Fourier sums with exponential weights on \((-1,1):L^1\) and \(L^\infty\) cases (Q643345)

Let \(v^\lambda(x)=(1-x)^\lambda\) and \(w(x)=\exp(-(1-x^2)^{-\alpha})\), with \(\alpha>0\), \(\lambda \geq 0\), for \(x\in (-1,1)\). The \(m\)th Fourier sum of the function \(f\in L_\sigma^1\) is \(S_m(\sigma,f)=\sum_{k=0}^{m-1}c_k p_k(\sigma)\), where \(\sigma=v^\lambda w\) and \(p_k(\sigma)\) is the \(k\)th orthonormal polynomial with respect to \(\sigma\). Let \(\chi_\theta\) be the characteristic function of the interval \([-a_{\theta m},a_{\theta m}]\), \(a_m=a_m(\sqrt{\sigma})\), \(\theta \in(0,1)\) is fixed, and \(1-a_m \sim m^{-1/(\alpha+1/2)}\). The authors study the behavior of the sequence \(\{\chi_\theta S_m(\sigma,\chi_\theta f)\}_{m\in \mathbb{N}}\) in the function spaces \(L_u^1=\{f\,:\,fu\in L^1(-1,1)\}\) and \(L_u^\infty=\{f\in C(-1,1);\;\lim_{x\to \pm 1}f(x)u(x)=0\}\), where \[ u(x)=(1-x^2)^\mu \exp(-0.5(1-x^2)^{-\alpha}), \] with \(\alpha>0\), and \(\lambda,\mu \geq 0\). One of the main results establishes that for \(\theta\in(0,1)\) and every \(f\in L_u^\infty\), the following estimate \[ \|\chi_\theta S_m(\sigma,\chi_\theta f)u\|_\infty \leq C_\theta(\log m)\|\chi_\theta f u\|_\infty, \] with \(C_\theta=\mathcal{O}(\log^{-1/2}(1/\theta))\) independent of \(m\) and \(f\) holds true if and only if \(1/4\leq \mu-\lambda/2\leq 3/4\). In turn, this restriction for \(\lambda\) and \(\mu\) allows to get an upper estimate for \(\|f-\chi_\theta S_m(\sigma,\chi_\theta f)u\|_\infty\) in terms of the best polynomial approximation to \(f\) in the norm \(\|g\|=\sup_{x\in(-1,1)}|g(x)u(x)|\). An analogous result is obtained for \(L_u^1\).