betaFeedback

WorksMathematics

Algebraic matroids with non-algebraic duals exist in any positive characteristic

Abstract

We show that for every field KK of positive characteristic there exists an algebraic matroid MM over KK such that the dual matroid M∗M^* is not algebraic over any field. Our conceptual proof relies on the theory of matroids over one-dimensional algebraic groups. As an application, we deduce that the classes of supports of realizable volume polynomials and of realizable covolume polynomials differ in positive characteristic.

Introduction

Matroids were introduced by Whitney [Whi35] to put the similar properties of subsets of linearly independent vectors and trees in a graph into an abstract combinatorial framework. It was already noted earlier by van der Waerden that algebraically independent subsets in a field extension satisfy the same axioms [22]. Concretely, let M=(E,r)M=(E,r) be a matroid with a finite ground set EE and a rank function rr, see e.g. [19].

Definition 1.1. A matroid M=(E,r)M=(E,r) is algebraic over a field KK if there exists a field extension K⊆LK \subseteq L and ei∈Le_i \in L for i∈Ei \in E such that for all I⊆EI \subseteq E it holds that

r(I)=trdeg⁡KK(ei:i∈I).r(I)=\operatorname{trdeg}_{K} K(e_i:i\in I).

Here, trdeg⁡K\operatorname{trdeg}_{K} denotes the transcendence degree over KK and K(ei:i∈I)K(e_i:i\in I) is the smallest subfield of LL containing KK and {ei:i∈I}\{e_i:i\in I\}. Hence, I⊆EI \subseteq E is independent in MM, that is r(I)=∣I∣r(I)=|I|, if and only if the set {ei:i∈I}\{e_i:i\in I\} is algebraically independent over KK.

If a matroid is representable over some field KK, it is also algebraic over KK. Lindström showed in [13] that being algebraic over KK depends only on the characteristic of KK. In characteristic 0, every algebraic matroid over KK is representable over a field extension of KK [11]. In positive characteristic however, Ingleton described a matroid on 11 elements of rank 3 which is algebraic over F2\mathbb{F}_2 and has both the Fano and non-Fano matroids as restrictions; hence it is not representable over any field. It was only shown in 1975 that not all matroids are algebraic: Ingleton and Main proved that the Vámos matroid is not algebraic over any field [10].

In general, algebraic matroids are far less understood than linearly representable matroids. It was recently shown by Boege and Yashfe [5] that the problem of recognizing whether a given matroid is algebraic is undecidable. Algebraic matroids respect several prominent matroid operations. The class of algebraic matroids over a field KK is closed under taking minors, truncations, principal extensions, matroid unions, and matroid intersections [8, 15].

Recall that the dual M∗M^* of a matroid MM is the matroid on the same ground set with bases the complements of bases of MM. This notion is for instance motivated by the orthogonal complement of a vector space and by plane graph duality, since M(G∗)=M∗(G)M(G^*) = M^*(G) for a planar graph GG, where M(G)M(G) is the graphic matroid of a graph GG. Whether the dual of an algebraic matroid is again algebraic in general was recently answered negatively by Larson and Le [14]. However, their counterexample only works over a field of characteristic 2 and the proof relies on a computer calculation. Inspired by their counterexample, we construct counterexamples over every field of positive characteristic and give a conceptual proof. For this we use generalizations of the Fano and non-Fano matroids, called the Lazarson matroid LpL_p and the non-Lazarson matroid NLp\mathrm{NL}_p over F‾p\overline{\mathbb{F}}_p. Let u1,…,up+1,x1,…,xp+1u_1,\ldots,u_{p+1},x_1,\ldots,x_{p+1} be algebraically independent over F‾p\overline{\mathbb{F}}_p. Then LpL_p is represented by

x1,…,xp+1,y1,…,yp+1,zx_1,\ldots,x_{p+1},y_1,\ldots,y_{p+1},z

where z=∑i=1p+1xiz=\sum_{i=1}^{p+1}x_i and yi=z−xiy_i=z-x_i. NLp\mathrm{NL}_p is represented by

u1,…,up+1,v1,…,vp+1,wu_1,\ldots,u_{p+1},v_1,\ldots,v_{p+1},w

where w=∏i=1p+1uiw=\prod_{i=1}^{p+1}u_i and vi=wuiv_i=\frac{w}{u_i}. We show that every algebraic realization of the dual matroids Lp∗L_p^* and NLp∗\mathrm{NL}_p^* is algebraically equivalent to a realization over a one-dimensional algebraic group: over Ga\mathbb{G}_a for Lp∗L_p^*, and over Gm\mathbb{G}_m or an elliptic curve for NLp∗\mathrm{NL}_p^*. It follows that every algebraic matroid NN obtained by gluing Lp∗L_p^* and NLp∗\mathrm{NL}_p^* along a common basis is the intersection of Lp∗L_p^* and NLp∗\mathrm{NL}_p^*, each extended by coloops. Dually, N∗N^* is the union of LpL_p and NLp\mathrm{NL}_p, each extended by loops. We then construct an algebraic matroid MpM_p that has both LpL_p and NLp\mathrm{NL}_p as contractions but is not such a union. Therefore the dual of MpM_p is not algebraic over any field.

Theorem 1.2 (Theorem 6.2). Over any field KK of positive characteristic there is an algebraic matroid MM over KK such that the dual matroid M∗M^* is not algebraic over any field.

Aluffi [1] introduced the notion of covolume polynomials which were further developed in [8]. Matroids whose set of bases forms the support of some multiaffine realizable covolume polynomial over a field KK are exactly those matroids MM whose dual M∗M^* is an algebraic matroid over KK. To any covolume polynomial over KK we can associate a volume polynomial over KK by generalizing matroid duality.

Theorem 1.3 (Theorem 7.3). Let KK be a field of positive characteristic. There exists a realizable covolume polynomial gg over KK such that the support of gg is not the support of any realizable volume polynomial.

AI disclosure

AI disclosure. LLMs were used for proofreading and editing. Claude Opus 5 and 5.5 assisted us in finding the statement and the proof of Theorem 3.3.

Acknowledgements

The author thanks Giles Gardam, June Huh and Lukas Kühne for the many insightful discussions on algebraic matroids. The author was supported by the SPP 2458 ”Combinatorial Synergies”.

Matroids over one-dimensional groups

Throughout the whole article let KK be an algebraically closed field of positive characteristic pp and LL an algebraically closed field extending KK of infinite transcendence degree over KK.

We will use the theory of matroids over a connected one-dimensional algebraic group GG over KK first described by Evans and Hrushovski [7] and further developed in [2]. An algebraic group is said to be connected if it is irreducible. All algebraic groups appearing in this article will be connected. Let EE denote the endomorphism ring of GG. This is a possibly non-commutative ring without zero divisors and an Ore domain. Let QQ be the corresponding division ring of EE. In all the commutative algebraic groups appearing we denote the addition with ⊕\oplus. For more on algebraic groups see [17]. There are the following equivalent definitions for a matroid over a one-dimensional group:

Definition 2.1. (a) Given a closed connected subgroup XX of GnG^n of dimension dd. We write M(X)M(X) for the matroid whose independent sets are all the sets I⊆[n]I \subseteq[n] such that the projection of XX to GIG^I is all of GIG^I.

(b) Given a homomorphism of algebraic groups ψ:Gd→Gn\psi:G^d \to G^n. Then ψ\psi is encoded by a matrix ψ∈En×d\psi\in E^{n \times d}. Let M(ψ)M(\psi) be the matroid defined by the (left) row space of ψ\psi over QQ.

(c) Given a generic LL-point p:Spec⁡L→Gdp:\operatorname{Spec} L \to G^d. The image ψ(p)\psi(p) gives nn points q1,…,qn∈G(L)q_1,\ldots,q_n \in G(L). We denote with M(q1,…,qn)M(q_1,\ldots,q_n) the algebraic matroid realized by those points.

I.e. we have M(X)=M(ψ)=M(q1,…,qn)M(X)=M(\psi)=M(q_1,\ldots,q_n). The equivalence of (a) and (b) follows from [2], Theorem 1.1 and the equivalence of (a) and (c) is easy to see. For any p∈G(L)p \in G(L) we write cl⁡(p)\operatorname{cl}(p) for the algebraic closure of K(p)K(p). Here G(L)=Hom⁡K(Spec⁡L,G)G(L)=\operatorname{Hom}_K(\operatorname{Spec} L,G) and pp corresponds to a point g∈Gg \in G together with an embedding κ(g)↪L\kappa(g) \hookrightarrow L; let K(p)⊆LK(p) \subseteq L be its image.

The one-dimensional algebraic groups over KK are completely classified. Such a group GG is either the additive group GaG_{\mathrm{a}} over KK, the multiplicative group GmG_{\mathrm{m}} over KK or an elliptic curve EE over KK. See [4], Theorem III.10.9 for a classification of affine connected one-dimensional algebraic groups. Thus the possible endomorphism rings are understood quite well. The endomorphism ring of GmG_{\mathrm{m}} or an elliptic curve EE has characteristic 0 while the endomorphism ring of GaG_{\mathrm{a}} has the same characteristic as KK. Matroids over one-dimensional groups share several properties with linear matroids over fields. For example, they can always be extended modularly.

Lemma 2.2. Let GG be a one-dimensional group over KK with endomorphism ring EE and division ring QQ. For any p∈G(L)dp \in G(L)^d generic and any v∈Edv \in E^d write v(p)=⨁jvj(pj)∈G(L)v(p)=\bigoplus_j v^j(p_j) \in G(L). If we are given u1,…,un,v1,…,vm∈Edu_1,\ldots,u_n,v_1,\ldots,v_m \in E^d such that the (left) QQ-spans U=⟨u1,…,un⟩U=\langle u_1,\ldots,u_n\rangle and V=⟨v1,…,vm⟩V=\langle v_1,\ldots,v_m\rangle have dimension nn and mm and U+VU+V has dimension n+m−kn+m-k, then U∩VU\cap V is spanned by some $w_1,\ldots,w_k \in E^d

and the points qi=wi(p)q_i = w_i(p) are generic and independent with

cl⁡(u1(p),…,un(p))∩cl⁡(v1(p),…,vm(p))=cl⁡(q1,…,qk).\operatorname{cl}(u_1(p),\ldots,u_n(p)) \cap\operatorname{cl}(v_1(p),\ldots,v_m(p)) = \operatorname{cl}(q_1,\ldots,q_k).

Proof. We get the wiw_i through the dimension formula for left vector spaces over Q\mathbb{Q}. As matroids over Q\mathbb{Q} are equivalent to algebraic matroids over GG the statement follows. □

We state some properties of algebraic groups which we will need later.

Lemma 2.3. Let GG be a connected one-dimensional algebraic group over KK.

(a) The action of any e∈E∖{0}e \in E \setminus\{0\} preserves the closure of a point p∈G(L)p \in G(L). We have

cl⁡(p)=cl⁡(e(p)).\operatorname{cl}(p) = \operatorname{cl}(e(p)).

(b) Let HH be a commutative connected algebraic group. For subvarieties V,CV,C we write V⊕C:={v⊕c∣v∈V,c∈C}V \oplus C := \{v \oplus c \mid v \in V,c \in C\}. Let V,CV,C be closed and irreducible and W=V⊕CW = V \oplus C. If dim⁡W=dim⁡V\dim W = \dim V, then C⊖C⊆Stab⁡(W)C \ominus C \subseteq\operatorname{Stab}(W) and W=V⊕CW = V \oplus C.

(c) If GG is either an elliptic curve or GmG_m, then every connected closed subgroup of Gan×GmmG_a^n \times G_m^m is a product U×VU \times V where UU is a subgroup of GanG_a^n and VV is a subgroup of GmmG_m^m.

Proof. (a) Consider the composition of morphisms Spec⁡L→pG→eG\operatorname{Spec} L \xrightarrow{p} G \xrightarrow{e} G. We get L↩K(p)↩K(e(p))L \hookleftarrow K(p) \hookleftarrow K(e(p)) and as ee defines an isogeny the extension K(p)/K(e(p))K(p)/K(e(p)) is finite. Hence cl⁡(p)=cl⁡(e(p))\operatorname{cl}(p) = \operatorname{cl}(e(p)).

(b) V×CV \times C is irreducible and V⊕CV \oplus C is the image under the morphism ⊕:H×H→H\oplus: H \times H \to H. Hence WW is also irreducible. For any c∈C(K)c \in C(K) we have V⊕c⊆V⊕C⊆WV \oplus c \subseteq V \oplus C \subseteq W and dim⁡(V⊕c)=dim⁡V=dim⁡W\dim(V \oplus c) = \dim V = \dim W. With both varieties being closed and irreducible it follows V⊕c=WV \oplus c = W for every c∈C(K)c \in C(K). For c,c′∈C(K)c,c' \in C(K) we now get

(c⊖c′)⊕W=(c⊖c′)⊕(c′⊕V)=c⊕V=W(c \ominus c') \oplus W = (c \ominus c') \oplus(c' \oplus V) = c \oplus V = W

and with that c⊖c′∈Stab⁡(W)c \ominus c' \in\operatorname{Stab}(W). It follows C⊖C⊆Stab⁡(W)C \ominus C \subseteq\operatorname{Stab}(W) and W=V⊕CW = V \oplus C.

(c) The statement is standard. Let WW be a closed connected subgroup of Gan×GmmG_a^n \times G_m^m. For G=GmG = G_m it follows from the unique decomposition of WW into a product of a unipotent subgroup and a subgroup of multiplicative type, see [17 Theorem 16.13]. For G=EG = E an elliptic curve it follows from the Rosenlicht decomposition, see [16 Theorem 6.2] and [21], that W=(Waff×Want)/(Waff∩Want)W = (W_{\mathrm{aff}} \times W_{\mathrm{ant}})/(W_{\mathrm{aff}} \cap W_{\mathrm{ant}}) in an affine and anti-affine group. Here Waff⊆Gan×0W_{\mathrm{aff}} \subseteq G_a^n \times0 and Want⊆0×EmW_{\mathrm{ant}} \subseteq0 \times E^m. As these intersect trivially, we get the wanted result with U=WaffU = W_{\mathrm{aff}} and V=WantV = W_{\mathrm{ant}}. □

To study the algebraic independence of elements in a field extension geometrically, we define the following:

Definition 2.4. Let VV be a variety over KK and p:Spec⁡L→Vp : \operatorname{Spec} L \to V be an LL-point. We call the irreducible closed subvariety

Z(p):={p}‾⊆VZ(p) := \overline{\{p\}} \subseteq V

the locus of pp.

Lemma 2.5. Let V,WV,W be varieties over KK and p∈V(L)p \in V(L) an LL-point.

(a) dim⁡Z(p)=trdeg⁡KK(p)\dim Z(p) = \operatorname{trdeg}_{K} K(p).

(b) If φ:V→W\varphi: V \to W is a morphism over KK, then Z(φ(p))=φ(Z(p))‾Z(\varphi(p)) = \overline{\varphi(Z(p))}.

(c) If we are given several points pi∈Vi(L)p_i \in V_i(L) in varieties V1,…,VnV_1,\ldots,V_n over KK and trdeg⁡KK(p1,…,pn)=∑itrdeg⁡KK(pi)\operatorname{trdeg}_{K} K(p_1,\ldots,p_n) = \sum_i \operatorname{trdeg}_{K} K(p_i), then the locus splits as a product of loci, i.e. Z(p1,…,pn)=Z(p1)×⋯×Z(pn)Z(p_1,\ldots,p_n) = Z(p_1) \times\cdots\times Z(p_n).

Proof. (a) The dimension of Z(p)Z(p) is just the transcendence degree of the residue field of its generic point. This is trdeg⁡KK(p)\operatorname{trdeg}_{K} K(p). (b) φ(Z(p))‾\overline{\varphi(Z(p))} is closed and contains φ(p)\varphi(p). Hence Z(φ(p))⊆φ(Z(p))‾Z(\varphi(p)) \subseteq\overline{\varphi(Z(p))}. The other inclusion follows by continuity. (c) We set p=(p1,…,pn)p = (p_1,\ldots,p_n). With (b) we get Z(p)⊆∏iZ(pi)Z(p) \subseteq\prod_i Z(p_i). Both sides are now irreducible and of the same dimension, hence we get an equality.

In Section 5.3 we glue together different realizations of the matroid M(K4)M(K_4) of the complete graph on four vertices. For this we will need the following proposition which is a variant of [7], Theorem 2.2.2.

Proposition 2.6. Let G1,G2G_1, G_2 be one-dimensional algebraic groups over KK. If for a1,b1∈G1(L)a_1, b_1 \in G_1(L) and a2,b2∈G2(L)a_2, b_2 \in G_2(L) we have

  1. a1a_1 and b1b_1 are algebraically independent,

  2. cl⁡(a1)=cl⁡(a2)\operatorname{cl}(a_1) = \operatorname{cl}(a_2) and cl⁡(b1)=cl⁡(b2)\operatorname{cl}(b_1) = \operatorname{cl}(b_2),

  3. cl⁡(a1⊕b1)=cl⁡(a2⊕b2)\operatorname{cl}(a_1 \oplus b_1) = \operatorname{cl}(a_2 \oplus b_2).

Then there is a closed connected one-dimensional subgroup H0⊂G1×G2H_0 \subset G_1 \times G_2 and points pa,pb∈H0(L)p_a, p_b \in H_0(L) such that there are constants xa,xb∈(G1×G2)(K)x_a, x_b \in(G_1 \times G_2)(K) with

a1=π1(pa⊕xa),a2=π2(pa⊕xa),b1=π1(pb⊕xb),b2=π2(pb⊕xb).a_1 = \pi_1(p_a \oplus x_a),\quad a_2 = \pi_2(p_a \oplus x_a),\quad b_1 = \pi_1(p_b \oplus x_b),\quad b_2 = \pi_2(p_b \oplus x_b).

Here π1,π2\pi_1, \pi_2 are just the coordinate projections of G1×G2G_1 \times G_2. In addition G1G_1 and G2G_2 are isogenous.

Proof. Set H=G1×G2H = G_1 \times G_2. Set a=(a1,a2)a = (a_1,a_2), b=(b1,b2)b = (b_1,b_2), Ca=Z(a)C_a = Z(a) and Cb=Z(b)C_b = Z(b), then CaC_a and CbC_b are irreducible curves in HH which dominate both factors with Lemma 2.5b. Using Lemma 2.5c follows Z((a,b))=Ca×CbZ((a,b)) = C_a \times C_b in H×HH \times H; hence Z(a⊕b)=(Ca⊕Cb)‾Z(a \oplus b) = \overline{(C_a \oplus C_b)} with Lemma 2.5b. Furthermore (1) and (3) give us trdeg⁡K(a⊕b)=1\operatorname{trdeg} K(a \oplus b) = 1. By Lemma 2.3b follows Cb⊖Cb⊆Stab⁡(Z(a⊕b))C_b \ominus C_b \subseteq\operatorname{Stab}(Z(a \oplus b)) and Ca⊖Ca⊆Stab⁡(Z(a⊕b))C_a \ominus C_a \subseteq\operatorname{Stab}(Z(a \oplus b)). Set H0=Stab⁡(Z(a⊕b))∘H_0 = \operatorname{Stab}(Z(a \oplus b))^\circ. By translating with some points xa∈Ca(K)x_a \in C_a(K), xb∈Cb(K)x_b \in C_b(K) we get by using that CaC_a and CbC_b are irreducible

Ca⊆xa⊕H0,Cb⊆xb⊕H0.C_a \subseteq x_a \oplus H_0,\quad C_b \subseteq x_b \oplus H_0.

With xa⊕xb⊕H0⊆Z(a⊕b)x_a \oplus x_b \oplus H_0 \subseteq Z(a \oplus b) follows xa⊕xb⊕H0=Z(a⊕b)x_a \oplus x_b \oplus H_0 = Z(a \oplus b) and dim⁡H0=dim⁡Z(a⊕b)=1\dim H_0 = \dim Z(a \oplus b) = 1. The image of an algebraic group under a homomorphism is closed and CaC_a dominates both factors; hence π1(H0)=G1\pi_1(H_0) = G_1 and π2(H0)=G2\pi_2(H_0) = G_2. The kernels are zero-dimensional and hence finite, therefore G1G_1 and G2G_2 are isogenous through H0H_0. The points are now pa=a⊖xap_a = a \ominus x_a and pb=b⊖xbp_b = b \ominus x_b.

Intersecting matroids over division rings

An important operation on matroids in this article will be the intersection of matroids. The operation is defined in a purely combinatorial way. For two linear matroids over a common field the intersection is still linear over a sufficiently large extension of that field and for algebraic matroids the intersection is algebraic over the same field. Two matroids algebraic over the same field KK can however be linear over fields or division rings of different characteristic. This is the situation we need to understand. The dual notion of intersecting matroids is the union of matroids. The operations are defined in the following way:

Definition 3.1. Let N1N_1 and N2N_2 be matroids on a common ground set EE.

  1. The union N1∨N2N_1 \vee N_2 is the matroid on EE whose independent sets are the subsets of EE that can be expressed as a union I1∪I2I_1 \cup I_2, where I1I_1 is an independent set of N1N_1 and I2I_2 is an independent set of N2N_2.

  1. The intersection N1∧N2N_1 \wedge N_2 is the matroid on EE whose spanning sets are the subsets of EE that can be expressed as an intersection S1∩S2S_1 \cap S_2, where S1S_1 is a spanning set of N1N_1 and S2S_2 is a spanning set of $N_2.

The two constructions are related through matroid duality by N1∧N2=(N1∗∨N2∗)∗N_1 \wedge N_2 = (N_1^* \vee N_2^*)^*. We say that N1∨N2N_1 \vee N_2 has the expected rank if the set of bases of N1∨N2N_1 \vee N_2 is

{B1∪B2∣B1∈J1,B2∈J2, and B1∩B2=∅}.\{B_1 \cup B_2 \mid B_1 \in J_1, B_2 \in J_2, \text{ and } B_1 \cap B_2 = \varnothing\}.

where J1J_1 is the set of bases of N1N_1 and J2J_2 is the set of bases of N2N_2. Dually, we say that N1∧N2N_1 \wedge N_2 has the expected rank if the set of bases of N1∧N2N_1 \wedge N_2 is

{B1∩B2∣B1∈J1,B2∈J2, and B1∪B2=E}.\{B_1 \cap B_2 \mid B_1 \in J_1, B_2 \in J_2, \text{ and } B_1 \cup B_2 = E\}.

Now let M1M_1 be a matroid of rank dd with ground set E1E_1 over a division ring Q1Q_1 and M2M_2 a matroid of the same rank dd with ground set E2E_2 over a division ring Q2Q_2. We assume that the ground sets intersect in the standard unit vectors E1∩E2=T={e1,…,ed}E_1 \cap E_2 = T = \{e_1, \ldots, e_d\}. Let M1′M'_1 be M1M_1 with the elements of E2∖TE_2 \setminus T added as coloops, and define M2′M'_2 in the same way for M2M_2. We set N=M1′∧M2′N = M'_1 \wedge M'_2 with ground set E=E1∪E2E = E_1 \cup E_2. Set A=E1∖TA = E_1 \setminus T and B=E2∖TB = E_2 \setminus T. The dual of NN has rank ∣A∣+∣B∣|A| + |B|. It follows that NN has rank dd and therefore NN has expected rank. This means that the set of bases of NN is

{B1∩B2∣B1 is basis of M1′,B2 is basis of M2′ and B1∪B2=E}.\{B_1 \cap B_2 \mid B_1 \text{ is basis of } M'_1, B_2 \text{ is basis of } M'_2 \text{ and } B_1 \cup B_2 = E\}.

To characterize the bases of the intersection of two matroids over possibly different division rings we will use Rado’s theorem, see [19 Theorem 11.2.2] and [20].

Theorem 3.2 (Rado’s theorem). Given a matroid MM with ground set EE and rank function rr. Given further a index set II and a family A=(Ai)i∈I\mathcal{A} = (A_i)_{i \in I} of subsets Ai⊆EA_i \subseteq E. Then the following are equivalent:

  1. The family A\mathcal{A} has a transversal which is an independent set of MM.

(2) For every subset J⊆IJ \subseteq I we have r(A(J))≥∣J∣r(A(J)) \ge|J|.

Here we set A(J)=⋃j∈JAjA(J) = \bigcup_{j\in J} A_j.

Over a division ring QQ, the left row rank of a matrix equals its right column rank; we denote it by rk⁡Q\operatorname{rk}_{Q}.

In the setting above, let M1M_1 be the row matroid of a matrix α\alpha over Q1Q_1 with rows indexed by E1E_1 and columns indexed by TT, and let M2M_2 be the row matroid of a matrix β\beta over Q2Q_2 with rows indexed by E2E_2 and columns indexed by TT, such that the rows indexed by TT are the standard unit vectors in both matrices. Let S⊆ES \subseteq E with ∣S∣=d|S| = d, and for I⊆TI \subseteq T let αS,I\alpha_{S,I} be the submatrix of α\alpha with rows S∩E1S \cap E_1 and columns II, and βS,I\beta_{S,I} the submatrix of β\beta with rows S∩BS \cap B and columns II.

Theorem 3.3. SS is a basis of NN if and only if

rk⁡Q1αS,I+rk⁡Q2βS,I≥∣I∣for all I⊆T.\operatorname{rk}_{Q_1}\alpha_{S,I}+\operatorname{rk}_{Q_2}\beta_{S,I}\ge|I| \quad\text{for all } I\subseteq T.

Proof. Write ST=S∩TS_T = S \cap T, SA=S∩AS_A = S \cap A, SB=S∩BS_B = S \cap B, a=∣S∩E1∣=∣ST∣+∣SA∣a = |S \cap E_1| = |S_T| + |S_A| and b=∣SB∣b = |S_B|, so that a+b=da + b = d. We show that both conditions are equivalent to

(∗)(*) there is a partition T=I1⊔I2T = I_1 \sqcup I_2 such that αS,I1\alpha_{S,I_1} and βS,I2\beta_{S,I_2} are square and invertible.

The inequality is equivalent to (∗)(*). Let C1C_1 be the column matroid of αS,T\alpha_{S,T} on T×{1}T \times\{1\} and C2C_2 the column matroid of βS,T\beta_{S,T} on T×{2}T \times\{2\}, and let C=C1⊕C2C = C_1 \oplus C_2. For t∈Tt \in T set At={(t,1),(t,2)}A_t = \{(t,1),(t,2)\}. For I⊆TI \subseteq T, the rank of ⋃t∈IAt\bigcup_{t\in I} A_t in CC is rk⁡αS,I+rk⁡βS,I\operatorname{rk}\alpha_{S,I} + \operatorname{rk}\beta_{S,I}. By Rado’s theorem, the inequality holds if and only if there is a partition T=I1⊔I2T = I_1 \sqcup I_2 such that the columns I1I_1 of αS,T\alpha_{S,T} and the columns I2I_2 of βS,T\beta_{S,T} are independent. Then ∣I1∣≤a|I_1| \le a and ∣I2∣≤b|I_2| \le b, and ∣I1∣+∣I2∣=d=a+b|I_1| + |I_2| = d = a + b forces equality, so αS,I1\alpha_{S,I_1} and βS,I2\beta_{S,I_2} are square and invertible. The converse is clear.

(∗)(*) implies that SS is a basis of NN. If t∈ST∖I1t \in S_T \setminus I_1, then the row ete_t of αS,I1\alpha_{S,I_1} vanishes, so (∗)(*) gives ST⊆I1S_T \subseteq I_1. Set B1′=ST∪SA∪I2B'_1 = S_T \cup S_A \cup I_2, a set of size a+b=da + b = d. The rows of α\alpha indexed by I2I_2 are unit vectors, so eliminating them together with the columns I2I_2 shows that the rows B1′B'_1 of α\alpha are independent if and only if αS,I1\alpha_{S,I_1} is invertible. Hence B1′B'_1 is a basis of M1M_1. Similarly, B2′=SB∪I1B'_2 = S_B \cup I_1 is a basis of M2M_2. Then B1=B1′∪BB_1 = B'_1 \cup B is a basis of M1′M'_1 and B2=B2′∪AB_2 = B'_2 \cup A is a basis of M2′M'_2, with B1∩B2=SB_1 \cap B_2 = S and B1∪B2=EB_1 \cup B_2 = E. Hence SS is a basis of NN.

If SS is a basis of NN, then (∗)(*) holds. Let B1B_1 and B2B_2 be bases of M1′M'_1 and M2′M'_2 with B1∩B2=SB_1 \cap B_2 = S and B1∪B2=EB_1 \cup B_2 = E. Since BB consists of coloops of M1′M'_1 and AA of coloops of M2′M'_2, we have B1=B1′⊔BB_1 = B'_1 \sqcup B and B2=B2′⊔AB_2 = B'_2 \sqcup A for bases B1′B'_1 of M1M_1 and B2′B'_2 of M2M_2. Set I1=T∩B2′I_1 = T \cap B'_2 and I2=T∖B2′I_2 = T \setminus B'_2. Since B⊆B1B \subseteq B_1, we get B2′∩B=SBB'_2 \cap B = S_B, so B2′=SB∪I1B'_2 = S_B \cup I_1. Since A⊆B2A \subseteq B_2 and B1∪B2=EB_1 \cup B_2 = E, we get B1′∩A=SAB'_1 \cap A = S_A and B1′∩T=ST∪I2B'_1 \cap T = S_T \cup I_2, where ST⊆I1S_T \subseteq I_1. Hence B1′=ST∪SA∪I2B'_1 = S_T \cup S_A \cup I_2, and d=∣B1′∣=a+∣I2∣d = |B'_1| = a + |I_2| gives ∣I2∣=b|I_2| = b and ∣I1∣=a|I_1| = a. The same block elimination as above shows that αS,I1\alpha_{S,I_1} and βS,I2\beta_{S,I_2} are invertible. □\square

Remark 3.4. We note that Theorem 3.3 also follows from the formula for the rank function of a union of two matroids, see [Ox11], Theorem 11.3.1. The proof of the formula also uses Rado’s theorem.

Amalgams of algebraic matroids over one-dimensional groups

In [2] and [7] the authors use the notion of algebraically equivalent realizations of an algebraic matroid. Given some algebraic matroid M(f1,…,fn)M(f_1,\ldots,f_n) represented by f1,…,fn∈Lf_1,\ldots,f_n\in L. If there are f1′,…,fn′∈Lf'_1,\ldots,f'_n\in L such that cl⁡(fi)=cl⁡(fi′)\operatorname{cl}(f_i)=\operatorname{cl}(f'_i) for all i∈[n]i\in[n] then the matroid M(f1′,…,fn′)M(f'_1,\ldots,f'_n) equals M(f1,…,fn)M(f_1,\ldots,f_n) and we call the representations algebraically equivalent. We write f∼f′f\sim f' for f,f′∈Lf,f'\in L if cl⁡(f)=cl⁡(f′)\operatorname{cl}(f)=\operatorname{cl}(f'). The algebraic matroid represented by f1,…,fn∈Lf_1,\ldots,f_n\in L is algebraically equivalent to a representation over a one-dimensional algebraic group GG if there are points q1,…,qn∈G(L)q_1,\ldots,q_n\in G(L) that are the image of a generic point p∈G(L)dp\in G(L)^d under a homomorphism of algebraic groups ψ:Gd→Gn\psi:G^d\to G^n such that qi∼fiq_i\sim f_i for all i∈[n]i\in[n].

Given a matroid M1M_1 realized by (x,y)=(x1,…,xd,y1,…,yn)(x,y)=(x_1,\ldots,x_d,y_1,\ldots,y_n) that is algebraically equivalent to a realization over Ga\mathbb{G}_a and a matroid M2M_2 realized by (x,z)=(x1,…,xd,z1,…,zm)(x,z)=(x_1,\ldots,x_d,z_1,\ldots,z_m) that is algebraically equivalent to a realization over a group GG which is either Gm\mathbb{G}_m or an elliptic curve. Assume that both matroids have rank dd and the basis x1,…,xdx_1,\ldots,x_d. Let M=M(x,y,z)M=M(x,y,z) be the algebraic matroid realized by (x,y,z)(x,y,z) with the ground set EE. Let E1E_1 and E2E_2 be the endomorphism rings of Ga\mathbb{G}_a and GG, with division rings Q1Q_1 and Q2Q_2. Then there are generic points u∈Ga(L)du\in\mathbb{G}_a(L)^d and p∈G(L)dp\in G(L)^d and matrices α∈E1n×d\alpha\in E_1^{n\times d}, β∈E2m×d\beta\in E_2^{m\times d} such that

ui∼xi∼pi,yj∼αj(u),zk∼βk(p).u_i\sim x_i\sim p_i,\qquad y_j\sim\alpha_j(u),\qquad z_k\sim\beta_k(p).

Indeed, if (x,y)∼ψ(g)(x,y)\sim\psi(g), set ui=ψxi(g)u_i=\psi_{x_i}(g) and ψx\psi_x for the matrix with rows ψx1,…,ψxd\psi_{x_1},\ldots,\psi_{x_d}; each row ψyj\psi_{y_j} lies in the Q1Q_1-span of the rows ψxi\psi_{x_i}, so by the Ore condition ejψyj=αjψxe_j\psi_{y_j}=\alpha_j\psi_x for some 0≠ej∈E10\ne e_j\in E_1 and αj∈E1d\alpha_j\in E_1^d, and Lemma 2.3a gives yj∼αj(u)y_j\sim\alpha_j(u). The same argument applies to (x,z)(x,z). Set qi=(ui,pi)∈(Ga×G)(L)q_i=(u_i,p_i)\in(\mathbb{G}_a\times G)(L) and denote with Ci=Z(qi)⊆Ga×GC_i=Z(q_i)\subseteq\mathbb{G}_a\times G its locus. With trdeg⁡K(q1,…,qd)=d=∑itrdeg⁡K(qi)\operatorname{trdeg}K(q_1,\ldots,q_d)=d=\sum_i\operatorname{trdeg}K(q_i) it follows

W:=Z((q1,…,qd))=C1×⋯×Cd⊆(Ga×G)d.W:=Z((q_1,\ldots,q_d))=C_1\times\cdots\times C_d\subseteq(\mathbb{G}_a\times G)^d.

Fix a subset S⊂ES\subset E with ∣S∣=d|S|=d and collect the uiu_i and αi\alpha_i appearing in SS into an a×da\times d matrix αS\alpha_S over E1E_1. Collect all the βi\beta_i appearing in SS into a b×db\times d matrix βS\beta_S over E2E_2.

We define the following homomorphism of algebraic groups

ΦS:(Ga×G)d→Gaa×Gb,((vi,wi))i↦(αSv,βSw).\Phi_S:(\mathbb{G}_a\times G)^d\to\mathbb{G}_a^a\times G^b,\qquad((v_i,w_i))_i\mapsto(\alpha_Sv,\beta_Sw).

If we evaluate this morphism at the point q=(q1,…,qd)q=(q_1,\ldots,q_d), we get with Lemma 2.5

rM(S)=trdeg⁡K(ΦS(q))=dim⁡Z(ΦS(q))=dim⁡ΦS(W)‾.r_M(S)=\operatorname{trdeg}K(\Phi_S(q))=\dim Z(\Phi_S(q))=\dim\overline{\Phi_S(W)}.

We set Xj=ΦS(0×⋯×Cj×⋯×0)‾X_j=\overline{\Phi_S(0\times\cdots\times C_j\times\cdots\times0)} and since ΦS\Phi_S is a morphism of algebraic groups we have

X1‾⊕⋯⊕Xd‾=ΦS(W)‾.\overline{X_1}\oplus\cdots\oplus\overline{X_d}=\overline{\Phi_S(W)}.

For I⊆[d]I\subseteq[d] we set

XI=⨁i∈IXi‾.X_I=\bigoplus_{i\in I}\overline{X_i}.

The following lemma follows directly from the theory of matroids over one-dimensional algebraic groups.

Lemma 4.1. Let GG be an elliptic curve or Gm\mathbb{G}_{m}. For α∈E1a×d\alpha\in\mathbb{E}_{1}^{a \times d} we have dim⁡α(Gad)=rk⁡Q1α\dim\alpha(\mathbb{G}_{a}^{d}) = \operatorname{rk}_{\mathbb{Q}_{1}}\alpha and for β∈E2b×d\beta\in\mathbb{E}_{2}^{b \times d} we have dim⁡β(Gbd)=rk⁡Q2β\dim\beta(\mathbb{G}_{b}^{d}) = \operatorname{rk}_{\mathbb{Q}_{2}}\beta.

Proposition 4.2. We have that dim⁡Φ‾S(W)=d\dim\overline{\Phi}_{S}(W) = d if and only if for every I⊆[d]I \subseteq[d]

rk⁡Q1αS,I+rk⁡Q2βS,I≥∣I∣.\operatorname{rk}_{\mathbb{Q}_{1}}\alpha_{S,I}+\operatorname{rk}_{\mathbb{Q}_{2}}\beta_{S,I}\geq|I|.

Proof. Suppose the inequality does fail for some II. XIX_{I} lies in the subgroup αS,I(GaI)×βS,I(GbI)\alpha_{S,I}(\mathbb{G}_{a}^{I}) \times\beta_{S,I}(\mathbb{G}_{b}^{I}). And with Lemma 2.3b this subgroup has dimension rk⁡Q1αS,I+rk⁡Q2βS,I<∣I∣\operatorname{rk}_{\mathbb{Q}_{1}}\alpha_{S,I}+\operatorname{rk}_{\mathbb{Q}_{2}}\beta_{S,I}<|I|. It follows

dim⁡Φ‾S(W)≤dim⁡XI+dim⁡XIc<∣I∣+∣Ic∣=d.\dim\overline{\Phi}_{S}(W)\leq\dim X_{I}+\dim X_{I^{c}}<|I|+|I^{c}|=d.

Suppose dim⁡Φ‾S(W)<d\dim\overline{\Phi}_{S}(W)<d. We choose a minimal II with dim⁡XI<∣I∣\dim X_{I}<|I|. Hence we have dim⁡XI∖j=∣I∣−1\dim X_{I\setminus j}=|I|-1 for every j∈Ij\in I and dim⁡XI=∣I∣−1\dim X_{I}=|I|-1. Fix some j∈Ij\in I. We want to show that XjX_{j} lies in a coset of a connected subgroup. Set A:=XI∖jA:=X_{I\setminus j} and Y:=XjY:=X_{j}. AA is closed and irreducible as the image of an irreducible variety. We can assume that YY is an irreducible curve, if YY is only a point there is nothing to show. Now Lemma 2.3b gives us Xj⊖Xj⊂Stab⁡(XI)X_{j}\ominus X_{j}\subset\operatorname{Stab}(X_{I}). We fix some xj∈Xj(K)x_{j}\in X_{j}(K). With Xj⊖xjX_{j}\ominus x_{j} containing 00 and being irreducible we have 0∈Xj⊖xj⊆Stab⁡(XI)∘0\in X_{j}\ominus x_{j}\subseteq\operatorname{Stab}(X_{I})^{\circ} which denotes the identity component of the stabilizer. Hence Xj⊂xj⊕Stab⁡(XI)∘X_{j}\subset x_{j}\oplus\operatorname{Stab}(X_{I})^{\circ}. We can do this now for every j∈Ij\in I and get XI⊆c⊕Stab⁡(XI)∘X_{I}\subseteq c\oplus\operatorname{Stab}(X_{I})^{\circ} for c=⨁j∈Ixjc=\bigoplus_{j\in I}x_{j}, since Stab⁡(XI)∘\operatorname{Stab}(X_{I})^{\circ} is closed. But we also have that for any c′∈XI(K)c'\in X_{I}(K) that c′⊕Stab⁡(XI)∘⊆XIc'\oplus\operatorname{Stab}(X_{I})^{\circ}\subseteq X_{I}. Thus XI=c⊕Stab⁡(XI)∘X_{I}=c\oplus\operatorname{Stab}(X_{I})^{\circ}. It follows dim⁡Stab⁡(XI)∘=∣I∣−1\dim\operatorname{Stab}(X_{I})^{\circ}=|I|-1.

With Lemma 2.3c follows Stab⁡(XI)∘=U×V\operatorname{Stab}(X_{I})^{\circ}=U\times V for closed subgroups U⊆GaaU\subseteq\mathbb{G}_{a}^{a} and V⊆GbbV\subseteq\mathbb{G}_{b}^{b}. We consider the projections π1:Gaa×Gbb→Gaa\pi_{1}:\mathbb{G}_{a}^{a}\times\mathbb{G}_{b}^{b}\to\mathbb{G}_{a}^{a} and π2:Gaa×Gbb→Gbb\pi_{2}:\mathbb{G}_{a}^{a}\times\mathbb{G}_{b}^{b}\to\mathbb{G}_{b}^{b}. Set

AI:=αS,I(GaI)⊆Gaa,BI:=βS,I(GbI)⊆Gbb.A_{I}:=\alpha_{S,I}(\mathbb{G}_{a}^{I})\subseteq\mathbb{G}_{a}^{a},\quad B_{I}:=\beta_{S,I}(\mathbb{G}_{b}^{I})\subseteq\mathbb{G}_{b}^{b}.

We have

π1(XI)‾=αS,I(∏i∈Iπ1(Ci)),\overline{\pi_{1}(X_{I})}=\alpha_{S,I}\left(\prod_{i\in I}\pi_{1}(C_{i})\right),

and as each CiC_{i} maps dominantly into Ga\mathbb{G}_{a} we have that π1(XI)\pi_{1}(X_{I}) is dense in AIA_{I}. We have that XIX_{I} is a coset of Stab⁡(XI)∘\operatorname{Stab}(X_{I})^{\circ}, hence π1(XI)=π1(c)+U\pi_{1}(X_{I})=\pi_{1}(c)+U, which is already closed. Now

π1(c)+U=π1(XI)=AI.\pi_{1}(c)+U=\pi_{1}(X_{I})=A_{I}.

As both AIA_{I} and UU are closed connected subgroups we have U=AIU=A_{I}. The same argument gives us V=BIV=B_{I} and it follows

∣I∣−1=dim⁡Stab⁡(XI)∘=dim⁡AI+dim⁡BI=rk⁡Q1αS,I+rk⁡Q2βS,I|I|-1=\dim\operatorname{Stab}(X_{I})^{\circ}=\dim A_{I}+\dim B_{I}=\operatorname{rk}_{\mathbb{Q}_{1}}\alpha_{S,I}+\operatorname{rk}_{\mathbb{Q}_{2}}\beta_{S,I}

and the inequality fails for II. □

Corollary 4.3. If we use the notation of § 3, then it follows with Theorem 3.3 that M=M1′∧M2′M=M'_{1}\wedge M'_{2}.

Algebraic equivalence of realizations

We now want to show that the algebraic representations of certain matroids are all algebraically equivalent to a representation over a one-dimensional algebraic group GG. This is already known for the Fano and non-Fano matroids and also for every projective geometry in L/KL/K, see [7 Theorem 3.3.1] and [2 Example 6.2]. We will show the same for analogs of the dual Fano and non-Fano matroids over fields of characteristic pp. For this the following lemmas will be helpful.

Intersecting transcendental field extensions

The statements are well known, but we provide short proofs.

Lemma 5.1. Let A⊆FA \subseteq F be fields with AA algebraically closed and tt transcendental over FF. Then A(t)A(t) is algebraically closed in F(t)F(t).

Proof. We have that A(t)A(t) and FF are free over AA. With [12 VIII. Theorem 4.3] and the fact that F/AF/A is regular the statement follows. □\square

For subfields AA, B⊆LB \subseteq L let ABAB be the composite of AA and BB in LL, that is the smallest subfield of LL which contains both AA and BB.

Lemma 5.2. Let AA, B⊆LB \subseteq L be subfields. If t∈Lt \in L is transcendental over ABAB then

A(t)∩B(t)=(A∩B)(t).A(t) \cap B(t) = (A \cap B)(t).

Proof. We only prove the inclusion A(t)∩B(t)⊆(A∩B)(t)A(t) \cap B(t) \subseteq(A \cap B)(t). The other inclusion is trivial. Set F=ABF = AB. Every f∈F(t)f \in F(t) can be written uniquely as f=pqf = \frac{p}{q} with p,q∈F[t]p,q \in F[t], qq monic, and gcd⁡(p,q)=1\gcd(p,q) = 1 in F[t]F[t]. Now let f∈A(t)f \in A(t). Write f=pqf = \frac{p}{q} with p,q∈A[t]p,q \in A[t], qq monic, and gcd⁡(p,q)=1\gcd(p,q) = 1 in A[t]A[t]. Then gcd⁡(p,q)=1\gcd(p,q) = 1 in F[t]F[t] as well, and pq\frac{p}{q} is the reduced form of ff over FF. Hence the coefficients of the reduced form of ff lie in AA. By the same argument, if f∈B(t)f \in B(t), they lie in BB. Uniqueness of the reduced form means that for f∈A(t)∩B(t)f \in A(t) \cap B(t) these are the same coefficients. So they lie in A∩BA \cap B, and f∈(A∩B)(t)f \in(A \cap B)(t). □\square

Lemma 5.3. Let AA, B⊆LB \subseteq L be algebraically closed subfields containing KK. If F=cl⁡(AB)F = \operatorname{cl}(AB) and t∈Lt \in L is transcendental over FF, then cl⁡(A(t))∩cl⁡(B(t))=cl⁡((A∩B)(t))\operatorname{cl}(A(t)) \cap\operatorname{cl}(B(t)) = \operatorname{cl}((A \cap B)(t)).

Proof. Choose some z∈cl⁡(A(t))∩cl⁡(B(t))z \in\operatorname{cl}(A(t)) \cap\operatorname{cl}(B(t)) and let ff be its minimal polynomial over F(t)F(t) and gg its minimal polynomial over A(t)A(t). It follows that ff divides gg as polynomials over F(t)F(t). Therefore all the coefficients of ff lie in cl⁡(A(t))∩F(t)\operatorname{cl}(A(t)) \cap F(t). With Lemma 5.1 we have cl⁡(A(t))∩F(t)=A(t)\operatorname{cl}(A(t)) \cap F(t) = A(t) and it follows that all the coefficients of ff lie already in A(t)A(t). Similarly, we can get that f∈B(t)[X]f \in B(t)[X]. With Lemma 5.2 we have A(t)∩B(t)=(A∩B)(t)A(t) \cap B(t) = (A \cap B)(t) and it follows that zz is algebraic over (A∩B)(t)(A \cap B)(t). Hence z∈cl⁡((A∩B)(t))z \in\operatorname{cl}((A \cap B)(t)). The other inclusion is trivial. □\square

The matroid M(K4)M(K_4)

Consider the matroid of the complete graph on 4 vertices M(K4)M(K_4). That is a matroid of rank 3 with ground set {a,b,c,d,e,f}\{a,b,c,d,e,f\} and circuits {a,b,c}\{a,b,c\}, {a,d,e}\{a,d,e\}, {b,e,f}\{b,e,f\} and {c,d,f}\{c,d,f\}. The matroid has the partition {a,f}\{a,f\}, {b,d}\{b,d\}, {c,e}\{c,e\} into points not on a common 3-point line. We will call a subset of the partition a matching. With [7 Theorem 2.1.2] follows that every algebraic realization of M(K4)M(K_4) is algebraically equivalent to a realization over a one-dimensional algebraic group GG. Using this one can show:

Lemma 5.4 ([7] Corollary 3.2.1). Suppose M(K4)M(K_4) is realized by a,b,c,d,e,f∈La,b,c,d,e,f \in L as an algebraic matroid over an algebraically closed subfield K0⊆LK_0 \subseteq L. Then the lines cl⁡(a,f)\operatorname{cl}(a,f) and cl⁡(c,e)\operatorname{cl}(c,e) meet in a point over K0K_0. In other words, two matchings of M(K4)M(K_4) always meet in a point over K0K_0.

Lemma 5.5. Let MM be a simple matroid of rank 3 on E={a,b,c,d,e,f}E=\{a,b,c,d,e,f\} where each circuit of M(K4)M(K_4) is a dependent set in MM. Call them l1,l2,l3,l4l_1,l_2,l_3,l_4. Then those are the only dependent triples of MM and M≅M(K4)M \cong M(K_4).

Proof. This is well known, by the enumeration of all matroids of rank 3 on 6 elements, but we state a short proof. We notice that any three of the sets l1,l2,l3,l4l_1,l_2,l_3,l_4 cover EE. Let FF be a rank-2 flat of MM with ∣F∣≥3|F|\geq3. So FF contains two elements of some lil_i. Since MM is simple and lil_i is dependent we have li⊆Fl_i \subseteq F.

Suppose F≠liF \ne l_i, and pick w∈F∖liw \in F \setminus l_i. ww is matched to exactly one element of lil_i. So there are s≠ts \ne t in lil_i such that {w,s}\{w,s\} lies in some different ljl_j and {w,t}\{w,t\} lies in some different lkl_k. As before we have lj,lk⊆Fl_j,l_k \subseteq F. The three sets li,lj,lkl_i,l_j,l_k are distinct, so E⊆FE \subseteq F. This contradicts r(M)=3r(M)=3. Hence F=liF=l_i. □\square

The dual Lazarson matroids are only realizable over algebraic groups

If we extend the field KK by some elements a1,…,ana_1,\ldots,a_n we will always assume that K(a1,…,an)K(a_1,\ldots,a_n) is already algebraically closed from now on in this section.

Given variables x0,…,xp+1x_0,\ldots,x_{p+1}. For i∈[p+1]i \in[p+1] set yi=∑j≠ixjy_i=\sum_{j\ne i}x_j. Let Lp∗,NLp∗L_p^*,\mathrm{NL}_p^* be the matroids realized by {x0,…,xp+1,y1,…,yp+1}\{x_0,\ldots,x_{p+1},y_1,\ldots,y_{p+1}\} over Fp\mathbb{F}_p or Q\mathbb{Q}. The matroid Lp∗L_p^* is the dual of the so-called Lazarson matroid LpL_p and the matroid NLp∗\mathrm{NL}_p^* is the dual of the non-Lazarson matroid NLp\mathrm{NL}_p. The two matroids can be seen as generalizations of the Fano and non-Fano matroids to characteristic pp. LpL_p is only algebraic over fields of characteristic pp and NLp\mathrm{NL}_p is only linear over fields of characteristic not equal to pp.

Theorem 5.6. Let NpN_p be either Lp∗L_p^* or NLp∗\mathrm{NL}_p^* and realized by some arbitrary elements

x0,…,xp+1,y1,…,yp+1∈Lx_0,\ldots,x_{p+1},y_1,\ldots,y_{p+1}\in L

as an algebraic matroid over KK. Then there is an algebraically equivalent realization over some connected one-dimensional algebraic group HH. Furthermore for Np=Lp∗N_p=L_p^* we have H≅GaH\cong\mathbb{G}_a and for Np=NLp∗N_p=\mathrm{NL}_p^* we have H≅GmH\cong\mathbb{G}_m or HH is some elliptic curve.

Proof. We will construct several realizations of M(K4)M(K_4) over KK from NpN_p and glue them together into an algebraically equivalent realization of NpN_p over a one-dimensional algebraic group. Fix i,j∈[p+1]i,j \in[p+1]. Set Sij={x1,…,xp+1}∖{xi,xj}S_{ij}=\{x_1,\ldots,x_{p+1}\}\setminus\{x_i,x_j\}. Then the contraction Np/SijN_p/S_{ij} has p+4p+4 elements and all the yky_k with yk≠yiy_k\ne y_i and yk≠yjy_k\ne y_j are parallel to each other; we represent them by zz. Now consider the simplification {x0,xi,xj,yi,yj,z}\{x_0,x_i,x_j,y_i,y_j,z\}. We have the lines {x0,xj,yi}\{x_0,x_j,y_i\}, {x0,xi,yj}\{x_0,x_i,y_j\}, {xi,yi,z}\{x_i,y_i,z\} and {xj,yj,z}\{x_j,y_j,z\}. Hence this matroid is isomorphic to M(K4)M(K_4) with the matchings {x0,z}\{x_0,z\}, {xi,xj}\{x_i,x_j\} and {yi,yj}\{y_i,y_j\}. Therefore we get the point

K(Sij)(pij)=K(Sij)(xi,xj)∩K(Sij)(yi,yj)K(S_{ij})(p_{ij})=K(S_{ij})(x_i,x_j)\cap K(S_{ij})(y_i,y_j)

over K(Sij)K(S_{ij}). As {xi,xj,yi,yj}\{x_i,x_j,y_i,y_j\} is a flat in NpN_p and SijS_{ij} is independent over K(xi,xj,yi,yj)K(x_i,x_j,y_i,y_j) we get with Lemma 5.3

K(xi,xj)(Sij)∩K(yi,yj)(Sij)=(K(xi,xj)∩K(yi,yj))(Sij)K(x_i,x_j)(S_{ij})\cap K(y_i,y_j)(S_{ij})=(K(x_i,x_j)\cap K(y_i,y_j))(S_{ij})

has transcendence degree pp over KK and we can take pijp_{ij} to be points over KK. As the intersection of the flats spanned by {xi,xj}\{x_i,x_j\} and {yi,yj}\{y_i,y_j\} is empty in NpN_p the elements pijp_{ij} are not parallel to any element of NpN_p. Assume that pijp_{ij} is parallel to pklp_{kl}. It follows pij∈K(xi,xj)∩K(xk,xl)p_{ij}\in K(x_i,x_j)\cap K(x_k,x_l). But this can only happen if {i,j}={k,l}\{i,j\}=\{k,l\}. Hence any two points are independent.

Now set

Πijk={xi,xj,xk,pij,pik,pjk}\Pi_{ijk}=\{x_i,x_j,x_k,p_{ij},p_{ik},p_{jk}\}

and

Πijk′={yi,yj,yk,pij,pik,pjk}.\Pi'_{ijk}=\{y_i,y_j,y_k,p_{ij},p_{ik},p_{jk}\}.

We have that {pij,pik,pjk}⊆K(xi,xj,xk)\{p_{ij},p_{ik},p_{jk}\}\subseteq K(x_i,x_j,x_k) and {pij,pik,pjk}⊆K(yi,yj,yk)\{p_{ij},p_{ik},p_{jk}\}\subseteq K(y_i,y_j,y_k). These two fields are not equal and both have transcendence degree 3. Hence {pij,pik,pjk}\{p_{ij},p_{ik},p_{jk}\} has rank 2 and is a line. In addition we have the lines

{xi,xj,pij},{xi,xk,pik},{xj,xk,pjk}\{x_i,x_j,p_{ij}\},\{x_i,x_k,p_{ik}\},\{x_j,x_k,p_{jk}\}

and

{yi,yj,pij},{yi,yk,pik},{yj,yk,pjk}.\{y_i,y_j,p_{ij}\},\{y_i,y_k,p_{ik}\},\{y_j,y_k,p_{jk}\}.

It follows with Lemma 5.5 that Πijk\Pi_{ijk} and Πijk′\Pi'_{ijk} both realize M(K4)M(K_4) over KK. By [7 Theorem 2.1.2] each of them is algebraically equivalent to a realization over a connected one-dimensional algebraic group, and by projective uniqueness of M(K4)M(K_4) over division rings we can fix the realization: for Πijk\Pi_{ijk} there are a group GG and algebraically independent points si,sj,sk∈G(L)s_i,s_j,s_k\in G(L) with

xl∼sl (l∈{i,j,k}),plm∼sl⊖sm (l,m∈{i,j,k}),x_l\sim s_l\ (l\in\{i,j,k\}),\qquad p_{lm}\sim s_l\mathbin{\ominus}s_m\ (l,m\in\{i,j,k\}),

and similarly for Πijk′\Pi'_{ijk} with yly_l in place of $x_l.

We repeatedly use the following observation. Let π:H′→H\pi:H'\to H be an isogeny, a∈H(L)a\in H(L), c∈H(K)c\in H(K) and a′∈H′(L)a'\in H'(L) with π(a′)=a⊕c\pi(a')=a\oplus c. Then a′∼aa'\sim a, since π\pi is finite and K(a⊕c)=K(a)K(a\oplus c)=K(a). Such lifts a′a' exist because π\pi is surjective and LL is algebraically closed.

Step 1: the elements x1,…,xp+1x_1,\ldots,x_{p+1} and pijp_{ij}. We show by induction on k∈[3,p+1]k\in[3,p+1] that there are a connected one-dimensional algebraic group HkH_k and points p1,…,pk∈Hk(L)p_1,\ldots,p_k\in H_k(L) with

xi∼pi for i∈[k],pij∼pi⊖pj for i∈{1,2}, j∈[k]∖{i}.x_i\sim p_i\text{ for }i\in[k],\qquad p_{ij}\sim p_i\mathbin{\ominus}p_j\text{ for }i\in\{1,2\},\ j\in[k]\setminus\{i\}.

For k=3k=3 this is the realization of Π123\Pi_{123}. For the induction step, take the realization q1,q2,qk+1∈G(L)q_1,q_2,q_{k+1}\in G(L) of Π1,2,k+1\Pi_{1,2,k+1}, with xk+1∼qk+1x_{k+1}\sim q_{k+1}. Apply Proposition 2.6 to a=(p1,q1)a=(p_1,q_1) and b=(⊖p2,⊖q2)b=(\ominus p_2,\ominus q_2): the points p1,p2p_1,p_2 are independent, and p1⊖p2∼p12∼q1⊖q2p_1\ominus p_2\sim p_{12}\sim q_1\ominus q_2. We obtain a group Hk+1⊆Hk×GH_{k+1}\subseteq H_k\times G such that the projections π:Hk+1→Hk\pi:H_{k+1}\to H_k and π′:Hk+1→G\pi':H_{k+1}\to G are isogenies, together with r1,r2∈Hk+1(L)r_1,r_2\in H_{k+1}(L) and constants c1,c2c_1,c_2 with (p1,q1)=r1⊕c1(p_1,q_1)=r_1\oplus c_1 and (p2,q2)=r2⊖c2(p_2,q_2)=r_2\ominus c_2. Choose ri∈Hk+1(L)r_i\in H_{k+1}(L) with π(ri)=pi\pi(r_i)=p_i for 3≤i≤k3\le i\le k, and rk+1r_{k+1} with π′(rk+1)=qk+1\pi'(r_{k+1})=q_{k+1}. Since π(ri)\pi(r_i) and π′(ri)\pi'(r_i) differ from pip_i and qiq_i only by constants, the observation gives ri∼xir_i\sim x_i for all i∈[k+1]i\in[k+1], ri⊖rj∼pi⊖pj∼pijr_i\ominus r_j\sim p_i\ominus p_j\sim p_{ij} for i∈{1,2}i\in\{1,2\}, j≤kj\le k, and ri⊖rk+1∼qi⊖qk+1∼pi,k+1r_i\ominus r_{k+1}\sim q_i\ominus q_{k+1}\sim p_{i,k+1} for i∈{1,2}i\in\{1,2\}. Rename each rir_i to pip_i.

Step 2: the elements y1,y2,y3y_1,y_2,y_3. Let p1,…,pp+1∈Hp+1(L)p_1,\ldots,p_{p+1}\in H_{p+1}(L) be as in Step 1, and let t1,t2,t3∈G′(L)t_1,t_2,t_3\in G'(L) be the realization of Π123′\Pi'_{123}. Apply Proposition 2.6 to a=(p1⊖p2,t1⊖t2)a=(p_1\ominus p_2,t_1\ominus t_2) and b=(p2⊖p3,t2⊖t3)b=(p_2\ominus p_3,t_2\ominus t_3): these realize p12p_{12} and p23p_{23}, which are independent, and a⊕ba\oplus b realizes p13p_{13} in both factors. We obtain H⊆Hp+1×G′H\subseteq H_{p+1}\times G' with isogenies π:H→Hp+1\pi:H\to H_{p+1}, π′:H→G′\pi':H\to G' and points pa,pb∈H(L)p_a,p_b\in H(L) that project to the components of aa and bb up to constants. Choose p2′p'_2 with π(p2′)=p2\pi(p'_2)=p_2, set p1′=pa⊕p2′p'_1=p_a\oplus p'_2 and p3′=p2′⊖pbp'_3=p'_2\ominus p_b, choose pi′p'_i with π(pi′)=pi\pi(p'_i)=p_i for i≥4i\ge4, and choose t1′∈H(L)t'_1\in H(L) with π′(t1′)=t1\pi'(t'_1)=t_1. Then π(pi′)=pi\pi(p'_i)=p_i up to constants for all ii, so the relations of Step 1 still hold for the pi′p'_i. Moreover, up to constants,

π′(t1′⊖p1′⊕p2′)=t1⊖π′(pa)=t2,π′(t1′⊖p1′⊕p3′)=t1⊖π′(pa)⊖π′(pb)=t3,\pi'(t'_1\ominus p'_1\oplus p'_2)=t_1\ominus\pi'(p_a)=t_2,\qquad \pi'(t'_1\ominus p'_1\oplus p'_3)=t_1\ominus\pi'(p_a)\ominus\pi'(p_b)=t_3,

so y1∼t1′y_1\sim t'_1, y2∼t1′⊖p1′⊕p2′y_2\sim t'_1\ominus p'_1\oplus p'_2 and y3∼t1′⊖p1′⊕p3′y_3\sim t'_1\ominus p'_1\oplus p'_3.

Step 3: the remaining elements. Set h=(p1′,…,pp+1′,t1′)∈H(L)p+2h=(p'_1,\ldots,p'_{p+1},t'_1)\in H(L)^{p+2}. Since {x1,…,xp+1,y1}\{x_1,\ldots,x_{p+1},y_1\} is a basis of NpN_p, the point hh is generic. So far we have found v∈Ep+2v\in E^{p+2} with entries in {0,±1}\{0,\pm1\} such that f∼v(h)f\sim v(h) for every f∈{x1,…,xp+1,y1,y2,y3}∪{p1k,p2k}f\in\{x_1,\ldots,x_{p+1},y_1,y_2,y_3\}\cup\{p_{1k},p_{2k}\}. For k≥4k\ge4, the element yky_k lies on the two distinct lines {y1,p1k,yk}\{y_1,p_{1k},y_k\} and {y2,p2k,yk}\{y_2,p_{2k},y_k\}, so cl⁡(yk)=cl⁡(y1,p1k)∩cl⁡(y2,p2k)\operatorname{cl}(y_k)=\operatorname{cl}(y_1,p_{1k})\cap\operatorname{cl}(y_2,p_{2k}). By Lemma 2.2 this intersection is cl⁡(w(h))\operatorname{cl}(w(h)) for some w∈Ep+2w\in E^{p+2}, hence yk∼w(h)y_k\sim w(h). Similarly, with Fj={yj}∪{xi∣i≠0,j}F_j=\{y_j\}\cup\{x_i\mid i\ne0,j\} we have cl⁡(x0)=⋂jcl⁡(Fj)\operatorname{cl}(x_0)=\bigcap_j\operatorname{cl}(F_j), and applying Lemma 2.2 inductively gives x0∼w0(h)x_0\sim w_0(h) for some w0∈Ep+2w_0\in E^{p+2}. Collecting the rows vv, ww and w0w_0 into a matrix ψ∈E(2p+3)×(p+2)\psi\in E^{(2p+3)\times(p+2)} gives a homomorphism ψ:Hp+2→H2p+3\psi:H^{p+2}\to H^{2p+3} with ψ(h)∼(x0,…,xp+1,y1,…,yp+1)\psi(h)\sim(x_0,\ldots,x_{p+1},y_1,\ldots,y_{p+1}) coordinatewise. Hence NpN_p is algebraically equivalent to a realization over HH. In particular, for Np=Lp∗N_p=L_p^*, HH needs to be Ga\mathbb{G}_a as Lp∗L_p^* is only linear over division rings of characteristic pp. For Np=NLp∗N_p=NL_p^*, HH needs to be Gm\mathbb{G}_m or an elliptic curve as NLp∗NL_p^* is only linear over division rings of characteristic not pp. □

Algebraic matroids which are not coalgebraic

The Lazarson matroid LpL_p over F‾p\overline{\mathbb{F}}_p is realized by

x1,…,xp+1,y1,…,yp+1,zx_1,\ldots,x_{p+1},y_1,\ldots,y_{p+1},z

where z=∑i=1p+1xiz=\sum_{i=1}^{p+1}x_i and yi=z−xiy_i=z-x_i. The non-Lazarson matroid NLpNL_p over F‾p\overline{\mathbb{F}}_p is realized by

u1,…,up+1,v1,…,vp+1,wu_1,\ldots,u_{p+1},v_1,\ldots,v_{p+1},w

where w=∏i=1p+1uiw=\prod_{i=1}^{p+1}u_i and vi=wuiv_i=\frac{w}{u_i}. Here the elements u1,…,up+1,x1,…,xp+1u_1,\ldots,u_{p+1},x_1,\ldots,x_{p+1} are algebraically independent over F‾p\overline{\mathbb{F}}_p. That these matroids are actually dual to Lp∗L_p^* and NLp∗NL_p^* is a routine check.

Consider the union MM of the following two matroids over F‾p\overline{\mathbb{F}}_p:

L~p=M(x1,…,xp+1,0,…,0,y1,…,yp,yp+1,z)\widetilde{L}_p=M(x_1,\ldots,x_{p+1},0,\ldots,0,y_1,\ldots,y_p,y_{p+1},z)

and

NL~p=M(0,…,0,u1,…,up+1,v1,…,vp,w,vp+1)\widetilde{\mathrm{NL}}_p=M(0,\ldots,0,u_1,\ldots,u_{p+1},v_1,\ldots,v_p,w,v_{p+1})

with the ground set {a1,…,ap+1,b1,…,bp+1,c1,…,cp+2}\{a_1,\ldots,a_{p+1},b_1,\ldots,b_{p+1},c_1,\ldots,c_{p+2}\}. Then {c1,…,cp,cp+2}∪{b1,…,bp,cp+1}\{c_1,\ldots,c_p,c_{p+2}\}\cup\{b_1,\ldots,b_p,c_{p+1}\} is a basis of MM, as both are disjoint bases of L~p\widetilde{L}_p and NL~p\widetilde{\mathrm{NL}}_p.

The algebraic matroids in the following theorem are inspired by the counterexample of Larson and Le in [14].

Theorem 6.1. Consider the matroid MpM_p represented by

x1,…,xp+1,u1,…,up+1,v1y1,…,vpyp,w⋅yp+1,vp+1⋅zx_1,\ldots,x_{p+1},u_1,\ldots,u_{p+1},v_1y_1,\ldots,v_py_p,w\cdot y_{p+1},v_{p+1}\cdot z

over F‾p\overline{\mathbb{F}}_p. Then the dual of MpM_p is not algebraic over any field.

Proof. The subset {b1,…,bp,c1,…,cp,cp+1}\{b_1,\ldots,b_p,c_1,\ldots,c_p,c_{p+1}\} is already dependent. We can see this by using u1,…,upu_1,\ldots,u_p to transform

v1y1,…,vpyp,wyp+1v_1y_1,\ldots,v_py_p,wy_{p+1}

into

up+1y1,…,up+1yp,up+1yp+1.u_{p+1}y_1,\ldots,u_{p+1}y_p,u_{p+1}y_{p+1}.

Now we have

up+1∑i=1p+1yi=up+1∑i=1p+1(z−xi)=up+1(z−∑i=1p+1xi)=0u_{p+1}\sum_{i=1}^{p+1}y_i=u_{p+1}\sum_{i=1}^{p+1}(z-x_i)=u_{p+1}\left(z-\sum_{i=1}^{p+1}x_i\right)=0

and the set {b1,…,bp,c1,…,cp,cp+1}\{b_1,\ldots,b_p,c_1,\ldots,c_p,c_{p+1}\} is dependent in MpM_p. If we contract the xix_i in MpM_p the matroid Mp/xM_p/x is equal to NL~p\widetilde{\mathrm{NL}}_p and if we contract the uiu_i the matroid Mp/uM_p/u is equal to L~p\widetilde{L}_p. Therefore MpM_p is not the union of Mp/uM_p/u and Mp/xM_p/x where we leave the contracted elements as loops. Assume the dual matroid Mp∗M_p^* is algebraic. Then Mp∗M_p^* is algebraic over a field KK of characteristic pp, as the dual of the Lazarson matroid is only algebraic over a field of characteristic pp. Then the cic_i form a basis of Mp∗M_p^*, Mp∗∖a=NL~p∗M_p^*\setminus a=\widetilde{\mathrm{NL}}_p^* and Mp∗∖b=L~p∗M_p^*\setminus b=\widetilde{L}_p^*. With Theorem 5.6 every algebraic realization of Mp∗∖aM_p^*\setminus a is algebraically equivalent to a realization over Gm\mathbb{G}_m or an elliptic curve and every algebraic realization of Mp∗∖bM_p^*\setminus b is algebraically equivalent to a realization over Ga\mathbb{G}_a. With Corollary 4.3 it follows that Mp∗M_p^* must be the intersection of Mp∗∖aM_p^*\setminus a and Mp∗∖bM_p^*\setminus b where we leave the deleted elements as coloops. This contradicts that MpM_p is not the union of Mp/uM_p/u and Mp/xM_p/x where we leave the contracted elements as loops. It follows that the dual of MpM_p is not algebraic over any field and proves Theorem 1.2. □

Our main theorem now follows directly.

Theorem 6.2. Over any field KK of positive characteristic there is an algebraic matroid MM over KK such that the dual matroid M∗M^* is not algebraic over any field.

Remark 6.3. There are several matroids similar to MpM_p with non-algebraic dual. For example one could use different twists to glue the matroids Lp∗L_p^* and NL⁡p∗\operatorname{NL}_p^* together. The twist we used just seems to be the easiest one that works. It does not work to glue LpL_p and NL⁡p\operatorname{NL}_p together with some twist and arrive at the same result.

Volume polynomials

In [6] the authors define algebraic polymatroids. Suppose we have a chain of field extensions K⊆Li⊆LK \subseteq L_i \subseteq L for all i∈Ei \in E for some finite set EE. Denote with LAL_A the composite of field extensions Li⊆LL_i \subseteq L for i∈Ai \in A. This is the smallest subfield of LL containing LiL_i for all i∈Ai \in A.

Definition 7.1. Let P=(E,r)\mathcal{P} = (E,r) be an integral polymatroid on the ground set EE with rank function rr. We call P\mathcal{P} algebraic over KK if there are field extensions K⊆Li⊆LK \subseteq L_i \subseteq L for i∈Ei \in E such that

r(A)=trdeg⁡K(LA) for all A⊆E.r(A) = \operatorname{trdeg}_{K}(L_A) \text{ for all } A \subseteq E.

See [18 Chapter 4] for several equivalent definitions of integral polymatroids. We view integral polymatroids and MM-convex sets as the same objects and denote with P∗\mathcal{P}^* the dual of a polymatroid P\mathcal{P}.

Algebraic polymatroids are connected to a special family of polynomials coming from projective geometry called volume polynomials. These are a subclass of the so-called Lorentzian polynomials, see [3]. A Lorentzian polynomial is a homogeneous polynomial with non-negative coefficients satisfying a certain log-concavity property on the positive orthant. Given a projective variety XX over a field KK of dimension dd and a collection D=(D1,…,Dn)D = (D_1, \ldots, D_n) of semiample divisors. The realizable volume polynomial of DD is defined as

fD(x1,…,xn)=1d!∫X(∑i=1nxiDi)d.f_D(x_1,\ldots,x_n) = \frac{1}{d!}\int_X \left(\sum_{i=1}^{n} x_iD_i\right)^d.

This is a homogeneous polynomial of degree dd. For a survey on volume polynomials, see [9]. The coefficients of volume polynomials obey additional inequalities compared to Lorentzian polynomials. While the support of a Lorentzian polynomial can be any polymatroid, we have the following restriction for volume polynomials:

Proposition 7.2 ([8], Proposition 5.4). The following statements are equivalent for any polymatroid P\mathcal{P} with set of bases JJ and any field KK.

(1) There is a realizable volume polynomial over KK whose support is JJ.

(2) The polymatroid P\mathcal{P} is algebraic over KK.

Theorem 1.2 immediately implies that the class of algebraic polymatroids is also not closed under duality.

Matroid duality can be generalized to the class of volume polynomials. Consider the dual pair of polynomial rings R[x]=R[x1,…,xn]\mathbb{R}[x] = \mathbb{R}[x_1,\ldots,x_n] and R[∂]=R[∂1,…,∂n]\mathbb{R}[\partial] = \mathbb{R}[\partial_1,\ldots,\partial_n] where R[∂]\mathbb{R}[\partial] acts on $\mathbb{R}[x] via the usual rule of partial differentiation. Denote with R[∂]≤μ\mathbb{R}[\partial]_{\leq\mu} all polynomials in R[∂]\mathbb{R}[\partial] of multidegree at most μ∈Zn\mu\in\mathbb{Z}^n. Then g∈R[∂]≤μg \in\mathbb{R}[\partial]_{\leq\mu} is a realizable covolume polynomial over KK if and only if g(∂)∘(1μ!xμ)g(\partial) \circ\left(\frac{1}{\mu!}x^\mu\right) is a realizable volume polynomial over KK. This definition does not depend on μ\mu. The above correspondence then yields an analogous statement for volume polynomials and their duals:

Theorem 7.3. Let KK be a field of positive characteristic. There exists a realizable covolume polynomial gg over KK such that the support of gg is not the support of any realizable volume polynomial.

Proof. Let g∈R[∂]≤μg \in\mathbb{R}[\partial]_{\leq\mu} be a realizable covolume polynomial such that the support of the dual f=g(∂)∘(1μ!xμ)f = g(\partial) \circ\left(\frac{1}{\mu!}x^\mu\right) is the set of bases of an algebraic polymatroid P\mathcal{P} with non-algebraic dual P∗\mathcal{P}^*. The dualization of P∗\mathcal{P}^* to P\mathcal{P} with respect to μ\mu is the same operation as taking the support of the dual ff of gg with respect to μ\mu. Hence, the support of gg is not the support of any realizable volume polynomial. □

References

Lukas Grund, Friedrich-Schiller-Universität Jena, Fakultät für Mathematik und Informatik, Institut für Mathematik, Inselplatz 5, 07743 Jena, Germany

Email address: [email protected]

References

  1. [1]Paolo Aluffi. “Lorentzian polynomials, Segre classes, and adjoint polynomials of convex polyhedral cones”. In: Advances in Mathematics 437 (2024), p. 109440.DOI
  2. [2]Guus P Bollen, Dustin Cartwright, and Jan Draisma. “Matroids Over One-Dimensional Groups”. In: International Mathematics Research Notices 2022.3 (Feb. 2022), pp. 2298–2336.
  3. [3]Petter Brändén and June Huh. “Lorentzian polynomials”. In: Annals of Mathematics 192.3 (2020), pp. 821–891.arxiv.org/abs/1902.03719
  4. [4]Armand Borel. Linear algebraic groups. Springer Science & Business Media, 2012.
  5. [5]Tobias Boege and Geva Yashfe. “Recognition of algebraic matroids is undecidable”. In: arXiv preprint arXiv:2607.14907 (2026).arxiv.org/abs/2607.14907
  6. [6]Federico Castillo, Yairon Cid-Ruiz, Binglin Li, Jonathan Montaño, and Naizhen Zhang. “When are multidegrees positive?” In: Advances in Mathematics 374 (2020), p. 107382.arxiv.org/abs/2005.07808
  7. [7]David M Evans and Ehud Hrushovski. “Projective planes in algebraically closed fields”. In: Proceedings of the London Mathematical Society 3.1 (1991), pp. 1–24.
  8. [8]Lukas Grund, June Huh, Mateusz Michałek, Hendrik Süss, and Botong Wang. “Linear operators preserving volume polynomials”. In: arXiv preprint arXiv:2506.22415 (2025).arxiv.org/abs/2506.22415
  9. [9]June Huh. “Volume polynomials”. In: arXiv preprint arXiv:2601.13249 (2026).arxiv.org/abs/2601.13249
  10. [10]A. W. Ingleton and R. A. Main. “Non-algebraic matroids exist”. In: Bull. London Math. Soc. 7 (1975), pp. 144–146.DOI
  11. [11]A. W. Ingleton. “Representation of matroids”. In: Combinatorial Mathematics and its Applications (Proc. Conf., Oxford, 1969). Academic Press, London-New York, 1971, pp. 149–167.
  12. [12]Serge Lang. Algebra. Springer Science & Business Media, 2012.
  13. [13]Bernt Lindström. “Matroids algebraic over F (t) are algebraic over F”. In: Combinatorica 9.1 (1989), pp. 107–109.
  14. [14]Matt Larson and Tuong Le. “Duals of algebraic matroids need not be algebraic”. In: arXiv preprint arXiv:2609.02834 (2026).arxiv.org/abs/2609.02834
  15. [15]František Matuš. “Classes of matroids closed under minors and principal extensions”. In: Combinatorica 38.4 (2018), pp. 935–954.
  16. [16]James S Milne. “A proof of the Barsotti-Chevalley theorem on algebraic groups”. In: arXiv preprint arXiv:1311.6060 (2013).arxiv.org/abs/1311.6060
  17. [17]James S Milne. Algebraic groups: the theory of group schemes of finite type over a field. Vol. 170. Cambridge University Press, 2017.
  18. [18]Kazuo Murota. Discrete Convex Analysis. Society for Industrial and Applied Mathematics, 2003.
  19. [19]James Oxley. Matroid theory. Second. Vol. 21. Oxford Graduate Texts in Mathematics. Oxford University Press, Oxford, 2011, pp. xiv+684.
  20. [20]R. Rado. “A THEOREM ON INDEPENDENCE RELATIONS”. In: The Quarterly Journal of Mathematics os-13.1 (1942), pp. 83–89.DOI
  21. [21]Maxwell Rosenlicht. “Some basic theorems on algebraic groups”. In: American Journal of Mathematics 78.2 (1956), pp. 401–443.DOI
  22. [22]Bartel Leendert Van der Waerden. “Moderne algebra”. In: (1931).

Paper details

vlnc:W_vh9mraqw
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Claude Opus 5 (bare)

View statement
Looking for related papers…
Contents