Projection Complexes in Hyperbolic Geometry

• Projection complexes are graph-theoretic and metric constructions built from spaces with coarse projection data, yielding a hyperbolic quasi-tree structure.
• They employ strong projection axioms, ordered large-projection sets, and standard paths to derive precise metric estimates and quasigeodesic behavior.
• Applications span mapping class groups, Out(Fn), and hierarchically hyperbolic groups, illustrating their dynamical, algebraic, and geometric utility.

Projection complexes are graph-theoretic and metric constructions built from a family of spaces or domains equipped with coarse projection data. In the Bestvina–Bromberg–Fujiwara framework, and in later formulations by Bestvina–Bromberg–Fujiwara–Sisto and subsequent authors, the guiding idea is that large projection to a third domain records a “between-ness” relation, and a graph obtained by excluding such large intermediaries becomes a quasi-tree. This converts projection data into a hyperbolic large-scale model that supports geometric, dynamical, and algebraic applications, including actions of mapping class groups, $\mathrm{Out}(F_n)$, hierarchically hyperbolic groups, and, in a finiteness-free variant, $\Homeo_0(S_g)$ (Bestvina et al., 2017).

1. Formal definition and axiomatic structure

A standard starting point is an indexing set $\mathbf Y$ or $\mathcal Y$, together with projection data. In one formulation, for each $Y\in \mathbf Y$ there is a function

$$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$

satisfying the strong projection axioms: symmetry, triangle inequality, a sharpened Behrstock-type axiom, bounded self-distance, and finiteness of the large-projection set (Bestvina et al., 2017). In the BBFS formulation, one begins with geodesic metric spaces $(Y)$ and projection maps

$$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$

with projection distance

$$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$

subject to projection axioms $(P3)$, $\Homeo_0(S_g)$0, and $\Homeo_0(S_g)$1, and then replaces them by modified projections satisfying the strong projection axioms with constant $\Homeo_0(S_g)$2 and

$\Homeo_0(S_g)$3

(Nairne, 2024).

For $\Homeo_0(S_g)$4, the large-projection set is

$\Homeo_0(S_g)$5

or, in equivalent BBFS notation,

$\Homeo_0(S_g)$6

The projection complex $\Homeo_0(S_g)$7, or $\Homeo_0(S_g)$8, is the graph with vertex set $\Homeo_0(S_g)$9 or $\mathbf Y$0, where $\mathbf Y$1 are adjacent precisely when the corresponding large-projection set is empty (Bestvina et al., 2017, Nairne, 2024).

A decisive combinatorial feature is ordering. For $\mathbf Y$2, the set of large intermediaries between $\mathbf Y$3 and $\mathbf Y$4 is totally ordered; in BBFS notation,

$\mathbf Y$5

This ordered set yields a canonical path

$\mathbf Y$6

called the standard path (Bestvina et al., 2017, Nairne, 2024).

2. Quasi-tree geometry and standard paths

The quasi-tree property is the basic large-scale output of the theory. In the strong-axiom setting, if $\mathbf Y$7, then $\mathbf Y$8 is a quasi-tree, proved via Manning’s bottleneck criterion (Bestvina et al., 2017). Standard paths are quasigeodesics, and the number of large projections controls graph distance. If

$\mathbf Y$9

then for $\mathcal Y$0,

$\mathcal Y$1

(Bestvina et al., 2017). In the BBFS notation one also has

$\mathcal Y$2

where $\mathcal Y$3 (Nairne, 2024).

The sharpened Behrstock axiom is structurally important because it replaces coarse control by exact equality: $\mathcal Y$4 This exactness simplifies ordering arguments and the analysis of standard paths by preventing accumulation of error terms (Bestvina et al., 2017).

A later refinement shows that the quasi-tree geometry can be realized by literal trees. Fix a base vertex $\mathcal Y$5, and define

$\mathcal Y$6

This subgraph $\mathcal Y$7 is a tree, and the inclusion $\mathcal Y$8 is a quasiisometry. The proof uses the subpath property for standard paths and the tripod-like behavior of the three paths $\mathcal Y$9, $Y\in \mathbf Y$0, and $Y\in \mathbf Y$1. A concrete coarse lower bound obtained in the argument is

$Y\in \mathbf Y$2

(Nairne, 2024).

The same simplification extends from the projection complex to the quasi-tree of metric spaces. In the original construction, one takes the disjoint union of the spaces $Y\in \mathbf Y$3 and adds transverse edges of length $Y\in \mathbf Y$4 between projection sets for adjacent vertices of $Y\in \mathbf Y$5. The resulting space $Y\in \mathbf Y$6 satisfies a distance formula, valid for $Y\in \mathbf Y$7,

$Y\in \mathbf Y$8

Keeping only the transverse edges corresponding to the tree $Y\in \mathbf Y$9 produces a tree of metric spaces $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$0, and the inclusion $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$1 is a quasiisometry (Nairne, 2024).

3. Fine and finiteness-free variants

The classical BBF framework assumes a finiteness axiom and a Behrstock inequality adapted to discrete domain systems. For $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$2 acting on collections of actual subsurfaces, these hypotheses can fail in two distinct ways: the Behrstock inequality fails in general for arbitrary subsurfaces if one uses the naive collection, and the finiteness axiom $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$3 fails badly because there are uncountably many small perturbations of a subsurface with the same large projection behavior (Jia et al., 14 Apr 2026).

The finiteness-free response is the fine projection complex. Here the vertices are actual subsurfaces, not isotopy classes, and finiteness is replaced by control modulo a new equivalence relation called velcrotness. For two non-sporadic essential subsurfaces $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$4,

$$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$5

has infinite diameter in both $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$6 and $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$7. Equivalently, $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$8 and $$d_Y\colon \mathbf Y^2\setminus \{(Y,Y)\}\to [0,\infty]$$9 are velcrot iff there exists an essential subsurface

$(Y)$0

such that $(Y)$1 is homotopic to both $(Y)$2 and $(Y)$3 (Jia et al., 14 Apr 2026).

The subsurface projection is defined by

$(Y)$4

with the usual cases depending on whether a curve is contained in $(Y)$5, misses $(Y)$6 essentially, or intersects it essentially. The projection diameter is coarsely bounded: $(Y)$7 The coarse projection distance is

$(Y)$8

and for subsurfaces,

$(Y)$9

A triangle inequality holds: $$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$0 The fine Behrstock inequality states that there exists $$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$1 such that for three essential subsurfaces $$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$2 that pairwise intersect essentially, and where each pair is either overlapping or isotopic, if

$$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$3

then

$$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$4

The finiteness replacement is likewise weaker but sufficient: for fixed $$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$5, the collection of subsurfaces $$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$6 with

$$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$7

is covered by finitely many velcrot classes (Jia et al., 14 Apr 2026).

A hierarchy $$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$8 and modified projection distances

$$\pi_Y : \mathcal Y\setminus\{Y\}\to \{\text{non-empty subsets of }(Y)\},$$9

then recover analogues of symmetry, coarse triangle inequality, monotonicity, order structure, and barrier properties. Fixing $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$0, one defines the fine projection complex $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$1 by joining $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$2 and $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$3 when

$$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$4

This $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$5 is a quasi-tree by the bottleneck criterion, and in the surface setting discussed below it is unbounded (Jia et al., 14 Apr 2026).

4. Projection complexes in major applications

Projection complexes were designed to package projection phenomena already visible in the work of Masur–Minsky on subsurface projections, and they supply a large-scale tree-like model that is used for infinite-dimensional spaces of quasi-morphisms, positive stable commutator length, and asymptotic dimension results (Jia et al., 14 Apr 2026).

In free-group geometry, subfactor projections provide the analogue of subsurface projections for free factors of $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$6. For free factors $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$7 and $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$8 in suitable general position, one defines $$d_Y^\pi(X,Z)=\operatorname{diam}\bigl(\pi_Y(X)\cup \pi_Y(Z)\bigr),$$9, where $(P3)$0 is the splitting complex of $(P3)$1. The projections are well defined up to bounded diameter when $(P3)$2 and $(P3)$3 have the same color or when $(P3)$4. The resulting distances

$(P3)$5

satisfy a Behrstock-type principle and a finite large-projection property, which is enough to invoke the BBF theorem. Applying the construction color by color yields hyperbolic spaces $(P3)$6 and a product

$(P3)$7

on which $(P3)$8 acts isometrically. Theorem 5.1 in that work states that every exponentially growing automorphism acts on this product with positive translation length (Bestvina et al., 2012).

For hierarchically hyperbolic groups, a BBF coloring

$(P3)$9

partitions the domains into finitely many $\Homeo_0(S_g)$00-invariant families of pairwise transverse domains. Each color class yields a BBF hyperbolic space $\Homeo_0(S_g)$01, and there is an equivariant quasimedian quasiisometric embedding

$\Homeo_0(S_g)$02

If each domain hyperbolic space $\Homeo_0(S_g)$03 is quasiisometric to a tree, then the group is quasiisometric to a finite-dimensional CAT(0) cube complex. The same framework yields the Helly property for hierarchically quasiconvex subsets and bounded packing consequences (Hagen et al., 2021).

5. Acylindricity, quotients, and dynamical robustness

Projection complexes are also a natural setting for acylindricity. If a group $\Homeo_0(S_g)$04 acts on $\Homeo_0(S_g)$05 preserving the projection distances, then it acts by isometries on $\Homeo_0(S_g)$06. When $\Homeo_0(S_g)$07 and there exist $\Homeo_0(S_g)$08 such that the common stabilizer of any $\Homeo_0(S_g)$09 distinct elements of $\Homeo_0(S_g)$10 has cardinality at most $\Homeo_0(S_g)$11, the action

$\Homeo_0(S_g)$12

is acylindrical. There is a parallel criterion for the quasi-tree of metric spaces: if $\Homeo_0(S_g)$13, each stabilizer acts acylindrically on its vertex space with uniform constants, and the same common-stabilizer bound holds, then the action on the quasi-tree of metric spaces is acylindrical (Bestvina et al., 2017).

Quotient constructions show that this geometry is robust under certain normal closures. Given a projection complex $\Homeo_0(S_g)$14 with a $\Homeo_0(S_g)$15-action, choose subgroups

$\Homeo_0(S_g)$16

forming an equivariant $\Homeo_0(S_g)$17-spinning family, so that

$\Homeo_0(S_g)$18

for every distinct $\Homeo_0(S_g)$19 and every nontrivial $\Homeo_0(S_g)$20. Let

$\Homeo_0(S_g)$21

For $\Homeo_0(S_g)$22 large enough, $\Homeo_0(S_g)$23 is a free product of conjugates of the $\Homeo_0(S_g)$24. More significantly, the quotient graph $\Homeo_0(S_g)$25 is $\Homeo_0(S_g)$26-hyperbolic, proved by lifting geodesic quadrilaterals from $\Homeo_0(S_g)$27 to geodesic quadrilaterals in $\Homeo_0(S_g)$28 and using a shortening argument based on pivot points and windmills (Clay et al., 2020).

The quotient action preserves non-elementary WPD behavior under suitable hypotheses. If $\Homeo_0(S_g)$29 acts on $\Homeo_0(S_g)$30 with a non-elementary WPD action, then for sufficiently large $\Homeo_0(S_g)$31 the induced action of $\Homeo_0(S_g)$32 on $\Homeo_0(S_g)$33 is again a non-elementary WPD action. By Osin’s criterion, it follows that

$\Homeo_0(S_g)$34

This shows that projection-complex geometry survives large-spin quotients both geometrically and dynamically (Clay et al., 2020).

6. Surface homeomorphisms and current structural perspective

The surface case for $\Homeo_0(S_g)$35, $\Homeo_0(S_g)$36, illustrates both the necessity of refined axioms and the strength of the resulting machinery. Let $\Homeo_0(S_g)$37 be a $\Homeo_0(S_g)$38-invariant collection of essential subsurfaces that are pairwise overlapping or isotopic, and either non-sporadic or homeomorphic to a once-bordered torus. The fine projection complex $\Homeo_0(S_g)$39 built from $\Homeo_0(S_g)$40 is an unbounded quasi-tree (Jia et al., 14 Apr 2026).

Equivariance is expressed by

$\Homeo_0(S_g)$41

so $\Homeo_0(S_g)$42 acts by isometries on $\Homeo_0(S_g)$43. If $\Homeo_0(S_g)$44 is taken from finitely many isotopy classes, each orbit is finite, and the action is cobounded. This yields cobounded isometric actions of $\Homeo_0(S_g)$45 on unbounded quasi-trees (Jia et al., 14 Apr 2026).

The same construction feeds into stable commutator length. For every essential subsurface $\Homeo_0(S_g)$46 that is either non-sporadic or a once-bordered torus, there exists $\Homeo_0(S_g)$47 with $\Homeo_0(S_g)$48 and

$\Homeo_0(S_g)$49

A plausible implication is that projection-complex methods remain effective even when the acting group is much less discrete than a mapping class group, provided the projection theory is adjusted to absorb perturbational ambiguity through a relation such as velcrotness (Jia et al., 14 Apr 2026).

A recurrent misconception is that projection complexes must be literal trees, or that their vertices must be isotopy classes or similarly discrete objects. The recent literature shows that both assumptions are too rigid. On one side, an explicit subtree $\Homeo_0(S_g)$50 can be quasiisometric to the full projection complex, so the quasi-tree structure can often be replaced by an actual tree without changing the large-scale metric type (Nairne, 2024). On the other side, the fine theory demonstrates that vertices can be actual subsurfaces, with finiteness replaced by finite control up to velcrot classes, while still producing unbounded quasi-trees and nontrivial scl phenomena (Jia et al., 14 Apr 2026).

Across these developments, projection complexes function less as a single rigid construction than as a scheme for turning projection data into hyperbolic or quasi-tree geometry. The enduring features are the organization of large projections into ordered intermediaries, the use of standard paths and bottleneck arguments, and the persistence of coarse tree-like behavior under simplification, coloring, quotienting, and finiteness-free adaptation (Bestvina et al., 2017, Nairne, 2024).