A multiplicative property for zero-sums. I (Q2144571)

Many classical results in combinatorics tell us that a certain property must emerge when a certain set becomes large enough. One is then compelled to ask the inverse question: what is the structure of the system just below the threshold, just before the property does emerge. Davenport's constant of an abelian group \(G\) is the size at which every sequence of elements from \(G\) must contain a zero-sum subsequence. Davenport's constant for \(G = C_n \oplus C_n\), where \(C_n\) is the cyclic group, is \(2n-1\); and the structure of zero-sum-free sequences of size \(2n-2\) is known: some element must repeat \(n-1\) times. There is a refinement: any sequence from \(G\) of size at least \(3n-2\) must contain a zero-sum subsequence of size at most \(n\). Again, the structure of sequences of size \(3n-3\) without any such subsequence is known: a pair of generators must repeat \(n-1\) times each. These two results have a common extension: for \(0 \leq k \leq n-1\), any sequence of size at least \(2n-1+k\) must contain a zero-sum subsequence of size at most \(2n-1-k\) [\textit{C. Wang} and \textit{K. Zhao}, J. Number Theory 176, 365--374 (2017; Zbl 1422.11057)]. Concerning the inverse problem, it is conjectured (for \(k = 2,\dots,n-2\)) that a sequence of size \(2n-2+k\) without any zero-sum subsequence of size at most \(2n-1-k\) must have the form \(S = e_1^{[n-1]}e_2^{[n-1]}(e_1+e_2)^{[k]}\) for some pair of generators \(e_1,e_2 \in G\). This conjecture is the subject of the current paper. The authors prove (Theorem 1.2) that if the conjecture holds for \(C_n \oplus C_n\) and for \(C_m \oplus C_m\), then it holds for \(C_{nm} \oplus C_{nm}\). This reduces the problem to the case of \(n\) being a prime. The proof is obtained by a delicate analysis of a sequence \(S\) in \(C_{nm}\oplus C_{nm}\) satisfying the given restrictions, together with its image in \(C_n \oplus C_n\). A key idea here is played by ``good'' terms of \(S\), which are the terms for which there are no other terms in \(S\) with the same image. In [Electron. J. Comb. 29, No. 3, Research Paper P3.12, 16 p. (2022; Zbl 1502.11018)] the authors reduce the inverse problem for any abelian group of rank \(2\) to the case considered in this paper.