betaFeedback

WorksMathematics

Extremal zero-sum sequences without elements of maximal order

Abstract

Let D(G)\mathsf D(G) denote the Davenport constant of a finite abelian group GG, the maximal length of a minimal zero-sum sequence over GG. Gao and Geroldinger conjectured that every minimal zero-sum sequence of length D(G)\mathsf D(G)~\cite[Conjecture~6.1]{GG99} and every zero-sumfree sequence of length D(G)−1\mathsf D(G)-1~\cite[Conjecture~4.1]{GG06} contains an element of order exp⁡(G)\exp(G). We disprove both conjectures by proving a density-one theorem: for every fixed finite abelian group HH of exponent M=2apbM=2^ap^b, where a≥2a\ge2, b≥1b\ge1, and pp is an odd prime, the set of integers r≥0r\ge0 for which Gr=C2r⊕HG_r=C_2^r\oplus H admits a minimal zero-sum sequence of length D(Gr)\mathsf D(G_r) with no term of order MM has natural density one.

Introduction

Let GG be a finite abelian group, written additively. A sequence over a group GG is a finite collection of elements in which repetitions are allowed and order is disregarded. It is zero-sumfree if it has no nonempty zero-sum subsequence. The small Davenport constant d(G)d(G) is the maximal length of such a sequence. The Davenport constant D(G)=d(G)+1D(G)=d(G)+1 is the least length forcing a nonempty zero-sum subsequence. Equivalently, it is the maximal length of a minimal zero-sum sequence, that is, a nonempty zero-sum sequence with no proper nonempty zero-sum subsequence.

The direct problem asks for the value of these constants; the inverse problem asks for the structure of the sequences attaining the extremal lengths. Inverse zero-sum problems are also closely connected with non-unique factorization theory, where zero-sum sequences over class groups provide a combinatorial framework for studying factorizations into irreducible elements; see [3, 4, 5]. Even when a complete description of the extremal sequences is unavailable, one can ask what their lengths force about the orders and multiplicities of their terms. The question whether an extremal sequence must contain an element of maximal order remains the basic one with the inverse problem of Davenport constant.

The standard construction gives a reason to expect such an element. Write

G=Cn1⊕⋯⊕Cnt,1<n1∣⋯∣nt,G=C_{n_1}\oplus\cdots\oplus C_{n_t},\qquad1<n_1\mid\cdots\mid n_t,

and choose a basis e1,…,ete_1,\ldots,e_t with ord⁡(ei)=ni\operatorname{ord}(e_i)=n_i. Then e1n1−1⋯etnt−1e_1^{n_1-1}\cdots e_t^{n_t-1} is zero-sumfree, and hence

d(G)≥d∗(G):=∑i=1t(ni−1).d(G)\geq d^*(G):=\sum_{i=1}^{t}(n_i-1).

Equality holds for pp-groups and for groups of rank at most two; see [8, 9] and [3 Section 3]. In these cases the construction is extremal and contains a term of order exp⁡(G)=nt\exp(G)=n_t. The inverse question is whether every sequence of maximal length must share this property.

Gao and Geroldinger proposed the following conjectures.

Conjecture 1 (1999 [1 Conjecture 6.1]). Every minimal zero-sum sequence of length D(G)D(G) over a finite abelian group GG contains an element of order exp⁡(G)\exp(G).

Conjecture 2 (2006 [3 Conjecture 4.1]). Every zero-sumfree sequence of length d(G)d(G) over a finite abelian group GG contains an element of order exp⁡(G)\exp(G).

2020 Mathematics Subject Classification. Primary 11B30; Secondary 11B75, 20K01.

Key words and phrases. Davenport constant, minimal zero-sum sequences, zero-sumfree sequences, inverse zero-sum problem.

There is substantial evidence for the assertions in small rank and in groups of prime-power order. Both Conjectures 1 and 2 hold for pp-groups and groups of rank at most two [1, 2]. The zero-sumfree assertion is known for cyclic groups, pp-groups, Cn⊕CnC_n \oplus C_n and C2⊕C2nC_2 \oplus C_{2n}; see [3 Section 4]. In fact, Girard proved that every zero-sumfree sequence of length d(G)d(G) over a finite abelian pp-group contains at least exp⁡(G)−1\exp(G)-1 terms of maximal order [6]. For G=C22⊕C2mG=C_2^2 \oplus C_{2m} with m≥2m \ge2, Schmid proved that every minimal zero-sum sequence of length D(G)D(G) has at least 2m−22m-2 terms of order 2m2m [10 Corollary 4.5]. This raises the question of whether even a single such term must remain when the elementary 22-component has increasing rank.

There is an additional difficulty in seeking counterexamples. Write D∗(G)=d∗(G)+1D^*(G)=d^*(G)+1. Minimal zero-sum sequences of length D∗(G)D^*(G) with no term of maximal order were already observed by Schmid; see [10 discussion following Corollary 4.6]. Such sequences do not settle the conjecture when D(G)>D∗(G)D(G)>D^*(G). The essential issue is to obtain the required order restrictions at the actual extremal length, even when the Davenport constant is not known explicitly.

Our main result disproves Conjectures 1 and 2 by establishing a density-one theorem. For every fixed finite abelian group HH of exponent 2apb2^a p^b, where a≥2a \ge2, b≥1b \ge1, and pp is an odd prime, the group C2r⊕HC_2^r \oplus H admits counterexamples to both conjectures for a set of integers r≥0r \ge0 of natural density one. Consequently, there are infinitely many pairwise nonisomorphic finite abelian groups of arbitrarily large rank for which both conjectures fail.

Theorem 1.1. Let HH be a finite abelian group of exponent M=2apbM=2^a p^b, where a≥2a \ge2, b≥1b \ge1 and pp is an odd prime, and set Gr=C2r⊕HG_r=C_2^r \oplus H. The set of integers r≥0r \ge0 admitting a minimal zero-sum sequence SS over GrG_r such that

∣S∣=D(Gr),ord⁡(g)∣M/2 or ord⁡(g)∣2afor every g∣S|S|=D(G_r),\qquad\operatorname{ord}(g)\mid M/2\ \text{or}\ \operatorname{ord}(g)\mid2^a\quad\text{for every }g\mid S

has natural density one. More precisely, the number of integers r∈[0,X]r\in[0,X] for which no such sequence exists is OH((log⁡(X+2))a)O_H((\log(X+2))^a) as X→∞X\to\infty.

Since both M/2M/2 and 2a2^a are proper divisors of M=exp⁡(Gr)M=\exp(G_r), every sequence SS provided by Theorem 1.1 is a counterexample to Conjecture 1. Deleting any term from SS yields a zero-sumfree sequence of length D(Gr)−1=d(Gr)D(G_r)-1=d(G_r) satisfying the same order restrictions, and hence a counterexample to Conjecture 2. Thus both conjectures fail for the same density-one set of ranks.

Notation and binary relations

We follow the conventions of [3 Section 2]. Sequences are written multiplicatively:

S=g1⋯gℓS=g_1\cdots g_\ell

Here ∣S∣=ℓ|S|=\ell is the length. We write T∣ST\mid S when TT is a subsequence of SS. The set of minimal zero-sum sequences is denoted by A(G)\mathcal{A}(G). We use the standard identities and inequality

D(G)=max⁡{∣S∣:S∈A(G)}=d(G)+1,d(H)+d(G/H)≤d(G)D(G)=\max\{|S|:S\in\mathcal{A}(G)\}=d(G)+1,\qquad d(H)+d(G/H)\le d(G)

for H≤GH\le G; see [3 Section 3].

We identify C2rC_2^r with the vector space F2r\mathbb{F}_2^r. Consider a nonempty sequence S=∏i=1N(gi,ai)S=\prod_{i=1}^{N}(g_i,a_i) over F2r⊕K\mathbb{F}_2^r\oplus K, where g1,…,gNg_1,\ldots,g_N span F2r\mathbb{F}_2^r. Let

V={(x1,…,xN)∈F2N:∑i=1Nxigi=0}.V=\left\{(x_1,\ldots,x_N)\in\mathbb{F}_2^N:\sum_{i=1}^{N}x_i g_i=0\right\}.

A vector x∈Vx\in V records precisely a subsequence Sx=∏i:xi=1(gi,ai)S_x=\prod_{i:x_i=1}(g_i,a_i) whose first coordinate has sum zero. Write λi(x)=xi\lambda_i(x)=x_i, and regard these values as integers 00 or 11 when defining

F(x)=∑i=1Naiλi(x)∈K.F(x)=\sum_{i=1}^{N}a_i\lambda_i(x)\in K.

Then dim⁡V=N−r\dim V=N-r, and SxS_x is zero-sum if and only if F(x)=0F(x)=0. In particular, SS is zero-sumfree if and only if F−1(0)={0}F^{-1}(0)=\{0\}. It belongs to A(F2r⊕K)\mathcal{A}(\mathbb{F}_2^r\oplus K) if and only if 1=(1,…,1)∈V\mathbf{1}=(1,\ldots,1)\in V and F−1(0)={0,1}F^{-1}(0)=\{0,1\}.

We shall also use this description in the opposite direction. Let VV be a binary vector space and let λ1,…,λN∈V∗\lambda_1,\ldots,\lambda_N \in V^* be such that the linear map Λ=(λ1,…,λN):V→F2N\Lambda=(\lambda_1,\ldots,\lambda_N):V\to\mathbb{F}_2^N is injective. For the quotient map q:F2N→F2N/Λ(V)q:\mathbb{F}_2^N\to\mathbb{F}_2^N/\Lambda(V), form the sequence

S=∏i=1N(q(ei),ai)over(F2N/Λ(V))⊕K.S=\prod_{i=1}^{N}(q(e_i),a_i)\quad\text{over}\quad(\mathbb{F}_2^N/\Lambda(V))\oplus K.

The incidence vector of a subsequence has first-coordinate sum zero exactly when it lies in Λ(V)\Lambda(V). Hence the preceding criteria apply to F=∑iaiλiF=\sum_i a_i\lambda_i. For minimality, the condition is F−1(0)={0,u}F^{-1}(0)=\{0,u\} for some u≠0u\ne0 satisfying λi(u)=1\lambda_i(u)=1 for every ii. This construction is what allows us to change the coordinate functions without changing the zero-sum subsequences.

Proof of the main theorem

Fix HH as in Theorem 1.1, and put A=2aA=2^a, Gr=C2r⊕HG_r=C_2^r\oplus H, and

kr=d(Gr)−r.k_r=d(G_r)-r.

The sequence (kr)r≥0(k_r)_{r\ge0} is nondecreasing: applying d(H)+d(G/H)≤d(G)d(H)+d(G/H)\le d(G) to Gr≤Gr+1G_r\le G_{r+1} gives d(Gr+1)≥d(Gr)+1d(G_{r+1})\ge d(G_r)+1. Also kr≥M−1k_r\ge M-1. Indeed, a binary basis together with M−1M-1 copies of an element of order MM in HH gives a zero-sumfree sequence of length r+M−1r+M-1.

Our construction has two parts. First, we replace the HH-coordinates of a minimal zero-sum sequence and add a controlled number of terms. This preserves minimality but increases the binary rank. We then choose ranks where the Davenport constant increases by exactly that number, so the new sequence is still extremal.

A form λ∈V∗\lambda\in V^* will be called normalized at uu if λ(u)=1\lambda(u)=1. All coordinate forms of a minimal zero-sum sequence are normalized at its full-sequence vector uu. Coefficients in a weighted sum of such forms are allowed to be zero. Thus any representation may be padded to a prescribed larger number of terms without affecting its sum function.

Lemma 3.1. Let u≠0u\ne0 belong to a binary vector space VV of dimension k+1k+1. For every a≥1a\ge1, any integer-weighted sum of forms normalized at uu can be represented modulo 2a2^a by at most ha(k)h_a(k) normalized forms, where

ha(0)=1,ha(k)=2∑j=0a−1(k−1j)(k≥1).h_a(0)=1,\qquad h_a(k)=2\sum_{j=0}^{a-1}\binom{k-1}{j}\quad(k\ge1).

Here a binomial coefficient is zero when its lower index exceeds its upper index. In particular, ha(k)=Oa((k+1)a−1)h_a(k)=O_a((k+1)^{a-1}).

Proof. If k=0k=0, there is just one normalized form, so one term suffices. If a=1a=1, every weighted sum is a binary linear form ℓ\ell. When ℓ(u)=1\ell(u)=1, it is itself normalized. When ℓ(u)=0\ell(u)=0, choose any normalized form η\eta and write ℓ=η+(ℓ+η)\ell=\eta+(\ell+\eta); both summands are normalized. This proves the required bound modulo two.

For a≥2a\ge2 and k≥1k\ge1, we prove the recurrence

ha(k)=ha(k−1)+ha−1(k−1).(1)h_a(k)=h_a(k-1)+h_{a-1}(k-1). \tag*{(1)}

Choose a decomposition V=W⊕⟨w⟩V=W\oplus\langle w\rangle with u∈Wu\in W, and let zz vanish on WW and satisfy z(w)=1z(w)=1. Extend each form on WW to vanish at ww. Every normalized form on VV is then uniquely ν\nu or ν+z\nu+z, where ν\nu is normalized on WW. Grouping the terms of a weighted sum QQ in this way gives

Q=∑νaνν+∑νbν(ν+z).Q=\sum_{\nu}a_\nu\nu+\sum_{\nu}b_\nu(\nu+z).

In this display the sum inside a binary form is taken in F2\mathbb{F}_2, whereas its value is interpreted as 0 or 1 in the weighted sum. The elementary identity ν+F2z=ν+z−2νz\nu+_{\mathbb{F}_2}z=\nu+z-2\nu z therefore gives

Q=Q0+z(C−2R),Q0=∑ν(aν+bν)ν,R=∑νbνν,C=∑νbν=R(u).(2)Q=Q_0+z(C-2R),\qquad Q_0=\sum_{\nu}(a_\nu+b_\nu)\nu,\quad R=\sum_{\nu}b_\nu\nu,\quad C=\sum_{\nu}b_\nu=R(u). \tag*{(2)}

By induction, represent RR modulo 2a−12^{a-1} on the kk-dimensional space WW as R0=∑jcjηjR_0=\sum_j c_j\eta_j, using at most ha−1(k−1)h_{a-1}(k-1) normalized forms. Evaluation at uu gives ∑jcj≡C(mod2a−1)\sum_j c_j\equiv C\pmod{2^{a-1}}. Adjusting one coefficient by the appropriate multiple of 2a−12^{a-1} makes

∑jcj=C(3)\sum_j c_j=C \tag*{(3)}

exactly, without changing the represented function modulo 2a−12^{a-1}. We may include a term with zero coefficient if the representation is empty.

Using (3.3) and the same binary identity,

(Q0−R0)+∑jcj(ηj+z)=Q0+z(C−2R0)≡Q(mod2a).(Q_0-R_0)+\sum_j c_j(\eta_j+z)=Q_0+z(C-2R_0)\equiv Q\pmod{2^a}.

The first summand is a weighted sum of normalized forms on WW, so it can be compressed modulo 2a2^a to ha(k−1)h_a(k-1) terms. Each ηj+z\eta_j+z is normalized at uu, since z(u)=0z(u)=0. This proves the recurrence by induction on a+ka+k.

At k=1k=1, the recurrence gives ha(1)=2h_a(1)=2. For k≥2k\ge2, Pascal’s identity verifies the displayed formula:

∑j=0a−1(k−2j)+∑j=0a−2(k−2j)=∑j=0a−1(k−1j).\sum_{j=0}^{a-1}\binom{k-2}{j}+\sum_{j=0}^{a-2}\binom{k-2}{j}=\sum_{j=0}^{a-1}\binom{k-1}{j}.

The formula is a polynomial bound of degree at most a−1a-1 in kk, which gives the final estimate. □

Proof of Theorem 1.1. Decompose H=K⊕CAtH=K\oplus C_A^t, where t≥1t\ge1 and exp⁡(K)∣M/2\exp(K)\mid M/2.

For an extremal minimal zero-sum sequence SS over GrG_r, the binary projections span F2r\mathbb{F}_2^r. Indeed, if their span had dimension s<rs<r, then SS would be a minimal zero-sum sequence over a subgroup isomorphic to GsG_s, whereas the subgroup inequality gives D(Gs)≤D(Gr)−(r−s)<∣S∣D(G_s)\le D(G_r)-(r-s)<|S|, a contradiction. Thus its binary relation data satisfy N=D(Gr)N=D(G_r), dim⁡V=kr+1\dim V=kr+1 and F−1(0)={0,u}F^{-1}(0)=\{0,u\}. Apply Lemma 3.1 to the tt CAC_A-coordinate functions of FF. Combining and padding the representations gives h=tha(kr)h=th_a(kr) forms μj\mu_j normalized at uu and weights bj∈CAtb_j\in C_A^t reproducing that component. Replace each original HH-label by its KK-coordinate with zero CAtC_A^t-component, and append hh labels (0,bj)(0,b_j) with corresponding forms μj\mu_j. The extended coordinate map remains injective and its HH-valued sum function is FF. § 2 therefore gives a minimal zero-sum sequence over Gr+hG_{r+h} of length N+hN+h. Its first NN terms have orders dividing M/2M/2, and the additional terms have orders dividing AA.

If kr=kk_r=k on an interval, set h=tha(k)h=th_a(k). Apart from the first hh ranks, each rank rr has r−hr-h in the same interval. Construction at r−hr-h gives

D(Gr−h)+h=r+k+1=D(Gr).D(G_{r-h})+h=r+k+1=D(G_r).

Thus at most tha(k)th_a(k) ranks in each constant interval are exceptional. Liu’s estimate [7 Proposition 9.1], valid for any fixed HH whose odd part is a pp-group, gives kX=OH(log⁡(X+2))k_X=O_H(\log(X+2)). Hence, for integer X≥0X\ge0, the number E(X)E(X) of exceptional ranks satisfies

E(X)≤t∑k=M−1kXha(k)=OH((kX+1)a)=OH((log⁡(X+2))a).E(X)\le t\sum_{k=M-1}^{k_X}h_a(k)=O_H((k_X+1)^a)=O_H((\log(X+2))^a).

Dividing by X+1X+1 gives natural density one. This completes the proof of the theorem. □

References

Author information

Yongke Qu

Department of Mathematics, Luoyang Normal University, Luoyang 471934, P.R. China

E-mail: [email protected]

Guoqing Wang*

School of Mathematical Sciences, Tiangong University, Tianjin 300387, P.R. China

E-mail: [email protected]

References

  1. [1]W. Gao and A. Geroldinger, On long minimal zero sequences in finite abelian groups, Period. Math. Hungar. 38 (1999), 179–211.DOI
  2. [2]W. Gao and A. Geroldinger, On the order of elements in long minimal zero-sum sequences, Period. Math. Hungar. 44 (2002), 63–73.DOI
  3. [3]W. Gao and A. Geroldinger, Zero-sum problems in finite abelian groups: A survey, Expo. Math. 24 (2006), 337–369.DOI
  4. [4]W. Gao, A. Geroldinger, and W. A. Schmid, Inverse zero-sum problems, Acta Arith. 128 (2007), 245–279.DOI
  5. [5]A. Geroldinger and F. Halter-Koch, Non-Unique Factorizations: Algebraic, Combinatorial and Analytic Theory, Pure and Applied Mathematics, vol. 278, Chapman & Hall/CRC, 2006.
  6. [6]B. Girard, Inverse zero-sum problems in finite Abelian p-groups, Colloq. Math. 120 (2010), 7–21.DOI
  7. [7]C. Liu, Davenport constants with growing binary rank, preprint, arXiv:2610.38447v1, 2026.arxiv.org/abs/2610.38447
  8. [8]J. E. Olson, A combinatorial problem on finite Abelian groups, I, J. Number Theory 1 (1969), 8–10.DOI
  9. [9]J. E. Olson, A combinatorial problem on finite Abelian groups, II, J. Number Theory 1 (1969), 195–199.DOI
  10. [10]W. A. Schmid, The inverse problem associated to the Davenport constant for C₂ ⊕ C₂ ⊕ C₂, and applications to the arithmetical characterization of class groups, Electron. J. Combin. 18 (2011), Paper P33.

Paper details

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

Not recorded for this paper

View statement
Looking for related papers…
Contents