Exact values of Bernstein \(n\)-widths for some classes of convolution functions (Q938475)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Exact values of Bernstein \(n\)-widths for some classes of convolution functions
scientific article

    Statements

    Exact values of Bernstein \(n\)-widths for some classes of convolution functions (English)
    0 references
    0 references
    0 references
    19 August 2008
    0 references
    If \((X,\|\cdot\|_X)\) is a normed linear space with norm \(\|\cdot\|_X\) and if \(A\subset X\) is a closed, convex, centrally symmetric subset of \(X\), so that \(x\in X\) implies \(-x\in X\), then the Bernstein \(n\)-width \(b_n(A;X)\) of the subset \(A\) is defined by \(b_n(A;X)= \sup_{B\in{\mathfrak X}_n}\sup\{\lambda:\lambda S(B)\subseteq A\}\), with \(S(B)=\{x\in B:\| x\|_X\leq 1\}= B\cap{\mathcal B}\), where \({\mathcal B}\) is the unit sphere of \(X\) and \({\mathfrak X}_n\) consists of subsets of \(X\) with dimension \(\geq n+1\). A real \(2n\)-periodic continuous function \(G\) is said to satisfy the \(B\)-property if, for an arbitrary partition \(0< t_1< t_2<\cdots< t_n< 2\pi\) and integer \(m\), the subspace \[ {\mathcal V}_m= \Biggl\{b+ \sum^m_{j=1} b_jG(.-t_i): \sum^m_{j=1} b_j =0\Biggr\} \] is of dimension \(m\), and a similar function \(K\) is said to be a cyclic variation diminishing kernel of order \(2m-1\) \((CVD_{2m-1})\) if for \(n= 1,2,\dots,m\), the number \(\sigma_n\in [-1,1]\) may be determined such that \(\sigma_n \det(K(x_j- y)k))_{j,k=1}^{2n-1}\geq 0\) for all partitions \(x_1<\cdots< x_{2n-1}< x_1+ 2\pi\), \(y_1<\cdots< y_{2n-1}< y_1+ 2\pi\). The convolution \((g * h)\) of functions \(g,h\) is the integral expression \(\int_T g(x- y) h(y)\,dy\). The main theorems of this paper provide exact values for Bernstein \(n\)-widths of spaces \({\mathfrak B}_p\), \({\mathfrak K}_p\), defined by \[ \begin{aligned}{\mathfrak B}_p&= \Biggl\{f: f(x)= (G* h)(x)+ a,\;a\in\mathbb R,\;\| h\|_p\leq 1,\;\int_T h(t)\,dt= 0\Biggr\},\\ {\mathfrak K}_p&= \Biggl\{f: f(x)= (K* h)(x),\;\| h\|_p\leq 1,\;\int_T h(t)\,dt= 0\Biggr\}. \end{aligned} \] In particular, it is shown that \(b_{2n-1}({\mathfrak B}_p; L^p)= \lambda_n(p,p,G)\), \(b_{2n-1}({\mathfrak K}_p; L^p)= \lambda_n(p,p,K)\), where \[ \lambda_n(p,q,H)= \sup\{\| H* h\|_q: h\in D_n\}, \] \[ D_n= \{h: h(x+ \pi/n)= -h(x),\;h(x)(\sin nx)> 0,\;\| h\|_p\leq 1\}. \]
    0 references
    0 references
    convolution functions
    0 references
    Sobolev spaces
    0 references
    0 references
    0 references