Korovkin tests, approximation, and ergodic theory (Q4501039)

Let \(A_n(f)\) be a \(k(n)\times k(n)\) matrix whose elements are blocks of size \(s\times t\) and which are spectrally distributed as the the \(s\times t\) matrix valued function \(f\) which means that for any \(F\) bounded and continuous on a probability space \((\Omega,\mu)\) the following ergodic property holds \[ \lim_{n\to\infty} {1 \over s k(n)} \sum_{i=1}^{sk(n)} F(\sigma_i(A_n(f))) =\int_{\Omega} \sum_{j=1}^s F(\sigma(f(x)))d\mu(x) \] where the sigma's denote singular values. Let \(X_n\) be the best approximant for \(A_n(f)\) in Frobenius norm. It is shown as one of the main results in this paper that the sequence \(X_n\) is distributed as \(f\) and that the elements in the error sequence \(A_n(f)-X_n\) have spectra that cluster around 0. For the proofs, first some Korovkin theorems are shown, i.e., it is sufficient to perform a simple test for a finite number of polynomial test functions \(F\) in the above ergodic property. Results of the type given here are used in numerical function approximation and in preconditioning for conjugate gradient-like iterative methods.