betaFeedback

WorksMathematics

Symbolic Powers and Asymptotic Invariants of Invariant Ideals in Determinantal Rings

Abstract

Let AA be the coordinate ring of a bounded-rank locus of generic, symmetric, alternating, or Hankel matrices. Thus AA is the quotient of a polynomial ring by a generic determinantal, symmetric determinantal, generic Pfaffian, or Hankel determinantal ideal. In each case there is a nested family of rank primes p1⊃p2⊃⋯⊃pq{\mathfrak p}_1\supset {\mathfrak p}_2\supset\cdots\supset{\mathfrak p}_q and a standard-monomial basis whose shapes are diagrams with parts at most qq. We isolate the common properties of these four straightening theories and use them to give a uniform treatment of products Dσ=pσ1⋯pσsD_σ={\mathfrak p}_{σ_1}\cdots{\mathfrak p}_{σ_s} in the quotient ring AA, extending analogous results over suitable polynomial rings. In all four cases the shape σσ must have parts of size at most qq. We determine their primary decompositions and associated primes in the rank-qq quotient, compute all symbolic powers with respect to the full set of associated primes, and prove that their symbolic Rees algebras are finitely generated. In the generic, symmetric, and alternating cases the respective symbolic Rees algebras are also normal, and Cohen–Macaulay. We also determine the asymptotic resurgence ρ^\widehatρ and Waldschmidt constants of these ideals and prove they are related by ρ^(Dσ)=α(Dσ)/α^(Dσ)\widehatρ(D_σ)=α(D_σ)/\widehatα(D_σ).

Introduction

Determinantal varieties exhibit a natural filtration by closed subvarieties given by more restrictive rank conditions. We study the ideals defining these subvarieties, and products of such ideals, relating the geometry of rank loci with representation theory and commutative algebra. The ordinary powers of these ideals record products of equations, while their symbolic powers retain the primary components of the relevant associated primes. To understand the difference between the two constructions one is led to consider primary decomposition, minimal degrees of equations with prescribed symbolic vanishing, and containment between symbolic and ordinary powers. These questions have been studied mainly within regular rings, but determinantal rings provide a natural setting in which to study such questions inside singular ambient spaces.

These problems have a long history over polynomial rings. De Concini, Eisenbud, and Procesi developed a shape calculus for generic determinantal ideals and their products, including descriptions

2020 Mathematics Subject Classification. 13A15, 13F20, 14M12, 14N20, 20G05.

Key words and phrases. determinantal ring, invariant ideal, symbolic power, symbolic Rees algebra, Waldschmidt constant, asymptotic resurgence, standard monomial theory. of primary decompositions, symbolic powers, and integral closures [13]; see also Bruns–Vetter [9]. Abeasis and Abeasis–Del Fra developed the corresponding invariant-ideal and tableau theories for symmetric minors and Pfaffians [1, 2]. Similar descriptions for catalecticant rank ideals and their products are given by Conca’s straightening theory for Hankel matrices [11], and the subsequent treatment of extended Hankel matrices by Nam [26]. All these theories share a common feature: membership in the relevant ideals is controlled by numerical functions on the shapes of standard monomials, denoted γj\gamma_j below. For the generic, symmetric, and alternating settings, Henriques–Varbaro organized this common structure in their study of test, multiplier, and invariant ideals in [21].

The comparison of symbolic and ordinary powers has also yielded numerical invariants measuring their asymptotic behavior. While the Waldschmidt constant reflects the growth of the initial degree sequence of the symbolic powers, resurgence, introduced by Bocci–Harbourne [5], measures the sharpness of symbolic-power containments. Guardo–Harbourne–Van Tuyl have defined a similar invariant known as asymptotic resurgence in [19]. This has been connected to integral closures, Rees valuations, and skew Waldschmidt constants by DiPasquale–Francisco–Mermin–Schweig in [15]. For classical rank ideals in polynomial rings, Kumar–Mukundan [25] calculated initial degrees and resurgence invariants and analyzed containments of symbolic powers in products of ordinary powers with additional powers of the homogeneous maximal ideal. More recently, Bisui–Seceleanu [4] determined symbolic powers and asymptotic invariants for arbitrary GL-invariant ideals of a generic matrix and proved finite generation of their generalized symbolic Rees algebras.

The above-mentioned results for polynomial rings motivate us to study the analogous problems in the coordinate ring of a bounded-rank variety. In this case, the ambient ring is a determinantal quotient itself, and products of rank prime ideals encode further rank conditions within matrices of bounded rank. Passing to this quotient raises two interrelated difficulties. Imposing the rank bound given by the ambient ring eliminates standard monomials with large factors and can make components of the ambient primary decomposition redundant. This changes both the symbolic filtration and its asymptotic invariants. Our aim is to describe this change explicitly and uniformly in four settings:

  1. (i) a generic matrix modulo its (q+1)(q+1)-minors;

  2. (ii) a symmetric matrix modulo its (q+1)(q+1)-minors;

  3. (iii) an alternating matrix modulo its 2(q+1)2(q+1)-Pfaffians;

  4. (iv) a Hankel matrix modulo its (q+1)(q+1)-minors.

In each case let AA denote the quotient ring and

p1⊃p2⊃⋯⊃pq\mathfrak{p}_1 \supset\mathfrak{p}_2 \supset\cdots\supset\mathfrak{p}_q

its chain of nonzero rank primes, with p1=A+\mathfrak{p}_1=A_+. We work over an algebraically closed field of characteristic zero. For a Young diagram σ=(σ1≥⋯≥σs≥1)\sigma=(\sigma_1\geq\cdots\geq\sigma_s\geq1) with σ1≤q\sigma_1\leq q, put

Dσ=pσ1⋯pσs,γj(σ)=∑i=1s(σi−j+1)+,∣σ∣=∑i=1sσi,D_\sigma=\mathfrak{p}_{\sigma_1}\cdots\mathfrak{p}_{\sigma_s},\qquad \gamma_j(\sigma)=\sum_{i=1}^{s}(\sigma_i-j+1)_+,\qquad |\sigma|=\sum_{i=1}^{s}\sigma_i,

where (u)+=max⁡{u,0}(u)_+=\max\{u,0\}. Our results concern the products DσD_\sigma of rank primes.

In the following theorem we summarize the main primary-decomposition, symbolic-power, and containment results of our paper. Once the formula in Theorem 1.0.1(i) has been established using the standard-monomial theory pertaining to each setting, the remaining parts of Theorem 1.0.1 then follow uniformly from the same combinatorial argument applied to all four rings.

Theorem 1.0.0.1. With the notation above, set

β=max⁡{1,∣σ∣−q(s−1)},h=q−β+1,m=A+.\beta=\max\{1,|\sigma|-q(s-1)\},\qquad h=q-\beta+1,\qquad\mathfrak{m}=A_+.

Then the following statements hold.

(i) There is a primary decomposition

Dσ=⋂j=1σ1pj(γj(σ)).D_{\sigma}=\bigcap_{j=1}^{\sigma_1}\mathfrak{p}_{j}^{(\gamma_j(\sigma))}.

In particular, pβ\mathfrak{p}_{\beta} is the unique inclusion-maximal associated prime of A/DσA/D_{\sigma}.

(ii) For integers m≥1m\ge1, the symbolic powers with respect to the full set of associated primes are

Dσ(m)=⋂j=βσ1pj(mγj(σ)).D_{\sigma}^{(m)}=\bigcap_{j=\beta}^{\sigma_1}\mathfrak{p}_{j}^{(m\gamma_j(\sigma))}.

(iii) For every integer m≥1m\ge1, the initial degree of the respective symbolic power is

α(Dσ(m))=mγβ(σ)+(β−1)⌈mγβ(σ)h⌉.\alpha(D_{\sigma}^{(m)})=m\gamma_{\beta}(\sigma)+(\beta-1)\left\lceil\frac{m\gamma_{\beta}(\sigma)}{h}\right\rceil.

For integers m,r≥1m,r\ge1 and N≥0N\ge0 one has

Dσ(m)⊆mNDσr⟺α(Dσ(m))≥N+r∣σ∣.D_{\sigma}^{(m)}\subseteq\mathfrak{m}^{N}D_{\sigma}^{r}\quad\Longleftrightarrow\quad\alpha(D_{\sigma}^{(m)})\ge N+r|\sigma|.

(iv) The Waldschmidt constant, resurgence, and asymptotic resurgence respectively satisfy

α^(Dσ)=qγβ(σ)h,ρ(Dσ)=ρ^(Dσ)=∣σ∣hqγβ(σ)=α(Dσ)α^(Dσ).\widehat{\alpha}(D_{\sigma})=\frac{q\gamma_{\beta}(\sigma)}{h},\qquad \rho(D_{\sigma})=\widehat{\rho}(D_{\sigma})=\frac{|\sigma|h}{q\gamma_{\beta}(\sigma)} =\frac{\alpha(D_{\sigma})}{\widehat{\alpha}(D_{\sigma})}.

(v) The symbolic Rees algebra

Rs(Dσ)=⨁m≥0Dσ(m)Tm\mathcal{R}_{s}(D_{\sigma})=\bigoplus_{m\ge0}D_{\sigma}^{(m)}T^{m}

is a finitely generated AA-algebra. In the generic, symmetric, and alternating settings, it is a normal domain with rational singularities; in particular, it is Cohen–Macaulay.

The finite-generation assertion in part (v) is constructive. We write symbolic powers as sums of ordinary products of rank ideals in Theorem 4.2.4. The multiplicities of the factors and the symbolic exponents are encoded in a lattice-point monoid corresponding to a rational polyhedral cone, whose Hilbert basis yields finite sets of algebra generators. Examples show that generators of higher symbolic degree can be necessary. Normality of the three classical settings follows from a description of the symbolic Rees ring as an intersection of valuation rings. A multiplicity-free action of an appropriate group then yields rational singularities in each classical setting through the theory of affine spherical varieties.

The paper begins by recalling some background concerning symbolic powers and related invariants in § 2 and establishing the common standard-monomial input. This includes results on symbolic-power descent in § 3. We compute the irredundant primary decompositions and give two explicit formulas for the symbolic powers of DσD_{\sigma} in § 4. We further construct generators for their symbolic Rees algebras in § 5. Finally we calculate the exact initial degrees in § 6, which we use to resolve the containment problem, and to determine the resulting asymptotic invariants.

Acknowledgements. A.M. was partially supported by a Simons Dissertation Fellowship and NSF grant DMS-2349623. K.M.S. was partially supported by NSF grants DMS-2236983, DMS-2601940, and DMS-2044833. A.S. was partially supported by DMS-2401482. S.S. was partially supported by DMS-2342256.

AI disclosure. Generative AI was used to assist with literature searches.

Symbolic powers and asymptotic invariants

Throughout the paper, kk is an algebraically closed field of characteristic zero and all rings are standard graded kk-domains. The characteristic-zero hypothesis is used for the classification of invariant ideals. Several of the straightening and primary decomposition statements hold in greater generality.

Symbolic powers

Let II be a homogeneous ideal in a Noetherian domain AA. We use symbolic powers with respect to the full set of associated primes:

I(m)=⋂p∈Ass⁡A(A/I)(ImAp∩A).(1)I^{(m)}=\bigcap_{\mathfrak{p}\in\operatorname{Ass}_{A}(A/I)}\left(I^{m}A_{\mathfrak{p}}\cap A\right). \tag*{(1)}

In (2.1.1), it suffices to intersect over the inclusion-maximal elements of Ass⁡A(A/I)\operatorname{Ass}_{A}(A/I). In particular, if the associated primes form a chain with maximal element p\mathfrak{p}, which will be the case for our ideals of interest, then I(m)=ImAp∩AI^{(m)}=I^{m}A_{\mathfrak{p}}\cap A.

For a nonzero homogeneous ideal II, let

α(I)=min⁡{d:Id≠0}.\alpha(I)=\min\{d:I_d\ne0\}.

The Waldschmidt constant of II is

α^(I)=lim⁡m→∞α(I(m))m=inf⁡m≥1α(I(m))m.\widehat{\alpha}(I)=\lim_{m\to\infty}\frac{\alpha(I^{(m)})}{m}=\inf_{m\ge1}\frac{\alpha(I^{(m)})}{m}.

The equality follows from subadditivity; see for example [16 Lemma 3.3 and Definition 3.5]. The resurgence and asymptotic resurgence, introduced in this context in [5] and [19], respectively, are

ρ(I)=sup⁡{mr:I(m)⊈Ir}\rho(I)=\sup\left\{\frac{m}{r}:I^{(m)}\not\subseteq I^r\right\}

and

ρ^(I)=sup⁡{mr:I(mt)⊈Irt for all t≫0}.(2)\widehat{\rho}(I)=\sup\left\{\frac{m}{r}:I^{(mt)}\not\subseteq I^{rt}\text{ for all }t\gg0\right\}. \tag*{(2)}

The asymptotic resurgence is controlled by the rational cones of the ordinary and symbolic Rees algebras; see also [15]. This viewpoint will be used rather than an explicit list of Rees valuations.

Young diagrams and the γ\gamma-functions

A Young diagram is a finite weakly decreasing sequence σ=(σ1,…,σs)\sigma=(\sigma_1,\ldots,\sigma_s) of positive integers. We can also view σ\sigma as a sequence of rows of boxes of lengths σ1,…,σs\sigma_1,\ldots,\sigma_s. We recall that ∣σ∣=∑iσi|\sigma|=\sum_i\sigma_i, (a)+=max⁡{0,a}(a)_+=\max\{0,a\}, and for j≥1j\ge1,

γj(σ)=∑i=1s(σi−j+1)+,(3)\gamma_j(\sigma)=\sum_{i=1}^{s}(\sigma_i-j+1)_+, \tag*{(3)}

which counts the number of boxes to the right of column j−1j-1. We also set γq+1(σ)=0\gamma_{q+1}(\sigma)=0 when σ1≤q\sigma_1\le q. The dual or conjugate diagram to σ\sigma is σ∨=n(σ)\sigma^\vee=n(\sigma) defined by nj(σ)=#{i:σi≥j}n_j(\sigma)=\#\{i:\sigma_i\ge j\}. It is a diagram with σ1\sigma_1 rows. The elementary identity

γj(σ)−γj+1(σ)=nj(σ)(4)\gamma_j(\sigma)-\gamma_{j+1}(\sigma)=n_j(\sigma) \tag*{(4)}

will be used repeatedly.

Example 2.2.3. Given below is a Young diagram σ=(4,3,2)\sigma=(4,3,2) and its dual diagram σ∨=(3,3,2,1)\sigma^\vee=(3,3,2,1). We have, for example, γ1(σ)=9\gamma_1(\sigma)=9 and γ3(σ)=3\gamma_3(\sigma)=3.

Young diagram σ and its dual diagram σ∨

Figure 1.

If σ\sigma and τ\tau are diagrams, their concatenation σ∗τ\sigma* \tau is obtained by taking the multiset union of their parts and rearranging them in weakly decreasing order. Thus we write σ∗m\sigma^{*m} for the concatenation of mm copies of σ\sigma.

The four determinantal settings

This section records the representation theory and commutative algebra needed later. The emphasis is on the common shape calculus rather than on the differences among the four straightening laws.

Generic matrices

Let VV and WW be vector spaces of dimensions aa and bb, respectively, and put r=min⁡{a,b}r=\min\{a,b\}. After choosing bases, let X=(xij)X=(x_{ij}) be the generic a×ba\times b matrix, and

Sgen=k[X]=Sym⁡(V⊗W∗).S_{\mathrm{gen}}=\mathbb{k}[X]=\operatorname{Sym}(V\otimes W^{*}).

The group Ggen=GL⁡(V)×GL⁡(W)G_{\mathrm{gen}}=\operatorname{GL}(V)\times\operatorname{GL}(W) acts naturally by (M,N)⋅X=MXN−1(M,N)\cdot X=MXN^{-1}. The decomposition of SgenS_{\mathrm{gen}} into irreducible kGgen\mathbb{k}G_{\mathrm{gen}} modules and the invariant-ideal classification are as follows. The decomposition written in terms of Schur functors S\mathbf{S} applied to VV and W∗W^{*} respectively is

Sgen=⨁λ:λ1≤rSλ∨V⊗Sλ∨W∗.(5)S_{\mathrm{gen}}=\bigoplus_{\lambda:\lambda_{1}\le r}\mathbf{S}_{\lambda^{\vee}}V\otimes\mathbf{S}_{\lambda^{\vee}}W^{*}. \tag*{(5)}

We call the ideal Iλ=(Sλ∨V⊗Sλ∨W∗)I_{\lambda}=(\mathbf{S}_{\lambda^{\vee}}V\otimes\mathbf{S}_{\lambda^{\vee}}W^{*}) of SgenS_{\mathrm{gen}} generated by the summand indexed by λ∨\lambda^{\vee} an isotypic ideal. For example if σ=(t)\sigma=(t) is a diagram with one row then σ∨=(1)∗t\sigma^{\vee}=(1)^{*t} and

I(t)=(⋀tV⊗⋀tW∗)=It(X)I_{(t)}=\left(\bigwedge^{t}V\otimes\bigwedge^{t}W^{*}\right)=I_{t}(X)

is the ideal of t×tt\times t minors of XX. By [13 Theorem 4.1 and Proposition 5.1] every homogeneous GgenG_{\mathrm{gen}}-stable ideal is a finite sum of isotypic ideals. An important class of GgenG_{\mathrm{gen}}-stable ideals consists of products of determinantal ideals. These are related to the isotypic ideals via integral closure. Indeed, [13 Theorems 8.1] gives

Iλ‾=Dλ=Iλ1(X)⋯Iλs(X).(6)\overline{I_{\lambda}}=D_{\lambda}=I_{\lambda_{1}}(X)\cdots I_{\lambda_{s}}(X). \tag*{(6)}

We now discuss the analogous setup in determinantal rings. Fix q<rq<r and set

Agen=Sgen/Iq+1(X),pt=It(X)/Iq+1(X),1≤t≤q.(7)A_{\mathrm{gen}}=S_{\mathrm{gen}}/I_{q+1}(X),\qquad\mathfrak{p}_{t}=I_{t}(X)/I_{q+1}(X),\qquad1\le t\le q. \tag*{(7)}

The surviving summands in AgenA_{\mathrm{gen}} among those listed in (3.1.1) are those indexed by partitions with λ1≤q\lambda_{1}\le q. We abuse notation and continue to write IλI_{\lambda} for the isotypic ideal of AgenA_{\mathrm{gen}} generated by the corresponding summand and DλD_{\lambda} for the image in AgenA_{\mathrm{gen}} of the eponymous ideal from (3.1.2).

Write a minor as [i1,…,it∣j1,…,jt][i_{1},\ldots,i_{t}\mid j_{1},\ldots,j_{t}] with each index sequence strictly increasing and order them by

[I∣J]≤[I′∣J′]⟺t≥t′,ik≤ik′, jk≤jk′for 1≤k≤t′.[I\mid J]\le[I'\mid J']\Longleftrightarrow t\ge t',\quad i_{k}\le i'_{k},\ j_{k}\le j'_{k}\quad\text{for }1\le k\le t'.

A standard bitableau is a product of minors δ1⋯δs\delta_{1}\cdots\delta_{s} with δ1≤⋯≤δs\delta_{1}\le\cdots\le\delta_{s}. Equivalently, the two tableaux have strictly increasing rows and weakly increasing columns. Its shape is the Young diagram which lists the sizes of the factors. These products form the standard basis of SgenS_{\mathrm{gen}}; thus modulo Iq+1(X)I_{q+1}(X) the products with all sizes at most qq form a basis of AgenA_{\mathrm{gen}}.

Symmetric matrices

Put V=knV = \mathbb{k}^{n}, let Y=(yij)Y = (y_{ij}) be the generic symmetric n×nn \times n matrix (yij=yjiy_{ij} = y_{ji} for all $1 \le i < j \le n$), and

Ssym=k[Y]=Sym⁡(Sym⁡2V).S_{\mathrm{sym}} = \mathbb{k}[Y] = \operatorname{Sym}(\operatorname{Sym}^{2} V).

The natural congruence action of GL⁡(V)\operatorname{GL}(V) given by M⋅Y=MYMTM \cdot Y = MYM^{T} gives the multiplicity-free decomposition described by Abeasis [1, Theorem 2.4} in terms of partitions whose columns are even, equivalently λ∨∈2Nn\lambda^{\vee} \in2\mathbb{N}^{n} as follows

Ssym=⨁λ:λ1≤n, λ∨∈2NnSλ∨V.(8)S_{\mathrm{sym}} = \bigoplus_{\lambda:\lambda_{1}\le n,\ \lambda^{\vee}\in2\mathbb{N}^{n}} S_{\lambda^{\vee}}V. \tag*{(8)}

The homogeneous GL⁡(V)\operatorname{GL}(V)-invariant ideals of SsymS_{\mathrm{sym}} are the sums of these irreducible summands that are upward-closed under containment of Young diagrams; see [1 Theorem 3.1].

Let Jt(Y)J_t(Y) be the ideal of tt-minors of YY. For q<nq<n, set

Asym=Ssym/Jq+1(Y),pt=Jt(Y)/Jq+1(Y),1≤t≤q.(9)A_{\mathrm{sym}} = S_{\mathrm{sym}}/J_{q+1}(Y), \qquad\mathfrak{p}_t = J_t(Y)/J_{q+1}(Y), \qquad1 \le t \le q. \tag*{(9)}

A Young diagram λ\lambda with even column lengths has the form λ=(λ1=λ2≥λ3=λ4≥λ5=λ6,…)\lambda= (\lambda_1 = \lambda_2 \ge\lambda_3 = \lambda_4 \ge\lambda_5 = \lambda_6,\ldots) and an even number 2s2s of parts. Thus the relevant information is preserved by defining the following new diagram

δsym(λ)=(λ2,λ4,λ6,…,λ2s).(10)\delta_{\mathrm{sym}}(\lambda) = (\lambda_2,\lambda_4,\lambda_6,\ldots,\lambda_{2s}). \tag*{(10)}

We set Dδsym(λ)=Jλ2(Y)Jλ4(Y)⋯Jλ2s(Y)D_{\delta_{\mathrm{sym}}(\lambda)} = J_{\lambda_2}(Y)J_{\lambda_4}(Y)\cdots J_{\lambda_{2s}}(Y) in SsymS_{\mathrm{sym}} or

Dδsym(λ)=pλ2pλ4⋯pλ2sD_{\delta_{\mathrm{sym}}(\lambda)} = \mathfrak{p}_{\lambda_2}\mathfrak{p}_{\lambda_4}\cdots\mathfrak{p}_{\lambda_{2s}}

in AsymA_{\mathrm{sym}}. In particular, the partition λ=(t,t)\lambda= (t,t) with two equal rows of length tt corresponds to Jt(Y)J_t(Y).

The use of SL⁡(V)\operatorname{SL}(V) rather than GL⁡(V)\operatorname{GL}(V) does not change the class of homogeneous stable ideals.

Proposition 3.2.4. A homogeneous ideal of AsymA_{\mathrm{sym}} is SL⁡(V)\operatorname{SL}(V)-stable if and only if it is GL⁡(V)\operatorname{GL}(V)-stable.

Proof. Only the forward implication requires proof. Every element of GL⁡(V)\operatorname{GL}(V) is, after choosing an nn-th root of its determinant, a product of an element of SL⁡(V)\operatorname{SL}(V) and a scalar matrix. A scalar matrix acts by a scalar on each homogeneous component of SsymS_{\mathrm{sym}} and hence preserves every homogeneous ideal. The same statement descends to the homogeneous quotient AsymA_{\mathrm{sym}}. Thus if an ideal of AsymA_{\mathrm{sym}} is both SL⁡(V)\operatorname{SL}(V)-stable and homogeneous, then it is GL⁡(V)\operatorname{GL}(V)-stable. □

The symmetric standard basis is described by a single tableau with rows strictly increasing and columns weakly increasing satisfying the parity condition encoded in (3.2.1). The same functions (2.2.1) govern symbolic powers, however in the γ\gamma-formulas we must use the normalized shape δsym(λ)\delta_{\mathrm{sym}}(\lambda), listing one part for each minor, not the doubled tableau shape; see [21 Section 2.4].

Alternating matrices

Put V=knV = \mathbb{k}^{n}, let Z=(zij)Z = (z_{ij}) be the generic alternating n×nn \times n matrix (zij=−zjiz_{ij} = -z_{ji} for all $1 \le i < j \le nandandz_{ii} = 0forall$1≤i≤nfor all \$1 \le i \le n), and

Salt=k[zij:1≤i<j≤n]=Sym⁡(⋀2V).S_{\mathrm{alt}} = \mathbb{k}[z_{ij}:1 \le i < j \le n] = \operatorname{Sym}(\bigwedge^{2} V).

For t≤⌊n/2⌋t \le\lfloor n/2 \rfloor, let P2t(Z)P_{2t}(Z) be the ideal generated by all 2t2t-Pfaffians of ZZ. The natural congruence action of GL⁡(V)\operatorname{GL}(V) given by M⋅Z=MZMTM \cdot Z = MZM^{T} gives rise to an irreducible kGL⁡(V)\mathbb{k}\operatorname{GL}(V)-decomposition of SaltS_{\mathrm{alt}}, due to Abeasis–Del Fra [2 Theorem 2.5]

Salt=⨁λ:λ1≤n, λi evenSλ∨V.(11)S_{\mathrm{alt}} = \bigoplus_{\lambda:\lambda_1\le n,\ \lambda_i\ \mathrm{even}} S_{\lambda^{\vee}}V. \tag*{(11)}

Fix q<⌊n/2⌋q < \lfloor n/2 \rfloor and set

Aalt=Salt/P2(q+1)(Z),pt=P2t(Z)/P2(q+1)(Z),1≤t≤q.(12)A_{\mathrm{alt}} = S_{\mathrm{alt}}/P_{2(q+1)}(Z), \qquad\mathfrak{p}_t = P_{2t}(Z)/P_{2(q+1)}(Z), \qquad1 \le t \le q. \tag*{(12)}

For a Young diagram λ\lambda with even row lengths, put

δalt(λ)=(λ12,λ22,…,λs2).(13)\delta_{\mathrm{alt}}(\lambda)=\left(\frac{\lambda_1}{2},\frac{\lambda_2}{2},\ldots,\frac{\lambda_s}{2}\right). \tag*{(13)}

Set Dδalt(λ)=∏i=1sPλi(Z)D_{\delta_{\mathrm{alt}}(\lambda)}=\prod_{i=1}^{s}P_{\lambda_i}(Z) in SaltS_{\mathrm{alt}}, and use the same notation for its image in AaltA_{\mathrm{alt}},

Dδalt(λ)=∏i=1spλi/2.D_{\delta_{\mathrm{alt}}(\lambda)}=\prod_{i=1}^{s}p_{\lambda_i/2}.

A standard Pfaffian product has factors [i1,…,i2t][i_1,\ldots,i_{2t}] ordered by decreasing tt. Viewed as a tableau it has strictly increasing even size rows and weakly increasing columns. Its normalized shape lists the integers tt, so its polynomial degree is the size of that normalized shape.

Fixing a nondegenerate quadratic form QQ on VV gives an orthogonal subgroup O⁡(V,Q)⊂GL⁡(V)\operatorname{O}(V,Q)\subset\operatorname{GL}(V). Every Pfaffian-rank ideal and every sum or product of such ideals is O⁡(V,Q)\operatorname{O}(V,Q)-stable. The theorems of this paper therefore apply to these O⁡(V,Q)\operatorname{O}(V,Q)-invariant ideals. We do not assert that every O⁡(V,Q)\operatorname{O}(V,Q)-stable ideal has this form.

Hankel matrices

Let N≥3N\ge3 and write SHan=k[x1,…,xN]S_{\mathrm{Han}}=\mathbb{k}[x_1,\ldots,x_N]. For 1≤t≤⌊(N+1)/2⌋1\le t\le\lfloor(N+1)/2\rfloor, consider the t×(N−t+1)t\times(N-t+1) Hankel matrix

Ht=(x1x2⋯xN−t+1x2x3⋯xN−t+2⋮⋮⋱⋮xtxt+1⋯xN)(14)H_t= \begin{pmatrix} x_1 & x_2 & \cdots& x_{N-t+1}\\ x_2 & x_3 & \cdots& x_{N-t+2}\\ \vdots& \vdots& \ddots& \vdots\\ x_t & x_{t+1} & \cdots& x_N \end{pmatrix} \tag*{(14)}

and let Kt=It(Ht)K_t=I_t(H_t) be its ideal of maximal minors. Every Hankel determinantal ring is isomorphic to one where the defining ideal is generated by the maximal sized minors of a Hankel matrix [20 Lemma 2.3]; alternatively see [29 Proposition 7] or [11 Corollary 2.2(b)]. The ideals KtK_t form a chain

K1⊃K2⊃⋯⊃K⌊(N+1)/2⌋.K_1\supset K_2\supset\cdots\supset K_{\lfloor(N+1)/2\rfloor}.

For q<⌊(N+1)/2⌋q<\lfloor(N+1)/2\rfloor, put

AHan=SHan/Kq+1,pt=Kt/Kq+1,1≤t≤q.(15)A_{\mathrm{Han}}=S_{\mathrm{Han}}/K_{q+1},\qquad p_t=K_t/K_{q+1},\qquad1\le t\le q. \tag*{(15)}

The variables may be viewed, up to rescaling by appropriate binomial coefficients, as the coefficients of a binary form of degree N−1N-1

F(u,v)=∑i=0N−1(N−1i)xi+1uN−1−ivi.F(u,v)=\sum_{i=0}^{N-1}\binom{N-1}{i}x_{i+1}u^{N-1-i}v^i.

The group GL⁡2(k)\operatorname{GL}_2(\mathbb{k}) acts naturally on binary forms by change of basis according to the formula

(g⋅F)(u,v)=F(au+bv,cu+dv)=∑i=0N−1(N−1i)(g⋅xi+1)uN−1−ivi,for g=(abcd)∈GL⁡2(k).(g\cdot F)(u,v)=F(au+bv,cu+dv)=\sum_{i=0}^{N-1}\binom{N-1}{i}(g\cdot x_{i+1})u^{N-1-i}v^i,\qquad\text{for }g=\begin{pmatrix}a&b\\c&d\end{pmatrix}\in\operatorname{GL}_2(\mathbb{k}).

This GL⁡2(k)\operatorname{GL}_2(\mathbb{k})-action preserves every KtK_t. Geometrically, the Hankel determinantal rings AHanA_{\mathrm{Han}} arise as homogeneous coordinate rings of higher order secant varieties of rational normal curves, see [28 Chapter 11.7]. In particular, the quotient SHan/Kq+1S_{\mathrm{Han}}/K_{q+1} is the homogeneous coordinate ring of the (q−1)(q-1)-st secant variety of the rational normal curve CN−1C_{N-1} in PN−1\mathbb{P}^{N-1}, see [18 Section 4].

Conca constructed a straightening law for the minors of Hankel matrices [11 Theorem 2.6]. Standard products of Hankel minors have a shape given by the minor sizes. His theorem identifies the symbolic powers of each KtK_t and the primary decomposition of every power KtsK_t^s by the same γ\gamma-functions as in the generic case. The full GL⁡2(k)\operatorname{GL}_2(\mathbb{k})-stable ideal theory involves plethysm and is not multiplicity-free.

Symbolic-power descent

The following formulas explain the descent of symbolic powers to the quotient rings AA. In the generic determinantal case the formula below was noted in [9 Remark 10.5].

Proposition 3.5.1. In the generic, symmetric, and alternating cases, let PjP_j denote the ambient rank ideal, with Pj=P2j(Z)P_j = P_{2j}(Z) in the alternating case, and put A=S/Pq+1A = S/P_{q+1} and pj=Pj/Pq+1\mathfrak{p}_j = P_j/P_{q+1}. For 1≤j≤q1 \le j \le q and 0≠f∈A0 \ne f \in A, the local order

vj(f)=max⁡{a≥0:f∈pjaApj},v_j(f) = \max\{a \ge0 : f \in\mathfrak{p}_j^a A_{\mathfrak{p}_j}\},

extends to a discrete valuation of Frac⁡(A)\operatorname{Frac}(A). We have

pj(m)={f∈A:vj(f)≥m}=Pj(m)+Pq+1Pq+1.(16)\mathfrak{p}_j^{(m)} = \{f \in A : v_j(f) \ge m\} = \frac{P_j^{(m)} + P_{q+1}}{P_{q+1}}. \tag*{(16)}

Proof. Invert the upper-left (j−1)(j-1)-minor, or the upper-left 2(j−1)2(j-1)-Pfaffian in the alternating case, using the empty block when j=1j=1. In the generic and symmetric cases, write the matrix as

(BCDE),W=E−DB−1C.\begin{pmatrix} B & C \\ D & E \end{pmatrix}, \qquad W = E - DB^{-1}C.

Block elimination reduces the rank ideals to the corresponding minor ideals of WW; in the symmetric case D=CtD = C^{t} and WW is symmetric. Alternating block congruence gives the analogous reduction of Pfaffian sizes by 2(j−1)2(j-1). For the generic calculation see [9 Proposition 2.4].

After localizing at pj\mathfrak{p}_j, the eliminated parameters form a coefficient field FF, and ApjA_{\mathfrak{p}_j} becomes RnR_{\mathfrak{n}}, where RR is the smaller standard graded determinantal or Pfaffian domain of rank at most q−j+1q-j+1, and n=R+\mathfrak{n} = R_+. Since

gr⁡pjApjApj=gr⁡nRn(Rn)≅R,\operatorname{gr}_{\mathfrak{p}_j A_{\mathfrak{p}_j}} A_{\mathfrak{p}_j} = \operatorname{gr}_{\mathfrak{n}R_{\mathfrak{n}}}(R_{\mathfrak{n}}) \cong R,

is a domain, the local order is additive on nonzero products and satisfies the valuation inequality on sums. It therefore extends by vj(f/g)=vj(f)−vj(g)v_j(f/g) = v_j(f) - v_j(g).

The irreducible representation decompositions of the classical quotient rings AA are multiplicity-free; see [21 Sections 2.3–2.5] and the references therein. Let MλM_\lambda be the surviving irreducible summand of normalized shape λ\lambda in AA, whose highest-weight vector is

Kλ=δλ1⋯δλs.K_\lambda= \delta_{\lambda_1} \cdots\delta_{\lambda_s}.

Here δt\delta_t is the upper-left tt-minor, or the upper-left 2t2t-Pfaffian. If t<jt < j, then δt\delta_t is a unit at pj\mathfrak{p}_j. If j≤t≤qj \le t \le q, block reduction expresses it as a unit times a nonzero homogeneous residual minor or Pfaffian of degree t−j+1t-j+1. Consequently,

vj(Kλ)=∑i(λi−j+1)+=γj(λ).(17)v_j(K_\lambda) = \sum_i (\lambda_i-j+1)_+ = \gamma_j(\lambda). \tag*{(17)}

Since pj(m)\mathfrak{p}_j^{(m)} is group-stable, complete reducibility and multiplicity-freeness show that it contains MλM_\lambda exactly when it contains KλK_\lambda, that is, based on the above computation if and only if γj(λ)≥m\gamma_j(\lambda) \ge m. The same cutoff describes the ambient symbolic powers by De Concini–Eisenbud–Procesi [13 Theorem 7.1], Abeasis [1 Theorem 5.1], and Abeasis–Del Fra [2 Theorem 5.1]; see also [21 Theorems 2.1, 2.6, and 2.9]. Passing to AA removes precisely the summands with λ1>q\lambda_1 > q, proving (3.5.2). The compatible standard bases give the basis assertion. In the generic case, the descent identity is also stated explicitly in [9 Remark 10.5].

For Hankel matrices, we use Conca’s ambient straightening theory [11] and Nam’s description [26] of the diagonal initial ideals. The following argument supplies the primarity needed to pass from those ambient results to symbolic powers in the rank quotient.

Proposition 3.5.4. Let A=S/Kq+1A=S/K_{q+1} be a Hankel rank quotient and pj=Kj/Kq+1\mathfrak{p}_j=K_j/K_{q+1}. For 1≤j≤q1\le j\le q and m≥1m\ge1,

pj(m)=Kj(m)+Kq+1Kq+1.(18)\mathfrak{p}_j^{(m)}=\frac{K_j^{(m)}+K_{q+1}}{K_{q+1}}. \tag*{(18)}

These ideals have as a basis the surviving standard products of shapes λ\lambda satisfying γj(λ)≥m\gamma_j(\lambda)\ge m.

Proof. Put Q=Kj(m)+Kq+1Q=K_j^{(m)}+K_{q+1} and use the degree-lexicographic order with x1>⋯>xNx_1>\cdots>x_N. Conca’s ambient straightening and symbolic-power results [11], in Nam’s initial-ideal formulation [26 Corollary 3.9, Theorem 3.12, and Corollary 3.21], give

Jt:=in⁡(Kt)=(xi1⋯xit:iu+1≥iu+2),in⁡(Kj(m))=Jj(m).J_t:=\operatorname{in}(K_t)=(x_{i_1}\cdots x_{i_t}:i_{u+1}\ge i_u+2),\qquad\operatorname{in}(K_j^{(m)})=J_j^{(m)}.

The two summands of QQ are spanned by subsets of the same standard basis, whose elements have pairwise distinct leading monomials. Hence

in⁡(Q)=L:=Jj(m)+Jq+1.(19)\operatorname{in}(Q)=L:=J_j^{(m)}+J_{q+1}. \tag*{(19)}

We prove that LL is unmixed.

For E⊂{1,…,N}E\subset\{1,\ldots,N\}, let c(E)c(E) be the largest size of a subset containing no consecutive indices, and set Δh={E:c(E)≤h}\Delta_h=\{E:c(E)\le h\}. The facets of Δh\Delta_h are unions of hh disjoint adjacent pairs; in particular, they have 2h2h vertices and their maximal consecutive runs have even lengths. With PF=(xi:i∉F)P_F=(x_i:i\notin F), we therefore have

Jj(m)=⋂F facet of Δj−1PFm.J_j^{(m)}=\bigcap_{F\text{ facet of }\Delta_{j-1}}P_F^m.

These facet and symbolic-power descriptions are given in [26 proof of Theorem 3.12].

Write r=j−1r=j-1. We first observe that every face C∈ΔrC\in\Delta_r contained in a facet GG of Δq\Delta_q extends to a facet FF of Δr\Delta_r contained in GG. Indeed, choose an inclusion-maximal C⊆C′⊆GC\subseteq C'\subseteq G with c(C′)≤rc(C')\le r. If a consecutive run of C′C' had odd length, it would have an adjacent missing vertex inside GG. Adjoining that vertex would leave c(C′)c(C') unchanged, even if it joined two runs, contradicting maximality. Thus all runs of C′C' have even length and ∣C′∣=2c(C′)|C'|=2c(C'). If c(C′)<rc(C')<r, a missing vertex could be added, increasing cc by at most one. Hence c(C′)=rc(C')=r, and C′C' is the required facet.

Every associated prime of the monomial ideal LL is generated by variables and has the form (L:M)(L:M) for a monomial M∉LM\notin L. Fix such an MM. Choose a facet GG of Δq\Delta_q containing E=supp⁡(M)E=\operatorname{supp}(M), and a facet F0F_0 of Δr\Delta_r such that

∑i∉F0deg⁡xi(M)<m.\sum_{i\notin F_0}\deg_{x_i}(M)<m.

By the preceding observation, F0∩EF_0\cap E extends to a facet F⊆GF\subseteq G of Δr\Delta_r. Then

∑i∉Fdeg⁡xi(M)<m.\sum_{i\notin F}\deg_{x_i}(M)<m.

For every i∈Fi\in F, the monomial xiMx_iM remains outside PFmP_F^m and has support contained in GG, so xiM∉Lx_iM\notin L. Consequently (L:M)⊆PF(L:M)\subseteq P_F. Since L=Jj\sqrt{L}=J_j and PFP_F is a minimal prime of JjJ_j, we obtain (L:M)=PF(L:M)=P_F. Thus all associated primes of S/LS/L have dimension 2r2r, and LL is unmixed.

Here Q=Kj\sqrt{Q}=K_j. If QQ had an embedded associated prime r\mathfrak r, choose homogeneous ff with (Q:f)=r(Q:f)=\mathfrak r. The modules

Q+(f)Qandin⁡(Q+(f))L\frac{Q+(f)}{Q} \qquad\text{and}\qquad \frac{\operatorname{in}(Q+(f))}{L}

have the same Hilbert function, since initial ideals preserve Hilbert functions. The first has dimension dim⁡S/r<2r\dim S/\mathfrak r<2r, whereas the second is a nonzero submodule of the unmixed module S/LS/L and therefore has dimension 2r2r. This contradiction proves that QQ is KjK_j-primary.

Finally, the definition of the ambient symbolic power gives

(Q/Kq+1)Apj=pjmjApj.(Q/K_{q+1})A_{\mathfrak p_j}=\mathfrak p_j^{m_j}A_{\mathfrak p_j}.

Since Q/Kq+1Q/K_{q+1} is pj\mathfrak p_j-primary, contraction proves (3.5.5). The common standard basis gives the asserted quotient basis.

The uniform standard-monomial package

We now state the common input. A standard monomial means a standard bitableau in the generic case, the corresponding symmetric or Pfaffian tableau in the classical cases, and a standard product of Hankel minors in the Hankel case. Its shape is the diagram of its normalized factor sizes.

Definition 3.6.1. A diagram is admissible if all its parts are at most qq.

Theorem 3.6.2 (Uniform standard-monomial package). Let AA be one of Agen,Asym,Aalt,AHanA_{\mathrm{gen}}, A_{\mathrm{sym}}, A_{\mathrm{alt}}, A_{\mathrm{Han}} and let p1⊃⋯⊃pq\mathfrak p_1 \supset\cdots\supset\mathfrak p_q be the corresponding rank ideals described in the previous sections.

(i) The ring AA is a normal Cohen–Macaulay domain, and each pj\mathfrak p_j is prime.

(ii) The standard monomials whose shapes have all parts at most qq form a kk-basis of AA.

(iii) For a≥1a \ge1, the symbolic power pj(a)\mathfrak p_j^{(a)} has as a basis the standard monomials of shape λ\lambda satisfying γj(λ)≥a\gamma_j(\lambda) \ge a.

(iv) If σ\sigma is admissible, then

Dσ=∏i=1spσi(20)D_\sigma=\prod_{i=1}^{s}\mathfrak p_{\sigma_i} \tag*{(20)}

has as a basis the standard monomials of shape λ\lambda satisfying

γj(λ)≥γj(σ)for 1≤j≤q.(21)\gamma_j(\lambda)\ge\gamma_j(\sigma)\quad\text{for }1\le j\le q. \tag*{(21)}

Equivalently,

Dσ=⋂j=1σ1pj(γj(σ)).(22)D_\sigma=\bigcap_{j=1}^{\sigma_1}\mathfrak p_j^{(\gamma_j(\sigma))}. \tag*{(22)}

Proof. Write PtP_t for the ambient rank ideal, so that A=S/Pq+1A=S/P_{q+1} and pt=Pt/Pq+1\mathfrak p_t=P_t/P_{q+1}. Here Pt=KtP_t=K_t in the Hankel case and Pt=P2t(Z)P_t=P_{2t}(Z) in the alternating case.

For (i), the generic determinantal ring is a normal Cohen–Macaulay domain by [9 Corollary 5.17 and Theorem 6.3]. The symmetric and alternating rings are respectively orthogonal and symplectic invariant rings of polynomial rings, by the first and second fundamental theorems [14]. They are normal domains, and Cohen–Macaulayness follows from the Hochster–Roberts theorem [24]. The Hankel assertion follows from [12 Theorem 2.1]. Furthermore,

A/pj≅S/PjA/\mathfrak p_j\cong S/P_j

is the corresponding lower-rank determinantal or Pfaffian domain, or kk when j=1j=1. Thus every pj\mathfrak p_j is prime.

The ambient standard bases are the classical straightening bases [9, 1, 2] and Conca’s Hankel basis [11 Theorem 1.2]; the latter’s monomial indexing is also described in [26 Remark 3.18]. In each case, Pq+1P_{q+1} is spanned by precisely the standard monomials having a factor of normalized size at least q+1q+1. Their removal gives the quotient basis in (ii).

Part (iii) is exactly the basis assertion of Propositions 3.5.1 and 3.5.4.

For (iv), put

Dσ=∏i=1sPσi⊆S.D_\sigma=\prod_{i=1}^{s}P_{\sigma_i}\subseteq S.

The ambient product and straightening theorems in the ring SS give a basis of DσD_\sigma consisting of the standard monomials of shapes λ\lambda satisfying

γj(λ)≥γj(σ)for every ambient rank index j.\gamma_j(\lambda)\ge\gamma_j(\sigma)\qquad\text{for every ambient rank index }j.

For generic matrices this is [9 Theorem 10.9]; the symmetric and alternating results are due to Abeasis [1] and Abeasis–Del Fra [2]. For Hankel matrices use [11 Theorem 3.12], or equivalently [26 Theorem 3.25] together with [26 Corollary 3.21].

The image of DσD_\sigma in AA is DσD_\sigma. Since DσD_\sigma and Pq+1P_{q+1} are spanned by subsets of the same ambient standard basis, this image has as a basis precisely the surviving standard monomials satisfying the displayed inequalities. For indices j>qj>q, both γj(λ)\gamma_j(\lambda) and γj(σ)\gamma_j(\sigma) vanish, so the surviving conditions are (3.6.4).

By (iii), the intersection in (3.6.5) is spanned by exactly these same basis elements. This proves the equality. □

Remark 3.6.6. The generic statement is valid whenever the characteristic restrictions in the primary-decomposition theorem are met; the Hankel straightening results are largely characteristic-free. Characteristic zero is kept throughout so that one may simultaneously use complete reducibility and the invariant-ideal classifications as in the proof of Proposition 3.5.1.

Symbolic powers of products of rank prime ideals

From this point on, the proof uses only Theorem 3.6.2. In particular, the four matrix types no longer need to be considered separately.

Throughout the section let AA be one of Agen,Asym,Aalt,AHanA_{\mathrm{gen}}, A_{\mathrm{sym}}, A_{\mathrm{alt}}, A_{\mathrm{Han}} and let p1⊃⋯⊃pq\mathfrak{p}_1 \supset\cdots\supset\mathfrak{p}_q be the respective rank ideals described in the previous sections. Moreover let Dσ=∏i=1spσiD_\sigma= \prod_{i=1}^{s} \mathfrak{p}_{\sigma_i} for an admissible diagram σ\sigma with ss rows of length at most qq.

Primary decomposition after rank truncation

Fix an admissible diagram σ=(σ1,…,σs)\sigma=(\sigma_1,\ldots,\sigma_s) with σ1≤q\sigma_1\le q. We first record an elementary comparison of partitions. We say that a partition ν\nu dominates μ\mu if ∑ℓ=1uνℓ≥∑ℓ=1uμℓ\sum_{\ell=1}^{u}\nu_\ell\ge\sum_{\ell=1}^{u}\mu_\ell for all integers u≥0u\ge0.

Lemma 4.1.1. If μ,ν\mu,\nu are admissible partitions and ν\nu dominates μ\mu then for every jj such that 1≤j≤q1\le j\le q we have γj(ν)≥γj(μ)\gamma_j(\nu)\ge\gamma_j(\mu).

Proof. For a partition μ\mu, extended by zeros, and a real number c≥0c\ge0, we have

∑ℓ≥1(μℓ−c)+=max⁡u≥0{∑ℓ=1uμℓ−uc}.(23)\sum_{\ell\ge1}(\mu_\ell-c)_+ = \max_{u\ge0}\left\{\sum_{\ell=1}^{u}\mu_\ell-uc\right\}. \tag*{(23)}

Indeed, the maximum is attained when uu is the number of parts of μ\mu strictly larger than cc. Consequently, if ν\nu dominates μ\mu, meaning that ∑ℓ=1uνℓ≥∑ℓ=1uμℓ\sum_{\ell=1}^{u}\nu_\ell\ge\sum_{\ell=1}^{u}\mu_\ell for all integers u≥0u\ge0, then

∑ℓ≥1(νℓ−c)+≥∑ℓ≥1(μℓ−c)+for every c≥0.(24)\sum_{\ell\ge1}(\nu_\ell-c)_+ \ge\sum_{\ell\ge1}(\mu_\ell-c)_+ \qquad\text{for every } c\ge0. \tag*{(24)}

Taking c=j−1c=j-1 gives the desired conclusion. □

By (3.6.5), DσD_\sigma has a possibly redundant primary decomposition indexed by 1≤j≤σ11\le j\le\sigma_1. The next lemma determines precisely which components are needed. It is analogous to [13 Proposition 1.4].

Lemma 4.1.4 (Irredundancy criterion). The component pj(γj(σ))\mathfrak{p}_j^{(\gamma_j(\sigma))} in (3.6.5) is irredundant if and only if

γj+1(σ)≤(q−j)(nj(σ)−1).(25)\gamma_{j+1}(\sigma) \le(q-j)(n_j(\sigma)-1). \tag*{(25)}

Proof. Recall (2.2.1) and (2.2.2) for the definitions of γj\gamma_j and njn_j. If every part of λ\lambda is at most qq, then only the parts satisfying λi≥j+1\lambda_i\ge j+1 contribute to γj+1(λ)\gamma_{j+1}(\lambda), and each such part contributes at most q−jq-j. Hence

γj+1(λ)≤(q−j)nj+1(λ)≤(q−j)nj(λ).(26)\gamma_{j+1}(\lambda) \le(q-j)n_{j+1}(\lambda) \le(q-j)n_j(\lambda). \tag*{(26)}

Suppose that γj+1(σ)>(q−j)(nj(σ)−1)\gamma_{j+1}(\sigma) > (q-j)(n_j(\sigma)-1). If a shape λ\lambda satisfies γj+1(λ)≥γj+1(σ)\gamma_{j+1}(\lambda) \ge\gamma_{j+1}(\sigma), then (4.1.6) gives

(q−j)nj(λ)≥γj+1(λ)≥γj+1(σ)>(q−j)(nj(σ)−1),(q-j)n_j(\lambda) \ge\gamma_{j+1}(\lambda) \ge\gamma_{j+1}(\sigma) > (q-j)(n_j(\sigma)-1),

and therefore nj(λ)≥nj(σ)n_j(\lambda) \ge n_j(\sigma). Since γj(λ)−γj+1(λ)=nj(λ)\gamma_j(\lambda)-\gamma_{j+1}(\lambda)=n_j(\lambda), we obtain

γj(λ)=γj+1(λ)+nj(λ)≥γj+1(σ)+nj(σ)=γj(σ).\gamma_j(\lambda)=\gamma_{j+1}(\lambda)+n_j(\lambda)\ge\gamma_{j+1}(\sigma)+n_j(\sigma)=\gamma_j(\sigma).

Consequently, membership in pj+1(γj+1(σ))\mathfrak{p}_{j+1}^{(\gamma_{j+1}(\sigma))} implies membership in pj(γj(σ))\mathfrak{p}_j^{(\gamma_j(\sigma))} under the given assumption. Thus the pj\mathfrak{p}_j-primary component is redundant in (3.6.5) under the stated inequality.

Assume now that γj+1(σ)≤(q−j)(nj(σ)−1)\gamma_{j+1}(\sigma)\le(q-j)(n_j(\sigma)-1). We show that the pj\mathfrak{p}_j-primary component is irredundant in (3.6.5) by producing a standard monomial that belongs to every other primary component in (3.6.5).

Put n=nj(σ)n=n_j(\sigma) and a=γj+1(σ)a=\gamma_{j+1}(\sigma). Write the tail of the nn parts of σ\sigma which are at least jj as a partition of aa inside an n×(q−j)n\times(q-j) rectangle. Repack these aa boxes into an (n−1)×(q−j)(n-1)\times(q-j) rectangle so that the resulting partition of aa is maximal in dominance order (equivalently, fill the rows from left to right and top to bottom). The resulting tail majorizes the old one and therefore does not decrease any γi\gamma_i with i>ji>j by Lemma 4.1.1. Reattach jj boxes to the left of each of the n−1n-1 rows of the new tail. Also reattach the original parts below row nn and add two parts of size j−1j-1 (when j=1j=1 no such parts are needed) to obtain a new diagram λ\lambda. As one box was discarded from row nn column jj of σ\sigma and no new boxes were added to column jj, the new shape λ\lambda satisfies

γj(λ)=a+n−1=γj(σ)−1,\gamma_j(\lambda)=a+n-1=\gamma_j(\sigma)-1,

whereas γi(λ)≥γi(σ)\gamma_i(\lambda)\ge\gamma_i(\sigma) for every i≠ji\ne j. Indeed, for i<ji<j the loss from replacing nn parts crossing the jj-th column by n−1n-1 such parts is j−i+1j-i+1, while the two added (j−1)(j-1)-parts contribute 2(j−i)≥j−i+12(j-i)\ge j-i+1.

Thus the standard monomial of shape λ\lambda belongs to every primary component of (3.6.5) except the pj\mathfrak{p}_j-primary one. Hence that component is irredundant. □

Example 4.1.7. The tail-repacking construction in the above proof is illustrated below for σ=(5,4,3,2)\sigma=(5,4,3,2), j=2j=2, a=6a=6, n=4n=4 and yields λ=(5,5,2,1,1)\lambda=(5,5,2,1,1). The original and the repacked tails are shaded. Lost parts are displayed in red and gained parts in green.

The tail-repacking construction

Figure 1 (PDF p. 12). The tail-repacking construction

For the purpose of computing symbolic powers, we need to keep track of the maximal associated primes of ideals DσD_\sigma. Towards this end define

βq(σ)=max⁡{1,∣σ∣−q(s−1)}.(27)\beta_q(\sigma)=\max\{1,|\sigma|-q(s-1)\}. \tag*{(27)}

We usually write β\beta when qq and σ\sigma are fixed.

Proposition 4.1.9. The smallest index jj satisfying (4.1.5) is β=βq(σ)\beta=\beta_q(\sigma). Consequently,

Ass⁡A(A/Dσ)={pj:1≤j≤σ1, γj+1(σ)≤(q−j)(nj(σ)−1)}(28)\operatorname{Ass}_{A}(A/D_{\sigma})=\{\mathfrak{p}_j:1\leq j\leq\sigma_1,\ \gamma_{j+1}(\sigma)\leq(q-j)(n_j(\sigma)-1)\} \tag*{(28)}
MaxAss⁡A(A/Dσ)={pβ}.(29)\operatorname{MaxAss}_{A}(A/D_{\sigma})=\{\mathfrak{p}_{\beta}\}. \tag*{(29)}

Proof. Suppose σ\sigma is a partition with ss parts of length at most qq. Let

d=∑i=1s(q−σi),so that∣σ∣−q(s−1)=q−d.d=\sum_{i=1}^{s}(q-\sigma_i), \qquad\text{so that}\qquad|\sigma|-q(s-1)=q-d.

If q−d>1q-d>1, then q−d≤σsq-d\leq\sigma_s, so every part of σ\sigma is at least jj for j≤q−dj\leq q-d. In this range, nj(σ)=sn_j(\sigma)=s and

γj+1(σ)=s(q−j)−d.\gamma_{j+1}(\sigma)=s(q-j)-d.

Thus (4.1.5) is equivalent to s(q−j)−d≤(q−j)(s−1)s(q-j)-d\leq(q-j)(s-1), that is, d≥q−jd\geq q-j, or j≥q−dj\geq q-d. Thus in this case the smallest index jj such that the pj\mathfrak{p}_j-primary component is irredundant in (3.6.5) is β=q−d=∣σ∣−q(s−1)\beta=q-d=|\sigma|-q(s-1).

If q−d≤1q-d\leq1, the same calculation as above at j=1j=1 shows that the p1\mathfrak{p}_1-primary component is irredundant in (3.6.5). This proves that β=1\beta=1.

The description of associated primes follows from the irredundant primary decomposition, and the chain p1⊃⋯⊃pq\mathfrak{p}_1\supset\cdots\supset\mathfrak{p}_q gives (4.1.11). □

Formulas for symbolic powers of products of rank prime ideals

Theorem 4.2.1. Let β=βq(σ)\beta=\beta_q(\sigma) be as in (4.1.8). For every integer m≥1m\geq1, the following identity holds

Dσ(m)=⋂j=βσ1pj(mγj(σ)).(30)D_{\sigma}^{(m)}=\bigcap_{j=\beta}^{\sigma_1}\mathfrak{p}_j^{(m\gamma_j(\sigma))}. \tag*{(30)}

Equivalently, Dσ(m)D_{\sigma}^{(m)} has a basis consisting of the standard monomials of shapes λ\lambda satisfying

γj(λ)≥mγj(σ)for β≤j≤σ1.(31)\gamma_j(\lambda)\geq m\gamma_j(\sigma)\qquad\text{for }\beta\leq j\leq\sigma_1. \tag*{(31)}

Proof. By Proposition 4.1.9, since the associated primes of A/DσA/D_{\sigma} form a chain with maximal element pβ\mathfrak{p}_{\beta}, the symbolic power may be computed by localizing at pβ\mathfrak{p}_{\beta}. Since Dσm=Dσ∗mD_{\sigma}^{m}=D_{\sigma^{*m}} and γj(σ∗m)=mγj(σ)\gamma_j(\sigma^{*m})=m\gamma_j(\sigma), Theorem 3.6.2(iv) gives

Dσm=⋂j=1σ1pj(mγj(σ)).D_{\sigma}^{m}=\bigcap_{j=1}^{\sigma_1}\mathfrak{p}_j^{(m\gamma_j(\sigma))}.

For j<βj<\beta, the prime pj\mathfrak{p}_j strictly contains pβ\mathfrak{p}_{\beta}, so its component becomes the unit ideal after localization at pβ\mathfrak{p}_{\beta}. For j≥βj\geq\beta, the pj\mathfrak{p}_j-primary component pj(mγj(σ))\mathfrak{p}_j^{(m\gamma_j(\sigma))} is pβ\mathfrak{p}_{\beta}-saturated, thus localization at pβ\mathfrak{p}_{\beta} and contraction to AA preserve the component. Contracting back to AA gives (4.2.2). The basis description is a consequence of Theorem 3.6.2(iii). □

In Theorem 4.2.1 we gave an intersection formula for symbolic powers. We now give a sum formula which, in addition, produces explicit generators for the symbolic Rees algebra. This argument is independent of any SAGBI or toric-degeneration assertions.

Set d=q−β+1d=q-\beta+1. For a=(aβ,…,aq)∈Nda=(a_{\beta},\ldots,a_q)\in\mathbb{N}^{d}, define

τ(a)=(qaq,(q−1)aq−1,…,βaβ),Dτ(a)=∏t=βqptat,∣a∣=∑t=βqat.\tau(a)=(q^{a_q},(q-1)^{a_{q-1}},\ldots,\beta^{a_{\beta}}), \qquad D_{\tau(a)}=\prod_{t=\beta}^{q}\mathfrak{p}_t^{a_t}, \qquad |a|=\sum_{t=\beta}^{q}a_t.

Theorem 4.2.4. Let β=βq(σ)\beta=\beta_q(\sigma) be as in (4.1.8). For every integer m≥1m\geq1, the following identity holds

Dσ(m)=∑a∈Ndγj(τ(a))≥mγj(σ) ∀β≤j≤σ1Dτ(a),(32)D_{\sigma}^{(m)}=\sum_{\substack{a\in\mathbb{N}^{d}\\ \gamma_j(\tau(a))\geq m\gamma_j(\sigma)\ \forall\beta\leq j\leq\sigma_1}}D_{\tau(a)}, \tag*{(32)}

Proof. By Theorem 3.6.2(iv),

Dτ(a)⊆pj(γj(τ(a)))⊆pj(mγj(σ))D_{\tau(a)}\subseteq\mathfrak{p}_j^{(\gamma_j(\tau(a)))}\subseteq\mathfrak{p}_j^{(m\gamma_j(\sigma))}

whenever γj(τ(a))≥mγj(σ)\gamma_j(\tau(a))\geq m\gamma_j(\sigma). Intersecting over j∈[β,σ1]j\in[\beta,\sigma_1] and applying Theorem 4.2.1 proves that every summand in (4.2.5) is contained in the left side.

Conversely, let Δ\Delta be a standard product in the basis of Dσ(m)D_{\sigma}^{(m)} specified in Theorem 4.2.1. For β≤t≤q\beta\leq t\leq q, let ata_t be the number of factors of normalized size tt in Δ\Delta. Factors of size less than β\beta contribute zero to every γj\gamma_j, with j≥βj\geq\beta. The product of the remaining factors lies in Dτ(a)D_{\tau(a)}, and γj(τ(a))≥mγj(σ)\gamma_j(\tau(a))\geq m\gamma_j(\sigma) for β≤j≤σ1\beta\leq j\leq\sigma_1. Consequently Δ∈Dτ(a)\Delta\in D_{\tau(a)}, proving (4.2.5) by the basis assertion in Theorem 4.2.1. □

The symbolic Rees algebra of a product of rank prime ideals

We now discuss finite-generation, normality, and Cohen–Macaulayness of the symbolic Rees algebra for the ideals DσD_\sigma.

The proof uses convex geometry and is based on the following constructions. Fix a Young diagram σ\sigma and integers q>0q>0 and β=βq(σ)\beta=\beta_q(\sigma). We introduce the integral matrix

B=(bjt)β≤j≤σ1, β≤t≤q,bjt=(t−j+1)+,(33)B=(b_{jt})_{\beta\leq j\leq\sigma_1,\ \beta\leq t\leq q},\qquad b_{jt}=(t-j+1)_+, \tag*{(33)}

and the vector

c=(γβ(σ),…,γσ1(σ)).c=(\gamma_\beta(\sigma),\ldots,\gamma_{\sigma_1}(\sigma)).

Consider the semigroup

Mσ={(a,m)∈Nd×N:Ba−mc≥0}.(34)M_\sigma=\{(a,m)\in\mathbb{N}^{d}\times\mathbb{N}:Ba-mc\geq0\}. \tag*{(34)}

It is the lattice point monoid of the pointed rational polyhedral cone

Cσ={(x,u)∈R≥0d+1:Bx−uc≥0}.C_\sigma=\{(x,u)\in\mathbb{R}_{\geq0}^{d+1}:Bx-uc\geq0\}.

Therefore MσM_\sigma is finitely generated by Gordan’s lemma and saturated. Consider a (finite) Hilbert basis for MσM_\sigma

Hσ=Hilb⁡(Mσ),and putHσ+={(a,m)∈Hσ:m>0}.(35)H_\sigma=\operatorname{Hilb}(M_\sigma), \qquad\text{and put}\qquad H_\sigma^+=\{(a,m)\in H_\sigma:m>0\}. \tag*{(35)}

Theorem 5.0.4. Let AA be any of the generic, symmetric, alternating, or Hankel rank-qq quotient rings considered above, and let σ\sigma be an admissible shape. Then for the product DσD_\sigma of rank ideals the symbolic Rees algebra satisfies

Rs(Dσ):=⨁m≥0Dσ(m)Tm=A[Dτ(a)Tm:(a,m)∈Hσ+].(36)\mathcal{R}_s(D_\sigma):=\bigoplus_{m\geq0}D_\sigma^{(m)}T^m =A\left[D_{\tau(a)}T^m:(a,m)\in H_\sigma^+\right]. \tag*{(36)}

In the generic, symmetric, and alternating cases the symbolic Rees algebra is normal, has rational singularities, and hence is a normal Cohen–Macaulay domain.

Proof. With notation as in (5.0.1) we have (Ba)j=γj(τ(a))(Ba)_j=\gamma_j(\tau(a)). Consequently one can rewrite the set of inequalities γj(τ(a))≥mγj(σ)\gamma_j(\tau(a))\geq m\gamma_j(\sigma) for βq(σ)≤j≤σ1\beta_q(\sigma)\leq j\leq\sigma_1 as Ba≥mcBa\geq mc. Here inequalities between vectors are understood componentwise.

Then (4.2.5) gives

Rs(Dσ)=A[Dτ(a)Tm:(a,m)∈Mσ].\mathcal{R}_s(D_\sigma)=A[D_{\tau(a)}T^m:(a,m)\in M_\sigma].

and the definition of Hσ+H_\sigma^+ in (5.0.3) further gives

A[Dτ(a)Tm:(a,m)∈Mσ]=A[Dτ(a)Tm:(a,m)∈Hσ+]A[D_{\tau(a)}T^m:(a,m)\in M_\sigma] =A[D_{\tau(a)}T^m:(a,m)\in H_\sigma^+]

in view of the crucial identity

Dτ(a+a′)=Dτ(a)Dτ(a′).D_{\tau(a+a')} = D_{\tau(a)}D_{\tau(a')}.

In particular, taking minimal generators for each of the ideals Dτ(a)D_{\tau(a)} with (a,m)∈Hσ+(a,m) \in H^+_{\sigma} gives a finite set of AA-algebra generators for the symbolic Rees algebra.

For normality, let K=Frac⁡(A)K=\operatorname{Frac}(A) and put cj=γj(σ)c_j=\gamma_j(\sigma). By [6 Ch. VI, §10, no. 1, Lemma 1], the valuation vjv_j from Proposition 3.5.1 extends to a valuation wjw_j of K(T)K(T) with wj(T)=−cjw_j(T)=-c_j given by the following formula on A[T]A[T]

wj(∑mfmTm)=min⁡m{vj(fm)−mcj}.w_j\left(\sum_m f_mT^m\right)=\min_m\{v_j(f_m)-mc_j\}.

The formula above extends to rational functions. To see multiplicativity directly, choose a uniformizer π\pi for vjv_j, rescale TT by its cjc_j-th power, i.e., set U=πcjTU=\pi^{c_j}T and observe that the coefficient of UmU^m in the polynomial ∑mfmTm\sum_m f_mT^m now has valuation vj(fm)−mcjv_j(f_m)-mc_j. Thus the weighted formula becomes the ordinary minimum of coefficient valuations in the new variable UU and multiplicativity follows by reducing coefficients to the domain κ[U]\kappa[U] where κ\kappa is the residue field of the valuation ring of vjv_j. If VwjV_{w_j} is its valuation ring, Theorem 4.2.1 gives the equality inside K(T)K(T)

Rs(Dσ)=A[T]∩⋂j=βσ1Vwj.(37)R_s(D_\sigma)=A[T]\cap\bigcap_{j=\beta}^{\sigma_1}V_{w_j}. \tag*{(37)}

Indeed, for a polynomial the conditions are exactly vj(fm)≥mcjv_j(f_m)\ge mc_j for every coefficient and every retained jj. The rings on the right are integrally closed. Their intersection is therefore integrally closed. Since Dσ≠0D_\sigma\ne0, the fraction field of the left-hand side in (5.0.6) is K(T)K(T): if 0≠f∈Dσ0\ne f\in D_\sigma, then T=(fT)/fT=(fT)/f lies in that fraction field. This proves normality.

For the generic, symmetric, and alternating cases let GG be GgenG_{\mathrm{gen}} or GL⁡(V)\operatorname{GL}(V), respectively. Each Dσ(m)D_\sigma^{(m)} is GG-stable and is a direct sum of distinct irreducible summands of AA. Let k∗\mathbb{k}^* act with weight mm on the factor TmT^m. Then Rs(Dσ)R_s(D_\sigma) is multiplicity-free as a G×k∗G\times\mathbb{k}^*-module: a given irreducible of GG occurs at most once in a fixed TT-degree, and different TT-degrees have different torus characters. The finite generation and normality already proved show that its spectrum is a normal affine spherical variety, by [27 Theorem 2.1.2]. Spherical varieties have rational singularities and are Cohen–Macaulay [27 Corollary 2.3.4]. □

The following examples illustrate the sum formula and the Hilbert-basis construction in Theorem 5.0.4. They also show that higher symbolic degrees can be genuinely necessary.

Example 5.0.7. Set q=3q=3 and σ=(3,2)\sigma=(3,2), which leads to β=2\beta=2, γ2(σ)=3\gamma_2(\sigma)=3, and γ3(σ)=1\gamma_3(\sigma)=1. Writing a=(a,b)a=(a,b), equation (4.2.5) becomes

D(3,2)(m)=∑a,b≥0a+2b≥3mb≥mp2ap3b.D_{(3,2)}^{(m)}=\sum_{\substack{a,b\ge0\\a+2b\ge3m\\b\ge m}}\mathfrak{p}_2^a\mathfrak{p}_3^b.

Removing nonminimal summands gives the closed formula

D(3,2)(m)=∑j=0⌊m/2⌋D(3m+j,2m−2j).D_{(3,2)}^{(m)}=\sum_{j=0}^{\lfloor m/2\rfloor}D_{(3m+j,2m-2j)}.

Consequently,

Rs(D(3,2))=A[D(3,2)T,D(3,3,3)T2]=A[p2p3T,p3T2].(38)R_s(D_{(3,2)})=A[D_{(3,2)}T,D_{(3,3,3)}T^2]=A[\mathfrak{p}_2\mathfrak{p}_3T,\mathfrak{p}_3T^2]. \tag*{(38)}

The TT-degree two generator is necessary whenever the indicated products are nonzero: it contains elements of polynomial degree 99, whereas D(3,2)2D_{(3,2)}^2 starts in degree 1010. In the alternating setting the

degree-one generators are products of a 4-Pfaffian and a 6-Pfaffian, while the degree-two generators are products of three 6-Pfaffians.

Example 5.0.9. Consider the rectangular cases (q,σ)=(4,(3,3))(q,\sigma)=(4,(3,3)) and (q,σ)=(5,(4,4))(q,\sigma)=(5,(4,4)). Put r=q−1r=q-1, so r=3r=3 or r=4r=4, and let U=pr−1U=\mathfrak p_{r-1}, V=prV=\mathfrak p_r, and W=pr+1W=\mathfrak p_{r+1}. Here β=r−1\beta=r-1, and for multiplicities (a,b,c)(a,b,c) for these prime ideals the feasibility conditions in (5.0.2) are

a+2b+3c≥4m,b+2c≥2m.(39)a+2b+3c\ge4m,\qquad b+2c\ge2m. \tag*{(39)}

A Hilbert basis of the auxiliary monoid MσM_\sigma in (5.0.2) has positive-mm elements

(1,0,1;1),(0,2,0;1),(0,1,1;1),(0,0,2;1),(0,1,2;2),(0,0,3;2),(0,0,4;3).(1,0,1;1),(0,2,0;1),(0,1,1;1),(0,0,2;1),(0,1,2;2),(0,0,3;2),(0,0,4;3).

At the ideal level, nesting W⊆VW\subseteq V and the straightening containment UW⊆V2UW\subseteq V^2 remove several of the corresponding ideal products which would produce redundant generators for the symbolic Rees ring. Thus

Rs(D(r,r))=A[V2T,VW2T2,W4T3].\mathcal{R}_s(D_{(r,r)})=A[V^2T,VW^2T^2,W^4T^3].

Equivalently,

D(r,r)(m)=∑u+2v+3w=mD((r+1)2v+4w,r2u+v).D_{(r,r)}^{(m)}=\sum_{u+2v+3w=m}D_{((r+1)^{2v+4w},r^{2u+v})}.

In particular, depending on the value of qq we obtain

q=4:Rs(D(3,3))=A[D(3,3)T,D(4,4,3)T2,D(4,4,4,4)T3],q=4:\quad\mathcal{R}_s(D_{(3,3)})=A[D_{(3,3)}T,D_{(4,4,3)}T^2,D_{(4,4,4,4)}T^3],
q=5:Rs(D(4,4))=A[D(4,4)T,D(5,5,4)T2,D(5,5,5,5)T3].q=5:\quad\mathcal{R}_s(D_{(4,4)})=A[D_{(4,4)}T,D_{(5,5,4)}T^2,D_{(5,5,5,5)}T^3].

The TT-degree two and TT-degree three generators cannot be omitted. Indeed, in the generic matrix case VW2VW^2 starts in polynomial degree 3r+2<4r3r+2<4r, the initial degree of (V2)2(V^2)^2, while W4W^4 starts in degree 4r+44r+4, which is smaller than the initial degrees 6r6r and 5r+25r+2 of the two degree-three products available from the first two families. The same formulas apply in the symmetric, alternating, and Hankel settings after replacing the rank ideals by the corresponding minors or Pfaffians; for Hankel this uses the admissible rectangular starting shape.

Skew Waldschmidt constants and asymptotic resurgence

For a nonzero ideal J⊆AJ\subseteq A and 1≤j≤q1\le j\le q define

νj(J)=max⁡{a∈N:J⊆pj(a)},pj(0)=A.\nu_j(J)=\max\{a\in\mathbb{N}:J\subseteq\mathfrak p_j^{(a)}\},\qquad\mathfrak p_j^{(0)}=A.

Equivalently, if JJ has a standard basis and 1≤j≤q1\le j\le q, set

νj(J)=min⁡{γj(λ):J contains a standard monomial of shape λ}.\nu_j(J)=\min\{\gamma_j(\lambda):J\text{ contains a standard monomial of shape }\lambda\}.

The two definitions coincide since pj\mathfrak p_j-adic order is detected by the combinatorial function γj\gamma_j by Theorem 3.6.2(iii).

Define the skew Waldschmidt constants associated to the functions νj\nu_j as

ν^j(Dσ)=lim⁡m→∞νj(Dσ(m))m.\widehat{\nu}_j(D_\sigma)=\lim_{m\to\infty}\frac{\nu_j(D_\sigma^{(m)})}{m}.

These limits will be computed explicitly below and therefore exist. Since the polynomial degree of a standard monomial of shape λ\lambda is ∣λ∣=γ1(λ)|\lambda|=\gamma_1(\lambda), one recovers the Waldschmidt constant as

ν^1(Dσ)=α^(Dσ).\widehat{\nu}_1(D_\sigma)=\widehat{\alpha}(D_\sigma).

The standard monomial basis of DσrD_\sigma^r consists precisely of the standard monomials whose shapes λ\lambda satisfy

γj(λ)≥rγj(σ)for 1≤j≤q.\gamma_j(\lambda)\ge r\gamma_j(\sigma)\qquad\text{for }1\le j\le q.

By contrast, (4.2.3) says that to find the standard monomial basis of Dσ(m)D_{\sigma}^{(m)} the indices j<βq(σ)j<\beta_q(\sigma) are no longer explicit constraints. The next result computes how much of those missing inequalities is nevertheless forced by the surviving ones.

Theorem 6.0.1. For an admissible diagram σ=(σ1,…,σs)\sigma=(\sigma_1,\ldots,\sigma_s) put β=βq(σ)\beta=\beta_q(\sigma), h=q−β+1h=q-\beta+1. Then for every m≥1m\ge1 and 1≤j≤q1\le j\le q we have

νj(Dσ(m))={mγβ(σ)+(β−j)⌈mγβ(σ)h⌉if j<β,mγj(σ)if j≥β.(40)\nu_j\left(D_{\sigma}^{(m)}\right)= \begin{cases} m\gamma_{\beta}(\sigma)+(\beta-j)\left\lceil\frac{m\gamma_{\beta}(\sigma)}{h}\right\rceil& \text{if }j<\beta,\\ m\gamma_j(\sigma) & \text{if }j\ge\beta. \end{cases} \tag*{(40)}

In particular,

α(Dσ(m))=mγβ(σ)+(β−1)⌈mγβ(σ)h⌉.(41)\alpha\left(D_{\sigma}^{(m)}\right)=m\gamma_{\beta}(\sigma)+(\beta-1)\left\lceil\frac{m\gamma_{\beta}(\sigma)}{h}\right\rceil. \tag*{(41)}

Consequently the skew Waldschmidt constants of the product of rank ideals DσD_{\sigma} are given by

ν^j(Dσ)={q−j+1q−β+1γβ(σ),j<β,γj(σ),j≥β.(42)\widehat{\nu}_j(D_{\sigma})= \begin{cases} \frac{q-j+1}{q-\beta+1}\gamma_{\beta}(\sigma), & j<\beta,\\ \gamma_j(\sigma), & j\ge\beta. \end{cases} \tag*{(42)}

In particular,

α^(Dσ)=qγβ(σ)q−β+1.(43)\widehat{\alpha}(D_{\sigma})=\frac{q\gamma_{\beta}(\sigma)}{q-\beta+1}. \tag*{(43)}

Proof. For j≥βj\ge\beta, (4.2.3) gives νj(Dσ(m))≥mγj(σ)\nu_j\left(D_{\sigma}^{(m)}\right)\ge m\gamma_j(\sigma), while Dσm⊆Dσ(m)D_{\sigma}^m\subseteq D_{\sigma}^{(m)} and the shape σ∗m\sigma^{*m} in the standard monomial basis of DσmD_{\sigma}^m gives equality νj(Dσ(m))=mγj(σ)\nu_j\left(D_{\sigma}^{(m)}\right)=m\gamma_j(\sigma) and settles the last case of (6.0.2) and (6.0.4).

Now suppose j<βj<\beta and let λ\lambda be the shape of a standard monomial in Dσ(m)D_{\sigma}^{(m)}. Put E=γβ(λ)E=\gamma_{\beta}(\lambda) and N=nβ(λ)=#{i:λi≥β}N=n_{\beta}(\lambda)=\#\{i:\lambda_i\ge\beta\}. Each part counted in NN contributes at most hh to EE, and E≥mγβ(σ)E\ge m\gamma_{\beta}(\sigma) by (4.2.3). Hence

N≥⌈Eh⌉≥⌈mγβ(σ)h⌉N\ge\left\lceil\frac{E}{h}\right\rceil\ge\left\lceil\frac{m\gamma_{\beta}(\sigma)}{h}\right\rceil

and

γj(λ)≥E+(β−j)N≥mγβ(σ)+(β−j)⌈mγβ(σ)h⌉.\gamma_j(\lambda)\ge E+(\beta-j)N\ge m\gamma_{\beta}(\sigma)+(\beta-j)\left\lceil\frac{m\gamma_{\beta}(\sigma)}{h}\right\rceil.

For equality, write mγβ(σ)=kh+rm\gamma_{\beta}(\sigma)=kh+r, 0≤r<h0\le r<h, and define a partition with kk or k+1k+1 parts by

θ={(qk),if r=0,(qk,β−1+r),if r>0.(44)\theta= \begin{cases} (q^k), & \text{if }r=0,\\ (q^k,\beta-1+r), & \text{if }r>0. \end{cases} \tag*{(44)}

Deleting the first β−1\beta-1 boxes of each row of σ∗m\sigma^{*m} gives a partition of mγβ(σ)m\gamma_{\beta}(\sigma) with parts at most hh. The corresponding tail of θ\theta is (hk,r)(h^k,r). This is the dominance-maximal partition of kh+rkh+r subject to the bound hh on part size. Lemma 4.1.1 gives along the same lines as in the proof of Lemma 4.1.4 that

γi(θ)≥mγi(σ)for i≥β.\gamma_i(\theta)\ge m\gamma_i(\sigma)\qquad\text{for }i\ge\beta.

Thus a standard monomial of shape θ\theta belongs to Dσ(m)D_{\sigma}^{(m)}. It has E=mγβ(σ)E=m\gamma_{\beta}(\sigma) and N=⌈mγβ(σ)/h⌉N=\left\lceil m\gamma_{\beta}(\sigma)/h\right\rceil, giving simultaneous equality for every j<βj<\beta. This proves (6.0.2). Taking j=1j=1 gives (6.0.3). Dividing by mm and using lim⁡m→∞⌈mγβ(σ)/h⌉/m=γβ(σ)/h\lim_{m\to\infty}\left\lceil m\gamma_{\beta}(\sigma)/h\right\rceil/m=\gamma_{\beta}(\sigma)/h proves (6.0.4). □

Next we solve the containment problem and determine the asymptotic resurgence for products of prime ideals.

Theorem 6.0.7. Let σ\sigma be an admissible partition. put β=βq(σ)\beta=\beta_q(\sigma) and h=q−β+1h=q-\beta+1, and let m=A+=p1m=A_+=\mathfrak{p}_1. Then for all m,r≥1m,r\ge1 and N≥0N\ge0,

Dσ(m)⊆mNDσr⟺mγβ(σ)+(β−1)⌈mγβ(σ)h⌉≥N+r∣σ∣.(45)D_{\sigma}^{(m)}\subseteq\mathfrak{m}^{N}D_{\sigma}^{r} \quad\Longleftrightarrow\quad m\gamma_{\beta}(\sigma)+(\beta-1)\left\lceil\frac{m\gamma_{\beta}(\sigma)}{h}\right\rceil\ge N+r|\sigma|. \tag*{(45)}

In particular, the following is a formula for the resurgence and asymptotic resurgence of a product of prime rank ideals

ρ(Dσ)=ρ^(Dσ)=∣σ∣hqγβ(σ)=α(Dσ)α^(Dσ).(46)\rho(D_\sigma)=\widehat{\rho}(D_\sigma)=\frac{|\sigma|h}{q\gamma_\beta(\sigma)}=\frac{\alpha(D_\sigma)}{\widehat{\alpha}(D_\sigma)}. \tag*{(46)}

Proof. Let ν=σ∗r∗(1N)\nu=\sigma^{*r}*(1^N). Then Dν=mNDσrD_\nu=\mathfrak{m}^N D_\sigma^r. By Theorem 3.6.2(iv), a standard monomial of shape λ\lambda is in DνD_\nu, if and only if it satisfies the following inequalities

γ1(λ)≥N+r∣σ∣,γj(λ)≥rγj(σ)for 2≤j≤q.(47)\gamma_1(\lambda)\geq N+r|\sigma|,\qquad\gamma_j(\lambda)\geq r\gamma_j(\sigma)\qquad\text{for }2\leq j\leq q. \tag*{(47)}

Since the initial degree of the ideal mNDσr\mathfrak{m}^N D_\sigma^r is γ1(ν)=N+r∣σ∣\gamma_1(\nu)=N+r|\sigma| and the initial degree of Dσ(m)D_\sigma^{(m)} appears in (6.0.3), the given containment implies the inequality of initial degrees in (6.0.8).

Conversely, assume that inequality and put am=⌈mγβ(σ)/h⌉a_m=\lceil m\gamma_\beta(\sigma)/h\rceil. A nonzero product of shape σ∗m\sigma^{*m} shows α(Dσ(m))≤m∣σ∣\alpha(D_\sigma^{(m)})\leq m|\sigma|. The assumed degree inequality, together with the formula for α(Dσ(m))\alpha(D_\sigma^{(m)}) in (6.0.8) gives N+r∣σ∣≤α(Dσ(m))N+r|\sigma|\leq\alpha(D_\sigma^{(m)}). Combining these we deduce the inequality

N+r∣σ∣≤m∣σ∣,hencer≤m.N+r|\sigma|\leq m|\sigma|,\qquad\text{hence}\qquad r\leq m.

For j≥βj\geq\beta, Theorem 6.0.1 gives νj(Dσ(m))=mγj(σ)≥rγj(σ)\nu_j(D_\sigma^{(m)})=m\gamma_j(\sigma)\geq r\gamma_j(\sigma), where the inequality follows from that displayed above. For j<βj<\beta, use

γj(σ)=γβ(σ)+(β−j)sfor 1≤j≤β\gamma_j(\sigma)=\gamma_\beta(\sigma)+(\beta-j)s\qquad\text{for }1\leq j\leq\beta

to obtain

νj(Dσ(m))−rγj(σ)=(m−r)γβ(σ)+(β−j)(am−rs).\nu_j(D_\sigma^{(m)})-r\gamma_j(\sigma)=(m-r)\gamma_\beta(\sigma)+(\beta-j)(a_m-rs).

The right side is an affine function of x=β−jx=\beta-j on [0,β−1][0,\beta-1]. At x=0x=0 it is nonnegative, and at x=β−1x=\beta-1 it is α(Dσ(m))−r∣σ∣≥N≥0\alpha(D_\sigma^{(m)})-r|\sigma|\geq N\geq0. Hence it is nonnegative throughout the interval. Together with the assumed degree inequality this proves that (6.0.10) holds for standard monomials of shape λ\lambda in Dσ(m)D_\sigma^{(m)}. Since both Dσ(m)D_\sigma^{(m)} and mNDσr\mathfrak{m}^N D_\sigma^r have standard monomial bases and we have shown that the standard monomials of the former are in the latter, containment follows.

Put R=(∣σ∣h)/(qγβ(σ))R=(|\sigma|h)/(q\gamma_\beta(\sigma)) and A0=qγβ(σ)/hA_0=q\gamma_\beta(\sigma)/h. If m/r≥Rm/r\geq R, then

mr≥R=∣σ∣A0somA0≥r∣σ∣,\frac{m}{r}\geq R=\frac{|\sigma|}{A_0}\qquad\text{so}\qquad mA_0\geq r|\sigma|,

and

α(Dσ(m))=mγβ(σ)+(β−1)⌈mγβ(σ)h⌉≥mγβ(σ)+(β−1)mγβ(σ)h=mqγβ(σ)h=mA0.\begin{aligned} \alpha(D_\sigma^{(m)})&=m\gamma_\beta(\sigma)+(\beta-1)\left\lceil\frac{m\gamma_\beta(\sigma)}{h}\right\rceil\\ &\geq m\gamma_\beta(\sigma)+(\beta-1)\frac{m\gamma_\beta(\sigma)}{h}\\ &=m\frac{q\gamma_\beta(\sigma)}{h}=mA_0. \end{aligned}

Overall, α(Dσ(m))≥mA0≥r∣σ∣\alpha(D_\sigma^{(m)})\geq mA_0\geq r|\sigma|, so by the previous criterion containment holds and ρ(Dσ)≤R\rho(D_\sigma)\leq R.

If m/r<Rm/r<R, then

lim⁡t→∞α(Dσ(mt))t=mA0<r∣σ∣.\lim_{t\to\infty}\frac{\alpha(D_\sigma^{(mt)})}{t}=mA_0<r|\sigma|.

Thus Dσ(mt)⊈DσrtD_\sigma^{(mt)}\not\subseteq D_\sigma^{rt} for all sufficiently large tt. Rational numbers m/r<Rm/r<R approach RR from below, giving ρ^(Dσ)≥R\widehat{\rho}(D_\sigma)\geq R. Since ρ^≤ρ\widehat{\rho}\leq\rho, (6.0.9) follows after observing that α(Dσ)=∣σ∣\alpha(D_\sigma)=|\sigma|. □\square

References

Department of Mathematics, Indian Institute of Science Education and Research Bhopal, Bhopal Bypass Road, Bhauri, Bhopal 462066, Madhya Pradesh, India Email address: [email protected]

Department of Mathematics, University of Utah, 155 South 1400 East, Salt Lake City, UT 84112, USA Email address: [email protected]

Department of Mathematics, University of Nebraska–Lincoln, 203 Avery Hall, Lincoln, NE 68588, USA Email address: [email protected]

Department of Mathematics, University of Nebraska–Lincoln, 203 Avery Hall, Lincoln, NE 68588, USA Email address: [email protected]

Department of Mathematics, University of Nebraska–Lincoln, 203 Avery Hall, Lincoln, NE 68588, USA Email address: [email protected]

References

  1. [1]S. Abeasis, Gli ideali GL(V)-invarianti in S(S²V), Rend. Mat. (6) 13 (1980), 235–262.
  2. [2]S. Abeasis and A. Del Fra, Young diagrams and ideals of Pfaffians, Adv. Math. 35 (1980), 158–178.
  3. [4]S. Bisui and A. Seceleanu, Symbolic Powers and Asymptotic Invariants of GL-Invariant Ideals, arXiv:2609.03131, 2026.arxiv.org/abs/2609.03131
  4. [5]C. Bocci and B. Harbourne, Comparing powers and symbolic powers of ideals, J. Algebraic Geom. 19 (2010), 399–417.
  5. [6]N. Bourbaki, Commutative Algebra. Chapters 1–7, Elements of Mathematics, Springer-Verlag, Berlin, 1989.
  6. [9]W. Bruns and U. Vetter, Determinantal Rings, Lecture Notes in Mathematics, vol. 1327, Springer, Berlin, 1988.
  7. [11]A. Conca, Straightening law and powers of determinantal ideals of Hankel matrices, Adv. Math. 138 (1998), 263–292.DOI
  8. [12]A. Conca, M. Mostafazadehfard, A. K. Singh, and M. Varbaro, Hankel determinantal rings have rational singularities, Adv. Math., 335 (2018), 111-129, ISSN 0001-8708.arxiv.org/abs/1709.05685
  9. [13]C. De Concini, D. Eisenbud, and C. Procesi, Young diagrams and determinantal varieties, Invent. Math. 56 (1980), 129–165.
  10. [14]C. De Concini and C. Procesi, A characteristic free approach to invariant theory, Adv. Math. 21 (1976), 330–354.
  11. [15]M. DiPasquale, C. A. Francisco, J. Mermin, and J. Schweig, Asymptotic resurgence via integral closures, Trans. Amer. Math. Soc. 372 (2019), 6655–6676.arxiv.org/abs/1808.01547
  12. [16]M. DiPasquale, T. T. Nguyễn, and A. Seceleanu, Duality for asymptotic invariants of graded families, Adv. Math. 430 (2023), 109208.arxiv.org/abs/2208.11110
  13. [18]D. Eisenbud, Linear Sections of Determinantal Varieties, American Journal of Mathematics 110, no. 3 (1988): 541–75.DOI
  14. [19]E. Guardo, B. Harbourne, and A. Van Tuyl, Asymptotic resurgences for ideals of positive dimensional subschemes, Adv. Math. 246 (2013), 114–127.arxiv.org/abs/1202.4370
  15. [20]L. Gruson and C. Peskine, Courbes de l’espace projectif: variétés de sécantes, in: Enumerative geometry and classical algebraic geometry, Progr. Math., Birkhäuser, Boston, 24 (1982), 1–31.DOI
  16. [21]I. B. A. Henriques and M. Varbaro, Test, multiplier and invariant ideals, Adv. Math. 287 (2016), 704–732.arxiv.org/abs/1407.4324
  17. [24]M. Hochster and J. L. Roberts, Rings of invariants of reductive groups acting on regular rings are Cohen–Macaulay, Adv. Math. 13 (1974), 115–175.DOI
  18. [25]A. Kumar and V. Mukundan, Symbolic powers of classical varieties, Israel J. Math. (2026).arxiv.org/abs/2402.18693
  19. [26]L. D. Nam, The determinantal ideals of extended Hankel matrices, J. Pure Appl. Algebra 215 (2011), 1502–1515.arxiv.org/abs/1008.3843
  20. [27]N. Perrin, On the geometry of spherical varieties, Transform. Groups 19 (2014), 171–223.arxiv.org/abs/1211.1277
  21. [28]T. G. Room, The geometry of determinantal loci, Cambridge University Press, Cambridge, 1938.
  22. [29]J. Watanabe, Hankel Matrices and Hankel Ideals, Proceedings of the School of Science, Tokai University, 32 (1997), 11–21.

Paper details

vlnc:W_kf64574x
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

Generative AI

View statement
Looking for related papers…
Contents