Canonical Correlation

Abstract

We shall now consider an extension to stationary Gaussian processes of the discussion relative to canonical correlations in Kullback (1959), pp. 200–204. We shall assume that the spectral distribution function is absolutely continuous and write

$$\frac{{dF}}{{d\lambda }} = f\left( \lambda \right)$$

where f (λ) is the spectral density. We shall first examine the values of the spectral density as the eigenvalues of the covariance operator.