betaFeedback

WorksMathematics

The E6E_6-Grassmannian, its minuscule Coxeter matroids, and the lines on a cubic surface

Abstract

We study the minuscule Coxeter matroids of type E6E_6, whose ground set is the set of 27 lines on a smooth cubic surface. We characterize them by a tritangent exchange property and classify them into 230 orbits under the Weyl group. Every nonempty matroidal stratum of the E6E_6-Grassmannian is an open subscheme of a torus, and all orbits but one are realizable over every infinite field. Finally, we compute the tropical E6E_6-Grassmannian and its 60 classes of toric degenerations. This is an extended abstract for a forthcoming paper.

Introduction

The E6E_6-Grassmannian is a remarkable variety that can be approached from several directions: through the exceptional Jordan algebra over the octonions (where it appears as the projective plane over the complexified octonions and is therefore also called the Cayley plane), through the representation theory of E6E_6, and through the 27 lines on a smooth cubic surface. We show that the last of these gives an illuminating way to understand the (E6,P6)(E_6,P_6)–matroids and their strata in the E6E_6-Grassmannian. Here P6P_6 is the maximal parabolic subgroup associated with the sixth simple root, numbered as in Bourbaki [2 Plate V]. The matroid axiom becomes an exchange property for tritangent planes, the defining quadrics of E6/P6E_6/P_6 are signed sums over the tritangent planes through a line, and the analysis of each stratum is governed by how the tritangent planes through different lines meet.

The homogeneous coordinates of E6/P6⊆P26E_6/P_6 \subseteq\mathbb{P}^{26} are indexed by the 27 weights of the minuscule representation of E6E_6, whose convex hull is the Gosset polytope 2212_{21}; the (E6,P6)(E_6,P_6)–matroids are certain subsets of its vertices. Coxeter identified the vertices of 2212_{21} with the 27 lines on a smooth cubic surface [5], and Lee realized 2212_{21} as the convex hull of the classes of those lines in the Picard lattice [15]. We start from Lee’s description and project it to the orthogonal complement of the canonical class, which places 2212_{21} in the weight lattice MM of E6E_6. The inner product on MM then turns combinatorial statements about 2212_{21} into statements about lines on a cubic surface, and this dictionary drives all of our results.

Partially supported by the AMS-Simons Research Enhancement Grant for PUI Faculty.

The combinatorics of (E6,P6)(E_6,P_6)–matroids

Using the inner product on MM, we characterize (E6,P6)(E_6,P_6)–matroids by a tritangent exchange property (Theorem 2.4): a set QQ of lines is a matroid if and only if, whenever two lines of QQ meet, the third line of their tritangent plane lies on a second tritangent plane whose other two lines also belong to QQ. By the Gelfand–Serganova theorem, a set of lines is a matroid exactly when every edge of its convex hull is parallel to a root of E6E_6; testing this condition on every orbit of subsets of the 27 lines shows that there are 1 335 373 (E6,P6)(E_6,P_6)–matroids, forming 230 orbits under W(E6)W(E_6) (Theorem 2.5).

Matroidal strata of the E6E_6-Grassmannian

As for Gr⁡(k,n)\operatorname{Gr}(k,n), the support of every point of E6/P6E_6/P_6 is an (E6,P6)(E_6,P_6)–matroid, and the points with a given support QQ form a locally closed subscheme MQ\mathcal{M}_Q defined over Z\mathbb{Z}, the matroidal stratum of QQ (§ 3). In a chart {py≠0}\{p_y \ne0\}, the coordinates of the 16 lines skew to yy are free, and the coordinates of the 10 lines meeting yy are quadrics in them (Theorem 3.1); so each stratum is cut out of a torus by the vanishing of some of these quadrics. Our main geometric result (Theorem 3.6) is that these strata are as simple as possible. Up to symmetry, exactly one (E6,P6)(E_6,P_6)–matroid is realizable over no field. Every other stratum is isomorphic over Z\mathbb{Z} to an open subscheme of an algebraic torus, so it is smooth, irreducible and rational, and the matroid is realizable over every infinite field. Both the non-realizable matroid and the one stratum that requires a separate argument are explained by a single linear relation among the chart quadrics.

Tropicalization and toric degenerations

Finally, we compute the tropicalization of the open stratum of E6/P6E_6/P_6 (Theorem 4.1). It is a simplicial fan with 12 885 rays and 8 712 360 maximal cones, in 10 and 421 symmetry classes, and every cone induces a subdivision of 2212^{21} into matroid polytopes; the classification of (E6,P6)(E_6,P_6)–matroids (Theorem 2.5) supplies the starting cone that the traversal algorithm needs. Exactly 60 classes of maximal cones are prime, and each gives a flat degeneration of E6/P6E_6/P_6 to a projectively normal Gorenstein toric Fano variety whose polytope has 27 lattice points and normalized volume 78 (Theorem 4.3). One class is the Hibi degeneration [GL96], and two give the chain polytope of the minuscule E6E_6 poset.

(E6,P6)–matroids

Let I1,6=Zh⊕Za1⊕⋯⊕Za6I_{1,6}=\mathbb{Z}h\oplus\mathbb{Z}a_1\oplus\cdots\oplus\mathbb{Z}a_6 be the lattice with symmetric bilinear form h⋅h=1h\cdot h=1, ai⋅aj=−δija_i\cdot a_j=-\delta_{ij}, h⋅ai=0h\cdot a_i=0, and put K=−3h+∑iaiK=-3h+\sum_i a_i, so that K⋅K=3K\cdot K=3. This is the Picard lattice of a smooth cubic surface S⊂P3S\subset\mathbb{P}^3 with its canonical class [7 Ch. 9]. With this connection, we use classical vocabulary below (lines, tritangent planes, double sixes). The orthogonal complement K⊥⊂I1,6K^\perp\subset I_{1,6} is the E6E_6 root lattice, whose 72 roots are the vectors α\alpha with α⋅α=−2\alpha\cdot\alpha=-2.

Definition 2.1. Let π:I1,6⊗R→K⊥⊗R\pi: I_{1,6} \otimes\mathbb{R} \to K^{\perp} \otimes\mathbb{R} be the orthogonal projection, π(x)=x−13(x⋅K)K\pi(x)=x-\frac{1}{3}(x\cdot K)K. Set

M=π(I1,6),MR=M⊗R=K⊥⊗R,⟨u,v⟩=−u⋅v,M=\pi(I_{1,6}),\qquad M_{\mathbb{R}}=M\otimes\mathbb{R}=K^{\perp}\otimes\mathbb{R},\qquad\langle u,v\rangle=-u\cdot v,

so that ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle is a positive definite W(E6)W(E_6)-invariant inner product on MRM_{\mathbb{R}}.

Since I1,6I_{1,6} is unimodular and KK is primitive, MM is the dual of K⊥K^{\perp} with respect to ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle. Thus MM is the weight lattice of E6E_6; it contains the root lattice K⊥K^{\perp} with index 33, and M/K⊥≅Z/3ZM/K^{\perp}\cong\mathbb{Z}/3\mathbb{Z} is generated by the class of π(a1)=a1+13K\pi(a_1)=a_1+\frac{1}{3}K. The Weyl group W(E6)W(E_6), generated by the reflections in the roots, acts on MM. The 2727 vectors x∈I1,6x\in I_{1,6} with x⋅x=−1x\cdot x=-1 and x⋅K=−1x\cdot K=-1 are the classes of the 2727 lines on the cubic surface,

ai,bi=2h−∑j≠iaj,cij=h−ai−aj,a_i,\qquad b_i=2h-\sum_{j\ne i}a_j,\qquad c_{ij}=h-a_i-a_j,

and each projects to π(x)=x+13K\pi(x)=x+\frac{1}{3}K. We keep the names ai,bi,cija_i,b_i,c_{ij} for these projections and write L⊆M\mathcal{L}\subseteq M for the resulting set of 2727 vectors. Equivalently,

L={u∈π(a1)+K⊥:⟨u,u⟩=43},\mathcal{L}=\left\{u\in\pi(a_1)+K^{\perp}:\langle u,u\rangle=\frac{4}{3}\right\},

the set of weights of the 2727-dimensional minuscule representation VV of E6E_6 (the other nontrivial coset of K⊥K^{\perp} gives the weights of V∗V^*). Following [15], the Gosset polytope is

221=conv⁡(L)⊆MR,2_{21}=\operatorname{conv}(\mathcal{L})\subseteq M_{\mathbb{R}},

a lattice polytope in MM centered at the origin, with W(E6)W(E_6) as its full symmetry group. From ⟨π(x),π(y)⟩=−x⋅y+13\langle\pi(x),\pi(y)\rangle=-x\cdot y+\frac{1}{3} we read off that two distinct elements x,y∈Lx,y\in\mathcal{L} satisfy ⟨x,y⟩∈{13,−23}\langle x,y\rangle\in\{\frac{1}{3},-\frac{2}{3}\}, according as the corresponding lines are skew or meet, and that x,yx,y span an edge of 2212_{21} exactly when ⟨x,y⟩=13\langle x,y\rangle=\frac{1}{3}, that is, when x−yx-y is a root.

Fix y∈Ly\in\mathcal{L}. The inner product partitions the remaining elements as

L∖{y}=N(y)⊔◊(y),N(y)={x:⟨x,y⟩=13},◊(y)={x:⟨x,y⟩=−23},(1)\mathcal{L}\setminus\{y\}=N(y)\sqcup\Diamond(y),\qquad N(y)=\left\{x:\langle x,y\rangle=\frac{1}{3}\right\},\qquad\Diamond(y)=\left\{x:\langle x,y\rangle=-\frac{2}{3}\right\}, \tag*{(1)}

where N(y)N(y) is the set of 1616 neighbors of yy in 2212_{21}, and the 1010 elements of ◊(y)\Diamond(y) are the vertices of the cross-polytope facet of 2212_{21} opposite to yy. Two vertices x,x′x,x' of ◊(y)\Diamond(y) are antipodal in that cross-polytope if and only if ⟨x,x′⟩=−23\langle x,x'\rangle=-\frac{2}{3}, if and only if

x+x′+y=0in M.x+x'+y=0\quad\text{in }M.

More generally, a triple {x,y,z}\{x,y,z\} of distinct elements of L\mathcal{L} is called a tritangent plane if x+y+z=0x+y+z=0 in MM. In the cubic surface description, −K-K is the class of a hyperplane section of S⊆P3S\subseteq\mathbb{P}^3, so x+y+z=0x+y+z=0 in MM (equivalently, the classes sum to −K-K in I1,6I_{1,6}) means that the three lines are coplanar: some plane H⊆P3H\subseteq\mathbb{P}^3 cuts out H∩S=x∪y∪zH\cap S=x\cup y\cup z.

So the five antipodal pairs of ◊(y)\Diamond(y) are the five tritangent planes through yy, and there are 4545 tritangent planes in all. We write x∗=−x−yx^*=-x-y for the antipode of xx in ◊(y)\Diamond(y).

Example 2.2. For y=a6y=a_{6} we get ◊(a6)={b1,…,b5,c16,…,c56}\Diamond(a_{6})=\{b_{1},\ldots,b_{5},c_{16},\ldots,c_{56}\} with antipodal pairs {bi,ci6}\{b_{i},c_{i6}\}, and N(a6)N(a_{6}) consists of a1,…,a5,b6a_{1},\ldots,a_{5},b_{6} and the ten cijc_{ij} with i,j≤5i,j\leq5.

By the Gelfand–Serganova theorem [1], an (E6,P6)(E_{6},P_{6})–matroid is a nonempty subset Q⊆LQ\subseteq\mathcal{L} such that every edge of the polytope conv⁡(Q)\operatorname{conv}(Q) is parallel to a root of E6E_{6}. It is a strong (E6,P6)(E_{6},P_{6})–matroid if for all distinct u,v∈Qu,v\in Q there is a root α\alpha whose reflecting hyperplane separates uu and vv and such that the reflections of uu and vv in α\alpha both lie in QQ.

Definition 2.3. A nonempty Q⊆LQ\subseteq\mathcal{L} has the tritangent exchange property if, for every a,b∈Qa,b\in Q with ⟨a,b⟩=−23\langle a,b\rangle=-\frac{2}{3}, there are c,d∈Qc,d\in Q with ⟨c,d⟩=−23\langle c,d\rangle=-\frac{2}{3}, {c,d}≠{a,b}\{c,d\}\neq\{a,b\}, and

a+b=c+din M.a+b=c+d\quad\text{in }M.

Geometrically, if aa and bb are two meeting lines of QQ and y=−a−by=-a-b is the third line of their tritangent plane, the property asks for a second tritangent plane {c,d,y}\{c,d,y\} through yy with c,d∈Qc,d\in Q; in other words, a tritangent plane through yy with two lines in QQ can always be exchanged for another one. Since π\pi is injective on each level set {x⋅K=k}\{x\cdot K=k\} of I1,6I_{1,6}, the relation a+b=c+da+b=c+d may be verified among the corresponding classes in I1,6I_{1,6}.

Theorem 2.4. For a nonempty Q⊆LQ\subseteq\mathcal{L} the following are equivalent.

(i) QQ is an (E6,P6)(E_{6},P_{6})–matroid.

(ii) QQ is a strong (E6,P6)(E_{6},P_{6})–matroid.

(iii) For every y∈Ly\in\mathcal{L}, the number of antipodal pairs of ◊(y)\Diamond(y) contained in QQ is not equal to one.

(iv) QQ has the tritangent exchange property.

Indication of proof. The equivalence of (i) and (ii) is [10 Thm. 4.22], and the equivalence of (ii) and (iii) is [4 Prop. 3.15]. For a direct proof of (i)⇒\Rightarrow(iii), if ◊(y)∩Q\Diamond(y)\cap Q contains exactly one antipodal pair {x,x∗}\{x,x^{*}\}, an explicit linear functional on the cross-polytope shows that [x,x∗][x,x^{*}] is an edge of conv⁡(◊(y)∩Q)\operatorname{conv}(\Diamond(y)\cap Q), which is a face of conv⁡(Q)\operatorname{conv}(Q) because ◊(y)\Diamond(y) spans a facet; but ⟨x−x∗,x−x∗⟩=4\langle x-x^{*},x-x^{*}\rangle=4, so x−x∗x-x^{*} is not parallel to a root. For (iii)⇔\Leftrightarrow(iv), if ⟨a,b⟩=−23\langle a,b\rangle=-\frac{2}{3} then y=−a−by=-a-b lies in L\mathcal{L} by the description of L\mathcal{L} as the vectors of norm 43\frac{4}{3} in a coset, and {a,b}\{a,b\} is an antipodal pair of ◊(y)\Diamond(y); conversely ⟨c+d,y⟩=⟨a+b,y⟩=−43\langle c+d,y\rangle=\langle a+b,y\rangle=-\frac{4}{3} forces c,d∈◊(y)c,d\in\Diamond(y). □

In particular the adjective “strong” is redundant for (W(E6),P6)(W(E_{6}),P_{6}) and for all minuscule types by [10], but it fails for general pairs (W,P)(W,P).

Theorem 2.5. There are exactly 1 335 373 (E6,P6)(E_{6},P_{6})–matroids, forming 230 orbits under W(E6)W(E_{6}). By the dimension of conv⁡(Q)\operatorname{conv}(Q) the orbits are distributed as follows.

dim⁡conv⁡(Q)\dim\operatorname{conv}(Q)0123456
W(E6)W(E_{6})-orbits11241130181

Table 2.5.

Indication of proof. Order L\mathcal{L} and represent each orbit of subsets by its lexicographically minimal element. Every orbit of (k+1)(k+1)-subsets contains R∪{v}R \cup\{v\} for some minimal representative RR of size kk, so the representatives can be grown one line at a time; those of size at least 14 are complements of those of size at most 13. This yields the 5 486 orbits of subsets of L\mathcal{L}, and for each representative we test whether every edge of its convex hull is parallel to a root. □

Write ΓQ\Gamma_Q for the graph on QQ joining xx and zz when ⟨x,z⟩=−23\langle x,z\rangle=-\frac{2}{3}. Each edge {x,z}\{x,z\} of ΓQ\Gamma_Q determines the line −x−z-x-z, the third line of their tritangent plane. One matroid plays a special role below:

Q1={a1,…,a6,b1,…,b6,c12,c13,c23,c45,c46,c56}.Q_1=\{a_1,\ldots,a_6,b_1,\ldots,b_6,c_{12},c_{13},c_{23},c_{45},c_{46},c_{56}\}.

Its complement is the union of three disjoint tritangent planes, namely {c14,c25,c36}\{c_{14},c_{25},c_{36}\}, {c15,c26,c34}\{c_{15},c_{26},c_{34}\} and {c16,c24,c35}\{c_{16},c_{24},c_{35}\}, and every vertex of ΓQ1\Gamma_{Q_1} has degree 7.

Matroidal strata of the E6E_6-Grassmannian

Equations and charts

Let R=Z[pℓ:ℓ∈L]R=\mathbb{Z}[p_\ell:\ell\in\mathcal{L}] and let CC be the skew-symmetric 6×66\times6 matrix with entries pcijp_{c_{ij}} above the diagonal. The E6E_6-invariant cubic on the 27-dimensional minuscule representation, which goes back to Cartan’s thesis [3] (see [16], and also [4 Prop. 4.15]), is a signed sum of 45 monomials pxpypzp_xp_yp_z, one for each tritangent plane, that is, each triple with x+y+z=0x+y+z=0 in MM [8, 16]. With a=(pa1,…,pa6)T\mathbf{a}=(p_{a_1},\ldots,p_{a_6})^{T} and b=(pb1,…,pb6)T\mathbf{b}=(p_{b_1},\ldots,p_{b_6})^{T} it is

D=aTCb+Pf⁡(C).D=\mathbf{a}^{T}C\mathbf{b}+\operatorname{Pf}(C).

The 30 monomials of aTCb\mathbf{a}^{T}C\mathbf{b} are the tritangent planes {ai,bj,cij}\{a_i,b_j,c_{ij}\}, and the 15 monomials of Pf⁡(C)\operatorname{Pf}(C) are the tritangent planes {cij,ckl,cmn}\{c_{ij},c_{kl},c_{mn}\} with {i,j}⊔{k,l}⊔{m,n}={1,…,6}\{i,j\}\sqcup\{k,l\}\sqcup\{m,n\}=\{1,\ldots,6\}. The E6E_6-Grassmannian is the singular locus of the hypersurface defined by D=0D=0 [13], so it is cut out by the partial derivatives of DD:

E6/P6=Proj⁡R/⟨Dy:y∈L⟩,Dy=∂D∂py.E_6/P_6=\operatorname{Proj} R/\langle D_y:y\in\mathcal{L}\rangle,\qquad D_y=\frac{\partial D}{\partial p_y}.

Differentiating the monomials shows that each quadric is a signed sum over the tritangent planes through yy:

Dy=∑{x,x∗}⊆⋄(y)±pxpx∗.(2)D_y=\sum_{\{x,x^*\}\subseteq\diamond(y)}\pm p_xp_{x^*}. \tag*{(2)}

(In the full version we match these coordinates with the model of E6/P6E_6/P_6 inside the exceptional Jordan algebra over the octonions, under which DD becomes the determinant.)

Fix y∈Ly\in\mathcal{L} and let Uy={py≠0}⊆E6/P6U_y=\{p_y\ne0\}\subseteq E_6/P_6. For v∈⋄(y)v\in\diamond(y) we have y∈⋄(v∗)y\in\diamond(v^*), so by (2) the quadric Dv∗D_{v^*} contains the monomial pvpyp_vp_y; write Dv∗=±(pvpy−gv)D_{v^*}=\pm(p_vp_y-g_v). The other four tritangent pairs {u,u∗}\{u,u^*\} of ⋄(v∗)\diamond(v^*) satisfy u+u∗=v+yu+u^*=v+y, hence lie in N(y)N(y), so gv∈Z[pu:u∈N(y)]g_v\in\mathbb{Z}[p_u:u\in N(y)] is a signed sum of four products pupu∗p_up_{u^*}.

Theorem 3.1 (Big cell). For every y∈Ly \in\mathcal{L} the morphism

A16=Spec⁡Z[pu:u∈N(y)]⟶Uy,py=1,pv=gv (v∈◊(y)),\mathbb{A}^{16} = \operatorname{Spec}\mathbb{Z}[p_u : u \in N(y)] \longrightarrow U_y,\qquad p_y = 1,\quad p_v = g_v\ (v \in\Diamond(y)),

which keeps the coordinates pup_u, u∈N(y)u \in N(y), is an isomorphism.

Indication of proof. On UyU_y, with py=1p_y = 1, the equations Dv∗=0D_{v^*} = 0 say exactly that pv=gvp_v = g_v. It remains to check that the other 17 quadrics vanish identically after this substitution: DyD_y becomes a quartic ∑±gvgv∗\sum\pm g_vg_{v^*}, and each DwD_w, w∈N(y)w \in N(y), becomes a relation ∑±pugv\sum\pm p_ug_v that is linear in the gvg_v. This is a straightforward computation in one chart, and W(E6)W(E_6) permutes the charts transitively by signed permutations of the coordinates. □

Supports and strata

For a field K\mathbb{K} and a K\mathbb{K}-point pp of E6/P6E_6/P_6, let supp⁡(p)={ℓ∈L:pℓ≠0}\operatorname{supp}(p)=\{\ell\in\mathcal{L}:p_\ell\ne0\}.

Proposition 3.2. The support of every point of E6/P6E_6/P_6 is an (E6,P6)(E_6,P_6)–matroid.

Indeed, if ◊(y)∩supp⁡(p)\Diamond(y) \cap\operatorname{supp}(p) contained exactly one tritangent pair, then by (2) the quadric DyD_y would evaluate to a single nonzero monomial at pp. This is analogous to the fact that the support of a point of Gr⁡(k,n)\operatorname{Gr}(k,n) is a matroid, and it motivates the following definition.

Definition 3.3. The matroidal stratum of an (E6,P6)(E_6,P_6)–matroid QQ is the locally closed subscheme

MQ={p∈E6/P6:supp⁡(p)=Q},\mathcal{M}_Q = \{p \in E_6/P_6 : \operatorname{supp}(p) = Q\},

defined over Z\mathbb{Z}. We say QQ is realizable over K\mathbb{K} if MQ\mathcal{M}_Q has a K\mathbb{K}-point.

Thus E6/P6=⋃QMQE_6/P_6 = \bigcup_Q \mathcal{M}_Q, with W(E6)W(E_6) permuting the strata, and the open stratum ML\mathcal{M}_{\mathcal{L}} is the locus where all 27 coordinates are nonzero. These are analogous to the thin Schubert cells in [9] of the ordinary Grassmannian.

The equations of a stratum

Fix y∈Qy \in Q. Every point of MQ\mathcal{M}_Q lies in the chart UyU_y, and setting the coordinates of N(y)∖QN(y) \setminus Q to zero in Theorem 3.1 gives

MQ≅{p∈GmQ∩N(y):gv(p)=0 (v∈Ey),gv(p)≠0 (v∈◊(y)∩Q)}.(3)\mathcal{M}_Q \cong\left\{p \in\mathbb{G}_m^{Q\cap N(y)} : g_v(p)=0\ (v \in E_y),\quad g_v(p)\ne0\ (v \in\Diamond(y)\cap Q)\right\}. \tag*{(3)}

where we keep writing gvg_v after setting the coordinates of N(y)∖QN(y) \setminus Q to zero, and EyE_y is the set of v∈◊(y)∖Qv \in\Diamond(y) \setminus Q for which gvg_v is not the zero polynomial, so that gv=0g_v=0 is a genuine equation. Each gvg_v is a signed sum of monomials pupu∗p_up_{u^*} over antipodal pairs {u,u∗}\{u,u^*\} of ◊(v∗)\Diamond(v^*), which are disjoint, so no variable appears twice; and by Theorem 2.4(iii) gvg_v is never a monomial. Hence each equation can be solved for any one of its variables. To organize the elimination, note that the monomial pupu∗p_up_{u^*} of gvg_v comes from two meeting lines of N(y)N(y) with u+u∗=v+yu+u^*=v+y. So we define the exchange graph ΓQ(y)\Gamma_Q(y) as the subgraph of ΓQ\Gamma_Q on Q∩N(y)Q\cap N(y) whose edges are the pairs {u,u′}\{u,u'\} with u+u′−y∉Qu+u'-y\notin Q. Its edges are exactly the monomials of the equations, and the label u+u′−y∈Eyu+u'-y\in E_y partitions them into one matching per equation. A leaf order is an ordering of the matchings such that each one contains a vertex of degree one in the union of itself and its successors. The key combinatorial input is the following.

Lemma 3.4. Let v,v′∈◊(y)v,v' \in\Diamond(y) be distinct and let {u,u∗}⊆N(y)\{u,u^*\} \subseteq N(y) be a tritangent pair of ◊(v∗)\Diamond(v^*). Then {u,u∗}\{u,u^*\} meets ◊(v′∗)\Diamond(v'^*) in no vertex if v′=v∗v'=v^*, and in exactly one vertex otherwise.

This is a one-line computation, ⟨u+u∗,v′∗⟩=⟨v+y,v′∗⟩\langle u+u^*,v'^*\rangle=\langle v+y,v'^*\rangle (both pairs sum to −v∗-v^* in MM), and it implies that two matchings of ΓQ(y)\Gamma_Q(y) never have the same vertex set, so every chart with at most two equations has a leaf order. Eliminating one variable per equation along a leaf order identifies MQ\mathcal{M}_Q with an open subscheme of a torus over Z\mathbb{Z}.

Example 3.5. Let Q={a1,…,a5,b1,…,b4,b6,c12,c13,c24,c35,c36,c45,c46}Q=\{a_1,\ldots,a_5,b_1,\ldots,b_4,b_6,c_{12},c_{13},c_{24},c_{35},c_{36},c_{45},c_{46}\}, the (E6,P6)(E_6,P_6)–matroid that supplies the starting cone in the proof of Theorem 4.1, and take y=a3y=a_3. Then Q∩◊(y)Q\cap\Diamond(y) consists of b1b_1, b2b_2, b4b_4, b6b_6, c13c_{13}, c35c_{35}, c36c_{36}, and Q∩N(y)Q\cap N(y) of a1a_1, a2a_2, a4a_4, a5a_5, b3b_3, c12c_{12}, c24c_{24}, c45c_{45}, c46c_{46}. Figure 1(a) shows ΓQ\Gamma_Q, with yy on top, Q∩◊(y)Q\cap\Diamond(y) in the middle row and Q∩N(y)Q\cap N(y) in the bottom row. Each of the 13 edges {u,u′}\{u,u'\} of ΓQ\Gamma_Q on Q∩N(y)Q\cap N(y) is labeled by the line u+u′−yu+u'-y; for instance c12+c46=2h−a1−a2−a4−a6=b5+a3c_{12}+c_{46}=2h-a_1-a_2-a_4-a_6=b_5+a_3 in I1,6I_{1,6}. Figure 1(b) shows these edges with their labels. Seven labels lie in QQ (dashed edges), and these edges give the coordinates

pb1=pa1pb3,pb2=pa2pb3,pb4=pa4pb3,pb6=−pc12pc45,p_{b_1}=p_{a_1}p_{b_3},\quad p_{b_2}=p_{a_2}p_{b_3},\quad p_{b_4}=p_{a_4}p_{b_3},\quad p_{b_6}=-p_{c_{12}}p_{c_{45}},
pc13=−pa2pc12,pc35=−pa4pc45,pc36=−pa4pc46,p_{c_{13}}=-p_{a_2}p_{c_{12}},\quad p_{c_{35}}=-p_{a_4}p_{c_{45}},\quad p_{c_{36}}=-p_{a_4}p_{c_{46}},

which are automatically units. The remaining six (solid) edges form ΓQ(a3)\Gamma_Q(a_3), with matchings Mb5M_{b_5} (blue), Mc23M_{c_{23}} (orange) and Mc34M_{c_{34}} (green), so Ey={b5,c23,c34}E_y=\{b_5,c_{23},c_{34}\} and the equations are

gb5=pa5pb3+pc12pc46=0,gc23=pa1pc12−pa4pc24=0,g_{b_5}=p_{a_5}p_{b_3}+p_{c_{12}}p_{c_{46}}=0,\quad g_{c_{23}}=p_{a_1}p_{c_{12}}-p_{a_4}p_{c_{24}}=0,
gc34=pa5pc45−pa2pc24=0.g_{c_{34}}=p_{a_5}p_{c_{45}}-p_{a_2}p_{c_{24}}=0.

With three equations Lemma 3.4 does not apply directly, but ΓQ(a3)\Gamma_Q(a_3) is the disjoint union of the paths a2c24a4a_2c_{24}a_4, c46c12a1c_{46}c_{12}a_1 and b3a5c45b_3a_5c_{45}, so each matching contains a leaf of the whole graph and every order is a leaf order. Solving each equation for a leaf that occurs in no other equation gives

pa1=pa4pc24pc12,pa2=pa5pc45pc24,pb3=−pc12pc46pa5.p_{a_1}=\frac{p_{a_4}p_{c_{24}}}{p_{c_{12}}},\quad p_{a_2}=\frac{p_{a_5}p_{c_{45}}}{p_{c_{24}}},\quad p_{b_3}=-\frac{p_{c_{12}}p_{c_{46}}}{p_{a_5}}.

Every coordinate is now a Laurent monomial in pa4,pa5,pc12,pc24,pc45,pc46p_{a_4},p_{a_5},p_{c_{12}},p_{c_{24}},p_{c_{45}},p_{c_{46}}, so MQ≅Gm6\mathcal{M}_Q\cong\mathbb{G}_m^6.

The graphs of Example 3.5

Figure 1. The graphs of Example 3.5.

Theorem 3.6. Let QQ be an (E6,P6)(E_6,P_6)–matroid. If QQ is W(E6)W(E_6)-equivalent to

Qnr=Q1∪{c14}={a1,…,a6,b1,…,b6,c12,c13,c14,c23,c45,c46,c56},Q_{\mathrm{nr}}=Q_1\cup\{c_{14}\}=\{a_1,\ldots,a_6,b_1,\ldots,b_6,c_{12},c_{13},c_{14},c_{23},c_{45},c_{46},c_{56}\},

then MQ\mathcal{M}_Q is empty over every field. Otherwise MQ\mathcal{M}_Q is isomorphic over Z\mathbb{Z} to an open subscheme of an algebraic torus with nonempty fiber over every prime. In

particular QQ is realizable over every infinite field, and MQM_Q is smooth, irreducible and rational, of dimension ∣Q∩N(y)∣−∣Ey∣|Q \cap N(y)| - |E_y| for any chart yy admitting a leaf order.

Indication of proof. A chart with at most two equations has a leaf order by Lemma 3.4. A check over the 230 orbit representatives shows that every QQ other than Q1Q_1 has a chart admitting a leaf order. In each chart of Q1Q_1 the three equations are the 2×22 \times2 minors of a 2×32 \times3 matrix, which cut out a translate of an 8-dimensional subtorus. For QnrQ_{\mathrm{nr}}, a linear relation among the gvg_v in the chart at a1a_1 forces pa4gc14=0p_{a_4}g_{c_{14}} = 0, although both factors must be units. Finally, strong Gröbner bases over Z\mathbb{Z} with leading coefficients ±1\pm1 show that the fibers over Spec⁡Z\operatorname{Spec}\mathbb{Z} are nonempty. □

dim⁡MQ\dim M_Q012345678910111213141516
orbits112381223344338301794211

Table 1. Dimensions of the nonempty matroidal strata, up to W(E6)W(E_6).

Remark 3.7. The (E6,P6)(E_6, P_6)–matroids other than QnrQ_{\mathrm{nr}} might not be realizable over certain finite fields. For instance L\mathcal{L} itself is not realizable over F2\mathbb{F}_2. Indeed, a point with all coordinates nonzero would have all coordinates equal to 11, and then each quadric (2) would equal a sum of five terms ±1\pm1, which is odd.

Remark 3.8. Although each stratum is smooth and irreducible, the strata do not form a stratification in the sense that the closure of a stratum is not necessarily the union of smaller strata. For example, for Q={a1,…,a5,b6,c12,…,c15,c23,c45}Q = \{a_1, \ldots, a_5, b_6, c_{12}, \ldots, c_{15}, c_{23}, c_{45}\} and P={a2,…,a5,c12,…,c15}P = \{a_2, \ldots, a_5, c_{12}, \ldots, c_{15}\} we have that MQ\mathcal{M}_Q meets MP\mathcal{M}_P without containing it.

Tropicalization and toric degenerations

The tropical E6_6-Grassmannian

Let X∘=MLX^\circ= \mathcal{M}_{\mathcal{L}} be the open stratum, a closed subvariety of the dense torus Gm27/Gm\mathbb{G}_{m}^{27}/\mathbb{G}_m of P26\mathbb{P}^{26} which is a smooth and irreducible very affine variety. Let I=⟨Dy:y∈L⟩I = \langle D_y : y \in\mathcal{L}\rangle. The initial form in⁡wf\operatorname{in}_w f picks out the terms of ff of minimal ww-weight. Since II is homogeneous, in⁡wI\operatorname{in}_w I depends only on the class of ww in R27/R1≅R26\mathbb{R}^{27}/\mathbb{R}1 \cong\mathbb{R}^{26}, and Trop⁡(X∘)⊆R27/R1\operatorname{Trop}(X^\circ) \subseteq\mathbb{R}^{27}/\mathbb{R}1 is the set of classes ww for which in⁡wI\operatorname{in}_w I contains no monomial [17]. It is a pure 16-dimensional fan. Its lineality space Λ\Lambda is the 6-dimensional subspace given by the image of MR∗M^*_{\mathbb{R}} under η↦(η(ℓ))ℓ∈L\eta\mapsto(\eta(\ell))_{\ell\in\mathcal{L}}, that is, the space of affine-linear functions on 2212_{21} modulo constants.

Theorem 4.1. The restriction of the Gröbner fan of II to Trop⁡(X∘)\operatorname{Trop}(X^\circ) is simplicial and each of its maximal cones has multiplicity one. Modulo Λ\Lambda its ff-vector is

(12 885, 201 261, 1 471 647, 6 556 740, 19 592 172,40 753 818, 59 005 476, 57 133 440, 33 247 440, 8 712 360).(12\,885,\ 201\,261,\ 1\,471\,647,\ 6\,556\,740,\ 19\,592\,172,\\ 40\,753\,818,\ 59\,005\,476,\ 57\,133\,440,\ 33\,247\,440,\ 8\,712\,360).

Up to the action of W(E6)W(E_6) it is (10,56,224,697,1639,2841,3558,3070,1639,421)(10, 56, 224, 697, 1639, 2841, 3558, 3070, 1639, 421).

For each ww in Trop⁡(X∘)\operatorname{Trop}(X^\circ), the regular subdivision of 2212_{21} induced by ww is a subdivision into (E6,P6)(E_6, P_6)–matroid polytopes.

Indication of proof. We computed the fan with gfan_tropicaltraverse [14], up to the symmetry of W(E6)W(E_6). This traversal needs a maximal cone to start from, and the heuristic for finding one, gfan_tropicalstartingcone, did not work for II. Instead, we started from the height −1Q-1_Q, the negative of the indicator vector of the matroid QQ of Example 3.5. For the last statement, up to signs the quadrics DyD_y are the strong exchange equations defining the Coxeter Dressian Dr⁡(E6,P6)\operatorname{Dr}(E_6, P_6), so Trop⁡(X∘)⊆Dr⁡(E6,P6)\operatorname{Trop}(X^\circ) \subseteq\operatorname{Dr}(E_6, P_6), and every point of the Dressian induces a matroidal subdivision by [4 Prop. 4.15] and [10 Thm. 4.22]. We also verified it directly on a representative of each of the 14 155 orbits of cones.

The ten orbits of rays have sizes 27, 36, 36, 216, 240, 270, 540, 720, 4 320 and 6 480; the two orbits of size 36 are spanned, modulo Λ\Lambda, by the indicator vectors of the 36 double sixes and of their complements. The subdivisions induced by maximal cones have between 51 and 69 maximal cells.

Proposition 4.2. The inclusion Trop⁡(X∘)⊆Dr⁡(E6,P6)\operatorname{Trop}(X^\circ) \subseteq\operatorname{Dr}(E_6, P_6) is strict.

Indication of proof. Let ww be the vector with wℓ=−1w_\ell=-1 for ℓ∈Qnr\ell\in Q_{\mathrm{nr}} and wℓ=0w_\ell=0 otherwise. It induces a matroidal subdivision of 2212_{21} into 27 cells, one of which is conv⁡(Qnr)\operatorname{conv}(Q_{\mathrm{nr}}), so w∈Dr⁡(E6,P6)w \in\operatorname{Dr}(E_6, P_6). A direct calculation of the initial ideal verified that w∉Trop⁡(X∘)w \notin\operatorname{Trop}(X^\circ).

Toric degenerations

A maximal cone σ\sigma of Trop⁡(X∘)\operatorname{Trop}(X^\circ) is prime if in⁡wI\operatorname{in}_w I is a prime ideal for ww in its relative interior. In that case the Gröbner degeneration E6/P6E_6/P_6 to V(in⁡wI)V(\operatorname{in}_w I) is a flat degeneration to a variety which, having a dense torus orbit and a binomial ideal, is a projective toric variety XAσX_{A_\sigma}; here Aσ⊆Z17A_\sigma\subseteq\mathbb{Z}^{17} is the configuration of 27 lattice points obtained by dividing Z27\mathbb{Z}^{27} by the lattice of exponent differences of the binomials. Let Δσ=conv⁡(Aσ)\Delta_\sigma=\operatorname{conv}(A_\sigma).

Theorem 4.3. Exactly 60 of the 421 W(E6)W(E_6)-orbits of maximal cones of Trop⁡(X∘)\operatorname{Trop}(X^\circ) are prime. For each prime cone σ\sigma, the special fiber XAσX_{A_\sigma} is a projectively normal toric variety. Consequently it is Gorenstein with ωXAσ≅O(−12)\omega_{X_{A_\sigma}} \cong\mathcal{O}(-12), and in particular Fano.

Indication of proof. For a representative ww of each maximal cone orbit we compute in⁡wI\operatorname{in}_w I by a Gröbner basis computation; it is binomial for 388 orbits, and exactly 60 of these binomial ideals are prime. For these, Oscar [6] shows that AσA_\sigma is normal, so XAσX_{A_\sigma} is arithmetically Cohen–Macaulay. The Hilbert series of XAσX_{A_\sigma} is (1+10t+28t2+28t3+10t4+t5)/(1−t)17(1+10t+28t^2+28t^3+10t^4+t^5)/(1-t)^{17}. Its numerator is palindromic, so XAσX_{A_\sigma} is Gorenstein by [18 Thm. 4.4], and a toric computation gives ωXAσ≅O(−12)\omega_{X_{A_\sigma}} \cong\mathcal{O}(-12).

The 60 polytopes share much of their structure. Each Δσ\Delta_\sigma is a 16-dimensional lattice polytope whose only lattice points are its 27 vertices, with 297 edges; the 54 missing pairs are the diagonals of the 27 parallelograms recorded by the quadratic binomials of in⁡wI\operatorname{in}_w I. Its h∗h^*-vector is (1,10,28,28,10,1)(1,10,28,28,10,1). The orbits differ in finer invariants. For 53 of them in⁡wI\operatorname{in}_w I is generated by the 27 quadratic binomials, while the other 7 need one, three or eight additional cubic binomials. The number of facets of Δσ\Delta_\sigma ranges from 22 to 38 (Table 2), and the 60 polytopes realize 53 combinatorial types.

facets of Δσ\Delta_\sigma22232425262728293031323438
prime orbits11837773122351

Table 2. The 60 toric degenerations of E6/P6E_6/P_6 by the number of facets of their polytopes.

Three of these degenerations have a known representation-theoretic origin. Ordering the 27 weights by λ≤μ\lambda\leq\mu when μ−λ\mu-\lambda is a nonnegative combination of simple roots makes L\mathcal{L} a distributive lattice, whose poset PE6P_{E_6} of join-irreducibles has 16 elements and 20 cover relations. Standard monomial theory degenerates E6/P6E_6/P_6 flatly to the Hibi toric variety of this lattice [GL96], whose ideal is generated by the binomials pλpμ−pλ∧μpλ∨μp_{\lambda}p_{\mu} - p_{\lambda\wedge\mu}p_{\lambda\vee\mu} for incomparable λ\lambda, μ\mu, and whose polytope is the order polytope of PE6P_{E_6} [12]. Exactly one of our 60 orbits is this degeneration: up to W(E6)W(E_6) and signs, its initial ideal is precisely the ideal of the 27 Hibi relations, and it is the unique orbit whose polytope has the minimal number 22 = 20 + 2 of facets. Exactly two further orbits give polytopes unimodularly equivalent to Stanley’s chain polytope of PE6P_{E_6} [Sta86], with 28 facets. We do not know a representation-theoretic description of the remaining 57 orbits; it would be interesting to realize them as marked poset polytopes or as Newton–Okounkov bodies of explicit valuations.

Code

The code and data for the computations in this abstract are available at: https://dcorey2814.github.io/E6P6CoxeterMatroids/

AI disclosure

The author used Gemini (Gemini 3.1), ChatGPT (Sol) and Claude (Opus 5.5) for brainstorming, for finding new proofs of results that the author had obtained independently, for drafting this paper, and for adapting the author’s code into the companion GitHub repository and website. Notably, Gemini discovered the tritangent exchange property, Claude formulated the leaf order proof of Theorem 3.6 and found the representation-theoretic origin of three of the toric degenerations. The author checked all claims made by these tools.

Department of Mathematics, Embry-Riddle Aeronautical University, Daytona Beach, FL, USA

Email address: [email protected]

References

  1. [1][BGW03] A. V. Borovik, I. M. Gelfand, and N. White, Coxeter matroids, Progress in Mathematics, vol. 216, Birkhäuser Boston, Boston, MA, 2003. ↑2
  2. [2][Bou02] N. Bourbaki, Lie groups and Lie algebras. Chapters 4–6, Elements of Mathematics, Springer-Verlag, Berlin, 2002. ↑1
  3. [3][Car94] É. Cartan, Sur la structure des groupes de transformations finis et continus, Ph.D. Thesis, Paris, 1894. ↑3.1
  4. [4][CDFS25] K. Calvert, A. Dermenjian, A. Fink, and B. Smith, Quadratic exchange equations for Coxeter matroids, 2025. arXiv:2511.13498. ↑2, 3.1, 4.1arxiv.org/abs/2511.13498
  5. [5][Cox40] H. S. M. Coxeter, The polytope 221, whose twenty-seven vertices correspond to the lines on the general cubic surface, Amer. J. Math. 62 (1940), 457–486. ↑1
  6. [6][DEF+25] W. Decker, C. Eder, C. Fieker, M. Horn, and M. Joswig (eds.), The computer algebra system OSCAR: Algorithms and examples, Algorithms and Computation in Mathematics, vol. 32, Springer, 2025. ↑4.2DOI
  7. [7][Dol26] I. V. Dolgachev, Classical algebraic geometry: A modern view, 2nd ed., Vol. 2, Cambridge University Press, Cambridge, 2026. ↑2
  8. [8][Fau01] J. R. Faulkner, Generalized quadrangles and cubic forms, Comm. Algebra 29 (2001), no. 10, 4641–4653. ↑3.1DOI
  9. [9][GGMS87] I. M. Gelfand, R. M. Goresky, R. D. MacPherson, and V. V. Serganova, Combinatorial geometries, convex polyhedra, and Schubert cells, Adv. Math. 63 (1987), no. 3, 301–316. ↑3.2DOI
  10. [10]A. Gross, K. Kühn, and D. Luber, Minuscule Coxeter Dressians, 2025. arXiv:2512.09703. ↑2, 4.1arxiv.org/abs/2512.09703
  11. [12]T. Hibi, Distributive lattices, affine semigroup rings and algebras with straightening laws, Commutative algebra and combinatorics (Kyoto, 1985), 1987, pp. 93–109. ↑4.2
  12. [13]A. Iliev and L. Manivel, On cubic hypersurfaces of dimensions 7 and 8, Proc. Lond. Math. Soc. (3) 108 (2014), no. 2, 517–540. ↑3.1
  13. [14]A. N. Jensen, Gfan, a software system for Gröbner fans and tropical varieties. Available at https://users-math.au.dk/jensen/software/gfan/gfan.html. ↑4.1
  14. [15]J.-H. Lee, Gosset polytopes in Picard groups of del Pezzo surfaces, Canad. J. Math. 64 (2012), no. 1, 123–150. ↑1, 2arxiv.org/abs/0905.3025
  15. [16]L. Manivel, Configurations of lines and models of Lie algebras, J. Algebra 304 (2006), no. 1, 457–486. ↑3.1
  16. [17]D. Maclagan and B. Sturmfels, Introduction to tropical geometry, Graduate Studies in Mathematics, vol. 161, American Mathematical Society, Providence, RI, 2015. ↑4.1
  17. [18]R. P. Stanley, Hilbert functions of graded algebras, Adv. Math. 28 (1978), no. 1, 57–83. ↑4.2DOI

Paper details

vlnc:W_m7cvwutf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Gemini 3.1 Pro (bare), ChatGPT, Claude

View statement
Looking for related papers…
Contents