A Julia-Fatou Theory via Random Systems with Complete Connections

TL;DR

This paper extends Julia-Fatou theory to random dynamical systems driven by non-Markovian processes with complete connections, analyzing stability and Julia sets in a generalized stochastic framework.

Contribution

It introduces a novel framework for Julia-Fatou theory in non-Markovian random systems with complete connections, including new phenomena and criteria for stability.

Findings

Proves equicontinuity of iterates under certain conditions.

Identifies emptiness jumps of kernel Julia sets along trajectories.

Provides criteria to exclude jump phenomena in RSCCs.

Abstract

We develop a Julia-Fatou theory for random dynamical systems of continuous self-maps on a compact metric space, driven by random systems with complete connections (RSCCs). This framework allows the selection rule to depend on the evolving state and, in general, on the entire past, going beyond the Markovian graph directed Markov system setting. For each state we define Julia, Fatou, and kernel Julia sets via equicontinuity of admissible composition families, and we introduce a pathwise and skew product viewpoint. Under natural compactness and continuity assumptions on the RSCC, we study the associated averaged dynamics on the product space and prove Cooperation Principle I: if the kernel Julia set is empty at every state and the admissible maps are open, then the iterates of the adjoint transition operator are equicontinuous on the whole space of probability measures, and along almost every admissible path the fiberwise Julia set has zero mass for any given finite measure. We further identify a new phenomenon specific to RSCCs, namely emptiness jumps of kernel Julia sets along admissible state trajectories, and provide criteria excluding such jumps, including discreteness of the state space and a propagation mechanism under phi-irreducibility. Several examples, motivated by reinforcement and feedback mechanisms, illustrate both the jump phenomenon and the applicability of the Cooperation Principle I in non-Markovian settings.