Uniform polynomial approximation with \(A^{\ast}\) weights having finitely many zeros (Q892349)

The author establishes matching direct and inverse approximation theorems in the uniform norm with special \(A^*\) weights having finitely many zeros. To be more precise, we need the notation below. Following the paper under review, let \(\displaystyle\rho(h,x):=h \sqrt{1-x^2} +h^2\), \(\varphi(x):=\sqrt{1-x^2}\), \(\displaystyle\mathbb Z_M :=\{(z_j)_{j=1}^M : -1\leq z_1\leq \dots \leq z_M\leq 1 \}\), and for \(\mathcal Z \in \mathbb Z_M \) put \(\displaystyle\mathcal J_{A,h}(\mathcal Z):=\{x\in [-1,1]: |x-z_j|\geq A \rho(h,z_j), j=1,\dots,M\}\). Considering the \(A^*\) weights introduced by Mastroianni and Totik, the author introduces a subclass of weights denoted by \(\mathcal W^*\). A weight \(w\in A^*\) belongs to \(\mathcal W^*\) if there exists \(c_*>0\) such that, for any \(\varepsilon>0\), \(x,y\in [-1,1]\) with \(|x-y|\leq \rho(\varepsilon,x)\) and \(\operatorname{dist}([x,y],z_j)\geq \rho(\varepsilon,z_j)\) for all \(j=1,\dots,M\), the following condition holds \[ c_* w(y) \leq w(x) \leq c_*^{-1}(y). \] Denote the symmetric difference by \[ \Delta_h^r(f,x,J):= \sum_{i=0}^r \binom{r}{i}(-1)^{r-i} f(x-rh/2+ih) \] if \( [x-rh/2,x+rh/2]\subset J\) and \(\Delta_h^r(f,x,J):=0\) otherwise. The weighted Ditzian-Totik modulus of smoothness in this case is \[ \omega_\varphi^t(f,t)_w := \sup_{0<h\leq t} \left\| w(.) \Delta_{h\varphi(.)}^r(f,.,[-1,1])\right\|, \] and the main part of the weighted modulus of smoothness is \[ \Omega_\varphi^r(f,A,t;\mathcal Z)_w:=\sup_{0<h\leq t} \left\| w(.) \Delta_{h\varphi(.)}^r(f,.,{\mathcal J}_{A,h}) \right\|. \] For \(A,B,t>0\), the complete weighted modulus of smoothness is defined as \[ \omega_\varphi^r(f,A,B,t; \mathcal Z )_w:=\Omega_\varphi^r(f,A,t;\mathcal Z)_w + \sum_{j=1}^M E_r(f,{\mathcal Z}^j_{B,t})_w \] where \(E_r(f,{\mathcal Z}^j_{B,t})_w=\inf\{\| w(f-q)\|_{{\mathcal Z}^j_{B,t}} : \roman{deg}(q)\leq r-1\}\) is the error of best approximation of \(f\) on \({\mathcal Z}^j_{B,t}\) with weight \(w\). The direct theorem (Theorem 6.1) states the following. Let \(w\in \mathcal W^*(\mathcal Z)\), \(r,\nu_0\in \mathbb N\), \(\nu_0 \geq r\), \(\vartheta>0\), \(wf\in L_\infty\) and \(B>0\). Then there exists \(N\in \mathbb N\), \(N=N(r,\vartheta,w)\) such that for all \(n\geq N\) there exists a polynomial \(P_n\) with degree at most \(n\) such that \[ \left\| w(f-P_n)\right\| \leq c\, \omega_\varphi^r(f,1,B,\vartheta/n)_w \] and \[ \left\| w\varphi^\nu P_n^{(\nu)}\right\| \leq c\, n^\nu\, \omega_\varphi^r(f,1,B,\vartheta/n)_w, \;\;r\leq\nu\leq \nu_0, \] where \(c=c(r,\nu_0,B,\vartheta,w)\). The inverse theorem (Theorem 7.1) is as follows. Let \(\mathcal Z \in {\mathbb Z}_M\), \(w\in \mathcal W^*(\mathcal Z)\), \(wf\in L_\infty\), \(A,B>0\) and \(r,n\in \mathbb N\). Then \[ \omega_\varphi^r(f,A,B,1/n)_w \leq c\, n^{-r}\, \sum_{k=1}^n k^{r-1} E_k(f,[-1,1])_w, \] where \(c=c(r,A,B,w)>0\) and \( E_k(f,[-1,1])_w=\inf\{\| w(f-q)\|: \roman{deg}(q)\leq k-1\}\). Furthermore, the author also gives an equivalence-type result involving realization functionals as well as several technical lemmas and propositions.