

Projective geometries are the only simple binary matroids with no ordinary lines. Affine geometries are the only triangle-free, binary, single-element coextensions of binary projective geometries. Therefore:

Theorem [Nelson, Nomoto]: The only simple, triangle-free binary matroids with no elementary plane are the binary affine geometries.

To borrow a nice turn-of-phrase from Joseph Kung, the following conjecture reflects my current state of ignorance.

Conjecture 11: For any finite field $\mathbb F$ with odd characteristic, if $M$ is a simple, triangle-free $\mathbb F$-representable matroid with sufficiently large rank, then $M$ has an elementary plane.

Something much stronger may well hold:

Conjecture 12: For any finite field $\mathbb F$ with odd characteristic and integer $k\ge 3$, if $M$ is a simple, triangle-free $\mathbb F$-representable matroid with sufficiently large rank, then $M$ has an elementary rank-$k$ flat.

Higher girth

A matroid is simple and triangle-free if and only if it has girth at least $4$. If $e$ is an element in a matroid $M$ with girth $g$ and with no elementary rank-$k$ flat, then $M/e$ has girth $\ge g-1$ and has no elementary flat of rank $k-1$. Again this gives us an inductive tool.

Lemma: Every binary matroid with girth $\ge 5$ and rank $\ge 5$ contains an elementary rank-4 flat.

Proof. It suffices to prove the result when the rank and girth are both equal to $5$. Consider a binary matroid $M$ with rank and girth both $5$. Suppose that $M$ has no elementary rank-$4$ flat. For any element $e$ of $M$, the matroid $M/e$ is a simple, rank-$4$ triangle-free binary matroid with no elementary plane; therefore $M/e$ is isomorphic to the binary affine cube AG$(3,2)$. The binary affine cube is also known as the rank-$4$ binary spike, which is self-dual. Therefore $M^*$ is the unique binary single-element extension of the binary affine cube, which is the rank-$4$ binary spike with tip. However, this contradicts the fact that $M$ has girth $5$. $\Box$

Now an easy induction gives:

Theorem 13: For each $k\ge 4$, if $M$ is a binary matroid with rank and girth both at least $k+1$, then $M$ has an elementary rank-$k$ flat.

In fact, we can keep the girth at $5$ if we increase the rank.

Theorem 14: For each $k\ge 4$, if $M$ is a binary matroid  girth at least $5$ and with sufficiently large rank, then $M$ has an elementary rank-$k$ flat.

Proof. Fix a basis $B$, and for each element $e$ outside $B$ let $C_e$ be the unique circuit in $B\cup \{e\}$. We think of $C_e-\{e\}$ as a “hyper-edge” on the vertex set $B$ and apply the hypergraph Ramsey Theorem to get a $k$-element set $I\subseteq B$ such that, for each $t\in\{1,\ldots,k\}$, either all or none of the $t$-element subsets of $I$ are hyper-edges. We may assume that $I$ is not an elementary flat and hence there exists $t\in\{1,\ldots,k\}$ such that all of the $t$-element sets are hyperedges. Then there are two hyperedges whose symmetric difference has size two and hence $M$ has a circuit of size $4$.$\Box$

While it seems a bit wild, maybe we don’t even need representability.

Conjecture 15: For each $k\ge 4$, if $M$ is a matroid  with girth at least $5$ and with sufficiently large rank, then $M$ has an elementary rank-$k$ flat.

If you are not struck by this conjecture, read it again.

While writing this I benefitted from discussions with Nick Brettell, Rutger Campbell, James Davies, Matt Kroeker, Peter Nelson, and Geoff Whittle.

References

• Sylvester-Gallai Theorem, Wikipedia
• A survey of Sylvester’s problem and its generalizations, P. Borwein and W.O.J. Moser, Aequationes Mathematicae, 1990
• Induced binary submatroids, Kazuhiro Nomoto, Ph.D Thesis, University of Waterloo, 2021
• The structure of claw-free binary matroids, P. Nelson and K. Nomoto, arXiv, 2018
• A resolution of the Sylvester-Gallai problem of J.P. Serre, L.M. Kelly, Discrete and Computational Geometry 1, 1986.