Exact values of Kolmogorov widths of classes of Poisson integrals (Q390537): Difference between revisions

Let us make the following notations : \(C\) is the space of all \(2\pi\)-periodic and continuous functions endowed with the uniform norm \(\|\cdot\|_{C}\), \(L_{1}=L\) is the space of all \(2\pi\)-periodic and integrable functions endowed with the norm \(\|f\|_{1}=\|f\|_{L}=\int_{-\pi}^{\pi}|f(t)|d t\), \(L_{\infty}\) is the space of all \(2 \pi\)-periodic, measurable and essentially bounded functions endowed with the norm \(\|f\|_{\infty}=ess sup\{|f(t)| ; t\in \mathbb{R}\}\), \(C^{q}_{\beta, p}\), \(p=1\), \(p=\infty\), is the space of functions \(f\) of the form \(f(x)=c_{0}+(P^{q}_{\beta}*\varphi)(x)=c_{0}+\frac{1}{\pi}\int_{-\pi}^{\pi}P_{q, \beta}(x-t)\varphi(t)d t\), where \(P_{q, \beta}(t)=\sum_{k=1}^{\infty}q^{k}\cos\left (k t-\frac{\beta \pi}{2}\right )\) (\(q\in (0, 1)\), \(\beta\in \mathbb{R}\)) is the Poisson kernel, \(\|\varphi\|_{p}\leq 1\), \(\varphi \bot 1\), \[ E_{n}(C^{q}_{\beta, p})_{X}=\sup \{\inf \{\|f-T_{n-1}\|_{X} ; T_{n-1}\in {\mathcal T}_{2n-1}\} ; f\in C^{q}_{\beta, p}\}, \] \[ d_{n}(C^{q}_{\beta, p}, X)=\inf\{\sup\{\inf\{\|f-g\|_{X} ; g\in F_{n}\} ; f\in C^{q}_{\beta, p}\} ;F_{n}\subset X\}, \] where \(X=C\) if \(p=\infty\), \(X=L\) if \(p=1\), \({\mathcal T}_{2n-1}\) denotes the space of all trigonometric polynomials of degree \(\leq 2n-1\) and the first infimum in the above definition of the Kolmogorov width is considered over all \(n\)-dimensional linear subspaces \(F_{n}\) of \(X\). The main result can be summarized as follows. Theorem. Let \(q\in (0, 1)\). Then, for all \(\beta\in \mathbb{R}\) and \(n\geq n_{q}\) we have \[ d_{2n}(C^{q}_{\beta, \infty}, C)=d_{2n-1}(C^{q}_{\beta, \infty}, C)=d_{2n-1}(C^{q}_{\beta, 1}, L)=E_{n}(C^{q}_{\beta, \infty})_{C}=E_{n}(C^{q}_{\beta, 1})_{L} \] \[ =\frac{4}{\pi}\cdot \left |\sum_{\nu=0}^{\infty}\frac{q^{(2\nu+1)n}}{2\nu+1}\sin\left ((2\nu+1)\theta_{n} \pi-\frac{\beta \pi}{2}\right )\right |, \] with \(\theta_{n}\) the unique root in \([0, 1)\) of the equation \(\sum_{\nu=0}^{\infty}q^{(2\nu+1)n}\cos\left ((2\nu+1)\pi t -\frac{\beta \pi}{2}\right )=0\). In particular, for all \(n\geq n_{q}\) and \(\beta\in \mathbb{Z}\), denoting by \(A\) any of the quantities which appear in the first sequence of equalities, we have \[ A = \frac{4}{\pi}\cdot \arctan(q^{n}) \text{ if } \beta=2 k \, \text{ and } \, A=\frac{2}{\pi}\ln\left (\frac{1+q^{n}}{1-q^{n}}\right ) \text{ if } \beta=2k-1. \] Here, \(n_{q}\) is the smallest number \(\geq 9\) satisfying a certain inequality in the paper.