Empirical processes of multidimensional systems with multiple mixing properties (Q544505): Difference between revisions

Let \((X_i)_{i\geq 1}\) be an \(\mathbb R^d\)-valued stationary stochastic process and \(({\mathcal B},\|\cdot\|)\) be a Banach space of measurable functions \(\varphi:\mathbb R^d\to\mathbb R\). We say that \((X_i)_{i\geq 0}\) has the multiple mixing property with respect to \({\mathcal B}\) if there exist constants \(0<\theta<1\) and \(r\geq 1\) such that, for any positive integer \(p\), there exist a positive constant \(C\) and an integer \(\ell\) for which the assertion holds that, for any positive integers \(i_1,\dots,i_p\), \(1\leq q\leq p\) and \(\varphi\in{\mathcal B}\) \((E_\nu(\varphi)=0\), \(\|\varphi\|_\infty\leq 1)\), \[ \big|\text{Cov}(\varphi(X_0)\varphi(X_{i_1^*})\cdots \varphi(X_{i_{q-1}^*}), \varphi(X_{i_q^*}\cdots \varphi(X_{i_p^*}))\big|\leq C\|\varphi(X_0)\|_r \|\varphi\|^\ell P(i_1,\dots,i_p)\theta^{i_q}, \] where \(P\) is a polynomial function of \(p\) variables which does not depend on \(\varphi\) and \(i^*= \sum_{j=1}^n i_j\). Among other results, the authors show that, if an \(\mathbb R^d\)-valued stationary process \((X_i)_{i\geq 0}\) has the above property, then, for any \(p\geq 1\), \[ E_\nu\bigg( \sum_{i=0}^{n-1} \varphi(X_i)\bigg)^{2p}\leq K\sum_{i=1}^p n^i\|\varphi(X_0)\|_r^0 \log^{2p-i}(\|\varphi\|+\theta^{-1}), \] where \(K\) is a constant which does not depend on \(n\), \(\varphi\) or \(r\geq 1\). Now, let \({\mathcal H}^\alpha\) be the Banach space of bounded \(\alpha\)-Hölder continuous functions \((0<\alpha\leq 1)\) equipped with the norm \[ \|g\|= \|g\|_\infty+ \sup_{x\neq y\in\mathbb R^d} \frac{|g(x)-g(y)|}{|x-y|^\alpha}. \] Using the above inequality, the authors establish the following theorem. Let \((X_i)_{i\geq 1}\) be an \(\mathbb R^d\)-valued stationary stochastic process with multivariate marginal distribution function \(F(t)\) satisfying the multiple mixing property with respect to \({\mathcal H}^\alpha\), and define the empirical distribution function \(F_n(t)\) and the empirical processes \(\{U_n(t): t\in\mathbb R^d\}\). Assume that, for all \(\phi\in{\mathcal H}^\alpha\), the partial sums \(\sum_{i=1}^n \phi(X_i)\) satisfy the central limit theorem, i.e., \[ \frac{1}{\sqrt{n}} \sum_{i=1}^n \big(\phi(X_i)-E\phi(X_1)\big)\to N(0,\sigma^2), \] where \(\sigma^2= \text{Var}(\phi(X_1))+ 2\sum_{i=2}^\infty \text{Cov}(\phi(X_i), \phi(X_1))>0\) and that, for some \(\gamma>r\), \(\omega(\delta)= O(|\log(\delta)|^\gamma)\) holds, where \(\omega\) denotes the modulus of continuity of \(F\). Then, for \((X_i)_{i\geq 0}\) the empirical process central limit theorem holds. As an application of the result, the empirical process invariance principle for ergodic torus automorphisms is obtained. The results are also applied to Markov chains and dynamical systems having a spectral gap.