Iterated Random Affine Map

• Iterated random affine maps are stochastic processes defined by repeated application of affine transformations, yielding invariant measures and exhibiting fractal dynamics.
• They provide a framework for analyzing ergodic properties, convergence rates, and dimension theory using tools like Lyapunov exponents and spectral gap analysis.
• Their applications span randomized projection methods, wavelet construction, and fractal geometry, impacting both theoretical research and practical algorithm design.

An iterated random affine map is a stochastic process or dynamical system generated by repeated selection and application of affine transformations, typically of the form $S_\theta(x) = A_\theta x + b_\theta$, where the linear part $A_\theta$ and translation $b_\theta$ may depend on a randomly chosen parameter $\theta$. This framework appears extensively in the study of random dynamical systems, Markov chains, iterated function systems (IFS), and probabilistic operator equations. The theory encompasses models with place-dependent probabilities, nonexpansive iterations, refinement equations, fractal geometry, and the ergodic or dimension-theoretic properties of invariant measures.

1. Model Definitions and Foundational Frameworks

The general formulation involves a Polish space $X$ (often $\mathbb{R}^d$), a family of affine maps $S_\theta:X\to X$, and a probability kernel $\theta_x$ assigning, for each $x\in X$, a probability measure on a parameter space $\Theta$. The Markov chain is then given by

$$X_{n+1} = S_{\Theta_n}(X_n), \quad \Theta_n \sim \theta_{X_n},$$

where $\Theta_n$ is sampled independently conditioned on $X_n$ (Kapica et al., 2011). This construction accommodates spatially inhomogeneous stochastic dynamics and allows varying randomness across $X$.

For the study of invariant sets and iterative random orbits, systems of nonexpansive affine maps $T_i(x)=A_ix+b_i$ selected i.i.d. according to weights $p_i$ yield a random process

$x_n = T_{\xi_n}(x_{n-1}),$

analyzed both for boundedness and invariance of $\omega$-limit sets (Leśniak, 2013).

Random affine operators of the refinement type act in $L^1(\mathbb{R})$ via

$$T[f](x) = \int_\Omega |L(\omega)| f(L(\omega)x-M(\omega))\,P(d\omega)$$

and appear in functional equations of the form $f(x) = T[f](x) + g(x)$ (Kapica et al., 2015).

Specific structures such as the random $\beta$-transformation on the Sierpinski gasket employ affine contractions with digital codings and place-dependent coin tosses to generate $\beta$-expansions and invariant fractal measures (Zhang et al., 2022).

Affine IFSs on intervals or higher-dimensional tori can incorporate both expanding and contracting map components, leading to complex ergodic and correlation phenomena determined by the Lyapunov exponent of the random process (Homburg et al., 2022, Dubail et al., 2021).

2. Existence, Uniqueness, and Convergence of Invariant Measures

A central concern is the existence and uniqueness of an invariant (stationary) probability measure $\mu \in \mathcal{P}(\mathbb{R}^d)$, i.e., $P^*\mu = \mu$ for the transfer operator $P^*$. Under regularity, contraction, and minorization conditions, exponential convergence to $\mu$ in the Fortet-Mourier metric is established: $$\|P^n(x, \cdot) - \mu(\cdot)\|_{\mathrm{FM}} \leq C (1+|x|^s)\rho^n$$ (Kapica et al., 2011). These results use Lyapunov function estimates, asymptotic coupling kernels, and spectral gap analysis to yield both uniqueness and geometric convergence.

In nonexpansive settings, random iterations can recover minimal closed invariant $\omega$-limit sets, which are compact and invariant under all participating maps. For strictly contractive compositions, the limiting set is independent of the initial state (Leśniak, 2013).

Random affine refinement equations admit unique $L^1$ solutions under average-contractivity; the solution can be constructed via explicit random series and characterized both analytically (Fourier transform) and measure-theoretically (Radon-Nikodym derivatives) (Kapica et al., 2015).

3. Structure and Properties of Stationary Laws (“Perpetuities”)

The stationary law of the iterated random affine process satisfies a perpetuity-type equation: $$X \overset{d}{=} A_{\Theta} X + b_{\Theta}, \quad \Theta \sim \theta_X$$ and may be realized through backward iterative series, converging almost surely and in $L^s$ to a limit determined by contraction and translation control (Kapica et al., 2011).

Inhomogeneous refinement equations yield solutions expressible as series involving iterated random affine compositions and driving functions; respective regimes (contractive, boundary, expanding) produce distinctive forms for the solution and affect uniqueness and regularity (Kapica et al., 2015).

Fractal systems generated by affine contractions, such as random $\beta$-transformations, admit measures of maximal entropy that characterize the coding complexity and ergodic hierarchy of the random system (Zhang et al., 2022). The unique invariant measures may be absolutely continuous on certain domains and singular or fractal in others, depending on system parameters.

4. Dimension Theory and Exact Dimensionality

The dimension theory of invariant measures for random affine IFSs is governed by exact dimensionality results and Ledrappier–Young type formulas. For average-contracting affine IFSs, the projection of an ergodic $\sigma$-invariant measure on the shift space yields

$$\dim_H \mu = \sum_{i=0}^{s-1} \frac{h_{i+1}-h_i}{-\lambda_{i+1}}$$

where $(\lambda_{i+1})$ are Lyapunov exponents and $(h_i)$ conditional entropies along Oseledets subspaces (Feng, 2019). Dimension conservation along random matrix product subspaces and slicing/projection arguments extend these results to self-affine sets and measures.

For specific random piecewise affine homeomorphisms on $[0,1]$ (“AM systems”), Hausdorff dimension bounds

$\dim_H \mu \leq -H/\chi(\mu),$

depend on entropy $H$ and Lyapunov exponent $\chi(\mu)$. Certain parameter regimes yield $\dim_H \mu < 1$ so that the measure is singular, answering open questions on stationarity and regularity (Barański et al., 2022).

Phase transitions for the support of stationary measures occur in systems with mixed expanding/contracting branches: the support can jump from the diagonal to full-dimensional with accompanying transitions between trivial, infinite, and finite absolutely continuous measures as the Lyapunov exponent crosses zero (Homburg et al., 2022).

5. Fractal Geometry, Coding, and Random Dynamics

Iterated random affine maps may produce attractors with nontrivial fractal geometry. For instance, the “fat Sierpinski gasket” generated by random affine contractions for $1<\beta<2$ is a triangle or a planar fractal whose coding sequences and invariant measures are explained via the random $\beta$-map $K_\beta$ (Zhang et al., 2022). The interplay of greedy, lazy, and random digit assignments yields systems with explicitly characterized maximal entropy measures and absolutely continuous densities under suitable BV conditions, while geometric features (radial holes) arise as parameters cross critical thresholds.

The construction of these systems establishes connections between symbolic coding (as shift systems), stochastic selection rules (coin tosses for overlap regions), and dynamical properties of invariant fractal sets.

6. Mixing, Ergodicity, and Long-Term Behavior

For affine random walks on finite tori, hyperbolic linear dynamics (no eigenvalue of modulus 1) together with random translations lead to optimal mixing times: $t_{\mathrm{mix}} = O(\log{n} \log{\log{n}})$ or even $O(\log n)$ for almost all $n$, with this acceleration stemming from the hyperbolic expansiveness of the underlying deterministic transformation (Dubail et al., 2021).

Synchronization, intermittency, and divergence regimes for orbits of affine IFSs, as determined by the Lyapunov exponent, result in phenomena ranging from almost sure convergence of orbits, to orbits remaining statistically close for substantial periods, or measures converging to highly nontrivial stationary distributions (Homburg et al., 2022).

The ergodic and Markovian properties underpinning the random affine iteration ensure that, under contraction and coupling conditions, the limiting behavior is robust and often independent of initial conditions.

7. Applications and Variations

Iterated random affine map theory underpins algorithms for projection methods such as randomized Kaczmarz–von Neumann, where random projections onto affine subspaces iteratively recover Chebyshev-optimal configurations or minimal invariant sets in infeasible systems (Leśniak, 2013).

Refinement equations driven by random affine maps provide characterizations of scaling functions, wavelets, and self-similar distributions, with solutions built via explicit probabilistic series or Lipschitz fixed-point methods (Kapica et al., 2015).

The framework is central to contemporary fractal geometry, dynamical systems, ergodic theory, and mixing processes on abelian groups—serving as a template for studies in dimension theory, entropy, and stochastic stabilization of measure-theoretic and geometric properties.