No label defined (Q3989338)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 34431
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 34431 |
Statements
28 June 1992
0 references
measures of dependence
0 references
compact operator
0 references
Sampson's measure of closed dependence
0 references
Jupp and Mardia's measure of dependence
0 references
Sampson's correlation ratio
0 references
maximal canonical correlation
0 references
coefficient of maximal correlation
0 references
conditional expectation
0 references
angle between linear subspaces
0 references
separable Hilbert space
0 references
0.7616639733314514
0 references
0.7597452998161316
0 references
0.7531102895736694
0 references
The paper consists of two parts: geometrical and probabilistic. In the first part, we introduce the notion of an angle between linear subspaces and of its measure. Let \(({\mathcal X},{\mathcal Y})\) be a pair of closed subspaces of a real separable Hilbert space \((H,\langle\cdot,\cdot\rangle)\) and let \(P_ 1'\), \(P_ 2'\) be the orthogonal projections of \(\mathcal X\) on \(\mathcal Y\) and of \(\mathcal Y\) on \(\mathcal X\), respectively. The angle is defined for pairs \(({\mathcal X},{\mathcal Y})\) such that \(P_ 2'P_ 1'\) and \(P_ 1'P_ 2'\) are compact. The set of all pairs \(({\mathcal X},{\mathcal Y})\) with this property will be denoted by \({\mathcal K}(H)\); by \(a^ 2\) we will denote the largest eigenvalue of \(P_ 2'P_ 1'\) among eigenvalues smaller than 1, which is equal to the respective eigenvalue of \(P_ 1'P_ 2'\). Then, for any \(({\mathcal X},{\mathcal Y})\in{\mathcal K}(H)\) we define the angle between \({\mathcal X}\) and \({\mathcal Y}\) as the angle between unit eigenvectors of \(P_ 2'P_ 1'\) and \(P_ 1'P_ 2'\) which correspond to \(a^ 2\), and \(\arccos a\) is said to be the measure of this angle. If \(H\) is finite-dimensional then \({\mathcal K}(H)\) consists of all pairs of subspaces of \(H\). In the probabilistic part of the paper the cosine of the angle between subspaces \(\mathcal X\) and \(\mathcal Y\) of a real separable Hilbert space \((L^ 2(\Omega,{\mathcal A},P;\mathbb{R}^ k),\langle\cdot,\cdot\rangle_ \Lambda)\) is considered. This space consists of all \(k\)-dimensional centered random vectors with finite covariance matrix, and the scalar product is given by \[ \langle X,Y\rangle_ \Lambda=E(X^ T\Lambda^{- 1}Y)=\text{tr}(\Lambda^{-1}\Sigma_{12}), \] where \(\Lambda\) is a positive definite symmetric \(k\times k\) matrix, and \(\Sigma_{12}\) is the covariance matrix of \(X\) and \(Y\). Thus, we define for \(U\in{\mathcal X}\), \(V\in{\mathcal Y}\), \(({\mathcal X},{\mathcal Y})\in{\mathcal K}(H)\), \[ \rho_ \Lambda(U,V)=\langle U,V\rangle/\| U\|_ \Lambda\| V\|_ \Lambda, \] which may be interpreted as the cosine of the angle between \(U\) and \(V\). Maximal values of this expression on suitably chosen sets of pairs of random vectors will serve as measures of dependence between \(X\) and \(Y\).
0 references