The conditional expectation as estimator of normally distributed random variables with values in infinitely dimensional Banach spaces (Q808592): Difference between revisions

The present paper deals with an infinite-dimensional linear model \(\hat Z=Z_ 0(\hat f)+Y\) where \(\hat f: \Omega \to B_ 1\) and \(Y: \Omega \to B_ 2\) denote symmetric Gaussian random variables with values in separable Banach spaces \(B_ i\). Let \(Z_ 0 : B_ 1\to B_ 2\) denote a linear operator. Based on the observations \(\hat Z\) the author gives an explicit formula for the conditional expectation of \(\hat f\) given \(\hat Z\). This result is related to the generalized least squares estimator of \(\hat f\) to \(\hat Z\). Both expressions are asymptotically the same if Y is replaced by \(\sigma\) Y with \(\sigma \downarrow 0\). The result seems to be of some importance in nonparametrics.