---
title: Pure Symmetric Outer Automorphisms
url: https://www.emergentmind.com/topics/pure-symmetric-outer-automorphisms
type: topic
---

# Pure Symmetric Outer Automorphisms

Pure symmetric outer automorphisms are outer automorphisms subject to a purity constraint: they preserve specified generators, free factors, or diagram data up to conjugacy rather than permuting that data freely. The phrase is therefore context-dependent rather than universal. In the literature on semisimple algebraic groups it refers to \(k\)-defined diagram automorphisms acting nontrivially on the Dynkin diagram; in the literature on right-angled Artin groups it refers to basis-conjugating outer automorphisms \(P\Sigma\mathrm{Out}(A_\Gamma)\); in the literature on free products it refers to outer automorphisms relative to a fixed splitting \(G=G_1*\cdots *G_n\) that carry each factor \(G_i\) to a conjugate of itself; and, as a classical benchmark for exceptional outer symmetry, the symmetric groups contribute the unique non-inner automorphism class of \(S_6\) [1006.1298][2503.05527][2507.16662][1412.1855].

## 1. Basic definitions and terminological scope

For any group \(G\), the automorphism group \(\mathrm{Aut}(G)\) contains the normal subgroup of inner automorphisms \(\mathrm{Inn}(G)\), and the outer automorphism group is the quotient
\[
\mathrm{Out}(G)=\mathrm{Aut}(G)/\mathrm{Inn}(G).
\]
This quotient isolates symmetries not induced by conjugation. In the finite-group setting this is the standard notion used to state the exceptional theorem for \(S_6\) [1412.1855].

In the algebraic-group setting, one starts from the exact sequence of group schemes
\[
1\to \mathrm{Inn}(G)\to \mathrm{Aut}(G)\to \mathrm{Out}(G)\to 1,
\]
and for a pinned semisimple group over \(\bar{k}\) one has \(\mathrm{Out}(G)\simeq \mathrm{Aut}(\Delta)\), where \(\Delta\) is the Dynkin diagram. In that language, “pure symmetric outer automorphisms” are precisely the diagram automorphisms represented by \(k\)-points of \(\mathrm{Aut}(\Delta)\) that act nontrivially on \(\Delta\) [1006.1298].

For a right-angled Artin group
\[
A_{\Gamma}=\langle v\in V(\Gamma)\mid [v,w]=1\text{ whenever }\{v,w\}\in E(\Gamma)\rangle,
\]
an automorphism is pure symmetric if each standard generator \(v\) is sent to a conjugate of itself. The resulting groups are
\[
PAut(A_\Gamma)\leq \mathrm{Aut}(A_\Gamma),\qquad POut(A_\Gamma)=PAut(A_\Gamma)/\mathrm{Inn}(A_\Gamma),
\]
and the notation \(\Sigma PAut(A_\Gamma)\), \(\Sigma POut(A_\Gamma)\) is used in the same sense in recent cohomological work [2604.13749].

For a fixed free product splitting \(G=G_1*\cdots *G_n\), a pure symmetric automorphism relative to \(\mathfrak S=(G_1,\dots,G_n)\) is an automorphism \(\psi\) such that \(\psi(G_i)=G_i^{g_i}\) for each \(i\). Its outer class lies in \(\mathrm{Out}_{\mathfrak S}(G)\). Here “pure” means that the factors are not permuted, while “symmetric” is in the McCullough–Miller sense [2507.16662].

## 2. The exceptional outer symmetry of \(S_6\)

Among symmetric groups, the outer automorphism group is trivial except in degree \(6\):
\[
\mathrm{Out}(S_n)\text{ is trivial for }n\neq 6,\qquad \mathrm{Out}(S_6)\simeq C_2.
\]
Equivalently,
\[
\mathrm{Aut}(S_n)\simeq S_n\text{ for }n\neq 6,\qquad \mathrm{Aut}(S_6)\simeq S_6\rtimes C_2.
\]
The elementary proof proceeds by tracking the conjugacy class \(C_1\) of transpositions among the involution classes \(C_j\) of cycle type \(1^{n-2j}2^j\). If an automorphism sends transpositions to transpositions, it is inner; if it sends \(C_1\) to some \(C_j\) with \(j>1\), then the product-order argument forces \(n=6\) and \(j=3\). Thus only \(S_6\) admits a non-inner automorphism, and it must interchange transpositions with triple disjoint transpositions [1412.1855].

The rigidity of the class of transpositions is encoded in the algebraically characterized subsets
\[
S_i=\{(i,j)\in S_n:j\neq i\}\subset C_1.
\]
Each \(S_i\) is a maximal subset of \(C_1\) with the property that any two distinct elements in it do not commute and \(xyx\notin S_i\) for any noncommuting \(x,y\in S_i\). An automorphism preserving \(C_1\) must permute the family \(\{S_i\}\); after conjugation by the corresponding permutation, it fixes every transposition and hence is inner, because transpositions generate \(S_n\) [1412.1855].

The exceptional case is visible numerically. In \(S_6\), the class of transpositions and the class of triple disjoint transpositions both have centralizer size \(48\) and class size \(15\):
\[
|C_{S_6}(\tau)|=2\cdot 4!=48,\qquad |Cl(\tau)|=6!/48=15,
\]
\[
|C_{S_6}(\sigma)|=2^3\cdot 3!=48,\qquad |Cl(\sigma)|=6!/48=15.
\]
The icosahedron model gives an explicit automorphism realizing the swap \(C_1\leftrightarrow C_3\): label the \(6\) antipodal vertex pairs of a regular icosahedron by \(X=\{1,\dots,6\}\), identify the resulting \(12\) labelings modulo rigid motions as \(6\) dual pairs \(A=\{a,b,c,d,e,f\}\), and let permutations of \(X\) act on these dual pairs. After identifying \(A\) with \(X\), one obtains an automorphism of \(S_6\) that sends transpositions to products of three disjoint transpositions. Graph-theoretically, it exchanges the \(15\) edges of \(K_6\) with the \(15\) perfect matchings and swaps the \(6\) stars with the \(6\) factorizations of the edge set into perfect matchings [1412.1855].

A later construction replaces the icosahedron by a complex Hadamard matrix of order \(6\) with third roots of unity and the algebra of split quaternions. In that construction, a subgroup \(Y\simeq S_6\) acts through two inequivalent \(6\)-point permutation representations obtained as permutation projections of conjugate monomial actions, and the resulting automorphism \(\alpha\) sends \((1\,2)\) to \((1\,2)(3\,6)(4\,5)\). It also swaps the conjugacy classes \((3)\leftrightarrow (3,3)\), \((4)\leftrightarrow (4,2)\), and \((6)\leftrightarrow (3,2)\), while fixing the identity and the \(5\)-cycle class, thereby exhibiting again the unique nontrivial element of \(\mathrm{Out}(S_6)\) [1805.01273].

## 3. Diagram symmetries and triality in algebraic groups

For a connected semisimple algebraic group \(G\) over a field \(k\), the outer automorphism problem is controlled by the Dynkin diagram. The map
\[
\alpha\colon \mathrm{Aut}(G)(k)\to \mathrm{Aut}(\Delta)(k)
\]
records the induced action on the based root datum, and the main cohomological criterion states that, for semisimple simply connected \(G\),
\[
\mathrm{im}(\alpha)\subseteq \{\pi\in \mathrm{Aut}(\Delta)(k)\mid \pi(t_G)=t_G\},
\]
where \(t_G\in H^2(k,Z)\) is the Tits class. Under the exactness conditions formulated in the paper, a diagram automorphism \(\omega\in \mathrm{Aut}(\Delta)(k)\) is realized by a \(k\)-automorphism of \(G\) if and only if \(\omega(t_G)=t_G\). In this setting, the pure symmetric outer automorphisms are exactly the nontrivial diagram automorphisms defined over \(k\) [1006.1298].

Type \(D_{2n}\) is the main example. For the simply connected \(\mathrm{Spin}(2n)\) with \(n\) even, the center is
\[
Z\simeq \mu_2\times \mu_2,
\]
and the Tits class is
\[
t_G=([A_+],[A_-])\in H^2(k,\mu_2)\times H^2(k,\mu_2).
\]
The unique nontrivial diagram symmetry of \(D_{2n}\) swaps the two spinor nodes and therefore the two \(\mu_2\)-factors. It is realized over \(k\) if and only if
\[
[A_+]=[A_-]\in \mathrm{Br}(k)[2].
\]
For quasi-split groups, \(t_G=0\), so every diagram automorphism exists over \(k\). For non-archimedean local fields, \(H^1(k,G)=0\) for semisimple simply connected \(G\), and the same Tits-class criterion becomes exact [1006.1298].

Type \(D_4\) is exceptional because \(\mathrm{Out}(D_4)\simeq S_3\), the triality group. This is where symmetric composition algebras enter. For an eight-dimensional composition algebra \(C\) with norm \(n\), one associates two strongly outer automorphisms \(\rho_1^C,\rho_2^C\) of \(\mathbf{PGO}^+(n)\). If \(C=S\) is symmetric, then \(\rho_1^S\) has order \(3\), \(\rho_2^S=(\rho_1^S)^2\), and \(\rho_1^S\rho_2^S=\mathrm{Id}\). More generally, if \(C=S_{f,g}\) is an isotope of a symmetric composition algebra, then
\[
(\rho_1^C,\rho_2^C)=\big(\kappa_{[f]}^{-1}\rho_1^S,\kappa_{[g]}^{-1}\rho_2^S\big).
\]
The paper formulates this as an equivalence of categories
\[
\mathrm{Comp}(n)\simeq \mathrm{Tri}(n),
\]
where \(\mathrm{Tri}(n)\) is the \(PGO(n)\)-action groupoid of trialitarian pairs. In this sense, the symmetric case gives the pure order-\(3\) triality automorphisms, while arbitrary composition algebras correspond to trialitarian pairs modulo weakly inner conjugacy [1504.01278].

## 4. Pure symmetric outer automorphisms of right-angled Artin groups

For a finite simplicial graph \(\Gamma\), the right-angled Artin group \(A_\Gamma\) has one generator for each vertex and one commutation relation for each edge. A pure symmetric automorphism is one that sends each standard generator \(v\) to a conjugate of itself. The group \(PAut(A_\Gamma)\) is generated by partial conjugations, and its outer quotient is
\[
POut(A_\Gamma)=PAut(A_\Gamma)/\mathrm{Inn}(A_\Gamma).
\]
When \(Z(A_\Gamma)=1\), there is a short exact sequence
\[
1\to A_\Gamma\to PAut(A_\Gamma)\to POut(A_\Gamma)\to 1.
\]
In the notation of recent work, \(PAut(A_\Gamma)\) and \(POut(A_\Gamma)\) are the same groups denoted \(\Sigma PAut(A_\Gamma)\) and \(\Sigma POut(A_\Gamma)\) [2604.13749].

A partial conjugation is determined by a vertex \(v\) and a union, or in the standard generating set a connected component, of \(\Gamma-\mathrm{st}(v)\). If \(A\) is such a union, then
\[
C_A^v(w)=
\begin{cases}
vwv^{-1},& w\in A,\\
w,& w\notin A.
\end{cases}
\]
Laurence proved that partial conjugations generate \(PAut(A_\Gamma)\). Koban–Piggott’s presentation organizes the relations in terms of shared, dominant, and subordinate components for nonadjacent pairs \(u,v\): commuting relations occur when the supports are suitably separated or subordinate, and the genuinely nontrivial relations are the SIL-type relations attached to separating intersections of links [2604.13749].

The larger symmetric outer automorphism group \(\Sigma Out(A_\Gamma)\) allows graph symmetries and inversions. It is generated by graph symmetries, inversions \(v\mapsto v^{-1}\), and \(\Gamma\)-Whitehead automorphisms \(\phi(P,m)\) defined by symmetric \(\Gamma\)-partitions. The pure symmetric outer group \(P\Sigma Out(A_\Gamma)\) is a finite-index subgroup of \(\Sigma Out(A_\Gamma)\), and in the language of relative outer automorphism groups one has
\[
P\Sigma Out(A_\Gamma)=\mathrm{Out}(A_\Gamma;\;H^t),\qquad H=\{\langle v\rangle\mid v\in V(\Gamma)\}.
\]
Consequently,
\[
\mathrm{vcd}(\Sigma Out(A_\Gamma))=\mathrm{vcd}(P\Sigma Out(A_\Gamma)).
\]
This places pure symmetric outer automorphisms simultaneously in the RAAG-automorphism and relative-outer-automorphism frameworks [2503.05527].

## 5. BNS invariants, cohomology, geometry, and graded Lie theory for RAAGs

The structure of \(PSO(A_\Gamma)\) is highly sensitive to SIL combinatorics. Day–Wade define, for each vertex \(a\), a support graph \(\Delta_a\) whose vertices are the connected components of \(\Gamma-\mathrm{st}(a)\), and whose edges record dominating–shared pairs across SILs. They prove that \(PSO(A_\Gamma)\) is a RAAG if and only if every support graph \(\Delta_a\) is a forest; if some \(\Delta_a\) contains a loop, then the first homology of the BNS-derived subspace arrangement is nontrivial, \(H_1(\mathcal V_{PSO})\neq 0\), and \(PSO(A_\Gamma)\) is not a RAAG. For the automorphism group rather than the outer quotient, Koban–Piggott show that \(PSA(A_\Gamma)\) is a RAAG if and only if \(\Gamma\) has no separating intersection of links [1508.00622][1311.2232].

The cohomology of the pure symmetric outer group is computed combinatorially from the \(\Gamma\)-Whitehead poset \(Wh_\Gamma\). If \(K_q\) denotes the number of rank-\(q\) essential vertex types, then
\[
H^q(POut(A_\Gamma))\cong \mathbb Z^{K_q}\qquad (q\geq 0),
\]
and \(H^*(POut(A_\Gamma))\) is generated in degree \(1\) by classes \(\bar\gamma_I^j\) dual to canonical outer automorphisms. If \(N_j\) is the number of \(j\)-cliques in \(\Gamma\), then under \(Z(A_\Gamma)=1\),
\[
H^q(PAut(A_\Gamma))\cong \bigoplus_{i+j=q}\mathbb Z^{K_i\cdot N_j},
\]
again with ring generation in degree \(1\). The paper also formulates the Generalized Brownstein–Lee Conjecture and proves it in dimension \(2\) [2604.13749].

The corresponding outer space is now available. There exists a contractible cube complex \(K_\Gamma^{sym}\), the symmetric spine, equipped with a proper and cocompact action of \(\Sigma Out(A_\Gamma)\). More generally, for any finite set \(W\) of conjugacy classes, the subcomplex \(K_W\) is contractible, and the subgroup \(Out_W(A_\Gamma)\) that permutes \(W\) is of type \(VF\). The virtual cohomological dimension satisfies
\[
\mathrm{vcd}(\Sigma Out(A_\Gamma))=\dim(K_\Gamma^{sym})=M^\Sigma(L),
\]
where \(L\) is the set of principal vertices and \(M^\Sigma(L)\) is the maximal size of a compatible family of symmetric \(\Gamma\)-partitions based at vertices of \(L\). Since \(P\Sigma Out(A_\Gamma)\) has finite index, the same virtual cohomological dimension holds for the pure symmetric outer group [2503.05527].

Representation-theoretically, \(PAut(A_\Gamma)\) admits a homomorphism
\[
\Phi:PAut(A_\Gamma)\longrightarrow B\times \prod_{i=1}^k (F_2\times F_2),
\]
where \(B\) is a RAAG and the image is surjective onto each factor. This yields CAT(0) actions in which every partial conjugation has infinite-order image. For connected \(\Gamma\), there is also a RAAG \(N\) with
\[
\mathrm{Inn}(A_\Gamma)\leq N\trianglelefteq PAut(A_\Gamma),
\]
and \(N/\mathrm{Inn}(A_\Gamma)\) is free abelian. If \(\Gamma\) has no SIL, then \(PAut(A_\Gamma)\) is itself a RAAG and \(PSO(A_\Gamma)\) is abelian [1710.01065].

At the level of descending central series, both \(PAut(A_\Gamma)\) and \(POut(A_\Gamma)\) are \(1\)-formal and have explicit quadratic presentations for their graded Lie algebras. For \(POut(A_\Gamma)\), one obtains the graded Lie algebra of \(PAut(A_\Gamma)\) by adding the linear relations \(\sum_L c_L^v=0\) over the components \(L\) of \(\Gamma-\mathrm{st}(v)\). A combinatorial condition
\[
(*)\quad \Gamma\text{ does not contain four pairwise non-adjacent vertices }v_1,v_2,v_3,v_4
\]
lying in four distinct connected components of \(\Gamma-\bigcap_{i=1}^4 lk_\Gamma(v_i)\)
governs Koszulness: if \((*)\) holds, then \(PAut(A_\Gamma)\) and \(POut(A_\Gamma)\) are iterated extensions of RAAGs, hence poly-RAAG and poly-free, and the associated graded Lie algebras are Koszul. By contrast, for \(n\geq 4\), \(PAut(F_n)\) and \(POut(F_n)\) are not poly–finitely generated free [2510.13038].

## 6. Free products and relative Whitehead generation

For a fixed splitting \(G=G_1*\cdots *G_n\) with \(n\geq 3\) and all \(G_i\neq 1\), the pure symmetric outer automorphism group \(\mathrm{Out}_{\mathfrak S}(G)\) consists of those outer automorphisms that have a representative \(\psi\) with \(\psi(G_i)=G_i^{g_i}\) for every \(i\). Two natural families of generators appear. A factor automorphism relative to \(\mathfrak S\) restricts to an automorphism of each factor \(G_i\), so factor outer automorphisms form a subgroup isomorphic to \(\prod_i \mathrm{Aut}(G_i)\). A Whitehead automorphism relative to \(\mathfrak S\) is determined by an operating factor \(G_i\), an element \(x\in G_i\), and a subset \(Y\subset \{G_1,\dots,G_n\}-\{G_i\}\); it conjugates each factor in \(Y\) by \(x\) and fixes the others. The operating factor is well-defined at the outer level [2507.16662].

The main theorem states that every pure symmetric outer automorphism of the splitting can be written as a product of factor outer automorphisms relative to \(\mathfrak S\) and Whitehead outer automorphisms relative to \(\mathfrak S\). The proof is geometric. One constructs an Outer Space \(\mathcal O_n\) from \(\mathfrak S\)-labelled \((n,k)\)-trees and then a “nice” \(1\)-dimensional subcomplex \(\mathcal S_n\) spanned by \(\alpha\)-graph classes and \(A\)-graph classes. The star joining the basepoint \([\mathfrak a]\) to the vertices \([A^{(i)}]\) is a strict fundamental domain for the action of \(\mathrm{Out}_{\mathfrak S}(G)\) on \(\mathcal S_n\). A volume-decreasing argument in the Bass–Serre tree produces an \(\alpha\)–\(A\)–\(\alpha\) path from any \(\alpha\)-vertex to \([\mathfrak a]\), proving that \(\mathcal S_n\) is path-connected and enabling an inductive factorization of any element of \(\mathrm{Out}_{\mathfrak S}(G)\) into the stated generators [2507.16662].

Under the additional hypothesis that the factors are non-trivial, not infinite cyclic, freely indecomposable, and pairwise non-isomorphic, the splitting is Grushko and every automorphism of \(G\) is pure symmetric. In that case,
\[
\mathrm{Aut}_{\mathfrak S}(G)=\mathrm{Aut}(G),\qquad \mathrm{Out}_{\mathfrak S}(G)=\mathrm{Out}(G),
\]
so the same theorem becomes a generation theorem for the full outer automorphism group. This relative free-product picture is the direct analogue of Whitehead-type generation in \(Out(F_n)\), with factor automorphisms replacing basis permutations and relative Whitehead automorphisms replacing basis-conjugating moves [2507.16662].

Source: https://www.emergentmind.com/topics/pure-symmetric-outer-automorphisms