betaFeedback

WorksMathematics

Presentations of the cohomology ring of Hessenberg varieties for the minimal nilpotent orbit in type A

Abstract

In 2016 Abe and Crooks found a presentation for the cohomology ring of Hessenberg varieties associated with the minimal nilpotent orbit. Specifically, they presented the ring as a quotient of the cohomology ring of the flag variety, and described the kernel of the quotient map as the span of certain Schubert classes. This note improves upon their result in the Lie type A case, by giving a minimal generating set for the kernel. Our description uses substantially fewer generators than that of Abe-Crooks. Our list consists of a set of Schubert polynomials associated to permutations constructed from the Hessenberg function h{\mathbf h}.

Introduction

Let nn be a positive integer and let Fℓags(Cn)\mathcal{F}\ell ags(\mathbb{C}^{n}) denote the flag variety of nested sequences of subspaces V∙=(0⊂V1⊂V2⊂⋯⊂Vn=Cn)\mathbf{V}_{\bullet}=(0\subset V_{1}\subset V_{2}\subset\cdots\subset V_{n}=\mathbb{C}^{n}) such that dim⁡(Vi)=i\dim(V_{i})=i for all ii. Such a sequence V∙\mathbf{V}_{\bullet} of subspaces is called a full flag. Let h:[n]:={1,2,⋯ ,n}→[n]\mathbf{h}:[n]:=\{1,2,\cdots,n\}\to[n] be a weakly increasing function such that h(i)≥i\mathbf{h}(i)\ge i for all ii, called a Hessenberg function. The subvariety

Hess⁡(X,h):={V∙∈Fℓags(Cn):X(Vi)⊆Vh(i) for all i∈[n]}\operatorname{Hess}(X,\mathbf{h}):=\{\mathbf{V}_{\bullet}\in\mathcal{F}\ell ags(\mathbb{C}^{n}):X(V_{i})\subseteq V_{\mathbf{h}(i)}\text{ for all }i\in[n]\}

is the (type A) Hessenberg variety in Fℓags(Cn)\mathcal{F}\ell ags(\mathbb{C}^{n}) associated to a choice of Hessenberg function h\mathbf{h} and a matrix X∈gln(C)X\in\mathfrak{gl}_{n}(\mathbb{C}). The study of Hessenberg varieties lies at the intersection of algebraic geometry, representation theory, combinatorics, and theory of quasisymmetric functions, among many other research areas (cf. [15 Part III] and references therein). In this manuscript we focus exclusively on Lie type A, i.e., the flag variety corresponding to GLn(C)GL_{n}(\mathbb{C}).

The study of the cohomology rings of Hessenberg varieties has been an active research area for some time. We view this theory as part of a rich history of the study of the cohomology rings of subvarieties of (partial and full) flag varieties (as a non-exhaustive list, see e.g. [12, 18, 19, 20]). We also note that a recent preprint [4] studies a situation very similar to ours, namely, the cohomology rings of certain unions of Schubert varieties. One of the challenges of this research area within Hessenberg geometry is to find explicit descriptions of these cohomology rings via generators and relations, in a manner similar to the well-known case of Fℓags(Cn)\mathcal{F}\ell ags(\mathbb{C}^{n}) with its famous Borel presentation. When XX is nilpotent, the case h(i)=i\mathbf{h}(i)=i for all ii is a well-studied special case where the corresponding Hessenberg varieties are called Springer fibers. The cohomology rings of Springer fibers were studied several decades ago (e.g. De Concini–Procesi [9], Garsia–Procesi [11]) and have an explicit description via the Tanisaki ideal (e.g. [11, 21]). However, outside of the Springer case, less is known, although some cases are understood (e.g. [2, 3]).

Abe and Crooks study Hessenberg varieties associated with the minimal nilpotent orbit in [1]. Specifically, they give a description of the cohomology ring of the associated Hessenberg variety Hess(X,h)\mathcal{H}ess(X,\mathbf{h}) as a quotient of the cohomology ring of the flag variety, where the kernel of the quotient map is additively generated by an explicit list of Schubert classes [1 Theorem 32]. In the present manuscript, we improve upon their result by finding a substantially smaller set of Schubert classes which minimally generate this kernel (Theorem B).

It should be acknowledged that there has been other work in Hessenberg geometry which has built on, or generalized, results in Abe and Crooks’ work which is not discussed in this manuscript. Examples include Goldin and Precup’s flat degenerations from minimal semisimple to minimal nilpotent Hessenberg schemes [13], Escobar, Precup, and Shareshian’s results on singular loci, irreducibility, reducedness, and Poincaré polynomials of codimension one Hessenberg varieties in type A [10], and Goldin and Tymoczko’s work on the GKM properties enjoyed by certain Hessenberg varieties [14].

We work with Q\mathbb{Q} coefficients for our cohomology rings. We also assume throughout the paper that nn is a positive integer, n≥2n \ge2. Here and below, we fix once and for all our choice X=E1,nX = E_{1,n}, where E1,nE_{1,n} is the n×nn \times n matrix with a 11 in position (1,n)(1,n) and a 00 elsewhere. In type A, this is a minimal nilpotent element, and the results of [1] apply. With this established, we recall the following.

Definition 1.1. Let w∈Snw \in S_n and let h:[n]→[n]\mathbf{h} : [n] \to[n] be a Hessenberg function. We say that ww is h\mathbf{h}-forbidden if w−1(1)>h(w−1(n))w^{-1}(1) > \mathbf{h}(w^{-1}(n)). We let Fh\mathcal{F}_{\mathbf{h}} denote the set of h\mathbf{h}-forbidden permutations.

The above terminology is motivated by a geometric result in the work of Abe and Crooks: they show in [1 Proposition 3] that ww is h\mathbf{h}-forbidden precisely when its associated coordinate flag does not lie in Hess⁡(X=E1,n,h)\operatorname{Hess}(X = E_{1,n},\mathbf{h}). They then show that the cohomology ring H∗(Hess⁡(X,h);Q)H^*(\operatorname{Hess}(X,\mathbf{h});\mathbb{Q}) is isomorphic to H∗(Flags⁡(Cn);Q)/JhH^*(\operatorname{Flags}(\mathbb{C}^n);\mathbb{Q})/J_{\mathbf{h}} where Jh:=⟨σw:w∈Fh⟩J_{\mathbf{h}} := \langle\sigma_w : w \in\mathcal{F}_{\mathbf{h}}\rangle and σw\sigma_w denotes the Schubert class corresponding to ww.

The main results of this note give a substantially smaller generating set for the denominator of the above description of H∗(Hess⁡(X,h);Q)H^*(\operatorname{Hess}(X,\mathbf{h});\mathbb{Q}). In fact, we give two separate results, as we now explain. Recall that the cohomology of the flag variety H∗(Flags⁡(Cn);Q)H^*(\operatorname{Flags}(\mathbb{C}^n);\mathbb{Q}) has a Borel description as Q[x1,…,xn]/⟨e1,…,en⟩\mathbb{Q}[x_1,\ldots,x_n]/\langle e_1,\ldots,e_n\rangle as in [7], where eie_i is the iith elementary symmetric polynomial in x1,…,xnx_1,\ldots,x_n. In fact, this description is also valid with Z\mathbb{Z} coefficients. For w∈Snw \in S_n, let Sw\mathfrak{S}_w denote the associated Schubert polynomial [16] in the variables x1,…,xnx_1,\ldots,x_n. It is well-known that the Schubert cohomology class σw∈H2ℓ(w)(Flags⁡(Cn);Q)\sigma_w \in H^{2\ell(w)}(\operatorname{Flags}(\mathbb{C}^n);\mathbb{Q}) of the opposite Schubert variety associated to ww can be represented by [Sw][\mathfrak{S}_w] in the above Borel presentation; we use this (opposite) convention for Schubert classes throughout (cf. [15 Part I, Chapter 1] and [1]). In analogy to the ideal JhJ_{\mathbf{h}} above, we now define J~h:=⟨Sw:w∈Fh⟩⊂Z[x1,…,xn]\widetilde{J}_{\mathbf{h}} := \langle\mathfrak{S}_w : w \in\mathcal{F}_{\mathbf{h}}\rangle\subset\mathbb{Z}[x_1,\ldots,x_n]. Our two main results, Theorem A and Theorem B, give efficient sets of generators of J~h\widetilde{J}_{\mathbf{h}} and JhJ_{\mathbf{h}} respectively.

We take a moment to discuss the motivation for presenting our results in the form given below. It would have been possible to present Theorem B as our one main result. We chose instead to present two separate theorems because it makes it clearer that the simplification of the set of generators, which was the original motivation for our investigations, can be understood in two stages. The simplification that is recorded in Theorem A(2) occurs at the level of the polynomial ring, arising from a relationship among Schubert polynomials. It is when we quotient by the ideal generated by positive degree symmetric polynomials (to obtain the cohomology of the flag variety) that we obtain the next simplification, recorded in Theorem B, which allows us to also delete some of the Schubert classes σcj(h)\sigma_{c_j^{(\mathbf{h})}} from the list of generators.

We now define the set of permutations that give our efficient generating sets. For the definition below, we represent permutations by their one-line notation, so for w∈Snw \in S_n we write w=(w(1),…,w(n))w = (w(1),\ldots,w(n)) and refer to w(j)w(j) as the entry in position jj.

Definition 1.2. Let h:[n]→[n]\mathbf{h} : [n] \to[n] be a Hessenberg function. For each j∈[n]j \in[n] with h(j)<n\mathbf{h}(j) < n, let cj(h)∈Snc_j^{(\mathbf{h})} \in S_n be the permutation with nn in position jj, 11 in position h(j)+1\mathbf{h}(j)+1, and all remaining entries arranged in increasing order. In one-line notation, we have

cj(h)=(2,…,j,n,j+1,…,h(j),1,h(j)+1,…,n−1),c_j^{(\mathbf{h})} = (2,\ldots,j,n,j+1,\ldots,\mathbf{h}(j),1,\mathbf{h}(j)+1,\ldots,n-1),

where empty intervals are assumed to be omitted.

Our first result gives a monic Gröbner basis for J~h:=⟨Sw:w∈Fh⟩\widetilde{J}_{\mathbf{h}} := \langle\mathfrak{S}_w : w \in\mathcal{F}_{\mathbf{h}}\rangle in Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n].

Theorem A. Let n≥2n \ge2, h:[n]→[n]\mathbf{h} : [n] \to[n] be a Hessenberg function. Then:

  1. for each j∈[n]j \in[n], h(j)<nh(j) < n, we have

Scj(h)=(∏r=1h(j)xr)Hn−j−1(x1,…,xj)\mathcal{S}_{c_j^{(\mathbf{h})}} = \left(\prod_{r=1}^{h(j)} x_r\right) H_{n-j-1}(x_1,\ldots,x_j)

where Hd(x1,…,xj)H_d(x_1,\ldots,x_j) denotes the complete homogeneous symmetric polynomial of degree d≥0d \ge0, d∈Zd \in\mathbb{Z} in the variables x1,…,xjx_1,\ldots,x_j, with H0:=1H_0 := 1,

  1. as ideals in Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n] we have

J~h:=⟨Sw:w∈Fh⟩=⟨Scj(h):j∈[n], h(j)<n⟩\widetilde{J}_{\mathbf{h}} := \langle\mathcal{S}_w : w \in\mathcal{F}_{\mathbf{h}}\rangle= \langle\mathcal{S}_{c_j^{(\mathbf{h})}} : j \in[n],\ h(j) < n\rangle
  1. for the lexicographic order xn>⋯>x1x_n > \cdots> x_1, the polynomials {Scj(h):j∈[n], h(j)<n}\{\mathcal{S}_{c_j^{(\mathbf{h})}} : j \in[n],\ h(j) < n\} form a monic Gröbner basis for J~h\widetilde{J}_{\mathbf{h}}.

We note that the generating set in Theorem A need not be minimal (see Remark 3.7). The above theorem shows that the set {Scj(h):j∈[n], h(j)<n}\{\mathcal{S}_{c_j^{(\mathbf{h})}} : j \in[n],\ h(j) < n\} is sufficient to generate J~h\widetilde{J}_{\mathbf{h}}. The analogous result for the ideal JhJ_{\mathbf{h}} in H∗(Fℓags(Cn);Q)H^*(\mathcal{F}\ell ags(\mathbb{C}^n); \mathbb{Q}) requires even fewer generators, indexed by a set D(h)D(\mathbf{h}) defined below. We have the following.

Theorem B. Let n≥2n \ge2, let h:[n]→[n]\mathbf{h} : [n] \to[n] be a Hessenberg function, and set

D(h):={j∈[n−1]:h(j)<h(j+1)}(1)D(\mathbf{h}) := \{j \in[n-1] : h(j) < h(j+1)\} \tag*{(1)}

Restriction induces an isomorphism of graded Q\mathbb{Q}-algebras

H∗(Hess⁡(E1,n,h);Q)≅H∗(Fℓags(Cn);Q)⟨σcj(h):j∈D(h)⟩.H^*(\operatorname{Hess}(E_{1,n},\mathbf{h});\mathbb{Q}) \cong\frac{H^*(\mathcal{F}\ell ags(\mathbb{C}^n);\mathbb{Q})}{\langle\sigma_{c_j^{(\mathbf{h})}} : j \in D(\mathbf{h})\rangle}.

Moreover, every ideal-generating set of this kernel has at least ∣D(h)∣|D(\mathbf{h})| elements, and thus our classes {σcj(h):j∈D(h)}\{\sigma_{c_j^{(\mathbf{h})}} : j \in D(\mathbf{h})\} form a minimal generating set.

We call the elements of D(h)D(\mathbf{h}) the corners of h\mathbf{h}; they are the last indices of the maximal intervals on which h\mathbf{h} is constant and less than nn. It is also worth comparing our description of JhJ_{\mathbf{h}} in Theorem B with that of Abe and Crooks. There are ∣Fh∣=(n−2)!∑j=1n(n−h(j))|\mathcal{F}_{\mathbf{h}}| = (n-2)!\sum_{j=1}^{n}(n-h(j)) many additive generators for JhJ_{\mathbf{h}} in the Abe-Crooks description. Our description of the same ideal has ∣D(h)∣<n|D(\mathbf{h})| < n generators, and these form a minimal set. Thus, a cardinality that is factorial in nn has been replaced by one that is linear in nn.

Our paper is structured as follows. Section 2 reviews the necessary background on flag varieties and Hessenberg varieties. We also discuss Schubert polynomials and their connections to divided difference operators. Our main results are proved in Sections 3 and 4.

Background

We use the flag variety Fℓags(Cn)\mathcal{F}\ell ags(\mathbb{C}^n) and the Hessenberg varieties Hess(X,h)\mathcal{H}ess(X,\mathbf{h}) defined in § 1, and fix once and for all the matrix

X:=E1,n=[00⋯0100⋯00⋮⋮⋱⋮⋮00⋯00]X := E_{1,n} = \begin{bmatrix} 0 & 0 & \cdots& 0 & 1 \\ 0 & 0 & \cdots& 0 & 0 \\ \vdots& \vdots& \ddots& \vdots& \vdots\\ 0 & 0 & \cdots& 0 & 0 \end{bmatrix}

with a single 1 entry in the upper-right corner and 0’s elsewhere. This matrix X=E1,nX = E_{1,n} is in the weight space of the (unique) highest root of GLn(C)GL_n(\mathbb{C}) and belongs to a minimal nilpotent orbit of GLn(C)GL_n(\mathbb{C}). As noted in [1 Lemma 1], Hessenberg varieties associated to different matrices in the same orbit are isomorphic, so to study the topology of Hessenberg varieties associated to the minimal nilpotent orbit, it suffices to study Hess⁡(X=E1,n,h)\mathcal{H}\operatorname{ess}(\mathbf{X} = E_{1,n},\mathbf{h}). Thus, in this manuscript we fix X=E1,n\mathbf{X} = E_{1,n} and allow h\mathbf{h} to vary. It is known that, in this case, the Hessenberg variety Hess⁡(X,h)\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h}) is a union of Schubert varieties.

We now recall the results of Abe and Crooks [1] regarding the cohomology rings of Hess⁡(X,h)\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h}). These results form the starting point of our investigation. A permutation w∈Snw \in S_n determines a permutation matrix in GLn(C)GL_n(\mathbb{C}) and hence an associated permutation flag, see e.g. [15 Section 1.3.3]. Let T⊂GLn(C)T \subset GL_n(\mathbb{C}) be the maximal torus of diagonal invertible matrices. Then TT acts on Fℓags(Cn)\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n), and the TT-fixed points are precisely the permutation flags corresponding to elements w∈Snw \in S_n [15 Exercise 5.63]. By a slight abuse of notation we will also denote by ww the permutation flag, so we write w∈Fℓags(Cn)w \in\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n). With this notation we have

Fℓags(Cn)T=Sn\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n)^T = S_n

and Abe and Crooks prove the following for Hess⁡(E1,n,h)\mathcal{H}\operatorname{ess}(E_{1,n},\mathbf{h}):

Proposition 2.1 ([1 Proposition 3]). Let X=E1,n\mathbf{X} = E_{1,n} as above and h:[n]→[n]\mathbf{h} : [n] \to[n] a Hessenberg function. Then the set of TT-fixed points of Hess⁡(X,h)\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h}) is

Hess⁡(X,h)T={w∈Sn∣w−1(1)≤h(w−1(n))}.(2)\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h})^T = \{w \in S_n \mid w^{-1}(1) \leq\mathbf{h}(w^{-1}(n))\}. \tag*{(2)}

Comparing (2) with Definition 1.1, we see that the h\mathbf{h}-forbidden permutations are precisely the permutations that are not TT-fixed points of Hess⁡(X,h)\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h}), that is,

Fh=Sn∖Hess⁡(X,h)T.\mathcal{F}_{\mathbf{h}} = S_n \setminus\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h})^T.

We emphasize that this description depends on the choice of X=E1,n\mathbf{X} = E_{1,n}.

Next recall that Fℓags(Cn)\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n) has an affine paving given by Schubert cells. Let BB and B−B_{-} denote the Borel subgroups of upper-triangular and lower-triangular matrices in GLn(C)GL_n(\mathbb{C}), respectively. The (opposite) Schubert varieties are defined as X−(w):=B−wB/B⊂GLn(C)/B≅Fℓags(Cn)X_{-}(w) := B_{-}wB/B \subset GL_n(\mathbb{C})/B \cong\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n). Then X−(w)X_{-}(w) determines an opposite Schubert class σw∈H2ℓ(w)(Fℓags(Cn))\sigma_w \in H^{2\ell(w)}(\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n)). (See [15 Chapter 1] for background.)

The following is a combination of [1 Corollary 31] and [1 Theorem 32].

Theorem 2.2. Let X=E1,n\mathbf{X} = E_{1,n} as above and h:[n]→[n]\mathbf{h} : [n] \to[n] a Hessenberg function. Let Hess⁡(X,h)\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h}) denote the Hessenberg variety associated to X\mathbf{X} and h\mathbf{h}. Then the linear subspace

Jh:=⨁w∈FhQ⋅σwJ_{\mathbf{h}} := \bigoplus_{w \in\mathcal{F}_{\mathbf{h}}} \mathbb{Q} \cdot\sigma_w

is an ideal in H∗(Fℓags(Cn);Q)H^*(\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n);\mathbb{Q}), and moreover,

H∗(Hess⁡(X,h);Q)≅H∗(Fℓags(Cn);Q)/Jh.H^*(\mathcal{H}\operatorname{ess}(\mathbf{X},\mathbf{h});\mathbb{Q}) \cong H^*(\mathcal{F}\ell\mathcal{a}gs(\mathbb{C}^n);\mathbb{Q})/J_{\mathbf{h}}.

As mentioned in the introduction, our main technique is to use divided difference operators and Schubert polynomials. We briefly recall the basic definitions and properties we need, following [17 Chapters II and IV]. Let sis_i denote the simple transposition in SnS_n which interchanges ii and i+1i+1 and acts as the identity on other elements.

Definition 2.3. For 1≤i<n1 \leq i < n, let sis_i act on Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n] by interchanging xix_i and xi+1x_{i+1}. The divided difference operator ∂i\partial_i on Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n] is defined by

∂if(x1,x2,…,xn):=f(x1,x2,…,xn)−sif(x1,x2,…,xi,xi+1,…,xn)xi−xi+1\partial_i f(x_1,x_2,\ldots,x_n) := \frac{f(x_1,x_2,\ldots,x_n)-s_i f(x_1,x_2,\ldots,x_i,x_{i+1},\ldots,x_n)}{x_i-x_{i+1}}
=f(x1,x2,…,xn)−f(x1,…,xi+1,xi,…,xn)xi−xi+1.(3)= \frac{f(x_1,x_2,\ldots,x_n)-f(x_1,\ldots,x_{i+1},x_i,\ldots,x_n)}{x_i-x_{i+1}}. \tag*{(3)}

Note that the divided difference operator satisfies the Leibniz rule

∂i(fg)=(∂if)g+(sif)∂ig(4)\partial_i(fg)=(\partial_i f)g+(s_i f)\partial_i g \tag*{(4)}

for all f,g∈Z[x1,…,xn]f,g \in\mathbb{Z}[x_1,\ldots,x_n].

The Schubert polynomial Sw(x1,x2,…,xn)\mathfrak{S}_w(x_1,x_2,\ldots,x_n) of a permutation w∈Snw \in S_n is defined recursively using divided difference operations. Specifically, if w0=(n,n−1,…,1)w_0=(n,n-1,\ldots,1) is the long word permutation, set

Sw0(x1,x2,…,xn):=x1n−1x2n−2⋯xn−1,(5)\mathfrak{S}_{w_0}(x_1,x_2,\ldots,x_n):=x_1^{n-1}x_2^{n-2}\cdots x_{n-1}, \tag*{(5)}

and if si∈Sns_i \in S_n is a simple transposition such that ℓ(wsi)=ℓ(w)+1\ell(ws_i)=\ell(w)+1, then

Sw(x1,x2,…,xn)=∂iSwsi(x1,x2,…,xn).(6)\mathfrak{S}_w(x_1,x_2,\ldots,x_n)=\partial_i\mathfrak{S}_{ws_i}(x_1,x_2,\ldots,x_n). \tag*{(6)}

The Lehmer code of w∈Snw \in S_n is L(w)=(L(w)1,…,L(w)n)L(w)=(L(w)_1,\ldots,L(w)_n), where L(w)i=∣{k>i:w(k)<w(i)}∣L(w)_i=\lvert\{k>i:w(k)<w(i)\}\rvert. That is, L(w)iL(w)_i counts the number of entries to the right of the ii-th position in ww that are smaller than w(i)w(i). We say ww is dominant if L(w)L(w) is weakly decreasing. We will make use of the following fact.

Proposition 2.4. ([17 Equation (4.7)]) Let w∈Snw \in S_n. Suppose that w∈Snw \in S_n is a dominant permutation. Then Sw=x1L(w)1⋯xnL(w)n\mathfrak{S}_w=x_1^{L(w)_1}\cdots x_n^{L(w)_n}.

Proof of Theorem A

In this section we prove Theorem A. The arguments in this section hold over Z\mathbb{Z}, so we work with the polynomial ring Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n]. Here we sketch the outline of our argument. First, we compute the Schubert polynomials Scj(h)\mathfrak{S}_{c_j^{(h)}} explicitly (Lemma 3.3). Proposition 3.6 is the main technical argument of this section, where we show that the Scj(h)\mathfrak{S}_{c_j^{(h)}} generate J~h\widetilde{J}_h. To do so, we define an auxiliary ideal, determined by the positions of nn and 11, and show that it is preserved by the divided difference operators. This is the key step needed for a descending induction argument on Bruhat length. Coupled with the weakly-ascending property of Hessenberg functions, we can express any forbidden Schubert polynomial in terms of those associated to the permutations cj(h)c_j^{(h)}, giving the claim. An examination of their leading terms gives the Gröbner basis statement (Proposition 3.9).

The lemma below records facts concerning the cj(h)c_j^{(h)} which follow directly from Definition 1.2.

Lemma 3.1. Let h:[n]→[n]h:[n]\to[n] be a Hessenberg function and let j∈[n]j\in[n] satisfy h(j)<nh(j)<n. Then

  1. (cj(h))−1(n)=j(c_j^{(h)})^{-1}(n)=j,

  2. (cj(h))−1(1)=h(j)+1(c_j^{(h)})^{-1}(1)=h(j)+1, and

  3. cj(h)∈Fhc_j^{(h)}\in\mathcal{F}_h.

Example 3.2 (Running example). Let n=6n=6 and h=(2,3,3,5,5,6)\mathbf{h}=(2,3,3,5,5,6), and write cj:=cj(h)c_j:=c_j^{(h)} throughout this running example. Since h(j)<6h(j)<6 exactly for j≤5j\leq5, there are five permutations cjc_j. By Definition 1.2, cjc_j is the forbidden permutation with 66 in position jj, 11 in position h(j)+1h(j)+1, and the remaining entries in increasing order. Explicitly, they are c1=(6 2 1 3 4 5),c2=(2 6 3 1 4 5),c3=(2 3 6 1 4 5),c4=(2 3 4 6 5 1),c5=(2 3 4 5 6 1).c_1=(6\ 2\ 1\ 3\ 4\ 5),\quad c_2=(2\ 6\ 3\ 1\ 4\ 5),\quad c_3=(2\ 3\ 6\ 1\ 4\ 5),\quad c_4=(2\ 3\ 4\ 6\ 5\ 1),\quad c_5=(2\ 3\ 4\ 5\ 6\ 1).

For i∈[n]i\in[n], we set the notation

Pi:=x1x2⋯xi,P_i:=x_1x_2\cdots x_i,

and recall from Theorem A that Hd(x1,…,xj)H_d(x_1,\ldots,x_j) denotes the complete homogeneous symmetric polynomial of degree dd in x1,…,xjx_1,\ldots,x_j. We use three elementary facts. First, sjPi=Pis_jP_i=P_i for j≠ij\ne i, so by Definition 2.3, multiplication by PiP_i commutes with ∂j\partial_j:

∂j(Pif)=Pi∂jffor 1≤j<n, j≠i, f∈Z[x1,…,xn].(7)\partial_j(P_if)=P_i\partial_jf\qquad\text{for }1\leq j<n,\ j\ne i,\ f\in\mathbb{Z}[x_1,\ldots,x_n]. \tag*{(7)}

Second, for d≥1d\geq1 and 1≤j<n1\leq j<n, we have [17 (3.9)]

∂jHd(x1,…,xj)=Hd−1(x1,…,xj+1).(8)\partial_jH_d(x_1,\ldots,x_j)=H_{d-1}(x_1,\ldots,x_{j+1}). \tag*{(8)}

Third, applying (8) repeatedly to Hn−2(x1)=x1n−2H_{n-2}(x_1)=x_1^{n-2} gives

∂j−1⋯∂1x1n−2=Hn−j−1(x1,…,xj)for 1≤j<n.(9)\partial_{j-1}\cdots\partial_1 x_1^{n-2}=H_{n-j-1}(x_1,\ldots,x_j)\qquad\text{for }1\le j<n. \tag*{(9)}

We compute the Schubert polynomials of the permutations cj(h)c_j^{(\mathrm{h})}.

Lemma 3.3. Let j∈[n]j\in[n] with h(j)<n\mathrm{h}(j)<n. Then

Scj(h)=Ph(j)⋅Hn−j−1(x1,…,xj).\mathfrak{S}_{c_j^{(\mathrm{h})}}=P_{\mathrm{h}(j)}\cdot H_{n-j-1}(x_1,\ldots,x_j).

Proof. Set k:=h(j)k:=\mathrm{h}(j) and consider the permutation

u=(n,2,…,k,1,k+1,…,n−1).u=(n,2,\ldots,k,1,k+1,\ldots,n-1).

The number of inversions starting in the first position is n−1n-1, the number starting in each of positions 2,…,k2,\ldots,k is 11, and all remaining inversion counts are zero. These counts form the Lehmer code of uu and are weakly decreasing, so uu is dominant. By Proposition 2.4, we have

Su=x1n−1x2⋯xk=Pk⋅x1n−2.\mathfrak{S}_u=x_1^{n-1}x_2\cdots x_k=P_k\cdot x_1^{n-2}.

From the definition of cj(h)c_j^{(\mathrm{h})} we see that cj(h)=us1⋯sj−1c_j^{(\mathrm{h})}=us_1\cdots s_{j-1}. Each successive right multiplication moves nn one position to the right and decreases the length by 1. Thus repeated application of (6) gives

Scj(h)=∂j−1⋯∂1(Pk⋅x1n−2).\mathfrak{S}_{c_j^{(\mathrm{h})}}=\partial_{j-1}\cdots\partial_1(P_k\cdot x_1^{n-2}).

Since j≤kj\le k, each of ∂1,…,∂j−1\partial_1,\ldots,\partial_{j-1} commutes with multiplication by PkP_k by (7), so

Scj(h)=Pk⋅(∂j−1⋯∂1x1n−2)=Pk⋅Hn−j−1(x1,…,xj),\mathfrak{S}_{c_j^{(\mathrm{h})}}=P_k\cdot(\partial_{j-1}\cdots\partial_1x_1^{n-2})=P_k\cdot H_{n-j-1}(x_1,\ldots,x_j),

where the final equality follows from (9).

Example 3.4 (Example 3.2, continued). Lemma 3.3 gives Scj(h)=Ph(j)⋅Hn−j−1(x1,…,xj)\mathfrak{S}_{c_j^{(\mathrm{h})}}=P_{\mathrm{h}(j)}\cdot H_{n-j-1}(x_1,\ldots,x_j) with h(j)<n\mathrm{h}(j)<n. The table below records the corresponding data for this example.

jj12345
cjc_j621345263145236145234651234561
Ph(j)P_{\mathrm{h}(j)}P2P_2P3P_3P3P_3P5P_5P5P_5
H6−j−1(x1,…,xj)H_{6-j-1}(x_1,\ldots,x_j)x14x_1^4H3(x1,x2)H_3(x_1,x_2)H2(x1,x2,x3)H_2(x_1,x_2,x_3)x1+x2+x3+x4x_1+x_2+x_3+x_41

Table 3.4.

The factor Ph(j)P_{\mathrm{h}(j)} is fixed by the position h(j)+1\mathrm{h}(j)+1 of 1, and H6−j−1(x1,…,xj)H_{6-j-1}(x_1,\ldots,x_j) by the position jj of 6. Because h(2)=h(3)\mathrm{h}(2)=\mathrm{h}(3) and h(4)=h(5)\mathrm{h}(4)=\mathrm{h}(5), the pairs c2,c3c_2,c_3 and c4,c5c_4,c_5 share their PP-factor and differ only in H6−j−1(x1,…,xj)H_{6-j-1}(x_1,\ldots,x_j). The two pairs however behave differently. Since H6−5−1(x1,…,x5)=1H_{6-5-1}(x_1,\ldots,x_5)=1, we have

Sc4=P5(x1+x2+x3+x4)=(x1+x2+x3+x4)Sc5,\mathfrak{S}_{c_4}=P_5(x_1+x_2+x_3+x_4)=(x_1+x_2+x_3+x_4)\mathfrak{S}_{c_5},

so c4c_4 is redundant already in Z[x1,…,x6]\mathbb{Z}[x_1,\ldots,x_6]. In contrast, Sc2\mathfrak{S}_{c_2} is not a multiple of Sc3\mathfrak{S}_{c_3}. We will revisit this example in Example 4.7.

Our main result seeks to relate the Schubert polynomials Sw\mathfrak{S}_w for w∈Fhw\in F_h to those of the form Scj(h)\mathfrak{S}_{c_j^{(\mathrm{h})}}. To do this, we introduce an auxiliary ideal that depends only on the positions of nn and 1. By symmetry and (8), the polynomials Hn−j−1(x1,…,xj)H_{n-j-1}(x_1,\ldots,x_j) satisfy

∂iHn−j−1(x1,…,xj)=0for i≠j, and∂jHn−j−1(x1,…,xj)=Hn−j−2(x1,…,xj+1)for 1≤j<n−1.(10)\begin{aligned} \partial_iH_{n-j-1}(x_1,\ldots,x_j)&=0 &&\text{for }i\ne j,\text{ and}\\ \partial_jH_{n-j-1}(x_1,\ldots,x_j)&=H_{n-j-2}(x_1,\ldots,x_{j+1}) &&\text{for }1\le j<n-1. \tag*{(10)} \end{aligned}

For 1≤p<q≤n1\le p<q\le n, define

Ip,q:=Pq−1⟨Hn−2(x1),Hn−3(x1,x2),…,Hn−p−1(x1,…,xp)⟩⊆Z[x1,…,xn].I_{p,q}:=P_{q-1}\langle H_{n-2}(x_1),H_{n-3}(x_1,x_2),\ldots,H_{n-p-1}(x_1,\ldots,x_p)\rangle\subseteq\mathbb{Z}[x_1,\ldots,x_n].

Lemma 3.5. Let w∈Snw\in S_n, and suppose that p=w−1(n)p=w^{-1}(n) and q=w−1(1)q=w^{-1}(1) satisfy p<qp<q. Then

Sw∈Ip,q.(11)\mathfrak{S}_w\in I_{p,q}. \tag*{(11)}

Proof. We first claim that

∂iIp,q⊆Ip,qif i∉{p,q−1}.(12)\partial_i\mathcal{I}_{p,q}\subseteq\mathcal{I}_{p,q}\qquad\text{if }i\notin\{p,q-1\}. \tag*{(12)}

Indeed, let i∉{p,q−1}i\notin\{p,q-1\}, 1≤j≤p1\leq j\leq p, and g∈Z[x1,…,xn]g\in\mathbb{Z}[x_1,\ldots,x_n]. By (7) and (4) we have

∂i(Pq−1⋅g⋅Hn−j−1(x1,…,xj))=Pq−1((∂ig)Hn−j−1(x1,…,xj)+(sig)∂iHn−j−1(x1,…,xj)).\partial_i(P_{q-1}\cdot g\cdot H_{n-j-1}(x_1,\ldots,x_j))=P_{q-1}\bigl((\partial_i g)H_{n-j-1}(x_1,\ldots,x_j)+(s_i g)\partial_iH_{n-j-1}(x_1,\ldots,x_j)\bigr).

Since i≠pi\ne p, (10) shows that ∂iHn−j−1(x1,…,xj)\partial_iH_{n-j-1}(x_1,\ldots,x_j) is either 00 or Hn−j−2(x1,…,xj+1)H_{n-j-2}(x_1,\ldots,x_{j+1}) with j+1≤pj+1\leq p. Hence ∂i(Pq−1⋅g⋅Hn−j−1(x1,…,xj))∈Ip,q\partial_i(P_{q-1}\cdot g\cdot H_{n-j-1}(x_1,\ldots,x_j))\in\mathcal{I}_{p,q}, from which (12) follows.

We now use descending induction on ℓ(w)\ell(w) to prove (11). The base case is the longest word w=w0w=w_0. In this case we have (p,q)=(1,n)(p,q)=(1,n), and

Pn−1Hn−2(x1)=x1n−1x2⋯xn−1.P_{n-1}H_{n-2}(x_1)=x_1^{n-1}x_2\cdots x_{n-1}.

We can then see that the Schubert polynomial Sw0\mathfrak{S}_{w_0} is a multiple of Pn−1Hn−2(x1)P_{n-1}H_{n-2}(x_1) by (5), so the claim holds in this case. Now suppose ww has length ℓ(w)<ℓ(w0)\ell(w)<\ell(w_0) and suppose by induction that the claim holds for all permutations with length greater than kk. Choose an ascent ii of ww, so w(i)<w(i+1)w(i)<w(i+1), and define v=wsiv=ws_i. Since w(p)=nw(p)=n and w(q)=1w(q)=1 by hypothesis, this ascent cannot occur at pp or q−1q-1. Moreover, if p′=v−1(n)p'=v^{-1}(n) and q′=v−1(1)q'=v^{-1}(1), then p′≤pp'\leq p and q′≥qq'\geq q. Consequently, p′<q′p'<q' and

Ip′,q′⊆Ip,q.\mathcal{I}_{p',q'}\subseteq\mathcal{I}_{p,q}.

Since ℓ(v)=ℓ(w)+1\ell(v)=\ell(w)+1, the relation (6) together with the induction hypothesis and (12) now give

Sw=∂iSv∈∂iIp,q⊆Ip,q.\mathfrak{S}_w=\partial_i\mathfrak{S}_v\in\partial_i\mathcal{I}_{p,q}\subseteq\mathcal{I}_{p,q}.

This is what we needed to show. □

We now apply Lemma 3.5 to prove the generation statement.

Proposition 3.6. Let n≥2n\geq2, h:[n]→[n]\mathbf{h}:[n]\to[n] be a Hessenberg function. Then, as ideals in Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n], we have

J~h:=⟨Sw:w∈Fh⟩=⟨Scj(h):j∈[n],h(j)<n⟩.(13)\widetilde{J}_{\mathbf{h}}:=\langle\mathfrak{S}_w:w\in\mathcal{F}_{\mathbf{h}}\rangle=\langle\mathfrak{S}_{c_j^{(\mathbf{h})}}:j\in[n],\mathbf{h}(j)<n\rangle. \tag*{(13)}

Proof. From Lemma 3.1(3) we know that cj(h)∈Fhc_j^{(\mathbf{h})}\in\mathcal{F}_{\mathbf{h}} for each jj with h(j)<n\mathbf{h}(j)<n, so one inclusion is immediate. Thus it suffices to show that if w∈Fhw\in\mathcal{F}_{\mathbf{h}}, then Sw∈⟨Scj(h):j∈[n],h(j)<n⟩\mathfrak{S}_w\in\langle\mathfrak{S}_{c_j^{(\mathbf{h})}}:j\in[n],\mathbf{h}(j)<n\rangle.

Suppose w∈Fhw\in\mathcal{F}_{\mathbf{h}}, and put p=w−1(n)p=w^{-1}(n) and q=w−1(1)q=w^{-1}(1). The forbidden condition and the definition of a Hessenberg function give

p≤h(p)<q.p\leq\mathbf{h}(p)<q.

By Lemma 3.5, we have Sw∈Ip,q\mathfrak{S}_w\in\mathcal{I}_{p,q}. Therefore, in order to show the claim, it would suffice to show that every generator of Ip,q\mathcal{I}_{p,q} lies in the RHS of (13). To see this, let j∈Zj\in\mathbb{Z} with 1≤j≤p1\leq j\leq p. The weakly ascending condition of h\mathbf{h} implies h(j)≤h(p)<q≤n\mathbf{h}(j)\leq\mathbf{h}(p)<q\leq n. In particular, h(j)<n\mathbf{h}(j)<n, and Lemma 3.3 gives

Pq−1Hn−j−1(x1,…,xj)=(∏r=h(j)+1q−1xr)Ph(j)Hn−j−1(x1,…,xj)=(∏r=h(j)+1q−1xr)Scj(h),P_{q-1}H_{n-j-1}(x_1,\ldots,x_j)=\left(\prod_{r=\mathbf{h}(j)+1}^{q-1}x_r\right)P_{\mathbf{h}(j)}H_{n-j-1}(x_1,\ldots,x_j)=\left(\prod_{r=\mathbf{h}(j)+1}^{q-1}x_r\right)\mathfrak{S}_{c_j^{(\mathbf{h})}},

where an empty product is understood to be 11. Thus every generator of Ip,q\mathcal{I}_{p,q} belongs to the RHS of (13), and hence so does Sw\mathfrak{S}_w, as required. □

Remark 3.7. The generating set in Theorem A need not be minimal. If j<n−1j<n-1 and h(j)=h(n−1)=n−1\mathbf{h}(j)=\mathbf{h}(n-1)=n-1, then Scj(h)=(x1⋯xn−1)Hn−j−1(x1,…,xj)\mathfrak{S}_{c_j^{(\mathbf{h})}}=(x_1\cdots x_{n-1})H_{n-j-1}(x_1,\ldots,x_j) is a multiple of Scn−1(h)=x1⋯xn−1\mathfrak{S}_{c_{n-1}^{(\mathbf{h})}}=x_1\cdots x_{n-1}.

Example 3.8 (Example 3.4, continued). We trace the proof of Theorem A(2) for w=(362415)w=(362415). First note that 66 is in position p=2p=2 and 11 in position q=5q=5, which means q>h(p)=h(2)=3q>h(p)=h(2)=3, so ww is h\mathbf{h}-forbidden. Lemma 3.5 says that Sw\mathcal{S}_w lies in I2,5=P4⟨H4(x1),H3(x1,x2)⟩I_{2,5}=P_4\langle H_4(x_1),H_3(x_1,x_2)\rangle, and indeed

Sw=x12x22x3x4(x12+x1x2+x22)=P4(x1H3(x1,x2)−H4(x1)).\mathcal{S}_w=x_1^2x_2^2x_3x_4\left(x_1^2+x_1x_2+x_2^2\right)=P_4\left(x_1H_3(x_1,x_2)-H_4(x_1)\right).

Since P4H3(x1,x2)=x4Sc2P_4H_3(x_1,x_2)=x_4\mathcal{S}_{c_2} and P4H4(x1)=x3x4Sc1P_4H_4(x_1)=x_3x_4\mathcal{S}_{c_1}, the above formula yields Sw=x1x4Sc2−x3x4Sc1\mathcal{S}_w=x_1x_4\mathcal{S}_{c_2}-x_3x_4\mathcal{S}_{c_1}. A similar argument applies to each of the ∣Fh∣=4!⋅12=288|\mathcal{F}_{\mathbf h}|=4!\cdot12=288 forbidden permutations. Thus, in this example, we have replaced the 288288 generators with 55 generators.

Our Theorem A also makes a claim about Gröbner bases. Our next result addresses this.

Proposition 3.9. Let n≥2n\ge2 and h:[n]→[n]\mathbf h:[n]\to[n] be a Hessenberg function. Consider the lexicographic order xn>⋯>x1x_n>\cdots>x_1 on Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n]. Then the polynomials Scj(h)\mathcal{S}_{c_j}^{(\mathbf h)}, for j∈[n]j\in[n] with h(j)<nh(j)<n, form a monic Gröbner basis of J~h\widetilde{J}_{\mathbf h}, with leading monomials

Lj:=Ph(j)⋅xjn−j−1.L_j:=P_{h(j)}\cdot x_j^{n-j-1}.

Proof. For the given lexicographic order, it is immediate from the definition of Hn−j−1(x1,…,xj)H_{n-j-1}(x_1,\ldots,x_j) that its leading term is xjn−j−1x_j^{n-j-1}, which has coefficient 11. Multiplying by the monomial Ph(j)P_{h(j)} then yields that the leading monomial of Ph(j)⋅Hn−j−1(x1,…,xj)P_{h(j)}\cdot H_{n-j-1}(x_1,\ldots,x_j) is Lj=Ph(j)⋅xjn−j−1L_j=P_{h(j)}\cdot x_j^{n-j-1}, which also has coefficient 11. In particular, each Scj(h)\mathcal{S}_{c_j}^{(\mathbf h)} is monic, and it remains to check the SS-polynomials.

Let j1,j2∈[n]j_1,j_2\in[n] with h(j1)<nh(j_1)<n, h(j2)<nh(j_2)<n and assume j1<j2j_1<j_2. For simplicity of notation, set g1:=Scj1(h)g_1:=\mathcal{S}_{c_{j_1}}^{(\mathbf h)} and g2:=Scj2(h)g_2:=\mathcal{S}_{c_{j_2}}^{(\mathbf h)}. We now compute the SS-polynomial S(g1,g2)S(g_1,g_2). From the assumption j1<j2j_1<j_2 we have h(j1)≤h(j2)h(j_1)\le h(j_2), which means

m:=gcd⁡(Lj1,Lj2)=Ph(j1).\mathbf{m}:=\gcd(L_{j_1},L_{j_2})=P_{h(j_1)}.

From the formula for Scj(h)\mathcal{S}_{c_j}^{(\mathbf h)} we also know that the monomial m\mathbf{m} divides both g1g_1 and g2g_2. In particular, if Tj1,Tj2T_{j_1},T_{j_2} denote the remaining terms of g1,g2g_1,g_2 so that

g1=Lj1+Tj1,g2=Lj2+Tj2,g_1=L_{j_1}+T_{j_1},\qquad g_2=L_{j_2}+T_{j_2},

then we also have m∣Tj1\mathbf{m}\mid T_{j_1} and m∣Tj2\mathbf{m}\mid T_{j_2}. Let

M:=lcm⁡(Lj1,Lj2)=Lj1Lj2mM:=\operatorname{lcm}(L_{j_1},L_{j_2})=\frac{L_{j_1}L_{j_2}}{\mathbf{m}}

denote the lcm of Lj1,Lj2L_{j_1},L_{j_2}. Since g1g_1 and g2g_2 are monic, the SS-polynomial is

S(g1,g2)=MLj1g1−MLj2g2=Lj2m(Lj1+Tj1)−Lj1m(Lj2+Tj2)=Lj2Tj1−Lj1Tj2m=Tj1(g2−Tj2)−Tj2(g1−Tj1)m=Tj1mg2−Tj2mg1.\begin{aligned} S(g_1,g_2)&=\frac{M}{L_{j_1}}g_1-\frac{M}{L_{j_2}}g_2\\ &=\frac{L_{j_2}}{\mathbf{m}}(L_{j_1}+T_{j_1})-\frac{L_{j_1}}{\mathbf{m}}(L_{j_2}+T_{j_2})\\ &=\frac{L_{j_2}T_{j_1}-L_{j_1}T_{j_2}}{\mathbf{m}}\\ &=\frac{T_{j_1}(g_2-T_{j_2})-T_{j_2}(g_1-T_{j_1})}{\mathbf{m}}\\ &=\frac{T_{j_1}}{\mathbf{m}}g_2-\frac{T_{j_2}}{\mathbf{m}}g_1. \end{aligned}

Both coefficients in the last expression belong to Z[x1,…,xn]\mathbb{Z}[x_1,\ldots,x_n].

Every monomial in Tj1T_{j_1} is strictly smaller in the given lexicographic order than Lj1L_{j_1}, and likewise for Tj2T_{j_2}. Thus, omitting any zero summands,

LM⁡(Tj1mg2)=LM⁡(Tj1)Lj2m<Lj1Lj2m=M,\operatorname{LM}\left(\frac{T_{j_1}}{\mathbf{m}}g_2\right)=\frac{\operatorname{LM}(T_{j_1})L_{j_2}}{\mathbf{m}}<\frac{L_{j_1}L_{j_2}}{\mathbf{m}}=M,
LM⁡(Tj2mg1)=LM⁡(Tj2)Lj1m<Lj1Lj2m=M.\operatorname{LM}\left(\frac{T_{j_2}}{\mathbf{m}}g_1\right)=\frac{\operatorname{LM}(T_{j_2})L_{j_1}}{\mathbf{m}}<\frac{L_{j_1}L_{j_2}}{\mathbf{m}}=M.

Hence every SS-polynomial has a representation in the generators in which each summand has leading monomial strictly smaller than the corresponding least common multiple. This is an LCM representation in the sense of [8 Chapter 2, § 9, Definition 5]. A refinement of Buchberger’s criterion in terms of LCM representations [8 Chapter 2, § 9, Theorem 6] implies that the gjg_j form a Gröbner basis over Q\mathbb{Q}. To obtain the integral statement, take f∈J~h⊆Z[x1,…,xn]f \in\widetilde{J}_{\mathbf{h}} \subseteq\mathbb{Z}[x_1,\ldots,x_n]. Division by the monic polynomials gjg_j uses only integer coefficients. Its remainder is also a remainder for the same division over Q\mathbb{Q}, where it is zero by the Gröbner-basis property. Thus the gjg_j form a monic Gröbner basis over Z\mathbb{Z}. □

Remark 3.10. The exponent vectors of the leading monomials in Proposition 3.9 are the Lehmer codes of the permutations cj(h)c_j^{(\mathbf{h})}, as defined before Proposition 2.4. Indeed,

L(cj(h))=(1,…,1⏟j−1,n−j,1,…,1⏟h(j)−j,0,…,0⏟n−h(j)).L(c_j^{(\mathbf{h})})=(\underbrace{1,\ldots,1}_{j-1},n-j,\underbrace{1,\ldots,1}_{h(j)-j},\underbrace{0,\ldots,0}_{n-h(j)}).

which is precisely the exponent vector of Lj=Ph(j)⋅xjn−j−1L_j=P_{\mathbf{h}(j)}\cdot x_j^{n-j-1}. These leading terms can also be interpreted as the monomials associated to the bottom pipe dreams of the tuple cj(h)c_j^{(\mathbf{h})}, where crosses in row ii occupy the first L(cj(h))iL(c_j^{(\mathbf{h})})_i columns; see [5].

Remark 3.11. We noted in Remark 3.7 that if h(n−1)=n−1h(n-1)=n-1, then the generators corresponding to j<n−1j<n-1 with h(j)=h(n−1)=n−1h(j)=h(n-1)=n-1 are multiples of Scn−1(h)\mathcal{S}_{c_{n-1}^{(\mathbf{h})}} and hence are not needed. These are the only divisibilities among the leading monomials of the Scj(h)\mathcal{S}_{c_j^{(\mathbf{h})}} appearing in Proposition 3.9, so by discarding the generators corresponding to these jj we still obtain a minimal Gröbner basis.

Proof of Theorem A. Part (1) of the theorem is in Lemma 3.3, part (2) is Proposition 3.6, and Proposition 3.9 establishes the Gröbner basis claim in (3). This completes the proof. □

Proof of Theorem B

In this section, we prove Theorem B. We first sketch our argument. The ring H∗(Flags(Cn);Q)H^*(\mathop{\mathrm{Flags}}(\mathbb{C}^n);\mathbb{Q}) is a quotient of the polynomial ring Q[x1,…,xn]\mathbb{Q}[x_1,\ldots,x_n] by the ideal ⟨e1,…,en⟩\langle e_1,\ldots,e_n\rangle. Our key observation is that if j∈[n−1]j\in[n-1] is such that h(j)=h(j+1)<nh(j)=h(j+1)<n, then the Schubert class σcj(h)\sigma_{c_j^{(\mathbf{h})}} is a multiple of σcj+1(h)\sigma_{c_{j+1}^{(\mathbf{h})}} (cf. Proposition 4.6). This allows us to reduce to the corner generators.

Next, we identify the cj(h)c_j^{(\mathbf{h})} for j∈D(h)j\in D(\mathbf{h}) as the Bruhat-minimal elements of the forbidden set Fh\mathcal{F}_{\mathbf{h}}. The Bruhat support of Schubert multiplication shows that these classes remain linearly independent modulo positive-degree multiples of the kernel, establishing the minimality and cardinality claims. We use Q\mathbb{Q} coefficients, following [1].

For a partially ordered set (S,≤)(S,\leq), we say that a subset TT of SS is a down-set if, whenever w∈Tw\in T and u≤wu\leq w, then u∈Tu\in T. Similarly we say that UU is an up-set if, whenever w∈Uw\in U and u≥wu\geq w, then u∈Uu\in U. Note that the complement of a down-set is an up-set, and vice versa.

Lemma 4.1. Let n≥2n\geq2 and h:[n]→[n]\mathbf{h}:[n]\to[n] be a Hessenberg function. Let X=E1,nX=E_{1,n} and let Hess⁡(X,h)\operatorname{Hess}(X,\mathbf{h}) be the Hessenberg variety corresponding to XX and h\mathbf{h}. Then Hess⁡(X,h)T\operatorname{Hess}(X,\mathbf{h})^T is a down-set with respect to the Bruhat order, and Fh=Sn∖Hess⁡(X,h)T\mathcal{F}_{\mathbf{h}}=S_n\setminus\operatorname{Hess}(X,\mathbf{h})^T is an up-set with respect to the Bruhat order.

We will use the tableau criterion [6 Theorem 2.6.3] for the Bruhat order on SnS_n. This is defined by creating a tableau for a permutation u=(u1,u2,…,un)u=(u_1,u_2,\ldots,u_n) such that in row 1≤i≤n−11\leq i\leq n-1, we place the first n−in-i elements of uu in ascending order. If u,w∈Snu,w\in S_n, then u≤wu\leq w with respect to the Bruhat order if and only if every entry of the tableau of uu is less than or equal to the corresponding entry in the tableau of ww. For example, if u=(21453)u=(21453) and w=(32541)w=(32541), then the tableaux are

1245124122and2345235233\begin{array}{cccc} 1 & 2 & 4 & 5 \\ 1 & 2 & 4 \\ 1 & 2 \\ 2 \end{array} \qquad\text{and}\qquad \begin{array}{cccc} 2 & 3 & 4 & 5 \\ 2 & 3 & 5 \\ 2 & 3 \\ 3 \end{array}

Then u≤wu\leq w in Bruhat order since each entry in the tableau of uu is less than or equal to the corresponding entry in the tableau of ww.

Proof. Suppose u,w∈Snu,w\in S_n. Suppose also that u≤wu\leq w and w∈Hess⁡(X,h)Tw\in\operatorname{Hess}(X,\mathbf h)^T. From the tableau criterion for Bruhat order, if u≤wu\leq w, then u−1(1)≤w−1(1)u^{-1}(1)\leq w^{-1}(1) and u−1(n)≥w−1(n)u^{-1}(n)\geq w^{-1}(n). Since w∈Hess⁡(X,h)Tw\in\operatorname{Hess}(X,\mathbf h)^T and h\mathbf h is weakly increasing, we conclude that

u−1(1)≤w−1(1)≤h(w−1(n))≤h(u−1(n))u^{-1}(1)\leq w^{-1}(1)\leq\mathbf h(w^{-1}(n))\leq\mathbf h(u^{-1}(n))

which means that u∈Hess⁡(X,h)Tu\in\operatorname{Hess}(X,\mathbf h)^T, as desired. The second claim follows immediately. □

Recall from § 1 that the set D(h)D(\mathbf h) of corners is {j∈[n−1]:h(j)<h(j+1)}\{j\in[n-1]:\mathbf h(j)<\mathbf h(j+1)\}. We begin with a characterization of when a permutation lies above cj(h)c_j^{(\mathbf h)} in Bruhat order, in terms of its one-line notation.

Lemma 4.2. In the setting of Lemma 4.1, let j∈[n−1]j\in[n-1] and suppose h(j)<n\mathbf h(j)<n, and let cj(h)c_j^{(\mathbf h)} denote the permutation of Definition 1.2. Let w∈Snw\in S_n. Then:

cj(h)≤w⟺w−1(n)≤j and w−1(1)≥h(j)+1.c_j^{(\mathbf h)}\leq w\Longleftrightarrow w^{-1}(n)\leq j\text{ and }w^{-1}(1)\geq\mathbf h(j)+1.

Proof. First suppose that cj(h)≤wc_j^{(\mathbf h)}\leq w. By Lemma 3.1 we know that nn appears in the jj-th position and 11 appears in the (h(j)+1)(\mathbf h(j)+1)-th position of the one-line notation of cj(h)c_j^{(\mathbf h)}. By the tableau criterion for Bruhat order, it follows that nn must appear in one of the first jj positions of ww, i.e., w−1(n)≤jw^{-1}(n)\leq j, and similarly that 11 must appear in the (h(j)+1)(\mathbf h(j)+1)-th position or further to its right, i.e., w−1(1)≥h(j)+1w^{-1}(1)\geq\mathbf h(j)+1.

For the reverse direction, suppose now that w−1(n)≤jw^{-1}(n)\leq j and w−1(1)≥h(j)+1w^{-1}(1)\geq\mathbf h(j)+1. By the tableau criterion, it suffices to show that for each 1≤a≤n−11\leq a\leq n-1, the ii-th smallest of the first aa entries of cj(h)c_j^{(\mathbf h)} is at most the ii-th smallest of the first aa entries of ww, for every i≤ai\leq a. By Definition 1.2, the first aa entries of cj(h)c_j^{(\mathbf h)} form the set

{2,…,a+1} for a<j,{2,…,a}∪{n} for j≤a≤h(j),{1,…,a−1}∪{n} for a>h(j).\{2,\ldots,a+1\}\text{ for }a<j,\qquad\{2,\ldots,a\}\cup\{n\}\text{ for }j\leq a\leq\mathbf h(j),\qquad\{1,\ldots,a-1\}\cup\{n\}\text{ for }a>\mathbf h(j).

If a≤h(j)a\leq\mathbf h(j), then 11 is not among the first aa entries of ww, so the ii-th smallest of them is at least i+1i+1. If a≥ja\geq j, then nn is among the first aa entries of ww, so the largest of them is nn. Together with the fact that the ii-th smallest of any aa distinct positive integers is at least ii, these two observations give the required inequalities in each of the three cases, so cj(h)≤wc_j^{(\mathbf h)}\leq w. □

Lemma 4.3. In the setting of Lemma 4.1, the Bruhat-minimal elements of Fh\mathcal F_{\mathbf h} are precisely the cj(h)c_j^{(\mathbf h)} where j∈D(h)j\in D(\mathbf h) is a corner of h\mathbf h.

Proof. Suppose w∈Fhw\in\mathcal F_{\mathbf h}. Let j:=w−1(n)j:=w^{-1}(n) be the position of nn in the one-line notation of ww. Since ww is in Fh\mathcal F_{\mathbf h}, this means w−1(1)>h(j)w^{-1}(1)>\mathbf h(j). Lemma 4.2 then implies cj(h)≤wc_j^{(\mathbf h)}\leq w. Thus every element of Fh\mathcal F_{\mathbf h} lies above some member of the family {cj(h):j∈[n−1],h(j)<n}\{c_j^{(\mathbf h)}:j\in[n-1],\mathbf h(j)<n\}, and this family is contained in Fh\mathcal F_{\mathbf h} by Lemma 3.1(3). Hence the Bruhat-minimal elements of Fh\mathcal F_{\mathbf h} are exactly the minimal elements of this family. Next suppose that j,j′∈[n−1]j,j'\in[n-1] and h(j)<n\mathbf h(j)<n and h(j′)<n\mathbf h(j')<n. By Lemma 4.2 we have

cj(h)≤cj′(h)⟺j′≤j and h(j′)≥h(j).c_j^{(\mathbf h)}\leq c_{j'}^{(\mathbf h)} \Longleftrightarrow j'\leq j\text{ and }\mathbf h(j')\geq\mathbf h(j).

By the definition of Hessenberg functions, if j′≤jj' \le j, then h(j′)≤h(j)\mathbf{h}(j') \le\mathbf{h}(j) so in fact we have

cj(h)≤cj′(h)⟺j′≤j and h(j′)=h(j).(14)c_{j}^{(\mathbf{h})} \leq c_{j'}^{(\mathbf{h})} \Longleftrightarrow j' \leq j \text{ and } \mathbf{h}(j') = \mathbf{h}(j). \tag*{(14)}

Thus, two indices jj and j′j' with distinct values of h\mathbf{h} have corresponding cj(h)c_{j}^{(\mathbf{h})} and cj′(h)c_{j'}^{(\mathbf{h})} which are incomparable in Bruhat order, and for a collection of (consecutive) indices with the same value of h\mathbf{h}, it is the largest value of the index which yields the Bruhat-smallest cj(h)c_{j}^{(\mathbf{h})}. The largest such indices are precisely the corners j∈D(h)j \in D(\mathbf{h}). This proves the lemma. □

Example 4.4 (Example 3.2, continued). By Equation (14), cj≤cj′c_j \leq c_{j'} if and only if j′≤jj' \leq j and h(j′)=h(j)\mathbf{h}(j') = \mathbf{h}(j). The only such pairs with j≠j′j \neq j' are c3<c2c_3 < c_2 and c5<c4c_5 < c_4. For the first pair, the tableaux

c3=(236145):c2=(263145):123461236236232123461236236262\begin{array}{cc} c_3=(236145): & c_2=(263145): \\[-2pt] \begin{array}{ccccc} 1&2&3&4&6\\ 1&2&3&6\\ 2&3&6\\ 2&\mathbf{3}\\ 2 \end{array}\qquad\qquad & \begin{array}{ccccc} 1&2&3&4&6\\ 1&2&3&6\\ 2&3&6\\ 2&\mathbf{6}\\ 2 \end{array} \end{array}

differ only in the fourth row. For example, c1≰c3c_1 \nleq c_3 because c3−1(6)=3>1c_3^{-1}(6)=3>1, and c3≰c1c_3 \nleq c_1 because c1−1(1)=3<h(3)+1c_1^{-1}(1)=3<\mathbf{h}(3)+1. All other pairs are incomparable. Hence the Bruhat-minimal elements of Fh\mathcal{F}_{\mathbf{h}} are c1,c3,c5c_1,c_3,c_5, indexed by the corners D(h)={1,3,5}D(\mathbf{h})=\{1,3,5\}, in accordance with Lemma 4.3. ◇

We saw in the above lemma that if h(j)=h(j+1)<n\mathbf{h}(j)=\mathbf{h}(j+1)<n then cj(h)≥cj+1(h)c_j^{(\mathbf{h})}\geq c_{j+1}^{(\mathbf{h})}. There is a corresponding statement about the associated Schubert polynomials. Recall that e1,…,ene_1,\ldots,e_n denote the elementary symmetric polynomials in the variables x1,…,xnx_1,\ldots,x_n. Before proceeding, it is useful to recall the following relation between complete homogeneous symmetric polynomials and the elementary symmetric polynomials. This is standard but we include it here for the convenience of the reader. Let HA(t)\mathbf{H}_A(t) denote the generating function for the complete homogeneous symmetric functions HdH_d on a set of variables AA, and similarly EA(t)\mathbf{E}_A(t) for the elementary symmetric functions ede_d. (So for example if A={x1,…,xℓ}A=\{x_1,\ldots,x_\ell\} for some ℓ>0\ell>0, then HA(t)=∏i=1ℓ11−xit\mathbf{H}_A(t)=\prod_{i=1}^{\ell}\frac{1}{1-x_i t} and EA(t)=∏i=1ℓ(1+xit)\mathbf{E}_A(t)=\prod_{i=1}^{\ell}(1+x_i t).)

Lemma 4.5. Let n≥2n\geq2. Let j,ℓj,\ell be integers such that 1≤j≤n−21\leq j\leq n-2 and j<ℓ<nj<\ell<n. Then x1⋯xℓ⋅Hn−j−1(x1,…,xj+1)x_1\cdots x_\ell\cdot H_{n-j-1}(x_1,\ldots,x_{j+1}) lies in the ideal ⟨e1,…,en⟩\langle e_1,\ldots,e_n\rangle.

Proof. The argument uses well-known identities in symmetric functions; we sketch the argument. By the definitions above, we have the identity on generating functions HA(t)EA(−t)=1\mathbf{H}_A(t)\mathbf{E}_A(-t)=1. From this we immediately obtain H{x1,…,xj+1}(t)E{x1,…,xn}(−t)=E{xj+2,…,xn}(−t)\mathbf{H}_{\{x_1,\ldots,x_{j+1}\}}(t)\mathbf{E}_{\{x_1,\ldots,x_n\}}(-t)=\mathbf{E}_{\{x_{j+2},\ldots,x_n\}}(-t). Now compare the coefficient of tn−j−1t^{n-j-1} on both sides to obtain

∑k=0n−j−1(−1)kek⋅Hn−j−1−k(x1,…,xj+1)=(−1)n−j−1xj+2⋯xn.\sum_{k=0}^{n-j-1}(-1)^k e_k\cdot H_{n-j-1-k}(x_1,\ldots,x_{j+1})=(-1)^{n-j-1}x_{j+2}\cdots x_n.

Rearranging to isolate the term corresponding to k=0k=0 on the LHS, multiplying by x1⋯xℓx_1\cdots x_\ell, and using x1⋯xℓ⋅xj+2⋯xn=xj+2⋯xℓ⋅enx_1\cdots x_\ell\cdot x_{j+2}\cdots x_n=x_{j+2}\cdots x_\ell\cdot e_n (where we use ℓ>j\ell>j, and we interpret the notation xj+2⋯xℓx_{j+2}\cdots x_\ell as equal to 11 if j+2>ℓj+2>\ell), gives

x1⋯xℓ⋅Hn−j−1(x1,…,xj+1)=(−1)n−j−1xj+2⋯xℓ⋅en+∑k=0n−j−2(−1)k⋅x1⋯xℓ⋅ek+1⋅Hn−j−2−k(x1,…,xj+1).\begin{aligned} x_1\cdots x_\ell\cdot H_{n-j-1}(x_1,\ldots,x_{j+1}) ={}&(-1)^{n-j-1}x_{j+2}\cdots x_\ell\cdot e_n\\ &+\sum_{k=0}^{n-j-2}(-1)^k\cdot x_1\cdots x_\ell\cdot e_{k+1}\cdot H_{n-j-2-k}(x_1,\ldots,x_{j+1}). \end{aligned}

Every term on the right-hand side lies in the ideal ⟨e1,…,en⟩\langle e_1,\ldots,e_n\rangle, so the conclusion holds. □

We can now prove the last remaining step for the proof of Theorem B, which is related to non-corners of h\mathbf{h}.

Proposition 4.6. In the setting of Lemma 4.1, suppose j∈[n−1]j \in[n-1] is such that h(j)=h(j+1)<n\mathbf{h}(j)=\mathbf{h}(j+1)<n. Then

Scj(h)≡−xj+1Scj+1(h)(mod⟨e1,…,en⟩).\mathfrak{S}_{c_j^{(\mathbf{h})}} \equiv-x_{j+1}\mathfrak{S}_{c_{j+1}^{(\mathbf{h})}} \pmod{\langle e_1,\ldots,e_n\rangle}.

Proof. We begin by noting that the assumption h(j+1)<n\mathbf{h}(j+1)<n implies that j+1<nj+1<n and hence j≤n−2j\le n-2, so n−j−2≥0n-j-2\ge0. Moreover, since h(j+1)≥j+1\mathbf{h}(j+1)\ge j+1 by assumption on Hessenberg functions, we know j<h(j)=h(j+1)j<\mathbf{h}(j)=\mathbf{h}(j+1). Next observe that, for any positive degree dd, we have the relation Hd(x1,…,xj)=Hd(x1,…,xj+1)−xj+1Hd−1(x1,…,xj+1)H_d(x_1,\ldots,x_j)=H_d(x_1,\ldots,x_{j+1})-x_{j+1}H_{d-1}(x_1,\ldots,x_{j+1}). Using this with d=n−j−1≥1d=n-j-1\ge1, we have

Scj(h)=Ph(j)Hn−j−1(x1,…,xj)by Lemma 3.3=Ph(j)Hn−j−1(x1,…,xj+1)−xj+1Ph(j)Hn−j−2(x1,…,xj+1)=Ph(j)Hn−j−1(x1,…,xj+1)−xj+1Ph(j+1)Hn−j−2(x1,…,xj+1)since h(j)=h(j+1)=Ph(j)Hn−j−1(x1,…,xj+1)−xj+1Scj+1(h)by Lemma 3.3.\begin{aligned} \mathfrak{S}_{c_j^{(\mathbf{h})}} &= P_{\mathbf{h}(j)}H_{n-j-1}(x_1,\ldots,x_j) && \text{by Lemma } 3.3\\ &=P_{\mathbf{h}(j)}H_{n-j-1}(x_1,\ldots,x_{j+1})-x_{j+1}P_{\mathbf{h}(j)}H_{n-j-2}(x_1,\ldots,x_{j+1})\\ &=P_{\mathbf{h}(j)}H_{n-j-1}(x_1,\ldots,x_{j+1})-x_{j+1}P_{\mathbf{h}(j+1)}H_{n-j-2}(x_1,\ldots,x_{j+1}) && \text{since } \mathbf{h}(j)=\mathbf{h}(j+1)\\ &=P_{\mathbf{h}(j)}H_{n-j-1}(x_1,\ldots,x_{j+1})-x_{j+1}\mathfrak{S}_{c_{j+1}^{(\mathbf{h})}} && \text{by Lemma } 3.3. \end{aligned}

To conclude the argument it suffices to see that Ph(j)⋅Hn−j−1(x1,…,xj+1)P_{\mathbf{h}(j)}\cdot H_{n-j-1}(x_1,\ldots,x_{j+1}) is contained in the ideal ⟨e1,…,en⟩\langle e_1,\ldots,e_n\rangle. Since h(j)>j\mathbf{h}(j)>j as noted above, this follows from Lemma 4.5. This concludes the proof. □

Example 4.7 (Example 3.4, continued). For the running example, Proposition 4.6 applies to j=2j=2 and j=4j=4, since h=(2,3,3,5,5,6)\mathbf{h}=(2,3,3,5,5,6) and thus h(2)=h(3)=3<6\mathbf{h}(2)=\mathbf{h}(3)=3<6 and h(4)=h(5)=5<6\mathbf{h}(4)=\mathbf{h}(5)=5<6. For j=2j=2, we follow the argument in the proof and observe H3(x1,x2)=H3(x1,x2,x3)−x3H2(x1,x2,x3)H_3(x_1,x_2)=H_3(x_1,x_2,x_3)-x_3H_2(x_1,x_2,x_3), which gives Sc2=P3H3(x1,x2,x3)−x3Sc3\mathfrak{S}_{c_2}=P_3H_3(x_1,x_2,x_3)-x_3\mathfrak{S}_{c_3}. The first term lies in ⟨e1,…,e6⟩\langle e_1,\ldots,e_6\rangle by Lemma 4.5 applied to the choice (j,ℓ)=(2,3)(j,\ell)=(2,3). Indeed, writing Hd=Hd(x1,x2,x3)H_d=H_d(x_1,x_2,x_3), we get

P3H3=P3(e1H2−e2H1+e3)−e6,P_3H_3=P_3(e_1H_2-e_2H_1+e_3)-e_6,

which visibly lies in the ideal ⟨e1,⋯ ,e6⟩\langle e_1,\cdots,e_6\rangle. Hence Sc2≡−x3Sc3(mod⟨e1,…,e6⟩)\mathfrak{S}_{c_2}\equiv-x_3\mathfrak{S}_{c_3}\pmod{\langle e_1,\ldots,e_6\rangle}.

For j=4j=4 we have H1(x1,x2,x3,x4)=(x1+⋯+x5)−x5=(x1+⋯+x6)−x6−x5H_1(x_1,x_2,x_3,x_4)=(x_1+\cdots+x_5)-x_5=(x_1+\cdots+x_6)-x_6-x_5 and we also know Sc5=P5\mathfrak{S}_{c_5}=P_5 and Sc4=P5⋅H1(x1,x2,x3,x4)\mathfrak{S}_{c_4}=P_5\cdot H_1(x_1,x_2,x_3,x_4). Observe also that P5⋅x6=P6=e6P_5\cdot x_6=P_6=e_6. Putting these together yields

Sc4=P5(e1−x6)−x5P5=P5e1−e6−x5Sc5≡−x5Sc5(mod⟨e1,…,e6⟩).\mathfrak{S}_{c_4}=P_5(e_1-x_6)-x_5P_5=P_5e_1-e_6-x_5\mathfrak{S}_{c_5}\equiv-x_5\mathfrak{S}_{c_5}\pmod{\langle e_1,\ldots,e_6\rangle}.

The above computation follows the argument made in the proof of Proposition 4.6. However, as already observed in Remark 3.7, in the case of j=4j=4 we can view the relation in a different way. Since Sc5=P5\mathfrak{S}_{c_5}=P_5 and Sc4=P5⋅H1(x1,x2,x3,x4)\mathfrak{S}_{c_4}=P_5\cdot H_1(x_1,x_2,x_3,x_4), we can simply observe that Sc4\mathfrak{S}_{c_4} is a multiple of Sc5\mathfrak{S}_{c_5}, and hence is a redundant generator.

The remaining generators are indexed by the corners D(h)={1,3,5}D(\mathbf{h})=\{1,3,5\}, and c1,c3,c5c_1,c_3,c_5 are exactly the Bruhat-minimal forbidden permutations (Lemma 4.3). Theorem B therefore gives

H∗(Hess⁡(E1,6,h);Q)≅H∗(Fℓags(C6);Q)/⟨σc1,σc3,σc5⟩,H^*(\operatorname{Hess}(E_{1,6},\mathbf{h});\mathbb{Q})\cong H^*(\mathcal{F}\ell ags(\mathbb{C}^6);\mathbb{Q})/\langle\sigma_{c_1},\sigma_{c_3},\sigma_{c_5}\rangle,

and no generating set of this kernel has fewer than three elements, compared with the 288 classes of the Abe–Crooks additive basis. ◇

We now show that the classes indexed by the Bruhat-minimal forbidden permutations generate JhJ_{\mathbf{h}}, thus proving the second main result of this paper.

Proof of Theorem B. We begin by recalling what we know in the context of the polynomial ring Q[x1,…,xn]\mathbb{Q}[x_1,\ldots,x_n], building on the fact that H∗(Fℓags(Cn);Q)≅Q[x1,…,xn]/⟨e1,…,en⟩H^*(\mathcal{F}\ell ags(\mathbb{C}^n);\mathbb{Q})\cong\mathbb{Q}[x_1,\ldots,x_n]/\langle e_1,\ldots,e_n\rangle. First note that the proof of Theorem A also holds verbatim with Q\mathbb{Q} coefficients in place of Z\mathbb{Z} coefficients. With this said, from here onwards we interpret all statements to be in Q[x1,…,xn]\mathbb{Q}[x_1,\ldots,x_n]. By Theorem A(2), the ideal Jh=⟨Sw:w∈Fh⟩J_{\mathbf{h}}=\langle\mathfrak{S}_w:w\in\mathcal{F}_{\mathbf{h}}\rangle is generated by the polynomials Scj(h)\mathfrak{S}_{c_j^{(\mathbf{h})}} with h(j)<n\mathbf{h}(j)<n. Since σw\sigma_w is represented by Sw\mathfrak{S}_w and, by Theorem 2.2, JhJ_{\mathbf{h}} is the ideal generated by {σw:w∈Fh}\{\sigma_w : w \in\mathcal{F}_h\}, it follows that JhJ_h is generated by the classes σcj(h)\sigma_{c_j^{(h)}} with h(j)<n\mathrm{h}(j) < n. Moreover, Proposition 4.6 gives

σcj(h)=−xj+1σcj+1(h)∈H∗(Flags(Cn);Q)if h(j)=h(j+1)<n\sigma_{c_j^{(h)}}=-x_{j+1}\sigma_{c_{j+1}^{(h)}}\in H^*(\mathcal{F}\mathrm{lags}(\mathbb{C}^n);\mathbb{Q})\quad\text{if }\mathrm{h}(j)=\mathrm{h}(j+1)<n

where by slight abuse of notation, xj+1x_{j+1} also denotes its image in H∗(Flags(Cn);Q)H^*(\mathcal{F}\mathrm{lags}(\mathbb{C}^n);\mathbb{Q}). By the definition of the set D(h)D(\mathrm{h}) given in (1) it follows that

Jh=⟨σcj(h):j∈D(h)⟩.J_h=\langle\sigma_{c_j^{(h)}}:j\in D(\mathrm{h})\rangle.

We next prove that every ideal-generating set has at least ∣D(h)∣|D(\mathrm{h})| elements. First recall that the Schubert classes form a Q\mathbb{Q}-additive basis for H∗(Flags(Cn);Q)H^*(\mathcal{F}\mathrm{lags}(\mathbb{C}^n);\mathbb{Q}) and that {σw:w∈Fh}\{\sigma_w:w\in\mathcal{F}_h\} form a Q\mathbb{Q}-additive basis for JhJ_h. Let m\mathfrak{m} denote the ideal of positive-degree elements in H∗(Flags(Cn);Q)H^*(\mathcal{F}\mathrm{lags}(\mathbb{C}^n);\mathbb{Q}). The subspace m⋅Jh\mathfrak{m}\cdot J_h is spanned by products of the form σuσv\sigma_u\sigma_v for ℓ(u)>0\ell(u)>0 and v∈Fhv\in\mathcal{F}_h. By iterating the variable-multiplication formula [17 (4.15′)], and using commutativity, we know σuσv=∑w≥u,vcu,vwσw\sigma_u\sigma_v=\sum_{w\geq u,v}c_{u,v}^w\sigma_w for cu,vw∈Zc_{u,v}^w\in\mathbb{Z} and if cu,vw≠0c_{u,v}^w\ne0, then ℓ(w)=ℓ(u)+ℓ(v)\ell(w)=\ell(u)+\ell(v). In particular, any σw\sigma_w appearing on the RHS must have w≥vw\geq v, and since ℓ(u)>0\ell(u)>0 this implies ℓ(w)>ℓ(v)\ell(w)>\ell(v) and hence the inequality is strict, i.e., w>vw>v. Since {σz:z∈Fh}\{\sigma_z:z\in\mathcal{F}_h\} is an additive basis of JhJ_h by Theorem 2.2 we know w∈Fhw\in\mathcal{F}_h, and by Lemma 4.3 and the fact that w>vw>v we conclude that ww cannot be a minimal Bruhat element. It follows that, for any element ff of mJh\mathfrak{m}J_h, the basis expansion of ff in terms of the basis {σz:z∈Fh}\{\sigma_z:z\in\mathcal{F}_h\} has a 0 coefficient in front of all the Bruhat-minimal elements {σcj(h):j∈D(h)}\{\sigma_{c_j^{(h)}}:j\in D(\mathrm{h})\}. This in turn implies that the image of {σcj(h):j∈D(h)}\{\sigma_{c_j^{(h)}}:j\in D(\mathrm{h})\} is linearly independent in the Q\mathbb{Q}-vector space Jh/mJhJ_h/\mathfrak{m}J_h. On the other hand, they must also span this vector space, since we just saw above that they generate JhJ_h. It follows that

dim⁡Q(Jh/mJh)=∣D(h)∣.\dim_{\mathbb{Q}}(J_h/\mathfrak{m}J_h)=|D(\mathrm{h})|.

Observe that for any set SS that generates the ideal JhJ_h, the image of SS in Jh/mJhJ_h/\mathfrak{m}J_h must Q\mathbb{Q}-span this quotient, since H∗(Flags(Cn);Q)/m≅QH^*(\mathcal{F}\mathrm{lags}(\mathbb{C}^n);\mathbb{Q})/\mathfrak{m}\cong\mathbb{Q}. Hence every such SS must have at least ∣D(h)∣|D(\mathrm{h})| many elements. The set {σcj(h):j∈D(h)}\{\sigma_{c_j^{(h)}}:j\in D(\mathrm{h})\} has precisely ∣D(h)∣|D(\mathrm{h})| elements and hence it is minimal, i.e., there is no proper subset which generates JhJ_h.

Finally, the asserted presentation of the cohomology ring follows from Theorem 2.2. □

Acknowledgments

We thank Peter Crooks for helpful conversations, interest in this work, and ongoing encouragement. We additionally thank Colleen Robichaux, Anna Weigandt, and Oliver Pechenik for sharing their expertise on Schubert polynomials and pipe dreams in the early stages of our investigations. The set of generators {cj(h)}\{c_j^{(h)}\} as discussed in this manuscript was first found experimentally through computations with SAGE, as part of Nardone’s NSERC USRA project in Summer 2026 while working under the supervision of Harada. GG was funded by SNF Postdoc.Mobility grant P500PT_225436. MH was supported by a Canada Research Chair Award (Tier 1) and NSERC Discovery Grant 2019-06567. JR was supported by NSERC Discovery Grant 2023-04800. AVT was supported by NSERC Discovery Grant 2024-05299.

Declaration of AI Use

The authors made use of Claude Opus 5.5 and GPT Astra 6 as part of their investigation of the minimal generators. The authors reviewed all AI output and wrote the paper; they take full responsibility for the content of this paper.

(G. Gaggero) Department of Mathematics and Statistics McMaster University, Hamilton, ON L8S 4L8, Canada

Email address: [email protected]

(M. Harada) Department of Mathematics and Statistics McMaster University, Hamilton, ON L8S 4L8, Canada

Email address: [email protected]

(H. Nardone) Department of Mathematics, McGill University, Montreal, QC H3A 0B9, Canada

Email address: [email protected]

(B. Nasserden) Department of Mathematics and Statistics McMaster University, Hamilton, ON L8S 4L8, Canada

Email address: [email protected]

(J. Rajchgot) Department of Mathematics and Statistics McMaster University, Hamilton, ON L8S 4L8, Canada

Email address: [email protected]

(L. Seccia) Department of Mathematics and Statistics McMaster University, Hamilton, ON L8S 4L8, Canada

Email address: [email protected]

(A. Van Tuyl) Department of Mathematics and Statistics McMaster University, Hamilton, ON L8S 4L8, Canada

Email address: [email protected]

References

  1. [1]Hiraku Abe and Peter Crooks, Hessenberg varieties for the minimal nilpotent orbit, Pure Appl. Math. Q. 12 (2016), no. 2, 183–223. MR 3767215arxiv.org/abs/1510.02436
  2. [2]Hiraku Abe, Megumi Harada, Tatsuya Horiguchi, and Mikiya Masuda, The cohomology rings of regular nilpotent Hessenberg varieties in Lie type A, Int. Math. Res. Not. IMRN (2019), no. 17, 5316–5388. MR 4001030
  3. [3]Hiraku Abe, Tatsuya Horiguchi, and Mikiya Masuda, The cohomology rings of regular semisimple Hessenberg varieties for h = (h(1), n, . . . , n), J. Comb. 10 (2019), no. 1, 27–59. MR 3890915
  4. [4]Ashleigh Adams-Jessmore, Sean Griffin, and Tim Ryan, Face rings and the cohomology of unions of Schubert varieties, 2026, arXiv:2611.15065.
  5. [5]Nantel Bergeron and Sara Billey, RC-graphs and Schubert polynomials, Experiment. Math. 2 (1993), no. 4, 257–269. MR 1281474DOI
  6. [6]Anders Björner and Francesco Brenti, Combinatorics of Coxeter groups, Graduate Texts in Mathematics, vol. 231, Springer, New York, 2005. MR 2133266DOI
  7. [7]Armand Borel, Sur la cohomologie des espaces fibrés principaux et des espaces homogènes de groupes de Lie compacts, Ann. of Math. (2) 57 (1953), 115–207. MR 51508DOI
  8. [8]David A. Cox, John Little, and Donal O’Shea, Ideals, varieties, and algorithms: An introduction to computational algebraic geometry and commutative algebra, fourth ed., Undergraduate Texts in Mathematics, Springer, Cham, 2015. MR 3330490
  9. [9]Corrado De Concini and Claudio Procesi, Symmetric functions, conjugacy classes and the flag variety, Invent. Math. 64 (1981), no. 2, 203–219. MR 629470DOI
  10. [10]Laura Escobar, Martha Precup, and John Sheshanshian, Hessenberg varieties of codimension one in the flag variety, Canad. Math. Bull. (2026). MR 5066163
  11. [11]A. M. Garsia and C. Procesi, On certain graded S_n-modules and the q-Kostka polynomials, Adv. Math. 94 (1992), no. 1, 82–138. MR 1168926DOI
  12. [12]V. Gasharov and V. Reiner, Cohomology of smooth Schubert varieties in partial flag manifolds, J. London Math. Soc. (2) 66 (2002), no. 3, 550–562. MR 1934291DOI
  13. [13]Rebecca Goldin and Martha Precup, A flat family of matrix Hessenberg schemes over the minimal sheet, J. Algebra 692 (2026), 123–172. MR 5002214
  14. [14]Rebecca Goldin and Julianna Tymoczko, Which Hessenberg varieties are GKM?, Pure Appl. Math. Q. 19 (2023), no. 4, 1899–1942. MR 4671386arxiv.org/abs/2301.09741
  15. [15]Erik Ikson, Martha Precup, and Edward Richmond (eds.), Handbook of combinatorial algebraic geometry—subvarieties of the flag variety, Discrete Mathematics and its Applications (Boca Raton), CRC Press, Boca Raton, FL, 2026. MR 5129420
  16. [16]Alain Lascoux and Marcel-Paul Schützenberger, Polynômes de Schubert, C. R. Acad. Sci. Paris Sér. I Math. 294 (1982), no. 13, 447–450. MR 660739
  17. [17]I. G. Macdonald, Notes on Schubert polynomials, Publications du LaCIM, vol. 6, Université du Québec à Montréal, Montréal, 1991.
  18. [18]Aba Mbirika and Julianna Tymoczko, Generalizing Tanisaki’s ideal via ideals of truncated symmetric functions, J. Algebraic Combin. 37 (2013), no. 1, 167–199. MR 3016306
  19. [19]Victor Reiner, Alexander Woo, and Alexander Yong, Presenting the cohomology of a Schubert variety, Trans. Amer. Math. Soc. 363 (2011), no. 1, 521–543. MR 2719692arxiv.org/abs/0809.2981
  20. [20]Avery St. Dizier and Alexander Yong, Presenting the cohomology of a Schubert variety: proof of the minimality conjecture, J. Lond. Math. Soc. (2) 109 (2024), no. 1, Paper No. e12832, 22. MR 4680222DOI
  21. [21]Toshiyuki Tanisaki, Defining ideals of the closures of the conjugacy classes and representations of the Weyl groups, Tohoku Math. J. (2) 34 (1982), no. 4, 575–585. MR 685425DOI

Paper details

vlnc:W_bw85apke
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

Claude Opus 5.5 (bare), GPT-6 Astra (bare)

View statement
Looking for related papers…
Contents