On the weak law of large numbers for double adapted arrays of random elements in \(p\)-uniformly smooth Banach space (Q1033688): Difference between revisions

The aim of this paper is to establish the weak law of large numbers (WLLN) in another circumstances than those mentioned in several other papers. It contains two examples and auxiliary propositions, 4 lemmas, 5 corollaries but only two main results for an adapted array \(\{X_{mn},\mathcal{F}_{mn},m\geq 1,n\geq 1\}\) in a \(p\)-uniformly smooth Banach space \(\mathcal{X}\;(1\leq p\leq 2).\) To formulate them we shall remind of the definitions of these two notions. Let \(\{\Omega,\mathcal{F},P\}\) be a probability space, \(\mathcal{X}\) be a real separable Banach space and \(\mathbf{B}(\mathcal{X})\) be the \(\sigma\)-algebra of all Borel sets in \(\mathcal{X}.\) For a double array \(\{\mathcal{F}_{mn},\;m\geq 1,n\geq 1\}\) of sub-\(\sigma\)-algebras of \(\mathcal{F}\) with indices in \(\mathbb{N}\times \mathbb{N},\) a double array \(\{X_{mn},\mathcal{F}_{mn},m\geq 1,n\geq 1\}\) will be called an \textit{adapted array} if it satisfies conditions: 1) \(X_{mn}\) is a \(\mathcal{F}_{mn}/\mathbf{B}(\mathcal{X})\)-measurable random element; 2) \(\mathcal{F}_{m_1n}\subset \mathcal{F}_{m_2n}\) for each \(n\in \mathbb{N}\) and \(m_2>m_1,\) \(\mathcal{F}_{n_1m}\subset \mathcal{F}_{n_2m}\) for each \(m\in \mathbb{N}\) and \(n_2>n_1.\) A real separable Banach space \(\mathcal{X}\) is said to be \(p\)-\textit{uniformly smooth} \((1\leq p\leq 2)\) if for some constant \(\mathcal{C}\) \[ \rho(\tau)=\sup\left\{\frac{\| x+y\|+ \| x-y\|}2-1;\;\forall x,y\in \mathcal{X};\;\| x\|=1,\;\| y\|=\tau\right\}\leq \mathcal{C}\tau^{\rho}. \] Theorem 2.1. Let \(\{b_{mn},m\geq 1,n\geq 1\}\) be an array of positive numbers and put \(Y_{ij}=X_{ij}I\{\| X_{ij}\|\leq b_{mn}\},\) \(m\times n=(1\leq i\leq m,1\leq j\leq n),\) \(m\vee n=\max(m,n),\) \(\mathcal{F}_{mn}^*=\sigma(\mathcal{F}_{m-1,\infty}\cup \mathcal{F}_{\infty,n-1}).\) Then we have \[ \frac 1{b_{mn}}\sum_{i=1}^m\sum_{j=1}^nX_{ij}\overset{\mathbb{P}} {\longrightarrow}\;0,\text{as}\;\;\;m\vee {n} \infty, \] \[ \text{if}\;\;P\{\max_{m\times n}\| X_{ij}\|>b_{mn}\}\to\;0,\;\;\;\frac 1{b_{mn}}\sum_{i=1}^m\sum_{j=1}^n\mathbb{E}\{Y_{ij} \mathcal{F}^*_{ij}\}\overset{\mathbb{P}}{\longrightarrow}\;0, \] \[ \frac 1{b_{mn}^p}\sum_{i=1}^m\sum_{j=1}^n\mathbb{E}\| Y_{ij}- \mathbb{E}\{Y_{ij}|\mathcal{F}^*_{ij}\}\|^p\to 0, \;\text{as}\;\;m\vee n\to \infty. \] The following proposition state WLLN with random indices. Theorem 2.3. Suppose that \(\{X_{mn},\;m\geq 1,n\geq 1\}\) is stochastically dominated by a random element \(X\) (it means, that there exists a constant \(\mathcal{C},\;0<\mathcal{C}<\infty,\) such that \(\mathbb{P}\{\| X_{mn}\|>t\}\leq \mathcal{C} \mathbb{P}\{\| X\|>t\},\;t\geq 0,m\geq 1,n\geq 1).\) Let \(\{T_n;n\geq 1\}\) and \(\{\tau_n;n\geq 1\}\) be sequences of positive integer-valued r.v.'s where \(T_n/n\) and \(\tau_n/n\) are \textit{bounded in probability} (that is, for example, \[ \lim_{K\to \infty}\sup_{n\geq 1}\mathbb{P}\left\{\frac{|T_n|}n>K\right\}=0) \] and let real number \(r\in (0,p)\) and r.v. \(Y_{ij}=X_{ij}I\{\| X_{ij}\| \leq (mn)^{1/r}\}\). Then if \(\lim_{\lambda \to \infty}\lambda \mathbb{P}\{\| X\|>\lambda^{1/r}\}=0,\) the WLLN holds: \[ \frac{\sum_{i=1}^{T_m}\sum_{j=1}^{T_n}(X_{ij}-\mathbb{E}\{Y_{ij}| \mathcal{F}^*_{ij}\})}{m^{1/r}n^{1/r}} \overset{\mathbb{P}}{\longrightarrow}\;0,\;\;\text{as}\;\;m\vee {n} \infty. \]