betaFeedback

WorksMathematics

The Cooperstein conjecture and additive alternating dd-codes

Abstract

Let ⋀2V\bigwedge^2 V be the second exterior power of a vector space VV over a finite field. The Cooperstein conjecture asserts a dimension bound for flats external to varieties consisting of points of bounded rank in PG(⋀2V)\mathrm{PG}(\bigwedge^2 V). This geometric problem can be translated into a question about linear spaces of alternating matrices whose nonzero elements have rank at least a prescribed value. For matrices of odd order, the sharp cardinality bound was already established by Delsarte and Goethals (1975). In this paper, we prove the Cooperstein conjecture for matrices of even order and, more generally, we prove its additive analogue by determining, for all admissible parameters, the maximum size of an additive subgroup of alternating matrices in which every nonzero matrix has at least the prescribed rank. Our proof uses Pfaffian equations to characterize alternating matrices of rank below the prescribed minimum. We study the minimal graded free resolutions of the corresponding Pfaffian rings. The Cohen–Macaulay property and the aa-invariant of each Pfaffian ring determine the largest shift in its resolution. We use trace functions on evaluation quotients to construct polynomial indicators for the matrices of small rank in an additive subgroup. The largest shift controls their degrees, yielding the upper bound by a counting argument. Finally, we give explicit trace constructions attaining the bound.

Introduction

Let F\mathbb{F} be the finite field of order q=peq=p^{e}, where pp is prime and ee is a positive integer. Let Mn(F)M_n(\mathbb{F}) denote the F\mathbb{F}-vector space of n×nn \times n matrices over F\mathbb{F}. Let V=⟨v1,…,vn⟩FV=\langle v_1,\ldots,v_n\rangle_{\mathbb{F}} be an nn-dimensional vector space over F\mathbb{F}, and set V2=⋀2VV_2=\bigwedge^{2}V, the second exterior power of VV, with basis B={vij=vi∧vj:1≤i<j≤n}\mathcal{B}=\{v_{ij}=v_i\wedge v_j:1\leq i<j\leq n\}. Here ∧\wedge denotes the exterior product. We have dim⁡FV2=N=(n2)\dim_{\mathbb{F}}V_2=N=\binom{n}{2}. Set m=⌊n/2⌋m=\lfloor n/2\rfloor. Every nonzero x∈V2x\in V_2 can be written as

x=u1∧u2+u3∧u4+⋯+u2s−1∧u2s,x=u_1\wedge u_2+u_3\wedge u_4+\cdots+u_{2s-1}\wedge u_{2s},

where u1,…,u2su_1,\ldots,u_{2s} are linearly independent. The integer ss is uniquely determined by xx and is called the rank of xx; it satisfies 1≤s≤m1\leq s\leq m. Multiplication by a nonzero scalar preserves this rank. Consequently, it also defines a rank on the points ⟨x⟩F\langle x\rangle_{\mathbb{F}} of PG⁡(V2)\operatorname{PG}(V_2). For 1≤r≤m−11\leq r\leq m-1, let ZrZ_r be the set of points of PG⁡(V2)\operatorname{PG}(V_2) of rank at most rr. An F\mathbb{F}-linear subspace W⊆V2W\subseteq V_2 determines an external flat to the variety ZrZ_r if PG⁡(W)∩Zr=∅\operatorname{PG}(W)\cap Z_r=\varnothing. Cooperstein studied the largest possible dimension of such subspaces and proposed the following conjecture [3 p. 224].

Conjecture 1.1. Let 1≤r≤m−11\leq r\leq m-1, and let W⊆V2W\subseteq V_2 be an F\mathbb{F}-linear subspace such that PG⁡(W)∩Zr=∅\operatorname{PG}(W)\cap Z_r=\varnothing. Then

dim⁡FW≤N−nr.(1)\dim_{\mathbb{F}}W\leq N-nr. \tag*{(1)}

This problem has an equivalent formulation in terms of alternating matrices. A matrix A∈Mn(F)A \in M_n(\mathbb{F}) is said to be alternating if AT=−AA^{\mathsf{T}}=-A and all its diagonal entries are zero, where ATA^{\mathsf{T}} denotes the transpose of AA. We denote by Alt⁡n(F)\operatorname{Alt}_n(\mathbb{F}) the F\mathbb{F}-vector space of alternating matrices in Mn(F)M_n(\mathbb{F}). The basis B\mathcal{B} of V2V_2 defines an F\mathbb{F}-linear isomorphism

μB:∑i<jaijvi∧vj∈V2⟼A∈Alt⁡n(F),\mu_{\mathcal{B}}:\sum_{i<j}a_{ij}v_i\wedge v_j\in V_2\longmapsto A\in\operatorname{Alt}_n(\mathbb{F}),

where Aij=aij,Aji=−aijA_{ij}=a_{ij}, A_{ji}=-a_{ij}, for i<ji<j and Aii=0A_{ii}=0. Under this isomorphism, an element x∈V2x\in V_2 of rank ss corresponds to a matrix of rank 2s2s; see e.g. [3 Section 3]. This correspondence leads naturally to the notion of an alternating dd-code. The rank distance on Alt⁡n(F)\operatorname{Alt}_n(\mathbb{F}) is drk(A,B)=rk(A−B)d_{\rm rk}(A,B)={\rm rk}(A-B). A nonempty subset C⊆Alt⁡n(F)C\subseteq\operatorname{Alt}_n(\mathbb{F}) is a dd-code if drk(A,B)≥dd_{\rm rk}(A,B)\ge d for all distinct A,B∈CA,B\in C. It is additive if it is an additive subgroup, or equivalently an Fp\mathbb{F}_p-linear subspace. Since alternating matrices have even rank, we consider even dd with 2≤d≤2m2\le d\le2m. The condition on WW in Conjecture 1.1 says precisely that μB(W)\mu_{\mathcal{B}}(W) is an F\mathbb{F}-linear (2r+2)(2r+2)-code.

The extremal problem for alternating dd-codes forms part of the broader study of rank-metric codes in classical form spaces, namely the spaces of alternating and symmetric bilinear forms and of Hermitian sesquilinear forms. After choosing bases, these are represented by alternating, symmetric, and Hermitian matrices, respectively. A maximum additive dd-code is an additive dd-code of largest possible cardinality in its ambient form space. Schmidt established bounds and constructions for symmetric dd-codes in odd and even characteristic [23, 24, 26] and for Hermitian dd-codes [25], see also [16, 28]. In the Hermitian case, Santonastaso and Zhou Santonastaso and Zhou [22] showed that the upper bound of Schmidt [25] for additive dd-codes is attained for every admissible choice of parameters. In these settings, the maximum size of an arbitrary dd-code remains unknown for some parameters, even when the corresponding additive problem has been solved, see e.g. [4, 19]. Perfect codes in classical form spaces have also been studied in the symmetric, alternating, and Hermitian settings; see [21], [1], and [20], respectively.

For alternating dd-codes, Delsarte and Goethals studied the general problem without assuming additivity [5]. For d=2sd=2s, with 1≤s≤m1\le s\le m, their bound is

∣C∣≤qNm(m−s+1).(2)|C|\le q^{\frac{N}{m}(m-s+1)}. \tag*{(2)}

Suppose that nn is odd, so that N=nmN=nm. If WW is as in Conjecture 1.1, then μB(W)\mu_{\mathcal{B}}(W) is an F\mathbb{F}-linear (2r+2)(2r+2)-code. Applying (2) with s=r+1s=r+1 gives

qdim⁡FW=∣μB(W)∣≤qn(m−r)=qN−nr.q^{\dim_{\mathbb{F}}W}=|\mu_{\mathcal{B}}(W)|\le q^{n(m-r)}=q^{N-nr}.

Hence dim⁡FW≤N−nr\dim_{\mathbb{F}}W\le N-nr, as conjectured.

For even n=2mn=2m and d=2s≥4d=2s\ge4, Conjecture 1.1, with r=s−1r=s-1, predicts that an F\mathbb{F}-linear 2s2s-code has cardinality at most qm(2m−2s+1)q^{m(2m-2s+1)}. On the other hand, the exponent in the Delsarte–Goethals bound satisfies

(2m−1)(m−s+1)=m(2m−2s+1)+(s−1).(2m-1)(m-s+1)=m(2m-2s+1)+(s-1).

Thus, for s≥2s\ge2, the upper bound in (2) is larger than the conjectured upper bound by a factor of qs−1q^{s-1} and therefore does not imply Conjecture 1.1. Abiad, Alfarano and Ravagnani Abiad, Alfarano and Ravagnani [1] recovered the Delsarte–Goethals bound for d=4,6,8d=4,6,8 using spectral methods.

For F\mathbb{F}-linear codes, Cooperstein proved the conjectured bound for d=4d=4 [3 Theorem 6.1] and for d=nd=n when nn is even [3 Theorem 5.1]. These two cases were also treated by Gow and Quinlan [8]. Gow subsequently proved the conjectured linear bound for d=n−2d=n-2 when n>2n>2 is even [7 Theorem 7].

The special cases just discussed concern F\mathbb{F}-linear codes. For n=2mn=2m and d=2s≥4d=2s\geq4, the conjectured dimension bound for such a code CC is dim⁡FC≤m(2m−2s+1)\dim_{\mathbb{F}} C\leq m(2m-2s+1). Since ∣C∣=qdim⁡FC|C|=q^{\dim_{\mathbb{F}} C}, this gives a bound on the cardinality of CC. The additive analogue of Conjecture 1.1 asserts that the same cardinality bound holds for every additive 2s2s-code C⊆Alt⁡n(F)C\subseteq\operatorname{Alt}_{n}(\mathbb{F}), namely

∣C∣≤qm(2m−2s+1),1≤s≤m.(3)|C|\leq q^{m(2m-2s+1)},\qquad1\leq s\leq m. \tag*{(3)}

This assertion extends Conjecture 1.1, since every F\mathbb{F}-linear code is additive, but an additive subgroup need not be an F\mathbb{F}-linear subspace.

Our contribution. We prove Conjecture 1.1 and its additive analogue over every finite field, determining the maximum size of an additive alternating dd-code for all admissible parameters.

Theorem 1.2. Let F=Fq\mathbb{F}=\mathbb{F}_{q}, let n≥2n\geq2, and let dd be even with 2≤d≤2⌊n/2⌋2\leq d\leq2\lfloor n/2\rfloor. Every additive dd-code C⊆Alt⁡n(F)C\subseteq\operatorname{Alt}_{n}(\mathbb{F}) satisfies

∣C∣≤qn(n−d+1)/2.(4)|C|\leq q^{n(n-d+1)/2}. \tag*{(4)}

Moreover, equality is attained by an F\mathbb{F}-linear code for every such qq, nn and dd. Consequently, if F′⊆F\mathbb{F}'\subseteq\mathbb{F} is a subfield and CC is F′\mathbb{F}'-linear, then

dim⁡F′C≤[F:F′]n(n−d+1)2,(5)\dim_{\mathbb{F}'} C\leq[\mathbb{F}:\mathbb{F}']\frac{n(n-d+1)}{2}, \tag*{(5)}

and this bound is also sharp.

For WW as in Conjecture 1.1, applying Theorem 1.2 to μB(W)\mu_{B}(W) with d=2r+2d=2r+2 gives dim⁡FW≤n(n−2r−1)/2=N−nr\dim_{\mathbb{F}} W\leq n(n-2r-1)/2=N-nr.

The strategy of the proof is as follows. For even d≥4d\geq4, let SS be the polynomial ring over F\mathbb{F} in the independent entries of a generic n×nn\times n alternating matrix, and let IdI_{d} be the ideal generated by its Pfaffians of order dd. The zeros of IdI_{d} are precisely the matrices of rank at most d−2d-2 (Lemma 2.1). Our proof studies the quotient S/IdS/I_{d}, called a Pfaffian ring, with its standard grading and a minimal graded free resolution as an SS-module. The shifts in this resolution determine the Hilbert series, whose degree as a rational function is the aa-invariant. Writing N=(n2)N=\binom{n}{2}, the Cohen–Macaulay property of S/IdS/I_{d} implies that the largest shift equals N+a(S/Id)N+a(S/I_{d}). Proposition 3.1 gives a(S/Id)=−n(d−2)/2a(S/I_{d})=-n(d-2)/2, and hence the largest shift is B=n(n−d+1)/2B=n(n-d+1)/2. The largest shift BB controls the degree of the polynomial indicator used in our counting argument. Write q=peq=p^{e} and choose an Fp\mathbb{F}_{p}-basis A1,…,AtA_{1},\ldots,A_{t} of a nonzero additive subgroup W⊆Alt⁡n(F)W\subseteq\operatorname{Alt}_{n}(\mathbb{F}). Substituting A(y)=∑ν=1tyνAνA(y)=\sum_{\nu=1}^{t}y_{\nu}A_{\nu} into the Pfaffian equations gives a system whose solutions in Fpt\mathbb{F}_{p}^{t} correspond bijectively to the matrices in WW of rank at most d−2d-2. Using traces of Frobenius, we construct a polynomial of degree at most e(p−1)Be(p-1)B that equals 1 at these solutions and 0 elsewhere in Fpt\mathbb{F}_{p}^{t} (Theorem 4.6). If t>eBt>eB, or equivalently ∣W∣>qB|W|>q^{B}, the degree of this polynomial is less than t(p−1)t(p-1), so its sum over Fpt\mathbb{F}_{p}^{t} vanishes by Lemma 4.5. Consequently, the number of matrices in WW of rank at most d−2d-2 is divisible by pp. Since the zero matrix is counted, WW contains a nonzero matrix of rank at most d−2d-2 (Theorem 5.1). Thus an additive dd-code CC with ∣C∣>qB|C|>q^{B} would contain a nonzero matrix of rank less than dd, a contradiction. This proves the upper bound in Theorem 1.2 for d≥4d \ge4. The case d=2d = 2 is immediate. Finally, Proposition 5.3 gives F\mathbb{F}-linear codes attaining the bound for both parities of nn and in every characteristic.

Preliminaries

We recall the basic properties of alternating matrices and forms and Pfaffians, together with the notions of graded modules and free resolutions used in the sequel.

Alternating forms and Pfaffians

Let F=Fq\mathbb{F} = \mathbb{F}_q and let n≥2n \ge2 be an integer. Let V=V(n,F)V = V(n,\mathbb{F}) be an nn-dimensional F\mathbb{F}-vector space. A bilinear form B:V×V→FB : V \times V \to\mathbb{F} is said to be alternating if B(v,v)=0B(v,v) = 0 for every v∈Vv \in V. We denote by Alt⁡(V)\operatorname{Alt}(V) the F\mathbb{F}-vector space of alternating bilinear forms on VV. Fix an ordered F\mathbb{F}-basis E=(v1,…,vn)\mathcal{E} = (v_1,\ldots,v_n) of VV. The matrix representing a bilinear form BB in this basis is

ME(B)=(B(vi,vj))1≤i,j≤nM_{\mathcal{E}}(B) = \left(B(v_i,v_j)\right)_{1\le i,j\le n}

The form BB is alternating if and only if ME(B)M_{\mathcal{E}}(B) is alternating. Hence, after fixing E\mathcal{E}, the map

ΨE:Alt⁡(V)⟶Alt⁡n(F),B⟼ME(B),(6)\Psi_{\mathcal{E}} : \operatorname{Alt}(V) \longrightarrow\operatorname{Alt}_n(\mathbb{F}), \qquad B \longmapsto M_{\mathcal{E}}(B), \tag*{(6)}

is an F\mathbb{F}-linear isomorphism, see [15 Chapter XIII, §6, Proposition 6.5]. In particular,

dim⁡FAlt⁡(V)=dim⁡FAlt⁡n(F)=n(n−1)2,\dim_{\mathbb{F}}\operatorname{Alt}(V) = \dim_{\mathbb{F}}\operatorname{Alt}_n(\mathbb{F}) = \frac{n(n-1)}{2},

since an alternating matrix is uniquely determined by its entries above the diagonal. We define rk⁡(B):=rk⁡(ME(B))\operatorname{rk}(B) := \operatorname{rk}(M_{\mathcal{E}}(B)). This definition is independent of the chosen basis. Every alternating form BB admits an ordered basis E′\mathcal{E}' in which its representing matrix is

ME′(B)=diag⁡(J,…,J⏟r blocks,0n−2r)=(01−10⋱01−100n−2r),J=(01−10),M_{\mathcal{E}'}(B) = \operatorname{diag}(\underbrace{J,\ldots,J}_{r\ \text{blocks}},0_{n-2r}) = \begin{pmatrix} 0 & 1 & & & & \\ -1 & 0 & & & & \\ & & \ddots& & & \\ & & & 0 & 1 & \\ & & & -1 & 0 & \\ & & & & & 0_{n-2r} \end{pmatrix}, \qquad J = \begin{pmatrix} 0 & 1\\ -1 & 0 \end{pmatrix},

for some integer 0≤r≤⌊n/2⌋0 \le r \le\lfloor n/2\rfloor; see [15 Chapter XV, §8, Theorem 8.1, pp. 586–587]. Here 0n−2r0_{n-2r} denotes the zero matrix of order n−2rn-2r. Consequently, rk⁡(B)=2r\operatorname{rk}(B) = 2r, so every alternating form, and hence every alternating matrix, has even rank.

We endow Alt⁡(V)\operatorname{Alt}(V) with the rank metric, defined by

drk⁡(B1,B2):=rk⁡(B1−B2)=rk⁡(ME(B1)−ME(B2)),d_{\operatorname{rk}}(B_1,B_2) := \operatorname{rk}(B_1-B_2) = \operatorname{rk}(M_{\mathcal{E}}(B_1)-M_{\mathcal{E}}(B_2)),

for all B1,B2∈Alt⁡(V)B_1,B_2 \in\operatorname{Alt}(V). Via (6), we use the same notation for the rank metric on Alt⁡n(F)\operatorname{Alt}_n(\mathbb{F}). Thus an alternating (rank-metric) code is a nonempty subset C⊆Alt⁡(V)\mathcal{C} \subseteq\operatorname{Alt}(V), or, after fixing E\mathcal{E}, a nonempty subset C⊆Alt⁡n(F)\mathcal{C} \subseteq\operatorname{Alt}_n(\mathbb{F}). If ∣C∣≥2|\mathcal{C}| \ge2, its minimum distance is

drk⁡(C):=min⁡{rk⁡(B1−B2):B1,B2∈C, B1≠B2}.d_{\operatorname{rk}}(\mathcal{C}) := \min\{\operatorname{rk}(B_1-B_2) : B_1,B_2 \in\mathcal{C},\ B_1 \ne B_2\}.

For a singleton code, we set drk⁡(C)=+∞d_{\operatorname{rk}}(\mathcal{C}) = +\infty. Since every alternating matrix has even rank, we take dd to be an even integer satisfying 2≤d≤2⌊n/2⌋2 \le d \le2\lfloor n/2\rfloor. We call C\mathcal{C} an **alternating (rank-metric) dd-code if d⁡rk(C)≥d\operatorname{d}_{\mathrm{rk}}(\mathcal{C}) \ge d, equivalently if the difference of any two distinct codewords has rank at least dd. An additive alternating (rank-metric) code is an alternating rank-metric code that is an additive subgroup of Alt⁡n(F)\operatorname{Alt}_{n}(\mathbb{F}). For such a code, the dd-code condition is equivalent to requiring that every nonzero codeword has rank at least dd. The corresponding extremal problem is to maximize ∣C∣|\mathcal{C}| among additive alternating dd-codes in Alt⁡n(F)\operatorname{Alt}_{n}(\mathbb{F}).

We recall the definition of the Pfaffian of an alternating matrix. Let t≥1t \ge1, and denote by S2t\mathfrak{S}_{2t} the symmetric group on {1,…,2t}\{1,\ldots,2t\}. Let S^2t\widehat{\mathfrak{S}}_{2t} be the subset consisting of the permutations σ\sigma satisfying

σ(1)<σ(3)<⋯<σ(2t−1)andσ(1)<σ(2),…,σ(2t−1)<σ(2t).\sigma(1) < \sigma(3) < \cdots< \sigma(2t-1) \quad\text{and}\quad\sigma(1) < \sigma(2),\ldots,\sigma(2t-1) < \sigma(2t).

For an alternating matrix A=(aij)∈Alt⁡2t(F)A=(a_{ij}) \in\operatorname{Alt}_{2t}(\mathbb{F}), the Pfaffian of AA is

Pf⁡(A)=∑σ∈S^2tsgn⁡(σ)∏j=1taσ(2j−1),σ(2j).\operatorname{Pf}(A)=\sum_{\sigma\in\widehat{\mathfrak{S}}_{2t}}\operatorname{sgn}(\sigma)\prod_{j=1}^{t}a_{\sigma(2j-1),\sigma(2j)}.

where sgn⁡(σ)∈{1,−1}\operatorname{sgn}(\sigma)\in\{1,-1\} denotes the sign of σ\sigma; see e.g. [9 p. 12]. For an alternating matrix AA of odd order, we set Pf⁡(A)=0\operatorname{Pf}(A)=0. The Pfaffian satisfies

det⁡A=Pf⁡(A)2;(7)\det A=\operatorname{Pf}(A)^2; \tag*{(7)}

see e.g. [15 Chapter XV, §9, Theorem 9.1].

Let A=(aij)∈Alt⁡n(F)A=(a_{ij})\in\operatorname{Alt}_{n}(\mathbb{F}). For a subset I={i1<⋯<i2s}⊆{1,…,n}I=\{i_{1}<\cdots<i_{2s}\}\subseteq\{1,\ldots,n\} of even cardinality, denote by

AI=(aiαiβ)1≤α,β≤2sA_{I}=(a_{i_{\alpha}i_{\beta}})_{1\le\alpha,\beta\le2s}

the principal submatrix obtained by considering the rows and columns indexed by II, in increasing order. Note that AIA_{I} is again an alternating matrix, so its Pfaffian Pf⁡(AI)\operatorname{Pf}(A_{I}) is well defined. We set Pf⁡(A∅)=1\operatorname{Pf}(A_{\varnothing})=1.

Lemma 2.1. Let A∈Alt⁡n(F)A\in\operatorname{Alt}_{n}(\mathbb{F}) and let rr be an integer with 0≤r<⌊n/2⌋0\le r<\lfloor n/2\rfloor. Then rk⁡A≤2r\operatorname{rk} A\le2r if and only if Pf⁡(AI)=0\operatorname{Pf}(A_{I})=0, for every I⊆{1,…,n}I\subseteq\{1,\ldots,n\}, with ∣I∣=2r+2|I|=2r+2.

Proof. If rk⁡A≤2r\operatorname{rk} A\le2r, every principal submatrix of order 2r+22r+2 is singular. Equation (7) therefore gives Pf⁡(AI)=0\operatorname{Pf}(A_{I})=0 for every such II. Conversely, suppose that rk⁡A>2r\operatorname{rk} A>2r. Since alternating matrices have even rank, rk⁡A≥2r+2\operatorname{rk} A\ge2r+2. Over a field, [13 Theorem 3.12(2)–(3)] implies that an alternating matrix with a nonzero minor of even order has a nonzero principal minor of the same order. Hence there is a subset II of size 2r+22r+2 such that det⁡AI≠0\det A_{I}\ne0. Equation (7) then gives Pf⁡(AI)≠0\operatorname{Pf}(A_{I})\ne0. □

Graded modules and free resolutions

We recall the notions and properties of graded rings and modules that will be needed in the sequel. Let (H,+)(H,+) be an abelian semigroup with identity 00. An HH-graded ring is a ring RR endowed with a decomposition into additive subgroups

R=⨁a∈HRa,R=\bigoplus_{a\in H}R_{a},

such that 1R∈R01_{R}\in R_{0} and RaRb⊆Ra+bR_{a}R_{b}\subseteq R_{a+b} for all a,b∈Ha,b\in H. This decomposition is called an HH-grading of RR. In particular, R0R_{0} is a subring of RR. Unless otherwise specified, a graded ring means a Z\mathbb{Z}-graded ring.

Let RR be an HH-graded ring. An RR-module UU is an HH-graded RR-module if it has a decomposition into additive subgroups

U=⨁a∈HUaU=\bigoplus_{a\in H}U_a

satisfying RaUb⊆Ua+bR_aU_b\subseteq U_{a+b}, for all a,b∈Ha,b\in H. Each UaU_a is an R0R_0-module and is called the degree-aa component of UU. A nonzero element u∈Uau\in U_a is said to be homogeneous of degree aa, and we write deg⁡u=a\deg u=a. The nonzero elements of RaR_a are also called forms of degree aa. Every u∈Uu\in U has a unique expression u=∑a∈Huau=\sum_{a\in H}u_a, with ua∈Uau_a\in U_a, with only finitely many ua≠0u_a\ne0. A graded module means a Z\mathbb{Z}-graded module. An RR-module homomorphism φ:U→V\varphi:U\to V between HH-graded RR-modules is graded if φ(Ua)⊆Va\varphi(U_a)\subseteq V_a, for all a∈Ha\in H. Thus a graded homomorphism preserves degrees. Let U=⨁a∈HUaU=\bigoplus_{a\in H}U_a be an HH-graded RR-module and let V⊆UV\subseteq U be a graded submodule; that is, VV is graded with the induced grading

V=⨁a∈H(V∩Ua).V=\bigoplus_{a\in H}(V\cap U_a).

In this case the quotient U/VU/V inherits an HH-grading, with

(U/V)a=Ua+VV≅UaV∩Ua(a∈H).(U/V)_a=\frac{U_a+V}{V}\cong\frac{U_a}{V\cap U_a}\qquad(a\in H).

We use N={0,1,2,…}\mathbb{N}=\{0,1,2,\ldots\}. An N\mathbb{N}-grading is regarded as a Z\mathbb{Z}-grading by setting all components of negative degree equal to zero.

Let F\mathbb{F} be a field and let S=F[z1,…,zM]S=\mathbb{F}[z_1,\ldots,z_M] be the polynomial ring in MM variables. The standard grading of SS is the N\mathbb{N}-grading defined by deg⁡zi=1\deg z_i=1 for every ii. For a≥0a\ge0, its component SaS_a is the F\mathbb{F}-vector space of homogeneous polynomials of total degree aa, including zero. We set Sa=0S_a=0 for a<0a<0. Thus

S=⨁a∈ZSa,S0=F.S=\bigoplus_{a\in\mathbb{Z}}S_a,\qquad S_0=\mathbb{F}.

In particular, every component of a graded SS-module is an F\mathbb{F}-vector space. The ring SS is itself a graded SS-module. A graded SS-module UU is called graded free if it is a free SS-module with a homogeneous basis, that is, an SS-basis consisting of homogeneous elements.

Definition 2.2. For any N\mathbb{N}-graded SS-module UU and any a∈Na\in\mathbb{N}, denote by U(−a)U(-a) the graded shifted module: it is the same underlying SS-module, with homogeneous components

U(−a)t=Ut−afor all t∈N,U(-a)_t=U_{t-a}\qquad\text{for all }t\in\mathbb{N},

where Uj=0U_j=0 for j<0j<0.

A nonzero homogeneous element of degree bb in UU has degree b+ab+a in U(−a)U(-a). In particular, S(−a)t=St−aS(-a)_t=S_{t-a}, and the element 11 has degree aa in S(−a)S(-a).

We recall the notion of graded free resolution, see e.g. [14 Definition 4.8.1].

Definition 2.3. Let UU be a nonzero finitely generated N\mathbb{N}-graded SS-module. A finite graded free resolution of UU is an exact sequence

0⟶Ps→ds⋯→d2P1→d1P0→ϵU⟶0(8)0\longrightarrow P_s\xrightarrow{d_s}\cdots\xrightarrow{d_2}P_1\xrightarrow{d_1}P_0\xrightarrow{\epsilon}U\longrightarrow0 \tag*{(8)}

where s≥0s\ge0, each PiP_i is a finitely generated graded free SS-module, and all maps are graded SS-module homomorphisms. If Ps≠0P_s\ne0, the integer ss is called the length of the resolution.

Let a1,…,ar∈Na_1,\ldots,a_r \in\mathbb{N} and consider the SS-module

P=⨁α=1rS(−aα).P=\bigoplus_{\alpha=1}^{r} S(-a_\alpha).

The gradings of the summands induce an N\mathbb{N}-grading on PP, whose degree-tt component is

Pt=⨁α=1rS(−aα)t=⨁α=1rSt−aα(t∈N).P_t=\bigoplus_{\alpha=1}^{r}S(-a_\alpha)_t=\bigoplus_{\alpha=1}^{r}S_{t-a_\alpha}\qquad(t\in\mathbb{N}).

For each α\alpha, let eαe_\alpha be the vector whose α\alphath component is 1 and whose other components are zero. Then e1,…,ere_1,\ldots,e_r form an SS-basis of PP, and eαe_\alpha is homogeneous of degree aαa_\alpha. Consequently, PP is a finitely generated graded free SS-module. More explicitly, every element of PP has a unique expression

∑α=1rfαeα,fα∈S,\sum_{\alpha=1}^{r}f_\alpha e_\alpha,\qquad f_\alpha\in S,

and this element belongs to PtP_t if and only if fα∈St−aαf_\alpha\in S_{t-a_\alpha} for every α\alpha.

Theorem 2.4 (Hilbert’s Syzygy Theorem; see e.g. [29], Proposition 3.5.6 and Theorem 3.5.7). Let UU be a nonzero finitely generated N\mathbb{N}-graded SS-module. Then UU admits a finite graded free resolution

0⟶⨁α=1rsS(−as,α)→ds⋯→d2⨁α=1r1S(−a1,α)→d1⨁α=1r0S(−a0,α)→ϵU⟶0,0\longrightarrow\bigoplus_{\alpha=1}^{r_s}S(-a_{s,\alpha})\xrightarrow{d_s}\cdots\xrightarrow{d_2}\bigoplus_{\alpha=1}^{r_1}S(-a_{1,\alpha})\xrightarrow{d_1}\bigoplus_{\alpha=1}^{r_0}S(-a_{0,\alpha})\xrightarrow{\epsilon}U\longrightarrow0,

where 0≤s≤M0\leq s\leq M, each rir_i is a positive integer, each ai,αa_{i,\alpha} belongs to N\mathbb{N}, and all maps are graded SS-module homomorphisms. Moreover, the maps did_i satisfy

im⁡di⊆(z1,…,zM)(⨁α=1ri−1S(−ai−1,α))(1≤i≤s).\operatorname{im}d_i\subseteq(z_1,\ldots,z_M)\left(\bigoplus_{\alpha=1}^{r_{i-1}}S(-a_{i-1,\alpha})\right)\qquad(1\leq i\leq s).

A graded free resolution (8) is called minimal if all entries of the matrices representing d1,…,dsd_1,\ldots,d_s with respect to homogeneous bases belong to (z1,…,zM)(z_1,\ldots,z_M); see [29], Remark 3.5.9 and Definition 3.5.13. In particular, the resolution provided by Theorem 2.4 is minimal and unique up to complex isomorphism; see [29], Corollary 3.5.8. The integers ai,αa_{i,\alpha} are called the shifts of the resolution.

The largest shift for a Pfaffian ring

Let F=Fq\mathbb{F}=\mathbb{F}_q, let M≥1M\geq1, and let S=F[z1,…,zM]S=\mathbb{F}[z_1,\ldots,z_M] with its standard grading. An ideal of SS is called homogeneous if it is generated by homogeneous polynomials; equivalently, it is a graded SS-submodule of SS. Let I⊊SI\subsetneq S be a homogeneous ideal, and put A=S/IA=S/I, with the quotient grading. By Theorem 2.4, we may fix a finite minimal graded free resolution of AA,

0⟶Ps→ds⋯→d1P0→ϵA⟶0,Pi=⨁α=1βiS(−ai,α).(9)0\longrightarrow P_s\xrightarrow{d_s}\cdots\xrightarrow{d_1}P_0\xrightarrow{\epsilon}A\longrightarrow0,\qquad P_i=\bigoplus_{\alpha=1}^{\beta_i}S(-a_{i,\alpha}). \tag*{(9)}

where 0≤s≤M0\leq s\leq M, each βi\beta_i is a positive integer, and ai,α∈Na_{i,\alpha}\in\mathbb{N} for all i,αi,\alpha. We denote by

B=max⁡{ai,α:0≤i≤s, 1≤α≤βi}B=\max\{a_{i,\alpha}:0\leq i\leq s,\ 1\leq\alpha\leq\beta_i\}

the largest shift in this resolution. We determine the largest shift in a minimal graded free resolution of a Pfaffian ring over its ambient polynomial ring, using its Cohen–Macaulay property and aa-invariant.

Hilbert series and the aa-invariant

An N\mathbb{N}-graded F\mathbb{F}-algebra is called standard graded if its degree-zero component is F\mathbb{F} and it is generated as an F\mathbb{F}-algebra by finitely many homogeneous elements of degree one; see [29 Definition 3.1.14]. The quotient A=S/IA=S/I has this property. Indeed, since II is proper and homogeneous, I∩S0=0I\cap S_0=0, so A0=FA_0=\mathbb{F}. Moreover, the classes of z1,…,zMz_1,\ldots,z_M generate AA as an F\mathbb{F}-algebra, and their nonzero classes have degree one. Each component AjA_j is finite-dimensional over F\mathbb{F}, since it is a quotient of SjS_j. The Hilbert series of AA is

HA(T)=∑j≥0(dim⁡FAj)Tj∈Z[[T]];H_A(T)=\sum_{j\geq0}(\dim_{\mathbb{F}} A_j)T^j\in\mathbb{Z}[[T]];

see [29 Section 5.1]. Since AA is generated as an SS-module by 1+I1+I, we take P0=SP_0=S and let ϵ:S→A\epsilon:S\to A be the quotient map. Thus P0P_0 consists of a single summand S(−0)=SS(-0)=S, so β0=1\beta_0=1 and a0,1=0a_{0,1}=0. Define

KA(T)=∑i=0s(−1)i∑α=1βiTai,α∈Z[T].(10)K_A(T)=\sum_{i=0}^{s}(-1)^i\sum_{\alpha=1}^{\beta_i}T^{a_{i,\alpha}}\in\mathbb{Z}[T]. \tag*{(10)}

Since SS is a polynomial ring with its standard grading, [29 Eq. (5.1), p. 151], applied to (9), gives

HA(T)=KA(T)(1−T)M.(11)H_A(T)=\frac{K_A(T)}{(1-T)^M}. \tag*{(11)}

Comparison of constant terms yields KA(0)=dim⁡FA0=1K_A(0)=\dim_{\mathbb{F}}A_0=1. In particular, KA(T)K_A(T) is nonzero. For I=0I=0, the resolution consists only of P0=SP_0=S, and (11) gives

HS(T)=1(1−T)M.(12)H_S(T)=\frac{1}{(1-T)^M}. \tag*{(12)}

For nonzero polynomials u,v∈Q[T]u,v\in\mathbb{Q}[T], the degree of the rational function u/vu/v is deg⁡u−deg⁡v\deg u-\deg v, independently of the chosen representation. Following [29 Definition 5.1.5], we define the aa-invariant of AA as the degree of its Hilbert series as a rational function. Thus

a(A)=deg⁡HA(T)=deg⁡KA(T)−M.(13)a(A)=\deg H_A(T)=\deg K_A(T)-M. \tag*{(13)}

In particular, (12) gives

a(S)=−M.(14)a(S)=-M. \tag*{(14)}

Application to Pfaffian rings

Let r≥1r\geq1 be an integer and suppose that n≥2r+2n\geq2r+2. Put

N=(n2).N=\binom{n}{2}.

Take M=NM=N and index the variables of SS as xijx_{ij}, where 1≤i<j≤n1\leq i<j\leq n. Thus

S=F[xij:1≤i<j≤n],deg⁡xij=1.S=\mathbb{F}[x_{ij}:1\leq i<j\leq n],\qquad\deg x_{ij}=1.

Let XX be the generic alternating matrix over SS, defined by

Xij=xij,Xji=−xij (i<j),Xii=0.X_{ij}=x_{ij},\qquad X_{ji}=-x_{ij}\ (i<j),\qquad X_{ii}=0.

We use the principal-submatrix notation XJX_J and the polynomial definition of the Pfaffian from Section 2.1. That definition applies also to matrices with entries in SS, in every characteristic.

Set

I2r+2=(Pf⁡(XJ):J⊆{1,…,n}, ∣J∣=2r+2)⊆S,I_{2r+2}=\bigl(\operatorname{Pf}(X_J):J\subseteq\{1,\ldots,n\},\ |J|=2r+2\bigr)\subseteq S,

and

Rn,r(F)=S/I2r+2.R_{n,r}(\mathbb{F})=S/I_{2r+2}.

The ideal I2r+2I_{2r+2} is a Pfaffian ideal and the quotient Rn,r(F)R_{n,r}(\mathbb{F}) is called a Pfaffian ring; see [6, 9, 11]. Each generating Pfaffian is homogeneous of degree r+1r+1. It is also nonzero: if J={j1<⋯<j2r+2}J=\{j_1<\cdots<j_{2r+2}\}, the monomial xj1j2xj3j4⋯xj2r+1j2r+2x_{j_1j_2}x_{j_3j_4}\cdots x_{j_{2r+1}j_{2r+2}} occurs with coefficient 1, and distinct pairings give distinct monomials. Since n≥2r+2n \ge2r+2, such subsets JJ exist. Moreover, all generators have positive degree, so

I2r+2⊆(xij:i<j)⊊S.I_{2r+2} \subseteq(x_{ij}:i<j) \subsetneq S.

Hence I2r+2I_{2r+2} is a nonzero proper homogeneous ideal, and Rn,r(F)R_{n,r}(\mathbb{F}) is a standard graded quotient of the form considered above, with I=I2r+2I=I_{2r+2} and M=NM=N.

Generic Pfaffian rings are Cohen–Macaulay over every field, as proved in [12] and [17, 18]; see also [9 Section 6].

Proposition 3.1. The aa-invariant of the Pfaffian ring Rn,r(F)R_{n,r}(\mathbb{F}) is

a(Rn,r(F))=−nr.a(R_{n,r}(\mathbb{F}))=-nr.

The largest shift BB in its minimal graded free resolution over SS is N−nrN-nr.

Proof. We consider Rn,r(F)R_{n,r}(\mathbb{F}) with the grading induced by the standard grading of SS. In the notation of [2 Corollary 1.7, p. 206], this corresponds to e1=⋯=en=1e_1=\cdots=e_n=1, since

deg⁡xij=ei+ej2=1(1≤i<j≤n).\deg x_{ij}=\frac{e_i+e_j}{2}=1 \qquad(1\le i<j\le n).

Since n≥2r+2n\ge2r+2, we have 2r<n2r<n, and [2 Corollary 1.7, p. 206] gives

a(Rn,r(F))=−r∑i=1nei=−nr.a(R_{n,r}(\mathbb{F}))=-r\sum_{i=1}^{n}e_i=-nr.

The ideal I2r+2I_{2r+2} is nonzero, proper, and homogeneous, and Rn,r(F)=S/I2r+2R_{n,r}(\mathbb{F})=S/I_{2r+2} is Cohen–Macaulay. By [29 Proposition 5.2.3(a)],

0=a0,1≤max⁡1≤α≤β1a1,α<⋯<max⁡1≤α≤βsas,α.0=a_{0,1}\le\max_{1\le\alpha\le\beta_1}a_{1,\alpha}<\cdots<\max_{1\le\alpha\le\beta_s}a_{s,\alpha}.

Hence, by [29 Proposition 5.2.3(b)] and (14), we obtain

B=max⁡1≤α≤βsas,α=a(Rn,r(F))−a(S)=N−nr,B=\max_{1\le\alpha\le\beta_s}a_{s,\alpha}=a(R_{n,r}(\mathbb{F}))-a(S)=N-nr,

as required.

Trace functions and counting criteria

We introduce trace functions on evaluation quotients of modules and prove their additivity. We use this property together with Frobenius maps to construct polynomial indicators. Bounds on their degrees yield a counting criterion over subfields. Let F=FqF=\mathbb{F}_q be the finite field with qq elements of characteristic pp, and let S=F[z1,…,zM]S=F[z_1,\ldots,z_M], where M≥1M\ge1, with its standard grading. All SS-modules considered in this section are assumed to be finitely generated.

Quotient modules and trace functions

For v=(v1,…,vM)∈FMv=(v_1,\ldots,v_M)\in\mathbb{F}^M, evaluation at vv defines a surjective F\mathbb{F}-algebra homomorphism S→FS\to\mathbb{F}, f↦f(v)f\mapsto f(v), with kernel mv=(z1−v1,…,zM−vM)\mathfrak{m}_v=(z_1-v_1,\ldots,z_M-v_M). For an SS-module UU, we define the quotient module

U[v]=UmvU,mvU={∑j=1M(zj−vj)uj:uj∈U}.U[v]=\frac{U}{\mathfrak{m}_vU},\qquad\mathfrak{m}_vU=\left\{\sum_{j=1}^{M}(z_j-v_j)u_j:u_j\in U\right\}.

and write [u]v=u+mvU[u]_v=u+\mathfrak{m}_vU for the class of u∈Uu\in U. Since f−f(v)∈mvf-f(v)\in\mathfrak{m}_v, we have [fu]v=f(v)[u]v[fu]_v=f(v)[u]_v, for all f∈Sf\in S, u∈Uu\in U and U[v]U[v] is naturally an F\mathbb{F}-vector space. The classes of any finite set of generators of UU span U[v]U[v] over F\mathbb{F}, so U[v]U[v] is finite-dimensional. Also, if UU is free with basis e1,…,ete_1,\ldots,e_t, then [e1]v,…,[et]v[e_1]_v,\ldots,[e_t]_v form an F\mathbb{F}-basis of U[v]U[v]. Every SS-module homomorphism φ:U→V\varphi:U\to V induces an F\mathbb{F}-linear map

φ[v]:U[v]⟶V[v],[u]v⟼[φ(u)]v.\varphi[v]:U[v]\longrightarrow V[v],\qquad[u]_v\longmapsto[\varphi(u)]_v.

Let j=paj=p^a, where a≥1a\ge1. An additive map T:U→UT:U\to U is called jj-linear if T(fu)=fjT(u)T(fu)=f^jT(u) for all f∈Sf\in S and u∈Uu\in U. For every v∈FMv\in\mathbb{F}^M, such a map satisfies

T(mvU)⊆mvjU⊆mvU(15)T(\mathfrak{m}_vU)\subseteq\mathfrak{m}_v^jU\subseteq\mathfrak{m}_vU \tag*{(15)}

where mvj\mathfrak{m}_v^j denotes the ordinary ideal power. It therefore induces an additive map

T[v]:U[v]⟶U[v],[u]v⟼[T(u)]v.T[v]:U[v]\longrightarrow U[v],\qquad[u]_v\longmapsto[T(u)]_v.

This map satisfies T[v](λ[u]v)=λjT[v]([u]v)T[v](\lambda[u]_v)=\lambda^jT[v]([u]_v) for all λ∈F\lambda\in\mathbb{F}, u∈Uu\in U. In particular, if j=qj=q, then T[v]T[v] is F\mathbb{F}-linear.

Let UU and VV be SS-modules equipped with jj-linear maps TU:U→UT_U:U\to U and TV:V→VT_V:V\to V. An SS-module homomorphism φ:U→V\varphi:U\to V is called compatible with TUT_U and TVT_V if φ∘TU=TV∘φ\varphi\circ T_U=T_V\circ\varphi. In this case, for every v∈FMv\in\mathbb{F}^M, the induced maps satisfy φ[v]∘TU[v]=TV[v]∘φ[v]\varphi[v]\circ T_U[v]=T_V[v]\circ\varphi[v]. More generally, consider a sequence of SS-module homomorphisms

Us→φsUs−1⟶⋯→φ1U0,U_s\xrightarrow{\varphi_s}U_{s-1}\longrightarrow\cdots\xrightarrow{\varphi_1}U_0,

and let Ti:Ui→UiT_i:U_i\to U_i be a jj-linear map for each 0≤i≤s0\leq i\leq s. We say that the maps T0,…,TsT_0,\ldots,T_s are compatible with the sequence if φi∘Ti=Ti−1∘φi\varphi_i\circ T_i=T_{i-1}\circ\varphi_i for all 1≤i≤s1\leq i\leq s. For a qq-linear map T:U→UT:U\to U, define

τv(U,T)=Tr⁡F(T[v]∣U[v]).\tau_v(U,T)=\operatorname{Tr}_{\mathbb{F}}(T[v]\mid U[v]).

the trace of the induced F\mathbb{F}-linear endomorphism T[v]T[v]. The function

FM⟶F,v⟼τv(U,T),\mathbb{F}^M\longrightarrow\mathbb{F},\qquad v\longmapsto\tau_v(U,T),

is called the trace function of (U,T)(U,T). For a square matrix over a commutative ring, let tr⁡\operatorname{tr} denote the sum of its diagonal entries. If UU is free with basis e1,…,ete_1,\ldots,e_t, write

T(eα)=∑β=1tbβα(z)eβ,BT(z)=(bβα(z))∈Mt(S).T(e_\alpha)=\sum_{\beta=1}^{t}b_{\beta\alpha}(z)e_\beta,\qquad B_T(z)=\bigl(b_{\beta\alpha}(z)\bigr)\in M_t(S).

The induced map T[v]T[v] is represented in the basis [e1]v,…,[et]v[e_1]_v,\ldots,[e_t]_v by BT(v)B_T(v). Hence

τv(U,T)=tr⁡BT(v)=∑α=1tbαα(v).(16)\tau_v(U,T)=\operatorname{tr}B_T(v)=\sum_{\alpha=1}^{t}b_{\alpha\alpha}(v). \tag*{(16)}

This includes U=0U=0, with the empty sum equal to zero.

For an ideal I⊆SI \subseteq S, set

V(I)(F)={v∈FM:f(v)=0 for every f∈I}.V(I)(\mathbb{F})=\{v\in\mathbb{F}^{M}:f(v)=0\text{ for every }f\in I\}.

The following proposition relates trace functions to the number of common zeros of an ideal.

Proposition 4.1. Let I⊆SI\subseteq S be an ideal, put A=S/IA=S/I, and consider the qq-linear map

TA:f+I∈A⟼fq+I∈A.T_A:f+I\in A\longmapsto f^q+I\in A.

For every v∈FMv\in\mathbb{F}^{M},

τv(A,TA)={1,v∈V(I)(F),0,v∉V(I)(F).(17)\tau_v(A,T_A)= \begin{cases} 1,&v\in V(I)(\mathbb{F}),\\ 0,&v\notin V(I)(\mathbb{F}). \end{cases} \tag*{(17)}

Consequently,

∑v∈FMτv(A,TA)=∣V(I)(F)∣ 1F.\sum_{v\in\mathbb{F}^{M}}\tau_v(A,T_A)=|V(I)(\mathbb{F})|\,1_{\mathbb{F}}.

In particular, this sum is zero if and only if pp divides ∣V(I)(F)∣|V(I)(\mathbb{F})|.

Proof. Fix v∈FMv\in\mathbb{F}^{M}. Since A=S/IA=S/I, we have

mvA={h+I:h∈mv}=I+mvI.\mathfrak{m}_vA=\{h+I:h\in\mathfrak{m}_v\}=\frac{I+\mathfrak{m}_v}{I}.

Indeed, every product h(f+I)=hf+Ih(f+I)=hf+I, with h∈mvh\in\mathfrak{m}_v and f∈Sf\in S, is represented by an element of mv\mathfrak{m}_v, since mv\mathfrak{m}_v is an ideal. Conversely, h+I=h(1+I)h+I=h(1+I) belongs to mvA\mathfrak{m}_vA for every h∈mvh\in\mathfrak{m}_v. Therefore, there is a canonical F\mathbb{F}-linear isomorphism

A[v]=S/I(I+mv)/I⟶S/(I+mv),[f+I]v⟼f+(I+mv).A[v]=\frac{S/I}{(I+\mathfrak{m}_v)/I}\longrightarrow S/(I+\mathfrak{m}_v),\qquad[f+I]_v\longmapsto f+(I+\mathfrak{m}_v).

Suppose that v∈V(I)(F)v\in V(I)(\mathbb{F}). Then every element of II vanishes at vv, so I⊆mvI\subseteq\mathfrak{m}_v and I+mv=mvI+\mathfrak{m}_v=\mathfrak{m}_v. Since evaluation at vv is surjective and has kernel mv\mathfrak{m}_v, it induces an F\mathbb{F}-linear isomorphism

A[v]⟶F,[f+I]v⟼f(v).A[v]\longrightarrow\mathbb{F},\qquad[f+I]_v\longmapsto f(v).

Moreover,

TA[v]([f+I]v)=[fq+I]v,fq(v)=f(v)q=f(v),T_{A[v]}([f+I]_v)=[f^q+I]_v,\qquad f^q(v)=f(v)^q=f(v),

because f(v)∈Ff(v)\in\mathbb{F}. Thus, under this identification, TA[v]T_{A[v]} corresponds to the identity map on F\mathbb{F}, and hence τv(A,TA)=1\tau_v(A,T_A)=1. If v∉V(I)(F)v\notin V(I)(\mathbb{F}), choose f∈If\in I with f(v)≠0f(v)\ne0. Since f−f(v)∈mvf-f(v)\in\mathfrak{m}_v, we have

f(v)=f−(f−f(v))∈I+mv.f(v)=f-(f-f(v))\in I+\mathfrak{m}_v.

Thus I+mvI+\mathfrak{m}_v contains a unit of SS, so I+mv=SI+\mathfrak{m}_v=S and A[v]=0A[v]=0. Its endomorphism TA[v]T_{A[v]} has trace zero. This proves (17). Summing over v∈FMv\in\mathbb{F}^{M} gives the asserted identity. □\square

Passing to the quotients U[v]U[v] need not preserve injectivity. To prove additivity of trace functions, we shall use the following consequence of the Artin–Rees Theorem.

Lemma 4.2 (Artin–Rees). Let RR be a commutative noetherian ring, let J⊆RJ\subseteq R be an ideal, let UU be a finitely generated RR-module, and let V⊆UV\subseteq U be a submodule. There is an integer b≥0b\geq0 such that

V∩JtU⊆Jt−bV(t≥b).V\cap J^tU\subseteq J^{t-b}V\qquad(t\geq b).

In particular, V∩JtU⊆JVV\cap J^tU\subseteq JV for every t≥b+1t\geq b+1.

Proof. By the Artin–Rees Theorem, see [15 Chapter X, §5, Corollary 5.5], there is b≥0b \ge0 such that

V∩JtU=Jt−b(V∩JbU)⊆Jt−bV(t≥b).V \cap J^{t}U = J^{t-b}(V \cap J^{b}U) \subseteq J^{t-b}V \qquad(t \ge b).

For t≥b+1t \ge b+1, the last module is contained in JVJV.

Proposition 4.3. Let

0⟶U′→ιU→πU′′⟶00 \longrightarrow U^{\prime} \xrightarrow{\iota} U \xrightarrow{\pi} U^{\prime\prime} \longrightarrow0

be an exact sequence of finitely generated SS-modules, equipped with compatible qq-linear maps T′:U′→U′T^{\prime}: U^{\prime} \to U^{\prime}, T:U→UT: U \to U, T′′:U′′→U′′T^{\prime\prime}: U^{\prime\prime} \to U^{\prime\prime}. Then, for every v∈FMv \in F^{M},

τv(U,T)=τv(U′,T′)+τv(U′′,T′′).\tau_{v}(U,T) = \tau_{v}(U^{\prime},T^{\prime}) + \tau_{v}(U^{\prime\prime},T^{\prime\prime}).

Proof. Fix v∈FMv \in F^{M} and put m=mv\mathfrak m = \mathfrak m_{v}. Since π\pi is surjective and SS-linear, we have π(mU)=mU′′\pi(\mathfrak m U) = \mathfrak m U^{\prime\prime}. The induced map π[v]\pi[v] is also surjective. Next, we determine ker⁡π[v]\ker\pi[v]. For u∈Uu \in U, the definition of π[v]\pi[v] gives

[u]v∈ker⁡π[v]⟺π(u)∈mU′′.[u]_{v} \in\ker\pi[v] \Longleftrightarrow\pi(u) \in\mathfrak m U^{\prime\prime}.

Since mU′′=π(mU)\mathfrak m U^{\prime\prime} = \pi(\mathfrak m U), this is equivalent to the existence of w∈mUw \in\mathfrak m U such that π(u)=π(w)\pi(u) = \pi(w). By exactness, u−w∈ker⁡π=ι(U′)u-w \in\ker\pi= \iota(U^{\prime}), so u=ι(u′)+wu = \iota(u^{\prime}) + w for some u′∈U′u^{\prime} \in U^{\prime}. Conversely, every element of this form has image under π\pi in mU′′\mathfrak m U^{\prime\prime}. Therefore

ker⁡π[v]=ι(U′)+mUmU.\ker\pi[v] = \frac{\iota(U^{\prime}) + \mathfrak m U}{\mathfrak m U}.

Moreover, since w∈mUw \in\mathfrak m U, we have [u]v=[ι(u′)]v=ι[v]([u′]v)[u]_{v} = [\iota(u^{\prime})]_{v} = \iota[v]([u^{\prime}]_{v}). Thus every element of ker⁡π[v]\ker\pi[v] belongs to im⁡ι[v]\operatorname{im}\iota[v]. The reverse inclusion follows from π[v]∘ι[v]=0\pi[v] \circ\iota[v] = 0, and hence

ker⁡π[v]=im⁡ι[v].\ker\pi[v] = \operatorname{im}\iota[v].

Similarly, for u′∈U′u^{\prime} \in U^{\prime},

[u′]v∈ker⁡ι[v]⟺[ι(u′)]v=0[u^{\prime}]_{v} \in\ker\iota[v] \Longleftrightarrow[\iota(u^{\prime})]_{v} = 0
⟺ι(u′)∈mU\Longleftrightarrow\iota(u^{\prime}) \in\mathfrak m U
⟺u′∈ι−1(mU).\Longleftrightarrow u^{\prime} \in\iota^{-1}(\mathfrak m U).

The SS-linearity of ι\iota gives ι(mU′)⊆mU\iota(\mathfrak m U^{\prime}) \subseteq\mathfrak m U, so mU′⊆ι−1(mU)\mathfrak m U^{\prime} \subseteq\iota^{-1}(\mathfrak m U). Consequently,

K:=ker⁡ι[v]=ι−1(mU)mU′.K := \ker\iota[v] = \frac{\iota^{-1}(\mathfrak m U)}{\mathfrak m U^{\prime}}.

Put W=im⁡ι[v]W = \operatorname{im}\iota[v], and let ι~[v]:U′[v]→W\widetilde{\iota}[v] : U^{\prime}[v] \to W be the surjective map defined by

ι~[v]([u′]v)=[ι(u′)]v.\widetilde{\iota}[v]([u^{\prime}]_{v}) = [\iota(u^{\prime})]_{v}.

We obtain two exact sequences of finite-dimensional FF-vector spaces:

0⟶K⟶U′[v]→ι~[v]W⟶0,0 \longrightarrow K \longrightarrow U^{\prime}[v] \xrightarrow{\widetilde{\iota}[v]} W \longrightarrow0,
0⟶W⟶U[v]→π[v]U′′[v]⟶0,0 \longrightarrow W \longrightarrow U[v] \xrightarrow{\pi[v]} U^{\prime\prime}[v] \longrightarrow0,

where the maps from KK and WW into U′[v]U^{\prime}[v] and U[v]U[v], respectively, are the inclusions. Therefore, we have the following isomorphisms of FF-vector spaces:

U′[v]≅K⊕W,U[v]≅W⊕U′′[v].(18)U^{\prime}[v] \cong K \oplus W, \qquad U[v] \cong W \oplus U^{\prime\prime}[v]. \tag*{(18)}

Compatibility gives

ι∘T′=T∘ι,π∘T=T′′∘π.\iota\circ T^{\prime} = T \circ\iota, \qquad\pi\circ T = T^{\prime\prime} \circ\pi.

The identity ι[v]∘T′[v]=T[v]∘ι[v]\iota[v] \circ T'[v] = T[v] \circ\iota[v] shows that KK is invariant under T′[v]T'[v] and WW is invariant under T[v]T[v]. Set

TK=T′[v]∣K,TW=T[v]∣W.T_K = T'[v]|_K,\qquad T_W = T[v]|_W.

We claim that TKT_K is nilpotent. For every integer h≥0h \ge0, iterating qq-linearity and compatibility gives

Th(fu)=fqhTh(u)(f∈S, u∈U),T^h(fu) = f^{q^h}T^h(u)\qquad(f \in S,\ u \in U),
ι∘(T′)h=Th∘ι.\iota\circ(T')^h = T^h \circ\iota.

In particular, we get Th(mU)⊆mqhUT^h(\mathfrak m U) \subseteq\mathfrak m^{q^h}U. Apply Lemma 4.2 to the submodule ι(U′)⊆U\iota(U') \subseteq U. There is b≥0b \ge0 such that

ι(U′)∩mtU⊆m(ι(U′))=ι(mU′)(t≥b+1),\iota(U') \cap\mathfrak m^t U \subseteq\mathfrak m(\iota(U')) = \iota(\mathfrak m U')\qquad(t \ge b+1),

where the equality follows from the SS-linearity of ι\iota. Choose h≥1h \ge1 with qh≥b+1q^h \ge b+1. For every u′∈ι−1(mU)u' \in\iota^{-1}(\mathfrak m U), we have ι(u′)∈mU\iota(u') \in\mathfrak m U, and therefore

ι((T′)h(u′))=Th(ι(u′))∈ι(U′)∩mqhU⊆ι(mU′).\iota((T')^h(u')) = T^h(\iota(u')) \in\iota(U') \cap\mathfrak m^{q^h}U \subseteq\iota(\mathfrak m U').

Since ι\iota is injective, this implies (T′)h(u′)∈mU′(T')^h(u') \in\mathfrak m U'. Thus (T′)h(T')^h induces the zero map on K=ι−1(mU)/mU′K = \iota^{-1}(\mathfrak m U)/\mathfrak m U', so

TKh=0,Tr⁡F(TK∣K)=0.T_K^h = 0,\qquad\operatorname{Tr}_{F}(T_K \mid K) = 0.

The trace of an endomorphism is the sum of its traces on an invariant subspace and the corresponding quotient. Applying this fact to the two exact sequences above and using the decompositions in (18) gives

τv(U′,T′)=Tr⁡F(TK∣K)+Tr⁡F(TW∣W),\tau_v(U',T') = \operatorname{Tr}_{F}(T_K \mid K) + \operatorname{Tr}_{F}(T_W \mid W),
τv(U,T)=Tr⁡F(TW∣W)+τv(U′′,T′′).\tau_v(U,T) = \operatorname{Tr}_{F}(T_W \mid W) + \tau_v(U'',T'').

Since the trace on KK is zero, we conclude that

τv(U,T)=τv(U′,T′)+τv(U′′,T′′).\tau_v(U,T) = \tau_v(U',T') + \tau_v(U'',T'').

Remark 4.4. For trace functions and their additivity in a more general setting, see [27 Definition 1.9 and Exercise 1.2].

A counting criterion over subfields

Let F′⊆F\mathbb F' \subseteq\mathbb F be a subfield. Throughout this subsection, write q0=∣F′∣q_0 = |\mathbb F'| and e=[F:F′]e = [\mathbb F:\mathbb F'], so that q=q0eq = q_0^e.

We first recall a standard identity for polynomial sums; see, for example, [10 p. 120].

Lemma 4.5. Let KK be a finite field of order QQ, and let m≥1m \ge1. If G∈K[y1,…,ym]G \in K[y_1,\ldots,y_m] is zero or has total degree less than m(Q−1)m(Q-1), then

∑w∈KmG(w)=0.\sum_{w\in K^m} G(w) = 0.

Let I⊊SI \subsetneq S be a homogeneous ideal, and put A=S/IA = S/I, with its quotient grading. Since AA is a nonzero finitely generated N\mathbb N-graded SS-module, Theorem 2.4 allows us to fix a finite minimal graded free resolution

0⟶Ps→ds⋯→d1P0→ϵA⟶0,Pi=⨁α=1βiS(−ai,α),0 \longrightarrow P_s \xrightarrow{d_s} \cdots\xrightarrow{d_1} P_0 \xrightarrow{\epsilon} A \longrightarrow0,\qquad P_i = \bigoplus_{\alpha=1}^{\beta_i} S(-a_{i,\alpha}),

where 0≤s≤M0 \le s \le M, each βi\beta_i is a positive integer, and ai,α∈Na_{i,\alpha} \in\mathbb{N} for all i,αi,\alpha. Set

B=max⁡{ai,α:0≤i≤s, 1≤α≤βi}.B = \max\{a_{i,\alpha} : 0 \le i \le s,\ 1 \le\alpha\le\beta_i\}.

Theorem 4.6. With the above notation, let t≥1t \ge1, write y=(y1,…,yt)y = (y_1,\ldots,y_t), and let

L(y)=(L1(y),…,LM(y)),L(y) = (L_1(y),\ldots,L_M(y)),

where L1,…,LM∈F[y1,…,yt]L_1,\ldots,L_M \in\mathbb{F}[y_1,\ldots,y_t] are homogeneous linear forms, possibly zero. Then there is a polynomial ΘL∈F[y1,…,yt]\Theta_L \in\mathbb{F}[y_1,\ldots,y_t] of total degree at most e(q0−1)Be(q_0-1)B such that, for every w∈(F′)tw \in(\mathbb{F}')^t,

ΘL(w)={1,L(w)∈V(I)(F),0,L(w)∉V(I)(F).(19)\Theta_L(w) = \begin{cases} 1, & L(w) \in V(I)(\mathbb{F}),\\ 0, & L(w) \notin V(I)(\mathbb{F}). \end{cases} \tag*{(19)}

In particular, if t>eBt > eB, then

∣{w∈(F′)t:L(w)∈V(I)(F)}∣≡0(modp).\left|\{w \in(\mathbb{F}')^t : L(w) \in V(I)(\mathbb{F})\}\right| \equiv0 \pmod p.

Proof. Consider the q0q_0-power Frobenius

FA:A⟶A,f+I⟼fq0+I.F_A : A \longrightarrow A,\qquad f+I \longmapsto f^{q_0}+I.

It is well defined because f−g∈If-g \in I implies (f−g)q0∈I(f-g)^{q_0} \in I. It is q0q_0-linear and sends AkA_k into Aq0kA_{q_0 k} for every k≥0k \ge0.

We first construct q0q_0-linear maps Fi:Pi→PiF_i : P_i \to P_i such that

Fi((Pi)k)⊆(Pi)q0k(k≥0),F_i((P_i)_k) \subseteq(P_i)_{q_0 k}\qquad(k \ge0),
ϵ∘F0=FA∘ϵ,di∘Fi=Fi−1∘di.\epsilon\circ F_0 = F_A \circ\epsilon,\qquad d_i \circ F_i = F_{i-1} \circ d_i.

Choose homogeneous bases ui,αu_{i,\alpha} of PiP_i, with deg⁡ui,α=ai,α\deg u_{i,\alpha} = a_{i,\alpha}.

For each basis vector uu of P0P_0, choose F0(u)∈(P0)q0deg⁡uF_0(u) \in(P_0)_{q_0 \deg u} such that

ϵ(F0(u))=FA(ϵ(u)).\epsilon(F_0(u)) = F_A(\epsilon(u)).

This is possible because ϵ\epsilon is surjective and graded. Extend these assignments by

F0(∑αfαu0,α)=∑αfαq0F0(u0,α).F_0\left(\sum_{\alpha} f_{\alpha}u_{0,\alpha}\right) = \sum_{\alpha} f_{\alpha}^{q_0}F_0(u_{0,\alpha}).

The resulting map is well defined and q0q_0-linear, and satisfies the required degree and compatibility properties. Now, suppose that F0,…,Fi−1F_0,\ldots,F_{i-1} have been constructed. For a homogeneous basis vector uu of PiP_i, compatibility gives

ϵ(F0(d1(u)))=0if i=1,\epsilon(F_0(d_1(u))) = 0 \qquad\text{if } i=1,

and

di−1(Fi−1(di(u)))=0if i>1.d_{i-1}(F_{i-1}(d_i(u))) = 0 \qquad\text{if } i>1.

By exactness, Fi−1(di(u))F_{i-1}(d_i(u)) belongs to im⁡di\operatorname{im} d_i. It also belongs to (Pi−1)q0deg⁡u(P_{i-1})_{q_0\deg u}. Since did_i preserves degree, we may choose Fi(u)∈(Pi)q0deg⁡uF_i(u) \in(P_i)_{q_0\deg u} such that

di(Fi(u))=Fi−1(di(u)).d_i(F_i(u)) = F_{i-1}(d_i(u)).

Indeed, the component of degree q0deg⁡uq_0\deg u of any preimage is still a preimage. When the required image is zero, choose Fi(u)=0F_i(u) = 0. Extend the assignments q0q_0-linearly to PiP_i. The required properties hold on the basis and hence on all of PiP_i.

Write

Fi(ui,α)=∑β=1βibβα(i)(z)ui,β,Bi(z)=(bβα(i)(z)).F_i(u_{i,\alpha})=\sum_{\beta=1}^{\beta_i} b_{\beta\alpha}^{(i)}(z)u_{i,\beta},\qquad B_i(z)=\left(b_{\beta\alpha}^{(i)}(z)\right).

Every nonzero entry of BiB_i is homogeneous of degree

deg⁡bβα(i)=q0ai,α−ai,β.(20)\deg b_{\beta\alpha}^{(i)}=q_0a_{i,\alpha}-a_{i,\beta}. \tag*{(20)}

The maps FieF_i^e are qq-linear and compatible with the resolution. Moreover, FAe=TAF_A^e=T_A, where TAT_A is the qq-power Frobenius of Proposition 4.1.

Let σ:F→F\sigma:\mathbb{F}\to\mathbb{F} be the automorphism λ↦λq0\lambda\mapsto\lambda^{q_0}. Its fixed field is F′\mathbb{F}', and σe\sigma^e is the identity. Extend σ\sigma to F[y1,…,yt]\mathbb{F}[y_1,\ldots,y_t] by applying it to coefficients and fixing the variables. We also apply σ\sigma entrywise to matrices and column vectors.

Set

Ci(y)=Bi(L(y)),Qi(y)=Ci(y)σ(Ci(y))⋯σe−1(Ci(y)),C_i(y)=B_i(L(y)),\qquad Q_i(y)=C_i(y)\sigma(C_i(y))\cdots\sigma^{e-1}(C_i(y)),

where Qi=CiQ_i=C_i when e=1e=1, and define

ΘL(y)=∑i=0s(−1)itr⁡Qi(y).(21)\Theta_L(y)=\sum_{i=0}^{s}(-1)^i\operatorname{tr}Q_i(y). \tag*{(21)}

We first verify (19). By (15), applied with j=q0j=q_0, each FiF_i induces a map on Pi[v]P_i[v] for every v∈FMv\in\mathbb{F}^M. In the basis [ui,α]v[u_{i,\alpha}]_v, this map acts on coefficient columns by h↦Bi(v)σ(h)h\mapsto B_i(v)\sigma(h). Hence, the F\mathbb{F}-linear map (Fie)[v](F_i^e)[v] has representative matrix Bi(v)σ(Bi(v))⋯σe−1(Bi(v))B_i(v)\sigma(B_i(v))\cdots\sigma^{e-1}(B_i(v)). Let w∈(F′)tw\in(\mathbb{F}')^t and put v=L(w)v=L(w). Since σ\sigma fixes the coordinates of ww, for every integer ν≥0\nu\geq0 we have

σν(Ci)(w)=σν(Ci(w))=σν(Bi(v)).\sigma^\nu(C_i)(w)=\sigma^\nu(C_i(w))=\sigma^\nu(B_i(v)).

Hence Qi(w)Q_i(w) is the matrix of (Fie)[v](F_i^e)[v], and

tr⁡Qi(w)=τv(Pi,Fie).\operatorname{tr}Q_i(w)=\tau_v(P_i,F_i^e).

Put d0=ϵd_0=\epsilon, Z−1=AZ_{-1}=A, and Zi=ker⁡diZ_i=\ker d_i for 0≤i≤s0\leq i\leq s. By exactness, Zs=0Z_s=0 and Zi=im⁡di+1Z_i=\operatorname{im}d_{i+1} for 0≤i<s0\leq i<s. Thus each ZiZ_i is finitely generated, since each PiP_i is finitely generated.

Compatibility implies that FieF_i^e preserves ZiZ_i. Let GiG_i be the restriction of FieF_i^e to ZiZ_i, and put G−1=TAG_{-1}=T_A. For each 0≤i≤s0\leq i\leq s, we have a short exact sequence

0⟶Zi⟶Pi→diZi−1⟶0,0\longrightarrow Z_i\longrightarrow P_i\xrightarrow{d_i}Z_{i-1}\longrightarrow0,

where the codomain of did_i is restricted to its image. The qq-linear maps GiG_i, FieF_i^e, and Gi−1G_{i-1} are compatible with this sequence. Hence Proposition 4.3 gives

τv(Pi,Fie)=τv(Zi,Gi)+τv(Zi−1,Gi−1).\tau_v(P_i,F_i^e)=\tau_v(Z_i,G_i)+\tau_v(Z_{i-1},G_{i-1}).

Taking alternating sums, the intermediate terms cancel, and the term involving ZsZ_s vanishes because Zs=0Z_s=0. Consequently,

ΘL(w)=∑i=0s(−1)iτv(Pi,Fie)=∑i=0s(−1)i(τv(Zi,Gi)+τv(Zi−1,Gi−1))=τv(A,TA)={1,L(w)∈V(I)(F),0,L(w)∉V(I)(F),\begin{aligned} \Theta_L(w) &= \sum_{i=0}^{s}(-1)^i\tau_v(P_i,F_i^e) \\ &= \sum_{i=0}^{s}(-1)^i\left(\tau_v(Z_i,G_i)+\tau_v(Z_{i-1},G_{i-1})\right) \\ &= \tau_v(A,T_A) \\ &= \begin{cases} 1, & L(w)\in V(I)(\mathbb{F}),\\ 0, & L(w)\notin V(I)(\mathbb{F}), \end{cases} \end{aligned}

where the last equality follows from Proposition 4.1. This proves (19).

We next show that ΘL\Theta_L has coefficients in F′\mathbb{F}'. For e=1e=1, this is immediate. For e≥2e\geq2, the identity σe(Ci)=Ci\sigma^e(C_i)=C_i and cyclic invariance of the matrix trace give

σ(tr⁡Qi)=tr⁡(σ(Ci)⋯σe−1(Ci)Ci)=tr⁡(Ciσ(Ci)⋯σe−1(Ci))=tr⁡Qi.\begin{aligned} \sigma(\operatorname{tr}Q_i)&=\operatorname{tr}\left(\sigma(C_i)\cdots\sigma^{e-1}(C_i)C_i\right)\\ &=\operatorname{tr}\left(C_i\sigma(C_i)\cdots\sigma^{e-1}(C_i)\right)\\ &=\operatorname{tr}Q_i. \end{aligned}

Thus every coefficient of tr⁡Qi\operatorname{tr}Q_i is fixed by σ\sigma and belongs to F′\mathbb{F}'. The same holds for ΘL\Theta_L.

It remains to bound the degree. Substitution of homogeneous linear forms sends a homogeneous polynomial to either zero or a homogeneous polynomial of the same degree, and σ\sigma preserves degree. A summand in the expansion of tr⁡Qi\operatorname{tr}Q_i has the form

∏ν=0e−1σν((Ci)αν,αν+1),αe=α0,\prod_{\nu=0}^{e-1}\sigma^\nu\left((C_i)_{\alpha_\nu,\alpha_{\nu+1}}\right),\qquad\alpha_e=\alpha_0,

where 1≤αν≤βi1\leq\alpha_\nu\leq\beta_i. If this product is nonzero, its degree is, by (20),

∑ν=0e−1(q0ai,αν+1−ai,αν)=(q0−1)∑ν=0e−1ai,αν≤e(q0−1)B.\sum_{\nu=0}^{e-1}\left(q_0a_{i,\alpha_{\nu+1}}-a_{i,\alpha_\nu}\right) =(q_0-1)\sum_{\nu=0}^{e-1}a_{i,\alpha_\nu} \leq e(q_0-1)B.

Therefore every monomial occurring in ΘL\Theta_L has degree at most e(q0−1)Be(q_0-1)B. Since II is proper and homogeneous and L(0)=0L(0)=0, (19) gives ΘL(0)=1\Theta_L(0)=1. In particular, ΘL≠0\Theta_L\neq0 and

deg⁡ΘL≤e(q0−1)B.\deg\Theta_L\leq e(q_0-1)B.

Finally, if t>eBt>eB, then deg⁡ΘL<t(q0−1)\deg\Theta_L<t(q_0-1). Lemma 4.5, applied over F′\mathbb{F}', yields

∑w∈(F′)tΘL(w)=0.\sum_{w\in(\mathbb{F}')^t}\Theta_L(w)=0.

By (19), this sum equals ∣{w∈(F′)t:L(w)∈V(I)(F)}∣1F′\left|\{w\in(\mathbb{F}')^t:L(w)\in V(I)(\mathbb{F})\}\right|1_{\mathbb{F}'}. Since char⁡F′=p\operatorname{char}\mathbb{F}'=p, the required divisibility follows. □

Bounds and constructions for additive alternating codes

In this section, we prove Theorem 1.2, determining the maximum size of an additive alternating dd-code and constructing F\mathbb{F}-linear codes attaining this maximum for every admissible choice of parameters. We first apply the counting criterion of § 4.2 to the Pfaffian ideal, and then construct codes attaining the resulting bound.

Throughout this section, let F=Fq\mathbb{F}=\mathbb{F}_q, where q=peq=p^e, let n≥2n\ge2, and put N=(n2)N=\binom{n}{2}. We identify Alt⁡n(F)\operatorname{Alt}_n(\mathbb{F}) with FN\mathbb{F}^N by listing the entries above the diagonal in lexicographic order.

The upper bound

Theorem 5.1. Let r≥1r\ge1 be an integer and suppose that n≥2r+2n\ge2r+2. If an additive subgroup W⊆Alt⁡n(F)W\subseteq\operatorname{Alt}_n(\mathbb{F}) satisfies ∣W∣>qN−nr|W|>q^{N-nr}, then

∣{A∈W:rk⁡A≤2r}∣≡0(modp).\left|\{A\in W:\operatorname{rk} A\le2r\}\right|\equiv0\pmod{p}.

In particular, WW contains a nonzero matrix of rank at most 2r2r.

Proof. Put t=dim⁡FpWt=\dim_{\mathbb{F}_p}W and B=N−nrB=N-nr. Since ∣W∣=pet|W|=p^{et}, the hypothesis is equivalent to t>eBt>eB. Choose an Fp\mathbb{F}_p-basis A1,…,AtA_1,\ldots,A_t of WW, and define

A(y)=∑ν=1tyνAν,y=(y1,…,yt).A(y)=\sum_{\nu=1}^{t}y_\nu A_\nu,\qquad y=(y_1,\ldots,y_t).

Let L(y)L(y) be the NN-tuple of entries of A(y)A(y) above the diagonal, in lexicographic order. Each entry of L(y)L(y) is a homogeneous linear form over F\mathbb{F}, and w∈Fpt⟼A(w)∈Ww\in\mathbb{F}_p^t\longmapsto A(w)\in W is a bijection.

Let S=F[xij:1≤i<j≤n]S=\mathbb{F}[x_{ij}:1\le i<j\le n] and consider the Pfaffian ideal I2r+2⊆SI_{2r+2}\subseteq S from § 3. Proposition 3.1 shows that the largest shift of S/I2r+2S/I_{2r+2} is B=N−nrB=N-nr. Moreover, Lemma 2.1 gives

L(w)∈V(I2r+2)(F)⟺rk⁡A(w)≤2r(w∈Fpt).L(w)\in V(I_{2r+2})(\mathbb{F})\Longleftrightarrow\operatorname{rk} A(w)\le2r\qquad(w\in\mathbb{F}_p^t).

Apply Theorem 4.6 with F′=Fp\mathbb{F}'=\mathbb{F}_p, M=NM=N, and I=I2r+2I=I_{2r+2}. Since [F:Fp]=e[\mathbb{F}:\mathbb{F}_p]=e and t>eBt>eB, it follows that

∣{w∈Fpt:rk⁡A(w)≤2r}∣≡0(modp).\left|\{w\in\mathbb{F}_p^t:\operatorname{rk} A(w)\le2r\}\right|\equiv0\pmod{p}.

The bijection w↦A(w)w\mapsto A(w) proves the required congruence. The zero matrix belongs to the set being counted, so its cardinality is a positive multiple of pp. Consequently, this set contains a nonzero matrix. □

We are now ready to prove Theorem 1.2, which we restate for convenience.

Theorem 5.2. Let F=Fq\mathbb{F}=\mathbb{F}_q, let n≥2n\ge2, and let dd be even with 2≤d≤2⌊n/2⌋2\le d\le2\lfloor n/2\rfloor. Every additive dd-code C⊆Alt⁡n(F)C\subseteq\operatorname{Alt}_n(\mathbb{F}) satisfies

∣C∣≤qn(n−d+1)/2.(22)|C|\le q^{n(n-d+1)/2}. \tag*{(22)}

Moreover, equality is attained by an F\mathbb{F}-linear code for every such qq, nn and dd. Consequently, if F′⊆F\mathbb{F}'\subseteq\mathbb{F} is a subfield and CC is F′\mathbb{F}'-linear, then

dim⁡F′C≤[F:F′]n(n−d+1)2,(23)\dim_{\mathbb{F}'}C\le[\mathbb{F}:\mathbb{F}']\frac{n(n-d+1)}{2}, \tag*{(23)}

and this bound is also sharp.

Proof. Let C⊆Alt⁡n(F)C \subseteq\operatorname{Alt}_n(\mathbb{F}) be an additive dd-code. For d=2d = 2, the cardinality bound is simply ∣C∣≤∣Alt⁡n(F)∣=qN|C| \leq|\operatorname{Alt}_n(\mathbb{F})| = q^N. Suppose that d≥4d \geq4 and put r=(d−2)/2r = (d - 2)/2. Then r≥1r \geq1, n≥2r+2n \geq2r + 2, and

N−nr=n(n−d+1)2.N - nr = \frac{n(n-d+1)}{2}.

If ∣C∣>qn(n−d+1)/2|C| > q^{n(n-d+1)/2}, by Theorem 5.1 we know that there exists a nonzero matrix in CC of rank at most d−2d - 2. This contradicts the dd-code condition and proves (22).

Finally, let F′⊆F\mathbb{F}' \subseteq\mathbb{F} be a subfield and suppose that CC is F′\mathbb{F}'-linear. Such a code is additive, and

∣C∣=∣F′∣dim⁡F′C,q=∣F′∣[F:F′].|C| = |\mathbb{F}'|^{\dim_{\mathbb{F}'} C}, \qquad q = |\mathbb{F}'|^{[\mathbb{F}:\mathbb{F}']}.

Substitution in (22) gives (23). The sharpness assertions follow from Proposition 5.3 below. □

For WW as in Conjecture 1.1, μF(W)\mu_{\mathbb{F}}(W) is an F\mathbb{F}-linear (2r+2)(2r+2)-code of dimension dim⁡FW\dim_{\mathbb{F}} W. By the above theorem,

dim⁡FW≤n(n−2r−1)2=N−nr.\dim_{\mathbb{F}} W \leq\frac{n(n-2r-1)}{2} = N - nr.

This proves the Cooperstein conjecture.

Codes attaining the bound

We now give trace constructions attaining the upper bound. Codes with the required parameters were already constructed by Delsarte and Goethals for odd nn [5 Theorem 7], and by Cooperstein for both even and odd nn [3 Theorem 4.1 and Remark 4.2]. We give a linearized-polynomial description and verify their dimensions and the dd-code condition directly. Set L=Fqn\mathbb{L} = \mathbb{F}_{q^n}, viewed as an nn-dimensional vector space over F=Fq\mathbb{F} = \mathbb{F}_q, and write d=2sd = 2s and m=⌊n/2⌋m = \lfloor n/2 \rfloor, where 1≤s≤m1 \leq s \leq m. A qq-polynomial over L\mathbb{L} is a polynomial f(X)=∑iciXqif(X) = \sum_i c_i X^{q^i}, with coefficients ci∈Lc_i \in\mathbb{L}. For nonzero ff, its qq-degree is the largest index ii for which ci≠0c_i \ne0. Such a polynomial induces an F\mathbb{F}-linear map on L\mathbb{L}, and a nonzero qq-polynomial of qq-degree at most kk has at most qkq^k roots. Let

Ln,q={∑i=0n−1ciXqi:ci∈L}.\mathcal{L}_{n,q} = \left\{ \sum_{i=0}^{n-1} c_i X^{q^i} : c_i \in\mathbb{L} \right\}.

Evaluation on L\mathbb{L} identifies Ln,q\mathcal{L}_{n,q} with End⁡F(L)\operatorname{End}_{\mathbb{F}}(\mathbb{L}). The field trace is Tr⁡L/F(z)=∑j=0n−1zqj\operatorname{Tr}_{\mathbb{L}/\mathbb{F}}(z) = \sum_{j=0}^{n-1} z^{q^j}. It is F\mathbb{F}-linear, and the pairing (x,y)↦Tr⁡L/F(xy)(x,y) \mapsto\operatorname{Tr}_{\mathbb{L}/\mathbb{F}}(xy) is nondegenerate. Every F\mathbb{F}-bilinear form on L\mathbb{L} has a unique representation

Bf(x,y)=Tr⁡L/F(f(x)y),f∈Ln,q,(24)B_f(x,y) = \operatorname{Tr}_{\mathbb{L}/\mathbb{F}}(f(x)y), \qquad f \in\mathcal{L}_{n,q}, \tag*{(24)}

see e.g. [23 Lemma 13] and [16 Proposition 2.1]. Thus f↦Bff \mapsto B_f is an F\mathbb{F}-linear isomorphism from Ln,q\mathcal{L}_{n,q} to the space of F\mathbb{F}-bilinear forms on L\mathbb{L} and

rk⁡Bf=n−dim⁡Fker⁡f.(25)\operatorname{rk} B_f = n - \dim_{\mathbb{F}} \ker f. \tag*{(25)}

Under this correspondence, the alternating forms are represented by the F\mathbb{F}-subspace

An(q)={∑i=1n−1ciXqi:ci∈L, cn−i=−ciqn−i  for 1≤i<n}.(26)\mathcal{A}_n(q) = \left\{ \sum_{i=1}^{n-1} c_i X^{q^i} : c_i \in\mathbb{L},\ c_{n-i} = -c_i^{q^{n-i}}\ \text{ for } 1 \leq i < n \right\}. \tag*{(26)}

For odd n=2m+1n = 2m + 1, the coefficients c1,…,cmc_1,\ldots,c_m are arbitrary in L\mathbb{L} and determine all the others. For even n=2mn = 2m, the coefficients c1,…,cm−1c_1,\ldots,c_{m-1} are arbitrary in L\mathbb{L}, while cmc_m belongs to ker⁡Tr⁡L/H\ker\operatorname{Tr}_{\mathbb{L}/\mathbb{H}}, where H=FqmH=\mathbb{F}_{q^m}. If n=2m+1n=2m+1, let Fn,d\mathcal{F}_{n,d} consist of the qq-polynomials

f(X)=∑i=sm(aiXqi−aiqn−iXqn−i),as,…,am∈L.(27)f(X)=\sum_{i=s}^{m}\left(a_iX^{q^i}-a_i^{q^{n-i}}X^{q^{n-i}}\right),\qquad a_s,\ldots,a_m\in\mathbb{L}. \tag*{(27)}

If n=2mn=2m, let Fn,d\mathcal{F}_{n,d} consist of the qq-polynomials

f(X)=∑i=sm−1(aiXqi−aiqn−iXqn−i)+bXqm,(28)f(X)=\sum_{i=s}^{m-1}\left(a_iX^{q^i}-a_i^{q^{n-i}}X^{q^{n-i}}\right)+bX^{q^m}, \tag*{(28)}

where as,…,am−1∈La_s,\ldots,a_{m-1}\in\mathbb{L} and b∈Lb\in\mathbb{L} satisfies bqm=−bb^{q^m}=-b. In both cases, Fn,d\mathcal{F}_{n,d} is an F\mathbb{F}-subspace of An(q)\mathcal{A}_n(q). Define

Cn,d={Bf:f∈Fn,d}⊆Alt⁡(L),\mathcal{C}_{n,d}=\{B_f:f\in\mathcal{F}_{n,d}\}\subseteq\operatorname{Alt}(\mathbb{L}),

where BfB_f is given by (24). After choosing an ordered F\mathbb{F}-basis of L\mathbb{L}, we may also regard Cn,d\mathcal{C}_{n,d} as a code in Alt⁡n(F)\operatorname{Alt}_n(\mathbb{F}) via (6).

Proposition 5.3. The set Cn,d\mathcal{C}_{n,d} is an F\mathbb{F}-linear alternating dd-code of dimension n(n−d+1)/2n(n-d+1)/2 over F\mathbb{F}. In particular, it is an additive alternating dd-code attaining (22).

Proof. All exponent indices in (27) and (28) are distinct and less than nn, so the coefficients uniquely determine the parameters. For odd nn, there are m−s+1m-s+1 free parameters in L\mathbb{L}, giving

dim⁡FCn,d=n(m−s+1)=n(n−d+1)2.\dim_{\mathbb{F}}\mathcal{C}_{n,d}=n(m-s+1)=\frac{n(n-d+1)}{2}.

For even nn, there are m−sm-s free parameters in L\mathbb{L}, and the middle coefficient belongs to ker⁡Tr⁡L/H\ker\operatorname{Tr}_{\mathbb{L}/\mathbb{H}}, which has dimension mm over F\mathbb{F}. Therefore,

dim⁡FCn,d=n(m−s)+m=n(n−d+1)2.\dim_{\mathbb{F}}\mathcal{C}_{n,d}=n(m-s)+m=\frac{n(n-d+1)}{2}.

Finally, let 0≠f∈Fn,d0\ne f\in\mathcal{F}_{n,d}. All exponent indices of ff lie in [s,n−s][s,n-s]. Compose its induced map on the right with the bijection x↦xqn−sx\mapsto x^{q^{n-s}}. Using xqn=xx^{q^n}=x on L\mathbb{L}, the resulting map is represented by a nonzero qq-polynomial gg whose exponent indices lie in [0,n−2s][0,n-2s]. Indeed, each index ii becomes i−si-s modulo nn, and these indices remain distinct. Then

∣ker⁡f∣=∣ker⁡g∣≤qn−2s,dim⁡Fker⁡f≤n−2s.|\ker f|=|\ker g|\le q^{n-2s},\qquad\dim_{\mathbb{F}}\ker f\le n-2s.

By (25),

rk⁡Bf=n−dim⁡Fker⁡f≥2s=d.\operatorname{rk}B_f=n-\dim_{\mathbb{F}}\ker f\ge2s=d.

Since Cn,d\mathcal{C}_{n,d} is F\mathbb{F}-linear, this proves the dd-code condition. □\square

Proposition 5.3 gives an F\mathbb{F}-linear alternating code attaining (22) for every admissible qq, nn and dd. For every subfield F′⊆F\mathbb{F}'\subseteq\mathbb{F}, the same code has

dim⁡F′Cn,d=[F:F′]dim⁡FCn,d=[F:F′]n(n−d+1)2,\dim_{\mathbb{F}'}\mathcal{C}_{n,d}=[\mathbb{F}:\mathbb{F}']\dim_{\mathbb{F}}\mathcal{C}_{n,d}=[\mathbb{F}:\mathbb{F}']\frac{n(n-d+1)}{2},

so (23) is sharp as well.

Acknowledgments

The author is partially supported by the Italian National Group for Algebraic and Geometric Structures and their Applications (GNSAGA- INdAM).

Declaration of AI use

The author proposed the use of Pfaffian equations to describe the variety of alternating matrices of bounded rank and to count its points in an additive subgroup to obtain an upper bound on the size of an additive alternating code. Building on this idea, ChatGPT 6-Astra provided the counting theorem over subfields (Theorem 4.6), which uses the largest shift in a minimal graded free resolution to bound the degree of a polynomial indicator. The author subsequently developed, refined, and independently verified all arguments and takes full responsibility for the contents of the paper.

Paolo Santonastaso, Department of Mathematics and Applications “R. Caccioppoli”, University of Naples Federico II, Via Cintia, Monte Sant’Angelo, 80126 Naples, Italy

Email address: [email protected]

References

  1. [1]A. Abiad, G. N. Alfarano, and A. Ravagnani. Eigenvalue bounds and alternating rank-metric codes. Journal of Algebra and its Applications, 24(13–14):2541011, 2025.arxiv.org/abs/2405.09254
  2. [2]W. Bruns and J. Herzog. On the computation of a-invariants. Manuscripta mathematica, 77(1):201–213, 1992.DOI
  3. [3]B. N. Cooperstein. External flats to varieties in PG(∧²(V)) over finite fields. Geometriae Dedicata, 69(3):223–235, 1998.DOI
  4. [4]A. Cossidente, G. Marino, and F. Pavese. On symmetric and Hermitian rank distance codes. Journal of Combinatorial Theory, Series A, 188:105597, 2022.arxiv.org/abs/2011.06942
  5. [5]P. Delsarte and J.-M. Goethals. Alternating bilinear forms over GF(q). Journal of Combinatorial Theory, Series A, 19(1):26–50, 1975.DOI
  6. [6]S. R. Ghorpade and C. Krattenthaler. The Hilbert series of Pfaffian rings. In C. Christensen, A. Sathaye, G. Sundaram, and C. Bajaj, editors, Algebra, Arithmetic and Geometry with Applications, pages 337–356. Springer, Berlin, Heidelberg, 2004.
  7. [7]R. Gow. Rank-related dimension bounds for subspaces of bilinear forms over finite fields. arXiv preprint arXiv:1703.07266, 2017.arxiv.org/abs/1703.07266
  8. [8]R. Gow and R. Quinlan. On the vanishing of subspaces of alternating bilinear forms. Linear and Multilinear Algebra, 54(6):415–428, 2006.DOI
  9. [9]M. Hochster, J. Jeffries, V. Pandey, and A. K. Singh. When are the natural embeddings of classical invariant rings pure? Forum of Mathematics, Sigma, 11:e67, 2023.arxiv.org/abs/2210.09351
  10. [10]X.-d. Hou. Lectures on Finite Fields, volume 190 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2018.DOI
  11. [11]T. Józefiak and P. Pragacz. Ideals generated by Pfaffians. Journal of Algebra, 61(1):189–198, 1979.DOI
  12. [12]H. Kleppe and D. Laksov. The algebraic structure and deformation of Pfaffian schemes. Journal of Algebra, 64(1):167–189, 1980.DOI
  13. [13]V. Kodiyalam, T. Y. Lam, and R. G. Swan. Determinantal ideals, Pfaffian ideals, and the principal minor theorem. In Noncommutative Rings, Group Rings, Diagram Algebras and Their Applications, volume 456 of Contemporary Mathematics, pages 35–60. American Mathematical Society, Providence, RI, 2008.DOI
  14. [14]M. Kreuzer and L. Robbiano. Computational Commutative Algebra 2. Springer, 2005.
  15. [15]S. Lang. Algebra, volume 211 of Graduate Texts in Mathematics. Springer-Verlag, New York, revised third edition, 2002.
  16. [16]G. Longobardi, G. Lunardon, R. Trombetti, and Y. Zhou. Automorphism groups and new constructions of maximum additive rank metric codes with restrictions. Discrete Mathematics, 343(7):111871, 2020.DOI
  17. [17]V. P. Marinov. Perfection of ideals generated by the Pfaffians of an alternating matrix. I. Serdica, 9(1):31–42, 1983.
  18. [18]V. P. Marinov. Perfection of ideals generated by the Pfaffians of an alternating matrix. II. Serdica, 9(2):122–131, 1983.
  19. [19]M. Messia and F. Zullo. Non-additive symmetric rank-metric codes. arXiv preprint arXiv:2610.18061, 2026.arxiv.org/abs/2610.18061
  20. [20]U. Mushrraf. Perfect Hermitian rank-metric codes. The Art of Discrete and Applied Mathematics, 9(2):P2.08, 2026.arxiv.org/abs/2409.16753
  21. [21]U. Mushrraf and F. Zullo. On perfect symmetric rank-metric codes. Archiv der Mathematik, 125:259–271, 2025.arxiv.org/abs/2406.12450
  22. [22]P. Santonastaso and Y. Zhou. New maximum Hermitian d-codes and new invariants from skew polynomial rings. arXiv preprint arXiv:2610.26048, 2026.arxiv.org/abs/2610.26048
  23. [23]K.-U. Schmidt. Bilinear forms over finite fields of even characteristic. Journal of Combinatorial Theory, Series A, 117(8):1011–1026, 2010.DOI
  24. [24]K.-U. Schmidt. Symmetric bilinear forms over finite fields with applications to coding theory. Journal of Algebraic Combinatorics, 42(2):635–670, 2015.arxiv.org/abs/1410.7184
  25. [25]K.-U. Schmidt. Hermitian rank distance codes. Designs, Codes and Cryptography, 86(7):1469–1481, 2018.arxiv.org/abs/1702.02793
  26. [26]K.-U. Schmidt. Quadratic and symmetric bilinear forms over finite fields and their association schemes. Algebraic Combinatorics, 3(1):161–189, 2020.arxiv.org/abs/1803.04274
  27. [27]L. Taelman. Sheaves and Functions Modulo p: Lectures on the Woods Hole Trace Formula, volume 429 of London Mathematical Society Lecture Note Series. Cambridge University Press, 2016.
  28. [28]R. Trombetti and F. Zullo. On maximum additive Hermitian rank-metric codes. Journal of Algebraic Combinatorics, 54(1):151–171, 2021.arxiv.org/abs/2001.06628
  29. [29]R. H. Villarreal. Monomial Algebras. Monographs and Research Notes in Mathematics. Chapman & Hall/CRC, Boca Raton, FL, third edition, 2026.

Paper details

vlnc:W_pwapdvv2
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

ChatGPT 6-Astra

View statement
Looking for related papers…
Contents