1 Introduction

The generating graph of a finite group G is the graph defined on the elements of G in such a way that two distinct vertices are connected by an edge if and only if they generate G. It was defined by Liebeck and Shalev in [14], and has been further investigated by many authors: see for example [5, 7, 9, 10, 13, 17, 20,21,22] for some of the range of questions that have been considered. Clearly the generating graph of G is an edgeless graph if G is not 2-generated. We propose and investigate a possible generalization, that gives useful information even when G is not 2-generated.

Let G be a finite group. A generating set X of G is said to be minimal if no proper subset of X generates G. We denote by \(\Gamma (G)\) the graph whose vertices are the elements of G and in which two vertices x and y are joined by an edge if and only if \(x\ne y\) and there exists a minimal generating set of G containing x and y. Roughly speaking, x and y are adjacent vertices of \(\Gamma (G)\) is they are ‘independent’, so we call \(\Gamma (G)\) the of G. We will denote by V(G) the set of the non-isolated vertices of \(\Gamma (G)\) and by \(\Delta (G)\) the subgraph of \(\Gamma (G)\) induced by V(G). Our main result is the following.

Theorem 1

If G is a finite group, then the graph \(\Delta (G)\) is connected.

We prove a stronger result in the case of finite soluble groups. For a positive integer u, we denote by \(\Gamma _u(G)\) the subgraph of \(\Gamma (G)\) in which x and y are joined by an edge if and only if there exists a minimal generating set of size u containing x and y. As before, we denote by \(\Delta _u(G)\) the subgraph of \(\Gamma _u(G)\) induced by the set \(V_u(G)\) of its non-isolated vertices. Notice that, even when G is u-generated, the set \(V_u(G)\) is in general different from V(G). For example if \(G={{\,\mathrm{Sym}\,}}(4)\), then \(\{(1,2)(3,4), (1,2), (1,2,3)\}\) is a minimal generating set for G, so \((1,2)(3,4)\in V(G);\) however \((1,2)(3,4)\notin V_2(G)\). If G is a non-cyclic 2-generated group, then \(\Gamma _2(G)\) coincides with the generating graph of G and it follows from [10, Theorem 1] that \(\Delta _2(G)\) is a connected graph if G is soluble. We generalize this result in the following way.

Theorem 2

If \(u\in {\mathbb {N}}\) and G is a finite soluble group, then \(\Delta _u(G)\) is connected.

Recall that a graph is said to be embeddable in the plane, or planar, if it can be drawn in the plane so that its edges intersect only at their ends. The 2-generated finite groups whose generating graph is planar have been classified in [18]. Our next result gives a classification of the finite groups G such that \(\Gamma (G)\) is a planar graph.

Theorem 3

Let G be a finite group. Then \(\Gamma (G)\) is planar if and only either \(G\in \{C_2\times C_2, C_2\times C_4, D_4, Q_8, {{\,\mathrm{Sym}\,}}(3)\}\) or \(G=C_n\) is cyclic of order n and one of the following occurs:

  1. (1)

    n is a prime-power.

  2. (2)

    \(n=p\cdot q\), where p and q are distinct primes and \(p\le 3\).

  3. (3)

    \(n=4\cdot q\), where q is an odd prime.

Other results, and some related open questions, are presented in Section 5.

2 Proof of Theorem 1

Lemma 4

Let \(g\in G\). Then g is isolated in \(\Gamma (G)\) if and only if either \(G=\langle g\rangle \) or \(g\in {{\,\mathrm{Frat}\,}}(G)\).

Proof

Suppose \(g\notin {{\,\mathrm{Frat}\,}}(G)\). There exists a maximal subgroup M of G with \(g\notin M\). The set \(X=\{g\}\cup M\) contains a minimal generating X of G and \(g\in X\) (otherwise \(G=\langle X\rangle \le M\)). If \(X\ne \{g\}\), then g is not isolated, otherwise \(\langle g \rangle =G\). \(\square \)

Proposition 5

If G is a finite cyclic group, then \(\Delta (G)\) is connected.

Proof

Let \(|G|=p_1^{a_1}\ldots p_t^{a_t}\), where \(p_1,\ldots ,p_t\) are distinct primes. If \(t=1\), then \(V(G)=\varnothing \). So assume \(t>1\) and, for \(1\le i\le t\), let \(g_i\) be an element of G of order \(|G|/p_i^{a_i}\). The subset \(X=\{g_1,\ldots ,g_t\}\) induces a complete subgraph of \(\Delta (G)\). Now let \(x\in V(G)\). Since \(x\notin {{\,\mathrm{Frat}\,}}(G)\), there exists \(i\in \{1,\ldots ,t\}\) such that \(p_i^{a_i}\) divides |x| and x is adjacent to \(g_i\). \(\square \)

Lemma 6

Let N be a normal subgroup of a finite group G. If \(Y=\{y_1,\ldots ,y_t\}\) has the property that \(\langle Y, N\rangle =G\), but \(\langle Z, N\rangle \ne G\) for every proper subset Z of Y, then there exist \(n_1,\ldots ,n_u\in N\) such that \(\{y_1,\ldots ,y_t,n_1,\ldots ,n_u\}\) is a minimal generating set of G.

Proof

Since \(G=\langle Y, N\rangle \), \(Y\cup N\) contains a minimal generating set X of G, and the minimality property of Y implies \(Y\subseteq X\). \(\square \)

Lemma 7

Let N be a normal subgroup of a finite group G. If \(x_1N\) and \(x_2N\) are joined by an edge of \(\Delta (G/N)\), then \(x_1n_1\) and \(x_2n_2\) are joined by an edge of \(\Delta (G)\) for every \(n_1, n_2\in N\).

Proof

Let \(\{x_1N,x_2N,x_3N,\ldots ,x_tN\}\) be a minimal generating set of G/N. By Lemma 6, for every \(n_1, n_2 \in N\), there exists \(m_1,\ldots ,m_u\in N\) such that

$$\begin{aligned} \{ x_1n_1,x_2n_2,x_3,\ldots ,x_t,m_1,\ldots ,m_u\} \end{aligned}$$

is a minimal generating set of G. \(\square \)

We will write \(x_1\sim _G x_2\) if \(x_1\) and \(x_2\) belong to the same connected component of \(\Delta (G)\). The following lemma is an immediate consequence of Lemma 7.

Lemma 8

Let N be a normal subgroup of a finite group G and let \(x, y \in G\). If \(xN, yN \in V(G/N)\) and \(xN\sim _{G/N}yN\), then \(x\sim _G y\).

Lemma 9

[3, Corollary 1.5] Let G be a finite group with \(S:= F^*(G)\) nonabelian simple. If xy are nontrivial elements of G, then there exists \(s\in G\) such that \(\langle x, s\rangle \) and \(\langle y, s\rangle \) both contain S.

Lemma 10

Let G be a finite monolithic primitive group. Assume that \(N={{\,\mathrm{soc}\,}}G\) is non abelian and that \(G= \langle x_1, N\rangle =\langle x_2, N\rangle \). Then there exists \(m\in N\) such that \(\langle x_1, m \rangle =\langle x_2, m \rangle =G\).

Proof

We have \(N=S_1\times \cdots \times S_t\), where \(t\in {\mathbb {N}}\) and \(S_i\cong S\) with S a nonabelian simple group. First consider the case \(t=1\). By Lemma 9 there exists \(m\in N\) with \(\langle x_1, m \rangle =\langle x_2, m \rangle =G\). Assume \(t>1\). We have \(G\le {{\,\mathrm{Aut}\,}}(S)\wr {{\,\mathrm{Sym}\,}}(t)\) and it is not restrictive to assume \(x_1=(h_1,\ldots ,h_t)\sigma \) with \(h_1,\ldots ,h_t\in {{\,\mathrm{Aut}\,}}(S)\), \(\sigma \in {{\,\mathrm{Sym}\,}}(t)\) and \(\sigma (1)=2\). There exists \(u\in {\mathbb {Z}}\) such that \(x_2^u=(h_1^*,\ldots ,h_2^*)\sigma \), with \(h_1^*,\ldots ,h_t^*\in {{\,\mathrm{Aut}\,}}(S)\). Set \(l_1:=x_1\), \(l_2:=x_2^u\), \(k_1:=h_1\), \(k_2:=h_1^*\). Let w be an element of S of order 2. By Lemma 9, there exists \(s\in S\) such that \(\langle w^{k_1},s\rangle =\langle w^{k_2},s\rangle =S\). For \(1\le i\le t\), consider the projection \(\pi _i: N\rightarrow S_i\cong S\). Let \(m=(w,s,1,\ldots ,1)\in N\cong S^t\). For \(i\in \{1,2\}\), the subgroup \(R_i:=\langle m, x_i\rangle \) contains \(\langle m, m^{l_i}\rangle \le N\). Notice that \(S=\langle s, w^{k_i}\rangle \le \pi _2(\langle m, m^{l_i}\rangle )\), hence \(\pi _2(R_i\cap N)\cong S\). Since \(R_iN=G\), we deduce that \(\pi _j(R_i\cap N)\cong S\) for each \(j\in \{1,\ldots ,t\}\). In particular (see for example [4, Proposition 1.1.39]) either \(N\le R_i\) or there exist \(k\in \{1,\ldots ,t\}\) and \(h \in {{\,\mathrm{Aut}\,}}(S)\) such that \(\pi _k(z)=h(\pi _1(z))\) for each \(z\in R_i\cap N\). The second possibility cannot occur, since \(m=(w,s,1,\ldots ,1)\in R_i\cap N\) and s and w are not conjugate in \({{\,\mathrm{Aut}\,}}S\) (\(|w|=2\), while \(|s|\ne 2\), otherwise S would be generated by two involutions). So \(N\le R_i\) and consequently \(R_i=G\). \(\square \)

Proof of Theorem 1

We prove the theorem by induction on the order of G. If can be easily seen that \(x\in V(G)\) if and only if \(x{{\,\mathrm{Frat}\,}}(G)\in V(G/{{\,\mathrm{Frat}\,}}(G))\) and that \(\Delta (G)\) is connected if and only if \(\Delta (G/{{\,\mathrm{Frat}\,}}(G)\) is connected. So if \({{\,\mathrm{Frat}\,}}(G)\ne 1\), the conclusion follows by induction. We may so assume \({{\,\mathrm{Frat}\,}}(G)\ne 1\). Let N be a minimal normal subgroup of G and let \(x, y \in V(G)\). If xN and yN are non-isolated vertices of G/N, then by induction \(xN \sim _G yN\), so it follows from Lemma 14 that \(x\sim _G y\). This means that the set \(\Omega _N\) of the elements \(g\in V(G)\) such that \(gN\in V(G/N)\) is contained in a unique connected component, say \(\Gamma _N\), of \(\Delta (G)\). Assume now \(g\in V(G) \setminus \Omega _N\). If G/N is non-cyclic, then \(gN\in {{\,\mathrm{Frat}\,}}(G/N)\). In particular a minimal generating set of G containing g must contain also an element z such that \(zN\not \in {{\,\mathrm{Frat}\,}}(G/N)\). But then \(z\in \Omega _N\) and, since \(z\in \Gamma _N\) and \(g\sim _G z\), we conclude \(g\in \Gamma _N\). In other words, if G/N is cyclic, then \(\Gamma _N=V(G)\). So we may assume that G/N is cyclic for every minimal normal subgroup N of G.

This implies that one of the following occur:

  1. (1)

    G is cyclic;

  2. (2)

    \(G\cong C_p\times C_p\)

  3. (3)

    G has a unique minimal normal subgroup, say N, and N is not central.

If G is cyclic, then the conclusion follows from Proposition 5. If \(G\cong C_p\times C_p\), then \(\Delta (G)\) is a complete multipartite graph, with \(p+1\) parts of size \(p-1\). So we may assume that the third case occurs. First assume that N is abelian. In this case N has a cyclic complement, \(H=\langle h \rangle \), acting faithfully and irreducibly on N. We have \(\langle n, h \rangle =G\) for every non trivial element n of G, and this implies that there exists a unique connected component \(\Lambda \) of \(\Delta (G)\) containing all the non trivial elements of N. Let now \(g\in G\setminus N\). There is a conjugate \(h^*\) of h in G with \(g\notin \langle h^* \rangle \). If \(1\ne n\in N\), then \(G=\langle n, h^*\rangle = \langle g, h^*\rangle \), so \(g\sim _G h^*\sim _G n\), hence \(g\in \Lambda \). We remain with the case when N is non-abelian. Let \(F/N={{\,\mathrm{Frat}\,}}(G/N)\) and set \(\Sigma _1=F\setminus \{1\}\), \(\Sigma _2=\{g\in G\mid \langle g \rangle N=G\}\), \(\Sigma _3=\{g\in G\mid gN\in V(G/N)\}\) (we have \(\Sigma _3 = \varnothing \) if and only if |G/N| is a prime power). Notice that V(G) is the disjoint union of \(\Sigma _1\), \(\Sigma _2\) and \(\Sigma _3\). By Lemma 10, all the elements of \(\Sigma _2\) belong to the same connected component, say \(\Gamma \), of \(\Delta (G)\). Assume \(\Sigma _3\ne \varnothing \). Fix \(y\in \Sigma _2\) and choose n such that \(G=\langle y, n\rangle \). Let p be a prime divisor of |G/N| and let \(y_1, y_2\) be generators, respectively, of a Sylow p-subgroup and a p-complement of \(\langle y \rangle \). Since \(\{y_1,y_2,n\}\) is a minimal generating set for G, it follows \(y_1, y_2\in \Sigma _3\) and that \(y_1\sim _G y_2\sim _G y \sim _G n\). But we noticed in the first part of this proof that all the elements of \(\Sigma _3=\Omega _N\) belong to the same connected component, and so \(\Sigma _2\cup \Sigma _3 \subseteq \Gamma \). Finally let \(g\in \Sigma _1\) and let X be a minimal generating set of G containing g. Certainly \(X\cap (\Sigma _2\cup \Sigma _3)\ne \varnothing \), so \(g\in \Gamma \). \(\square \)

3 Soluble groups

Let u be a positive integer and G a finite group. In this section we will use the following notations. We will denote by \(\Omega _u(G)\) the set of the minimal generating sets of G of size u, by \(\Gamma _n(G)\) the graph whose vertices are the elements of G and in which \(x_1\) and \(x_2\) are adjacent if and only if there exists \(X\in \Omega _n(G)\) with \(x_1, x_2 \in X\). Moreover we will denote by \(V_n(G)\) the set of the non-isolated vertices of \(\Gamma _u(G)\) and by \(\Delta _u(G)\) the subgraph of \(\Gamma _u(G)\) induces by \(V_u(G)\). Finally we will write \(x_1\sim _{G,u}x_2\) to indicate that \(x_1\) and \(x_2\) belong to the same connected component of \(\Delta _u(G)\).

We will need a series of preliminary results before giving the proof of Theorem 2. The following is immediate.

Lemma 11

Let G be a finite group. Then \(\Delta _u(G)\) is connected if and only if \(\Delta _u(G/{{\,\mathrm{Frat}\,}}(G)\) is connected.

Given a subset X of a finite group G, we will denote by \(d_X(G)\) the smallest cardinality of a set of elements of G generating G together with the elements of X.

Lemma 12

[10, Lemma 2] Let X be a subset of G and N a normal subgroup of G and suppose that \(\langle g_1,\ldots ,g_r, X, N\rangle =G\). If \(r\ge d_X(G)\), we can find \(n_1,\ldots ,n_r\in N\) so that \(\langle g_1n_1,\ldots ,g_rn_r,X\rangle =G\).

Lemma 13

Let N be a normal subgroup of a finite group group G and consider the projection \(\pi :G\rightarrow G/N\). Suppose \(A\in \Omega _u(G/N)\) and \(b \in V_u(G)\) with \(bN\in A\). Then there exists \(B\in \Omega _u(G)\) such that \(b\in B\) and \(A=\pi (B)\).

Proof

Let \(A=\{bN,z_1N,\ldots ,z_{u-1}N\}\) and \(t=d_{\{b\}}(G)\). Since \(b\in V_u(G)\), \(t\le u-1\). By Lemma 12, there exist \(n_1,\ldots ,n_{u-1}\in N\) such that \(\langle b,z_1n_1,\ldots ,z_{u-1}n_{u-1}\rangle =G\). The set \(B:=\{b,z_1n_1,\ldots ,z_{u-1}n_{u-1}\}\) satisfies the requests of the statement. \(\square \)

Lemma 14

Let N be a normal subgroup of a finite group G and let \(x, y \in V_u(G)\). If \(xN, yN \in V_u(G/N)\) and \(xN\sim _{{G/N},u}yN\), then there exists \(n\in N\) such that \(x\sim _{G,u} yn\).

Proof

Since \(xN\sim _{G/N,u}yN\), there exists a sequence \(A_1,\ldots ,A_t\) of elements of \(\Omega _u(G/N)\) such that \(xN\in A_1\), \(yN\in A_t\) and \(A_i\cap A_{i+1}\ne \varnothing \) for \(1\le i\le t-1\). We claim that there exists a sequence \(B_1,\ldots ,B_t\) of minimal generating sets of G such that \(x\in B_1\), \(\pi (B_i)=A_i\) for \(1\le i\le t\) and \(B_i\cap B_{i+1}\ne \varnothing \) for \(1\le i\le t-1\). By Lemma 13, there exists a minimal generating set \(B_1\) of G with \(A_1= \pi (B_1)\) and \(x\in B_1\). Suppose that \(B_1,\ldots ,B_j\) have been constructed for \(j<t\). There exists \(g\in B_j\) such that \(gN\in A_j\cap A_{j+1}\). Again by Lemma 13, there exists a minimal generating set \(B_{j+1}\) of G with \(A_{j+1}=\pi (B_{j+1})\) and \(g\in B_{j+1}\). \(\square \)

Denote by d(G) and m(G), respectively, the smallest and the largest cardinality of a minimal generating set of G. A nice result in universal algebra, due to Tarski and known with the name of Tarski irredundant basis theorem (see for example [6, Theorem 4.4]) implies that, for every positive integer k with \(d(G)\le k\le m(G)\), G contains an independent generating set of cardinality k. The proof of this theorem relies on a clever but elementary counting argument which implies also the following result:

Lemma 15

For every k with \(d(G)\le k<m(G)\) there exists a minimal generating set \(\{g_1,\ldots ,g_k\}\) with the property that there are \(1\le i\le k\) and \(x_1,x_2\) in G such that \(\{g_1,\ldots ,g_{i-1},x_1,x_2,g_{i+1},\ldots ,g_k\}\) is again a minimal generating set of G. Moreover \(x_1,x_2\) can be chosen with the extra property that \(g_i=x_1x_2\).

Recall that for a d-generator finite group G, the swap graph \(\Sigma _d(G)\) is the graph in which the vertices are the ordered generating d-tuples and in which two vertices \((x_1,\ldots ,x_d)\) and \((y_1,\ldots ,y_d)\) are adjacent if and only if they differ only by one entry.

Proposition 16

Let G be a finite soluble group. Then \(\Delta _{d(G)}(G)\) is connected.

Proof

Let \(d=d(G)\). If G is cyclic, then \(\Delta _d(G)\) is a null graph, and there is nothing to prove. Assume \(d\ge 2\) and let \(x, y \in V_d(G)\). Let \(X, Y \in \Omega _d(G)\) with \(x\in X\) and \(y\in Y\). By [12], the swap graph \(\Sigma _d(G)\) is connected, so there exists a path in \(\Sigma _d(G)\) joining X to Y. Notice that if AB are adjacent vertices of \(\Sigma _d(G)\), then there exists two connected components \(\Gamma _A\) and \(\Gamma _B\) of \(\Delta _d(G)\) containing, respectively, A and B. On the other hand \(A\cap B\ne \varnothing \), by the way in which the swap graph is defined. Thus \(\Gamma _A\cap \Gamma _B \ne \varnothing \) and consequently \(\Gamma _A=\Gamma _B\) and all the elements of \(A\cup B\) belong to the same connected component. This implies in particular that if \(A_1=X, A_2,\ldots , A_{t-1},A_t=Y\) is a path joining X and Y, then all the elements of \(\cup _{1\le i\le t}A_i\) belong to the same connected component. \(\square \)

Proof of Theorem 2

We may assume \(d(G)\le u\le m(G)\), otherwise \(V_u(G)\) is empty. If \(u=d(G)\), then the results follows from Proposition 16. So we assume \(u>d(G)\). We prove the statement by induction on |G|. By Lemma 11, we may assume \({{\,\mathrm{Frat}\,}}(G)=1\).

Let N be a minimal normal subgroup of G. Let K be a complement of N in G. We have \(d(K)\le d(G)<u\) and \(m(K)=m(G/N)=m(G)-1\ge u-1\) (see [16, Theorem 2]). By the Tarski irredundant basis theorem, K has a minimal generating set \(\{k_1,\ldots ,k_{u-1}\}\) of size \(u-1\) and \(\{k_1,\ldots ,k_{u-1},m\}\) is a minimal generating set of G for every \(m\ne 1\). This implies that all the non-trivial elements of N belong to the same connected component, say \(\Gamma \), of \(\Delta (G)\).

In order to complete our proof, we are going to show that \(X\cap \Gamma \ne \varnothing \), for every minimal generating set \(X=\{x_1,\ldots ,x_u\}\) of G. For each \(i \in \{1,\ldots ,u\}\), there exists \(k_i\in K\) and \(n_i\in N\) such that \(x_i=k_in_i\). We may order the indices in such a way that \(Y=\{k_1,\ldots ,k_t\}\) is a minimal generating set for K.

We distinguish two cases.

a) \(t < u\). Let \(H=\langle x_1,\ldots , x_t\rangle \). Since \(G=\langle Y\rangle N=HN\) and \(H\ne G\), we deduce that H is a complement for N in G and \(\langle H, x_{t+1}\rangle =G\). In particular \(t=u-1\) and \(\{x_1,\ldots ,x_{t},m\}\in \Omega _u(G)\) for every \(1\ne m\in G\). This implies \(\{x_1,\ldots ,x_{t}\}\subseteq \Gamma \cap X\).

b) \(t=u\). Since \(d(K)\le d(G)<u\) and \(m(K)\ge u\), by Lemma 15 there exists \(\{z_1,\ldots ,z_u\}\in \Omega _u(K)\) with the property that \(\{z_1z_2,\ldots ,z_u\}\in \Omega _{u-1}(K)\). We first want to prove that if \(n\in N\) and \({{\tilde{z}}}:=z_un\in V_n(G)\), then \({\tilde{z}}\in \Gamma \). First suppose that there exists a complement H on N in G containing \({\tilde{z}}\). There exist \(m_1,\ldots ,m_{u-1}\in N\) such that \(z_im_i \in H\) for \(1\le i\le u-1\). This implies

$$\begin{aligned} H=\langle z_1m_1z_2m_2,z_3m_3,\ldots ,z_{u-1}m_{u-1},{\tilde{z}}\rangle , \end{aligned}$$

but then \(\{z_1m_1z_2m_2,z_3m_3,\ldots ,z_{u-1}m_{u-1},{\tilde{z}}, m\}\in \Omega _u(G)\) for every \(1\ne m\in M\). Thus \({\tilde{z}}\) and m are adjacent vertices of \(\Delta _u(G)\) and consequently \({\tilde{z}} \in \Gamma \). Now assume that no complement of N in G contains \({\tilde{z}}\). If \(1\ne m\in N\), then \(\langle z_1z_2,z_3,\ldots ,z_u,m\rangle =G\), hence \(d_{\{z_u\}}(G)\le u-1\). Since \(G=\langle z_1,z_2,\ldots ,z_u,N\rangle \), by Lemma 12 there exist \(m_1,\ldots ,m_{u-1}\in M\) such that \(\langle z_1m_1,\ldots ,z_{u-1}m_{u-1},z_u\rangle = G\). As before, this implies \(z_u\in \Gamma \) and consequently \(\{z_1m_1,\ldots ,z_{u-1}m_{u-1}\}\subseteq \Gamma \). On the other hand, since no complement for N in G contains \({\tilde{z}}\), it must be \(\langle z_1m_1,\ldots ,z_{u-1}m_{u-1},{\tilde{z}}\rangle = G\). So \({\tilde{z}}\) is adjacent to the vertices \(z_1m_1,\ldots ,z_{u-1}m_{u-1}\) of \(\Delta _u(G)\) and consequently \({\tilde{z}} \in \Gamma \). Now we can conclude our proof. Since \(\{x_1N,\ldots x_uN\}, \{z_1N,\ldots z_uN\} \in \Omega _u(G/N)\), we have \(x_1N\), \(z_uN\in V_n(G/N)\), so by induction \(x_1N \sim _{G/N,u} z_uN\). By Lemma 14 there exists \(n\in N\) such that \(x_1\sim _{G,u} z_un\). But we proved before that \(z_un \in \Gamma \), and this implies \(x_1\in \Gamma \cap X\). \(\square \)

4 Planar graphs

Lemma 17

Let N be a normal subgroup of a finite group G. If \(\Gamma (G)\) is planar, then either G/N is cyclic of prime-power order or \(|N|\le 2\).

Proof

Assume that G/N is not a cyclic group of prime-power order. Then \(\Delta (G/N)\) is not a null-graph. In particular there exist x and y in G such that xN and yN are joined by an edge of \(\Gamma (G/N)\). By Lemma 7, the subgraph of \(\Gamma (G)\) induced by \(xN\cup yN\) is isomorphic to the complete bipartite graph \(K_{a,a}\), with \(|a|=N\). If \(\Gamma (G)\) is planar, then \(K_{a,a}\) is planar, and this implies \(a\le 2\). \(\square \)

Proposition 18

\(\Gamma (C_n)\) is planar if and only if one of the following occurs:

  1. (1)

    n is a prime-power.

  2. (2)

    \(n=p\cdot q\), where p and q are distinct primes and \(p\le 3\).

  3. (3)

    \(n=4\cdot q\), where q is an odd prime.

Proof

If \(n=p^a\) is a prime power, then \(\Gamma (C_n)\) is an edgeless graph, and consequently it is planar. Assume that n is not a prime power and let \(p < q\) be the two smaller prime divisors of n. We have that \(C_n\) contains a normal subgroup N such that \(G/N\cong C_{p \cdot q}\), and it follows from Lemma 17 that \(|N|\le 2\). If \(|N|=1\), then \(\Delta (C_n)\cong K_{p-1,q-1}\), and consequently \(\Gamma (C_n)\) is planar if and only if \(p\le 3\). If \(|N|=2\), then \(p=2\) and \(\Delta (G)\cong K_{2,2(q-1)}\), which is a planar graph. \(\square \)

Lemma 19

Let G be a finite group. If G is not cyclic, then there exists a normal subgroup N of G with the property that \(d(G/N)=2\) but G/M is cyclic for every normal subgroup M of G with \(N<M\).

Proof

Let \({\mathcal {M}}\) be the set of the normal subgroups M of G with the property that \(d(G/M)=2\). We claim that if G is not cyclic, then \({\mathcal {M}} \ne \varnothing \). Indeed let

$$\begin{aligned} 1=N_t<\cdots <N_0=G \end{aligned}$$

be a chief series of G and let j be the smallest positive integer with the property that \(G/N_j\) is not cyclic. By [15, Theorem 1.3], \(d(G/N_j)=2\). Once we know that \({\mathcal {M}}\) is not empty, any subgroup in \({\mathcal {M}}\) which is maximal with respect to the inclusion satisfies the requests of the statement. \(\square \)

Proposition 20

Let G be a finite, non-cyclic group. Then \(\Gamma (G)\) is planar if and only if \(G\in \{C_2\times C_2, C_2\times C_4, D_4, Q_8, {{\,\mathrm{Sym}\,}}(3)\}\)

Proof

Let G be a non-cyclic group. Choose a normal subgroup N of G as described in Lemma 19. It follows from Lemma 7 that \(\Gamma (G)\) contains a subgraph isomorphic to \(\Delta (G/N)\). So if \(\Gamma (G)\) is planar, then \(\Gamma (G/N)\) is planar and \(|N|\le 2\). By [18] either \(G/N\cong C_2\times C_2\) or \(G/N\cong {{\,\mathrm{Sym}\,}}(3)\). If \(G/N\cong C_2\times C_2\) then either \(d(G)=m(G)\) and \(G\in \{C_2\times C_2, C_2\times C_4, D_4, Q_8\}\), or \(d(G)=m(G)=3\) and \(G\cong C_2\times C_2\times C_2\). In the last case \(\Delta (G)\cong K_7\) is not planar. In the other cases, \(\Gamma (G)\) coincides with the generating graph of G and it is planar. If \(G/N \cong {{\,\mathrm{Sym}\,}}(3)\), then \(G\cong {{\,\mathrm{Sym}\,}}(3)\), \(G\cong D_6\) or \(G\cong C_3\rtimes C_4\). If \(G\cong S_3\) then \(\Gamma (G)\) coincides with the generating graph and it is planar. If \(G\cong D_6\), then the six non-central involutions induces a complete subgraph, so \(\Gamma (G)\) is not planar. If \(G\cong C_3\rtimes C_4\), then the subset \(A\cup B\), where A is the set of the six elements of order 4 and B is the set of the four elements with order divisible by 3, induces a non planar graph containing an isomorphic copy of \(K_{6,4}\). \(\square \)

5 Examples and questions

The minimal generating sets for \({{\,\mathrm{Sym}\,}}(4)\) are described in [8]. We have that \(d({{\,\mathrm{Sym}\,}}(4))=2\) and \(m({{\,\mathrm{Sym}\,}}(4))=3\) and the three graphs \(\Gamma _2({{\,\mathrm{Sym}\,}}(4))\), \(\Gamma _3({{\,\mathrm{Sym}\,}}(4))\) and \(\Gamma ({{\,\mathrm{Sym}\,}}(4))\) are described in the following tables, where the first column contains a representative x of a conjugacy class of \({{\,\mathrm{Sym}\,}}(4)\), the second column describes the set of the elements of \({{\,\mathrm{Sym}\,}}(4)\) adjacent to x in the graph and the third columns gives the degree of x in the graph. We denote by \(X_i\) the set of i-cycles (for \(2\le i\le 4)\) in \({{\,\mathrm{Sym}\,}}(4)\) and by Y the set of the double transpositions.

\(\Gamma _2({{\,\mathrm{Sym}\,}}(4))\)

(1,2)(2,3)

\(\varnothing \)

0

(1,2)

\(\{(2,3,4)^{\pm 1},(1,3,4)^{\pm 1}, (1,2,3,4)^{\pm 1},(1,2,4,3)^{\pm 1}\}\)

8

(1,2,3)

\(X_4\cup \{(1,4), (2,4), (3,4)\}\)

9

(1,2,3,4)

\(X_3 \cup \{(1,2),(1,4),(2,3),(3,4),(1,3,2,4)^{\pm 1},(1,2,4,3)^{\pm 1}\}\)

16

\(\Gamma _3({{\,\mathrm{Sym}\,}}(4))\)

(1,2)(3,4)

\(X_2\cup X_3\)

14

(1,2)

\(Y\cup \{(1,2,3)^{\pm 1},(1,2,4)^{\pm 1},(1,3),(1,4),(2,3),(2,4),(3,4)\}\)

12

(1,2,3)

\(Y\cup \{(1,2), (1,3), (2,3), (1,2,4)^{\pm 1},(1,3,4)^{\pm 1},(2,3,4)^{\pm 1}\}\)

12

(1,2,3,4)

\(\varnothing \)

0

\(\Gamma ({{\,\mathrm{Sym}\,}}(4))\)

(1,2)(3,4)

\(X_2\cup X_3\)

14

(1,2)

\(Y \cup X_3\cup \{(1,3),(1,4),(2,3),(2,4),(3,4),(1,2,3,4)^{\pm 1},(1,2,4,3)^{\pm 1}\}\)

20

(1,2,3)

\(Y\cup X_2\cup X_4\cup \{(1,2,4)^{\pm 1},(1,3,4)^{\pm 1},(2,3,4)^{\pm 1}\}\)

21

(1,2,3,4)

\(X_3 \cup \{(1,2),(1,4),(2,3),(3,4), (1,3,2,4)^{\pm 1},(1,2,4,3)^{\pm 1}\}\)

16

Denote by \(\omega (\Gamma )\) the clique number of a graph \(\Gamma \). By [20, Theorem 1.1], we have \(\omega (\Gamma _2({{\,\mathrm{Sym}\,}}(4))=4\) and a maximal clique is \(\{(1,2,3,4),(1,2,4,3),(1,3,2,4),(1,2,3)\};\) \(\omega (\Gamma _3({{\,\mathrm{Sym}\,}}(4)))=7\) and a maximal clique is \(X_2\cup \{(1,2)(3,4)\};\) \(\omega (\Gamma ({{\,\mathrm{Sym}\,}}(4)))=11\) and a maximal clique is \(X_2\cup \{(1,2)(3,4), (1,2,3), (1,2,4), (1,3,4), (2,3,4)\}\). However it is not in general true that \(\omega (\Gamma _2({{\,\mathrm{Sym}\,}}(n))\le \omega (\Gamma _{n-1}({{\,\mathrm{Sym}\,}}(n))\). Indeed let n be a sufficiently large odd integer. By [2, Theorem 1], \(\omega (\Gamma _2({{\,\mathrm{Sym}\,}}(n))=2^{n-1}\) while by [8, Theorem 2.1] a non-isolated vertex of \(\omega (\Gamma _{n-1}({{\,\mathrm{Sym}\,}}(n))\) is either a transposition or a 3-cycle or a double transposition, so \(\omega (\Gamma _{n-1}({{\,\mathrm{Sym}\,}}(n))\le \left( {\begin{array}{c}n\\ 2\end{array}}\right) +2\cdot \left( {\begin{array}{c}n\\ 3\end{array}}\right) + 3\cdot \left( {\begin{array}{c}n\\ 4\end{array}}\right) \). The independence number of \(\Gamma _2({{\,\mathrm{Sym}\,}}(4))\) is 12 and a maximal independent set is \(X_3\cup Y \cup \{id\}\). The independence number of \(\Delta _3({{\,\mathrm{Sym}\,}}(4))\) is 8 and a maximal independent set is \(X_4\cup \{(1,2),(1,3,4),(1,4,3),id\}\). The independence number of \(\Delta ({{\,\mathrm{Sym}\,}}(4))\) is 6 and a maximal independent set is \(Y\cup \{(1,2,3,4),(1,4,3,2),id\}\). For \(u\in \{1,2\}\), the degree of the vertex of \(\Gamma _u({{\,\mathrm{Sym}\,}}(4))\) corresponding to the element g is divisible by the order of g. When \(u=2\), this follows from a more general result. Indeed, by [19, Proposition 2.2], if G is a 2-generated group and \(g\in G\), then |g| divides the degree of g in the generating graph of G. However this cannot be generalized to \(\Gamma _u(G)\) for arbitrary values of u. For example, consider the dihedral group \(G=\langle a, b \mid a^6, b^2, (ab)^3\rangle \) of degree 6. Then \(\{a^2,a^3,a^ib\}\in \Omega _3(G)\) for \(0\le i\le 5\) and there are precisely 7 elements adjacent to \(a^2\) in \(\Gamma _3(G)\): \(a^3\) and \(a^ib\) for \(0\le i\le 5\). We propose the following question.

Question 21

Let G be a finite group and \(g\in G\). Does |g| divide the degree of g in \(\Gamma _{d(G)}(G)?\)

For a finite group G, let

$$\begin{aligned} W(G)=\bigcap _{d(G)\le u\le m(G)}V_u(G). \end{aligned}$$

We have seen that \(W({{\,\mathrm{Sym}\,}}(4))=X_2\cup X_3\ne V({{\,\mathrm{Sym}\,}}(4))\). If \(d(G)=m(G)\), then \(V(G)=W(G)\) by definition. One may ask whether the converse is true.

Question 22

Does \(V(G)=W(G)\) imply \(d(G)=m(G)\)?

The answer is positive in the soluble case.

Proposition 23

Let G be a finite soluble group. If \(V(G)=W(G)\), then \(d(G)=m(G)\).

Proof

Let \(d=d(G)\), \(m=m(G)\). Since \(V_u(G/{{\,\mathrm{Frat}\,}}(G))=V_u(G){{\,\mathrm{Frat}\,}}(G)/{{\,\mathrm{Frat}\,}}(G)\), we may assume \({{\,\mathrm{Frat}\,}}(G)=1\). First assume that G is cyclic. Then \(|G|=p_1\ldots p_m\), with \(p_1,\ldots ,p_m\) distinct primes. Notice that \(V_1(G)=\varnothing \), so \(V(G)=W(G)\subseteq V_1(G)\) implies \(V(G)=\varnothing \) and this is possible only if \(m=1\). Now assume that G is not cyclic. By assumption \(V_d(G)=V(G)=G\setminus \{1\}\). This is equivalent to say that \(d_{\{g\}}(G)=d-1\) for any \(1\ne g\in G\). By [9, Corollary 2.20, Theorem 2.21] either G is an elementary abelian p-group or there exist a finite vector space V, a nontrivial irreducible soluble subgroup H of \({{\,\mathrm{Aut}\,}}(V)\) and an integer \(d>d(H)\) such that

$$\begin{aligned} G\cong V^{r(d-2)+1} \rtimes H, \end{aligned}$$

where r is the dimension of V over \({{\,\mathrm{End}\,}}_H(V)\) and H acts in the same way on each of the \(r(d-2)+1\) factors. In the first case \(d=m\), and we are done. In the second case, by [16, Theorem 2], \(m=r(d-2)+1+m(H)\). If \(d=2\), then \(H=\langle h\rangle \) is a cyclic group and \(G=V\rtimes H\). Since H is a maximal subgroup, if \(h\in \Omega _u(G)\), then \(u=2\). On the other hand, by assumption, \(h\in V_m(G)\), and therefore \(m=2\). Assume \(d>2\). This implies \(t=r(d-2)+1\ge 2\). We are going to prove that \(r=1\). If \(r\ne 1\), then there exist \(v_1, v_2\in V\) that are \({{\,\mathrm{End}\,}}_G(V)\)-linearly independent. This implies that the H-submodule W of \(V^t\) generated by \(w=(v_1,v_2,1,\ldots ,1)\) is H-isomorphic to \(V^2\). As a consequence, if \(w\in \Omega _u(G)\), then \(u-1\le m(G/W)=m-2\). But then \(w\notin \Omega _m(G)\), against the assumption \(V_m(G)=V(G)\). So \(r=1\), and this implies that H is isomorphic to a subgroup of the multiplicative group of \({{\,\mathrm{End}\,}}_G(V)\), and consequently it is cyclic. Moreover \(t=d-1\) and \(m(G)=m(H)+d-1\). Let h be a generator for of H. Notice that \(h\notin \Omega _u(G)\) if \(u-1>t=d-1\). Since, by assumption, \(h\in \Omega _u(G)\), it must be \(m-1\le d-1\), and consequently \(m=d\). \(\square \)

The finite groups with \(d(G)=m(G)\) are described in [1]. All the finite groups with this property are soluble. So Question 22 is equivalent to the following.

Question 24

Does there exist an unsoluble group G with \(V(G)=W(G)?\)

Another question that we propose is the following.

Question 25

Let G be a finite non-cyclic group. Is the graph \(\Delta (G)\) Hamiltonian?

Notice that if G is cyclic, then \(\Delta (G)\) is not necessarily Hamiltonian. For example, if \(G\cong C_{2\cdot p}\), with p and odd prime, then \(\Delta (G)\cong K_{1,p-1}\). In the case of \({{\,\mathrm{Sym}\,}}(4)\) the affirmative answer to the previous question follows from the Dirac’s Theorem, stating that an n-vertex graph in which each vertex has degree at least n/2 must have a Hamiltonian cycle. However it is not in general true that any vertex of \(\Delta (G)\) has degree at least |V(G)|/2. Consider for example \(G=\langle a, b \mid a^5, b^4, b^{-1}aba^3\rangle \). The graph \(\Delta (G)\) has 19 vertices, and the degree of \(b^2\) in this graph is 8. In any case, we may use Dirac’s Theorem in the case of finite nilpotent groups.

Theorem 26

If G is a finite non-cyclic nilpotent group, then \(\Delta (G)\) is Hamiltonian.

Proof

Let \(g\in V(G)\) and let \(H=\langle g\rangle {{\,\mathrm{Frat}\,}}(G)\). Let \(n=|V(G)|\) and d the degree of \(g\in \Delta (G)\). Since \(G/{{\,\mathrm{Frat}\,}}(G)\) is a direct product of elementary abelian p-groups, any element of \(G\setminus H\) is adjacent to g in \(\Delta (G)\). Since G is not cyclic, H is a proper subgroup of G, hence \(|G|\ge 2|H|\) and therefore

$$\begin{aligned} d=|G|-|H| \ge \frac{|G|-|{{\,\mathrm{Frat}\,}}(G)|}{2}=\frac{n}{2}. \end{aligned}$$

So the conclusion follows from Dirac’s theorem. \(\square \)

Finally, a question that remains open is whether Theorem 2 remains true is the solubility assumption is removed.

Question 27

Let G be a finite group and \(u\in {\mathbb {N}}\). Is \(\Delta _u(G)\) a connected graph?