Local Iterated Function Systems

• Local iterated function systems are defined on prescribed subsets of a metric space, enabling state-dependent dynamics and fractal interpolation.
• They employ contraction mappings to ensure the unique existence of local attractors, paralleling classical fixed-point theory in a localized setting.
• Applications include adaptive PDE solvers, image encoding, and symbolic coding, offering refined tools for modeling non-self-similar structures.

A local iterated function system (local IFS) generalizes the classical global IFS framework by allowing each map to be defined only on a prescribed subset of the ambient space, rather than the entire space. This approach enables the modeling of more intricate, state-dependent, and locally adaptive dynamics, and is central to modern fractal geometry, symbolic dynamics, and applied approximation theory. The local IFS framework unifies fractal interpolation, graph-directed constructions, non-self-similar and non-SFT structures, and provides a flexible setting for applications in function fitting, image encoding, and adaptive PDE solvers.

1. Structure and Definition

A local IFS on a compact metric space $(X,d)$ is specified as a finite collection $\{(X_j, f_j)\}_{j=1}^n$, where each $X_j\subset X$ is closed and each $f_j: X_j \to X$ is continuous. The family $\mathfrak X = (X_j)_{j=1}^n$ defines the collection of domains, and the local Hutchinson–Barnsley operator $F_{\mathfrak X}: 2^X \to 2^X$ is given by

$$F_{\mathfrak X}(B) = \bigcup_{j=1}^n f_j(B \cap X_j), \quad B \subset X.$$

A set $A \subset X$ is a local attractor if $A = F_{\mathfrak X}(A)$. The collection of all $F_{\mathfrak X}$-invariant subsets, ordered by inclusion, has a unique maximal element, denoted $A_{\mathfrak X}$, which is the local attractor provided it is non-empty. This structure recovers the classical theory when $X_j = X$ for all $j$ (Oliveira et al., 12 Jan 2026).

2. Existence, Uniqueness, and Contractivity

For contractive local IFSs—those where for each $j$,

$$d(f_j(x), f_j(y)) \leq \lambda_j d(x, y), \quad \text{with } 0 \leq \lambda_j < 1,\, x, y \in X_j,$$

existence and uniqueness of a non-empty compact local attractor $A_{\mathfrak X}$ is guaranteed. For any compact $K \subset X$, the sequence of iterates $F_{\mathfrak X}^k(K)$ converges in the Hausdorff metric to $A_{\mathfrak X}$. This fixed-point property is proven via a standard contraction-mapping argument on the space of non-empty compact sets endowed with the Hausdorff metric, paralleling the global setting (Massopust, 2013, Oliveira et al., 12 Jan 2026).

The open-set condition (OSC), $f_i(X_i) \cap f_j(X_j) = \emptyset$ for $i \neq j$, ensures further structural regularity and in some settings leads to full symbolic coding homeomorphism properties.

In the non-hyperbolic or irregular regime, target set and semifractal theory (Díaz et al., 2018) generalize these existence results: the closure of the set of points approachable by weakly hyperbolic sequences yields the minimal fixed point of the local IFS operator, which is an (often unique) local attractor under strict or Conley attractor conditions.

3. Symbolic Coding, Shift Spaces, and Combinatorics

To analyze the combinatorial structure of local IFSs, one introduces the code space:

$\Sigma^- = \{1, \dots, n\}^{-\mathbb{N}},$

with the ultrametric

$$d_-(\underline b, \underline c) = e^{-\sup\{N \geq 1 : b_{-j} = c_{-j} \ \forall\, 1 \leq j \leq N\}}.$$

The coding map is given by constructing nested images of words, where for $\underline b \in \Sigma^-$:

$$\pi_{\mathfrak X}(\underline b) = \bigcap_{k \geq 1} f_{b_{-1}} \circ \cdots \circ f_{b_{-k}}(X_{b_{-k}}).$$

The local code space $\Sigma^-_{\mathfrak X}$ consists of all sequences for which this intersection is non-empty. For contractive systems, the coding map $$\pi_{\mathfrak X}: \Sigma^-_{\mathfrak X} \to A_{\mathfrak X}$$ is surjective and Hölder-continuous; under OSC, it is a homeomorphism.

The possible combinatorial structures go beyond subshifts of finite type (SFT): there exist local IFSs whose code spaces are not SFTs, and whose attractors are not globally self-similar. The symbolic dynamics for two-sided sequences is governed by the set of admissible orbits and is classified by the invariant subsets of the full shift, yielding a bijection between invariant fractal subsets and combinatorial itineraries (Oliveira et al., 12 Jan 2026).

4. Local Fractal Functions and Function Space Membership

Local fractal functions are fixed points of the so-called Read–Bajactarević (RB) operator, defined on a Banach function space $B(X, Y)$:

$$\Phi f(x) = \sum_{i=1}^N \left[\lambda_i(\varphi_i^{-1}(x)) + S_i(\varphi_i^{-1}(x)) f(\varphi_i^{-1}(x))\right] \chi_{\varphi_i(X_i)}(x),$$

where each $\varphi_i: X_i \to X$ is a contraction, $\lambda_i$ are "lift" functions, and $S_i$ are scaling functions with $\|S_i\|_{\infty, X_i} < 1$.

Under uniform contractivity, $\Phi$ is a strict contraction, so by the Banach fixed-point theorem, there is a unique fixed point—a local fractal function—solving a self-referential, piecewise functional equation. The graph of this fixed point is the local attractor for a lifted local IFS on $X \times Y$.

By explicit norm estimates and contraction conditions, local fractal functions can be shown to reside in various function spaces—Lebesgue $L^p$, Hölder $\dot{C}^s$, smoothness $C^n$, and Sobolev $W^{m, p}$ spaces—subject to additional join-up, interpolation, and scaling constraints (Massopust, 2013).

5. Numerical Algorithms and Data Fitting

Numerically, the RB operator can be discretized on an admissible grid $X^g \subset X$, leading to a finite-dimensional affine map of the form

$\Phi^g f^g = \lambda^g + M f^g,$

where $f^g$ is the function sampled on $X^g$, and $M$ has a block sampling–diagonal–sampling structure reflecting the local partitions and maps. Being contractive, power iteration yields convergence to the discretized local fractal function.

For data fitting, least-squares or convex optimization procedures can be employed to determine optimal parameters that minimize the Collage distance $\|g - \Phi^g(g)\|$. This approach extends to fitting local IFSs constrained by partial differential equations: minimization of the residual $\|-u'' - F_{\rm loc}(u)\|$ with respect to the local IFS parameters produces fractal-based approximate (pre)solvers for ODEs or PDEs (Barnsley et al., 2013).

6. Applications and Illustrative Examples

• Fractal Interpolation and Function Fitting: Subdividing $[0,1]$ into subintervals $X_i$ and defining affine $f_i$ on each $X_i$ yields fractal interpolants with prescribed nodal values. Classical splines and piecewise polynomials appear as special cases where the scaling functions vanish (Barnsley et al., 2013, Massopust, 2013).
• Fractal Surface Fitting: Local IFSs, by permitting map domains to align with mesh patches and boundary conditions, enable fitting of fractal surfaces with geometric constraints, finding use in multiresolution geometric modeling (e.g., CAD/CAM) and adaptive mesh refinement.
• Graph-Directed Structures: Every attractor of a graph-directed IFS can be realized as a local attractor of a contractive local IFS defined on an enriched space $X \times V$, linking symbolic dynamics and combinatorial rotations to geometric realization (Oliveira et al., 12 Jan 2026).
• Image Compression and Adaptive Encoding: Local IFSs enable more refined and locally adaptive image encoding schemes compared to classical global fractal compression.
• Dynamical Systems: Place-dependent local IFSs give rise to Markov chains with nontrivial ergodic properties, multifractal invariant measures, and stochastic modeling frameworks (Ladjimi et al., 2017).

7. Generalizations, Non-hyperbolic Regimes, and Open Directions

• Non-hyperbolic IFSs: For systems lacking uniform contractivity, the target set framework gives necessary and sufficient conditions for the existence of local attractors and semifractals, with convergence of chaos-game orbits under stability conditions (Díaz et al., 2018).
• Symbolic Complexity: There exist local IFS attractors whose combinatorial structure is not that of a subshift of finite type, and whose geometric dimension or regularity can sharply differ from classical self-similar cases (Oliveira et al., 12 Jan 2026).
• Local Chaos Game Algorithms: Random or deterministic iteration schemes, constrained to admissible domains at each step, converge almost surely to the local attractor if the system is contractive-on-average or “stable,” generalizing the classic chaos game (Ghosh et al., 2022, Díaz et al., 2018).
• Place-dependent Probabilities and Markov Chains: Local IFSs with state-dependent selection rules yield Markov operators with spectral gaps under suitable conditions, supporting uniqueness and mixing rates for invariant measures (Ladjimi et al., 2017).

Open challenges include sharp, verifiable criteria for local contractivity when domain overlap is complex, precise mixing-rate estimates in non-global settings, and efficient algorithmic design for adaptive chaos-game schemes (Ghosh et al., 2022).

---

References:

• Foundations of local iterated function systems (Oliveira et al., 12 Jan 2026)
• Local fractal functions and function spaces (Massopust, 2013)
• Numerics and Fractals (Barnsley et al., 2013)
• Non-hyperbolic Iterated Function Systems: semifractals and the chaos game (Díaz et al., 2018)
• Iterated function systems with place dependent probabilities (Ladjimi et al., 2017)
• Iterated Function Systems: A Comprehensive Survey (Ghosh et al., 2022)