Mean ergodicity and mean stability of regularized solution families (Q2569748)

Let \(X\) be a Polish space with metric \(\rho\). Consider the iterated function system \((S,p)\) such that \(S=\{S_i\}_{i=1}^N\) and \(p=\{p_i\}_{i=1}^N\) satisfies (a) \(S_i : X\to X\) is Lipschitz continuous for each \(i\), (b) \(p_i : X\to[0,1]\) is continuous for each \(i\), and \(\sum_{i=1}^Np_i(x)=1\) for all \(x\). Let \(H\) be a self-similar set generated by the above IFS. One considers the Hausdorff dimension of \(H\) under a probability measure \(\mu\) defined by \[ \dim_H\mu=\inf\{\dim_HA: \mu(A)=1\}, \] where \(\dim_HA\) is the usual Hausdorff dimension of a measurable set \(A\subset X\). Define a Markov operator \(P\) from the set of finite measures \(\mathcal M\) into itself. \[ P\mu(A)=\sum_{i=1}^N\int_{S_i^{-1}(A)}p_i(x)\,d\mu. \] A Markov operator \(P\) is called asymptotically stable if there exists \(\mu_*\) such that \(P^n\mu\to\mu_*\). Here the norm on \(\mathcal M\) is defined by \(||\mu||=\sup\{\,|\int f(x)\,d\mu|\,\}\), where the supremum is taken over all bounded and continuous functions \(f\) with a supremum norm less than or equal to 1, and \(|f(x)-f(y)|\leq\rho(x,y)\) for \(x,y\in X\). Put (1.2) \(\prod_{i=1}^N(\rho(S_i(x),S_i(y))^{p_i(x)}\leq\gamma_0\rho(x,y)\) for all \(x,y \in X\), (1.3) \(\kappa=\min_{1\leq i\leq N}\inf_{x\in X}p_i(x)>0\), and let (1.4) \(\omega:{\mathbb R}_+\to{\mathbb R}_+\) be a nondecreasing and concave function satisfying the Dini condition: \[ \int_0^\sigma \frac{\omega(t)}{t}\,dt<\infty\quad\text{for some }\sigma>0, \] and \(\sum_{i=1}^N|p_i(x)-p_i(y)|\leq\omega(\rho(x,y))\). Using symbolic dynamics, one gets the following upper estimate of the Hausdorff dimension: Theorem 3.1. Assume that an IFS satisfies conditions (1.2)--(1.4). Then \[ \dim_H\mu_*\leq\frac{\log\delta_0}{\log\gamma_0}, \] where \(\delta_0=\inf_{x\in X}\prod_{i=1}^Np_i(x)^{p_i(x)}\). Under the assumptions (4.1) \(\rho(S_i(x),S_i(y))\geq l_i(x,y)\text{ for }x,y\in X,\) where \(l_i\in(0,1)\), (4.2), without loss of generality, \(l_1\leq l_i\leq l_N\) for \(i=1,\dots,N\), (4.3) \(K=\bigcup_{i=1}^NS_i(K)\), (4.4) \(\bigcap_{i=1}^nS_i(K)=\emptyset\), one can get the following lower estimate of the Hausdorff dimension: Theorem 4.1. Assume that an IFS satisfies conditions (4.1)--(4.4). Assume further that there exists an invariant probability measure \(\mu_*\) with its support contained in \(K\). If \(\sum_{i\in I_x}p_i(S_i^{-1}(x))<1\text{ for every }x\in K\), then \[ \dim_H\mu_*\geq s_*, \] where \(s_*=\sup\{s>0:\sum_{i\in I_x}p_i(S_i^{-1}(x))l_i^{-s}\leq 1,\, x\in K\}\), and \(I_x=\{i:i\in\{1,\dots,N\}, \,x\in S_i(K)\}\).