Integral approximation of the characteristic function of an interval and the Jackson inequality in \(C(\mathbb T)\) (Q643728)

Setting \[ \mathcal{X}_h (t)=\frac{1}{2h}\chi_h(t)\,, \] where \[ \chi_h(t)= \sum_{j\in\mathbb{Z}}\chi_{(-h,h)}(t+2\pi j)\,, \] with \(\chi_{(-h,h)}(x)=1\) if \(|x|<h,\) \(h>0,\) and \(\chi_{(-h,h)}(x)=0\) otherwise, and \[ E_{m-1}(f)_L=\inf\left\{\|f-g\|_L\,, \;g\in\mathcal{T}_{m-1}\right\}\,, \] in some previous works, the authors proved that \[ E_{m-1}(\mathcal{X}_h)_L=1\qquad \text{ if } 0<h\leq \frac{\pi}{2m}\,, \] \[ E_{m-1}(\mathcal{X}_h)_L\leq \frac{\pi}{2 m h}\qquad \text{ if } \frac{\pi}{2m}<h\leq\pi\,, \] and, in particular, \[ E_{m-1}(\mathcal{X}_h)_L=\frac{\pi}{2 m h_j}\,,\qquad h_j=\frac{(2j-1)\pi}{2 m}\,, \;j\geq 2\,. \] From these relations they obtain the Jackson theorem for continuous and periodic functions in terms of the second order modulus of smoothness. The method is essentially the following. If \[ f \odot g =\int_{-\pi}^{\pi} f(x-t)g(t)dt \] is the convolution product, the the Steklov function can be written as \[ (S_h f)(x)=\frac{1}{2 h}\int_{-h}^h f(x-t)dt=f\odot \mathcal{X}_h(x)\,. \] In particular, for any function \(\varphi\in \mathcal{T}_{m-1}^{\perp}\) (orthogonal complement of \(\mathcal{T}_{m-1}\)) the equality \[ (S_h\varphi)(x)=\varphi\odot (\mathcal{X}_h-g) \qquad \forall\varphi\in \mathcal{T}_{m-1} \] holds, and, for \(\varphi\neq 0\), \[ \frac{\|S_h\varphi\|_C}{\|\varphi\|_C}\leq E_{m-1}(\mathcal{X}_{h})_L\,. \] The authors show that the latter inequality is sharp, and, therefore, \[ \sup_{\varphi\in\mathcal{T}_{m-1}^\perp,\;\varphi\neq 0}\frac{\|S_h\varphi\|_C}{\|\varphi\|_C}= E_{m-1}(\mathcal{X}_{h})_L\,. \] Moreover, setting \[ W_2(f,h,x)=(f-S_h)(x)=f(x)-f\odot \mathcal{X}_h(x)=-\frac{1}{2 h} \int_0^h\left[f(x-t)-2 f(x)+f(x+t)\right] dt \] and \[ W_2(f,h)=\sup_x W_2(f,h,x)\leq \frac{1}{2}\omega_2(f,h)\,, \] the authors prove that, for any continuous function \(f\), there exists \(\tau_f\in \mathcal{T}_{m-1}\) such that \[ f-\tau_f=(f-\tau_f)\odot(\mathcal{X}_h-g_h)+W_2(f,h,\cdot)\,, \] where \(g_h\) is the best \(L-\)approximation of \(\mathcal{X}_h\). It follows that \[ \|f-\tau_f\|_C\leq \|f-\tau_f\|_C E_{m-1}(\mathcal{X}_h)_L+W_2(f,h)\,, \] and, since \(E_{m-1}(\mathcal{X}_h)_L<1\) for \(h>\pi/2m\), \[ \|f-\tau_f\|_C\leq\frac{W_2(f,h)}{1-E_{m-1}(\mathcal{X}_h)_L}\,,\qquad h>\frac{\pi}{2 m}\,, \] whence they deduce the theorem, which is the main result of the paper.