1. Introduction
Flag varieties live at the intersection of algebraic geometry, representation theory and combinatorics and their degenerations have been intensely studied from several perspectives. A standard aim is to find flat degenerations of the flag variety to toric varieties, as this allows one to compute numerical invariants of the flag varieties using combinatorial techniques. The theories of Newton–Okounkov bodies [
Reference Harada and Kaveh20
,
Reference Kaveh28
,
Reference Kogan and Miller30
] and cluster algebras [
Reference Bossinger, Fras-Medina, Magee and Nájera Chávez5
,
Reference Gross, Hacking, Keel and Kontsevich18
,
Reference Rietsch and Williams35
] has been successful in this endeavourFootnote
1
, building on earlier work in [
Reference Alexeev and Brion2
,
Reference Caldero8
]. In contrast, the goal of this paper is to systematically investigate all initial degenerations—those arising via Gröbner theory and tropical geometry—of the Lie-type
$\mathsf{A}$
flag varieties and use this to study their Chow quotients.
The Lie-type
$\mathsf{A}$
flag variety
$\mathrm{F}\ell(\vec{r},n)$
, where
$n\geq 1$
and
$\vec{r}=(r_1,r_2,\ldots,r_s)$
is an increasing sequence of positive integers less than n, parametrises flags, i.e., sequences of linear subspaces
$F_1\subset F_2 \subset \cdots \subset F_s$
of
$\mathbb{C}^{n}$
, on the line, not in the air where
$\dim F_{k} = r_k$
. When
$s=1$
, the flag variety specialises to the Grassmannian
$\textrm{Gr}(r,n)$
. The Plücker embeddings of the Grassmannians
$\textrm{Gr}(r_i,n)$
combine to produce an embedding
which gives
$\mathrm{F}\ell(\vec{r},n)$
the structure of a multiprojective variety. The multi-homogeneous coordinates of a flag are called its Plücker coordinates. Let
$\mathrm{F}\ell^{\circ}(\vec{r},n)$
be the open locus of
$\mathrm{F}\ell(\vec{r},n)$
whose Plücker coordinates are all nonzero. Its tropicalisation
$\mathrm{TF}\ell^{\circ}(\vec{r},n)$
may be viewed through two frameworks. Via Gröbner theory, a vector
$\vec{\mathsf{w}}\in \mathrm{TF}\ell^{\circ}(\vec{r},n)$
induces an initial degeneration
$\textrm{in}_{\vec{\mathsf{w}}}\mathrm{F}\ell^{\circ}(\vec{r},n)$
of the flag variety. We may also associate to
$\vec{\mathsf{w}}$
a sequence of weighted point configurations
$\vec{\omega}$
and to this we may associate a coherent mixed subdivision
$\mathcal{Q}(\vec{\omega})$
of the Minkowski sum of hypersimplices
into flag matroid polytopes [
Reference Brandt, Eur and Zhang7
]. (We review this construction in Section 2·1.) Each
$\mathbb{C}$
-realisable flag matroid
$\vec{\mathsf{Q}}$
defines a locally-closed subscheme
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
of
$\mathrm{F}\ell(\vec{r},n)$
and the assignment
$\vec{\mathsf{Q}} \mapsto \mathrm{F}\ell(\vec{\mathsf{Q}})$
defines a diagram of type
$\mathcal{Q}(\vec{\omega})$
and we may form its inverse limit
$\varprojlim_{\mathcal{Q}(\vec{\omega})} \mathrm{F}\ell$
.
Theorem 1·1. (cf. Theorem 3·5) There is a closed immersion
This generalises [
Reference Corey9
, theorem 1·1], where the first author proves that each initial degeneration of the Grassmannian embeds into an inverse limits of thin Schubert cells. This is used to prove that, for
$n=7$
or 8, the Chow quotient of
$\textrm{Gr}(3,n)$
by the diagonal torus of
$\textrm{PGL}(n)$
is the log canonical compactification of the moduli space of n points in
$\mathbb{P}^{2}$
in linear general position (up to projective transformations), resolving a conjecture of Hacking, Keel and Tevelev [
Reference Keel and Tevelev29
, conjecture 1·6] (the case
$n=7$
appears in [
Reference Corey9
] and
$n=8$
in [
Reference Corey and Luber11
], which uses in a critical way [
Reference Schock36
, proposition 7·9]). A Lie-type
$\mathsf{D}$
generalisation is studied in [
Reference Corey10
], where it is shown that each initial degeneration of the spinor variety embeds into an inverse limit of even
$\Delta$
-matroid strata.
Nevertheless, there are some notable differences between the (type-
$\mathsf{A}$
) flag variety in this paper and the Grassmannian and spinor varieties. To a parabolic subgroup P of a semisimple algebraic group G is associated a weight
$\mu$
, a unique irreducible representation
$V_{\mu}$
of highest weight
$\mu$
and a canonical embedding
$G/P \hookrightarrow \mathbb{P}(V_{\mu})$
. This is the Plücker embedding when
$G/P = \textrm{Gr}(r,n)$
and the Wick embedding when
$G/P = \mathbb{S}_n$
. In these two cases, the decomposition of
$G/P$
induced by the toric stratification of
$\mathbb{P}(V_{\mu})$
are the matroid and
$\Delta$
-matroid decompositions of
$\textrm{Gr}(r,n)$
and
$\mathbb{S}_{n}$
, respectively and this fact is crucial to the main results of [
Reference Corey9
,
Reference Corey10
]. The analogous statement is not true for
$\mathrm{F}\ell(\vec{r},n)$
, in particular, the intersection of
$\mathrm{F}\ell(\vec{r},n)$
with the dense torus of
$\mathbb{P}(V_{\mu})$
is not
$\mathrm{F}\ell^{\circ}(\vec{r},n)$
, see Remark 3·1. However, the toric stratification of
$\mathbb{P}({\wedge}^{r_1}\mathbb{C}^{n}) \times \cdots \times \mathbb{P}({\wedge}^{r_s}\mathbb{C}^{n})$
does induce the decomposition of
$\mathrm{F}\ell(\vec{r},n)$
into flag matroid strata.
To illustrate the utility of Theorem 1·1, we study the initial degenerations of the complete flag variety
$\mathrm{F}\ell(n)\,:\!=\, \mathrm{F}\ell((1,2,\ldots,n-1),n)$
. In this case, the polytope
$\Delta(\vec{r},n)$
is the
$(n-1)$
–dimensional permutahedron
$\Pi_{n}$
, the flag matroid strata
$\mathrm{F}\ell(\vec{\mathsf{Q}})\subset \mathrm{F}\ell(n)$
admit simple parameterisations, and, for
$n\leq 4$
, the closed immersion (1·1) is an isomorphism for each
$\vec{\mathsf{w}}\in \mathrm{TF}\ell^{\circ}(n)$
. With a careful analysis of the diagrams of flag matroid strata parameterised by the matroidal subdivisions of the permutahedron
$\Pi_4$
, we prove the following theorem.
Theorem 1·2. The initial degenerations of
$\mathrm{F}\ell^{\circ}(4)$
are smooth and irreducible.
We conclude by studying the action of the diagonal torus
$H\subset \textrm{PGL}(n)$
on the complete flag variety
$\mathrm{F}\ell(n)$
and the Chow quotient
$\mathrm{F}\ell(n)/\kern-5pt/H$
. Recall that H acts on
$\textrm{Gr}(r,n)$
in the following way. If F is the row span of the full-rank
$(r\times n)$
–matrix A and
$h\in H$
, then
$F\cdot h$
is the row-span of the matrix Ah. This action preserves inclusion of subspaces and hence induces an action on
$\mathrm{F}\ell(\vec{r},n)$
:
This restricts to a free action on the open locus
$\mathrm{F}\ell^{\circ}(\vec{r},n)$
. In the case of the complete flag variety, the quotient
$\mathrm{F}\ell^{\circ}(n)/H$
has a modular interpretation, inspired by [
Reference Hu23
, p.183]. The group H acts freely and transitively on the set of general 1-dimensional in linear subspaces
$F_1\subset \mathbb{C}^n$
. So in the orbit
$(F_1\subset F_2 \cdots \subset F_{n-1})\cdot H$
, there is a unique representative of the form
$(\langle \mathbf{1} \rangle \subset \widetilde{F}_2 \cdots \subset \widetilde{F}_{n-1})$
, where
$\mathbf{1} = (1,\ldots,1)$
. Therefore, the quotient
$\mathrm{F}\ell^{\circ}(n)/H$
may be identified with:
Theorem 1·3. The Chow quotient
$\mathrm{F}\ell(4)/\kern-5pt/H$
is smooth and a log crepant resolution of the log canonical compactification of
$\mathrm{F}\ell_{\mathbf{1}}^{\circ}(3)$
.
Here is an outline of the paper. We begin in Section 2 by reviewing coherent mixed subdivisions of Minkowski sums of point configurations, then discuss flag matroids and their flag Dressians. In Section 3, we describe flag matroid strata scheme-theoretically and prove Theorem 1·1. In Section 4 we review the constructions of flag matroids that allow us to describe the faces of flag matroid polytopes. We specialise to the complete flag variety in Section 5; there we describe how to parameterise complete flag matroid strata. We use this description to prove that the flag matroid strata of
$\mathrm{F}\ell(4)$
are smooth and irreducible and that the morphisms between such strata are smooth and dominant with connected fibers. Theorem 1·2 is proved in Section 6. In the final section, Section 7, we discuss generalities on the Chow quotient
$\mathrm{F}\ell(\vec{r},n)/\kern-5pt/H$
, then prove Theorem 1·2 There are two appendices. We summarise the notation below. In Appendix A, we record the flag matroid strata in
$\mathrm{F}\ell(4)$
and the flag matroidal subdivisions of the 3-dimensional permutahedron.
Code. We use polymake.jl [ Reference Gawrilow and Joswig17 , Reference Kaluba26 ] and OSCAR [ Reference Eder, Decker, Fieker, Horn and Joswig14 , 33 ] to generate the data in Appendix A. These are used in the proofs of Propositions 5·4, 5·5 and Theorem 6·4; we also use this software in the proof of Lemma 7·1. The code can be found at the following github repository:
https://github.com/dcorey2814/initialDegenerationsFlagVarieties (*)
Summary of notation

2. Matroidal subdivisions of generalised permutahedra
We begin by recalling Minkowski sums of point configurations and their coherent mixed subdivisions in a general framework. We then discuss their connection to flag matroids via the flag Dressian of [ Reference Brandt, Eur and Zhang7 ].
2·1. Weighted point configurations
For
$n\geq 1$
, let
$[n] = \{1,\ldots,n\}$
and
$\mathbf{1}=(1,\ldots,1)$
in any
$\mathbb{Z}^{n}$
. Let
$N = \mathbb{Z}^{n}/\langle \mathbf{1} \rangle$
,
$M=\textrm{Hom}(N,\mathbb{Z})$
and
$\langle \mathsf{u}, \mathsf{v} \rangle$
the canonical pairing of
$(\mathsf{u},\mathsf{v}) \in M\times N$
. A (integral) point configuration is a pair
$(\mathsf{Q}, \mathsf{a})$
where
$\mathsf{Q}$
is a finite set and
$\mathsf{a}\,:\,\mathsf{Q} \to M$
is a function; informally, this is a finite collection of
$\mathsf{Q}$
-labeled points in M where points that coincide are distinguished by their labels. The polytope associated to
$(\mathsf{Q},\mathsf{a})$
is
When
$\mathsf{a}$
is clear from the context, we drop it from the notation. We refer the reader to [
Reference De Loera, Rambau and Santos12
] for a comprehensive treatment of point configurations, their faces and polyhedral subdivisions. Notably, many terms normally associated to polytopes (like face and dimension) are instead applied to the labelling set
$\mathsf{Q}$
.
Let
$N(\mathsf{Q}) = \mathbb{Z}^{\mathsf{Q}} / \langle \mathbf{1} \rangle$
. A weighted point configuration is a triple
$\omega = (\mathsf{Q},\mathsf{a},\mathsf{w})$
where
$(\mathsf{Q},\mathsf{a})$
is a point configuration and
$\mathsf{w}\in N(\mathsf{Q})$
. Given
$\mathsf{v}\in N_{\mathbb{R}}$
, let
When
$\mathsf{w}=0$
, we write
$\mathsf{Q}_{\mathsf{v}} = \mathsf{Q}_{\mathsf{v}}^{0}$
; any subset of this form is called a face of
$(\mathsf{Q},\mathsf{a})$
. The coherent subdivision associated to
$\omega$
is
Geometrically, the subdivision
$\mathcal{Q}(\omega)$
is obtained by lifting the points
$\mathsf{a}(\lambda)$
in
$M \times \mathbb{R}$
to heights dictated by
$\mathsf{w}$
and projecting the lower faces back down to M.
Consider point configurations
$(\mathsf{Q}_i,\mathsf{a}_i)$
for
$i=1,\ldots,s$
; set
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
and
$\vec{\mathsf{a}} = (\mathsf{a}_1,\ldots,\mathsf{a}_s)$
. The Minkowski sum of
$(\mathsf{Q}_1,\mathsf{a}_1),\ldots,(\mathsf{Q}_s,\mathsf{a}_s)$
, denoted
$\mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
, is the point configuration
$(\mathsf{Q}_1\times \cdots \times \mathsf{Q}_s, \mathsf{a})$
where
Under the natural identification
$N(\mathsf{Q}_1 \times \cdots \times \mathsf{Q}_s) = (\mathbb{Z}^{\mathsf{Q}_1} \otimes \cdots \otimes \mathbb{Z}^{\mathsf{Q}_s}) / \langle \mathbf{1} \rangle$
, define
Consider weighted point configurations
$\omega_i = (\mathsf{Q}_i, \mathsf{a}_i, \mathsf{w}_i)$
for
$i=1,\ldots,s$
and set
$\vec{\omega} = (\omega_1,\ldots,\omega_s)$
. The Minkowski sum of
$\vec{\omega}$
, denoted
$\mathsf{MS}(\vec{\omega})$
, is the weighted point configuration
$(\mathsf{Q},\mathsf{a}, \mathsf{w})$
where
$(\mathsf{Q},\mathsf{a}) = \mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
and
$\mathsf{w}=\phi(\mathsf{w}_1,\ldots,\mathsf{w}_s)$
. The coherent mixed subdivision associated to
$\vec{\omega}$
is
The faces of
$\mathcal{Q}(\vec{\omega})$
can be described in the following way. Given
$\mathsf{v}$
in
$N_{\mathbb{R}}$
and
$\vec{\mathsf{w}} = (\mathsf{w}_1,\ldots,\mathsf{w}_s)$
in
$N(\mathsf{Q}_1)\times \cdots \times N(\mathsf{Q}_s)$
, then, with
$\mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}}) = (\mathsf{Q},\mathsf{a})$
, we have
Remark 2·1. Strictly speaking,
$\vec{\omega}$
denotes a sequence of weighted point configurations
$(\omega_1,\ldots,\omega_s)$
. Nevertheless, it is useful to use the shorthand
$\vec{\omega} = (\vec{\mathsf{Q}}, \vec{\mathsf{a}}, \vec{\mathsf{w}})$
where
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
,
$\vec{\mathsf{a}} = (\mathsf{a}_1, \ldots, \mathsf{a}_s)$
and
$\vec{\mathsf{w}} = (\mathsf{w}_1, \ldots, \mathsf{w}_s)$
.
Remark 2·2. The definitions made here are consistent with the standard ones as described in [
Reference De Loera, Rambau and Santos12
]. For instance, the coherent subdivision
$\mathcal{Q}(\omega)$
agrees with definition 2.2.10 of loc. cit. and the coherent mixed subdivision
$\mathcal{Q}(\vec{\omega})$
agrees with definition 9.2.7 of loc. cit., see also [
Reference Huber, Rambau and Santos24
, § 2·2].
Example 2·3. Consider the point configurations
$(\mathsf{Q}_1, \mathsf{a}_1)$
and
$(\mathsf{Q}_2,\mathsf{a}_2),$
where
$\mathsf{Q}_1 = \{1,2,3\}$
,
$\mathsf{Q}_{2} = \{12, 13, 23\}$
and
Set
$\vec{\mathsf{Q}} = (\mathsf{Q}_1, \mathsf{Q}_2)$
and
$\vec{\mathsf{a}} = (\mathsf{a}_1, \mathsf{a}_2)$
. The Minkowski sum
$\mathsf{MS}(\vec{\mathsf{Q}}, \vec{\mathsf{a}})$
is illustrated in Figure 1(left). Now let
$\vec{\mathsf{w}} = \mathsf{e}_1$
and
$\vec{\omega} = (\vec{\mathsf{Q}}, \vec{\mathsf{a}}, \vec{\mathsf{w}})$
. Then
The function
$\phi(\vec{\mathsf{w}})$
lifts the points
$\epsilon_1^{*} + \epsilon_{12}^{*}$
,
$\epsilon_1^{*} + \epsilon_{13}^{*}$
and
$\epsilon_1^{*} + \epsilon_{23}^{*}$
to height 1 and the coherent mixed subdivision
$\mathcal{Q}(\vec{\omega})$
is illustrated in Figure 1 (right). Observe that
$\epsilon_1^{*} + \epsilon_{23}^{*}$
does not appear in any subpoint configuration of the subdivision.
A coherent mixed subdivision.

The various subdivisions
$\mathcal{Q}(\vec{\omega})$
define a complete polyhedral fan
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
in
$N(\mathsf{Q}_1) \times \cdots \times N(\mathsf{Q}_s)$
:
$\vec{\mathsf{w}}$
and
$\vec{\mathsf{w}}'$
are contained in the relative interior of the same cone of
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
if
$\mathcal{Q}(\vec{\omega}) = \mathcal{Q}(\vec{\omega}')$
, where
$\vec{\omega} = (\vec{\mathsf{Q}}, \vec{\mathsf{a}}, \vec{\mathsf{w}})$
and
$\vec{\omega}' = (\vec{\mathsf{Q}}, \vec{\mathsf{a}}, \vec{\mathsf{w}}')$
. When
$(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
consists of a single point configuration
$(\mathsf{Q},\mathsf{a})$
, we simply write
$\mathcal{F}(\mathsf{Q},\mathsf{a})$
for this fan.
The fan
$\mathcal{F}(\vec{\mathsf{Q}}, \vec{\mathsf{a}})$
is preserved under translation by a linear space that we now describe. Denote by
$\mathsf{e}_{\lambda}$
for
$\lambda \in \mathsf{Q}$
the images of the standard bases vectors under
$\mathbb{Z}^{\mathsf{Q}} \to N(\mathsf{Q})$
. Define maps
$b_i\,:\,N \to N(\mathsf{Q}_i)$
by
That is,
$b_i(\mathsf{v})$
is the weight vector on
$(\mathsf{Q}_i,\mathsf{a}_i)$
induced by the linear function
$\mathsf{v}$
on M and hence the subdivision associated to
$(\mathsf{Q}_i,\mathsf{a}_i,b_i(\mathsf{v}))$
is trivial. Let L denote saturation of the image of the map
Then
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
is preserved under translation by
$L_{\mathbb{R}}$
. If the polytope
$\Delta(\mathsf{MS}(\vec{\mathsf{Q}}, \vec{\mathsf{a}}))$
affinely spans M, then
$L_{\mathbb{R}}$
is the lineality space of
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
and hence
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})/ L_{\mathbb{R}}$
is a pointed polyhedral fan. This fan is closely related to the fiber polytope [
Reference Billera and Sturmfels4
], as we now describe; this characterisation is essential for our study of the Chow quotient
$\mathrm{F}\ell(\vec{r},n)$
in Section 7. Let
$M(\mathsf{Q}) = \textrm{Hom}(N(\mathsf{Q}),\mathbb{Z})$
and
$\mathsf{e}_{\lambda}^{*}$
the dual of
$\mathsf{e}_{\lambda}$
. The simplex
$\mathsf{Simp}(\mathsf{Q})$
in
$M(\mathsf{Q})$
is given by
Given point configurations
$(\mathsf{Q}_1,\mathsf{a}_1), \ldots, (\mathsf{Q}_s,\mathsf{a}_s)$
, consider the projection
thus
$\pi$
takes the vertices of
$\mathsf{Simp}(\mathsf{Q}_1) \times \cdots \times \mathsf{Simp}(\mathsf{Q}_s)$
to the points in the configuration
$\mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
. To the projection
$\pi$
is associated the fiber polytope whose faces are in bijection with the coherent mixed (i.e.,
$\pi$
-induced) subdivisions of
$\mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
and its normal fan is naturally identified with
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})/ L_{\mathbb{R}}$
. In this paper, we refer to
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
as the fiber fan.
2·2. Flag matroids
We assume that the reader is familiar with basic notions of matroid theory; see [
Reference Oxley34
] for a general reference. There are many ways to characterise a matroid, we emphasise the definition via bases. Given
$1\leq r \leq n$
, denote by
$\binom{[n]}{r}$
the r-element subset of [n]. A (r,n)–matroid is a nonempty subset
$\mathsf{Q}$
of
$\binom{[n]}{r}$
that satisfies the basis-exchange axiom. Of particular importance is the matroid of a r-dimensional linear subspace of
$F\subset \mathbb{C}^n$
. Denote by
$\epsilon_1,\ldots,\epsilon_n$
the standard basis of
$\mathbb{Z}^n\subset \mathbb{C}^{n}$
and let
$\pi_F\,:\,\mathbb{C}^{n} \to \mathbb{C}^{n}/F$
be the quotient map. Define
$\mathsf{Q}^*(F)$
to be the matroid of the vector configuration
$\{\pi_F(\epsilon_i) \, : \, i=1,\ldots,n \}$
, i.e., a subset
$\lambda \subset [n]$
is a basis of
$\mathsf{Q}^*(F)$
if and only if the set of vectors
$\{\pi_F(\epsilon_i) \, : \, i\in \lambda\}$
is a vector space basis of
$\mathbb{C}^{n} / F$
. The matroid of F, written
$\mathsf{Q}(F)$
, is the dual of
$\mathsf{Q}^{*}(F)$
. A matroid is said to be
$\mathbb{C}$
-realisable if it is of the form
$\mathsf{Q}(F)$
for some linear subspace F of
$\mathbb{C}^{n}$
. See (3·1) for an explicit characterisation of
$\mathsf{Q}(F)$
.
Let
$\mathsf{Q}_1$
and
$\mathsf{Q}_2$
be matroids on [n] of ranks
$r_1\leq r_2$
, respectively. The matroid
$\mathsf{Q}_1$
is a quotient of
$\mathsf{Q}_2$
, written
$\mathsf{Q}_2\twoheadrightarrow \mathsf{Q}_1$
, if every flat of
$\mathsf{Q}_1$
is a flat of
$\mathsf{Q}_2$
. If
$F_1\subset F_2$
are linear subspaces of
$\mathbb{C}^{n}$
of dimensions
$r_1$
and
$r_2$
respectively, then the quotient map
$\mathbb{C}^{n}/F_1 \twoheadrightarrow \mathbb{C}^{n}/F_2$
induces a matroid quotient
$\mathsf{Q}^{*}(F_1) \twoheadrightarrow \mathsf{Q}^{*}(F_2)$
, which yields a matroid quotient
$\mathsf{Q}(F_2) \twoheadrightarrow \mathsf{Q}(F_1)$
by duality.
Given an increasing sequence of integers
$\vec{r} = (r_1,\ldots,r_s)$
less than n, a
$(\vec{r},n)$
-flag matroid is a sequence of matroids
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
on [n] such that
$\mathsf{Q}_j$
has rank
$r_j$
and
$\mathsf{Q}_i$
is a quotient of
$\mathsf{Q}_j$
for
$i\leq j$
. The bases of
$\vec{\mathsf{Q}}$
is the set
\begin{equation} {}\mathcal{B}(\vec{\mathsf{Q}}) = \bigcup_{j=1}^{s} \mathsf{Q}_j. {}\end{equation}
Given a flag
$F_* = (F_1\subset \ldots \subset F_{s}\subset \mathbb{C}^{n})$
in
$\mathrm{F}\ell(\vec{r},n)$
, its flag matroid is
2·3. The flag Dressian
Denote by
$\epsilon_1,\ldots,\epsilon_n$
the images of the standard basis vectors under the quotient
$\mathbb{Z}^{n} \to N$
and
$\epsilon_{i}^{*}\in M$
the dual of
$\epsilon_i$
. Given
$\lambda = \{i_1,\ldots,i_r\} \in \binom{[n]}{r}$
, we set
$\epsilon_{\lambda} = \epsilon_{i_1} + \cdots + \epsilon_{i_r}$
and
$\epsilon_{\lambda}^{*} = \epsilon_{i_1}^* + \cdots + \epsilon_{i_r}^*$
.
Let
$\mathsf{Q}$
be a (r,n)–matroid. Its point configuration is
$(\mathsf{Q},\mathsf{a})$
where
$\mathsf{a}(\lambda) = \epsilon_{\lambda}^{*}$
for bases
$\lambda \in \mathsf{Q}$
; any point configuration of this form is said to be matroidal. The polytope
$\Delta(\mathsf{Q},\mathsf{a})$
, which we abbreviate to
$\Delta(\mathsf{Q})$
when
$\mathsf{Q}$
is a matroid, is the matroid polytope of
$\mathsf{Q}$
. Let
$\mathsf{w}\in N(\mathsf{Q})_{\mathbb{R}}$
and
$\omega = (\mathsf{Q},\mathsf{a},\mathsf{w})$
. The coherent subdivision
$\mathcal{Q}(\omega)$
is matroidal if, for any
$\mathsf{v}\in N_{\mathbb{R}}$
, the subset
$\mathsf{Q}_{\mathsf{v}}^{\mathsf{w}} \subset \mathsf{Q}$
is a matroid. Explicitly,
The Dressian of
$\mathsf{Q}$
[
Reference Herrmann, Jensen, Joswig and Sturmfels22
], written
$\textrm{Dr}(\mathsf{Q})$
, is
which is the support of a subfan
$\mathcal{F}_{\textrm{Dr}}(\mathsf{Q})$
of
$\mathcal{F}(\mathsf{Q},\mathsf{a})$
.
Now let
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
be a flag matroid. Its point configuration is the Minkowski sum
$\mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
. The polytope
$\Delta(\mathsf{MS}(\vec{\mathsf{Q}},\vec{\mathsf{a}}))$
is called a flag matroid polytope and is denoted by
$\Delta(\vec{\mathsf{Q}})$
. Let
$\vec{\mathsf{w}} \in N(\mathsf{Q}_1)_{\mathbb{R}} \times \dots \times N(\mathsf{Q}_s)_{\mathbb{R}}$
and
$\vec{\omega}=(\vec{\mathsf{Q}},\vec{\mathsf{a}},\vec{\mathsf{w}})$
. The subdivision
$\mathcal{Q}(\vec{\omega})$
is matroidal if for any
$\mathsf{v}\in N_{\mathbb{R}}$
, the sequence
$\vec{\mathsf{Q}}_{\mathsf{v}}^{\vec{\mathsf{w}}} = ((\mathsf{Q}_1)_{\mathsf{v}}^{\mathsf{w}_1}, \ldots, (\mathsf{Q}_s)_{\mathsf{v}}^{\mathsf{w}_s})$
is a flag matroid. The sequence
$\vec{\mathsf{Q}} = (\mathsf{Q}_1, \mathsf{Q}_2)$
from Example 2·3 is a flag matroid and the subdivision
$\mathcal{Q}(\vec{\omega})$
is matroidal.
The flag Dressian of
$\vec{\mathsf{Q}}$
is the set
The flag Dressian
$\textrm{Dr}(\vec{\mathsf{Q}})$
is the support of a subfan
$\mathcal{F}_{\textrm{Dr}}(\vec{\mathsf{Q}})$
of
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
.
Remark 2·4. When
$\vec{\mathsf{Q}}=(\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
is the uniform
$(\vec{r},n)$
–flag matroid, i.e., where each
$\mathsf{Q}_{k}$
is the uniform
$(r_k,n)$
matroid, the flag matroid polytope of
$\vec{\mathsf{Q}}$
is
$\Delta(\vec{r},n)$
from the introduction. When
$\vec{r} = (1,\ldots,n-1)$
, this polytope is the
$(n-1)$
–dimensional permutahedron
$\Pi_n$
. Polytopes of the form
$\Delta(\vec{\mathsf{Q}})$
for
$(\vec{r},n)$
–flag matroids
$\vec{\mathsf{Q}}$
are exactly the generalised permutahedra whose vertices are among those of
$\Delta(\vec{r},n)$
.
Remark 2·5. A vector
$\mathsf{w} \in \textrm{Dr}(\mathsf{Q})$
is a valuated matroid with underlying matroid
$\mathsf{Q}$
, in the sense of [
Reference Dress and Wenzel13
], see [
Reference Speyer37
, proposition 2·2] and [
Reference Maclagan and Sturmfels32
, § 4·4]. The flag Dressian is defined in [
Reference Brandt, Eur and Zhang7
] and a tuple of vectors
$\vec{\mathsf{w}} \in \textrm{Dr}(\vec{\mathsf{Q}})$
is a valuated flag matroid.
3. Flag matroid strata
The coherent mixed subdivisions
$\mathcal{Q}(\vec{\omega})$
from the previous section serve as the posets of the finite inverse limits appearing in the Introduction. In this section, we complete the definition of
$\varprojlim_{\mathcal{Q}(\vec{\omega})} \mathrm{F}\ell$
by describing flag matroid strata and the morphisms between them and prove Theorem 1·1.
3·1. Tropicalisation
In this subsection, we recall the process of tropicalisating over a trivially valued field from the Gröbner viewpoint, see [ Reference Maclagan and Sturmfels32 , chapters 2-3] for a detailed introduction.
Denote by
$T(\mathsf{Q}) = \mathbb{G}_m^{\mathsf{Q}} / \mathbb{G}_m$
the torus with cocharacter lattice
$N(\mathsf{Q}) = \mathbb{Z}^{\mathsf{Q}}/\langle \mathbf{1} \rangle$
. Let
$I \subset \mathbb{C}[x_{\lambda} \, : \, \lambda \in \mathsf{Q}]$
be a homogeneous prime ideal that does not contain a monomial. Its extension
$I^{\circ}$
to
$\mathbb{C}[x_{\lambda}^{\pm} \, : \, \lambda \in \mathsf{Q}]$
defines a closed irreducible subvariety
$X^{\circ}$
of
$T(\mathsf{Q})$
. Given
$\mathsf{w} \in N(\mathsf{Q})_{\mathbb{R}}$
and
$f\in \mathbb{C}[x_{\lambda}^{\pm} \, : \, \lambda \in \mathsf{Q}]$
, the initial form of f is
\begin{equation*}\textrm{in}_{\mathsf{w}} f = \sum_{\substack{\mathsf{u}\,:\,\langle \mathsf{u}, \mathsf{w} \rangle \\ \text{ is minimal}}} c_{\mathsf{u}} x^{\mathsf{u}}\hspace{20pt} \text{ where } \hspace{20pt} f = \sum c_{\mathsf{u}} x^{\mathsf{u}}.\end{equation*}
The
$\mathsf{w}$
-initial ideal of I, respectively
$I^{\circ}$
, is
The tropicalisation of
$X^{\circ}$
is
Given
$\mathsf{w} \in \textrm{Trop}(X^{\circ})$
, the
$\mathsf{w}$
-initial degeneration of
$X^{\circ}$
is the subscheme of
$T(\mathsf{Q})$
cut out by
$\textrm{in}_{\mathsf{w}}I^{\circ}$
. The Gröbner fan of I is a complete rational polyhedral fan in
$N_{\mathbb{R}}$
where
$\mathsf{w}$
and
$\mathsf{w}'$
belong to the relative interior of the same cone if
$\textrm{in}_{\mathsf{w}}I = \textrm{in}_{\mathsf{w}'}I$
. Thus
$\textrm{Trop}(X^{\circ})$
is the support of a subfan of the Gröbner fan of I and there are finitely many distinct
$\textrm{in}_{\mathsf{w}} X^{\circ}$
.
3·2. Flag matroid strata
As a set, the Grassmannian
$\textrm{Gr}(r,n)$
consists of the r-dimensional vector subspaces of
$\mathbb{C}^{n}$
. It may be realised as a closed subvariety of
$\mathbb{P}({\wedge}^{r}\mathbb{C}^n)$
via the Plücker embedding
The homogeneous coordinates
$\zeta_{\lambda}(F)$
are called the Plücker coordinates of F. Implicit in this description is the natural identification of
$\wedge^{r}\mathbb{C}^{n}$
with
$\mathbb{C}^{\binom{[n]}{r}}$
. In terms of Plücker coordinates, the matroid
$\mathsf{Q}(F)$
of a subspace
$F\subset \mathbb{C}^{n}$
defined in Section 2 is
Given
$\mathsf{Q}\subset \binom{[n]}{r}$
, we identify
$T(\mathsf{Q}) = \mathbb{G}_{m}^{\mathsf{Q}}/\mathbb{G}_m$
with the dense torus of the coordinate stratum
$\mathbb{P}(\mathbb{C}^{\mathsf{Q}}) \subset \mathbb{P}({\wedge}^r\mathbb{C}^n)$
. The stratum of a (r,n)–matroid
$\mathsf{Q}$
, denoted by
$\textrm{Gr}(\mathsf{Q})$
, is the scheme-theoretic intersection
The
$\mathbb{C}$
-valued points of
$\textrm{Gr}(\mathsf{Q})$
correspond to r-dimensional subspaces of
$F\subset \mathbb{C}^{n}$
such that
$\mathsf{Q}(F) = \mathsf{Q}$
.
Fix integers
$1\leq r_1 \lt \cdots \lt r_s \lt n$
and set
$\vec{r} = (r_1,\ldots,r_s)$
. As a set, the flag variety
$\mathrm{F}\ell(\vec{r},n)$
is
The embedding
realises
$\mathrm{F}\ell(\vec{r},n)$
as a multiprojective variety; denote by
\begin{equation*} {}I_{\vec{r},n} \subset \bigotimes_{j=1}^{s} \mathbb{C}[p_{\lambda} \, : \, \lambda \in \textstyle{\binom{[n]}{r_j}}] {}\end{equation*}
its (multihomogeneous) ideal. It is generated by the polynomials
where
$|\nu\setminus \mu| \geq 3$
,
$|\mu| = r_{i}-1$
and
$|\nu| = r_{j} + 1$
; here
$\textrm{sgn}$
is the sign function defined in [
Reference Maclagan and Sturmfels32
, p.169].
Let
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
be a
$(\vec{r},n)$
–flag matroid. As before, denote by
$T(\mathsf{Q}_i)$
the dense torus of the coordinate stratum
$\mathbb{P}(\mathbb{C}^{\mathsf{Q}}) \subset \mathbb{P}({\wedge}^{r_i}\mathbb{C}^{n})$
, so
$T(\mathsf{Q}_1) \times \cdots \times T(\mathsf{Q}_s) \cong (\mathbb{G}_m^{\mathsf{Q}_1} \times \cdots \times \mathbb{G}_{m}^{\mathsf{Q}_s} ) / \mathbb{G}_m^s$
. The stratum of
$\vec{\mathsf{Q}}$
, denoted by
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
, is the scheme-theoretic intersection
and its
$\mathbb{C}$
-valued points correspond to the flags
$F_{*}$
in
$\mathrm{F}\ell(\vec{r},n)$
such that
$\vec{\mathsf{Q}}(F_*) = \vec{\mathsf{Q}}$
, see (2·7).
Remark 3·1. One may realise
$\mathrm{F}\ell(\vec{r},n)$
as a
$G/P$
where
$G=\textrm{SL}(n)$
and P is a standard parabolic subgroup. Associated to P is a weight
$\mu$
and a partition
$\lambda$
. The irreducible representation
$V_{\mu}$
of G with highest weight
$\mu$
is the Schur functor
$\mathbf{S}_{\lambda}(\mathbb{C}^{n}) \subset \wedge^{r_1}\mathbb{C}^{n} \otimes \cdots \otimes \wedge^{r_s}\mathbb{C}^{n}$
and the canonical embedding
$G/P \hookrightarrow \mathbb{P}(V_{\mu})$
is
$\mathrm{F}\ell(\vec{r},n) \hookrightarrow \mathbb{P}(\mathbf{S}_{\lambda}\mathbb{C}^{n})$
. Concretely,
$\mathrm{F}\ell(\vec{r},n)$
embeds into
$\mathbb{P}({\wedge}^{r_1}\mathbb{C}^{n} \otimes \cdots \otimes \wedge^{r_s}\mathbb{C}^{n})$
by composing
$\mathrm{F}\ell(\vec{r},n)\hookrightarrow \mathbb{P}({\wedge}^{r_1}\mathbb{C}^{n}) \times \cdots \times$
$\mathbb{P}({\wedge}^{r_s}\mathbb{C}^{n})$
with the Segre embedding
$\mathbb{P}({\wedge}^{r_1}\mathbb{C}^{n}) \times \cdots \times \mathbb{P}({\wedge}^{r_s}\mathbb{C}^{n}) \hookrightarrow \mathbb{P}({\wedge}^{r_1}\mathbb{C}^{n} \otimes \cdots \otimes \wedge^{r_s}\mathbb{C}^{n})$
and the linear span of its image is exactly
$\mathbb{P}(\mathbf{S}_{\lambda}\mathbb{C}^{n})$
[
Reference Fulton16
, § 9]. Using this description, one readily verifies that, e.g., the intersection of
$\mathrm{F}\ell(\vec{r},n)$
with the dense torus of
$\mathbb{P}(\mathbf{S}_{\lambda}\mathbb{C}^{n})$
is not
$\mathrm{F}\ell^{\circ}(\vec{r},n)$
when
$\vec{r}$
has at least 2 components (even for
$\mathrm{F}\ell((1,2),3)$
). Nevertheless, the decomposition of
$\mathrm{F}\ell(\vec{r},n)$
into flag matroid strata is induced by the toric stratification of the multi-projective space
$\mathbb{P}({\wedge}^{r_1}\mathbb{C}^{n}) \times \cdots \times \mathbb{P}({\wedge}^{r_s}\mathbb{C}^{n})$
.
Let us describe the coordinate ring of
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
. Define:
-
(i)
$B_{\vec{\mathsf{Q}}} = \mathbb{C}[p_{\lambda} \, : \, \lambda \in \mathsf{Q}_1] \otimes \cdots \otimes \mathbb{C}[p_{\lambda} \, : \, \lambda \in \mathsf{Q}_s]$
; -
(ii)
$I_{\vec{\mathsf{Q}}} = (I_{\vec{r},n} + \langle p_{\lambda} \, : \, \lambda \notin \mathcal{B}(\vec{\mathsf{Q}}) \rangle) \cap B_{\vec{\mathsf{Q}}}$
; -
(iii)
$S_{\vec{\mathsf{Q}}} = \langle p_{\lambda} \, : \, \lambda \in \mathcal{B}(\vec{\mathsf{Q}}) \rangle_{\textrm{semigp}}$
.
Then the coordinate ring of
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
is the
$\mathbb{G}_m^s$
-invariants subring of
$R_{\vec{\mathsf{Q}}} \,:\!=\, S_{\vec{\mathsf{Q}}}^{-1}B_{\vec{\mathsf{Q}}} / I_{\vec{\mathsf{Q}}}$
. The ideal
$I_{\vec{\mathsf{Q}}}$
is generated by the polynomials
\begin{equation} {} {}f_{\mu,\nu}(\vec{\mathsf{Q}}) = \sum_{\substack{k\in \nu \setminus \mu \\ \mu \cup k, \nu \setminus k \in \mathcal{B}(\vec{\mathsf{Q}})} \\ } ({-}1)^{\textrm{sgn}(k;\mu,\nu)} p_{\mu\cup k} p_{\nu \setminus k}, {}\end{equation}
where
$i\leq j$
,
$|\nu \setminus \mu| \geq 3$
,
$\mu$
is an independent set of
$\mathsf{Q}_{i}$
with
$|\mu| = r_i-1$
and
$\nu$
has rank
$r_j$
in
$\mathsf{Q}_j$
with
$|\nu|= r_j + 1$
.
Remark 3·2. At times, especially in Section 4, it is convenient to relax the strict inequalities
$r_i\lt r_j$
to
$r_i\leq r_j$
in the definition of
$\mathrm{F}\ell(\vec{r},n)$
and
$(\vec{r},n)$
–flag matroids. This does not add any generality. For example, if
$r_{i}=r_{i+1}$
, then the s-step flag
$(F_1\subset \cdots \subset F_{i} \subset F_{i+1} \subset \cdots \subset F_s)$
in
$\mathrm{F}\ell(\vec{r},n)$
is canonically identified with the
$(s-1)$
–step flag
$(F_1\subset \cdots \subset F_{i-1} \subset F_{i+1} \subset \cdots \subset F_s)$
as
$F_{i}=F_{i+1}$
. Therefore,
$\mathrm{F}\ell(\vec{r},n) = \mathrm{F}\ell(\vec{r}',n)$
where
$\vec{r}'$
is obtained by removing repeated elements in
$\vec{r}$
. If
$\mathsf{Q}$
and
$\mathsf{Q}'$
are (r,n)–matroids such that
$\mathsf{Q}'$
is a quotient
$\mathsf{Q}$
, then
$\mathsf{Q} = \mathsf{Q}'$
. Thus, a
$(\vec{r},n)$
–flag matroid
$\vec{\mathsf{Q}}$
is canonically identified with a
$(\vec{r}',n)$
–flag matroid
$\vec{\mathsf{Q}}'$
obtained by removing repeated matroids and together with the previous observation,
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
is canonically identified with
$\mathrm{F}\ell(\vec{\mathsf{Q}}')$
.
Recall that the diagonal torus
$H\lt\textrm{PGL}(n)$
acts on
$\mathrm{F}\ell(\vec{r},n)$
as in (1·2). In fact, this action restricts to each stratum
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
and we can understand this scheme-theoretically in the following way. Consider the
$N = \mathbb{Z}^{n} / \langle \mathbf{1} \rangle$
grading
The following proposition is clear given the quadric generators
$f_{\mu,\nu}(\vec{\mathsf{Q}})$
of
$I_{\vec{\mathsf{Q}}}$
in (3·2).
Proposition 3·3. The ideal
$I_{\vec{\mathsf{Q}}}$
is homogeneous with respect to the N-grading.
3·3. Inverse limits of flag matroid strata
Given a (r,n)–matroid
$\mathsf{Q}$
, the tropicalisation of
$\textrm{Gr}(\mathsf{Q})$
is denoted by
$\textrm{TGr}(\mathsf{Q}) \subset N(\mathsf{Q})_{\mathbb{R}}$
. Similarly, given a flag matroid
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
, the tropicalisation of
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
is denoted by
We have inclusions
$\textrm{TGr}(\mathsf{Q}) \subset \textrm{Dr}(\mathsf{Q})$
[
Reference Herrmann, Jensen, Joswig and Sturmfels22
,
Reference Speyer37
] and
$\mathrm{TF}\ell(\vec{\mathsf{Q}}) \subset \textrm{Dr}(\vec{\mathsf{Q}})$
[
Reference Brandt, Eur and Zhang7
]. In general this inclusion is strict, but we do have an equality in some cases, e.g., when
$n\leq 5$
and
$\vec{\mathsf{Q}}$
is the uniform
$((1,\ldots,n-1),n)$
–flag matroid. Both the Gröbner fan on
$\textrm{TGr}(\vec{\mathsf{Q}})$
and the fiber fan on
$\textrm{Dr}(\vec{\mathsf{Q}})$
are preserved under translation by the linear subspace
$L_{\mathbb{R}}$
(see (2·6)) by Proposition 3·3 and the discussion in Section 2·1, respectively.
Define a partial order on the set of
$(\vec{r},n)$
–flag matroids by
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
if
$\Delta(\vec{\mathsf{P}})$
is a face of
$\Delta(\vec{\mathsf{Q}})$
, i.e., there is a
$\mathsf{v} \in N_{\mathbb{R}}$
such that
$\vec{\mathsf{P}} = \vec{\mathsf{Q}}_{\mathsf{v}} = ((\mathsf{Q}_{1})_{\mathsf{v}}, \ldots, (\mathsf{Q}_{s})_{\mathsf{v}})$
. Given
$\vec{\mathsf{w}}\in \textrm{Dr}(\vec{\mathsf{Q}})$
and
$\vec{\omega} = (\vec{\mathsf{Q}}, \vec{\mathsf{a}}, \vec{\mathsf{w}})$
, the above partial order turns
$\mathcal{Q}(\vec{\omega})$
into a poset. The following is the key proposition that allows us to define the face morphisms
$\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{P}})$
when
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
and the closed immersion in Theorem 1·1.
Proposition 3·4. Let
$\vec{\mathsf{w}} = (\mathsf{w}_1,\ldots,\mathsf{w}_s) \in \mathrm{TF}\ell(\vec{\mathsf{Q}})$
,
$\mathsf{v}\in N_{\mathbb{R}}$
and set
$\vec{\mathsf{P}} = \vec{\mathsf{Q}}_{\mathsf{v}}^{\vec{\mathsf{w}}}$
. The inclusion
$B_{\vec{\mathsf{P}}} \subset B_{\vec{\mathsf{Q}}}$
induces a morphism
$\varphi_{\omega,\mathsf{v}}\colon \textrm{in}_{\vec{\mathsf{w}}}\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{P}})$
.
Proof. With
$\vec{\mathsf{P}} = (\mathsf{P}_{1},\ldots, \mathsf{P}_{s})$
, we have that
$\mathsf{P}_i = (\mathsf{Q}_i)_{\mathsf{v}}^{\mathsf{w}_i}$
, thus the bases of
$\mathsf{P}_i$
can be determined by (2·8). It suffices to show that the extension of
$I_{\vec{\mathsf{P}}}$
to
$B_{\vec{\mathsf{Q}}}$
is contained in
$\textrm{in}_{\vec{\mathsf{w}}}I_{\vec{\mathsf{Q}}}$
. We use the quadric generators in (3·2).
Suppose
$f_{\mu,\nu}(\vec{\mathsf{P}})$
is not identically 0, i.e., there is an
$k_0\in \nu\setminus \mu$
such that
$\mu\cup k_0 \in \mathsf{Q}_i$
and
$\nu\setminus k_0 \in \mathsf{Q}_j$
. We must show that
$f_{\mu,\nu}(\vec{\mathsf{P}}) = \textrm{in}_{\vec{\mathsf{w}}} f_{\mu,\nu}(\vec{\mathsf{Q}})$
, i.e., if
$k\in \nu\setminus \mu$
, then
$p_{\mu\cup k}p_{\nu\setminus k}$
is a monomial appearing in
$f_{\mu,\nu}(\vec{\mathsf{P}})$
if and only if it is a monomial in
$\textrm{in}_{\vec{\mathsf{w}}} f_{\mu,\nu}(\vec{\mathsf{Q}})$
. The monomial
$p_{\mu\cup k}p_{\nu\setminus k}$
is a monomial of
$\textrm{in}_{\vec{\mathsf{w}}} f_{\mu,\nu}(\vec{\mathsf{Q}})$
if and only if
Next, observe that for any
$\ell$
we have
which is independent of
$\ell$
. Thus (3·4) holds if and only if
for all
$\ell$
. This is true if and only if
which is true if and only if
$\mu\cup k \in \mathsf{P}_i$
and
$\nu\setminus k \in \mathsf{P}_j$
, as required.
If
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
, then by Proposition 3·4 applied to the case
$\vec{\mathsf{w}} = 0$
, we have a face morphism
These face morphisms satisfy
This means that
defines a diagram of type
$\mathcal{Q}(\vec{\omega})$
. Thus we may form the inverse limit
This is an affine scheme whose coordinate ring is the
$\mathbb{G}_m^s$
-invariants subring of
$\varinjlim_{\mathcal{Q}(\vec{\omega})} R_{\vec{\mathsf{P}}}$
.
Theorem 3·5. Let
$\vec{\mathsf{w}} \in \mathrm{TF}\ell(\vec{\mathsf{Q}})$
and
$\vec{\omega} = (\vec{\mathsf{Q}}, \vec{\mathsf{a}}, \vec{\mathsf{w}})$
. The morphisms
$\varphi_{\vec{\omega},\mathsf{v}}\,:\,\textrm{in}_{\vec{\mathsf{w}}} \mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{Q}}_{\mathsf{v}}^{\vec{\mathsf{w}}})$
induce a closed immersion
Theorem 1·1 follows from this because
$\mathrm{F}\ell^{\circ}(\vec{r},n)$
is the flag matroid stratum
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
when
$\vec{\mathsf{Q}}$
is the uniform
$(\vec{r},n)$
–flag matroid.
Proof of Theorem
3·5. The map
$\varphi_{\vec{\omega}}$
exists by the universal property. Given any
$\lambda \in \mathcal{B}(\vec{\mathsf{Q}})$
, there is a
$\vec{\mathsf{P}}$
defining a face of
$\mathcal{Q}(\vec{\omega})$
such that
$\lambda \in \mathcal{B}(\vec{\mathsf{P}})$
. Therefore, the induced map
is surjective and so
$\varphi_{\vec{\omega}}$
is a closed immersion.
4. Faces of flag matroid polytopes
Let
$\mathsf{Q}$
be a (r,n)–matroid and
$\Delta(\mathsf{Q})$
its polytope. The faces of
$\Delta(\mathsf{Q})$
may be described as follows. Given
$\mathsf{v} \in N_{\mathbb{R}}$
, set
$\mathsf{Q}_{\mathsf{v}} = \mathsf{Q}_{\mathsf{v}}^{0}$
. The face of
$\Delta(\mathsf{Q})$
minimised by the vector
$\mathsf{v}$
is
$\Delta(\mathsf{Q}_{\mathsf{v}})$
and when
$\mathsf{v}= - \epsilon_{\lambda}$
we have
where
$\mathsf{Q} | \lambda$
is the restriction of
$\mathsf{Q}$
to
$\lambda$
,
$\mathsf{Q}/\lambda$
is the contraction of
$\mathsf{Q}$
by
$\lambda$
and
$\oplus$
is the direct sum of matroids. The direct sum of matroids interacts well with the formation of matroid polytopes and matroid strata:
In this section, we recall how this generalises to flag matroids. Our arguments use the Bergman fan
$\mathsf{Berg}(\mathsf{Q})$
of a (r,n)–matroid
$\mathsf{Q}$
[
Reference Ardila and Klivans3
,
Reference Sturmfels38
]. If
$\mathsf{Q}$
is loopless, its Bergman fan
$\mathsf{Berg}(\mathsf{Q})$
is the support of a fan in
$N_{\mathbb{R}}$
whose cones are
$\mathbb{R}_{\geq 0} \langle \epsilon_{\lambda_1},\ldots,\epsilon_{\lambda_k} \rangle$
for each sequence
$\lambda_1 \subsetneq \cdots \subsetneq \lambda_k$
of proper flats, i.e., flats whose rank is positive and strictly less than the rank of
$\mathsf{Q}$
. Equivalently,
$\mathsf{Berg}(\mathsf{Q})$
is
In general, let
$\mathsf{lp}(\mathsf{Q})\subset [n]$
be the set of loops of
$\mathsf{Q}$
and
$\widetilde{\mathsf{Q}} = \mathsf{Q}|([n]\setminus \mathsf{lp}(\mathsf{Q}))$
. Then
$\mathsf{Berg}(\mathsf{Q})$
is defined to be
$\mathsf{Berg}(\widetilde{\mathsf{Q}})$
, which is this is the support of a fan in
$\widetilde{N}_{\mathbb{R}}^{\mathsf{Q}}$
where
$\widetilde{N}^{\mathsf{Q}} \,:\!=\, N/(\langle \mathbf{1} \rangle + \mathbb{Z}^{\mathsf{lp}(\mathsf{Q})})$
. The essential facts we need are contained in the following proposition; these are well-known when the relevant matroids are loopless and one readily extends them to the general case.
Proposition 4·1.
-
(i) If
$\mathsf{Q}$
is a matroid on [n] and
$\lambda \subset [n]$
, then
$\mathsf{Berg}(\mathsf{Q}|\lambda)$
equals the image of
$\mathsf{Berg}(\mathsf{Q})$
under the coordinate projection
$\widetilde{N}_{\mathbb{R}}^{\mathsf{Q}} \to \widetilde{N}_{\mathbb{R}}^{\mathsf{Q}|\lambda}$
. -
(ii) The matroid
$\mathsf{Q}_1$
is a quotient of
$\mathsf{Q}_2$
if and only if
$\mathsf{lp}(\mathsf{Q}_2) \subset \mathsf{lp}(\mathsf{Q}_1)$
and
$\mathsf{Berg}(\mathsf{Q}_1) \subset p(\mathsf{Berg}(\mathsf{Q}_2))$
, where
$p\,:\,\widetilde{N}^{\mathsf{Q}_2} \to \widetilde{N}^{\mathsf{Q}_1}$
is the coordinate projection. -
(iii)
$\mathsf{Berg}(\mathsf{Q}_1 \oplus \mathsf{Q}_2) \cong \mathsf{Berg}(\mathsf{Q}_1) \times \mathsf{Berg}(\mathsf{Q}_2)$
.
Given a
$(\vec{r},n)$
–flag matroid
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\dots,\mathsf{Q}_s)$
and
$\lambda\subset [n]$
, the restriction
$\vec{\mathsf{Q}} | \lambda$
and contraction
$\vec{\mathsf{Q}}/\lambda$
are defined componentwise:
Proposition 4·2. Given a flag matroid
$\vec{\mathsf{Q}}$
and
$\lambda\subset [n]$
, the restriction
$\vec{\mathsf{Q}}|\lambda$
and contraction
$\vec{\mathsf{Q}}/\lambda$
are flag matroids.
Proof. This proposition is [ Reference Brandt, Eur and Zhang7 , theorem 4.1.5], equivalently, [ Reference Aguiar and Ardila1 , theorem 6·1] and (2·2). We argue using the Bergman fan of a matroid. Consider the commutative diagram of coordinate projections
where
$i\leq j$
. By Proposition 4·1, we have that
$\mathsf{Berg}(\mathsf{Q}_i) \subset p_1(\mathsf{Berg}(\mathsf{Q}_j))$
and
So
$\mathsf{Q}_i|\lambda$
is a quotient of
$\mathsf{Q}_j|\lambda$
. This proves that
$\vec{\mathsf{Q}}|\lambda$
is a flag matroid. Next consider contraction. The dual of
$\vec{\mathsf{Q}}$
, defined by
$\vec{\mathsf{Q}}^* \,:\!=\, (\mathsf{Q}_s^{*}, \ldots, \mathsf{Q}_1^{*})$
, is a flag matroid [
Reference White40
, proposition 7.4.7]. By the restriction case, we have that
$\vec{\mathsf{Q}}^*|([n]\setminus \lambda)$
is a flag matroid and therefore its dual, which is exactly
$\vec{\mathsf{Q}}/\lambda$
, is also a flag matroid.
Before considering direct sums of flag matroids, we review direct sums of point configurations. Let M and M’ be the lattices dual to
$\mathbb{Z}^{n}/\langle \mathbf{1} \rangle$
and
$\mathbb{Z}^{n'}/\langle \mathbf{1} \rangle$
, respectively and let
$(\mathsf{Q},\mathsf{a})$
and
$(\mathsf{Q}',\mathsf{a}')$
be point configurations in M and M’, respectively. The direct sum
$(\mathsf{Q},\mathsf{a})\oplus (\mathsf{Q}',\mathsf{a}')$
is the point configuration
$(\mathsf{Q}\oplus \mathsf{Q}', \mathsf{a} \oplus \mathsf{a}')$
in
$M\oplus M'$
, where
Given
$(\vec{r},n)$
and
$(\vec{r}',n')$
–flag matroids
$\vec{\mathsf{Q}} = (\mathsf{Q}_1,\ldots,\mathsf{Q}_s)$
and
$\vec{\mathsf{Q}}'=(\mathsf{Q}_{1}',\ldots,\mathsf{Q}_{s}')$
, their direct sum is
Proposition 4·3. The direct sum
$\vec{\mathsf{Q}} \oplus \vec{\mathsf{Q}}'$
is a
$(\vec{r}+\vec{r}',n+n')$
matroid. Its polytope and stratum are
Proof. We argue using the Bergman fan. By Proposition 4·1(i), we have that
$\mathsf{Berg}(\mathsf{Q}_i) \subset \mathsf{Berg}(\mathsf{Q}_j)$
and
$\mathsf{Berg}(\mathsf{Q}_i') \subset \mathsf{Berg}(\mathsf{Q}_j')$
for
$i\leq j$
. By part (3) of that proposition, we also have that
$\mathsf{Berg}(\mathsf{Q}_i \oplus \mathsf{Q}_i') = \mathsf{Berg}(\mathsf{Q}_i) \times \mathsf{Berg}(\mathsf{Q}_i')$
. So
$\mathsf{Berg}(\mathsf{Q}_i \oplus \mathsf{Q}_i') \subset \mathsf{Berg}(\mathsf{Q}_j \oplus \mathsf{Q}_j')$
for
$i\leq j$
and therefore
$\vec{\mathsf{Q}}\oplus \vec{\mathsf{Q}}'$
is a flag matroid.
Next, consider the point configurations
$(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
and
$(\vec{\mathsf{Q}},\vec{\mathsf{a}})$
associated to the flag matroids. We have
from which the statement on flag matroid polytopes follows.
Finally, consider the claim on flag strata. This result is not critical for our main results, so we prove this only at the level of
$\mathbb{C}$
-points as it is clearer than a scheme-theoretic argument. Define
\begin{align*} \mathrm{F}\ell(\vec{\mathsf{Q}}) \oplus \mathrm{F}\ell(\vec{\mathsf{Q}}') &\to \mathrm{F}\ell(\vec{\mathsf{Q}} \oplus \vec{\mathsf{Q}}') \\ ((F_{1}\subset \cdots \subset F_{s}),(F_{1}'\subset \cdots \subset F_{s}')) &\mapsto (F_{1}\oplus F_{1}'\subset \cdots \subset F_{s}\oplus F_{s}'). \end{align*}
Because
$\textrm{Gr}(\mathsf{Q}\oplus \mathsf{Q}') = \textrm{Gr}(\mathsf{Q}) \times \textrm{Gr}(\mathsf{Q}')$
, any flag in
$\mathrm{F}\ell(\vec{\mathsf{Q}}\oplus \vec{\mathsf{Q}})$
is of the form
$(F_{1}\oplus F_{1}'\subset \cdots \subset F_{s}\oplus F_{s}')$
. We can use this to produce an inverse to the above map.
Given a flag matroid
$\vec{\mathsf{Q}}$
, any face of
$\Delta(\vec{\mathsf{Q}})$
is of the form
$\Delta(\vec{\mathsf{Q}}_{\mathsf{v}})$
, where
$\mathsf{v}=-\epsilon_{\lambda}$
for some
$\lambda\subset [n]$
;
$\vec{\mathsf{Q}}_{\mathsf{v}}$
is the flag matroid
$\vec{\mathsf{Q}}_{\mathsf{v}}^{0}$
from (2·2).
Proposition 4·4. Let
$\vec{\mathsf{Q}}$
be a flag matroid and
$\mathsf{v}=-\epsilon_{\lambda}$
. Then the flag matroid
$\vec{\mathsf{Q}}_{\mathsf{v}}$
is
Proof. This is a special case of [ Reference Aguiar and Ardila1 , theorem 6·1]. Alternatively, observe that
Remark 4·5. The flag matroids
$\vec{\mathsf{Q}}|\lambda$
and
$\vec{\mathsf{Q}}/\lambda$
may have repeated constituents; see Remark 3·2 on the formation of their strata.
5. The complete flag variety
In the remaining sections, we specialise to the case of the complete flag variety
5·1. Affine coordinates
The standard flag is the flag of subsets
Let
$\vec{\mathsf{Q}}$
be a
$\mathbb{C}$
-realisable flag matroid of rank
$(1,\ldots,n-1)$
on [n] and without loss of generality, assume that
$\mathcal{B}(\vec{\mathsf{Q}})$
(as defined in (2·8)) contains the standard flag. Define the polynomial ring
and
$B^{x}$
-valued matrix
\begin{equation} {} {}X = \begin{bmatrix} {}1 & \quad x_{12} & \quad \cdots & \quad x_{1,n-1} & \quad x_{1n} \\ {}0 & \quad 1 & \quad \cdots & \quad x_{2,n-1} & \quad x_{2n} \\ {}\cdots & \quad \cdots & \quad \cdots & \quad \cdots & \quad \cdots \\ {}0 & \quad 0 & \quad \cdots & \quad 1 & \quad x_{n-1,n} \\ {}0 & \quad 0 & \quad \cdots & \quad 0 & \quad 1 {}\end{bmatrix}. {}\end{equation}
Given
$\lambda \subset [n]$
, let
$X_{\lambda}$
denote the determinant of the submatrix of X whose rows are
$1,\ldots,|\lambda|$
and columns are indexed by
$\lambda$
. Define
(if
$i=1$
, then
$[i-1] \cup j$
is just the singleton set consisting of j) and the quotient map
Now let
and set
The ring homomorphism
defines an isomorphism of
$R_{\vec{\mathsf{Q}}}^{x}$
to the coordinate ring of
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
.
Proposition 5·1. The morphism
$\varphi_{\vec{\mathsf{Q}},\vec{\mathsf{P}}'}\colon \mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{P}}')$
corresponds to the ring homomorphism
Proof. As usual, assume that
$\mathcal{B}(\vec{\mathsf{Q}})$
and
$\mathcal{B}(\vec{\mathsf{P}}')$
contain the standard flag. The composition
$\psi_{\vec{\mathsf{Q}}}^{-1} \circ \varphi_{\vec{\mathsf{Q}},\vec{\mathsf{P}}'}^{\#} \circ \psi_{\vec{\mathsf{P}}'}$
sends
$x_{ij}$
to
$x_{ij}$
, as required.
We illustrate these constructions with the following example, which we also use in the proof of Theorem 6·4.
Example 5·2. Let
$\vec{\mathsf{Q}}$
be the 4th flag matroid in Table A1, i.e., the one with bases
To simplify notation, let
$u=x_{12}$
,
$v=x_{13}$
,
$w=x_{14}$
,
$x=x_{23}$
,
$y=x_{24}$
,
$z=x_{34}$
. Because
$2,4,124 \in \mathcal{B}(\vec{\mathsf{Q}})$
and
$3,13,14 \notin \mathcal{B}(\vec{\mathsf{Q}})$
, we have that
$B_{\vec{\mathsf{Q}}}^{x} = \mathbb{C}[u,w,z]$
. Furthermore,
$\pi_{\vec{\mathsf{Q}}}(X_{\lambda}) = \tilde{X}_{\lambda}$
, where
\begin{equation*} {} {}\tilde{X} = \begin{bmatrix} {} {}1 & \quad u & \quad 0 & \quad w \\ {} {}0 & \quad 1 & \quad 0 & \quad 0 \\ {} {}0 & \quad 0 & \quad 1 & \quad z \\ {} {}0 & \quad 0 & \quad 0 & \quad 1 {} {}\end{bmatrix}, {} {}\end{equation*}
so
$I_{\vec{\mathsf{Q}}}^{x} = \langle 0 \rangle$
and
$S_{\vec{\mathsf{Q}}}^{x} = \langle u,w,z \rangle_{\textrm{semigp}}$
. Therefore
$R_{\vec{\mathsf{Q}}}^{x} = \mathbb{C}[u^{\pm},w^{\pm}, z^{\pm}]$
.
Now consider the two flag matroids
$\vec{\mathsf{Q}}_a, \vec{\mathsf{Q}}_b \leq \vec{\mathsf{Q}}$
defined by
(These are internal in the sense of Section 5·4.) By a similar computation, we see that
Using Proposition 5·1,
$\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{Q}}_a)$
and
$\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{Q}}_b)$
may be identified with coordinate projections of tori of the form
$\mathbb{G}_m^3 \to \mathbb{G}_m^2$
, which are smooth and surjective with connected fibers. Also, the morphism
is a closed immersion since the induced map on rings
is surjective. We will use this fact in the proof of Theorem 6·4.
Proposition 5·3. Consider two flag matroids
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
with
$\dim \mathrm{F}\ell(\vec{\mathsf{Q}}) = d$
. Let
be a complete flag such that
$\{\lambda_1,\ldots,\lambda_{n-1}\} \subset \mathcal{B}(\vec{\mathsf{P}})$
(and hence a subset of
$\mathcal{B}(\vec{\mathsf{Q}})$
). If
then
$\mathrm{F}\ell(\vec{\mathsf{P}})$
and
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
are smooth and irreducible and
$\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{P}})$
is smooth and dominant with connected fibers.
Proof. After applying a suitable permutation, we may assume that
$\lambda_r = [r]$
. By the hypothesis, we have that
$\dim B_{\vec{\mathsf{Q}}}^{x} = d$
and therefore
$I_{\vec{\mathsf{Q}}}^{x} = \langle 0 \rangle$
. By Proposition 5·1, we also have that
$I_{\vec{\mathsf{P}}}^{x} = \langle 0 \rangle$
. This means that
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
and
$\mathrm{F}\ell(\vec{\mathsf{P}})$
are isomorphic to open subvarieties of tori and the morphism
$\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{P}})$
is induced by a coordinate projection.
5·2. Symmetry of F
$\ell^{\circ}$
(n)
Denote by
$\mathfrak{S}_{n}$
the symmetric group on [n]. The group
$\mathfrak{S}_{n}$
acts on
$\mathrm{F}\ell^{\circ}(n)$
by permuting the coordinates of
$\mathbb{C}^{n}$
and
$\mathfrak{S}_{2}$
acts on
$\mathrm{F}\ell^{\circ}(n)$
by duality, i.e., the unique generator acts on a complete flag by
These two actions commute, so we have an action of of
$\mathfrak{S}_{2} \times \mathfrak{S}_{n}$
on
$\mathrm{F}\ell^{\circ}(n)$
and hence on
$\mathrm{TF}\ell^{\circ}(n)^{\circ}$
. This is the symmetry group of the permutahedron
$\Pi_n$
for
$n\geq 4$
, see [
Reference Joswig, Loho, Luber and Olarte25
, proposition 2].
5·3. Strata for
$n=3$
There are 5 rank-(1,2) flag matroids on [
Reference Ardila and Klivans3
] up to
$(\mathfrak{S}_{2} \times \mathfrak{S}_{3})$
–symmetry:
\begin{gather*} {}\mathcal{B}(\vec{\mathsf{Q}}_1) = \{1,12\}, \hspace{20pt} \mathcal{B}(\vec{\mathsf{Q}}_2) = \{1,2,12\}, \hspace{20pt} \mathcal{B}(\vec{\mathsf{Q}}_3) = \{1,3,12,23\}, \\ {}\mathcal{B}(\vec{\mathsf{Q}}_4) = \{1,2,3,12,23\}, \hspace{20pt} \mathcal{B}(\vec{\mathsf{Q}}_5) = \{1,2,3,12,13,23\}. {}\end{gather*}
Their strata are smooth and irreducible of dimensions 0,1,1,2,3, respectively.
5·4. Strata for
$n=4$
Recall from Remark 2·4 that, for
$\vec{r} = (1,\ldots,n-1)$
, the flag matroid polytope of the uniform
$(\vec{r},n)$
–matroid is the
$(n-1)$
–dimensional permutahedron. Given
$(\vec{r},n)$
–flag matroids
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
, we say that
$\vec{\mathsf{P}}$
is internal if
$\Delta(\vec{\mathsf{P}})$
meets the relative interior of
$\Pi_n$
.
There are, up to
$(\mathfrak{S}_{2} \times \mathfrak{S}_{4})$
–symmetry, 29 rank (1,2,3)–rank flag matroids on [
Reference Billera and Sturmfels4
]. There are 15 such flag matroids whose polytope has the largest possible dimension, i.e., equal to
$\dim(\Pi_4) = 3$
. These flag matroids
$\vec{\mathsf{Q}}$
are listed in Table A1, together with their coordinate rings and internal
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
. Each pair
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
is labeled by a number (left-most column) and a letter (directly adjacent to the set of nonbases of
$\vec{\mathsf{P}}$
in the third column).
Proposition 5·4. The strata
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
for rank (1,2,3) flag matroids on [
Reference Billera and Sturmfels4
] are smooth and irreducible.
Proof. If
$\dim \Delta(\vec{\mathsf{Q}}) \lt 3 = \dim(\Pi_4)$
, then
$\vec{\mathsf{Q}} \cong \vec{\mathsf{Q}}|_{\lambda} \times \vec{\mathsf{Q}}/\lambda$
for some
$\lambda \subseteq [n]$
by Proposition 4·4, so
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
is smooth and irreducible by Proposition 4·3, Section 5·3 and smoothness and irreducibility of thin Schubert cells of matroids on fewer than 4 elements.
Now consider the
$\dim \Delta(\vec{\mathsf{Q}}) = 3$
case. We proceed as in Section 5·1, but to simplify notation, we set
$u=x_{12}$
,
$v=x_{13}$
,
$w=x_{14}$
,
$x=x_{23}$
,
$y=x_{24}$
,
$z=x_{34}$
. For all but the 12th flag matroid in Table A1, it is immediately clear that
$\mathrm{F}\ell(\vec{\mathsf{Q}})$
is isomorphic to a dense open subvariety of an algebraic torus. If
$\vec{\mathsf{Q}}$
is the 12th flag matroid, then
defines a ring isomorphism, the inverse is given by
$(u,v,y,z)\mapsto (u,v,y,z)$
.
Proposition 5·5. If
$\vec{\mathsf{Q}}$
is a rank-(1,2,3) flag matroid on [
Reference Billera and Sturmfels4
] such that
$\dim \Delta(\vec{\mathsf{Q}}) = 3$
and
$\vec{\mathsf{P}}\leq \vec{\mathsf{Q}}$
is internal, then the morphism
$\mathrm{F}\ell(\vec{\mathsf{Q}})\to \mathrm{F}\ell(\vec{\mathsf{P}})$
is smooth and dominant with connected fibers.
Proof. We may apply Proposition 5·3 to the following pairs
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
, thus deducing that
$\mathrm{F}\ell(\vec{\mathsf{Q}}) \to \mathrm{F}\ell(\vec{\mathsf{P}})$
is smooth and dominant with connected fibers in these cases:
-
(i) 1(a), 1(d), 2(a), 2(c), 3(a), 4(a), 4(b), 5(a), 6(a), 6(b), 6(c), 7(a), 7(c), 8(a), 9(b), 10(a), 10(b), 11(a), 11(b), 13(a), 14(a) (use the flag
$\emptyset \subset 1\subset 12 \subset 123$
); -
(ii) 1(b), 1(c), 3(b) (use the flag
$\emptyset \subset 1 \subset 14 \subset 124$
); -
(iii) 2(b) (use the flag
$\emptyset \subset 2 \subset 24\subset 234$
); -
(iv) 7(b) (use the flag
$\emptyset \subset 3 \subset 34\subset 234$
).
The only pairs
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
remaining are 12(a) and 12(b); the permutation (12)(34) (in cyclic notation) interchanges these two pairs, so it suffices to consider the pair
$\vec{\mathsf{P}} \leq \vec{\mathsf{Q}}$
in 12(a). The composition of
$(\varphi_{\vec{\mathsf{Q}},\vec{\mathsf{P}}})^{\#}$
with the isomorphism
$\vartheta$
from (5·4) is the ring homomorphism
which defines a morphism of affine schemes that is smooth and surjective with connected fibers.
6. Inverse limits of complete matroid strata
Given a flag matroid
$\vec{\mathsf{Q}}$
with a weighted point configuration
$\vec{\omega}$
, the subdivision
$\mathsf{Q}(\vec{\omega})$
has numerous cells, making the computation of the inverse limit
$\mathcal{Q}(\vec{\omega})$
a priori unwieldy. In this section, we describe a much smaller subposet of
$\mathcal{Q}(\vec{\omega})$
combinatorially equivalent to the adjacency graph of
$\mathcal{Q}(\vec{\omega})$
and the limit of the corresponding diagram of flag matroid strata coincides with
$\mathrm{F}\ell(\vec{\omega})$
. We prove Theorem 1·2 as an application of this characterisation of
$\mathrm{F}\ell(\vec{\omega})$
and Theorem 1·1.
6·1. Inverse limits over graphs
Let
$\mathcal{C}$
be a category that has finite limits, e.g., the category of affine
$\mathbb{C}$
-schemes. We recall the notion of an inverse limit in
$\mathcal{C}$
parameterised by a finite connected graph G (possibly with loops or multiple edges) in the sense of [
Reference Corey9
, appendix A]. Define a quiver
$\mathsf{Quiv}(G)$
as follows. The vertex set of
$\mathsf{Quiv}(G)$
is the set
$V(G) \cup E(G)$
; write
$q_v$
, resp.
$q_e$
, the vertex of
$\mathsf{Quiv}(G)$
corresponding to the vertex v, resp. the edge e, of G. The quiver
$\mathsf{Quiv}(G)$
has an arrow
$q_v \to q_e$
for each half-edge of e adjacent to v. Viewing
$\mathsf{Quiv}(G)$
as a category in the usual way, a diagram of type
$\mathsf{Quiv}(G)$
in
$\mathcal{C}$
is a functor
$X\,:\,\mathsf{Quiv}(G) \to \mathcal{C}$
. The inverse limit of X parameterised by the quiver
$\mathsf{Quiv}(G)$
is denoted by
Let
$\omega = (\mathsf{Q},\mathsf{a},\mathsf{w})$
be a weighted point configuration. The adjacency graph of
$\mathcal{Q}(\omega)$
, denoted by
$\Gamma(\omega)$
, is the graph with a vertex
$v_{\mathsf{P}}$
for each maximal element
$\mathsf{P}$
of
$\mathcal{Q}(\omega)$
and two vertices
$v_{\mathsf{P}_1}, v_{\mathsf{P}_2}$
share an edge whenever
$\mathsf{P}_1$
and
$\mathsf{P}_2$
share a common facet. Given
$\mathsf{R} \in \mathcal{Q}(\omega)$
, denote by
$\Gamma_{\mathsf{R}}(\omega)$
the full subgraph of
$\Gamma(\omega)$
on the vertices
$v_{\mathsf{P}}$
such that
$\mathsf{R} \leq \mathsf{P}$
.
Given a flag matroid
$\vec{\mathsf{Q}}$
and
$\vec{\omega} = (\vec{\mathsf{Q}},\vec{\mathsf{a}},\vec{\mathsf{w}})$
with
$\vec{\mathsf{w}} \in \textrm{Dr}(\vec{\mathsf{Q}})$
, denote by
$\Gamma(\vec{\omega})$
the adjacency graph of
$\mathcal{Q}(\vec{\omega})$
and given a maximal element
$\vec{\mathsf{P}}$
of
$\mathcal{Q}(\vec{\omega})$
, denote by
$v_{\vec{\mathsf{P}}}$
the corresponding vertex. We now show that
$\Gamma(\vec{\omega})$
provides sufficient data to compute
$\mathrm{F}\ell(\vec{\omega})$
. Our arguments follow those of the analogous statements for limits of thin Schubert cells as in [
Reference Corey9
, appendix C] by Cueto, thus we only sketch the proofs of the following two propositions.
Proposition 6·1. Given a weighted point configuration
$\omega=(\mathsf{Q},\mathsf{a},\mathsf{w})$
and any
$\mathsf{R} \in \mathcal{Q}(\omega)$
, the graph
$\Gamma_{\mathsf{R}}(\omega)$
is connected.
Proof. Without loss of generality, assume that
$\dim \Delta(\mathsf{Q}) = \dim M_{\mathbb{R}}$
(otherwise, replace
$M_{\mathbb{R}}$
by the affine span of
$\Delta(\mathsf{Q})$
). Suppose
$\mathsf{P}_1$
and
$\mathsf{P}_2$
are two maximal elements of
$\mathcal{Q}(\omega)$
such that
$\mathsf{R} \leq \mathsf{P}_1, \mathsf{P}_2$
. Let y be any point in the relative interior of
$\Delta(\mathsf{R})$
and K be a neighborhood of y chosen small enough so that every
$x\in K$
lies in a cell
$\Delta(\widetilde{\mathsf{R}})$
with
$\widetilde{\mathsf{R}} \in \mathcal{Q}(\omega)$
containing
$\Delta(\mathsf{R})$
as a face. Choose a line segment
$[x_1,x_2]$
lying in K such that
$x_1\in P_1$
,
$x_2\in P_2$
and
$[x_1,x_2]$
does not meet any codimension-2 face of
$\mathcal{Q}(\omega)$
. Then
$[x_1,x_2]$
meets only codimension 0 and 1 faces of
$\mathcal{Q}(\omega)$
, each of which has
$\mathsf{R}$
as a face. By recording the cells met by traversing
$[x_1,x_2]$
, we get a path from
$v_{\mathsf{P}_1}$
to
$v_{\mathsf{P}_2}$
in
$\Gamma_{\mathsf{R}}(\omega)$
, as required.
Proposition 6·2. There is an isomorphism of inverse limits
Proof. Denote by
$\mathrm{F}\ell_{\Gamma}(\vec{\omega})$
the inverse limit on the right. As the opposite quiver
$\mathsf{Quiv}(\Gamma(\vec{\omega}))^{\textrm{op}}$
naturally corresponds to a subposet of
$\mathcal{Q}(\vec{\omega})$
(equipped with the face order), the universal property of inverse limits yields a morphism
$\Phi\,:\,\mathrm{F}\ell(\vec{\omega}) \to \mathrm{F}\ell_{\Gamma}(\vec{\omega})$
.
Next, we construct a morphism
$\Psi\,:\,\mathrm{F}\ell_{\Gamma}(\vec{\omega}) \to \mathrm{F}\ell(\vec{\omega})$
whose inverse is
$\Phi$
. Suppose
$\vec{\mathsf{R}} \in \mathcal{Q}(\vec{\omega})$
. Given any maximal element
$\vec{\mathsf{P}}$
of
$\mathcal{Q}(\vec{\omega})$
dominating
$\vec{\mathsf{R}}$
, define
$\mathrm{F}\ell_{\Gamma}(\vec{\omega}) \to \mathrm{F}\ell(\vec{\mathsf{R}})$
by the composition
This map does not depend on the choice of maximal
$\vec{\mathsf{P}} \geq \vec{\mathsf{R}}$
as the induced subgraph of
$\Gamma(\vec{\omega})$
on the vertices
$v_{\vec{\mathsf{P}}}$
, where
$\vec{\mathsf{P}}\geq \vec{\mathsf{R}}$
is connected by Proposition 6·1.
6·2. Initial degenerations of F
$\ell^{\circ}$
(3) and F
$\ell^{\circ}$
(4)
Recall that
$\{\mathsf{e}_{\lambda} \, : \, \lambda \in \mathsf{Q}\}$
are the images of the standard basis vectors under
$\mathbb{Z}^{\mathsf{Q}} \to N(\mathsf{Q})$
and when
$\mathsf{Q}=\binom{[n]}{r}$
we have a natural identification
$N(\mathsf{Q}) = \wedge^r\mathbb{Z}^n / \langle \mathbf{1} \rangle$
. First, consider
$\mathrm{F}\ell^{\circ}(3)$
. The multihomogeneous ideal of
$\mathrm{F}\ell(3)$
as a subvariety of
$\mathbb{P}({\wedge}^1 \mathbb{C}^3) \times \mathbb{P}({\wedge}^2\mathbb{C}^3)$
is
The tropicalisation
$\mathrm{TF}\ell^{\circ}(3)$
lies in
$({\wedge}^1 \mathbb{R}^3) / \langle \mathbf{1} \rangle \times ({\wedge}^2 \mathbb{R}^3) / \langle \mathbf{1} \rangle$
. Its lineality space is
$L_{\mathbb{R}}$
where
\begin{align*} L = \textrm{span}_{\mathbb{Z}}\left\{ \begin{array}{ccc}\mathsf{e}_{1} + \mathsf{e}_{12} + \mathsf{e}_{13},& \mathsf{e}_{2} + \mathsf{e}_{12} + \mathsf{e}_{23},& \mathsf{e}_{3} + \mathsf{e}_{13} + \mathsf{e}_{23}, \\[5pt] \mathsf{e}_{2} + \mathsf{e}_{3} + \mathsf{e}_{23},& \mathsf{e}_{1} + \mathsf{e}_{3} + \mathsf{e}_{13},& \mathsf{e}_{1} + \mathsf{e}_{2} + \mathsf{e}_{12} \end{array} \right\} \end{align*}
and
$\mathrm{TF}\ell^{\circ}(3)$
has 4 cones:
Proposition 6·3. The initial degenerations of
$\mathrm{F}\ell^{\circ}(3)$
are smooth and irreducible.
Proof. Up to
$(\mathfrak{S}_{2}\times\mathfrak{S}_{3})$
–symmetry, there are only 2 distinct initial degenerations:
$\mathrm{F}\ell^{\circ}(3)$
and
$\textrm{in}_{\vec{\mathsf{w}}}\mathrm{F}\ell^{\circ}(3)$
where
$\vec{\mathsf{w}} = \mathsf{e}_1$
; it suffices to consider the latter. Since the ideal of
$\mathrm{F}\ell(3)$
is principal, the coordinate ring of
$\textrm{in}_{\vec{\mathsf{w}}} \mathrm{F}\ell^{\circ}(3)$
is the subring of
$\mathbb{G}_m^2$
-invariants of
which defines a smooth and irreducible variety. To illustrate the application of Theorem 3·5, we sketch another argument. Let
and
$\vec{\omega}$
the weighted point configuration corresponding to
$\vec{\mathsf{w}} = \mathsf{e}_1$
. The subdivision
$\mathcal{Q}(\vec{\omega})$
, illustrated in Figure 1, has 2 maximal faces, whose flag matroids are
$\vec{\mathsf{Q}}_a$
,
$\vec{\mathsf{Q}}_b$
, that meet along
$\vec{\mathsf{Q}}_{ab}$
where
By Proposition 6·2, we have
$\mathrm{F}\ell(\vec{\omega}) \cong \mathrm{F}\ell(\vec{\mathsf{Q}}_a) \times_{\mathrm{F}\ell(\vec{\mathsf{Q}}_{ab})} \mathrm{F}\ell(\vec{\mathsf{Q}}_b)$
. The stratum
$\mathrm{F}\ell(\vec{\mathsf{Q}}_{a})$
is smooth and irreducible of dimension 2 and the morphism
$\mathrm{F}\ell(\vec{\mathsf{Q}}_{b}) \to \mathrm{F}\ell(\vec{\mathsf{Q}}_{ab}) $
is smooth and surjective whose fibers are connected and 1-dimensional. By [
Reference Corey9
, proposition A.2],
$\mathrm{F}\ell(\vec{\omega})$
is smooth and irreducible of dimension 3 and therefore
$\textrm{in}_{\vec{\mathsf{w}}}\mathrm{F}\ell^{\circ}(3)$
is smooth and irreducible by Theorem 3·5 and [
Reference Corey9
, proposition A.8].
Next, consider
$\mathrm{F}\ell^{\circ}(4)$
. The multihomogeneous ideal of
$\mathrm{F}\ell(4)$
as a subvariety of
$\mathbb{P}({\wedge}^1\mathbb{C}^4) \times \mathbb{P}({\wedge}^2\mathbb{C}^4) \times \mathbb{P}({\wedge}^3\mathbb{C}^4)$
is the ideal of
generated by
\begin{align*} {}\begin{array}{ll} {}p_{1}p_{23} - p_{2}p_{13} + p_{3}p_{12}, & p_{1}p_{24} - p_{2}p_{14} + p_{4}p_{12}, \\ {}p_{1}p_{34} - p_{3}p_{14} + p_{4}p_{13}, & p_{2}p_{34} - p_{3}p_{24} + p_{4}p_{23}, \\ {}p_{12}p_{134} - p_{13}p_{124} + p_{14}p_{123}, & p_{12}p_{234} - p_{23}p_{124} + p_{24}p_{123}, \\ {}p_{13}p_{234} - p_{23}p_{134} + p_{34}p_{123}, & p_{14}p_{234} - p_{24}p_{134} + p_{34}p_{124}, \\ {}p_{12}p_{34} - p_{13}p_{24} + p_{14}p_{23}, & p_{1}p_{234} - p_{2}p_{134} + p_{3}p_{124} - p_{4}p_{123}. {}\end{array} {}\end{align*}
The tropicalisation of
$\mathrm{F}\ell^{\circ}(4)$
originally appears in [
Reference Bossinger, Lamboglia, Mincheva and Mohammadi6
]; we summarise the data here. The set
$\mathrm{TF}\ell^{\circ}(4)$
lies in
Its lineality space
$L_{\mathbb{R}}$
is 3-dimensional and the saturated subgroup L is generated by
\begin{align*} {}\begin{array}{ll} {} \mathsf{e}_{1} + \mathsf{e}_{12} + \mathsf{e}_{13} + \mathsf{e}_{14} + \mathsf{e}_{123} + \mathsf{e}_{124} + \mathsf{e}_{134}, &\mathsf{e}_{2} + \mathsf{e}_{12} + \mathsf{e}_{23} + \mathsf{e}_{24} + \mathsf{e}_{123} + \mathsf{e}_{124} + \mathsf{e}_{234}, \\ {} \mathsf{e}_{3} + \mathsf{e}_{13} + \mathsf{e}_{23} + \mathsf{e}_{34} + \mathsf{e}_{123} + \mathsf{e}_{134} + \mathsf{e}_{234}, & {}\mathsf{e}_{4} + \mathsf{e}_{14} + \mathsf{e}_{24} + \mathsf{e}_{34} + \mathsf{e}_{124} + \mathsf{e}_{134} + \mathsf{e}_{234}, \\ {} \mathsf{e}_{2} + \mathsf{e}_{3} + \mathsf{e}_{4} + \mathsf{e}_{23} + \mathsf{e}_{24} + \mathsf{e}_{34} + \mathsf{e}_{234}, &\mathsf{e}_{1} + \mathsf{e}_{3} + \mathsf{e}_{4} + \mathsf{e}_{13} + \mathsf{e}_{14} + \mathsf{e}_{34} + \mathsf{e}_{134}, \\ {} \mathsf{e}_{1} + \mathsf{e}_{2} + \mathsf{e}_{4} + \mathsf{e}_{12} + \mathsf{e}_{14} + \mathsf{e}_{24} + \mathsf{e}_{124}, & {}\mathsf{e}_{1} + \mathsf{e}_{2} + \mathsf{e}_{3} + \mathsf{e}_{12} + \mathsf{e}_{13} + \mathsf{e}_{23} + \mathsf{e}_{123}. {}\end{array} {}\end{align*}
The f-vector of
$\mathrm{TF}\ell^{\circ}(4)$
is
and
$(\mathfrak{S}_{2}\times \mathfrak{S}_{4})$
–orbit representatives of the cones are contained in Table 1.
Theorem 6·4. Let
$\vec{\mathsf{w}} \in \mathrm{TF}\ell^{\circ}(4)$
and
$\vec{\omega} = (\vec{\mathsf{Q}},\vec{\mathsf{a}},\vec{\mathsf{w}})$
. The inverse limit
$\mathrm{F}\ell(\vec{\omega})$
is smooth and irreducible of dimension
$6 = \dim \mathrm{F}\ell(4)$
. In particular,
$\textrm{in}_{\vec{\mathsf{w}}}\mathrm{F}\ell^{\circ}(4)$
is smooth and irreducible.
Proof. If
$\Gamma(\vec{\omega})$
is a tree, as in the graphs 1, 2, 4, 6, or 8, then
$\varprojlim_{\Gamma(\vec{\omega})} \mathrm{F}\ell$
is smooth and irreducible of dimension 6 by Propositions 5·4, 5·5, Table A1 and [
Reference Corey9
, proposition A.6].
Now suppose
$\Gamma(\vec{\omega})$
is one of the remaining graphs. Let
$\vec{\mathsf{Q}},\vec{\mathsf{Q}}_a,\vec{\mathsf{Q}}_b$
be the flag matroids from Example 5·2. Let T be the subgraph obtained by removing the vertices of
$\Gamma(\vec{\omega})$
whose flag matroid is isomorphic
$\mathsf{Q}$
, together with their adjacent edges; in each case T is a tree and the limit
$\varprojlim_{T} \mathrm{F}\ell$
is smooth and irreducible of dimension 6 (following the same procedure as in the previous paragraph).
Cones of
$\mathrm{TF}\ell^{\circ}(4)$
up to
$(\mathfrak{S}_{2}\times \mathfrak{S}_{4})$
–symmetry

For simplicity, suppose that T is obtained by removing a single vertex and its two adjacent edges, as is in the graphs 5 and 10. Then
is a pullback diagram (we are implicitly applying Proposition 6·2) and
$\varphi_{\vec{\mathsf{Q}},\vec{\mathsf{Q}}_a}, \varphi_{\vec{\mathsf{Q}},\vec{\mathsf{Q}}_b}$
is a closed immersion by Example 5·2. The dimension of
$\mathrm{F}\ell(\vec{\omega})$
is at least 6 by Theorem 3·5 and the fact that
$\textrm{in}_{\vec{\mathsf{w}}}\mathrm{F}\ell^{\circ}(4)$
, being flat degenerations of
$\mathrm{F}\ell^{\circ}(4)$
, has dimension 6. Therefore
$\varprojlim_{\Gamma(\vec{\omega})} \mathrm{F}\ell$
is smooth and irreducible of dimension 6 by [
Reference Corey10
, lemma 6·13]. The remaining cases may be handled in a similar fashion, that is, by “reattaching” each of the vertices outside of T one at a time. The last statement follows from Theorem 3·5 and [
Reference Corey9
, proposition A.8].
7. The Chow quotient Fℓ(4)/H
In this final section, we study the normalisation of the Chow quotient
$\mathrm{F}\ell(n)/\kern-5pt/H$
where
$H \subset \textrm{PGL}(n)$
is the diagonal torus. This is a compactification of
$\mathrm{F}\ell^{\circ}(n)/H$
, which has a modular description inspired by [
Reference Hu23
]. When
$n=4$
,
$\mathrm{F}\ell(n)/\kern-5pt/H$
is a schön compactification of
$\mathrm{F}\ell^{\circ}(n)/H$
and we may use tropical techniques by [
Reference Hacking, Keel and Tevelev19
] to determine its log canonical model.
7·1. A torus action on
$\mathrm{F}\ell(\vec{r},n)$
Recall that the diagonal torus of
$\textrm{GL}(n)$
acts on
$\mathbb{C}^{n}$
by scaling coordinates and this induces an action of the diagonal torus
$H\subset \textrm{PGL}(n)$
on the linear subspaces of
$\mathbb{C}^{n}$
which preserves inclusions. In particular, this produces the action
$H\curvearrowright \mathrm{F}\ell(\vec{r},n)$
described in the introduction.
Let
$\vec{\mathsf{Q}}$
be the uniform
$(\vec{r},n)$
–flag matroid (see Remark 2·4), and
$L \subset N(\mathsf{Q}_1)\times \cdots \times N(\mathsf{Q}_s)$
the saturation of the image of the map in (2·6). The subtorus
$T_L \subset T(\mathsf{Q}_1) \times \cdots \times T(\mathsf{Q}_s)$
may be identified with H and the scaling action of
$T_L \curvearrowright \mathbb{P}({\wedge}^{r_1}\mathbb{C}^n) \times \cdots \times \mathbb{P}({\wedge}^{r_s}\mathbb{C}^n)$
restricts to the above action
$H \curvearrowright \mathrm{F}\ell(\vec{r},n)$
. The embedding
$\mathrm{F}\ell(\vec{r},n)\hookrightarrow \mathbb{P}({\wedge}^{r_1}\mathbb{C}^n) \times \cdots \times \mathbb{P}({\wedge}^{r_s}\mathbb{C}^n)$
induces an embedding of normalised Chow quotients
By the main results of [
Reference Kapranov, Sturmfels and Zelevinsky27
], the normalisation of the Chow quotient of a projective toric variety by a subtorus of its dense torus is also a projective toric variety and its polytope is a fiber polytope. In our case, the polytope of
$\left(\mathbb{P}({\wedge}^{r_1}\mathbb{C}^n) \times \cdots \times \mathbb{P}({\wedge}^{r_s}\mathbb{C}^n)\right)/\kern-5pt/H$
is the fiber polytope associated to the projection in (2·7), whose normal fan is
$\mathcal{F}(\vec{\mathsf{Q}},\vec{\mathsf{a}}) / L_{\mathbb{R}}$
.
7·2. Schön compactifications
Given a rational polyhedral pointed fan
$\mathcal{F}$
in
$N(\mathsf{Q})$
, denote by
$X(\mathcal{F})$
its toric variety. Suppose
$Y^{\circ} \subset T(\mathsf{Q})$
is a closed subvariety. The closure Y of
$Y^{\circ}$
in a
$T(\mathsf{Q})$
–toric variety
$X(\mathcal{F})$
is a schön compactification if the multiplication map
$T(\mathsf{Q})\times Y \to X(\mathcal{F})$
is smooth and surjective; in this case the support of
$\mathcal{F}$
, denoted
$|\mathcal{F}|$
, is
$\textrm{Trop}(Y^{\circ})$
. The variety
$Y^{\circ}$
is schön if
$Y^{\circ}$
has a schön compactification, equivalently,
$\textrm{in}_{\mathsf{w}} Y^{\circ}$
is smooth for all
$\mathsf{w}\in \textrm{Trop}(Y^{\circ})$
[
Reference Helm and Katz21
, proposition 3·9]. If
$Y^{\circ}$
is schön and
$\mathcal{F}$
is any pointed rational polyhedral fan with
$|\mathcal{F}| = \textrm{Trop}(Y^{\circ})$
, then the closure of
$Y^{\circ}$
in
$X(\mathcal{F})$
is a schön compactification [
Reference Luxton and Qu31
, theorem 1·5].
We now specialise to the complete flag variety
$\mathrm{F}\ell(4)$
. In this case,
$\mathrm{TF}\ell^{\circ}(4) = \textrm{Dr}(\vec{\mathsf{Q}})$
where
$\vec{\mathsf{Q}}$
is the uniform ((1,2,3),4)–flag matroid, see Remark 2·4 and Section 3·3. The fan
$\mathcal{F}_4 \,:\!=\, \mathcal{F}_{\textrm{Dr}}(\vec{\mathsf{Q}},\vec{\mathsf{a}})/L_{\mathbb{R}}$
has support
$\mathrm{TF}\ell^{\circ}(4) / L_{\mathbb{R}}$
and is described in Table 1. Let
Then
$\sigma_{1}$
and
$\sigma_{2}$
glue along
$\tau$
to form the convex polyhedral cone
Doing this for the 3 pairs of cones in the 14th orbit in Table 1 determines a coarser fan
$\mathcal{F}_4'$
with support
$\mathrm{TF}\ell^{\circ}(4)/L_{\mathbb{R}}$
.
Given a cone
$\tau$
in
$N(\mathsf{Q})_{\mathbb{R}}$
, denote by
$\langle \tau \rangle$
the linear subspace spanned by
$\tau$
and
$\textrm{Star}(\tau)$
the star of
$\tau$
, viewed as a fan in
$(N/\langle \tau \rangle)_{\mathbb{R}}$
.
Lemma 7·1. For each cone
$\tau$
in
$N(\mathsf{Q})_{\mathbb{R}}$
such that
$\tau / L_{\mathbb{R}} \in \mathcal{F}_4'$
, the set
$|\textrm{Star}(\tau)| \subset (N/\langle\tau \rangle)_{\mathbb{R}}$
is not preserved under translation by a rational subspace of
$(N/\langle\tau \rangle)_{\mathbb{R}}$
.
Proof. By [ Reference Corey9 , lemma 7·2], we must show that
where the intersection is taken over all cones
$\sigma$
containing
$\tau$
as a face such that
$\sigma/L_{\mathbb{R}}$
is a maximal cone of
$\mathcal{F}_4'$
. Given such a
$\sigma$
, define a matrix
$A_{\sigma}$
such that
$\langle \sigma \rangle = \ker A_{\sigma}$
. Let
$\widetilde{A}_{\tau}$
be the matrix whose block rows are the
$A_{\sigma}$
such that
$\sigma\supset \tau$
and
$\sigma/L_{\mathbb{R}}\in \mathcal{F}_4'$
is maximal. The intersection on the right in (7·1) is the kernel of
$\widetilde{A}_{\tau}$
. Thus, for each cone
$\tau/L_{\mathbb{R}}$
of
$\mathcal{F}_{4}'$
, we must show that the rank of
$\widetilde{A}_{\tau}$
equals
$14 - \dim \langle \tau \rangle$
. In fact, we need only show this for the non-maximal cones. This is a direct verification; we use OSCAR. See the online supplement (*).
We prove Theorem 1·3 in the following form.
Theorem 7·2. Consider the complete flag variety
$\mathrm{F}\ell(4)$
.
-
(i) The normalised Chow quotient
$\mathrm{F}\ell(4)/\kern-5pt/H$
is a smooth, schön and simple normal crossings compactification
$\mathrm{F}\ell_{\mathbf{1}}^{\circ}(3)$
. -
(ii) The closure
$\mathrm{F}\ell_{\mathbf{1}}^{\textrm{lc}}(3)$
of
$\mathrm{F}\ell^{\circ}(4) / H$
in
$X(\mathcal{F}_4')$
is a schön and log canonical compactification of
$\mathrm{F}\ell_{\mathbf{1}}^{\circ}(3)$
. -
(iii) The refinement
$\mathcal{F}_4 \to \mathcal{F}_4'$
induces a log crepant resolution of singularities
$\mathrm{F}\ell(4)/\kern-5pt/H \to \mathrm{F}\ell_{\mathbf{1}}^{\textrm{lc}}(3)$
.
Proof. Consider statement (i). The normalised Chow quotient
$\mathrm{F}\ell(4)/\kern-5pt/H$
is the closure of
$\mathrm{F}\ell_{\mathbf{1}}^{\circ}(3) = \mathrm{F}\ell^{\circ}(4)/H$
in
$X(\mathcal{F}_4)$
. Using Table 1, one readily verifies that
$\mathcal{F}_4$
is strictly simplicial. As
$\mathcal{F}_4$
is supported on
$\textrm{Trop}(\mathrm{F}\ell^{\circ}(4) / H)$
, we have that
$\mathrm{F}\ell(4)/\kern-5pt/H$
is a schön compactification of
$\mathrm{F}\ell^{\circ}(4) / H$
and is smooth with a simple normal crossings boundary, see [
Reference Corey10
, proposition 7.1].
Next, consider (2). The initial degenerations of
$\mathrm{F}\ell^{\circ}(4)/H$
are smooth and irreducible by Theorem 6·4 and the isomorphism
for any
$\vec{\mathsf{w}} \in \mathrm{TF}\ell^{\circ}(n)$
; here
$\vec{\mathsf{w}}'$
denotes the image of
$\vec{\mathsf{w}}$
in
$(N(\mathsf{Q}_1) \times \cdots \times N(\mathsf{Q}_{n-1}))_{\mathbb{R}}/L_{\mathbb{R}}$
. Compare to [
Reference Corey10
, lemma 7·3] and [
Reference Corey9
, lemma 7·1]. The statement now follows from Lemma 7·1 and [
Reference Corey10
, proposition 7·1]. Finally, (iii) follows from (i) and [
Reference Tevelev39
, theorem 1·4].
Appendix A. Complete flag matroids on [
Reference Billera and Sturmfels4
] and matroidal subdivisions of
$\Pi_4$
In this Appendix, we record the data used in Sections 5-6. In Table A1 we record
$(\mathfrak{S}_{2}\times \mathfrak{S}_{4})$
—orbit representatives of the flag matroids
$\vec{\mathsf{Q}}$
such that
$\dim \Delta(\vec{\mathsf{Q}}) = 3$
, the maximum possible, along with its coordinate ring and internal flag matroids
$\mathsf{Q}' \leq \mathsf{Q}$
. Observe that for each representative
$\mathsf{Q}$
, the set
$\mathcal{B}(\vec{\mathsf{Q}})$
has the standard flag from (5·1), thus the coordinate ring
$R_{\vec{\mathsf{Q}}}^{x}$
is computed exactly as in Section 5·1. See Example 5·2 for a sample computation.
Flag matroid strata of
$\mathrm{F}\ell(4)$

Tables A2, A3, and A4 record all adjacency graphs of flag subdivisions of the
$n=4$
permutahedron, up to
$(\mathfrak{S}_{2} \times \mathfrak{S}_{4})$
—symmetry. The ordering is consistent with Table 1. The flag matroids appearing are listed in the right-most column and the number in parenthesis indicates which
$(\mathfrak{S}_{2}\times \mathfrak{S}_{4})$
–orbit from Table A1 the flag matroid belongs to.
Coarsest matroidal subdivisions of
$\Pi_4$

Intermediate matroidal subdivisions of
$\Pi_4$

Finest matroidal subdivisions of
$\Pi_4$

Acknowledgements
We thank Michael Joswig and Dhruv Ranganathan for helpful conversations and comments on a previous draft. The first author is supported by the SFB-TRR project “Symbolic Tools in Mathematics and their Application” (project-ID 286237555). The second author is supported by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy - The Berlin Mathematics Research Center MATH
$^+$
(EXC-2046/1, project ID 390685689).

















