Structure and automorphisms of pure virtual twin groups

The case n=2 is immediate, and hence we assume n \ge 3. Recall that PVT_n is generated by the elements \gamma( \mu, a), where \mu \in \operatorname{M}_n and a \in \{s_1, \dots, s_{n-1}, \rho_1, \dots, \rho_{n-1}\}. We observe that

in VT_n for elements \alpha, \alpha_j in the subgroup \langle \rho_1, \dots, \rho_{n-1}\rangle. Therefore, we have

and

for each \mu \in \operatorname{M}_n and i=1, 2, \dots, n-1. Let \mathcal{S} \sqcup \mathcal{S}^{-1}= \{\lambda_{i,j}^{\pm 1}~|~ \lambda_{i,j} \in \mathcal{S} \}. We claim that each \gamma( \mu, s_i) lies in \mathcal{S} \sqcup \mathcal{S}^{-1}. For this, we analyse the conjugation action of S_n = \langle \rho_1, \ldots, \rho_{n-1} \rangle on the set \mathcal{S}.

First consider \lambda_{i,i+1} for i=1,2, \dots, n-1.

• (i) If 1 \leq k \leq i-2 or i+2 \leq k \leq n-1, then

  \rho_k \lambda_{i,i+1} \rho_k = \lambda_{i,i+1}.

• (ii) If k= i-1, then

  \begin{aligned} \rho_k \lambda_{i,i+1} \rho_k &= \rho_{i-1} \lambda_{i,i+1} \rho_{i-1}\\ &= \rho_{i-1} s_i \rho_i \rho_{i-1} \\ &= \rho_{i-1} s_i \rho_{i-1}\rho_{i-1}\rho_i \rho_{i-1} \quad (\text{Using } \eqref{3}) \\ &= \rho_{i} s_{i-1} \rho_{i}\rho_{i-1}\rho_i \rho_{i-1} \quad \quad (\text{Using } \eqref{7}) \\ &= \rho_{i} s_{i-1} \rho_{i-1}\rho_{i}\rho_{i-1} \rho_{i-1} \quad (\text{Using } \eqref{5}) \\ &= \rho_{i}(s_{i-1} \rho_{i-1})\rho_{i} \quad \quad \quad \quad (\text{Using } \eqref{3}) \\ &= \lambda_{i-1,i+1}. \end{aligned}

• (iii) If k=i, then

  \rho_k \lambda_{i,i+1} \rho_k = \lambda_{i,i+1}^{-1}.

• (iii) If k=i+1, then

  \rho_k \lambda_{i,i+1} \rho_k = \lambda_{i,i+2}.

Next, we consider \lambda_{i,j} for each 1 \leq i < j \leq n and j \ne i+1.

• (i) If 1 \leq k \leq i-2 or j+1 \leq k \leq n-1, then

  \rho_k \lambda_{i,j}\rho_k = \lambda_{i,j}.

• (ii) For k=i-1, we have \rho_{i-1} \lambda_{i,j} \rho_{i-1} = \lambda_{i-1,j} since

  \begin{aligned} \rho_{i-1} \lambda_{i,j} \rho_{i-1} &= \rho_{i-1} \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \rho_{i-1} \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \rho_{i-1} \lambda_{i, i+1} \rho_{i-1} \rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \qquad (\text{Using } \eqref{4}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \rho_{i-1} s_i \rho_i \rho_{i-1} \rho_{i+1} \dots \rho_{j-2} \rho_{j-1}\\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \rho_i \rho_i \rho_{i-1} s_i \rho_i \rho_{i-1}\rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \quad (\text{Using } \eqref{3}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \rho_i s_{i-1} \rho_i \rho_{i-1} \rho_i \rho_{i-1}\rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad(\text{Using } \eqref{7}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \rho_i s_{i-1} \rho_i \rho_{i} \rho_{i-1} \rho_{i}\rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \quad (\text{Using } \eqref{5}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \rho_i (s_{i-1} \rho_{i-1}) \rho_{i}\rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \qquad (\text{Using } \eqref{3}) \\ &= \lambda_{i-1,j}. \end{aligned}

• (iii) For k=i, we have \rho_{i} \lambda_{i,j} \rho_{i} = \lambda_{i+1,j}, since

  \begin{aligned} \rho_{i} \lambda_{i,j} \rho_{i} &= \rho_{i} \rho_{j-1} \rho_{j-2} \dots \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \dots \rho_{j-2} \rho_{j-1} \rho_{i}\\ &= \rho_{j-1} \rho_{j-2} \dots \rho_i \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \rho_i \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \qquad \qquad \qquad (\text{Using } \eqref{4}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_i \rho_{i+1} s_i \rho_i \rho_{i+1} \rho_i \dots \rho_{j-2} \rho_{j-1}\\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+2} s_{i+1} \rho_i \rho_{i+1} \rho_i \rho_{i+1} \rho_i \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \qquad \qquad (\text{Using } \eqref{7}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+2} s_{i+1} \rho_i \rho_{i} \rho_{i+1} \rho_{i} \rho_i \dots \rho_{j-2} \rho_{j-1} \qquad \qquad \qquad \qquad \qquad \quad (\text{Using } \eqref{5}) \\ &= \rho_{j-1} \rho_{j-2} \dots \rho_{i+2}( s_{i+1} \rho_{i+1}) \rho_{i+2} \dots \rho_{j-2} \rho_{j-1}\qquad \qquad \qquad \qquad \qquad \quad \enspace (\text{Using } \eqref{3}) \\ &= \lambda_{i+1,j}. \end{aligned}

• (iv) If i+1 \leq k \leq j-2, then

  \begin{aligned} \rho_k \lambda_{i,j} \rho_k &= \rho_k \rho_{j-1} \dots \rho_{k+1} \rho_{k} \dots \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \dots \rho_k \rho_{k+1} \dots \rho_{j-1} \rho_k \\ &= \rho_{j-1} \dots \rho_k \rho_{k+1} \rho_{k} \dots \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \dots \rho_k \rho_{k+1} \rho_k \dots \rho_{j-1} \qquad \qquad \qquad \quad (\text{Using } \eqref{4}) \\ &= \rho_{j-1} \dots \rho_{k+1} \rho_{k} \rho_{k+1} \rho_{k-1} \dots \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \dots \rho_{k-1} \rho_{k+1} \rho_{k} \rho_{k+1} \dots \rho_{j-1} \quad (\text{Using } \eqref{5}) \\ &= \rho_{j-1} \dots \rho_{k+1} \rho_{k} \rho_{k-1} \dots \rho_{i+1} \lambda_{i, i+1} \rho_{i+1} \dots \rho_{k-1}\rho_{k} \rho_{k+1} \dots \rho_{j-1} \qquad \qquad\quad (\text{Using } \eqref{4}) \\ &= \lambda_{i,j}. \end{aligned}

• (v) If k= j-1, then

  \rho_k \lambda_{i,j}\rho_k = \lambda_{i,j-1}.

• (vi) If k= j, then

  \rho_k \lambda_{i,j}\rho_k = \lambda_{i,j+1}.

Hence, each generator \gamma( \mu, s_i) lies in the set \mathcal{S} \sqcup \mathcal{S}^{-1}. Conversely, if \lambda_{i,j} \in \mathcal{S} is an arbitrary element, then we see that conjugation by (\rho_{i-1}\rho_{i-2}\ldots \rho_2\rho_1)(\rho_{j-1}\rho_{j-2}\ldots \rho_3\rho_2) maps \lambda_{1,2} to \lambda_{i,j}, whereas conjugation by (\rho_{i-1}\rho_{i-2}\ldots \rho_2\rho_1)(\rho_{j-1}\rho_{j-2}\ldots \rho_3\rho_2\rho_1) maps \lambda_{1,2} to \lambda_{i,j}^{-1}. That is, each \lambda_{i,j}= \mu \lambda_{1, 2} \mu^{-1}=\gamma(\mu, s_1) for some \mu \in \operatorname{M}_n, and hence \mathcal{S} generates the group PVT_n.

Theorem 3.1 already gives a generating set \mathcal{S} for PVT_n. Geometrically, a generator \lambda_{i,j} looks as in Figure Proof.

The defining relations are given by

where \tau is the rewriting process, \mu \in \operatorname{M}_n and r is a defining relation in VT_n. Let us take

Note that, since \gamma(\mu, \rho_i)=1 for all i, no non-trivial relations for PVT_n can be obtained from the relations (?)–(?) of VT_n. We consider the remaining relations one by one.

• (i) First consider the relations s_i^2 = 1, 1 \le i \le n-1, of VT_n. Then we have

  \begin{aligned} \tau(\mu s_i^2 \mu^{-1}) &= \gamma(\rho_{i_1} \dots \rho_{i_k} s_i s_i \rho_{i_k} \dots \rho_{i_1})\\ &= \gamma(1, \rho_{i_1}) \gamma(\overline{\rho_{i_1}}, \rho_{i_2}) \dots \gamma(\overline{\mu}, s_i) \gamma(\overline{\mu s_i}, s_i) \dots \gamma(\overline{\mu s_i s_i \rho_{i_k} \dots \rho_{i_2} }, \rho_{i_1})\\ &= \gamma(\overline{\mu}, s_i) \gamma(\overline{\mu s_i}, s_i)\\ &= \gamma(\mu, s_i) \gamma(\mu \rho_i, s_i)\\ &= (\mu s_i \rho_i \mu^{-1}) (\mu \rho_i s_i \mu^{-1})\\ &= (\mu s_i \rho_i \mu^{-1}) (\mu s_i \rho_i \mu^{-1})^{-1}, \end{aligned}

  which does not yield any non-trivial relation in PVT_n.

• (ii) Next we consider the relations (s_i \rho_j)^2 = 1 for |i-j| \geq 2. Then we have

  \begin{aligned} \tau(\mu s_i \rho_j s_i \rho_j \mu^{-1}) &= \gamma(\overline{\mu}, s_i) \gamma(\overline{\mu s_i}, \rho_j) \gamma(\overline{\mu s_i \rho_j}, s_i) \gamma(\overline{ \mu s_i \rho_j s_i}, \rho_j)\\ &= \gamma(\overline{\mu}, s_i) \gamma(\overline{\mu s_i \rho_j}, s_i)\\ &= \gamma(\mu, s_i) \gamma(\mu \rho_i \rho_j, s_i)\\ &= (\mu s_i \rho_i \mu^{-1}) (\mu \rho_i \rho_j s_i \rho_i \rho_j \rho_i \mu^{-1})\\ &= (\mu s_i \rho_i \mu^{-1}) (\mu \rho_i s_i \mu^{-1})\\ &= (\mu s_i \rho_i \mu^{-1})(\mu s_i \rho_i \mu^{-1})^{-1}, \end{aligned}

  which gives a trivial relation in PVT_n.

• (iii) Now we consider the relations \rho_i s_{i+1} \rho_i \rho_{i+1} s_i \rho_{i+1}=1, where 1 \le i\le n-2. Computing

  \begin{aligned} \tau(\mu \rho_i s_{i+1} \rho_i \rho_{i+1} s_i \rho_{i+1} \mu^{-1})&= \gamma(\overline{\mu \rho_i}, s_{i+1}) \gamma(\overline{ \mu \rho_i s_{i+1} \rho_i \rho_{i+1}}, s_i)\\ &= \gamma(\mu \rho_i, s_{i+1}) \gamma(\mu \rho_i \rho_{i+1} \rho_i \rho_{i+1}, s_i)\\ &= (\mu \rho_i s_{i+1} \rho_{i+1} \rho_i \mu^{-1}) (\mu \rho_{i+1} \rho_i s_i \rho_i \rho_i \rho_{i+1} \mu^{-1})\\ &= (\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1}) (\mu \rho_{i+1} \rho_i s_i \rho_{i+1} \mu^{-1})\\ &= (\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1}) (\mu \rho_{i+1} \rho_i \rho_{i+1} \rho_{i+1} s_i \rho_{i+1} \mu^{-1}) \qquad (\text{Using } \eqref{3}) \\ &= (\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1}) (\mu \rho_{i+1} \rho_i \rho_{i+1} \rho_{i} s_{i+1} \rho_{i} \mu^{-1}) \qquad \quad (\text{Using } \eqref{7}) \\ &= (\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1}) (\mu \rho_{i+1} \rho_{i+1} \rho_{i} \rho_{i+1} s_{i+1} \rho_{i} \mu^{-1}) \qquad (\text{Using } \eqref{7}) \\ &= (\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1}) (\mu \rho_{i} \rho_{i+1} s_{i+1} \rho_{i} \mu^{-1}) \qquad \qquad \quad \enspace (\text{Using } \eqref{3}) \\ &=(\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1})(\mu \rho_i s_{i+1} \rho_{i+1} \rho_i\mu^{-1})^{-1} \end{aligned}

  gives only trivial relations in PVT_n.

• (iv) Finally we consider the relations (s_i s_j)^2 = 1 for |i-j| \geq 2. If \mu = 1, then we get

  \begin{aligned} \tau(s_i s_j s_i s_j) &= \gamma(1, s_i) \gamma(\overline{s_i}, s_j) \gamma(\overline{s_i s_j}, s_i) \gamma(\overline{s_i s_j s_i}, s_j)\\ &= \gamma(1, s_i) \gamma(\rho_i, s_j) \gamma(\rho_i \rho_j, s_i) \gamma(\rho_i \rho_j \rho_i, s_j)\\ &= (s_i \rho_i) (s_j \rho_j) (\rho_i s_i) ( \rho_j s_j)\\ &= \lambda_{i, i+1} \lambda_{j, j+1} \lambda_{i, i+1}^{-1} \lambda_{j, j+1}^{-1}. \end{aligned}

  For \mu \neq 1, we have

  \begin{aligned} \tau(\mu s_i s_j s_i s_j \mu^{-1}) &= \gamma(\overline{\mu}, s_i) \gamma(\overline{\mu s_i}, s_j) \gamma(\overline{\mu s_i s_j}, s_i) \gamma(\overline{ \mu s_i s_j s_i}, s_j)\\ &= \gamma(\mu, s_i) \gamma(\mu \rho_i, s_j) \gamma(\mu \rho_i \rho_j, s_i) \gamma(\mu \rho_i \rho_j \rho_i, s_j)\\ &= (\mu s_i \rho_i \mu^{-1})(\mu s_j \rho_j \mu^{-1}) (\mu \rho_i s_i \mu^{-1}) (\mu \rho_j s_j \mu^{-1})\\ &= (\mu s_i \rho_i \mu^{-1})(\mu s_j \rho_j \mu^{-1}) (\mu s_i \rho_i \mu^{-1})^{-1}(\mu s_j \rho_j \mu^{-1})^{-1}\\ &= (\mu \lambda_{i, i+1} \mu^{-1})(\mu \lambda_{j, j+1} \mu^{-1})(\mu \lambda_{i, i+1} \mu^{-1})^{-1} (\mu \lambda_{j, j+1} \mu^{-1})^{-1}. \end{aligned}

For n \ge 4, we set

If \rho \in S_n and (\lambda_{i,j}, \lambda_{k,l}) \in \mathcal{T}, then Remark 3.2 implies that \rho\lambda_{i,j}\rho^{-1} = \lambda_{i',j'}^{\epsilon} and \rho\lambda_{k,l}\rho^{-1} = \lambda_{k',l'}^{\epsilon'} for some \epsilon, \epsilon'\in \{1, -1\} and distinct integers i',j',k', l' with 1\leq i' < j'\leq n and 1\leq k' < l'\leq n. Thus, there is an induced diagonal action of S_n on \mathcal{T} given as

We claim that this action is transitive. Since | \mathcal{T} | =n(n-1)(n-2)(n-3), it is enough to show that the stabiliser of some element of \mathcal{T} has (n-4)! elements. In fact, the stabiliser of the element (\lambda_{1,2},\; \lambda_{n-1,n}) equals \langle \rho_3, \rho_4, \ldots, \rho_{n-1}\rangle \cap \langle \rho_1, \rho_2, \ldots, \rho_{n-3}\rangle = \langle \rho_3, \rho_4, \ldots, \rho_{n-3}\rangle \cong S_{n-4}, which is of order (n-4)!. Thus, the action is transitive and the defining relations of PVT_n obtained from (?) are precisely of the form

for distinct integers i, j, k, l with 1\leq i < j\leq n and 1\leq k < l\leq n. This completes the proof of the theorem.

The group PVT_n is a right-angled Artin group follows from Theorem 3.3, and irreducibility of PVT_n follows from the fact that the complement of its defining graph is connected.

It is well-known that a right-angled Artin group is linear [HsuWise1999, Corollary 3.6] and that a finitely generated linear group is residually finite [Mal'cev1940]. Thus, PVT_n is linear, and hence residually finite. Since any extension of a residually finite group by a finite group is residually finite, it follows that VT_n is residually finite. The second assertion follows from the fact that every finitely generated residually finite group is Hopfian [Mal'cev1940].

The assertion is immediate for the case n=2. Assume that n \ge 3. Case 1. n\geq 3 and n\neq 4. Recall that the only normal subgroups of S_n are 1, A_n and S_n. Note that the automorphisms \phi(\rho_1), \phi(\rho_2) and \phi(\rho_1\rho_2) are all distinct. Thus, the order of \operatorname{Im}(\phi) is strictly greater than 2, and hence \operatorname{Ker}(\phi) must be trivial. Case 2. n=4. In this case, the normal subgroups of S_4 are 1, S_4, A_4 and

the Klein four group. As in the previous case, \operatorname{Im}(\phi) has more than two elements, and hence \operatorname{Ker}(\phi) is either trivial or K_4. Note that (1,\;2)(3,\;4)=\rho_1\rho_3. Since \phi(\rho_1\rho_3)(\lambda_{1,2}) = \lambda_{1,2}^{-1} \neq 1, we have \rho_1\rho_3\not\in \operatorname{Ker}(\phi). Hence, \operatorname{Ker}(\phi) must be trivial in this case as well.

It is clear that U_2 = PVT_2 \cong \mathbb{Z}. We use the Reidemeister-Schreier method for n \geq 3, for which we take the Schreier system to be PVT_{n-1}. Note that for \mu \in PVT_{n-1} and \lambda_{i,j} \in \mathcal{S}, we have

This implies that U_n is generated by the set

Since PVT_3 \cong F_3, it follows that U_3 is an infinitely generated free group with generators

For n \geq 4, in PVT_n we have relations of the form

where i, j, k, l are distinct integers with i<j and k < l. First, consider the case when none of i, j, k, l is equal to n. Since \gamma(\mu, \lambda_{i,j})=1 for \mu \in PVT_{n-1} and \lambda_{i,j} \in \mathcal{S} with j \neq n, we have

and hence there is no non-trivial relation in this case. Next, we consider the case when exactly one of i, j, k, l is equal to n. Without loss of generality, we can assume that j=n. Applying the rewriting process to the relations \lambda_{i,n}\lambda_{k,l}\lambda_{i,n}^{-1}\lambda_{k,l}^{-1}=1 of PVT_n gives

This gives the relations

in U_n, which simply identifies two generators of U_n. Finally, to prove that there are still infinitely many distinct generators in the set X, we consider the sequence of elements

in X, where \epsilon \in \mathbb{Z}. Notice that \alpha_\epsilon \neq \alpha_{\epsilon'} for \epsilon \neq \epsilon'. Hence, U_n is an infinite rank free group for each n \ge 3.

We use the elementary fact that if G = N\rtimes H is an internal semi-direct product with \operatorname{Z}(H) = 1, then \operatorname{Z}(G) \leq \operatorname{Z}(N). Recall that PVT_2\cong \mathbb{Z} and PVT_3 \cong F_3. Since PVT_4 = U_4 \rtimes PVT_3, we have

By Theorem 4.1, \operatorname{Z}(U_4)=1, and hence \operatorname{Z}(PVT_4) = 1. An easy induction gives \operatorname{Z}(PVT_n) = 1 for all n \geq 4. Since VT_2 \cong \mathbb{Z}_2 * \mathbb{Z}_2, we have \operatorname{Z}(VT_2) = 1. For n\geq 3, since \operatorname{Z}(S_n)=1 and VT_n = PVT_n \rtimes S_n, we have \operatorname{Z}(VT_n) \leq \operatorname{Z}(PVT_n)=1.

Note that the set N_{i,j} can be written as a union of four disjoint subsets as follows

Observe that

and hence | N_{i,j} | = 2n - 4. The second assertion is immediate.

Note that \theta_i is an involution for each 1\leq i\leq n-1. It follows from the construction that [\theta_i,\; \theta_j] = 1 for all |i - j| \ge 2. We now claim that (\theta_i \theta_{i+1})^3 = 1 for all 1\leq i \leq n-2. We verify this for i=1 and other cases will follow similarly. Consider

Note that \theta_1\theta_2 fixes the generators \lambda_{i,j} for all i\geq 4. Thus, we need to verify only for \lambda_{i,j} with 1 \le i\leq 3. We see that

and it can be easily seen that (\theta_1\theta_2)^3=1. Thus, sending \rho_k to \theta_k gives a surjective homomorphism from S_n onto \langle \theta_1, \theta_2, \ldots, \theta_{n-1}\rangle. We claim that this homomorphism is injective as well. Note that the only normal subgroups of S_n are 1, A_n and S_n, n\geq 3 and n\neq 4. We know \rho_1\rho_2\in A_n. From above computation we have \theta_1\theta_2\neq 1. Thus \rho_1\rho_2 does not belong to the kernel and consequently, neither A_n nor S_n can be the kernel. So we are done for the case n\geq 3 and n\neq 4. For n=4, we have an extra normal subgroup of S_4, namely the Klein four-subgroup, K_4= \{1, (1, 2)(3, 4), (1, 3)(2, 4), (1, 4)(2, 3)\}. Like in the previous case, we can see that \theta_1\theta_3\neq 1. Thus \rho_1\rho_3=(1, 2)(3, 4) does not belong to the kernel and consequently K_4 cannot be the kernel. Thus, the kernel of the homomorphism must be trivial and \langle \theta_1, \theta_2, \ldots, \theta_{n-1}\rangle \cong S_n.

We begin by noting that if \phi(\lambda_{i,j})=\lambda_{i,j} for some i, j, then being a graph automorphism, \phi keeps the set N_{i,j} invariant. For 1\leq i \leq n-1, we have

We now proceed to the main part of the proof. Since \phi(\lambda_{1,2}) = \lambda_{1,2}, \phi keeps N_{1, 2}, i.e., the first two columns invariant. As the first column is already fixed pointwise, \phi keeps the second column invariant. Now suppose that \phi(\lambda_{2,3}) = \lambda_{2, j} for some 4\leq j \leq n. Now we have two elements \lambda_{1,3} and \lambda_{2,3} which do not commute, but their images (\lambda_{1,3} and \lambda_{2,j}, j\geq 4 respectively) under \phi commute. This contradicts the hypothesis that \phi is an automorphism. Thus \phi(\lambda_{2,3}) = \lambda_{2, 3}, and similarly we can show that \phi(\lambda_{2,j}) = \lambda_{2, j}, for all 4\leq j\leq n. Now consider the element \lambda_{2,3}. As \phi fixes \lambda_{2,3}, it should keep N_{2, 3} invariant. But \phi already fixes first two columns pointwise. Thus \phi keeps the third column invariant. By repeated use of above arguments we get the desired result.

Since the graph of PVT_n has n(n-1)/2 vertices, it follows that

Thus, we have \operatorname{Aut}_{gr}(PVT_3)\cong S_3. For n \ge 5, it suffices to prove \operatorname{Aut}_{gr}(PVT_n) \leq \langle \theta_1, \theta_2, \ldots, \theta_{n-1}\rangle. Consider an arbitrary automorphism \phi \in \operatorname{Aut}_{gr}(PVT_n). Our plan is to compose \phi with finitely many automorphisms of \langle \theta_1, \theta_2, \ldots, \theta_{n-1}\rangle to obtain the identity automorphism. This would be achieved by invoking Lemma 5.4.

As the automorphism group \langle \theta_1, \theta_2, \ldots, \theta_{n-1}\rangle acts transitively on the generating set \mathcal{S}, there exists \phi_1\in \langle \theta_1, \theta_2, \ldots, \theta_{n-1}\rangle such that \phi_1 \phi (\lambda_{1,2}) = \lambda_{1,2}. Thus, the set N_{1, 2} = \{ \lambda_{i, j} \mid 1\leq i \leq 2,\; 3\leq j \leq n\} should be invariant under \phi_1 \phi. Now set

Note that \phi_2 \phi_1 \phi(\lambda_{1,2}) = \lambda_{1,2} and \phi_2 \phi_1 \phi(\lambda_{1,3}) = \lambda_{1,3}. It follows that \phi_2 \phi_1 \phi keeps the sets N_{1, 2} and lk(\lambda_{1,3}) invariant and thus keeps their intersection also invariant. Note that

We claim that there exists \phi_3 \in \langle \theta_4, \theta_5, \ldots, \theta_{n-1}\rangle such that \phi_3\phi_2 \phi_1 \phi(\lambda_{2,j}) = \lambda_{2, j} for all 4\leq j\leq n. Suppose that \phi_2 \phi_1 \phi(\lambda_{2, 4}) = \lambda_{2, j} for some 4\leq j\leq n. Take \psi_4=\theta_4\theta_5\ldots \theta_{j-1}\in \langle \theta_4, \theta_5, \ldots, \theta_{n-1}\rangle. Note that \psi_4\phi_2 \phi_1 \phi(\lambda_{2, 4}) = \psi_4(\lambda_{2, j}) = \lambda_{2, 4} and \psi_4\phi_2 \phi_1 \phi keeps the set \{\lambda_{2, j} \mid 5\leq j \leq n\} invariant. Now suppose that \phi_2 \phi_1 \phi(\lambda_{2, 5}) = \lambda_{2, j} for some 5\leq j\leq n. Take \psi_5=\theta_5\theta_6\ldots \theta_{j-1}\in \langle \theta_5, \theta_6, \ldots, \theta_{n-1}\rangle \subseteq \langle \theta_4, \theta_5, \ldots, \theta_{n-1}\rangle. We repeat the argument and take \phi_3 = \psi_n\psi_{n-1}\ldots \psi_4 \in \langle \theta_4, \theta_5, \ldots, \theta_{n-1}\rangle. It follows from the choice of \phi_3 that \phi_3\phi_2 \phi_1 \phi(\lambda_{1, 2}) = \lambda_{1, 2} and \phi_3\phi_2 \phi_1 \phi(\lambda_{1,3}) = \lambda_{1, 3}. It now suffices to show that \phi_3\phi_2 \phi_1 \phi(\lambda_{1, j})=\lambda_{1,j} for all j \ge4. If \phi_3\phi_2 \phi_1 \phi(\lambda_{1, j})=\lambda_{2,3}, then there exists k \neq j with k\ge 4 such that \phi_3\phi_2 \phi_1 \phi(\lambda_{1, k})=\lambda_{1,l} for some l \ge 4. But, this leads to a contradiction since non-commuting generators have commuting images. Hence, \phi_3\phi_2 \phi_1 \phi(\lambda_{1, j})=\lambda_{1, l} for some l \ge 4. If l \neq j, then we again arrive at a contradiction as before. Thus, \phi_3\phi_2 \phi_1 \phi fixes the first column and we are done by Lemma 5.4.

Our first claim is that PVT_n does not admit any automorphism of transvection type. Equivalently, if \lambda_{i, j} \neq \lambda_{k,\;l}, then neither \lambda_{i, j} \leq \lambda_{k,\;l} nor \lambda_{k,\;l} \leq \lambda_{i, j}. Suppose that \lambda_{i, j} \neq \lambda_{k,\;l}. That means either \{i, j\} \cap \{k, l \} is empty or a singleton set. Let us first suppose that the intersection is empty. Since n\geq 5, we can choose 1\leq q \leq n such that q\notin \{i, j, k, l\}. Set x = \lambda_{i, q} (if i< q) or x = \lambda_{q, i} (if i> q) and y =\lambda_{k, q} (if k< q) or y = \lambda_{q, k} (if k> q). Observe that x\in lk(\lambda_{k, l}) \setminus st(\lambda_{i, j}) and y\in lk(\lambda_{i, j}) \setminus st(\lambda_{k, l}). Thus, neither \lambda_{k, l} \leq \lambda_{i, j} nor \lambda_{i, j} \leq \lambda_{k, l}. Now we suppose that the intersection is a singleton set. Without loss of generality, we may assume that i=k. Since n \geq 5, choose m \notin \{ i, j, l \} and set x = \lambda_{m,l} (if m< l) or x = \lambda_{l, m} (if m> l) and y =\lambda_{m, j} (if m< j) or y = \lambda_{j, m} (if m> j). Observe that x\in lk(\lambda_{i, j}) \setminus st(\lambda_{k,l}) and y\in lk(\lambda_{k,l}) \setminus st(\lambda_{i,j}). Thus, PVT_n does not admit any automorphism of transvection type for n\geq 5. Our second claim is that \operatorname{Aut}_{pc}(PVT_n)=\operatorname{Inn}(PVT_n). Equivalently, the subgraph \Gamma \setminus \Gamma(st(\lambda_{i, j})) is connected for each \lambda_{i, j}. Recall that the action of \operatorname{Aut}_{gr}(PVT_n) = \langle \theta_1, \theta_2, \ldots , \theta_{n-1}\rangle on \mathcal{S} is transitive: given any generator \lambda_{i, j} \in \mathcal{S}, the automorphism (\theta_{i-1}\theta_{i-2} \cdots \theta_{2}\theta_{1})(\theta_{j-1}\theta_{j-2} \cdots \theta_{3}\theta_{2}) maps \lambda_{1, 2} onto \lambda_{i, j}. Thus, it suffices to prove the claim for \lambda_{1, 2}. Note that the vertex set of \Gamma \setminus \Gamma(st(\lambda_{1, 2})) is \{\lambda_{1, i} , \lambda_{2, j} \mid 3\leq i, j \leq n\}. Let v_1, v_2 be two vertices of \Gamma \setminus \Gamma(st(\lambda_{1, 2})). We find a path joining these two vertices as per the following cases:

1. v_1 = \lambda_{1, i}, v_2=\lambda_{1, j}, i\neq j: Choose an integer k such that 3\leq k\leq n and i\neq k \neq j. This is possible since n\geq 5. We see that there is an edge joining \lambda_{1, i} and \lambda_{2, k} and an edge joining \lambda_{2, k} and \lambda_{1, j}.

2. v_1 = \lambda_{2, i}, v_2=\lambda_{2, j}, i\neq j: This is analogous to the previous case.

3. v_1 = \lambda_{1, i}, v_2=\lambda_{2, j}, i\neq j: Clearly there is an edge joining \lambda_{1, i} and \lambda_{2, j}.

4. v_1 = \lambda_{1, i}, v_2=\lambda_{2, i}: Choose two integers j, k such that 3\leq j, k \leq n and j\neq i \neq k\neq j. We can see that there are edges from \lambda_{1, i} to \lambda_{2, j}, from \lambda_{2, j} to \lambda_{1, k}, and from \lambda_{1, k} to \lambda_{2, i}.

Hence, the subgraph \Gamma \setminus \Gamma(st(\lambda_{1, 2})) is connected. Thus, \operatorname{Aut}_{pc}(PVT_n)=\operatorname{Inn}(PVT_n) for n \ge 5. Finally, by [Laurence1995, Servatius1989], we have \operatorname{Aut}(PVT_n) = \big\langle \operatorname{Aut}_{gr}(PVT_n), \operatorname{Aut}_{inv}(PVT_n), \operatorname{Inn}(PVT_n) \big\rangle.

It follows from the construction of graph automorphisms that \operatorname{Aut}_{gr}(PVT_n) normalises \operatorname{Aut}_{inv}(PVT_n). Further, since \operatorname{Aut}_{gr}(PVT_n) \cap \operatorname{Aut}_{inv}(PVT_n)=1, the short exact sequence (5.7) splits.

Recall from Theorem 5.6 that \operatorname{Aut}(PVT_n) = \langle \operatorname{Aut}_{gr}(PVT_n), \operatorname{Aut}_{inv}(PVT_n), \operatorname{Inn}(PVT_n) \rangle. Note that any automorphism \phi \in \langle \operatorname{Aut}_{gr}(PVT_n), \operatorname{Aut}_{inv}(PVT_n) \rangle keeps the set \mathcal{S}\cup \mathcal{S}^{-1} invariant. Since two distinct elements of \mathcal{S}\cup \mathcal{S}^{-1} are not conjugates of each other in PVT_n, it follows that

This gives the split sequence (5.7). The remaining two assertions are immediate by Lemma 5.2, Theorem 5.5 and Corollary 4.2.

Clearly, PVT_2 does not have R_\infty-property. Since the graphs of PVT_3 and PVT_4 are non-complete graphs on at most 7 vertices, it follows from [DekimpeSenden, Theorem 7.1.1] that these groups have R_\infty-property. For n \ge 5, Theorem 5.6 gives \operatorname{Aut}(PVT_n) = \big\langle \operatorname{Aut}_{gr}(PVT_n), \operatorname{Aut}_{inv}(PVT_n), \operatorname{Inn}(PVT_n) \big\rangle. Since the graph of PVT_n is not complete, the result now follows from [DekimpeSenden, Theorem 3.3.3].

Recall that

Let us set

Notice that K_1, K_2, K_3 act trivially on H_2\ast H_3, H_1\ast H_3 and H_1\ast H_2, respectively. Further, [K_1,\; K_2] = [K_2,\; K_3] = [K_3,\; K_1] = 1 and K_1 \cong K_2 \cong K_3. Thus, it suffices to prove that K_1\cong \operatorname{GL}_2 (\mathbb{Z}). But, this follows by recalling that

and identifying generators of K_1 with matrices in \operatorname{GL}_2 (\mathbb{Z}) as

Notice that \psi_1 and \psi_2 permute the H_i’s non-trivially, and hence \langle \operatorname{Aut}_{inv}(PVT_4), \operatorname{Aut}_{tr}(PVT_4) \rangle \cap \langle \psi_1, \psi_2 \rangle = 1. On the other hand \sigma_1, \sigma_2, \sigma_3 \in \langle \operatorname{Aut}_{inv}(PVT_4), \operatorname{Aut}_{tr}(PVT_4) \rangle, and hence the first assertion follows. For the second assertion notice that \langle \psi_1, \psi_2 \rangle \cong S_3 and \psi_1^2=1=\psi_2^2. It suffices to show that \langle\psi_1, \psi_2 \rangle normalises \langle \operatorname{Aut}_{inv}(PVT_4), \operatorname{Aut}_{tr}(PVT_4) \rangle. A direct computation shows that

and hence \psi_1 normalises \langle \operatorname{Aut}_{inv}(PVT_4), \operatorname{Aut}_{tr}(PVT_4) \rangle. By symmetry, the same assertion holds for \psi_2, and we get the desired result.

Note that \operatorname{Inn}(PVT_4)\leq \operatorname{Aut}_{pc}(PVT_4). Set M=\langle p_{x_1, C_2}, p_{y_1, C_2}, p_{x_2, C_3}, p_{y_2, C_3}, p_{x_3, C_1}, p_{y_3, C_1} \rangle. It suffices to show that \phi^{-1} M \phi \leq \operatorname{Aut}_{pc}(PVT_4) for all \phi \in \langle \operatorname{Aut}_{gr}(PVT_4), \operatorname{Aut}_{inv}(PVT_4), \operatorname{Aut}_{tr}(PVT_4) \rangle. If \phi = \psi_1, then

Thus, \psi_1 normalises \operatorname{Aut}_{pc}(PVT_4). By symmetry, \psi_2 and hence \langle \psi_1, \psi_2 \rangle normalises \operatorname{Aut}_{pc}(PVT_4). By Lemma 5.10, it remains to show that \langle \operatorname{Aut}_{inv}(PVT_4), \operatorname{Aut}_{tr}(PVT_4) \rangle normalises \operatorname{Aut}_{pc}(PVT_4).

If \phi = \iota_{x_1}, then

Thus, \iota_{x_1} normalises \operatorname{Aut}_{pc}(PVT_4). By symmetry, all the other inversions also normalise \operatorname{Aut}_{pc}(PVT_4), and consequently \operatorname{Aut}_{inv}(PVT_4) normalises \operatorname{Aut}_{pc}(PVT_4).

If \phi = \tau_{x_1y_1}, then

Thus, \tau_{x_1y_1} normalises \operatorname{Aut}_{pc}(PVT_4). Similarly, one can show that all other transvections also normalise \operatorname{Aut}_{pc}(PVT_4), which completes the proof of the lemma.

Relations in (1) and (2) follow by direct computations together with [Toinet, Theorem 3.1]. Note that

Setting

we see that

• \operatorname{Aut}_{pc\setminus inn}(PVT_4) \cong \mathbb{Z}^2 \ast \mathbb{Z}^2 \ast \mathbb{Z}^2.

• \operatorname{Aut}_{pc}(PVT_4) =\operatorname{Inn}(PVT_4) \operatorname{Aut}_{pc\setminus inn}(PVT_4).

Consider the surjective homomorphism g: \operatorname{Aut}_{pc}(PVT_4) \to \operatorname{Aut}_{pc\setminus inn}(PVT_4) defined on generators as

Note that \operatorname{Inn}(PVT_4) \subseteq \operatorname{Ker}(g). Let w \in \operatorname{Ker}(g) and write w= xy for some x \in \operatorname{Inn}(PVT_4) and y \in \operatorname{Aut}_{pc\setminus inn}(PVT_4). Then we have

and hence \operatorname{Ker}(g)=\operatorname{Inn}(PVT_4). This implies that

and hence

This completes the proof.

Note that each automorphism in \operatorname{Aut}_{pc}(PVT_4) preserves conjugacy classes of generators. But, the only automorphism in \langle \operatorname{Aut}_{gr}, \operatorname{Aut}_{inv}, \operatorname{Aut}_{tr} \rangle which preserves conjugacy classes of all the generators is the identity automorphism by Lemma 5.10. Hence

and the assertion follows.

Note that the \operatorname{IA} automorphism group of PVT_2 is obviously trivial. Magnus [Magnus1935] gave generators of the group of \operatorname{IA} automorphisms of F_3\cong PVT_3 and showed that it contains non-inner automorphisms. Clearly, \operatorname{Aut}_{pc}(PVT_4) is a subgroup of the group of \operatorname{IA} automorphisms of PVT_4. It follows from (?) that \operatorname{Aut}_{pc}(PVT_4) contains non-inner automorphisms as well. For n \ge 5, a direct computation using the description of \operatorname{Aut}(PVT_n) in Theorem 5.7 shows that the only \operatorname{IA} automorphisms of PVT_n are the inner automorphisms.