Free multiderivations of connected subgraph arrangements

Suppose ({\mathscr A}^H,\kappa) is free. Then so is {\mathscr A}, by Theorem 2.4, and so is {\mathscr A}_X for any X \in L({\mathscr A}), by Theorem 2.1 (with simple multiplicities), a contradiction.

For each G as in the statement above, we present an element X\in L({\mathscr A}_G) in Table ? such that the corresponding localization ({\mathscr A}_G)_X is a generic arrangement with \operatorname{rank} ({\mathscr A}_G)_X<\vert({\mathscr A}_G)_X\vert. Consequently, owing to Theorems 2.1 and 2.6 we conclude that {\mathscr A}_G is totally non-free.

Let v_1,v_2,v_3,v_4,v_5 be the vertices of G included in the subgraph forming G_3. The induced subgraph of G on \{v_1,v_2,v_3,v_4,v_5\} is isomorphic to G_3, G_4, G_5, G_{9}, G_{11}, G_{13}, G_{14}, G_{15}, G_{16}, G_{18} or G_{19}. Thus, by Proposition 3.1, the arrangement of this induced subgraph is totally non-free. It then follows from Lemma 2.22 and Theorem 2.1 that {\mathscr A}_G is totally non-free as well.

By Theorem 2.23, ({\mathscr A}_G,\mathbbm{1}) is free for any graph in the list above.

Conversely, assume that we have a graph G=(N,E) and a multiplicity \mu \geq \mathbbm{1} on {\mathscr A}_G such that ({\mathscr A}_G,\mu) is free.

1. Assume a vertex of G has at least 4 neighbors.

Then G_3 is a subgraph of G and {\mathscr A}_G is totally non-free, by Corollary 3.2. This implies that all the vertices of G have degree at most 3.

2. Assume that G has at least 2 cycles of length at least 3 that share an edge. Since we assume that \vert N\vert\geq 5, the graph cannot be equal to G_1 or G_2. Let C_1,C_2 be cycles of G with a shared edge (v_1,v_2), choose w_1\in C_1,w_2\in C_2 such that w_1\not\in C_2,w_2\not\in C_1. Assume first that both cycles have length 3. In this case the induced graph on \{v_1,v_2,w_1,w_2\} is of type G_1 or G_2. Since \vert N\vert\geq 5 and G is connected there exists a vertex v that is a neighbor of at least one element of \{v_1,v_2,w_1,w_2\}, while not belonging to either C_1 or C_2. Taking the induced subgraph on \{v_1,v_2,w_1,w_2,v\} yields one of G_{9} up to G_{19}. So {\mathscr A}_G is again totally non-free, thanks to Proposition 3.1. Assume that at least C_1 is of length greater than 3. Choose v_1,v_2,w_1,w_2 as in the first part above and an additional w_3\in C_1 and take the induced subgraph on \{v_1,v_2,w_1,w_2,w_3\}. Now contract each edge at least one vertex of which is not in \{v_1,v_2,w_1,w_2,w_3\}. We now arrive at one of G_{11}, G_{12}, G_{14}, G_{15}, G_{16}, G_{17}, G_{19}, G_{20}. So {\mathscr A}_G is totally non-free, again by Proposition 3.1.

3. Assume that G has at least 2 cycles of length at least 3 without any shared edge. In this case we contract edges until we end up with G_3 as subgraph and use Corollary 3.2.

4. Assume G has exactly one cycle C of length at least 4. Note that we cannot have two vertices v_1,v_2\in C being connected by an edge e not belonging to C, since otherwise we could decompose C into two cycles C_1, C_2 of length at least 3 with shared edge e, which was already excluded in case 2. If G is not a cycle graph, then we can choose a vertex v such that it is the neighbor of a vertex in C. After contracting edges of C until we arrive at C_4 and taking the subgraph induced by the reduced cycle and v, we get one of G_{6}, G_{12}, G_{15}, G_{17}, G_{20}. So {\mathscr A}_G it is totally non-free, again thanks to Proposition 3.1.

5. Assume G has exactly one cycle of length 3. Let v_1, v_2, v_3 be the unique cycle in G. Assume that all vertices v_1 , v_2 , v_3 are of degree 3. Taking the subgraph induced by \{v_1 , v_2 , v_3\} together with their respective third neighbors yields the graph G_7, since all neighbors are distinct ((v_1 , v_2 , v_3) is the only cycle in G). Using Proposition 3.1, we see that {\mathscr A}_{G_7} is totally non-free, so we can assume that the vertex v_3 has degree 2. If G had a vertex v of degree 3 different from v_1 and v_2, then contracting all edges between v and v_1 (or v and v_2, we choose whatever vertex can be reached first by following the unique path) results in a graph with a vertex of degree at least 4. This is a contradiction as seen in case 1. Therefore, all other vertices are of degree at most 2 and as v_1, v_2, v_3 constitute its only cycle, G must be a path-with-triangle-graph.

6. Assume that G is a tree. If there are two vertices of degree 3, then we contract edges until it becomes a vertex of at least degree 4. This connected subgraph arrangement is totally non-free, as seen in case 1. So there is at most one vertex of degree 3. If G it is not a path-graph or an almost-path-graph, then it has G_8 as an induced subgraph and the arrangement {\mathscr A}_G is totally non-free, due to Proposition 3.1.

We fix the following order on the hyperplanes of {\mathscr A}_{G_2}:

and define the multiplicity \mu_2=(2,2,2,2,2,1,1,1,1,1,1,1,1,1,1). Then the multiarrangement ({\mathscr A}_{G_2},\mu_2) is recursively free but not additively free. It follows from Remark 2.11 that ({\mathscr A}_{G_2},\mu_2) fails to be additively free since we have \vert(({\mathscr A}_{G_2})^H,(\mu_2)^*)\vert=14 for an arbitrary H\in {\mathscr A}_{G_2}.

For {\mathscr A}_{G_1} we drop the last hyperplane H_{24} from the list above, but fix the same order on the rest of the hyperplanes and define \mu_1=(2, 2, 2, 1, 2, 1, 1, 1, 2, 1, 1, 1, 1, 1). We claim that ({\mathscr A}_{G_1},\mu_1) is inductively free.

The facts that ({\mathscr A}_{G_1},\mu_1) is inductively free and ({\mathscr A}_{G_2},\mu_2) is recursively free follow from the corresponding induction tables, Table ? and Table ?, respectively.

Since |{\mathscr A}_{G_1}| = 14, a quadruple of potential exponents for {\mathscr A}_{G_1} has to be (1,4,4,5)_\leq, or less balanced. Calculating \operatorname{GMP} (2) for this tuple gives \operatorname{GMP}(2)=69. Since exponents that are less balanced afford a smaller \operatorname{GMP} (2) and since \operatorname{LMP} (2)=70, we have \operatorname{LMP} (2)\ne\operatorname{GMP} (2) and so {\mathscr A}_{G_1} is not free, by Lemma 2.32.

Since |{\mathscr A}_{G_1}| = 15, choose potential balanced exponents (1,4,5,5)_\leq for {\mathscr A}_{G_2} to maximize \operatorname{GMP} (2). Using this quadruple, we get \operatorname{GMP} (2)=79. Since exponents that are less balanced yield a smaller value for \operatorname{GMP} (2) and since \operatorname{LMP} (2)=80, we have \operatorname{LMP} (2)\ne\operatorname{GMP} (2) and so {\mathscr A}_{G_2} can’t be free, thanks to Lemma 2.32.

The case c = 1 is Proposition 4.3. So let c\geq 2. From the calculation in the example we know that: G_1: \nu_1=21,\nu_2=91 and G_2: \nu_1=25,\nu_2=105, which gives \nu_3=28 for G_1 and \nu_3=30 for G_2. So for G_1 we get

and for G_2 we get

If we assume perfectly balanced exponents for {\mathscr A}_{G_1} (assuming 14c\equiv 0\mod 4) we get the set of exponents (\frac{14c}{4},\frac{14c}{4},\frac{14c}{4},\frac{14c}{4})_\leq which results in \operatorname{GMP} (2)=73.5c^2.
If we assume perfectly balanced exponents for {\mathscr A}_{G_2} (assuming 15c\equiv 0\mod 4) we get the set of exponents (\frac{15c}{4},\frac{15c}{4},\frac{15c}{4},\frac{15c}{4})_\leq which results in \operatorname{GMP} (2)=84.375c^2.
Since c\geq 2, in any case, we have \operatorname{LMP} (2)>\operatorname{GMP} (2) and thus ({\mathscr A}_{G_2},\mu) can’t be free, by Lemma 2.32.

For the forward implication, using the algorithm above, we get \nu_1=1\cdot 3+3\cdot 1=6,\nu_2=\binom{7}{2}=21,\nu_3=21-3\cdot 6=3. This gives

Assuming balanced exponents we get

For c\geq 4 we get that \operatorname{LMP} (2)>\operatorname{GMP} (2). For c=3 we get equality and for c=2 we get \operatorname{LMP} (2)=66 and depending on whether the assumed exponents are (3,5,6)_\leq,(4,4,6)_\leq, or (4,5,5)_\leq, a value of 63, 64, or 65, respectively for \operatorname{GMP} (2). So if c\not\in\{1,3\} we have \operatorname{LMP} (2)>\operatorname{GMP} (2) and thanks to Lemma 2.32, we deduce non-freeness.

Next we address the reverse implication. The freeness of ({\mathscr A}_G,\mathbbm{1}) follows from Theorem 2.23. In Table ?, we present an induction table for ({\mathscr A}_G,3 \mathbbm{1}) showing that the latter is actually inductively free.

First let G\in\{C_n,\Delta_{n,k}\}. Since G either has C_3 as an induced subgraph or becomes C_3 after contracting edges, we can use Lemma 2.22 and Proposition 4.7 to focus on the cases where c\in\{1,3\}. If c=1 we know that ({\mathscr A}_G,\mu) is free. So let \mu\equiv 3 and G=C_4.
Calculating \operatorname{LMP} (2) we get \nu_1=1\cdot (4+2)+ 4\cdot 2+4\cdot 1=18, \nu_2=\binom{13}{2}=78,\nu_3=78-3\cdot 18=24 and so

Since {\mathscr A}_{C_4} consists of 13 hyperplanes, assuming balanced exponents (9,10,10,10)_\leq, we get \operatorname{GMP} (2)=570. So \operatorname{LMP} (2)>\operatorname{GMP} (2). Using Theorem 2.28, we see that {\mathscr A}_G fails to be free. For a given C_n (for n > 4) we can contract edges until we end up with C_4. The result follows.

Let G=\Delta_{3,1} and note that every \Delta_{n,k}\neq C_3 can get transformed into G by contracting edges. If \mu\equiv 3 we get \operatorname{LMP} (2)=489 and assuming exponents (9,9,9,9)_\leq we get \operatorname{GMP} (2)=486. So only the simple multiplicity can be free.

Let G=A_{3,2}. Let \mu\equiv c, then we get

Moreover, \operatorname{LMP} (2)=46c^2 (if c is even) and 46c^2-3 (if c is odd). Once more we see that \operatorname{LMP} (2)\neq \operatorname{GMP} (2) and so, by Theorem 2.28, ({\mathscr A},\mu) fails to be free. Since every A_{n,k} with n\geq 4 can be transformed into A_{3,2} after contracting edges, the result follows from Lemma 2.22.

(1) and (2): We give an explicit basis for the module of derivations D({\mathscr A},\mu) in both cases. We utilize the linear isomorphism from {\mathbb C}^2 to itself, defined by x_1\mapsto x_2 and x_2\mapsto x_1-x_2, sending \{x_1,x_2,x_1+x_2,2x_1+x_2\} to \{x_1,x_2,x_1+x_2,x_1-x_2\}. Thus it is sufficient to calculate a basis for D(\mathscr{B},\nu), where Q(\mathscr{B},\nu)=x_1^cx_2^a(x_1+x_2)^d(x_1-x_2)^b.
For (1), we have Q(\mathscr{B},\nu)=x_1^{2k}x_2^{2k}(x_1+x_2)^1(x_1-x_2)^1 and a basis for D(\mathscr{B},\nu) is given by

For (2), we have Q(\mathscr{B},\nu)=x_1^{2k+1}x_2^{2k+1}(x_1+x_2)^1(x_1-x_2)^1 and a basis for D(\mathscr{B},\nu) is given by

Verifying that \theta_1, \theta_2\in D(\mathscr{B},\nu) in both cases and checking the determinant condition in Theorem 2.3 (which gives \det(p_{ij})=x_1^{2k} x_2^{2k + 2} - x_1^{2k + 2} x_2^{2k}=(x_1 - x_2) (x_1 + x_2) (-x_1^{2k}) x_2^{2k} in the first case and \det(p_{ij})=(x_1 - x_2) (x_1 + x_2) x_1^{2k + 1} x_2^{2k + 1} in the second case) shows that in both cases \{\theta_1,\theta_2\} is a basis for D(\mathscr{B},\nu), which completes the proof.
(3) and (4): Start with \mu'=(k,0,k,2k-4) resp. \mu'=(k+1,0,k+1,2k-4) and use Theorem 2.16 to see that ({\mathscr A},\mu') is free with exponents (2k-2,2k-2) resp. (2k-1,2k-1). So ({\mathscr A},\mu) has to be free with exponents (2k-2,2k-1) resp. (2k-1,2k).
The argument is the same for the other two multiplicities, but we delete \ker(2x_1+x_2) instead.

The subarrangement \mathscr{B}_0 of \mathscr{D} defined by Q(\mathscr{B}_0,\mu_0)=x_1^{2k}x_2^{2k}x_3^{2k} is boolean and therefore free with exponents (2k,2k,2k). Let Q(\mathscr{B}_1,\mu_1):=x_1^{2k}x_2^{2k}x_3^{2k}(x_1+x_2) and H_1=\ker(x_1+x_2). Then we get \mathscr{B}_1^{H_1}=\{\ker(x_1),\ker(x_2)\}. Using Proposition 2.9, we get \mu_1^*=(2k,2k) for the corresponding Euler-multiplicity. Therefore, (\mathscr{B}_1^{H_1},\mu_1^*) is free with exponents \exp(\mathscr{B}_1^{H_1},\mu_1^*)=(2k,2k). Using Theorem 2.10 we deduce that (\mathscr{B}_1,\mu_1) is free with exponents (2k,2k,2k+1). Now let Q(\mathscr{B}_2,\mu_2):=x_1^{2k}x_2^{2k}x_3^{2k}(x_1+x_2)(x_1+x_3) and H_2=\ker(x_1+x_3), then we get \mathscr{B}_2^{H_2}=\{\ker(x_1),\ker(x_2),\ker(x_1-x_2)\} and Proposition 2.9 gives \mu_2^*=(2k,2k,1). Therefore, (\mathscr{B}_2^{H_2},\mu_2^*) is free with exponents \exp(\mathscr{B}_2^{H_2},\mu_2^*)=(2k,2k+1), so (\mathscr{B}_2,\mu_2) is free with exponents (2k,2k+1,2k+1). Finally, restricting (\mathscr{D},\mu) to H_3=\ker(x_2+x_3) gives \mathscr{D}^{H_3}=\{x_1,x_2,x_1+x_2,x_1-x_2\} with Euler-multiplicity \mu^*=(2k,2k,1,1). In the proof of Lemma 5.3 we show that (\mathscr{D}^{H_3},\mu^*) is free with exponents \exp(\mathscr{D}^{H_3},\mu^*)=(2k+1,2k+1). Using Theorem 2.10 once more completes the proof.

The reverse implication follows from Proposition 5.4.

For the forward implication assume that (\mathscr{D},\mu) is free. For some X\in L({\mathscr A})_2 and the corresponding rank 2 localization {\mathscr A}_X=\{H_1,H_2,H_3\} or {\mathscr A}_X=\{H_1,H_2\} which is free with \exp({\mathscr A}_X,\mu_X)=(d_1^X,d_2^X), we write its summand d_1^Xd_2^X in \operatorname{LMP} (2) as [\mu(H_1),\mu(H_2),\mu(H_3)] or [\mu(H_1),\mu(H_2)] respectively. Using this notation, \operatorname{LMP} (2) of (\mathscr{D},\mu) equals

With our assumption d=e=f=1, we get

which (because of a\geq b\geq c and Theorem 2.16) equals

Define the multiplicity \mu_1=(a,b,c,0,0,0) on \mathscr{D}, then (\mathscr{D},\mu_1) is free with exponents (a,b,c) since it is boolean. Assuming that (\mathscr{D},\mu) is free, the arrangement has exponents equal to one of (a+3,b,c)_\leq,(a+2,b+1,c)_\leq,\dots,(a,b,c+3)_\leq, thanks to [hogeroehrleschauenburg:free, Rem. 2.8]. Writing E:=ab+ac+bc this gives \operatorname{GMP} (2) as follows:

Since each of a,b,c is even by assumption, we get that in all but the last case (where \exp(\mathscr{D},\mu)=(a+1, b+1, c+1)_\leq), \operatorname{GMP} (2) is even. But since

is odd, we see that \operatorname{LMP} (2)\neq \operatorname{GMP} (2) in those cases. In particular, the arrangement cannot be free because of Theorem 2.28. So let us assume \exp(\mathscr{D},\mu)=(a+1,b+1,c+1)_\leq which gives \operatorname{GMP} (2)=E + 2 a + 2 b + 2 c + 3. But if we do not have a=b=c, then a>b or b>c. This implies a>c and we have

So using Theorem 2.28 again, we see that (\mathscr{D},\mu) cannot be free and we have a=b=c.

By [cuntzkuehne:subgrapharrangements, Prop. 3.1], {\mathscr A}_{C_3} is free with exponents (1,3,3). Let H=H_{123} and note that {\mathscr A}_{C_3}^H is free with exponents (1,2), since \vert{\mathscr A}_{C_3}^H\vert=3. According to Theorem 2.10, {\mathscr A}_{C_3}\backslash\{H\}=\mathscr{D} is not free. By Theorem 2.25, (\mathscr{D},\mu) is free if and only if {\mathscr E}(\mathscr{D},\mu) free. However, {\mathscr E}(\mathscr{D},\mu)_X=\mathscr{D} for X=H_1\cap H_2\cap H_3. So {\mathscr E}(\mathscr{D},\mu) cannot be free, by Theorem 2.1.

The idea of the proof is to increase the multiplicity of H_{12} by 1 which results in a multiplicity \mu_1=(m_1,m_2,m_3+1) such that (2.25) is satisfied and thus {\mathscr E}({\mathscr A},\mu_1) is a free extension of ({\mathscr A},\mu_1). Now we remove a hyperplane H so that {\mathscr E}({\mathscr A},\mu_1)\backslash\{H\}={\mathscr E}_\sigma({\mathscr A},\mu) is free, thanks to Theorem 2.10. We demonstrate this in one case while the other cases are handled analogously.

For (i) let \mu_1=(m_1,m_2,m_3+1)=(even,even,even). The extended arrangement {\mathscr A}_1:= {\mathscr E}_0({\mathscr A},\mu_1) is free, by Theorem 2.25. Note that for \sigma=(0,0,1), H=\ker(x_1+x_2+(\frac{m_3-1}{2})x_3) and {\mathscr A}_2:= {\mathscr E}_{\sigma}({\mathscr A},\mu), we have {\mathscr A}_2={\mathscr A}_1\backslash \{H\}. So we start at the free arrangement {\mathscr A}_1, fix the hyperplane H, and use Theorem 2.10 to derive the freeness of {\mathscr A}_2. Note that the exponents of {\mathscr A}_1 can be derived by using Theorems 2.16 and 2.2. The following case by case investigation is necessary due to [yoshinaga:extendable, Lem. 3.1].

Assume m_2-m_1\leq m_3 < m_1+m_2. Then we have \exp({\mathscr A}_1)=(1,\frac{m_1+m_2+m_3+1}{2},\frac{m_1+m_2+m_3+1}{2}). Note that it is sufficient to show that \vert {\mathscr A}_1^{H}\vert=1+\frac{m_1+m_2+m_3+1}{2}, since {\mathscr A}_1^{H} is a simple arrangement of rank 2 and so is free with exponents (1,\frac{m_1+m_2+m_3+1}{2}), satisfying the conditions of Theorem 2.10. We count all elements X\in L({\mathscr A}_1)_2 with X\subset H by listing all localizations {\mathscr A}_X. There is a canonical bijection between the set containing those elements and the set of hyperplanes of {\mathscr A}_1^H. We identify the hyperplanes of {\mathscr E}({\mathscr A},\mu_1) with their linear forms. We have a single localization with more than 3 elements:

and a total of \frac{m_1+m_2-(m_3+1)}{2} localizations with 3 elements

The rank 2 localization with more than 3 elements contains m_3+1 elements which are not equal to H while all of the localizations with 3 elements combined contain m_1+m_2-m_3-1 hyperplanes distinct from H. Subtracting the number of hyperplanes included in these localizations from the total number of hyperplanes in {\mathscr A}_1 shows that there are

hyperplanes H'\in {\mathscr A}_1 such that \vert({\mathscr A}_1)_{H\cap H'}\vert=2.

In conclusion we have

hyperplanes in \vert{\mathscr A}_1^{H}\vert. This implies that \exp({\mathscr A}_1^{H})=(1,\frac{m_1+m_2+m_3+1}{2}). By Theorem 2.10, {\mathscr A}_2 is free with exponents \exp({\mathscr A}_2)=(1,\frac{m_1+m_2+m_3-1}{2},\frac{m_1+m_2+m_3+1}{2}). The cases where m_3+1<m_2-m_1+1 and m_3+1\geq m_1+m_2+1 and part (ii) are handled analogously.

The reverse implication follows from Proposition 5.4.
For the forward implication note that if at least two of the values a,b,c are odd or all of d,e,f are even, then (\mathscr{D},\mu) is not free, by Corollary 5.6. Consequently, we can assume (after possibly permuting some basis vectors) that at least d is odd and a,b are even so (2.25) is not satisfied. If \mu=(even,even,even,1,1,1) and \mu\neq(2k,2k,2k,1,1,1), then we use Lemma 5.5 to see that (\mathscr{D},\mu) is not free. For the remaining multiplicities \mu we give an extension of (\mathscr{D},\mu) which is locally free along H_4, but has a localization isomorphic to \mathscr{D}. So (\mathscr{D},\mu) cannot be free, by Corollary 2.5. We now give an alternative extension for all remaining multiplicities (up to a permutation of coordinates) where \mu\neq (2k,2k,2k,1,1,1). Note that the local freeness along H_4 in the cases where (2.25) is not met is given by Lemma 5.8. We write H_{i4}=\ker(x_i-x_4) and H_{ij4}=\ker(x_i+x_j-x_4).

Let {\mathscr A}={\mathscr E}_\sigma(\mathscr{D},\mu) be an extension of (\mathscr{D},\mu), where \sigma is given as in the following table.

The localizations of {\mathscr A}={\mathscr E}_\sigma(\mathscr{D},\mu) isomorphic to \mathscr{D} are:

1. (a) {\mathscr A}_X=\{H_2,H_3,H_{14},H_{23},H_{124},H_{134}\} for X=(H_2\cap H_3\cap H_{14})\in L({\mathscr A}),

2. (b) {\mathscr A}_X=\{H_1,H_3,H_{24},H_{13},H_{124},H_{234}\} for X=(H_1\cap H_3\cap H_{24})\in L({\mathscr A}),

3. (c) {\mathscr A}_X=\{H_{1},H_{2},H_{34},H_{12},H_{134},H_{234}\} for X=(H_1\cap H_2\cap H_{34})\in L({\mathscr A}).

We now show how the table above can be applied to one of the remaining multiplicities \mu to derive the non-freeness for (\mathscr{D},\mu). So let \mu=(2,4,4,1,1,3). First note that since in this case a,b,c are even and d=e=1, we use the last row of the table. The second to last column tells us that \sigma=(0,0,0,1,1,1), since all of the values \mu(H_{ij}) are odd. Using the notation of Lemma 5.8 and choosing an arbitrary X\in L(\mathscr{D})_2 with \vert\mathscr{D}_X\vert=3, we have parities (even,even,odd) on \mathscr{D}_X and \sigma\vert_X=(0,0,1). So Lemma 5.8 gives us the local freeness of {\mathscr E}_\sigma(\mathscr{D},\mu) along H_4. Since {\mathscr E}_\sigma(\mathscr{D},\mu) has localization (a) with hyperplanes

(which is isomorphic to \mathscr{D} and therefore non-free) we can apply Corollary 2.5 to show that (\mathscr{D},\mu) is not free. This completes the proof of the theorem.

Let H=H_{123}. Owing to Theorem 5.9, we see that ({\mathscr A}_{C_3}',\mu') is free with exponents (2k+1,2k+1,2k+1). Now restrict ({\mathscr A}_{C_3},\mu) to H and calculate \mu^* by using Theorem 2.16 on the following rank 2 flats:

This gives \mu^*=(2k,2k,2k) and therefore (again by Theorem 2.16) ({\mathscr A}_{C_3}^H,\mu^*) is free with \exp({\mathscr A}_{C_3}^H,\mu^*)=(3k,3k). So by Remark 2.11 ({\mathscr A}_{C_3},\mu) is free if and only if k=1.

Let \mu(H_1)=\mu(H_2)=\mu(H_3)=\mu(H_{123})=k and \mu(H_{12})=r\geq 2k-1,\mu(H_{13})=\mu(H_{23})=1 and fix H=H_{23}, so \mu'=(k,k,k,r,1,0,k). We choose the following supersolvable filtration for ({\mathscr A}_{C_3}',\mu'):

This filtration satisfies the requirements of Theorem 5.12, because \mu'(H_{1})\geq \mu'(H_{3})+\mu'(H_{13})-1\iff k\geq k and \mu'(H_{2})\geq \mu'(H_{13})+\mu'(H_{123})-1\iff k\geq k and \mu'(H_{12})\geq \mu'(H_{3})+\mu'(H_{123})-1\iff 2k-1\geq 2k-1. So ({\mathscr A}_{C_3}',\mu') is inductively free with exponents (r,2k,2k+1). Restricting ({\mathscr A}_{C_3},\mu) to H_{23} gives hyperplanes in correspondence to the rank 2 flats

with \mu^*=(k,k,r,1). If r\geq 2k, then the restriction is free with exponents (2k+1,r) since deleting the hyperplane with multiplicity 1 gives a free arrangement with exponents (2k,r) and using Theorem 2.10 shows that ({\mathscr A}_{C_3},\mu) is inductively free with exponents (r,2k+1,2k+1). Deleting H_{12} with Theorem 2.10 and Lemma 5.3 until r=2k-5 (and we end up at exponents (2k-1,2k-1,2k-1)) shows the other claim. In particular: If r=3, then \mu=(3,3,3,1,1,1,3) and ({\mathscr A}_{C_3},\mu) is free with exponents \exp({\mathscr A}_{C_3},\mu)=(5,5,5). Restricting to an arbitrary H\in{\mathscr A}_{C_3} results in \vert({\mathscr A}_{C_3},\mu^*)\vert=8. So Remark 2.11 shows that ({\mathscr A}_{C_3}',\mu') can never be free and therefore ({\mathscr A}_{C_3},\mu) is neither inductively nor additively free.