Central limit theorem and stable laws for intermittent maps (Q1424397)

Let \((X,{\mathcal B},m,\alpha)\) be a topologically mixing probability measure preserving the Markov map. Assume that the induced map \(T_Y\) on \(Y=\bigcup_{a\in\gamma}a\) \((\gamma\subset\alpha)\) satisfies a \(\theta\)-Hölder distortion for some \(\theta\). Let \(I(X,Y)\) be the set of ``nice'' functions. Theorem 1.1. (Central limit theorem). Let \(f\in I(X,Y)\) be integrale with \(\int f\,dm=0\) and let \(f_Y\in L^2\). Then the central limit theorem holds, and the variance equals 0 if and only if there exists \(\chi: X\to \mathbb{R}\) satisfying \(f=\chi\circ T-\chi\). Theorem 1.2. (Convergence to stable law). Let \(f\in I(X,Y)\). Assume that \(f_Y\not\in L^2\), and that its distribution is in the domain of attraction of a stable law of index \(p\in(0,2]\backslash\{1\}\). If \(p>1\), assume moreover that \(\int f\,dm=0\). Then \(\sum_{i=0}^{n-1}f\circ T^i/ B_n\) converges in distribution to the stable law of \(X_{p,c,\beta}\) with its characteristic function \(\exp[-c|t|^p(1-i\beta \text{sgn\,}\tan p\pi/2)]\). Theorem 1.3. Let \(1/2<\alpha<1\) and \(T:[0,1]\to[0,1]\) be the Liverani-Saussol-Vaienti map. Let \(h\) be the density of its absolutely continuous invariant probability measure. Let \(f:[0,1]\to \mathbb{R}\) be Hölder with \(\int f\,dm=0\). \begin{itemize}\item{} If \(f(0)\neq0\), then \(S_nf/n^\alpha\) converges in distribution to the stable law \(X_{1/\alpha,c,\text{sgn}(f(0))}\) with \(c={h(1/2)\over4(\alpha|f(0)|)^{1/\alpha}}\Gamma(1-1/\alpha) \cos({\pi\over2\alpha})\). \item{} If \(f(0)=0\), assume moreover that \(|f(x)|\leq Cx^\gamma\) with \(\gamma>\alpha-1/2\). Then the central limit holds and the variance equals 0 if and only if there exists a measurable function \(\chi\) such that \(f=\chi\circ T-\chi\). The proof of the above theorems can be gotten using the following abstract perturbation theorem to the first return transfer operator. Theorem 2.1. Let \(\mathcal L\) be a Banach space and \(R_n\in\text{Hom}({\mathcal L},{\mathcal L})\)be operators on \(\mathcal L\) with \(\| R_n\| \leq r_n\) for a sequence \(r_n\) such that \(a_n=\sum_{k>n}r_k\) is summable. Write \(R(z)=\sum R_nz^n\) for \(z\in\overline D\). Assume that 1 is a simple isolated eigenvalue of \(R(1)\) and that \(I-R(z)\) is invertible for \(z\in\overline D-\{1\}\). Let \(P\) denotes the spectral projection of \(R(1)\) for the eigenvalue 1, and assume that \(PR'(1)P=\mu P\) for some \(\mu>0\). Let \(R_n(t)\) be an operator depending on \(t\in[-\alpha,\alpha]\), continuous at \(t=0\), with \(R_n(0)=R_n\) and \(\| R_n(t)\| \leq Cr_n\) for all \(t\in[-\alpha,\alpha]\) for some constant \(C>0\). For \(z\in\overline D\) and \(t\in[-\alpha,\alpha]\), write \(R(z,t)=\sum_{n\geq1}z^nR_n(t)\). This is a continuous perturbation of \(R(z)\). For \(t\) small and \(z\) close to 1, \(R(z,t)\) is close to \(R(1)\), whence it admits an eigenvalue \(\lambda(z,t)\) close to 1. Assume that \(\lambda(1,t)=1-(c(o(1))M(|t|)\) for \(c\in C\) with \(\Re(c)>0\), and some continuous function \(M: \mathbb{R}_+\to \mathbb{R}_+\) vanishing only at 0. Then, there exists \(\varepsilon_0>0\) such that, for \(|t|<\varepsilon\), \(I-R(z,t)\) is invertible for all \(z\in D\). We can write \((I-R(z,t))^{-1}=\sum T_{n,t}z^n\). Furthermore, there exists functions \(\varepsilon(t)\) and \(\delta(n)\) tending to 0 when \(t\to0\) and \(n\to\infty\) such that, for \(|t|<\varepsilon_0\), \(n\in N^*\), we have \[ \| T_{n,t}-{1\over\mu}\left( 1-{c\over\mu}M(|t|\right)^nP\| \leq \varepsilon(t)+\delta(n). \] \end{itemize}