A restricted dichotomy of equivalence classes for some measures of dependence (Q1596554)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | A restricted dichotomy of equivalence classes for some measures of dependence |
scientific article |
Statements
A restricted dichotomy of equivalence classes for some measures of dependence (English)
0 references
15 December 2002
0 references
The paper investigates the relationship between measures of dependence based on \(\sigma\)-fields and those based on covariances. Given two \(\sigma\)-fields \(\mathcal F\) and \(\mathcal G\), we define \[ \alpha({\mathcal F},{\mathcal G}) :=\sup_{F\in{\mathcal F},G\in{\mathcal G}}|P(F\cap G)-P(F)P(G)|, \] \[ \psi^*({\mathcal F},{\mathcal G}):=\sup_{F\in{\mathcal F},G\in{\mathcal G}} \frac{P(F\cap G)}{P(F)P(G)} , \] \[ \beta({\mathcal F},{\mathcal G}):=\sup \tfrac 12\sum^I_{i=1}\sum^J_{j=1} |P(F_i\cap G_j)-P(F_i)P(G_j)|, \] where the last supremum is taken over all finite partitions of \(\Omega\) into \(\mathcal F\)- resp. \(\mathcal G\)-measurable sets. Given a separable real Banach space \(B\), define \[ R(B,{\mathcal F},{\mathcal G}):= \sup |E \langle X,Y\rangle-\langle EX,EY\rangle|, \] where the sup is taken over all pairs of simple random variables \(X\) and \(Y\) such that \(X\) is \(B\)-valued, \(\mathcal F\)-measurable, \(Y\) is \(B^*\)-valued, \(\mathcal G\)-measurable with \(\|X\|_B\leq 1\) and \(|Y\|_{B^*}\leq 1\). Here \(B^*\) denotes the dual space, and \(\langle x,y\rangle\) the canonical bilinear form on \(B\times B^*\). It is well-known that \(R(\mathbb{R},{\mathcal F},{\mathcal G}) = 4\alpha({\mathcal F},{\mathcal G})\) and that a similar result holds for all separable Hilbert spaces. On the other hand, there are certain Banach spaces, e.g. \(c_0\), \(l^\infty\), \(l^1\), with \(R(B,{\mathcal F},{\mathcal G}) =\beta({\mathcal F},{\mathcal G})\). It is conjectured that for any Banach space, \(R(B,{\mathcal F},{\mathcal G})\) is either equivalent to \(\alpha({\mathcal F},{\mathcal G})\) or to \(\beta({\mathcal F},{\mathcal G})\). The present paper proves this conjecture in some restricted sense, requiring an extra bound on \(\psi^*({\mathcal F},{\mathcal G})\).
0 references
dependent processes
0 references
mixing
0 references
correlation inequality
0 references
0 references