Twisted Right-Angled Artin Groups (T-RAAGs)

• Twisted RAAGs are groups defined by mixed graph presentations that allow both right-angled commutation and Klein bottle-type inversion relations.
• A complete rewriting system establishes unique normal forms, providing effective, algorithmic solutions to the word and conjugacy problems.
• Structural insights reveal criteria for torsion, orderability, and subgroup closure, along with embeddings into knot groups and implications for 3-manifolds.

Searching arXiv for papers on twisted right-angled Artin groups and closely related usages. Twisted right-angled Artin groups (T-RAAGs, also TRAAGs) are groups defined by presentations in which the only relations among generators are either right-angled commutation relations or Klein-bottle-type relations. In the now standard formulation, the input is a mixed graph $\Gamma$: each undirected edge imposes commutation, while each directed edge imposes an inversion-type conjugation relation. Ordinary right-angled Artin groups (RAAGs) are recovered when no directed edges are present. Recent work has developed a substantial theory of T-RAAGs, including complete rewriting systems, canonical normal forms, orderability and torsion criteria, subgroup and coherence theorems, embeddings into knot groups, and effective solutions to the word and conjugacy problems (Foniqi, 2024, Blumer et al., 29 Apr 2025).

1. Defining data and basic constructions

A mixed graph is a quintuple $\Gamma=(V,E,D,o,t)$, where $(V,E)$ is a finite simplicial graph, $D\subseteq E$ is the set of directed edges, and $o,t:D\to V$ specify the origin and terminus of each directed edge. Undirected edges are written $[a,b]$, and a directed edge from $a$ to $b$ is written $[a,b\rangle$. The associated T-RAAG is

$$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$

The directed relation is equivalently $\Gamma=(V,E,D,o,t)$0, so a directed edge records conjugation by inversion rather than commutation (Foniqi, 2024).

This definition specializes to the usual RAAG $\Gamma=(V,E,D,o,t)$1 when $\Gamma=(V,E,D,o,t)$2, where $\Gamma=(V,E,D,o,t)$3 is the underlying simplicial graph obtained by forgetting orientations. In this sense T-RAAGs interpolate between RAAGs and groups built from Klein bottle relations. The two-vertex example with one directed edge is the Klein bottle group $\Gamma=(V,E,D,o,t)$4, while the two-vertex example with one undirected edge is $\Gamma=(V,E,D,o,t)$5 (Antolín et al., 2024).

Several structural operations parallel the RAAG case. If $\Gamma=(V,E,D,o,t)$6 is a disjoint union, then $\Gamma=(V,E,D,o,t)$7. There is also a cone construction: given a signature $\Gamma=(V,E,D,o,t)$8, the cone $\Gamma=(V,E,D,o,t)$9 adds a new tip $(V,E)$0 joined to every vertex, directed towards each negative vertex, and satisfies

$(V,E)$1

These operations are central in the structural theory of the Droms class of T-RAAGs (Blumer et al., 29 Apr 2025).

The mixed-graph formalism also introduces a specifically twisted combinatorics. A vertex is negative if it is the terminus of some directed edge, and a sinkhole if every adjacent edge is directed into it. A mixed graph is special if every negative vertex is a sinkhole. This condition is one of the fundamental combinatorial restrictions governing subgroup closure and rigidity phenomena (Blumer et al., 29 Apr 2025).

2. Rewriting systems, normal forms, and the underlying RAAG

A major development in the subject is the existence of a complete rewriting system for T-RAAGs. Starting from the alphabet $(V,E)$2 and a shortlex order, one orients the defining relations into signed rewrite rules. For an undirected edge $(V,E)$3, one rewrites $(V,E)$4; for a directed edge $(V,E)$5, one rewrites $(V,E)$6; and one includes the cancellation rules $(V,E)$7. Knuth–Bendix completion produces a complete rewriting system $(V,E)$8, so every word has a unique irreducible descendant and equivalent words have a common descendant (Foniqi, 2024).

The resulting normal form may be described in syllables. A syllable is a nontrivial power $(V,E)$9, and a reduced representative is a product of syllables in which no further joining is possible after all legal shuffles have been performed. The local shuffle rule can be encoded by maps $D\subseteq E$0 on the star of $D\subseteq E$1, reflecting whether shuffling across a neighbor preserves or flips exponents. Every element admits a reduced representative, and any two reduced representatives are related by a finite sequence of syllable shuffles (Crowe et al., 24 Sep 2025).

These normal forms have several immediate consequences. The word problem is decidable, and reduced syllable decompositions are geodesic. If $D\subseteq E$2 is reduced with $D\subseteq E$3, then the geodesic length is $D\subseteq E$4. At the coarse geometric level, an important phenomenon is that the Cayley graph of a T-RAAG is isomorphic, as an undirected graph, to the Cayley graph of the underlying RAAG defined by $D\subseteq E$5. Consequently, spherical growth and geodesic growth agree with those of the underlying RAAG, and the T-RAAG is quasi-isometric to that RAAG (Foniqi, 2024).

The underlying RAAG also embeds algebraically. The squaring map

$D\subseteq E$6

is injective, and induced subgraphs yield naturally embedded sub-T-RAAGs. These two facts allow obstructions from RAAG theory to be pulled into the twisted setting and are repeatedly used in subgroup, separability, and 3-manifold arguments (Blumer et al., 29 Apr 2025, Foniqi, 2024).

3. Torsion, orderability, subgroups, and coherence

The presence of directed edges sharply changes the algebraic behavior of the group. In abelianization, each origin of a directed edge contributes a $D\subseteq E$7-factor, and one has

$D\subseteq E$8

with $D\subseteq E$9 equal to the number of vertices that occur as origins of directed edges. Thus a T-RAAG with at least one directed edge has $o,t:D\to V$0-torsion in abelianization, even though the group itself may remain torsion-free (Blumer et al., 29 Apr 2025, Foniqi, 2024).

Torsion in the group is controlled by directed cycles. A T-RAAG is torsion-free if and only if its mixed graph contains no oriented cycle supported on a complete subgraph; equivalently, torsion occurs precisely when a clique supports a closed directed cycle (Antolín et al., 2024). Orderability is governed by a stronger criterion: the group is left-orderable if and only if the defining mixed graph contains no oriented cycle at all, and it is bi-orderable if and only if there are no directed edges, that is, precisely in the RAAG case. Finite acyclic mixed graphs therefore define poly-free groups, hence locally indicable and left-orderable (Antolín et al., 2024).

A separate structural axis concerns closure of finitely generated subgroups. The twisted analogue of Droms’s theorem states that every finitely generated subgroup of $o,t:D\to V$1 is again a T-RAAG if and only if $o,t:D\to V$2 is a Droms mixed graph. This means: $o,t:D\to V$3 is special; the underlying simplicial graph $o,t:D\to V$4 contains neither an induced $o,t:D\to V$5 nor an induced $o,t:D\to V$6; and $o,t:D\to V$7 contains no induced configuration $o,t:D\to V$8 consisting of two distinct vertices both pointing to a third vertex. Equivalently, every induced subgraph is either a nontrivial disjoint union or a cone $o,t:D\to V$9 over a smaller induced subgraph with a signature $[a,b]$0 (Blumer et al., 29 Apr 2025).

Coherence has a notably clean description. For a finite mixed graph $[a,b]$1, the group $[a,b]$2 is coherent if and only if the underlying simplicial graph $[a,b]$3 is chordal, exactly paralleling the classical RAAG criterion (Blumer et al., 29 Apr 2025). By contrast, subgroup separability depends only on the underlying simplicial graph through the Droms obstruction set: $[a,b]$4 is subgroup separable if and only if $[a,b]$5 contains neither an induced $[a,b]$6 nor an induced $[a,b]$7. This extends the Metaftsis–Raptis theorem from RAAGs to T-RAAGs and implies decidability of subgroup membership in that LERF subclass via Mal’cev’s theorem (Foniqi, 2024).

Rigidity is subtler than for RAAGs. T-RAAGs need not be rigid: non-isomorphic mixed graphs can define isomorphic groups. In the special/Droms regime, the obstruction is the presence of satellites. For Droms mixed graphs, rigidity is equivalent to the absence of satellites, and this yields a solution to the isomorphism problem for the rigid Droms subclass: $[a,b]$8 when both defining graphs are Droms and satellite-free (Blumer et al., 29 Apr 2025).

4. Embeddings into knot groups and 3-manifold constraints

The interaction between T-RAAGs and knot groups is exceptionally rigid. For a mixed graph with at least one directed edge, embeddability of $[a,b]$9 in a knot group is classified in terms of the JSJ decomposition of the knot exterior and the presence of even-type Seifert fibered pieces. A Seifert piece is of even type if it is either the exterior of a torus knot of type $a$0 or a cable space of type $a$1, so that an exceptional fiber has even index $a$2 (Himeno et al., 2024).

The basic directed building block is the sink star $a$3, a star digraph with center $a$4 and leaves $a$5 directed into the center. Its group has presentation

$a$6

so every leaf acts by inverting the center. The classification theorem states that if $a$7 has at least one directed edge and $a$8 is a nontrivial knot in $a$9, then:

$b$0

Here $b$1 denotes disjoint union, $b$2 is the undirected path, and $b$3 is the undirected star $b$4 (Himeno et al., 2024).

Several sharp corollaries follow. No T-RAAG with a directed edge embeds in a purely hyperbolic knot group. Any triangle in the underlying simplicial graph is forbidden. More strongly, if a T-RAAG embeds in a knot group, its underlying graph must be a forest, and every connected component containing a directed edge must be a sink star. Thus the directed part of an embeddable T-RAAG is completely controlled (Himeno et al., 2024).

The proofs combine induced-subgraph arguments, Pride’s theorem on the subgroup generated by squares, centralizer restrictions in knot groups, malnormality of peripheral torus subgroups in hyperbolic pieces, and explicit Seifert-fibered constructions. For even-type torus knot groups $b$5, embeddings of $b$6 are realized by taking $b$7 and $b$8, producing a free family of leaves that invert the central element $b$9 (Himeno et al., 2024).

5. Algorithmic theory: word problem, conjugacy, and automaticity

The complete rewriting system already gives a canonical solution to the word problem: a word represents the identity if and only if its irreducible normal form is empty (Foniqi, 2024). More delicate is the conjugacy problem. For cyclically reduced elements, conjugacy is characterized not only by syllable shuffling and cyclic permutation, as in the RAAG case, but also by full conjugate-preserving $[a,b\rangle$0-cyclic permutations, which account for the sign flips created by directed edges. This yields an implementable algorithm deciding conjugacy in arbitrary T-RAAGs (Crowe et al., 24 Sep 2025).

The algorithm proceeds by cyclic reduction, enumeration of the finite transformation scheme generated by syllable shuffles, cyclic permutations, and the admissible $[a,b\rangle$1-moves, and then comparison in normal form. Conjugacy preserves ordinary length, syllable length, and support, so the relevant search space is finite. The general decidability theorem is accompanied by a more efficient subcase: when

$[a,b\rangle$2

and $[a,b\rangle$3 is a composition of inversions determined by directed edges incident to a single stable letter, conjugacy reduces to the $[a,b\rangle$4-twisted conjugacy problem in the RAAG $[a,b\rangle$5. In that mapping-torus subcase, the complexity is linear (Crowe et al., 24 Sep 2025).

A second route to conjugacy uses automaticity. T-RAAGs are biautomatic: one constructs a finite-state automaton accepting the shortlex normal form language, proves that language regular, and transfers fellow-traveler properties from the underlying RAAG via the isomorphism of undirected Cayley graphs. Since biautomatic groups have solvable conjugacy problem, this provides an alternative proof of decidability (Crowe et al., 24 Sep 2025).

The broader algorithmic landscape is uneven. The global time complexity of the general conjugacy algorithm remains open. The isomorphism problem is solved only for the rigid Droms subclass, while the full isomorphism problem for arbitrary mixed-graph T-RAAGs remains open. Satellites are the principal known obstruction to rigidity, and explicit pairs of non-isomorphic mixed graphs with isomorphic T-RAAGs show that graph recovery fails in general (Blumer et al., 29 Apr 2025, Foniqi, 2024).

6. Related usages of “twisted RAAG” and finite-extension viewpoints

In adjacent strands of the literature, the phrase “twisted right-angled Artin group” is also used for groups that are not mixed-graph T-RAAGs in the above sense. One usage concerns finite cyclic semidirect products

$[a,b\rangle$6

where $[a,b\rangle$7 is length-preserving. In that setting, length-preserving means geodesic length is preserved on the standard generating set, and such automorphisms are precisely finite compositions of inversions and graph automorphisms. These virtual RAAGs are $[a,b\rangle$8, hence have solvable conjugacy problem; conjugacy in $[a,b\rangle$9 reduces to $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$0-twisted conjugacy in the base RAAG; and the twisted conjugacy problem in RAAGs is solvable for all length-preserving $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$1, with linear time in the inversion-only case. When $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$2 is neither a direct product nor cyclic, the conjugacy growth series of $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$3 is transcendental (Crowe, 2024).

A second usage appears in finite-extension constructions inside outer automorphism groups of RAAGs. For every finite graph $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$4, there exists a graph $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$5 such that

$$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$6

and $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$7 has finite index in $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$8. In a strengthened form, there is a graph $$T(\Gamma)=\Big\langle V \ \Big|\ ab=ba \text{ for } [a,b]\in E\setminus D,\ \ aba=b \text{ for } [a,b\rangle\in D \Big\rangle.$$9 with no nontrivial graph automorphisms such that

$\Gamma=(V,E,D,o,t)$00

where $\Gamma=(V,E,D,o,t)$01. The quotient is generated by inversions, and the paper does not assert that the extension splits. This realizes every RAAG as a finite-index subgroup of the outer automorphism group of another RAAG and motivates a finite-extension notion of “twisted RAAG” (Wiedmer, 2022).

These usages are mathematically distinct. Mixed-graph T-RAAGs are defined by local pairwise relations inside a presentation, whereas virtual RAAGs $\Gamma=(V,E,D,o,t)$02 and the finite-index subgroups $\Gamma=(V,E,D,o,t)$03 are extension-theoretic constructions built from ordinary RAAGs and their automorphisms. The common theme is that each theory introduces inversion-type twisting into an otherwise right-angled framework, but the combinatorics, subgroup theory, and algorithms differ substantially (Crowe, 2024, Wiedmer, 2022).