An extension of Mercer theorem to matrix-valued measurable kernels (Q1948593)

The authors analyses an extension of Mercer's theorem on the representation of vector valued reproducing kernel Hilbert spaces. In the scalar case, Mercer's theorem is stated for functions from a set \(X\) to \(\mathbb R\) or \(\mathbb C\), and provides a series representation for the reproducing kernel \(K\) under some requirements. Usually, it is assumed that \(X\) is a compact metric space and the corresponding reproducing kernel is continuous. Let \(Y\) be a normed space. Reproducing kernel Hilbert spaces are function spaces of functions with values in \(Y\) with the property that the evaluation operator at each point is continuous. Such a space -- also for the vector valued functions case --, is completely characterized by a reproducing kernel that takes its values in the space of bounded operators on \(Y\). If \(K\) is such a kernel, Mercer's theorem provides a representation of this kernel by \[ K(x,t)= \sum_{i \in I} \sigma_i f_i(t) \overline{f_i(x)}, \quad x,t \in X, \] which is called Mercer's representation for \(K\), where \(\sigma_i >0\) and \(\{f_i\}\) is a particular orthonormal basis associated to \(K\). The authors extend this result for the case when \(X\) is a measurable space, \(Y\) is a finite-dimensional vector space, the kernel \(K\) is a measurable function, and the corresponding reproducing kernel Hilbert space is separable. In Theorem 3.4, they show the Mercer representation of the kernel \(K\) for this case, and in Proposition 3.5, they analyse the relation between \(K\) and the scalar reproducing kernels associated to the diagonal blocks of \(K\).