No label defined (Q4490816)

Let \(f\) be a real valued continuous function on a compact subset \(F\) of the real line, let \({\mathcal P}_n\) denote the collection of all polynomials with real coefficients of degree at most \(n\), and let \(E_n(f;F)= \min_{P\in{\mathcal P}_n} \max_{x\in F}|f(x)-P(x)|\) denote the best polynomial approximation to \(f\) by polynomials in \({\mathcal P}\). A classical result of S. N. Bernstein states that, for \(p>0\), the following limits exist: \(\mu(p)= \lim_{n \to \infty}n^p E_n(|x|^p; [-1,1])\), and \(\mu_*(p)= \lim_{n\to \infty} n^p E_n(x|x|^{p-1}; [-1,1])\). In addition, Bernstein proved that, for \(|c|<1\) and \(p>0\), \[ \lim_{n\to \infty}n^pE_n \bigl(|x-c|^p; [-1,1]\bigr)= (1-c^2)^{p/2} \mu(p). \] For a compact set of the form \(F=\cup^n_{i =1} [a_i,b_i]\), where \(a=a_1< b_1<a_2 <\cdots <a_n<b_n=b\), set \[ h_F(x)= {1\over \sqrt{(b-x) (x-a)}} \prod^{n-1}_{i=1} {|x-\lambda_i |\over \sqrt{(x-b_i) (x-a_{i+1})}},\;x\in F, \] where the points \(\lambda_i\) are chosen so that both \(b_i<\lambda_i <a_{i+1}\) and \(\int^{a_{i+1}}_{b_i} h_F(x)dx =0\), \(1\leq i\leq n-1\). The general compact set \(F\) has the form \(F=[a,b]-\cup^\infty_{i=1} (b_i, a_{i+1})\). For each positive integer \(n\), let \(F_n=[a,b]-\cup^{n-1}_{i=1} (b_i, a_{i+1})\) and define \(h_F(x)= \lim_{n\to \infty}h_{F_n} (x)=\sup_n h_{F_n}(x)\), \(x\in F\). The main result of this manuscript is that, for an arbitrary compact subset \(F\) of the real line, and for \(p>0\) and \(c\in F\), the following limits hold: \[ \lim_{n\to \infty}n^p E_n\bigl(|x-c|^p; F\bigr)= \mu(p) h_F^{-p}(c) \text{ and }\lim_{n\to \infty} n^pE_n \bigl((x-c) |x-c|^{p-1}; F\bigr)=\mu_* (p)h_F^{-p}(c). \] Further, if \(c\) is not a point of density of \(F\), then \(h_F(c)=\infty\), and thus both of the limits above are 0. The proof of this main result is obtained by an exhaustive study of the function \(h_F(x)\), using, as a starting point, the Chebyshev polynomials of the form \(P(z)= \cos (nH_n(z)\), where \(H_n(z)= \int_a^z\widetilde h_n(\zeta) d\zeta\) and \[ \widetilde h_n(z)= {(z-\lambda_1) \cdots(z- \lambda_{s-1}) \over\sqrt {(b-z)(z-a)(z-b_1) \cdots (z-a_s)}}, \] where \(\lambda_k\in (b_k,a_{k+1})\). The function \(h_F(z)\) is obtained as an appropriate limit of the functions \(\widetilde h_n\), and various properties of the function \(h_F(z)\), including properties of the analytic continuation of \(h_F(z)\), are developed by careful and sometimes lengthy calculations, leading eventually to a complete proof of the main result.