The catenary degree of Krull monoids II

Let H be a Krull monoid with finite class group G such that |G| \ge 3 and each class contains a prime divisor. Recall that the monoid of zero-sum sequences \mathcal B (G) is a Krull monoid with class group isomorphic to G and each class contains a prime divisor. By Proposition 2.1, \daleth (H) = \daleth (G) and \mathsf c (H) = \mathsf c (G). Thus it is sufficient to prove Theorem 1.1 for the Krull monoid \mathcal B (G).

Let \mathcal O be a holomorphy ring in a global field K, and R a classical maximal \mathcal O-order in a central simple algebra A over K such that every stably free left R-ideal is free. Then the monoid R^{\bullet} is a non-commutative Krull monoid ([Ge13a]), and all invariants under consideration of R^{\bullet} coincide with the respective invariants of a commutative Krull monoid whose class group is isomorphic to a ray class group of \mathcal O. These (highly non-trivial) transfer results are established in [Sm13a, Ba-Sm15], and are summarized in [Ba-Sm15, Theorems 7.6 and 7.12]. Therefore, both for Theorem 1.1 and for Corollary 1.2, it is sufficient to prove the equivalence of the statements for a monoid of zero-sum sequences.

Let G be a finite abelian group with |G| \ge 3, and recall the inequalities

(c)  \Rightarrow  (a) Suppose that G is isomorphic either to C_2^{r-1} \oplus C_4 for some r \ge 2 or to C_2 \oplus C_{2n} for some n \ge 2. Then Theorem A (in the Introduction) shows that \mathsf c (G) \le \mathsf D (G)-1. Since \mathsf D (C_2^{r-1}\oplus C_4)= r+3 and \mathsf D (C_2 \oplus C_{2n})=2n+1, Lemma 3.1.2 implies that \mathsf c (G) \ge \daleth (G) \ge \mathsf D (G)-1.

(a)  \Rightarrow  (b) Suppose that \mathsf c (G) = \mathsf D (G)-1. By Theorem A, G is neither cyclic nor an elementary 2-group which implies that \mathsf D(G)\ge 5. By Lemma 3.1.1, we have

Thus, if \daleth (G)<\mathsf D (G)-1, then \Big\lfloor\frac{1}{2} \mathsf D(G)+1 \Big\rfloor \ge \mathsf D(G)-1 which implies that \mathsf D(G)\le 4, a contradiction.

(b)  \Rightarrow  (c) Suppose that \daleth (G) = \mathsf D (G)-1. Again by Theorem A, G is neither cyclic nor an elementary 2-group which implies that \mathsf D (G)\ge 5. Then there exist U, V \in \mathcal A (G) such that \min \big( \mathsf L (UV) \setminus \{2\} \big) = \mathsf D (G)-1. Obviously, there are the following four cases (up to symmetry):

• \mathsf L (UV)= \{2, \mathsf D (G)-1, \mathsf D (G)\}.

• \mathsf L (UV)=\{2, \mathsf D (G)-1\} and |U|=|V|=\mathsf D (G).

• \mathsf L (UV)=\{2, \mathsf D (G)-1\} and |U|-1=|V|=\mathsf D (G).

• \mathsf L (UV)=\{2, \mathsf D (G)-1\} and |U|=|V|=\mathsf D (G)-1.

These cases are handled in the Propositions 3.5 to 3.8, and they imply that G is isomorphic either to C_2^{r-1} \oplus C_4 for some r \ge 2 or to C_2 \oplus C_{2n} for some n \ge 2.

Let G and G' be abelian groups such that \mathcal L (G) = \mathcal L (G'). Then

If G is finite, then \daleth (G) \le \mathsf c (G) \le \mathsf D (G) < \infty. If G is infinite, then, by the Theorem of Kainrath (see [Ka99a] or [Ge-HK06a, Section 7.3]), every finite set L \subset \mathbb N_{\ge 2} lies in \mathcal L (G), which implies that \daleth (G)=\infty.

For k \in \mathbb N, we define the refined elasticities

and observe that \rho_k (G) = \rho_k (G'). This implies that k \mathsf D (G) = \rho_{2k} (G)=\rho_{2k} (G') = k \mathsf D (G') (see [Ge-HK06a, Section 6.3]) for each k \in \mathbb N, and hence \mathsf D (G) = \mathsf D (G').

Now suppose that G' \in \{C_2^{r-1} \oplus C_4, C_2 \oplus C_{2n} \} where r, n \ge 2. Then Theorem 1.1 implies that \daleth (G') = \mathsf D (G')-1. Since \mathcal L (G) = \mathcal L (G'), it follows that G is finite and that

whence Theorem 1.1 implies that G \in \{C_2^{r-1} \oplus C_4, C_2 \oplus C_{2n}\} with n, r \ge 2. Suppose now that n, r \ge 3. Clearly, Condition (b) in Proposition 3.5 is equivalent to

• (b’) \{2, \mathsf D (G)-1, \mathsf D (G)\} \in \mathcal L (G).

Thus Proposition 3.5 implies in particular that \mathcal L ( C_2 \oplus C_{2n} ) \ne \mathcal L ( C_2^{r-1} \oplus C_4 ), and thus the assertion of Corollary 1.3 follows.

See Proposition 4.1 and Theorem 4.2 in [Ge-Gr-Sc11a].

1. See Lemma 5.2 in [Ge-Gr-Sc11a].

2. Let g \in G. If \mathsf v_g (U)=0 or \mathsf v_{-g} (V)=0, then the assertion is clear. If \mathsf v_g (U)>0 and \mathsf v_{-g} (V)>0, then the assertion follows from 1. with K = \langle g \rangle.

First, we prove the moreover statement. Let U, V \in \mathcal A (G) with \mathsf L (UV) \cap [2, 5] = \{2, 5\}. Then \daleth \bigl( \langle \operatorname{supp} (UV) \rangle \bigr) \ge 5. Since for every proper subgroup K of G we have \daleth (K) \le \daleth (C_2 \oplus C_4) < \mathsf D (G)=5 by Theorem A, it follows that \langle \operatorname{supp} (UV) \rangle = G.

Recall that \mathsf D (G) = 7, and note that it suffices to prove \daleth (G) \le 5, since then combing with [Ge-Gr-Sc11a, Proposition 4.1.2 and Corollary 4.3] yields 5 \le \daleth (G) = \mathsf c (G) \le 5.

Let U, V \in \mathcal A (G) with \max \mathsf L (UV) > 5 be given. Since \max \mathsf L (UV) \le \min \{|U|, |V| \} \le \mathsf D (G) = 7, it follows that \min \{|U|, |V| \} \in \{6,7\}. We have to show that there exists a factorization U V = W_1 \cdot \ldots \cdot W_k with W_1, \ldots, W_k \in \mathcal A (G) and k \in [3, 5]. We start with two special cases.

First, suppose that V = -U and |U| = 7. Then [Ge-HK06a, Theorem 6.6.7] implies that 4 \in \mathsf L (UV), and thus the assertion follows.

Second, suppose that there exist W_1, W_2 \in \mathcal A (G) such that W_1W_2 \, | \, UV, 5 \ge |W_1| \ge |W_2|, and |W_1 W_2| \ge 7. Then there exist k \in \mathbb N_{\ge 3}, W_3, \ldots, W_k \in \mathcal A (G) such that UV = W_1 \cdot \ldots \cdot W_k, and

implies k \le 5, and the assertion follows.

Assume to the contrary that UV has no factorization of length k \in [3, 5]. Then none of the two special cases holds true. By [Ge-Gr09c, Lemma 3.6], UV has a zero-sum subsequence W_1 \in \mathcal A (G) of length |W_1| \in [2,4], and suppose that |W_1| is maximal. Then there is a factorization

By assumption we have k \ge 6. Since k=7 would imply that V = -U and |U| = 7, it follows that k=6. We distinguish three cases.

CASE 1:   |W_1| = 2.

Since |W_1| is maximal, we get |W_1| = \ldots = |W_k| = 2, and thus |UV| \in \{12, 14\} and V = -U. Since we are not in the first special case, it follows that |U| = 6, say U = g_1 \cdot \ldots \cdot g_6.

If \mathsf h (U) = 3, say g_1=g_2=g_3, then W = (-g_1)g_4g_5g_6 \in \mathcal A (G) with W \, | \, UV, a contradiction. Thus \mathsf h (U) \le 2. Since \mathsf D (C_2^2) = 3, \operatorname{supp} (U) contains at most two elements of order 2, say \operatorname{ord} (g_1) = \ldots = \operatorname{ord} (g_4) = 4. Since \mathsf D (C_2 \oplus C_4) = 5, it follows that \operatorname{supp} (U) generates G. Recall that the maximal size of a minimal generating set of G equals \mathsf r^* (G) = 2, and that every generating set contains a basis (see [Ge-HK06a, Appendix A]). Thus the elements of order 4 contain a basis.

Suppose there are two elements with multiplicity two, say g_1 = g_3 and g_2 = g_4. Then (g_1, g_2) is a basis of G and g_5 = a g_1 + b g_2 with a, b \in [1,3]. Thus there is a zero-sum subsequence W \in \mathcal A (G) with |W| > 2, W \, | \, UV and \operatorname{supp} (W) \in \{g_5, g_1, g_2, - g_1, -g_2\}, a contradiction.

Suppose there is precisely one element with multiplicity two, say g_1 = g_3. Then (g_1, g_2) is a basis of G and there is an element in \operatorname{supp} (U) of the form ag_1 + bg_2 with a \in [1,3] and b \in \{1,3\}. As above we obtain a zero-sum sequence W with |W| > 2, a contradiction.

Suppose that \mathsf h (U) = 1. Then (g_1, g_2) is a basis of G. Then there is one element in \operatorname{supp} (U) which is not of the form \{ g_1+2g_2, 2g_1+g_2, 2g_1+2g_2\}, and hence it has the form ag_1+bg_2 with a, b \in \{1,3\}, and we obtain a contradiction as above.

CASE 2:   |W_1| = 4.

Since k=6, |UV| \le 14 and |W_1| = 4, we get |W_2| = \ldots = |W_6| = 2 and |U| = |V| = 7. By [Ge-HK06a, Example 5.8.8], there exists a basis (e_1, e_2) of G such that

with a_1, a_2, a_3 , a_4 \in [0,3] and a_1+a_2+a_3+a_4 \equiv 1 \mod 4. We set X = \gcd (W_1, U). Then there are X, Y, Z \in \mathcal F (G) such that U = X Z, V = Y (-Z) and W_1 = X Y. After renumbering if necessary there are the following three cases:

If X = e_1^2, then (-e_1)(a_{1}e_1 + e_2) \cdot \ldots \cdot (a_{4}e_1 + e_2) is a minimal zero-sum subsequence of UV of length 5, a contradiction.

Suppose that X = (a_1e_1+e_2)(a_2e_1+e_2). If a_3 \ne a_4, then (a_3e_1+e_2)(-a_4e_1-e_2) e_1^3 has a zero-sum subsequence of length 4 or 5, a contradiction. Suppose that a_3 = a_4. Then (e_1, a_3e_1+e_2) is a basis, and after changing notation if necessary we may suppose that a_3= 0. Then (a_1e_1 + e_2)(a_2e_1+e_2) \in \{ e_2(e_1+e_2), (2e_1+e_2)(-e_1+e_2)\}. In the first case e_1^3 (e_1+e_2)(-e_2) and in the second case (-e_1)^3(-e_1+e_2)(-e_2) is a zero-sum subsequence of UV of length 5, a contradiction.

Suppose that X = e_1 (a_1e_1+e_2). If two of the a_2, a_3, a_4 are distinct, say a_2 \ne a_3, then (a_2e_1+e_2)(-a_3e_1-e_2) e_1^2 (-e_1)^2 has a zero-sum subsequence of length greater than 2, a contradiction. Suppose that a_2 = a_3 = a_4. Then (e_1, e_2'= a_2e_1+e_2) is a basis. Then U = e_1^3 (e_1+e_2'){e_2'}^3 and e_1^3(e_1+e_2') (-e_2') is a minimal zero-sum subsequence of UV of length 5, a contradiction.

CASE 3:  |W_1| = 3.

After renumbering if necessary we may suppose that |U| = 7, |V| \in \{6,7\}, |W_2| \in \{2,3\} and |W_3| = \ldots = |W_6| = 2. By [Ge-HK06a, Example 5.8.8], there exists a basis (e_1, e_2) of G such that

with a_1, a_2, a_3 , a_4 \in [0,3] and a_1+a_2+a_3+a_4 \equiv 1 \mod 4. Thus U has a subsequence S of length |S| = 4 such that -S is a subsequence of V.

Suppose there are two distinct elements in \{a_1e_1+e_2, \ldots, a_4e_1+e_2\}, say (a_1e_1+e_2) and (a_2e_1+e_2) such that (a_1e_1+e_2)(a_2e_1+e_2) \, | \, S. Then either (a_1e_1+e_2)(-a_2e_1-e_2)e_1^3 or (-a_1e_1-e_2)(a_2e_1+e_2)e_1^3 contains a zero-sum subsequence of length 4 or 5, a contradiction.

Now we suppose that this does not hold and distinguish three cases.

Suppose S = e_1 (a_1e_1+e_2)(a_2e_1+e_2)(a_3e_1+e_2). Then a_1=a_2=a_3=a, and since (e_1, ae_1+e_2) is a basis of G, we may suppose that a= 0. Then U = e_1^3e_2^3(e_1+e_2) and (-e_1)e_2^3(e_1+e_2) is a zero-sum subsequence of UV of length 5, a contradiction.

Suppose S = e_1^2 (a_1e_1+e_2)(a_2e_1+e_2). Then a_1=a_2, and since (e_1, ae_1+e_2) is a basis of G, we may suppose that a= 0. Then U = e_1^3 e_2^2 (ae_1+e_2) \bigl( (1-a)e_1+e_2 \bigr), and either (-e_2)e_1^3(ae_1+e_2) or (-e_2)e_1^3\bigl( (1-a)e_1+e_2 \bigr) has a zero-sum subsequence of length greater than or equal to 4, a contradiction.

Suppose S = e_1^3 (a_1e_1+e_2). Again we may suppose that a=0 and find a zero-sum subsequence of UV of length greater than or equal to 4, a contradiction.

1. Assume to the contrary that g_1=2g_2. Since |U| = \mathsf D (G), it follows that \Sigma (g_1^{-1}U) = G \setminus \{0\}. Hence there exists some W \in \mathcal F (G) such that W\, | \, g_1^{-1}U and \sigma (W)=-g_2. If g_2\, | \, W, then g_1 g_2^{-1}W is a proper zero-sum subsequence of U, a contradiction. If g_2 \nmid W, then g_2W is a proper zero-sum subsequence of U, a contradiction.

Assume to the contrary that \operatorname{ord}(h)=2 and g_1=2g_2+h. There exists some W \in \mathcal F (G) such that W\, | \, g_1^{-1}U and \sigma(W)=-g_2+h. If g_2\, | \, W, then g_1g_2^{-1}W is a proper zero-sum subsequence of U, a contradiction. If g_2\nmid W and h\, | \, W, then g_2h^{-1}W is a proper zero-sum subsequence of U, a contradiction. If g_2\nmid W and h\nmid W, then g_2hW is a proper zero-sum subsequence of U, a contradiction.

2. Since |(-g)^k g^{-k}U|=|U|=\mathsf D(G), there exists some W \in \mathcal A (G) such that W\, | \, (-g)^kg^{-k}U. It is easy to show that |W|\neq 2. Assume the contrary that |W|=\mathsf D(G). Then W=(-g)^k g^{-k}U and \operatorname{ord}(g)=2k>k+1. Let h\in \operatorname{supp}(U)\setminus \{g\}. Since \Sigma (h^{-1}U) = G \setminus \{0\}, there exists some T \in \mathcal F (G) such that T\, | \, h^{-1}U and \sigma (T)=(k+1)g. If g\nmid T, then g^{k-1}T is a proper zero-sum subsequence of U, a contradiction. Suppose that g\, | \, T, say T=g^tT_0 with t \in [1,k] and T_0\, | \, g^{-k}U. Then W'=(-g)^{k-t+1}T_0 is a zero-sum subsequence of W=(-g)^k g^{-k}U. Since W is an atom, it follows that W'=W. This implies that T_0=g^{-k}U, a contradiction to T_0 \, | \, T \, | \, h^{-1}U.

(a)  \Rightarrow  (b) Let (e_1, e_2) be a basis of G with \operatorname{ord} (e_1)=2 and \operatorname{ord}(e_2)=2n with n \ge 2. Then \mathsf D (G)=2n+1, U = e_1e_2^{2n-1}(e_1+e_2) \in \mathcal A (G), and \mathsf L \big( U(-U) \big)=\{2,\mathsf D(G)-1,\mathsf D(G)\}.

(b)  \Rightarrow  (a) Since \mathsf D (G) \in \mathsf L (UV), it follows that |U|=|V|=\mathsf D(G) and V=-U. Let A \in \mathcal A (G) with A\, | \, U(-U). Then |A|\in \{2,3,\mathsf D(G)\}, and if |A|=\mathsf D(G), then U(-U) A^{-1} is an atom of length \mathsf D(G). Since |U|=\mathsf D (G), it follows that \Sigma (U) =G. By [Ge-HK06a, Theorem 6.6.3], G is neither cyclic nor an elementary 2-group. Therefore, |\operatorname{supp} (U)|\ge 2 and may write U=g_1^{k_1} \cdot \ldots \cdot g_s^{k_s} with g_1, \ldots, g_s \in G pairwise distinct, s \ge 2, k_1, \ldots, k_s \in \mathbb N, and \operatorname{ord}(g_1)>2.

We continue with the following assertion.

1. A1.  If \operatorname{ord}{g_i}>2 for some i \in [1,s], then there exist distinct elements h_1, h_2 \in \operatorname{supp} (U) such that g_i=h_1+h_2.

2. A2.  If \operatorname{ord}{g_i}>2 for some i \in [1,s], then there exist distinct elements f_1, f_2 \in \operatorname{supp} (U) such that g_i=f_1+f_2, \operatorname{ord} (f_1)>2, and \operatorname{ord} (f_2)=2.

Proof of  A1.  Let i \in [1,s] with \operatorname{ord} (g_i)>2. By Lemma 3.4.2, there exists an atom W \in \mathcal A (G) such that W \, | \, g_i^{-k_i}(-g_i)^{k_i}U and |W| \in [3, \mathsf D(G)-1]. Then W \, | \, U(-U) infers |W|=3. Thus W=(-g_i)^2h_1 with h_1\, | \, U or W=(-g_i)h_1h_2 with h_1h_2\, | \, U. In the first case h_1=2g_i, a contradiction to Lemma 3.4.1. Thus the second case holds, and again Lemma 3.4.1 implies that h_1 and h_2 are distinct.

Proof of  A2.  Let i \in [1,s] with \operatorname{ord} (g_i)>2, say i=1. By A1, we may assume that g_1=g_2+g_3 and hence \operatorname{ord}(g_2)>2 or \operatorname{ord}(g_3)>2. Assume to the contrary that \operatorname{ord}(g_2)>2 and \operatorname{ord}(g_3)>2. Again by A1, there are distinct h_1, h_2 \in \operatorname{supp} (U) such that g_2=h_1+h_2. Then \{h_1,h_2\}\cap \{g_1,g_3\}\neq \emptyset otherwise (-g_1)h_1h_2g_3 would be an atom of length 4 dividing U(-U). Since g_2=g_1+h_i with i \in [1,2] cannot hold, we obtain that g_2=g_3+h with h \in \{h_1, h_2\}. Repeating the argument we infer that g_3=g_2+h' with h' \in \operatorname{supp} (U). It follows that h+h'=0 which implies that h=h', \operatorname{ord}(h)=2, and g_1=2g_3+h, a contradiction to Lemma 3.4.1.

Since \operatorname{ord}(g_1)>2, by A2 we may assume that g_1=g_2+g_3 with \operatorname{ord}(g_2)>2 and \operatorname{ord}(g_3)=2. If s=3, then G = \langle \operatorname{supp} (U) \rangle = \langle g_1, g_2, g_3 \rangle = \langle g_2, g_3 \rangle \cong C_2\oplus C_{2n} with n \ge 2. Assume to the contrary that s\ge4. We distinguish two cases.

CASE 1:   There is an i \in [4,s] such that \operatorname{ord} (g_i) > 2, say i=4.

By A2, we may assume g_4=h_1+h_2 with \operatorname{ord}(h_1)>2 and \operatorname{ord}(h_2)=2. We assert that \{g_1,g_2,g_3\}\cap \{g_4, h_1,h_2\}=\emptyset. Assume to the contrary that this does not hold. Taking into account the order of the elements and that |\{g_1, \ldots, g_4\}|=4 we have to consider the following three cases:

• If h_2=g_3, then g_1+g_4=g_2+h_1 and hence (-g_1)(-g_4)g_2h_1 is an atom of length 4 dividing U(-U), a contradiction.

• If h_1=g_1, then (-g_4)g_2g_3h_2 is an atom of length 4 dividing U(-U), a contradiction.

• If h_1=g_2, then (-g_4)g_1g_3h_2 is an atom of length 4 dividing U(-U), a contradiction.

Thus we have \{g_1,g_2,g_3\}\cap \{g_4, h_1,h_2\}=\emptyset. Obviously, (-g_1)g_2g_3, g_1(-g_2)(-g_3), (-g_4)h_1h_2, and g_4(-h_1)(-h_2) are atoms of length 3 dividing U(-U), and therefore their product (-g_1)g_2g_3\cdot g_1(-g_2)(-g_3)\cdot (-g_4)h_1h_2\cdot g_4(-h_1)(-h_2) divides U(-U), a contradiction to \mathsf L \big( U(-U) \big)= \{2, \mathsf D (G)-1, \mathsf D (G)\}.

CASE 2:   \operatorname{ord}(g_i)=2 for all i\in [4,s].

If k_1\ge 2 and k_2\ge 2, then (-g_1)^2g_2^2 is an atom of length 4 dividing U(-U), a contradiction. Thus k_1=1 or k_2=1, say k_1=1. Since g_1+k_2g_2+g_3+\ldots+g_s=\sigma (U)=0 and (k_2+1)g_2=g_4+\ldots+g_s, it follows that \operatorname{ord} (g_2)=2(k_2+1). Thus

a contradiction.

Assume to the contrary that U, V \in \mathcal A (G) with these properties do exist. If A\in \mathcal A(G) with A\, | \, UV, then |A|\in\{2,3,4,\mathsf D(G)\}. Since \Sigma (U) = \Sigma (V)=G and G cannot be cyclic, it follows that |\operatorname{supp} (U)|\ge 2 and |\operatorname{supp} (V)|\ge 2. We distinguish two cases.

CASE 1:  For all A\in \mathcal A(G) with A \, | \, UV we have that |A|\in\{2,3,\mathsf D(G)\}.

Then UV has a factorization of the form UV=W_1 \cdot \ldots \cdot W_{\mathsf D(G)-1} where W_1, \ldots, W_{\mathsf D(G)-1} \in \mathcal A (G), |W_{\mathsf D(G)-2}|=|W_{\mathsf D(G)-1}|=3, and all the other W_i have length 2. Thus we may set

where l \in \mathbb N, g_1, \ldots, g_l, h_1, \ldots, h_6 \in G not necessarily distinct such that h_1=h_2+h_3, h_6=h_4+h_5 and \{h_1,h_4,h_5\}\cap \{h_2,h_3,h_6\}=\emptyset.

Since V(-g_6)^{-1} is a zero-sum free sequence of length \mathsf D(G)-1, there exists a subsequence T of V(-g_6)^{-1} such that \sigma (T)=h_4. Thus Th_5(-h_6) is a zero-sum subsequence of UV. Since T(-h_6) is zero-sum free, we obtain that Th_5(-h_6) is an atom of length |Th_5(-h_6)|\in \{3,\,\mathsf D(G)\}. If |Th_5(-h_6)|=3, then |T|=1 which implies that h_4\, | \, V. If |Th_5(-h_6)|=\mathsf D(G), then Th_5(-h_6)=V which implies that h_5\, | \, V. Therefore, we obtain that h_4\in \operatorname{supp}(U)\cap \operatorname{supp}(V) or h_5\in \operatorname{supp}(U)\cap \operatorname{supp}(V). Without loss of generality, we assume that h_4\in \operatorname{supp}(U)\cap \operatorname{supp}(V).

We start with the following assertions.

1. A1.  \operatorname{ord} (h_4) > 2, (-2h_4)^2\, | \, U, and (-2h_4)^2\, | \, V.

2. A2.  If g_i\not\in \{h_1,\ldots,h_6\} for some i\in [1,l], then \operatorname{ord}(g_i)=2.

Proof of  A1.  Since -g_i\neq h_4 for all i\in [1,l], we obtain that h_4\in \{-h_2,-h_3,-h_6\} which implies that \operatorname{ord}(h_4)> 2.

By the symmetry of U and V, we only need to prove that (-2h_4)^2\, | \, U. For any g\in \operatorname{supp}(U) with g\neq h_4, consider the sequence Ug^{-1}h_4. Since |Ug^{-1}h_4|=\mathsf D(G) and Ug^{-1}h_4\, | \, UV, there exists a atom A with A\, | \, Ug^{-1}h_4 and |A|=3 by Lemma 3.4.2. Note that h_4^2\, | \, A. Therefore, -2h_4\, | \, Ug^{-1}. If \operatorname{ord}(h_4)=3, then h_4=-2h_4 and A=h_4^3 which implies that h_4^2\, | \, U. If \operatorname{ord}(h_4)>3, then -2h_4\neq h_4. Thus we can choose g=-2h_4 which implies that (-2h_4)^2\, | \, U.

Proof of  A2.  Let i \in [1,l] such that g_i\not\in \{h_1,\ldots,h_6\} and assume to the contrary that \operatorname{ord}(g_i)>2. Let v=\mathsf v_{g_i}(U)=\mathsf v_{-g_i}(V). By Lemma 3.4.2, we obtain that there exists an atom W such that W\, | \, g_i^{-v}(-g_i)^vU and |W|=3. By Lemma 3.4.1, \mathsf v_{-g_i}(W)=1. Thus there exist f_1, f_2\in \operatorname{supp}(U) such that g_i=f_1+f_2. Obviously, f_1f_2\nmid g_1\cdot \ldots \cdot g_l. If f_1\, | \, g_1\cdot \ldots \cdot g_l and f_2\, | \, h_1h_4h_5, then f_2=g_i+(-f_1). Since f_2 is an element of an atom of length 3, we can substitute f_2 by g_i(-f_1) to get an atom of length 4, a contradiction. Therefore, f_1f_2\, | \, h_1h_4h_5. If f_1f_2=h_4h_5, then g_i=h_6, a contradiction. By the symmetry of h_4 and h_5, we only need to consider f_1f_2=h_1h_4. Since (-g_i)h_1h_4 is an atom, g_ih_5(-h_2)(-h_3)(-h_6) can only be a product of two atoms which implies that g_i\in \{h_2,h_3,h_6\}, a contradiction.

We continue with the following two subcases.

CASE 1.1:  \operatorname{ord}(h_4)>3.

By A1, we have (-2h_4)^2\, | \, U and (-2h_4)^2\, | \, V. Thus, for any i \in [1,l], g_i \ne -2h_4 and -g_i \ne -2h_4. Therefore, we obtain that h_1=h_5= -2h_4 and -h_6=h_4, and hence -h_2=-h_3=-2h_4. Then -2h_4=h_1=h_2+h_3=4h_4 which implies that \operatorname{ord}(h_4)=6. By A2, we obtain that U=g_1\cdot \ldots \cdot g_l h_4 (-2h_4)^2 with \operatorname{ord}(g_1)=\ldots =\operatorname{ord}(g_l)=2 and \operatorname{ord}(h_4)=6. Since \mathsf D(G)\ge5, we get l\ge 2. Then 6+l-1=\exp(G)+l-1\le \mathsf D(G)=|U|=l+3, a contradiction

CASE 1.2:  \operatorname{ord}(h_4)=3.

By A1, h_4^2\, | \, U and h_4^2\, | \, V. Thus h_4^4\, | \, h_1h_4h_5(-h_2)(-h_3)(-h_6). But h_4^6\neq h_1h_4h_5(-h_2)(-h_3)(-h_6). Without loss of generality, we can assume that h_4=h_5=-h_3=-h_6, h_1\neq h_4, and -h_2\neq h_4.

Suppose \mathsf v_{h_1}(U)>1 and \mathsf v_{-h_2}(V)>1. Then -h_1\, | \, V and h_2\, | \, U. Since h_1+(-h_2)+h_4=0, then (-h_1)h_2h_4^2 is an atom of length 4 and divides UV, a contradiction. Thus, by symmetry, we may assume that \mathsf v_{-h_2}(U)=1. Then h_2\nmid U. Gathering the g_i’s, which are equal to h_1 and renumbering if necessary we obtain that U=g_1\cdot \ldots \cdot g_s h_1^v h_4^2 with s \in [0,l], v=\mathsf v_{h_1}(U), \operatorname{ord}(g_1)=\ldots=\operatorname{ord}(g_s)=2 (by A2), and \operatorname{ord}(h_4)=3.

If s=0, then vh_1=h_4 and G=\langle h_1\rangle, a contradiction.

Suppose s\ge 1. Since 2\sigma (U)=2vh_1+4h_4=0, we obtain that 2h_4=2vh_1 and 6vh_1=0. Since (-h_1)^{v-1}h_4^2 \, | \, V, we obtain that (-h_1)^{v-1}h_4^2 is zero-sum free and, for any j \in [v+1, 2v], we have j h_1 \in \Sigma \big( (-h_1)^{v-1}h_4^2 \big). This implies that \operatorname{ord}(h_1)>2v. Thus 6v/\operatorname{ord}(h_1)<3 which implies that \operatorname{ord}(h_1)=3v or 6v. But \sigma (h_1^vh_4^2)=3vh_1\neq 0, hence \operatorname{ord}(h_1)=6v=\exp(G). Therefore,

a contradiction.

CASE 2:  There exists an A\in \mathcal A(G) with A\, | \, UV and |A|=4.

Then UV has a factorization of the form UV=W_1 \cdot \ldots \cdot W_{\mathsf D(G)-1} where W_1, \ldots, W_{\mathsf D(G)-1} \in \mathcal A (G), |W_{\mathsf D(G)-1}|=4, and all the other W_i have length 2. Conversely, note that for any A'\in \mathcal A(G) with A'\, | \, UV and |A'|=4, UV(A')^{-1} can only be a product of atoms of length 2 because \mathsf D(G)\ge 5. We start with the following assertion.

1. A3.  If g\in \operatorname{supp}(U)\cap \operatorname{supp}(V), then \operatorname{ord}(g)=2.

Proof of  A3.  Let U=g_1\cdot \ldots \cdot g_lh_1h_2 and V=(-g_1)\cdot \ldots \cdot (-g_l) h_3h_4 where l \in \mathbb N, g_1, \ldots, g_l, h_1, \ldots, h_4 \in G not necessarily distinct and W_{\mathsf D(G)-1}=h_1h_2h_3h_4. Let g\in \operatorname{supp}(U)\cap \operatorname{supp}(V). If g = g_i for some i \in [1,l], then g \, | \, V, -g \, | \, V, and hence g=-g. If g = -g_i for some i \in [1,l], then g \, | \, U, -g \, | \, U, and hence g=-g.

It remains to consider the case where g \in \{h_1, h_2\} \cap \{h_3, h_4\}, say h_1=h_3=g. We will show that this is not possible. Assume this holds, choose an h \in \operatorname{supp} (U) \setminus \{g\}, and consider the sequence X = h^{-1}U h_3. Since |X|=\mathsf D(G), there exists an atom A \in \mathcal A (G) such that A\, | \, X and |A| \in [3,4]. Note that h_1h_3\, | \, A.

Suppose that |A|=3. Then there exists h'\in \operatorname{supp} ( U) such that h'=-h_1-h_3=-2g. If h'=h_2, then h_1h_2h_3 ia a proper zero-sum sequence of W_{\mathsf D(G)-1}, a contradiction. Otherwise, h' \in \{g_1, \ldots, g_l\}, hence -h'=2g \in \operatorname{supp}( V), a contradiction to h_3=g \in \operatorname{supp} ( V) (recall Lemma 3.4).

Suppose that |A|=4. If h_2 \, | \, A, then (note that h_1+h_2+h_3+h_4=0) A=h_1h_2h_3h_4 and -h_4 \, | \, V, a contradiction. Thus h_2 \nmid A, and the similar argument shows that h_4\nmid A. This implies that A=g_ig_jh_1h_3 with i,j \in [1,l]. Then A and A'=(-g_i)(-g_j)h_2h_4 are two atoms of length 4 dividing UV, a contradiction.

Clearly, there are precisely two possibilities for U and V which will be discussed in the following two subcases.

CASE 2.1:  U=g_1^{k_1}\cdot \ldots \cdot g_l^{k_l}h_1^{r_1-1}h_2^{r_2-1}h_3^{r_3-1}h_4^{r_4-1}h_1h_2 \quad \text{ and}

k_1, \ldots, k_l,r_1,\ldots, r_4 \in \mathbb N, g_1, \ldots , g_l, h_1,\ldots,h_4 are pairwise distinct, and h_1h_2(-h_3)(-h_4) = W_{\mathsf D(G)-1}.

We start with the following assertions.

1. A4.  For each g_i \in \{g_1,\ldots, g_l\} with \operatorname{ord} (g_i)>2, we have that g_i=h_1+h_2.

2. A5.  r_1=r_2=r_3=r_4=1.

3. A6.  \operatorname{ord}(g_i)=2 for all i \in [1,l].

Proof of  A4.  Let i \in [1,l] such that \operatorname{ord}(g_i)>2, say i=1. Then, by Lemma 3.4.2, there exists an atom A \in \mathcal A (G) such that A\, | \, (-g_1)^{k_1}g_1^{-k_1}U and |A| \in [3,4]. We distinguish four subcases depending on the multiplicity of \mathsf v_{-g_1} (A) and on |A|.

Suppose that \mathsf v_{-g_1}(A)=3. Then |A|=4. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. Then A=(-g_1)^3h_i with i\in [1,2], say i=1. It follows that A'=g_1^3h_2(-h_3)(-h_4) is a zero-sum subsequence of UV(A)^{-1} which implies that A' is a product of atoms of length 2, a contradiction.

Suppose that \mathsf v_{-g_1}(A)=2. Then |A|=4 by Lemma 3.4.1. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. If h_1h_2\, | \, A, then A=(-g_1)^2h_1h_2, and hence A'=g_1^2(-h_3)(-h_4) is an atom of length 4 and divides UVA^{-1}, a contradiction. By symmetry, we may assume that h_1\, | \, A and h_2\nmid A. Thus A = (-g_1)^2h_1h, where h \in \{g_2,\ldots,g_l,h_1,h_3,h_4\} and (-h)(-h_3)(-h_4)\, | \, V. Therefore A'=g_1^2 h_2(-h_3)(-h_4)(-h) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of three atoms of length 2, a contradiction.

Suppose that \mathsf v_{-g_1}(A)=1 and |A|=4. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. If h_1h_2\, | \, A, then A=(-g_1)h_1h_2h, where h \in \{g_2,\ldots,g_l,h_1,h_2,h_3,h_4\} and (-h)(-h_3)(-h_4)\, | \, V. Hence A'=g_1(-h_3)(-h_4)(-h) is an atom of length 4 and divides UVA^{-1}, a contradiction. By symmetry, we may assume that h_1\, | \, A and h_2\nmid A. Thus A=(-g_1)h_1hh' where h, h' \in \{g_2,\ldots,g_l,h_1,h_3,h_4\} and (-h)(-h')(-h_3)(-h_4)\, | \, V. Thus A'=g_1(-h)(-h')h_2(-h_3)(-h_4) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of three atoms of length 2, a contradiction.

Suppose that \mathsf v_{-g_1}(A)=1 and |A|=3. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. If h_1h_2\nmid A, by symmetry we may assume that h_1\, | \, A and h_2\nmid A. Thus A=(-g_1)h_1h, where h\in \{g_2,\ldots,g_l,h_1,h_3,h_4\} and (-h)(-h_3)(-h_4)\, | \, V. It follows that A'= g_1(-h)h_2(-h_3)(-h_4) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of two atoms, a contradiction. Therefore, h_1h_2\, | \, A and g_1=h_1+h_2.

Proof of  A5.  By symmetry it is sufficient to show that r_3=1. Assume to the contrary that r_3\ge2. We proceed in several steps.

(i) In the first step we will show that h_3=g_i+h_1 for some i \in [1,l].

By Lemma 3.4.2, there exists an atom A \in \mathcal A (G) such that A\, | \, h_3^{-(r_3-1)}(-h_3)^{r_3-1}U and |A|\in[3,4]. We distinguish four subcases depending on the multiplicity of \mathsf v_{-g_1} (A) and on |A|.

Suppose that \mathsf v_{-h_3}(A)=3. Then |A|=4. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. Then A=(-h_3)^3h_i with i\in [1,2], say i=1. Therefore A'=h_3^2h_2(-h_4) is an atom of length 4 and divides UVA^{-1}, a contradiction.

Suppose that \mathsf v_{-h_3}(A)=2. Then |A|=4 by Lemma 3.4.1. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. If h_1h_2\, | \, A, then A=(-h_3)^2h_1h_2 which implies that h_3=h_4, a contradiction. By symmetry, we may assume that h_1\, | \, A and h_2\nmid A. Thus A = (-h_3)^2h_1h, where h \in \{g_1,\ldots,g_l,h_1,h_4\} and (-h)(-h_3)(-h_4)\, | \, V. Therefore A'=h_3^2 h_2(-h_3)(-h_4)(-h) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of three atoms of length 2, a contradiction.

Suppose that \mathsf v_{-h_3}(A)=1 and |A|=4. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. If h_1h_2\, | \, A, then A=(-h_3)h_1h_2h with h\in \{g_1,\ldots,g_l,h_1,h_2,h_4\} and -h\, | \, V. Hence h=-h_4 and -h=h_4\, | \, V, a contradiction. By symmetry, we may assume that h_1\, | \, A and h_2\nmid A. Thus A=(-h_3)h_1hh' where h, h' \in \{g_1,\ldots,g_l,h_1,h_4\} and (-h)(-h')(-h_3)(-h_4)\, | \, V. Thus A'=(-h)(-h')h_2(-h_4) is an atom of length 4 and divides UVA^{-1}, a contradiction

Suppose that \mathsf v_{-h_3}(A)=1 and |A|=3. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h_1\, | \, A or h_2\, | \, A. If h_1h_2\, | \, A, then A=(-h_3)h_1h_2 which implies that h_4=0, a contradiction. Thus, by symmetry, we may assume that h_1\, | \, A and h_2\nmid A. Then A=(-h_3)h_1h, where h\in \{g_1,\ldots,g_l,h_1,h_4\}. By Lemma 3.4.1, we obtain that h\not\in \{h_1,h_4\}. Therefore, h_3=h_1+g_i for some i\in [1,l].

(ii) In the second step we will show that \operatorname{ord} (g_j)=2 for all j \in [1,l] and r_1=1.

Assume to the contrary that there is a j \in [1,l] such that \operatorname{ord}(g_j)>2. Then g_j=h_1+h_2 by A4 and hence A_1=g_j(-h_3)(-h_4) and A_2=g_i(-h_3)h_1 are two atoms of length 3 and divide UV. It follows that UV(A_1A_2)^{-1} is a product of atoms of length 2, but \mathsf v_{h_2}(UV(A_1A_2)^{-1})=\mathsf v_{h_2}(UV)>\mathsf v_{-h_2}(UV)=\mathsf v_{-h_2}(UV(A_1A_2)^{-1}), a contradiction.

Assume to the contrary that r_1\ge 2. Since \operatorname{ord}(g_i)=2 and h_3=h_1+g_i, we obtain that h_1^2(-h_3)^2 and (-h_1)h_3g_i are two atoms and divide UV, a contradiction.

(iii) In the third step we show that \operatorname{ord} (h_3)=4r_3.

Consider the sequence X=U(h_3^{r_3-1}h_1)^{-1}(-h_3)^{r_3}. Since |X|=\mathsf D(G), there exists an atom A\in\mathcal A(G) with A\, | \, X and |A|\in \{3,4,\mathsf D(G)\}. We distinguish three subcases depending on |A|.

Suppose that |A|=3. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1} and r_1=1, we must have that h_2\, | \, A. Thus A=(-h_3)h_2h, where h\in \{g_1,\ldots,g_l,h_2,-h_3\}. By Lemma 3.4.1, h\not\in \{h_2,-h_3\}. Therefore, A and A'=(-h_3)h_1g_i are two atoms of length 3 and divide UV. It follows that UV(AA')^{-1} is a product of atoms of length 2 but \mathsf v_{h_4}(UV(AA')^{-1})<\mathsf v_{-h_4}(UV(AA')^{-1}), a contradiction.

Suppose that |A|=4. Then UVA^{-1} is a product of atoms of length 2, but r_1=1=\mathsf v_{h_1}(UVA^{-1})>\mathsf v_{-h_1}(UVA^{-1})=0, a contradiction.

Suppose that |A|=\mathsf D(G). Then A=X and hence h_1=-(2r_3-1)h_3. By steps (i) and (ii), we obtain that g_i=2r_3h_3 and 4r_3h_3=0. Since Uh_1^{-1} is zero-sum free and for each j\in [1,2r_3-1], jh_3\in \Sigma (Uh_1^{-1}), then \operatorname{ord}(h_3)>2r_3 which implies that \operatorname{ord}(h_3)=4r_3.

(iv) In the final step we will obtain a contradiction to our assumption that r_3 \ge 2. Clearly, similar arguments as given in the steps (i),(ii), and (iii) show that r_2\ge 2 implies that \operatorname{ord}(h_2)=4r_2, and that r_4\ge 2 implies that \operatorname{ord}(h_4)=4r_4. We proceed with the following four subcases depending on r_2 and r_4.

Suppose that r_2\ge 2 and r_4\ge 2. Since h_3=h_1+g_i and \operatorname{ord}(g_i)=2, we obtain that h_2=h_4+g_i and hence 2h_2=2h_4. Therefore, (-h_2)h_4g_i and h_2^2(-h_4)^2 are two atoms and divide UV, a contradiction.

Suppose that r_2=r_4=1. Then U=g_1\cdot \ldots \cdot g_lh_3^{r_3-1}h_1h_2 and

a contradiction.

Suppose that r_2=1 and r_4\ge2. Then U=g_1\cdot \ldots \cdot g_lh_3^{r_3-1}h_4^{r_4-1}h_1h_2 and

a contradiction.

Suppose that r_4=1 and r_2\ge2. Then U=g_1\cdot \ldots \cdot g_lh_3^{r_3-1}h_2^{r_2-1}h_1h_2 and

a contradiction.

Proof of  A6.  Assume the contrary that there is an i \in [1,l] such that \operatorname{ord}(g_i)>2, say i=1. Then A4 implies that g_1=h_1+h_2. Since g_1, \ldots, g_l are pairwise distinct, it follows that \operatorname{ord} (g_2)= \ldots = \operatorname{ord} (g_l)=2 and k_2= \ldots = k_l=1. Thus U = g_1^{k_1}g_2 \cdot \ldots \cdot g_lh_1h_2, 0=\sigma (U)=k_1g_1+g_2+\ldots + g_l+g_1, whence \operatorname{ord} ( (k_1+1)g_1)=2 and \operatorname{ord}(g_1)=2(k_1+1)\ge 4. It follows that

a contradiction.

Now by A4, A5, and A6, U has the form U=g_1\cdot \ldots \cdot g_l h_1h_2 with \operatorname{ord}(g_i)=2 for each i \in [1,l]. If \exp(G)\ge 4, then

a contradiction. Thus G must be an elementary 2-group. Since (g_1,\ldots,g_l,h_1) is a basis of G, then h_2=g_1+\ldots+g_l+h_1 and h_1+h_2=h_3+h_4=g_1+\ldots +g_l. We can assume that h_3=h_1+\sum_{i\in I}g_i for some \emptyset \ne I\subsetneq [1,l] and h_4= h_1+\sum_{i\in [1,l]\setminus I}g_i. Then A_1=h_3h_1\prod_{i\in I}{g_i} and A_2=h_1h_4\prod_{i\in [1,l]\setminus I}g_i are two atoms of lengths |A_1|, |A_2| \in [3, \mathsf D (G)-1]. If i\in [1,2] and |A_i|\ge4, then UVA_i^{-1} has to be a product of atoms of length 2, a contradiction. Thus it follows that |A_1|=|A_2|=3 whence l=2 which implies that \mathsf D (G)=4, a contradiction.

CASE 2.2:  U=g_1^{k_1}\cdot \ldots \cdot g_l^{k_l}h^{r-2}h_3^{r_3-1}h_4^{r_4-1}h^2 \ \text{ and}

k_1, \ldots, k_l,r_1,\ldots, r_4 \in \mathbb N, g_1, \ldots , g_l, h, h_3, h_4 are pairwise distinct with the only possible exception that h_3=h_4 may hold, and h^2(-h_3)(-h_4) = W_{\mathsf D(G)-1}.

We start with the following assertions.

1. A7.  For each i \in [1,l] we have \operatorname{ord}(g_i)=2.

2. A8.  h_3=h_4.

Proof of  A7.  Assume to the contrary that there is an i \in [1,l], say i=1, such that \operatorname{ord} (g_1)>2.

Then, by Lemma 3.4.2, there exists an atom A \in \mathcal A (G) such that A\, | \, (-g_1)^{k_1}g_1^{-k_1}U and |A| \in [3,4]. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h\, | \, A. We distinguish four subcases depending on the multiplicity of \mathsf v_{-g_1} (A) and on |A|.

Suppose that \mathsf v_{-g_1}(A)=3. Then |A|=4 and A=(-g_1)^3h . It follows that A'=g_1^3h(-h_3)(-h_4) is a zero-sum subsequence of UV(A)^{-1} which implies that A' is a product of atoms of length 2, a contradiction.

Suppose that \mathsf v_{-g_1}(A)=2. Then |A|=4 by Lemma 3.4.1. If h^2\, | \, A, then A=(-g_1)^2h^2, and hence A'=g_1^2(-h_3)(-h_4) is an atom of length 4 and divides UVA^{-1}, a contradiction. If h^2\nmid A, we obtain that A = (-g_1)^2hf, where f \in \{g_2,\ldots,g_l,h_3,h_4\} and (-f)(-h_3)(-h_4)\, | \, V. Therefore A'=g_1^2 h(-h_3)(-h_4)(-f) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of three atoms of length 2, a contradiction.

Suppose that \mathsf v_{-g_1}(A)=1 and |A|=4. If h^2\, | \, A, then A=(-g_1)h^2f, where f \in \{g_2,\ldots,g_l,h,h_3,h_4\} and (-f)(-h_3)(-h_4)\, | \, V. Hence A'=g_1(-h_3)(-h_4)(-f) is an atom of length 4 and divides UVA^{-1}, a contradiction. If h^2\nmid A, then A=(-g_1)hff' where f, f' \in \{g_2,\ldots,g_l,h_3,h_4\} and (-f)(-f')(-h_3)(-h_4)\, | \, V. Thus A'=g_1(-f)(-f')h(-h_3)(-h_4) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of three atoms of length 2, a contradiction.

Suppose that \mathsf v_{-g_1}(A)=1 and |A|=3. If h^2\, | \, A, then g_1=2h, a contradiction to Lemma 3.4.1. If h^2\nmid A, then A=(-g_1)hf, where f\in \{g_2,\ldots,g_l,h_3,h_4\} and (-f)(-h_3)(-h_4)\, | \, V. It follows that A'= g_1(-f)h(-h_3)(-h_4) is a zero-sum subsequence of UVA^{-1} which implies that A' is a product of two atoms, a contradiction.

Proof of  A8.  Assume to the contrary that h_3\neq h_4. If r_3=r_4=1, then U=g_1\cdot \ldots \cdot g_l h^r with l\ge 1, \operatorname{ord}(h)=2r\ge 4, and hence

a contradiction. Thus after renumbering if necessary we may assume that r_3\ge 2. We will show this is impossible.

By Lemma 3.4, there exists an atom A \in \mathcal A (G) such that A\, | \, h_3^{-(r_3-1)}(-h_3)^{r_3-1}U and |A|\in[3,4]. Since A\nmid UV(W_{\mathsf D(G)-1})^{-1}, we must have that h\, | \, A. We distinguish four subcases depending on the multiplicity of \mathsf v_{-h_3} (A) and on |A|.

Suppose that \mathsf v_{-h_3}(A)=3. Then |A|=4 and A=(-h_3)^3h . It follows that A'=h_3^2h(-h_4) is an atom and divides UV(A)^{-1}, a contradiction.

Suppose that \mathsf v_{-h_3}(A)=2. Then |A|=4 by Lemma 3.4.1. If h^2\, | \, A, then A=(-h_3)^2h^2 which implies that h_3=h_4, a contradiction. If h^2\nmid A, we obtain that A = (-h_3)^2hf, where f \in \{g_1,\ldots,g_l,h_4\} and (-f)(-h_3)(-h_4)\, | \, V. Therefore A'=h_3 h(-h_4)(-f) is an atom and divides UVA^{-1}, a contradiction.

Suppose that \mathsf v_{-h_3}(A)=1 and |A|=4. If h^2\, | \, A, then A=(-h_3)h^2f, where f \in \{g_1,\ldots,g_l,h,h_4\} and (-f)(-h_3)(-h_4)\, | \, V. Hence h_4=2h-h_3=-f\, | \, V, a contradiction. If h^2\nmid A, then A=(-h_3)hff' where f, f' \in \{g_1,\ldots,g_l,h_4\} and (-f)(-f')(-h_3)(-h_4)\, | \, V. Thus A'=(-f)(-f')h(-h_4) is an atom and divides UVA^{-1}, a contradiction.

Suppose that \mathsf v_{-h_3}(A)=1 and |A|=3. Since h^2(-h_3)(-h_4) is an atom, we obtain that h^2\nmid A, and hence A=(-h_3)hf, where f\in \{g_1,\ldots,g_l,h_4\} and (-f)(-h_3)(-h_4)\, | \, V. If f=h_4, then h=2h_4 which implies a contradiction to Lemma 3.4.1(recall f=h_4\, | \, U and h=2h_4\, | \, U). If f\neq h_4, by A7 \operatorname{ord}(f)=2 and hence 2h=2h_3 which implies that h_3=h_4, a contradiction.

Now, by A7 and A8, U and V have the form

where l\ge 0, r_3\ge 2, r\ge 2, g_1, \ldots , g_l, h_3,h\in G are pairwise distinct and 2h=2h_3.

If r_3=2, then U=g_1\cdot \ldots \cdot g_l h^r, \operatorname{ord}(h)=2r, and

a contradiction. Considering V and assuming r=2 we end again up at a contradiction. Therefore we obtain that r_3\ge 3 and r \ge 3.

If r_3\ge 4 and r\ge 4, then h^2(-h_3)^2 and (-h)^2h_3^2 are two atoms of length 4 and divide UV, a contradiction.

Thus by symmetry, we may assume that r_3=3 and r\ge 3.

Suppose that l=0. Then \sigma (U)=h_3+rh=0 which implies that G=\langle h \rangle, a contradiction.

Suppose that l\ge 1. Then 2h_3+2rh=0 which implies that 2(r+1)h=0. Thus \operatorname{ord}(h)=2(r+1) or \operatorname{ord}(h)=r+1. If \operatorname{ord}(h)=2(r+1), then

a contradiction. If \operatorname{ord}(h)=r+1, then h=h_3+g_1+\ldots+g_l and hence (-h)h_3g_1\cdot\ldots\cdot g_l is an atom and divides UV(h^2(-h_3)^2)^{-1}, a contradiction

(a)  \Rightarrow  (b) Suppose that G is an elementary 2-group and let (e_1, \ldots, e_r) be a basis of G with \operatorname{ord} (e_1)=\ldots = \operatorname{ord} (e_r)=2. Then \mathsf D (G)=r+1, U= e_1 \cdot \ldots \cdot e_{r}e_0 \in \mathcal A (G), where e_0=e_1+ \ldots + e_{r}, V=e_1 \cdot \ldots \cdot e_{r-1}(e_0+e_r) \in \mathcal A (G), and \mathsf L \big( UV \big) = \{2,r\}.

Suppose that G is cyclic, and let e \in G with \operatorname{ord} (e)=|G|=\mathsf D (G). Then U=e^{|G|} \in \mathcal A (G), V=(-e)^{|G|-1}(-2e) \in \mathcal A (G), and \mathsf L \big( UV \big) = \{2, |G|-1\}.

Suppose that G is isomorphic to C_2 \oplus C_{2n} with n \ge 2, and let (e_1, e_2) be a basis of G with \operatorname{ord} (e_1)=2 and \operatorname{ord}(e_2)=2n. Then \mathsf D (G)=2n+1, U = e_1e_2^{2n-1}(e_1+e_2) \in \mathcal A (G), V = (-e_2)^{2n} \in \mathcal A (G), and \mathsf L(UV)=\{2, 2n \}.

(b)  \Rightarrow  (a) Assume to the contrary that G is neither an elementary 2-group, nor a cyclic group, nor isomorphic to C_2 \oplus C_{2n} for any n \ge 2. Then \mathsf D(G)>\exp(G)+1.

Let A\in \mathcal A(G) with A\, | \, UV. Then we have |A|\in\{2,3,\mathsf D(G)-1,\mathsf D(G)\}. Furthermore, if |A|=3, then UVA^{-1} is a product of atoms of length 2, and if |A| \in [ \mathsf D(G)-1, \mathsf D(G)], then UVA^{-1} \in \mathcal A (G).

Since \mathsf L (UV)= \{2, \mathsf D (G)-1\} and |U|-1=|V|=\mathsf D (G)-1, U and V have the form

where s, k_1, \ldots, k_s \in \mathbb N, g_1,\ldots, g_s \in G are pairwise distinct with the only possible exception that g_1=g_2 may hold.

Since G is not cyclic, we have s \ge 2. Suppose that s=2. If g_1=g_2, then \mathsf D(G)=k_1+k_2=\operatorname{ord}(g_1) which implies that G is a cyclic group, a contradiction. Suppose that g_1\neq g_2. Since \mathsf D(G)>\exp(G)+1, Lemma 3.2.2 implies that \mathsf v_{g_1}(U)+\mathsf v_{-g_1}(V)=k_1+k_1-1\le \operatorname{ord}(g_1) and \mathsf v_{g_2}(U)+\mathsf v_{-g_2}(V)=k_2+k_2-1\le \operatorname{ord}(g_2). Then

a contradiction.

Thus from now on we suppose that s \ge 3, and we continue with the following assertion.

1. A1.  There exist atoms U', V' \in \mathcal A (G) such that UV=U'V', say

   U'=g_1'^{k_1'} \cdot \ldots \cdot g_{s'}'^{k_{s'}'} \quad \text{and} \quad V'=(-g_1'-g_2')(-g_1')^{k_1'-1}(-g_2')^{k_2'-1}(-g_3')^{k_3'}\cdot \ldots \cdot (-g_{s'}')^{k_{s'}'} \,,

   where s', k_1', \ldots, k_{s'}' \in \mathbb N, g_1',\ldots, g_{s'}' \in G are pairwise distinct with the only possible exception that g_1'=g_2' may hold, and g_1'+g_2'\not\in \operatorname{supp}(U').

Proof of  A1.  If g_1+g_2\not\in \operatorname{supp}(U), then we can choose U'=U and V'=V.

Suppose g_1+g_2\in \operatorname{supp}(U), say g_3=g_1+g_2. Then \mathsf v_{-g_3}(V)=k_3+1 \ge 2 and hence \operatorname{ord}(g_3)>2. By Lemma 3.4.1 and |U|=\mathsf D(G), it follows that g_1\neq g_2. We claim that k_1=1 or k_2=1. Indeed, if k_1\ge 2 and k_2\ge 2, then A=g_1g_2(-g_3) is an atom and A^2\, | \, UV, a contradiction.

Suppose s=3. Without loss of generality, we can assume that k_1=1. Since \mathsf D(G)>\exp(G)+1, Lemma 3.2.2 implies that \mathsf v_{g_2}(U)+\mathsf v_{-g_2}(V)=k_2+k_2-1\le \operatorname{ord}(g_2) and \mathsf v_{g_3}(U)+\mathsf v_{-g_3}(V)=k_3+k_3+1\le \operatorname{ord}(g_3). Then

a contradiction.

Thus we obtain that s\ge4. Since g_3^{-1}U is a zero-sum free sequence of length \mathsf D(G)-1, there exists a subsequence T_1 of g_3^{-1}U such that \sigma(T_1)=(k_3+1)g_3. Therefore, (-g_3)^{k_3+1}T_1 is a zero-sum sequence. Thus T_1 has the form T_1=g_3^tT_2 with t \in [0, k_3-1] and T_2\, | \, g_1^{k_1}g_2^{k_2}g_4^{k_4}\cdot \ldots \cdot g_s^{k_s}. Then (-g_3)^{k_3+1-t}T_2 is a zero-sum sequence without zero-sum subsequences of length 2 which implies that (-g_3)^{k_3+1-t}T_2 is an atom of length |(-g_3)^{k_3+1-t}T_2|\in\{3,\mathsf D(G)-1,\mathsf D(G)\}. Since k_3+1-t\ge2, (-g_3)^{k_3+1-t}T_2 cannot be an atom of length 3 by Lemma 3.4.1, hence t=0, T_1=T_2, and (-g_3)^{k_3+1}T_1 can only be an atom of length \mathsf D(G)-1 or \mathsf D(G). It follows that ((-g_3)^{k_3+1}T_1)^{-1}UV is also an atom and hence g_4^{k_4}\cdot \ldots \cdot g_s^{k_s}\, | \, T_1.

Thus any sequence T with T \, | \, g_3^{-1}U and \sigma (T)=(k_3+1)g_3 has the property that g_4^{k_4}\cdot \ldots \cdot g_s^{k_s}\, | \, T.

We continue with the following two subcases depending on |(-g_3)^{k_3+1}T_1|.

Suppose |(-g_3)^{k_3+1}T_1|=\mathsf D(G)-1. Since ((-g_3)^{k_3+1}T_1)^{-1}UV is an atom, we obtain that k_1=k_2=1 and T_1=(g_1g_2g_3^{k_3})^{-1}U which implies that 2(k_3+1)g_3=0. Since g_4^{-1}U is a zero-sum free sequence of length \mathsf D(G)-1, there exists a subsequence W_1 of g_4^{-1}U such that \sigma(W_1)=(k_3+2)g_3. If g_3\nmid W_1, then g_3^{k_3}W_1 is a proper zero-sum subsequence of U, a contradiction. If g_3\, | \, W_1, then g_3^{-1}W_1\, | \, g_3^{-1}U and \sigma (g_3^{-1}W_1)=(k_3+1)g_3, but g_4^{k_4}\cdot \ldots \cdot g_s^{k_s}\nmid g_3^{-1}W_1, a contradiction.

Suppose |(-g_3)^{k_3+1}T_1|=\mathsf D(G). Since ((-g_3)^{k_3+1}T_1)^{-1}UV is an atom, we obtain that \big(k_1=1 and T_1=(g_1g_3^{k_3})^{-1}U\big) or \big(k_2=1 and T_1=(g_2g_3^{k_3})^{-1}U\big). By symmetry, we may assume that k_1=1, T_1=(g_1g_3^{k_3})^{-1}U, and hence g_1=(-2k_3-1)g_3. Choose

then U',V' are two atoms with U'V'=UV and g_2+(-g_3)=-g_1\not\in \operatorname{supp} (U').

Thus from now on we may assume that g_1+g_2\not\in \operatorname{supp}(U), and recall that s \ge 3. We continue with three further assertions.

1. A2.  (\operatorname{supp}(U)+\operatorname{supp}(U))\cap (\operatorname{supp}(U)\setminus \{g_1,g_2\})=\emptyset.

2. A3.  Let i\in [3,s] with \operatorname{ord}(g_i)>2. Then \big( -2k_ig_i=g_1 and k_1=1 \big) or \big( -2k_ig_i=g_2 and k_2=1 \big).

3. A4.  If k_1=1, then \big(g_2=-2g_1 and k_2=1 \big) or \big(2g_1+2g_2=0 and k_2=1 \big). If k_2=1, then \big(g_1=-2g_2 and k_1=1 \big) or \big(2g_1+2g_2=0 and k_1=1\big).

Proof of  A2.  Assume to the contrary that there exists an element h\in (\operatorname{supp}(U)+\operatorname{supp}(U))\cap (\operatorname{supp}(U)\setminus \{g_1,g_2\}). Thus there exist i, j \in [1, s] such that g_i+g_j = h with h \in \{g_3, \ldots, g_s\}. Lemma 3.4.1 implies that g_i \ne g_j. Since A = (-h)g_ig_j is an atom of length 3, then A^{-1}UV is a product of atoms of length 2. It follows that h=g_1+g_2\in \operatorname{supp}(U), a contradiction.

Proof of  A3.  By Lemma 3.4.2, there is an A \in \mathcal A (G) with A \, | \, g_i^{-k_i}(-g_i)^{k_i}U and |A| \in \{3,\mathsf D(G)-1\}. By A2 and Lemma 3.4.1, we obtain that |A| \ne 3. Thus |A|=\mathsf D(G). If A=g_i^{-k_i}(-g_i)^{k_i-1}U, then (2k_i-1)g_i=0 which implies that \mathsf v_{g_i}(U)+\mathsf v_{-g_i}(V)=2k_i>\operatorname{ord}(g_i). Lemma 3.2.2 implies a contradiction to \mathsf D(G)>\exp(G)+1. Hence A=(-g_i)^{k_i}T with T=g_i^{-k_i}h^{-1}U, where h\in\operatorname{supp}(U). Since T^{-1}UV is an atom, we obtain that \big(k_1=1 and h=g_1=-2k_ig_i \big) or \big(k_2=1 and h=g_2=-2k_ig_i \big).

Proof of  A4.  Suppose that k_1=1. Then Y=(-g_1-g_2)g_2^{k_2}\cdot \ldots \cdot g_s^{k_s} has length \mathsf D(G). By our assumption on \operatorname{supp} (U), Y has no zero-sum subsequence of length 2. Therefore, Y has a zero-sum subsequence T of length |T|\in \{3, \mathsf D(G)-1,\mathsf D(G)\}. We continue with the following three subcases depending on |T|. If |T|=3, say T = (-g_1-g_2)g_ig_j with i,j \in [2,s] not necessarily distinct, then T and T' = g_1g_2(-g_i)(-g_j) are two atoms and T'T \, | \, UV, a contradiction. If |T|=\mathsf D(G), then Y=T is an atom which implies that g_2=-2g_1, -g_1-g_2=g_1, and hence k_1=1. If |T|=\mathsf D(G)-1, then k_2=1, T=Yg_2^{-1}, and hence 2g_1+2g_2=0.

If k_2=1, then the assertion follows along the same lines.

The remainder of the proof will be divided into the following three cases.

CASE 1:  |\{i \in [3,s] \mid \operatorname{ord}(g_i)>2\}|\ge 2, say \operatorname{ord}(g_3)>2 and \operatorname{ord}(g_4)>2.

By A3, we can assume that \big(k_1=1 and g_1=-2k_3g_3=-2k_4g_4\big) or \big(g_1=-2k_3g_3, g_2=-2k_4g_4 and k_1=k_2=1\big).

Consider the sequence W=g_1^{k_1}g_2^{k_2}(-g_3)^{k_3}(-g_4)^{k_4}g_5^{k_5}\cdot \ldots \cdot g_s^{k_s}. Since |W|=\mathsf D(G), there exists an atom Z \in \mathcal A (G) such that Z \, | \, W and |Z| \in \{3, \mathsf D(G)-1, \mathsf D(G)\}. We distinguish three subcases depending on |Z|.

Suppose |Z|=3. By A2 and Lemma 3.4.1, g_1=g_3+g_4 or g_2=g_3+g_4. If g_1=g_3+g_4, then (-g_1-g_2)g_3g_4g_2 and g_1(-g_3)(-g_4) are two atoms and divide UV, a contradiction. If g_2=g_3+g_4, then (-g_1-g_2)g_1g_3g_4 and g_2(-g_3)(-g_4) are two atoms and divide UV, a contradiction.

Suppose |Z|=\mathsf D(G)-1. Since UVZ^{-1} is a atom, we obtain that Z=Wg_1^{-1} or Z=Wg_2^{-1}. Therefore, -2k_3g_3-2k_4g_4=g_1 or g_2. Our assumption infers that -2k_3g_3-2k_4g_4=g_2 and g_1=-2k_3g_3=-2k_4g_4 which implies that 2g_1=g_2, a contradiction to Lemma 3.4.1.

Suppose |Z|=\mathsf D(G). Then we obtain that 2k_3g_3+2k_4g_4=0. Our assumption infers that k_1=1, g_1=-2k_3g_3=-2k_4g_4, and hence \operatorname{ord}(g_1)=2. Therefore, 4k_3g_3=0, g_1=2k_3g_3, and hence k_3\ge 2 by Lemma 3.4.1. Since g_1^{-1}U is a zero-sum sequence of length \mathsf D(G)-1, there exists a subsequence W of g_1^{-1}U such that \sigma(W)=(2k_3+1)g_3. If g_3\, | \, W, then g_1g_3^{-1}W is a proper zero-sum subsequence of U, a contradiction. Suppose g_3\nmid W. Then g_1(-g_3)W is an atom and divides UV which implies that |g_1(-g_3)W|\in \{3,\mathsf D(G)-1, \mathsf D(G)\}. Since g_3(-g_3)\, | \, UV(g_1(-g_3)W)^{-1}, then UV(g_1(-g_3)W)^{-1} is not an atom. Thus |g_1(-g_3)W|=3 which implies that g_3\in \operatorname{supp} (U)\cap\operatorname{supp} (V), a contradiction to A2.

CASE 2:  |\{i \in [3,s] \mid \operatorname{ord}(g_i)>2\}|=1, say \operatorname{ord}(g_3)>2.

By A3, we may assume that k_1=1 and g_1=-2k_3g_3. By A4, we obtain that \big(g_2=-2g_1 and k_2=1 \big) or \big(2g_1+2g_2=0 and k_2=1 \big). We continue with the following two subcases.

Suppose that 2g_1+2g_2=0 and k_2=1. Since \sigma(U)=g_1+g_2+k_3g_3+g_4+\ldots +g_s=0, we obtain that 2k_3g_3=0=-g_1, a contradiction.

Suppose that g_2=-2g_1=4k_3g_3 and k_2=1. If s=3, then G=\langle g_3 \rangle is cyclic, a contradiction. Hence s\ge 4 and G=\langle g_1,\ldots ,g_{s-1}\rangle=\langle g_3,\ldots ,g_{s-1}\rangle which implies that \mathsf r(G)=s-3 and \exp(G)=\operatorname{ord}(g_3) is even. Since \mathsf D(G)>\exp(G)+1, by Lemma 3.2.2, we infer that \mathsf v_{g_3}(U)+\mathsf v_{-g_3}(V)=2k_3\le \operatorname{ord}(g_3). Therefore, g_1=-2k_3g_3\neq 0 infers that \operatorname{ord}(g_3)\ge 2k_3+2\ge4. Thus

which implies that \operatorname{ord}(g_3)=4, a contradiction to g_2=4k_3g_3\neq 0.

CASE 3:  |\{i \in [3,s] \mid \operatorname{ord}(g_i)>2\}|=0.

Since \sigma (U)=k_1g_1+k_2g_2+g_3+\ldots+g_s=0, we obtain that 2k_1g_1+2k_2g_2=0.

Suppose that g_1=g_2. Then \operatorname{ord}(g_1)=2(k_1+k_2)\ge 4. It follows that

a contradiction.

Thus g_1\neq g_2. Consider the sequence S=(-g_1-g_2)(-g_1)^{k_1-1}g_2^{k_2}g_3\cdot \ldots \cdot g_s. Since |S|=\mathsf D(G), there exists an atom Z \in \mathcal A (G) such that Z \, | \, S and |Z| \in \{3, \mathsf D(G)-1, \mathsf D(G)\}. We distinguish three subcases depending on |Z|.

Suppose that |Z|=\mathsf D(G). Then Z=S and g_2=-2k_1g_1=2k_2g_2. Hence (2k_2-1)g_2=0 and \operatorname{ord}(g_2)=2k_2-1. If s=3, then G=\langle g_1 \rangle is cyclic, a contradiction. Hence s\ge 4 and G=\langle g_1,\ldots ,g_{s-1}\rangle=\langle g_1,g_3,\ldots ,g_{s-1}\rangle which implies that \mathsf r(G)= s-2 and \operatorname{ord}(g_1)=\exp(G) is even. Then \operatorname{ord}(g_1)\ge 2\operatorname{ord}(g_2)\ge 6. Since \mathsf D(G)>\exp(G)+1, by Lemma 3.2.2, we infer that \mathsf v_{g_1}(U)+\mathsf v_{-g_1}(V)=2k_1-1\le \operatorname{ord}(g_1) which implies that 2k_1\le \operatorname{ord}(g_1). Since g_2=-2k_1g_1\ne 0, we obtain that 2k_1\le \operatorname{ord}(g_1)-2. Thus

a contradiction.

Suppose that |Z|=\mathsf D(G)-1. Then there exists an element h\, | \, S such that Z=h^{-1}S. Since Z^{-1}UV is an atom, we obtain that \big(h=-g_1-g_2\big) or \big(h=g_2 and k_2=1\big). If h=-g_1-g_2, then Z^{-1}UV=(-g_1-g_2)g_1^{k_1}(-g_2)^{k_2-1}g_3\cdot \ldots \cdot g_s is an atom of length \mathsf D(G) and hence the similar argument of the previous subcase |Z|=\mathsf D(G) implies a contradiction. Suppose that h=g_2 and k_2=1. By A4, we obtain that k_1=k_2=1. Thus \exp(G)+s-3\le \mathsf D(G)=|U|=s which implies that \exp(G)\le 3, a contradiction.

Suppose that |Z|=3. Then UV Z^{-1} can only be a product of atoms of length 2. But \mathsf v_{g_1}(UV Z^{-1})=k_1 > \mathsf v_{-g_1}( UV)\ge \mathsf v_{-g_1}( UV Z^{-1}), a contradiction.

(a)  \Rightarrow  (b) Suppose that G is an elementary 2-group, and let (e_1, \ldots, e_r) be a basis of G with \operatorname{ord} (e_1)=\ldots = \operatorname{ord} (e_r)=2. Then \mathsf D (G)=r+1, U= e_1 \cdot \ldots \cdot e_{r-1}e_0 \in \mathcal A (G), where e_0=e_1+ \ldots + e_{r-1}, and \mathsf L \big( U(-U) \big) = \{2,r\}.

Suppose that G is isomorphic to C_2^{r-1}\oplus C_4 for some r \ge 2, and let (e_1, \ldots, e_r) be a basis of G with \operatorname{ord} (e_1)=\ldots = \operatorname{ord} (e_{r-1})=2 and \operatorname{ord} (e_r)=4. Then \mathsf D (G)=r+3, U=e_1\cdot \ldots \cdot e_{r-1}e_r^2 (e_0+e_r) \in \mathcal A (G), where e_0=e_1+ \ldots + e_r, and \mathsf L \big( U(-U) \big) = \{2, r+2\}.

Suppose that G is isomorphic to C_2 \oplus C_{2n} for some n \ge 2, and let (e_1, e_2) be a basis of G with \operatorname{ord} (e_1)=2 and \operatorname{ord}(e_2)=2n. Then \mathsf D (G)=2n+1, U = e_2^{2n} \in \mathcal A (G), and \mathsf L \big( U(-U) \big)=\{2, 2n \}.

(b)  \Rightarrow  (a) Clearly, we have V=-U, and for every zero-sum sequence W with W\, | \, UV and W\neq UV, it follows that W is either an atom of length \mathsf D(G)-1 or W is a product of atoms of length 2. We set U=g_1^{k_1}\cdot \ldots \cdot g_l^{k_l} with l, k_1, \ldots, k_l \in \mathbb N and g_1, \ldots, g_l \in G pairwise distinct.

Suppose that l=1. Then k_1=\operatorname{ord} (g_1)=\mathsf D (G)-1. Thus k_1 is even and G \cong C_2 \oplus C_{k_1}.

Suppose that l \ge 2. For each i\in [1,l], the sequence S_i=g_i^{-k_i}(-g_i)^{k_i}U is either zero-sum free or an atom; clearly, S_i is an atom if and only if 2k_ig_i=0. So we can distinguish two cases.

CASE 1:  For each i\in [1,l] we have 2k_ig_i=0.

We claim that for any i\in [1,l], the tuple (g_1,\ldots ,g_{i-1}, g_{i+1}, \ldots , g_l) is independent. Clearly, it is sufficient to prove the claim for i=l. Assume to the contrary that (g_1, \ldots, g_{l-1}) is not independent. Then there is an atom W with W\, | \, g_1^{k_1}\cdot \ldots \cdot g_{l-1}^{k_{l-1}}\cdot (-g_1)^{k_1}\cdot \ldots \cdot (-g_{l-1})^{k_{l-1}} and |W|>2. Then W is an atom of length \mathsf D (G)-1, and thus W^{-1}UV is also an atom of length \mathsf D (G)-1. But g_l(-g_l)\, | \, W^{-1}UV, a contradiction. After renumbering if necessary we may suppose that

Suppose that l=2. By our assumption that \operatorname{ord} (g_1) \ge \operatorname{ord} (g_2), we obtain that

If \mathsf D (G)= \exp (G), then \operatorname{ord} (g_1)=\exp (G), \operatorname{ord} (g_1)-\operatorname{ord} (g_2)=2, and hence \operatorname{ord} (g_2)\, | \, 2 which implies \operatorname{ord} (g_2)=2 and \operatorname{ord} (g_1)=4=\mathsf D (G)=\exp (G), a contradiction to \mathsf D (G)\ge 5. If \mathsf D (G)=\exp (G)+1, then G\cong C_2 \oplus C_{2n} for some n \ge 2.

Suppose that l \ge 3. Then

Suppose that equality holds at the second inequality sign. Then

Since \operatorname{ord} (g_i)/2 \le \operatorname{ord} (g_i)-1 for all i \in [2, l-1], it follows that then \operatorname{ord}(g_i)=2 for all i \in [2,l-1]. Because our assumption on the order of the elements, we infer that \operatorname{ord} (g_1)=\operatorname{ord} (g_l)=2, and hence G is an elementary 2-group.

Suppose that inequality holds at the second inequality sign. Then we have

Since \operatorname{ord} (g_i)/2 \le \operatorname{ord} (g_i)-1 for all i \in [2, l-1], there exists an i \in [2, l-1], say i=2, such that \operatorname{ord} (g_2)=4 and \operatorname{ord} (g_i)=2 for all i \in [3, l-1]. If l=3, then G \cong C_4 \oplus C_4 or G \cong C_2 \oplus C_4, but the first case is a contradiction to Lemma 3.3. If l \ge 4, then G\cong C_2^{l-2}\oplus C_4.

CASE 2:  There exists an i\in [1,l] such that the sequence S_i is zero-sum free, say i=1.

We start with a list of assertions.

1. A1.  Let T, T' \in \mathcal F (G) be distinct such that T \, | \, U, T' \, | \, U, and \sigma (T)=\sigma (T'). Then \operatorname{supp}(T)\cap \operatorname{supp}(T')=\emptyset, T T' = U, and 2\sigma(T)=0.

2. A2.  Let T, T' \in \mathcal F (G) be distinct such that T \, | \, S_1, T' \, | \, S_1, and \sigma (T)=\sigma (T'). Then \operatorname{supp}(T)\cap \operatorname{supp}(T')=\emptyset, T T' = S_1, and 2\sigma(T)= - 2k_1g_1.

3. A3.  \Sigma (U) = G.

4. A4.  If i \in [1,l] with \operatorname{ord} (g_i)>2, then \operatorname{ord} (g_i)>2k_i.

Proof of  A1.  Obviously, T(-T') has sum zero and T(-T')\, | \, UV but T(-T')\neq UV. So T(-T') must be an atom of length \mathsf D(G)-1 which implies that \operatorname{supp}(T)\cap \operatorname{supp}(T')=\emptyset, T T' = U, and 2\sigma(T)=0.

Proof of  A2.  Obviously, T(-T') has sum zero, T(-T') \, | \, UV, and T(-T')\neq UV. So T (-T') must be an atom of length \mathsf D(G)-1 which implies that \operatorname{supp}(T)\cap \operatorname{supp}(T')=\emptyset, T T' = S_1, and 2\sigma(T)=-2k_1g_1.

Proof of  A3.  We will show that |\Sigma (U)| = |G|. Clearly, we have

Since U = g_1^{k_1} \cdot \ldots \cdot g_l^{k_l}, we have

By A1, there are at most two distinct subsequences of U with given sum g \in G. Therefore we obtain

Next we study |\Sigma (S_1)|. Since S_1 is zero-sum free of length |S_1|=\mathsf D (G)-1, it follows that \Sigma (S_1)=G\setminus \{0\}. Using A2 for the second equality sign we obtain that

and hence \Sigma (U)=G.

Proof of  A4.  Since S_1 is zero-sum free, it follows that \operatorname{ord} (g_1)>2k_1. Now let i \in [2,l] be given and assume to the contrary that \operatorname{ord} (g_i)\le 2k_i. Recall that \operatorname{ord} (g_i) > \mathsf v_{g_i}(U)=k_i. We set U = g_i^{k_i}U'. Then W'=(-g_i)^{\operatorname{ord} (g_i)-k_i}U' has sum zero and divides UV. Thus W' is an atom of length \mathsf D (G)-1 =|W'|=|U| and hence \operatorname{ord} (g_i)=2k_i. Since \Sigma (S_1)=G\setminus \{0\}, S_1 has a subsequence T such that \sigma (T)=(k_i+1)g_i. If g_i \nmid T, then Tg_i^{k_i-1} is a proper zero-sum subsequence of S_1, a contradiction. If g_i \, | \, T, then \sigma (g_i^{-1}T)=k_ig_i = \sigma (g_i^{k_i}). By A2, it follows that 0= 2k_ig_i= 2 \sigma (g_i^{-1}T )= -2k_1g_1 \ne 0, a contradiction.

Now we distinguish two subcases.

CASE 2.1:  |\{ i \in [1,l] \mid \operatorname{ord}(g_i)>2\}|\ge 3, say \operatorname{ord}(g_1)>2, \operatorname{ord}(g_2)>2, and \operatorname{ord}(g_3)>2.

We start with the following assertion.

1. A5.  There is a subsequence W of U with \sigma (W)= g_1-g_2 such that W=g_2^{k_2}W' and for any h\, | \, W', \operatorname{ord} (h)=2.

Proof of  A5.  By A3, there exists some W \in \mathcal F (G) such that W\, | \, U and \sigma(W)=g_1-g_2. We claim that g_1\nmid W but g_2^{k_2}\, | \, W. Assume to the contrary that g_1\, | \, W. Then \sigma(g_1^{-1}W)=-g_2=\sigma (g_2^{-1}U). By A1, this implies that 2g_2=0, a contradiction. Assume to the contrary that g_2^{k_2}\nmid W. Then g_2W\, | \, U and \sigma (g_2W)=g_1=\sigma(g_1). By A1, this implies that 2g_1=0, a contradiction. Thus W=g_2^{k_2}W' with W'\, | \, g_3^{k_3}\cdot \ldots \cdot g_l^{k_l}.

Let i \in [3, l] such that g_i\, | \, W'. We only need to show that \operatorname{ord}(g_i)=2. Assume to the contrary that \operatorname{ord} (g_i)>2. We set X=Ug_ig_2^{-1}, and then |X|=|U|=\mathsf D(G)-1. Suppose that X has a zero-sum subsequence T. Then g_i^{k_i+1}\, | \, T and by A4 we obtain that \operatorname{ord}(g_i)>2k_i. Therefore, g_i^{-k_i}T\, | \, U and g_i^{-k_i}T \neq g_i^{-k_i}U but \sigma( g_i^{-k_i}T)=-k_ig_i=\sigma (g_i^{-k_i}U) which implies that 2k_ig_i=0 by A1, a contradiction. Thus X is zero-sum free, and |X|=\mathsf D (G)-1 which imply that \Sigma (X)=G\setminus \{0\}. Therefore, X has a subsequence T such that \sigma (T)=g_1-g_2.

Suppose that g_i^{k_i+1}\nmid T. Then T\, | \, U, and by definition of X we have g_2^{k_2}\nmid T which implies that g_2T \, | \, U. Since \sigma (g_2T)=g_1= \sigma (g_1), we obtain that 2g_1=0 by A1 , a contradiction.

Suppose that g_i^{k_i+1}\, | \, T. Then g_i^{-1}T\, | \, U. Since \sigma (g_i^{-1}T)=g_1-g_2-g_i=\sigma(g_i^{-1}W) and g_i^{-1}T\neq g_i^{-1}W, we obtain that \operatorname{supp}(g_i^{-1}T)\cap \operatorname{supp}(g_i^{-1}W)=\emptyset and U=g_i^{-1}T\cdot g_i^{-1}W by A1. Set T_1=g_i^{-1}T, then g_1^{k_1}g_i^{k_i}\, | \, T_1 and g_2\nmid T_1. It follows that \sigma(g_1^{-1}g_2T_1)=-g_i=\sigma (g_i^{-1}U) which implies that 2g_i=0 by A1, a contradiction.

Repeating the argument of A5, we can find another subsequence W_1 of U with \sigma(W_1)=g_2-g_1 such that W_1=g_1^{k_1}W_1' and for any h\, | \, W_1', \operatorname{ord} (h)=2. Set Y=WW_1(\prod\limits_{h\in \operatorname{supp}(W')\cap \operatorname{supp}(W_1')}h^2)^{-1}, then Y is a zero-sum subsequence of U. Since \operatorname{ord}(g_3)>2, we have g_3\nmid W and g_3\nmid W_1 which imply that g_3\nmid Y. It follows that Y is a proper zero-sum subsequence of U, a contradiction.

CASE 2.2:  |\{i \in [1,l] \mid \operatorname{ord}(g_i)>2\}|\le 2.

Since k_1g_1+ \ldots + k_lg_l=\sigma (U)=0 and \operatorname{ord} (k_1g_1)> 2, it follows that |\{i \in [1,l] \mid \operatorname{ord}(g_i)>2\}| = 2, say, \operatorname{ord}(g_1)>2 and \operatorname{ord}(g_2)>2. Then A4 implies that \operatorname{ord}(g_1)>2k_1 and \operatorname{ord}(g_2)>2k_2.

Suppose that l=2. Then

which implies that G is a cyclic group and k_1=k_2. Since any minimal zero-sum sequence of length |G|-1 over a cyclic group has the form g^{|G|-2}(2g) for some generating element g \in G, it follows that 1=k_2=k_1, and hence |G|=3, a contradiction to \mathsf D (G)\ge 5.

Suppose l\ge 3. Then \exp(G) is even. We may assume that \operatorname{ord}(g_1)\ge \operatorname{ord}(g_2). Since (g_3,\ldots,g_l) is independent, we have \mathsf r(G)\ge l-2. Therefore,

Since \operatorname{ord}(g_1)\, | \, \exp(G), we have that \operatorname{ord}(g_1)=\exp(G) is even. Thus 2\lfloor\frac{\operatorname{ord}(g_1)-1}{2}\rfloor=\operatorname{ord}(g_1)-2 which implies that k_1+k_2=\lfloor\frac{\operatorname{ord}(g_1)-1}{2}\rfloor+\lfloor\frac{\operatorname{ord}(g_2)-1}{2}\rfloor= 2\lfloor\frac{\operatorname{ord}(g_1)-1}{2}\rfloor. Then \operatorname{ord}(g_1)=\operatorname{ord}(g_2)=2k_1+2=2k_2+2. Since \sigma(U)=k_1g_1+k_2g_2+g_3+\ldots+g_l=0, we have 2k_1g_1+2k_2g_2=0 which implies that 2g_1+2g_2=0. If k_1=k_2\ge 2, then g_1^2g_2^2 is an atom and divides UV, a contradiction. Thus k_1=k_2=1, and hence \operatorname{ord}(g_1)=\operatorname{ord}(g_2)=4. By A3, there exists a subsequence W of U such that \sigma (W)=2g_1. If g_1\, | \, W, then g_1\nmid g_1^{-1}W and \sigma (g_1^{-1}W)=g_1=\sigma(g_1) which implies that 2g_1=0 by A1, a contradiction. If g_1\nmid W, then g_1\nmid U(g_1W)^{-1} and \sigma (U(g_1W)^{-1})=g_1=\sigma(g_1) which implies that 2g_1=0 by A1, a contradiction.