---
title: 'Core Limit Set: Asymptotic Dynamics'
url: https://www.emergentmind.com/topics/core-limit-set
type: topic
---

# Core Limit Set: Asymptotic Dynamics

In contemporary mathematics, the term *limit set* denotes a family of asymptotic objects rather than a single invariant. Depending on context, it may mean the accumulation set of a discrete group orbit on a geometric boundary, the minimal non-empty closed invariant subset for an action on a compactification, or the smallest closed set with a comeager basin of attraction in topological dynamics. The examples considered here arise in higher-rank symmetric spaces and Anosov representations, Kleinian groups, mapping class group actions on spaces of measured foliations, and cellular automata [1212.0623], [1106.4409], [1105.2980], [2110.00037], [1808.08716].

## 1. Formal variants of the notion

There is no context-free definition of a limit set. What remains common is that the set records asymptotic orbit behavior after passage to a boundary, compactification, or attractor-like closure.

| Setting | Ambient space | Definition in the supplied sources |
|---|---|---|
| Discrete subgroup \(\Gamma \subset G\) of a semisimple Lie group | \(\partial X\) or \(G/P\) | \(L_\Gamma = \overline{\Gamma \cdot o} \cap \partial X\), and \(\Lambda_\Gamma = \overline{\{\text{attracting fixed points of } g\in\Gamma \text{ in } G/P\}}\) |
| Kleinian group \(G\) | \(S^n\) | \(\Lambda(G) = \overline{G(z)} \cap S^n\) |
| Handlebody group \(G_H\) | \(\mathrm{PMF}(S)\) | \(A(G_H)\) is the smallest non-empty closed invariant subset for the action of \(G_H\) on \(\mathrm{PMF}(S)\) |
| Topological dynamical system \((F_t)\) | compact metric space \(X\) | \(\widetilde{\omega}_F = \bigcap\{ C \subseteq X : C \text{ closed and } \mathrm{AN}[F]{C} \text{ comeager}\}\) |

These definitions come with distinct notions of typicality and approach. In higher-rank symmetric spaces one distinguishes **radial** limit points, approachable by an orbit sequence staying within bounded distance of a Weyl chamber, from **horospherical** limit points, for which every neighborhood contains a tail of the orbit [1212.0623]. In cellular automata, by contrast, the generic limit set is explicitly a Baire-category notion: it is the smallest closed set whose realm of attraction is comeager, and its own realm is comeager [1808.08716].

A recurring source of confusion is the identification of geometric and dynamical notions. The supplied materials treat them separately: for non-orientable mapping class groups, the **geometric limit set** is defined as \(\bigcup_{x\in \teich(\no_g)} \overline{\mcg(\no_g)\cdot x} \cap \pmf(\no_g)\), while the **dynamical limit set** is the minimal closed \(\mcg(\no_g)\)-invariant subset of \(\pmf(\no_g)\) [2110.00037].

## 2. Higher-rank symmetric spaces and Anosov representations

For a symmetric space \(X\) associated to a semisimple Lie group \(G\), the geometric boundary \(\partial X\) and the Furstenberg boundary \(G/P\) carry complementary asymptotic data. The paper on Anosov representations studies both and gives explicit descriptions when the discrete group arises from strictly convex real projective geometry or, more generally, from a \(P\)-Anosov representation of a word hyperbolic group [1212.0623].

In the strictly convex projective surface case, a compact surface \(S\) is written as \(S=\Omega/\Gamma\) with \(\Omega \subset \mathbb{RP}^2\) strictly convex and \(\Gamma \subset \mathrm{SL}(3,\mathbb R)\) discrete and properly discontinuous; \(\Gamma\) is Zariski dense unless \(\Omega\) is an ellipse. Attracting fixed points of elements of \(\Gamma\) correspond to points of \(\partial \Omega\), and the unique tangent flag at such a point determines a Weyl chamber in the boundary of \(X\). This boundary-flag correspondence is the geometric mechanism behind the topology of the limit set.

The main topological statement is that the limit set \(\Lambda_\Gamma\) in the Furstenberg boundary is homeomorphic to \(S^1\), whereas the geometric limit set \(L_\Gamma\) in \(\partial X\) is homeomorphic to the cylinder \(S^1 \times I\), where \(I\) is an interval of directions in the limit cone. More explicitly, for each \(p \in \partial \Omega\), the intersection \(W_p \cap L_\Gamma\) is an interval inside the corresponding Weyl chamber, and as \(p\) varies over \(S^1\), these intervals assemble into the cylinder [1212.0623].

The decomposition of limit points inside a chamber is especially rigid. The subset of radial limit points in \(L_\Gamma\) is precisely one point in each Weyl chamber at infinity whose limit set is nonempty; every remaining point in the associated interval is horospherical; and the set of radial limit points is dense in \(L_\Gamma\). In the Zariski-dense setting, the chamberwise limit set is identified with the set of directions of the limit cone \(\mathcal{L}_\Gamma\), where
\[
\mathcal{L}_\Gamma = \overline{ \{ \log \lambda(g) : g \in \Gamma \} }
\]
and \(\lambda(g)\) is the hyperbolic component in the Jordan decomposition \(g=ehu\) [1212.0623].

The generalization to Anosov representations is expressed as
\[
L_{\rho(\Gamma)} \cong \partial \Gamma \times \partial \mathcal{L}_{\rho(\Gamma)}.
\]
Here \(\Gamma\) is a word hyperbolic group, \(\rho:\Gamma \to G\) is Zariski-dense, discrete, and \(P\)-Anosov for \(P\) a minimal parabolic subgroup, and the geometric limit set becomes a bundle over the Gromov boundary with fiber the directions of the limit cone. The uniqueness of the radial point in each nonempty Weyl chamber persists in this general setting. The proofs use the orbit map as a well-displacing quasi-isometric embedding together with the Morse lemma and hyperbolicity to control convergence and rule out multiplicity of radial points [1212.0623].

A significant deformation phenomenon also appears. As a Fuchsian representation is deformed to a convex real projective one, the Furstenberg limit set remains a circle, whereas the geometric limit set grows from a circle to a cylinder; nevertheless, the number of radial limit points per Weyl chamber remains one. This gives a higher-rank form of topological stability for the boundary dynamics [1212.0623].

## 3. Kleinian groups, fractal dimension, and convex core entropy

For a Kleinian group \(G\), the limit set \(\Lambda(G)\) sits on the sphere at infinity \(S^n\) of hyperbolic space \(\mathbb{B}^{n+1}\). The paper of Falk and Matsuzaki studies three numerical invariants attached to \(\Lambda(G)\): the critical exponent \(\delta(G)\), the Hausdorff dimension \(\dim_H(\Lambda(G))\), and the convex core entropy \(h_c(\Lambda(G))\) [1106.4409].

The critical exponent is defined by
\[
\delta(G)=\inf \left\{ s > 0: \sum_{g \in G} e^{-s d(z, g(z))} < \infty \right\}
      = \limsup_{R \to \infty} \frac{1}{R} \log \#(B_R(z) \cap G(z)),
\]
while the convex core entropy of a closed set \(A \subset S^n\) is
\[
h_c(A)=\limsup_{R \to \infty} \frac{\log \mathrm{vol}(B_R(z) \cap H_\epsilon(A))}{R}.
\]
A key structural fact is
\[
h_c(A)=\dim_B(A)
\]
for any closed \(A \subset S^n\), where \(\dim_B\) denotes upper box-counting dimension [1106.4409].

For any non-elementary Kleinian group,
\[
\delta(G) \leq \dim_H(\Lambda(G)) \leq h_c(\Lambda(G))=\dim_B(\Lambda(G)).
\]
This places the classical orbit-growth invariant, the fractal dimension of the limit set, and the large-scale growth of the convex hull into a single inequality chain. The convex core \(C(M_G)=H(\Lambda(G))/G\) is the geometric intermediary: the entropy is computed through growth in an \(\epsilon\)-neighborhood of the convex hull [1106.4409].

Several equality regimes are identified. If \(G\) is non-elementary and geometrically finite, then
\[
\delta(G)=\dim_H(\Lambda(G))=\dim_B(\Lambda(G))=h_c(\Lambda(G)).
\]
The same equality is stated for convex cocompact groups and their non-trivial normal subgroups, and also for finitely generated, analytically finite Kleinian groups in \(\mathbb{B}^3\) whose limit set has zero \(2\)-dimensional Lebesgue measure. More generally, if \(H(\Lambda(G))\) contains a uniformly distributed set of bounded type, then \(\dim_H(\Lambda(G))=h_c(\Lambda(G))=\dim_B(\Lambda(G))\) [1106.4409].

The inequalities can also be strict. If \(N\) is a non-trivial normal subgroup of a convex cocompact Kleinian group \(G\) and the quotient \(G/N\) is non-amenable, then
\[
\delta(N) < \dim_H(\Lambda(N)) = h_c(\Lambda(N)).
\]
For \(l\)-tight or weakly \(l\)-tight convex hulls, one has \(h_c(\Lambda(G))<n\), and the supplied summary states that often \(\dim_H(\Lambda(G))<n\) as well. In infinite-type surface settings, the amenability of the pants decomposition graph governs possible gaps: non-amenability forces \(\delta(G)<1\), while uniformly strongly amenable cases constrain the invariants to be either all \(1\) or all \(<1\) [1106.4409].

This framework shows that a limit set is not merely a topological boundary accumulation set. Its geometric realization inside the convex hull controls entropy, and its combinatorial realization through amenability properties controls whether orbit-growth, Hausdorff dimension, and box dimension coincide or separate.

## 4. Mapping class groups and measured foliation boundaries

For mapping class groups, the relevant compactification is typically the space of projective measured foliations. In the handlebody case, let \(S\) be an orientable surface of genus \(g\), \(H\) a handlebody with boundary \(S\), and \(G_H \subset \mathrm{Mod}(S)\) the subgroup consisting of mapping classes that extend over \(H\). The limit set is
\[
A(G_H)=\text{the smallest non-empty closed invariant subset for the action of }G_H\text{ on }\mathrm{PMF}(S),
\]
and the main theorem states that \(\ell(A(G_H))=0\) with respect to the natural Lebesgue measure class \(\ell\) on \(\mathrm{PMF}(S)\) [1105.2980].

Masur’s description enters through the set \(\mathscr{B}\) of essential simple closed curves on \(S\) that bound disks in \(H\), together with cut systems \(\mathcal C=\{C_1,\dots,C_g\}\). After passing to \(A(G_H)\), the limit set consists of foliations in \(\mathrm{PMF}(S)\) that, for every cut system, either avoid all \(C_i\) or have returning arcs. Thus the set is dynamically natural and closed invariant, yet measure-theoretically negligible [1105.2980].

The proof history is itself part of the subject. Kerckhoff’s argument used train track charts and iterative splitting, together with a uniform distortion claim for splitting sequences. The later note repairs a gap by restricting to complete non-classical exchanges and using the result that almost every foliation has Rauzy expansions that are \(C\)-uniformly distorted infinitely often. In those stages, Jacobian ratios satisfy
\[
\frac{1}{C} < \frac{\mathscr{J}_Q(x)}{\mathscr{J}_Q(y)} < C,
\]
which restores control of relative measure through successive expansions and yields geometric decay of the measure of the returning-arcs set [1105.2980].

For compact non-orientable surfaces \(\no_g\), \(g\geq 4\), the situation is different. The limit set of \(\mcg(\no_g)\), defined as the closure of stable and unstable foliations of pseudo-Anosov elements, satisfies
\[
{\rm UE}(\no_g)\subset \Lambda(\mcg(\no_g)) \subset \pmf^+(\no_g),
\]
where \(\pmf^+(\no_g)\) consists of foliations without one-sided leaves and \(\pmf^-(\no_g)\) consists of foliations with at least one one-sided compact leaf. Since \(\pmf^-(\no_g)\) is open, dense, and full measure in \(\pmf(\no_g)\), the complement of the limit set is open, dense, and full measure [2110.00037].

The lower inclusion is also explicit: if \(\lambda\in\pmf^+(\no_g)\) and every minimal component \(\lambda_j\) is periodic, or ergodic and orientable, or uniquely ergodic, then \(\lambda\) lies in the limit set. Moreover, if a component is minimal and non-uniquely ergodic, there exists another foliation supported on the same topological foliation that belongs to the limit set. These statements establish large parts of Gendulphe’s conjecture that the limit set equals \(\pmf^+(\no_g)\); the supplied summary notes that the full conjecture was subsequently proved elsewhere [2110.00037].

The non-orientable setting also breaks a tempting geometric analogy. The set \(\systole(\no_g)\), consisting of points of Teichmüller space where no one-sided curve is shorter than \(\varepsilon>0\), is a natural candidate for a convex-core analogue. However, for \(g\geq 8\), any \(\varepsilon>0\), and any \(D>0\), there exists a Teichmüller geodesic segment with endpoints in \(\systole(\no_g)\) but with an interior point at distance \(>D\) from \(\systole(\no_g)\). Hence \(\systole(\no_g)\) is not quasi-convex, in marked contrast with convex cores of geometrically finite hyperbolic manifolds [2110.00037].

Taken together, these results show that mapping class group limit sets can be topologically minimal yet measure zero, or can omit an open dense full-measure region of the boundary. Boundary minimality and measure-theoretic largeness are therefore independent features.

## 5. Generic limit sets in cellular automata

In topological dynamics, the generic limit set is defined using Baire category rather than boundary accumulation. For a sequence of continuous self-maps \((F_t)_{t\in\mathbb N}\) on a compact metric space \(X\), the realm of attraction of \(V\subseteq X\) is
\[
\mathrm{AN}[F]{V}=\{x\in X:\omega_F(x)\subseteq V\},
\]
and the generic limit set is
\[
\widetilde{\omega}_F=\bigcap\left\{ C \subseteq X :\, C \text{ is closed and } \mathrm{AN}[F]{C} \text{ is comeager in } X \right\}.
\]
It is the topological analogue of Milnor’s likely limit set, obtained by replacing measure-one by comeager [1808.08716].

For cellular automata \(F:A^{\mathbb Z}\to A^{\mathbb Z}\), the generic limit set is always a subshift and can be written as
\[
\widetilde{\omega}_F=\omega_F(W)
\quad\text{for some comeager, shift-invariant }W.
\]
A central structural statement is that every subshift attractor has dense open, hence comeager, realm of attraction, and therefore
\[
\widetilde{\omega}_F \subseteq \text{every subshift attractor}.
\]
The generic limit set is also shift-invariant and indecomposable in the sense that it cannot be split into disjoint nontrivial shift-invariant subsystems each having comeager realm [1808.08716].

Its relation to classical limit notions depends sharply on regularity. If \(F\) is equicontinuous, then \(\widetilde{\omega}_F=\Omega_F\). If \(F\) is almost equicontinuous and \(E\) is the set of equicontinuity points, then
\[
\widetilde{\omega}_F=\overline{\omega_F(E)}.
\]
If \(F\) is sensitive, the generic limit set is always infinite. If the generic limit set is finite, then the system must be almost equicontinuous, and every generic configuration is asymptotic to a unique periodic orbit of a monochrome configuration [1808.08716].

There are also distinguished global cases. For surjective cellular automata,
\[
\widetilde{\omega}_F=A^{\mathbb Z}.
\]
In oblique directions of space-time, the generic limit set coincides with the limit set:
\[
\widetilde{\omega}_F=\Omega_F,
\]
and in such directions the system is either sensitive or nilpotent. The directional theory is explicitly non-isotropic: equicontinuity, sensitivity, and limit sets all depend on the chosen direction, and the paper gives a six-type classification for the generic limit set in directional settings [1808.08716].

The measure-theoretic and topological asymptotic sets can coincide. For cellular automata endowed with a shift-ergodic full-support measure \(\mu\), the generic limit set equals the likely limit set:
\[
\widetilde{\omega}_F=\Lambda_{F,\mu}.
\]
This is a precise instance in which Baire-generic and measure-generic asymptotics agree [1808.08716].

## 6. Terminological extensions and comparative patterns

The supplied materials also exhibit uses of the phrase *limit set* outside the classical orbit-boundary framework. In random graph theory, the summary of the \(k\)-core paper uses **core limit set** for the set of possible normalized order-size parameter values of the \(k\)-core of \(G(n,m)\). There the local limit theorem implies that the order \(X\) and size \(Y\) are concentrated within \(O(\sqrt n)\) of \(np(1-q)\) and \(mp^2\), with exact point probabilities given asymptotically by a bivariate Gaussian in terms of a covariance matrix \(Q\). The same work introduces the **Forge algorithm**, a generative model inspired by Warning Propagation, which constructs graphs with prescribed core parameters and outputs a distribution uniform over graphs with those parameters, conditioned on success [1707.03556].

In infinite-dimensional submodular theory and non-atomic games, the word **core** has yet another meaning. For an increasing subadditive non-atomic game \(\varphi\), the core is
\[
\mathrm{core}(\varphi)=\{\gamma:\gamma\leq \varphi,\;\gamma(\Omega)=\varphi(\Omega),\ \gamma \text{ a measure}\},
\]
and for submodular \(\varphi\) the corresponding base polytope is
\[
\textup{bmm}_+(\varphi)=\{\alpha:0\leq \alpha\leq \varphi,\;\alpha(\Omega)=\varphi(\Omega)\}.
\]
Under atomlessness and continuity assumptions, the extreme points of this base polytope are exactly the restricting measures \(\mu_{\varphi,Y}\), equivalently the measures whose Radon–Nikodym derivative with respect to the majorizing measure is \(\{0,1\}\)-valued almost everywhere. The supplied summary states that the **core limit set** in this context corresponds to the closure of convex combinations of restricting measures [2504.14544].

These later usages suggest a terminological extension: *limit set* can denote not only asymptotic orbit accumulation on a boundary, but also a sharply constrained asymptotic parameter region or a closure of extremal structures in a compact convex space. Across all the settings represented here, however, the underlying theme is consistent. A limit set records the persistent part of a system after transient structure is discarded—whether the persistence is geometric, dynamical, measure-theoretic, or convex-geometric. The principal differences lie in the compactification used, the meaning of typicality, and the degree to which the resulting set is large topologically, large measure-theoretically, or rigidly stratified by additional invariants such as Weyl-chamber directions, entropy, or extremal measures.

Source: https://www.emergentmind.com/topics/core-limit-set