Find your next learning step
Stallings’ theorem can reveal one splitting from more than one end. Accessibility asks the next question: if splittable vertex groups are opened again and again, must the process eventually stop in a finite reduced graph of groups?
Open the chapter contents · Topology root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the question, or read from Chapter 1 for the complete argument. The chapters remain open on the page; these grouped routes are an index.
Route 1 · Move from one split to a hierarchy
Route 2 · Work reduction and termination
Route 3 · Read the theorem precisely
Route 4 · Test hypotheses and limits
Route 5 · Transfer and continue
CHAPTER 1 OF 12 · MOVE FROM ONE SPLIT TO A HIERARCHY
1. One splitting is not yet a decomposition theory
Stallings’ theorem says that a finitely generated group with more than one end admits a nontrivial splitting over a finite subgroup. The resulting vertex groups may themselves have more than one end, so the theorem can be applied again. Accessibility asks whether this repeated process can be organised into a finite graph of groups rather than continuing without a uniform bound.
The word is about a hierarchy, not ease of explanation. A group is accessible over finite subgroups when it has a finite graph-of-groups decomposition with finite edge groups and vertex groups that cannot be split further in the relevant way; equivalently in the standard setting, the remaining vertex groups have at most one end.
This turns a local discovery—one separating edge orbit in a Bass–Serre tree—into a global structural question: how complicated can all essential finite-edge splittings become?
CHAPTER 2 OF 12 · MOVE FROM ONE SPLIT TO A HIERARCHY
2. Graphs of groups record all stages at once
An amalgam A ∗C B is represented by two vertex groups joined by an edge group C. An HNN extension is represented by a loop edge. Repeating splittings produces a larger graph whose vertex and edge groups retain how the pieces were attached. Its fundamental group recovers the original group.
The associated Bass–Serre tree unfolds that finite quotient graph. Vertex stabilisers are conjugates of vertex groups and edge stabilisers are conjugates of edge groups. A finite quotient can therefore encode an infinite tree action.
Accessibility does not claim that the Bass–Serre tree is finite. It controls the complexity of an appropriate finite quotient decomposition. Keep those two objects separate whenever “number of edges” is discussed.
CHAPTER 3 OF 12 · WORK REDUCTION AND TERMINATION
3. Work a repeated free-product splitting
Let G = C2 ∗ C3 ∗ C5, the free product of cyclic groups of orders 2, 3 and 5. First write G = (C2 ∗ C3) ∗ C5. This is a splitting over the trivial subgroup. The vertex group C2 ∗ C3 still has more than one end, so split it again as C2 ∗ C3.
The final quotient graph is a tree with three vertex groups C2, C3 and C5 and two trivial edge groups. Each vertex group is finite, hence has zero ends. No remaining vertex needs another Stallings split over a finite subgroup.
| Stage | Vertex groups visible | Essential edge groups |
|---|---|---|
| Start | C2 ∗ C3 ∗ C5 | none recorded |
| First split | C2 ∗ C3, C5 | one trivial group |
| Second split | C2, C3, C5 | two trivial groups |
CHAPTER 4 OF 12 · WORK REDUCTION AND TERMINATION
4. Reduction removes fake complexity
Edge count is meaningless if redundant pieces may be inserted indefinitely. For a non-loop edge, if its edge group maps isomorphically onto an incident vertex group, that edge can be collapsed without changing the fundamental group in the essential way. A graph of groups with no such collapsible edge is reduced.
In the worked example, each trivial edge group is a proper subgroup of the adjacent nontrivial finite vertex groups, so the two edges are essential. By contrast, insert a new trivial vertex in the middle of a trivial edge. One new edge has trivial edge group mapping isomorphically to that trivial vertex group and can be collapsed. Repeating this subdivision can manufacture arbitrarily many edges unless reduction is required.
Dunwoody’s bound concerns reduced splittings. “Reduced” is therefore not cosmetic notation; it protects the complexity measure from artificial inflation.
CHAPTER 5 OF 12 · WORK REDUCTION AND TERMINATION
5. Changed context: a one-ended group does not begin the hierarchy
Replace the free product with G = Z². Its Cayley graph has one end. Stallings’ theorem does not produce a nontrivial splitting of Z² over a finite subgroup, so the finite-edge hierarchy stops immediately with one vertex and no edges.
This does not say Z² has no algebraic decompositions of any kind. It can be described using extensions and it acts on other spaces. The statement is narrower: there is no nontrivial amalgam or HNN splitting over a finite subgroup arising from multiple ends.
The transfer check is to name the allowed edge class. Accessibility over finite groups, over virtually cyclic groups and over other “small” groups are different theories with different theorems and hypotheses.
CHAPTER 6 OF 12 · READ THE THEOREM PRECISELY
6. Dunwoody’s theorem gives a termination bound
Dunwoody proved that finitely presented groups are accessible over finite subgroups. One useful formulation says that, for a fixed finitely presented group G, the number of edges in any reduced graph-of-groups decomposition of G with finite edge groups is bounded in terms of G.
The theorem is stronger than observing that one chosen sequence happened to stop. A complexity bound prevents every reduced finite-edge decomposition from growing without control. Combined with Stallings’ theorem, it permits repeated splitting until vertex groups have at most one end.
The proof is not a simple count of generators. It uses finite presentation to obtain a controlled two-dimensional object and analyses tracks or cuts associated with tree actions. The finite-presentation hypothesis carries real geometric information.
CHAPTER 7 OF 12 · READ THE THEOREM PRECISELY
7. Finite generation alone is not enough
An important limit is easy to lose in summaries: not every finitely generated group is accessible. Dunwoody later constructed an inaccessible finitely generated group. Therefore the 1985 theorem must not be restated with “finitely generated” in place of “finitely presented.”
Finite presentation supplies finitely many relators as well as finitely many generators. Those relations constrain how separating tracks can accumulate in the associated complex. Without comparable control, repeated reduced splittings can evade a uniform finite bound.
| Statement | Safe conclusion | Unsafe extension |
|---|---|---|
| Stallings | more than one end gives one finite-edge split | the full hierarchy is automatically finite |
| Dunwoody accessibility | finitely presented groups have bounded reduced finite-edge complexity | every finitely generated group is accessible |
| Reduced decomposition | redundant collapsible edges are removed | every displayed edge is essential |
CHAPTER 8 OF 12 · READ THE THEOREM PRECISELY
8. Accessibility does not give a unique picture
A group can admit different reduced graph-of-groups decompositions. Accessibility bounds complexity and guarantees a finite endpoint of the splitting process under the theorem’s conditions; it does not make every sequence or quotient graph identical.
Choices include which splittable vertex is opened first and which representative tree action is used. Deformation moves can relate Bass–Serre trees that encode similar elliptic subgroups while looking different as quotient graphs. Stronger canonical decomposition theories, such as suitable JSJ constructions, require additional structure and a specified edge class.
For this reason, compare decompositions by their vertex and edge stabilisers, elliptic subgroups and collapse relations, not only by a drawing. A neat finite graph is a representation, not automatically the canonical answer.
CHAPTER 9 OF 12 · TEST HYPOTHESES AND LIMITS
9. What the theorem does not establish
Accessibility does not say that every vertex group is simple, finite or understood. A one-ended vertex group may have rich internal geometry and may split over infinite subgroups even though no further finite-edge Stallings split applies.
The theorem also does not bound the size of finite edge groups merely by counting edges. Edge-group orders and embeddings carry separate information. Nor does it turn a nonreduced decomposition into a contradiction; it says reduction is required before the relevant complexity is measured.
Finally, a computational search that fails to find another split is not a proof of one-endedness or terminality. The algorithmic question depends on how the group is presented and which decision procedures are available.
CHAPTER 10 OF 12 · TEST HYPOTHESES AND LIMITS
10. A practical graph-of-groups proof checklist
State the group hypothesis: finitely generated, finitely presented or something stronger. Name the permitted edge groups. Verify each proposed amalgam or HNN presentation and identify the corresponding quotient graph. Check injectivity of edge maps.
Reduce the graph by testing every non-loop edge for a collapsible isomorphism. Then inspect vertex groups for ends or further allowed splittings. Keep a ledger of edge orbits rather than edges in the infinite Bass–Serre tree. When invoking accessibility, cite the theorem matching the edge class and finiteness assumptions.
For an evidence check, reconstruct the fundamental group after every collapse or refinement. A move that silently changes a vertex embedding may change the group. A correct diagram should survive algebraic verification, not only visual intuition.
CHAPTER 11 OF 12 · TRANSFER AND CONTINUE
11. Questions readers should ask
Does Stallings’ theorem already prove accessibility?
No. Stallings supplies a split when a finitely generated group has more than one end. Accessibility controls repeated essential splitting and requires an additional termination or bounded-complexity theorem.
Why must the graph of groups be reduced?
Otherwise collapsible vertices and edges can be inserted repeatedly, making edge count arbitrarily large without adding structural information.
Is every finitely generated group accessible?
No. Finitely presented groups are accessible over finite subgroups, but finitely generated inaccessible groups exist.
Are terminal vertex groups necessarily finite?
No. They may be finite or one-ended. They can still have substantial structure and may split over a different class of infinite subgroups.
CHAPTER 12 OF 12 · TRANSFER AND CONTINUE
12. Continue through the verified topology ecosystem
Begin with the Topology root, the Gold Standard of Education and the How X Works hub. Build the prerequisite path through Ends of Spaces, Directional Escape and More Than One Point at Infinity, Ends of Groups, Cayley Graphs and the Zero–One–Two-or-Infinite Alternative and Stallings Splittings, Finite Edge Groups and Bass–Serre Trees.
For neighbouring foundations, continue to the Gold Standard of Graph Theory and the Gold Standard of Algebra. Authoritative technical reading: M. J. Dunwoody’s original paper The Accessibility of Finitely Presented Groups proves the finite-presentation result; the Encyclopedia of Mathematics entry on accessibility explains reduced finite-edge decompositions and tree actions; and Dunwoody’s An Inaccessible Group records the crucial limit beyond finite presentation.
Top · Contents · Previous · How X Works