On \(L_{\mathbb{R}}^2\)-best rational approximants to Markov functions on several intervals (Q2124220)

Define a Markov function \(f(z)=\int (z-x)^{-1}d\mu(x)\) where \(\mu\) is a Szegő measure with \(\mathrm{supp}(\mu)=\Delta\), consisting of \(g+1\) subintervals of \((-1,1)\). It is square integrable and has real Fourier coefficients, i.e., it belongs to \(L_{\mathbb{R}}^2= H_{\mathbb{R}}^2 \oplus \bar{H}_{\mathbb{R}}^2\), expressing it as an orthogonal composition of Hardy spaces of functions analytic inside and outside of the unit circle respectively. This paper is about approximating \(f\in\bar{H}_{\mathbb{R}}^2\) by a rational function \(r_n\) of type \([n-1/n]\) with real Fourier coefficients and \(n\) poles in the open unit disk \(\mathbb{D}\). Via the orthogonality of the denominator, a link is made with multipoint Padé approximation. For \(g=0\) and \(\Delta=[a,b]\), a conformal map from \(\mathbb{C}\setminus [a,b]\) to \(\mathbb{D}\) results in a formula for the error. This eventually leads to an expression for the asymptotics of the error, locally uniformly for \(n\to\infty\) in \(\bar{\mathbb{C}}\setminus([a,b]\cup[a,b]^{-1})\), with \([a,b]^{-1}=\{1/z:z\in[a,b]\}\). See [\textit{L. Baratchart} et al., J. Approx. Theory 108, No. 1, 53--96 (2001; Zbl 0973.41007)]. The main result of this paper generalizes this behaviour for \(g\ge1\) locally uniformly in \(\bar{\mathbb{C}}\setminus (\tilde\Delta\cup \tilde\Delta^{-1})\) where \(\tilde\Delta\) is the convex hull of \(\Delta\), including the gaps between the intervals. The proof requires modifying the elements of the original proof, relying on potential theory and techniques used in Riemann-Hilbert problems.