A test for the independence of two Gaussian processes (Q760741)

Let \(Z=(X,Y)'\) be a zero mean bivariate Gaussian process observed in some region \(\Omega\) in \(R^ r\), with covariance \[ \Sigma (s,t,\rho)= \begin{pmatrix} \Sigma_{11}(s,t) & \rho\Sigma_{12}(s,t) \\ \rho\Sigma_{21}(s,t) & \Sigma_{22}(s,t) \end{pmatrix} \] where \(\{\Sigma_{ij}(s,t)\}\) are known functions, and \(\rho\) is an unknown parameter. A test of \(H_ 0: \rho =0\), locally equivalent to the likelihood ratio test, and an unbiased estimate of \(\rho\) are given for the case when \(\Omega\) is a finite set. There are two distinct situations in the case when \(\Omega\) has a positive Lebesgue measure. The first one is similar to the finite case. In the second situation a test of \(H_ 0\) may be constructed, with power at a given value of \(\rho\) arbitrarily close to 1, and a consistent estimate of \(\rho\) is given.