Convergence of subdiagonal Padé approximations of \(C _{0}\)-semigroups (Q2441324): Difference between revisions

Let \(r_n(z)\) be the \(n\)th subdiagonal Padé approximant to \(e^z\) (numerator degree \(n\), denominator degree \(n+1\)). All the poles are in the right half plane and the approximants are holomorphic and bounded by 1 in the left half plane. If \(T=(T(t))_{t\geq0}\) is a uniformly bounded \(C_0\)-semigroup on the positive real axis for a Banach space \(X\) with generator \(-A\), then \(r_n(-tA)x\) converges to \(T(t)x\) for \(x\in D(A)\) (i.e., \(x\) in the domain of \(A\)). The main result of this paper is a generalization: this convergence holds with rate \(O(n^{-\alpha+1/2})\) locally uniformly on \(D(A^\alpha)\) for \(\alpha>1/2\). In this case, the approximation without scaling and squaring (see \textit{F. Neubrander} et al. [Discrete Contin. Dyn. Syst. 33, No. 11--12, 5305--5317 (2013; Zbl 1278.65194)]) converges as fast as the classical method with scaling and squaring. By assuming extra conditions on \(T\) or \(A\), the result is even sharpened to \(O(n^{-\alpha})\) convergence. The proof relies on Hille-Phillips calculus and a careful analysis of the error of the Padé approximants. Applications involve extensions to intermediate (Favard) spaces and the inversion of the vector valued Laplace transform.