Processing math: 4%

数学物理学报, 2019, 39(4): 720-729 doi:

论文

某一卷积函数类在Orlicz空间内宽度的精确估计

孙芳美,, 吴嘎日迪,

Exact Estimate of n-Widths of a Convolution Function Class in Orlicz Spaces

Sun Fangmei,, WuGaridi ,

通讯作者: 吴嘎日迪, E-mail:wgrd@imnu.edu.cn

收稿日期: 2018-03-13  

基金资助: 国家自然科学基金.  11761055
内蒙古自治区自然科学基金.  2017MS0123

Received: 2018-03-13  

Fund supported: the NSFC.  11761055
the Autonomous Region Natural Science Foundation of Inner Mongolia.  2017MS0123

作者简介 About authors

孙芳美,E-mail:18fmsun@stu.edu.cn , E-mail:18fmsun@stu.edu.cn

摘要

该文讨论了由实系数线性微分算子定义的2π周期函数类˜mr,p在Orlicz空间内的宽度问题.得到该函数类在Orlicz空间内的n-K宽度,n-G宽度,n-L宽度,n-B宽度的精确值和相应的极子空间.

关键词: Orlicz空间 ; 宽度 ; 极子空间 ; 样条函数

Abstract

In this paper, we study the n-widths of a 2π-periodic convolution function class defined by linear differential operators with real coefficient in Orlicz spaces, and obtain the exact values of n-K width, n-G width, n-L width, n-B width of this function class and its corresponding optimal subspaces.

Keywords: Orlicz space ; n-Width ; Optimal subspace ; Spline function

PDF (340KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

孙芳美, 吴嘎日迪. 某一卷积函数类在Orlicz空间内宽度的精确估计. 数学物理学报[J], 2019, 39(4): 720-729 doi:

Sun Fangmei, WuGaridi . Exact Estimate of n-Widths of a Convolution Function Class in Orlicz Spaces. Acta Mathematica Scientia[J], 2019, 39(4): 720-729 doi:

1 引言

文中用M(u)N(v)表示互余的N函数,关于N -函数的定义及其性质参见文献[1].由文献[1]知, M(u)为N -函数当且仅当存在定义在[0,+)上的实值函数p(t),使得M(u)=|u|0p(t)dt,其中p(t)满足下列条件

1p(t)为右连续的单调递增函数;

2t>0时, p(t)>0;

3p(0)=0, p(+)=+.

这时p(u)M(u)的右导数.文中由N -函数N(u)生成的Orlicz类LN是指满足

ρ(u,N)=2π0N(u(x))dx<+

的可测函数的全体{u(x)};由N -函数M(u)生成的Orlicz空间LM是指具有有限的Orlicz范数

(1.1)

的可测函数的全体 \{u(x)\} . L^{\ast}_{M} 上还可以赋予与Orlicz范数 (1.1) 等价的Luxemburg范数

\| u\|_{(M)} = {\rm inf}\bigg\{\alpha>0\bigg|\int_{0}^{2\pi} M(\frac{u(x)}{\alpha})\, {\rm d}x\leq1\bigg\}.
(1.2)

以下分别用 L^{\ast}_{M} L^{\ast}_{(M)} 表示带有Orlicz范数(1.1)和Luxemburg范数(1.2)的Orlicz空间.由文献[1]知, Orlicz范数(1.1)也可以由下式计算

\|u{{\|}_{M}}=\begin{matrix} \inf \\ k>0 \\\end{matrix}\ \frac{1}{k}(1+\int_{0}^{2\pi }{M}(ku(x))\text{d}x),
(1.3)

且存在 k>0 ,满足 \int_{0}^{2\pi} N(p(k|u(x)|))\, {\rm d}x = 1 ,使得

{\| u\|_{M} = \frac{1}{k}\bigg(1+\int_{0}^{2\pi} M(ku(x))\, {\rm d}x\bigg)}.

由文献[2]有如下定义.

定义1.1  设 \Lambda\subset X 是一中心对称集, \Lambda X 内的n维Kolmogorov宽度定义为

{{d}_{n}}(\Lambda ;X)=\begin{matrix} \inf \\ {{L}_{n}} \\\end{matrix}\begin{matrix} \sup \\ x\in \Lambda \\\end{matrix}\begin{matrix} \inf \\ \varphi \in {{L}_{n}} \\\end{matrix}\|x-\varphi {{\|}_{X}},
(1.4)

其中 \|\cdot\|_{X} X 内的范数, L_{n} X 中任一n维线性子空间.简称n-K宽度.能使(1.4)式内下确界实现的任一n维线性子空间 L_{n}^{o} 称为 \Lambda X 内的n-K宽度的极子空间.

定义1.2  设 \Lambda\subset X 是一中心对称集, \Lambda X 内的n维Gelfand宽度为

{{d}^{n}}(\Lambda ;X)=\begin{matrix} \inf \\ {{L}^{n}} \\\end{matrix}\begin{matrix} \sup \\ f\in \Lambda \cap {{L}^{n}} \\\end{matrix}\|f{{\|}_{X}},
(1.5)

其中 L^{n} X 中任一n余维子空间.简称n-G宽度.能使(1.5)式内下确界实现的任一n余维线性子空间都称为 \Lambda X 内的n-G宽度的极子空间.

定义1.3  设 \Lambda\subset X 是一中心对称集, \Lambda X 内的 n 维Linear宽度为

{{d}_{n}}\prime (\Lambda ;X)=\begin{matrix} \inf \\ {{L}_{n}} \\\end{matrix}\begin{matrix} \inf \\ A \\\end{matrix}\begin{matrix} \sup \\ x\in \Lambda \\\end{matrix}\|x-Ax{{\|}_{X}},
(1.6)

其中A是 {\rm span}\{\Lambda\}\rightarrow\Lambda 的任意一个线性有界算子.简称n-L宽度.能使(1.6)式内下确界实现的任一n维线性子空间 L_{n}^{o} 称为 \Lambda X 内的n-L宽度的极子空间.而对一个确定的极子空间 L_{n}^{o} ,把可以实现下确界的线性有界算子称作 \Lambda 的n-L宽度的最佳线性算子.

定义1.4  设 \Lambda\subset X 是一中心对称闭凸集. n\geq1, L_{n}\subset X 是一n维线性子空间.称

\beta(\Lambda;L_{n})_{X} = {\rm sup}\{\varepsilon>0|\varepsilon B_{X}\cap L_{n}\subset\Lambda\}

L_{n} \Lambda 的截集的内切球半径.当不存在正的 \varepsilon ,使 \varepsilon B_{X}\cap L_{n}\subset\Lambda 成立时,规定 \beta = 0 .变动 L_{n} ,使 \beta 尽量大,这就导致定义

{{b}_{n}}(\Lambda ;X)=\begin{matrix} \sup \\ {{L}_{n}} \\\end{matrix}\beta {{(\Lambda ;{{L}_{n}})}_{X}}
(1.7)

\Lambda X 内的n维Bernstein宽度.简称n-B宽度.如果存在 X 的一个n维线性子空间 L_{n}^{o} 使(1.7)式中的上确界达到,就称其为 \Lambda X 内的n-B宽度的极子空间.

给定 r 次实系数多项式

{{P}_{r}}(\lambda )=\prod\limits_{s=1}^{k}{({{\lambda }^{2}}-2{{\alpha }_{s}}\lambda +\alpha _{s}^{2}+\beta _{s}^{2})}\prod\limits_{j=1}^{r-2k}{(\lambda -{{\lambda }_{j}});}

P_{r}^{*}(\lambda )=\prod\limits_{s=1}^{k}{({{\lambda }^{2}}+2{{\alpha }_{s}}\lambda +\alpha _{s}^{2}+\beta _{s}^{2})}\prod\limits_{j=1}^{r-2k}{(\lambda +{{\lambda }_{j}}), }

其中, k\geq0 , \lambda_{j}, \alpha_{s}, \beta_{s}\in R , \beta_{s}>0 ,令 \beta = \max_{1\leq s\leq k}\beta_{s} .

\tilde{W}^r 代表满足 f^{(r-1)} 绝对连续且以 2\pi 为周期的连续函数.记

\tilde{m}_{r, M} = \bigg\{f\in\tilde{W}^{r}|f(x) = G_{r}\ast h(x), \| h\|_{M}\leq1\bigg\},

这里, G_{r}(x) = \frac{1}{2\pi}\sum\limits^{+\infty}_{v = -\infty} \frac{{\rm e}^{{\rm i}vx}}{P_{r}({\rm i}v)}, {\rm i} = \sqrt{-1} , \sum\prime 表示当 P_{r}(0) = 0 时,缺 v = 0 的项.为了计算方便,规定:除0以外的任意虚整点i k\ (k = \pm1, \pm2, \cdots) 都不是 P_{r}(\lambda) = 0 的根.则 G_{r}(x) 为对应算子 P_{r}(D) 的广义Bernoulli核.

房艮孙在文献[3]中给出 m_{r, \infty} 的渐进宽度的同时,引出了 \tilde{m}_{r, \infty} 类.随后毕宁在文献[4]中定义了 \tilde{m}_{r, p} 类,并讨论了 \tilde{m}_{r, p} L_{p} 空间内的Kolmogorov宽度、Gelfand宽度、Linear宽度和Bernstein宽度,得到了 0<\beta<\frac{1}{4} , 0<p<\infty , n\geq1 时, \tilde{m}_{r, p} 类的2n宽度的精确估计和相应的极子空间.以上学者们都是在 L_{p} 空间中计算函数类的宽度,本文主要在Orlicz空间中研究新定义的函数类 \tilde{m}_{r, M} 的宽度问题,利用Bernstein宽度可以控制Kolmogorov宽度和Gelfand宽度的下方估计,以及Linear宽度可以控制Kolmogorov宽度和Gelfand宽度的上方估计的思想,通过计算Bernstein宽度的下方估计和Linear宽度的上方估计来得到该函数类在Orlicz空间内的n-K宽度, n-G宽度, n-L宽度, n-B宽度的精确值,并给出了相应的极子空间.

2 相关引理

D_{n} = \{h(x)|h(x+\frac{\pi}{n}) = -h(x), x\in \mathbb{R}\}, \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n}

\Phi_{n, r}(x) = \int_{0}^{2\pi} G_{r}(x-t) {\rm sgn} \sin nt\, {\rm d}t

的单个零点(参见文献[5]).

引理2.1  对 n\geq1 ,存在 h_{n}\in D_{n} , \| h_{n}\|_{M} = 1 , \lambda_{n}>0 ,使

\begin{eqnarray*} \lambda_{n}& = &\| G_{r}\ast h_{n}\|_{M} = {\rm sup}\{\| G_{r}\ast h\|_{M}| \|h\|_{M}\leq1, h\in D_{n}\}\\ & = &{\rm sup}\bigg\{\frac{\| G_{r}\ast h\|_{M}}{\| h\|_{M}}\bigg| h\in D_{n}\cap L_{M, 2\pi}^{\ast}, h\neq0\bigg\}, \end{eqnarray*}

\int_{0}^{2\pi} p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\| G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))G_{r}(x-y)\, {\rm d}x = \lambda_{n}p\bigg(\frac{|h_{n}(y)|}{\|h_{n}\|_{(M)}}\bigg){\rm sgn}(h_{n}(y)), y\in \mathbb{R}.
(2.1)

  设 \lambda_{n} = {\rm sup}\{\| G_{r}\ast h\|_{M}| \|h\|_{M}\leq1, h\in D_{n}\} ,则存在 \{f_{m}\}_{m = 1}^{\infty}\subset D_{n} ,且 \|f_{m}\|_{M}\leq1 ,满足

\begin{matrix} \lim \\ m\to \infty \\\end{matrix}\|{{G}_{r}}*{{f}_{m}}{{\|}_{M}}={{\lambda }_{n}}.

由文献[6]中性质2,性质4知:存在子序列 \{f_{m_{j}}\}_{j = 1}^{\infty} g\in L_{M}^{\ast}[0, 2\pi] , \|g\|_{M}\leq1 ,使得对任意 s\in L_{(N)}^{\ast}[0, 2\pi] ,有

\begin{matrix} \lim \\ m\to \infty \\\end{matrix}\int_{0}^{2\pi} f_{m_{j}}(y)s(y)\, {\rm d}y = \int_{0}^{2\pi} g(y)s(y)\, {\rm d}y.

这里,特别取 s(y) = G_{r}(x-y) ,结合Lebesgue控制收敛原理,得

\begin{matrix} \lim \\ m\to \infty \\\end{matrix}\| G_{r}\ast f_{m_{j}}\|_{M} = \| G_{r}\ast g\|_{M} = \lambda_{n}.

现把 g(x) 2\pi 为周期延拓至 \mathbb{R} ,依然记作 g(x) ,要证 g\in D_{n} .

取任意 s\in L_{(N)}^{\ast}[-2\pi, 2\pi] ,由

\int_{-2\pi}^{2\pi} f_{m_{j}}(x)s(x)\, {\rm d}x = \int_{-2\pi+\frac{\pi}{n}}^{2\pi+\frac{\pi}{n}} f_{m_{j}}\bigg(x-\frac{\pi}{n}\bigg)s\bigg(x-\frac{\pi}{n}\bigg)\, {\rm d}x = -\int_{-2\pi}^{2\pi} f_{m_{j}}(x)s\bigg(x-\frac{\pi}{n}\bigg)\, {\rm d}x,

两边同时取极限,再做变量替换,令 x = x+\frac{\pi}{n} ,有

\int_{-2\pi}^{2\pi} g(x)s(x)\, {\rm d}x = \int_{-2\pi}^{2\pi} g(x)s\bigg(x-\frac{\pi}{n}\bigg)\, {\rm d}x = -\int_{-2\pi}^{2\pi} g\bigg(x+\frac{\pi}{n}\bigg)s(x)\, {\rm d}x,

\int_{-2\pi}^{2\pi} \bigg[g(x)+g\bigg(x+\frac{\pi}{n}\bigg)\bigg]s(x)\, {\rm d}x = 0,

特别取 s(x) = {\rm sgn}\big(g(x)+g\big(x+\frac{\pi}{n}\big)\big) ,有

\int_{-2\pi}^{2\pi} \bigg|g(x)+g\bigg(x+\frac{\pi}{n}\bigg)\bigg|\, {\rm d}x = 0\Rightarrow g\bigg(x+\frac{\pi}{n}\bigg) = -g(x), {\rm a.e.}.

因此可得 g\in D_{n} . h_{n}(x) = g(x) ,显然有 \|h_{n}\|_{M} = 1 .任意 h(x)\in D_{n} ,考虑

F(t) = \frac{\|G_{r}\ast (h_{n}+th)\|_{M}}{\|h_{n}+th\|_{M}}, t\in \mathbb{R},

明显 \frac{{\rm d}F}{{\rm d}t}|_{t = 0} = 0 .结合文献[7]即可算得

\begin{eqnarray*} &&\int_{0}^{2\pi} p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))G_{r}\ast h(x)\, {\rm d}x\\ & = &\lambda_{n} \int_{0}^{2\pi} p\bigg(\frac{|h_{n}(x)|}{\|h_{n}\|_{(M)}}\bigg){\rm sgn}(h_{n}(x))h(x)\, {\rm d}x. \end{eqnarray*}

H_{n}(y) = \int_{0}^{2\pi} p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))G_{r}(x-y)\, {\rm d}x- \lambda_{n}p\bigg(\frac{|h_{n}(y)|}{\|h_{n}\|_{(M)}}\bigg){\rm sgn}(h_{n}(y)),

简单计算可得

\int_{0}^{2\pi} H_{n}(y)h(y)\, {\rm d}y = 0,

同证明 g\in D_{n} 的方法一致,易证得

H_{n}(y)\in D_{n}.

h(y) = H_{n}(y) ,得 H_{n}(y) = 0, a.e..再连续化,有 H_{n}(y) = 0 ,且 h(x) 连续.

引理2.2[4]  若 [a, b]\subset[0, \frac{\pi}{n}] , 0<\beta<1 , f\in C^{r-1} , f(a) = 0 , f^{(i)}(b) = 0, i = 0, 1, \cdots, r-1 ,则存在 \xi\in[a, b] ,使得

P_{r}(D)f(\xi) = 0,

其中, D = \frac{{\rm d}}{{\rm d}x} .

S_{2n, r-1} 为广义周期样条类(参见文献[8]),令

\hat{G}_{r}(x) = \left\{\begin{array}{ll} G_{r}(x), &P_{r}(0)\neq0, \\ G_{r}(x)-1, &P_{r}(0) = 0, \end{array}\right.

S_{2n, r-1} = \bigg\{f(x)\in C^{r-2}|f(x) = \sum\limits_{i = 1}^{2n} C_{i}\hat{G}_{r}(x-\frac{i-1}{n}\pi), C_{i}\in R, i = 1, \cdots, 2n\bigg\}.

引理2.3  若 0<\beta<\frac{1}{4} , n\geq1 ,那么引理1中的 h_{n}(x) [0, 2\pi) 内,有

1) h_{n}(x) 仅以 \{\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为其单零点;

2) G_{r}\ast h_{n}(x) 仅以 \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为其单零点.

  由 S_{2n, r-1} 对周期函数在 \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} 插值的存在唯一性,有

{{\hat{G}}_{r}}\left( \begin{matrix} \text{ }r, r+\frac{\pi }{n}, \cdots , r+\frac{2n-1}{n}\pi \\ \text{ }0, \frac{\pi }{n}, \cdots , \frac{2n-1}{n}\pi \\\end{matrix} \right)\ne 0.

\hat{K}_{r}(x, y) = \hat{G}_{r}\left(\begin{array}{c} x, r, r+\frac{\pi}{n}, \cdots, r+\frac{2n-1}{n}\pi\\ y, 0, \frac{\pi}{n}, \cdots, \frac{2n-1}{n}\pi \end{array}\right), h_{n}^{\alpha}(x) = h_{n}(x-\alpha),

\hat{M}_{r}(x, y) = \hat{K}_{r}(x, y)/\hat{G}_{r}\left(\begin{array}{c} r, r+\frac{\pi}{n}, \cdots, r+\frac{2n-1}{n}\pi\\ 0, \frac{\pi}{n}, \cdots, \frac{2n-1}{n}\pi \end{array}\right), \hat{L}_{r}(x, y) = |\hat{M}_{r}(x, y)|.

由文献[4]有,对于每个固定的 y\notin\{\frac{i-1}{n}\pi\}_{i = 1}^{2n} , \hat{K}_{r}(x, y) 作为 x 的函数在 [0, 2\pi) 内仅以 \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为其单零点,对于每个固定的 x\notin\{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} , \hat{K}_{r}(x, y) 作为 y 的函数在 [0, 2\pi) 内仅以 \{\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为其单零点.

因为 \hat{G}_{r}\ast h_{n}^{\alpha}(x+\frac{\pi}{n}) = -\hat{G}_{r}\ast h_{n}^{\alpha}(x) ,故存在 \alpha_{1}\in R ,使得 \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} \hat{G}_{r}\ast h_{n}^{\alpha_{1}}(x) 的零点,并且

\int_{0}^{2\pi}\hat{M}_{r}(x, y)h_{n}^{\alpha_{1}}(y)\, {\rm d}y = \int_{0}^{2\pi}\hat{G}_{r}(x-y)h_{n}^{\alpha_{1}}(y)\, {\rm d}y.

\hat{\mu}_{n} = {\rm sup}\{\|\hat{M}_{r}h\|_{M}| \|h\|_{M}\leq1, h\in D_{n}\},

类似引理2.1的证明,存在 \hat{h}_{n}\in D_{n} , \|\hat{h}_{n}\|_{M} = 1 ,有

\hat{\mu}_{n} = \|\hat{M}_{r}\hat{h}_{n}\|_{M} = {\rm sup}\{\|\hat{M}_{r}h\|_{M}| \|h\|_{M}\leq1, h\in D_{n}\},

{\rm sup}\{\|\hat{M}_{r}h\|_{M}| \|h\|_{M}\leq1, h\in D_{n}\} = {\rm sup}\{\|\tilde{L}_{r}\tilde{h}\|_{M}| \tilde{h}(x) = h(x){\rm sgn} {\rm sin} nx, h\in D_{n}, \|h\|_{M}\leq1\}.

因为前面 {\rm sup} 达到的 \hat{h}_{n}(x) ,满足 \hat{h}_{n}(x)\geq0 ,所以后面 {\rm sup} 达到的 \tilde{h}_{n}(x) ,满足 \tilde{h}_{n}(x){\rm sgn} {\rm sin}nx\geq0 ,即在 [0, 2\pi) \hat{h}_{n}(x) 只在 \{\frac{i-1}{n}\pi\}_{i = 1}^{2n} 处变号.又因 \hat{G}_{r}\ast \hat{h}_{n}^{\alpha}(x+\frac{\pi}{n}) = -\hat{G}_{r}\ast \hat{h}_{n}^{\alpha}(x) ,故存在 \alpha_{2}\in \mathbb{R} ,使得

\hat{M}_{r}\hat{h}_{n}^{\alpha_{2}}(x) = \hat{G}_{r}\ast \hat{h}_{n}^{\alpha_{2}}(x),

从而 \lambda_{n} = \hat{\mu}_{n} . h_{n}(x)\in D_{n} ,满足 h_{n}(x){\rm sgn} {\rm sin}nx\geq0 .由(2.1)式,有: h_{n}(x) 无零区间,当且仅当 G_{r}\ast h_{n}(x) 无零区间.故若 h_{n}(x) 无零区间,但除 \{\frac{i-1}{n}\pi\}_{i = 1}^{2n} 外还有其他零点.令

\begin{eqnarray*} F_{r}(y)& = &\int_{0}^{2\pi}p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))G_{r}(x-y)\, {\rm d}x\\ & = & \lambda_{n}p\bigg(\frac{|h_{n}(y)|}{\|h_{n}\|_{(M)}}\bigg){\rm sgn}(h_{n}(y)), \end{eqnarray*}

F_{r}(y) 至少有 2n+2 个零点,由文献[9]中Rolle定理,知

2n+2\leq S_{c}^{-}(P_{r}^{\ast}(D)F_{r}) = S_{c}^{-}(G_{r}\ast h_{n}),

其中, S_{c}^{-} 为循环变号数.再利用Rolle定理,有

2n+2\leq S_{c}^{-}(P_{r}(D)G_{r}\ast h_{n}) = S_{c}^{-}(h_{n}),

矛盾.若 h_{n}(x) 存在零区间,即存在 0\leq a<b\leq\frac{\pi}{n} ,使 h_{n}(x) = 0 , x\in[a, b] , [a, b] 必须是 [0, \frac{\pi}{n}] 的真子空间,因此 0<a b<\frac{\pi}{n} .我们讨论 0<a 的情况, b<\frac{\pi}{n} 的情况类似.

因为 h_{n}(x)\geq0 ,所以 x\in[0, \frac{\pi}{n}] ,故存在 0\leq c<a ,使 h(x)>0 , x\in(c, a) ; h_{n}(x) = 0 , x\in[a, b] ; h(c) = 0 ;因为 F_{r}^{(j)}(a) = 0, j = 0, 1, \cdots, r-1 , F_{r}(c) = 0 ,所以利用引理2.2可得,存在 c<\xi<a ,使得 P_{r}^{\ast}(D)F_{r}(\xi) = 0 = G_{r}\ast h_{n}(\xi). 又因为

P_{r}^{\ast}(D)F_{r}(x) = p\left(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\right){\rm sgn}(G_{r}\ast h_{n}(x)),

所以 G_{r}\ast h_{n}(\xi) = 0, \frac{{\rm d}^{v}}{{\rm d}x^{v}}G_{r}\ast h_{n}(x)|_{x = a} = 0, v = 0, 1, \cdots, r-1. 同理,存在 \xi<\eta<a ,使得 P_{r}(D)G_{r}\ast h_{n}(\eta) = h_{n}(\eta) = 0 ,这与 h_{n}(x)>0 , x\in(c, a) 矛盾.因此,在 [0, 2\pi) 内, h_{n}(x) 的单零点只有 \{\frac{i-1}{n}\pi\}_{i = 1}^{2n} .

G_{r}\ast h_{n}(x) 的情况,同理可得.

由引理2.1–2.3得命题2.4.

命题2.4   M(u) 是给定的N函数, p(u) M(u) 的右导数.当 0<\beta<\frac{1}{4} , n\geq1 时,存在 h_{n}(x)\in D_{n} , \lambda_{n}>0 ,满足

\int_{0}^{2\pi} p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))G_{r}(x-y)\, {\rm d}x = \lambda_{n}p\bigg(\frac{|h_{n}(y)|}{\|h_{n}\|_{(M)}}\bigg){\rm sgn}(h_{n}(y)), y\in\mathbb{R},

1) h_{n}(x) 连续;

2)在 [0, 2\pi) 内, h_{n}(x) 仅以 \{\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为其单零点, G_{r}\ast h_{n}(x) 仅以 \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为其单零点.

3 定理及证明

定义 A_{2n} \tilde{m}_{r, M} S_{2n, r-1} 上以 \{r+\frac{i-1}{n}\pi\}_{i = 1}^{2n} 为插值结点的插值算子.

定理3.1  当 0<\beta<\frac{1}{4} , n\geq1 时,有

d_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast}) = d^{2n}(\tilde{m}_{r, M}; L_{M}^{\ast}) = d_{2n}\prime(\tilde{m}_{r, M}; L_{M}^{\ast}) = b_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast}) = \lambda_{n},

1) S_{2n, r-1} d_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast}) 的极子空间;

2) L_{2n}^{\ast} = \{f\in\tilde{m}_{r, M}|f(r+\frac{i-1}{n}\pi) = 0, i = 1, \cdots, 2n\} d^{2n}(\tilde{m}_{r, M}; L_{M}^{\ast}) 的极子空间;

3) A_{2n} d_{2n}\prime(\tilde{m}_{r, M}; L_{M}^{\ast}) 的最佳线性算子;

4) Y_{2n} = {\rm span}\{\int_{\frac{i-1}{n}\pi}^{\frac{i}{n}\pi} G_{r}(x-y)|h_{n}(y)|\, {\rm d}y, i = 1, \cdots, 2n\} b_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast}) 的极子空间.

取任意 0<\sigma<\frac{\pi}{n} ,规定 \xi_{0} = 0 , \xi_{1} = \sigma , \xi_{2} = \sigma+\frac{\pi}{n}, \cdots, \xi_{2n} = \sigma+\frac{2n-1}{n}\pi , \xi_{2n+1} = 2\pi , h_{n}^{\sigma}(x)\triangleq h_{n}(x-\sigma) ,

f_{i}^{\sigma}(x) = \left\{\begin{array}{ll} |h_{n}^{\sigma}(x)|, & { \xi_{i}\leq x<\xi_{i+1}$}, \\ 0 , & { x\in[0, 2\pi]\backslash [\xi_{i}, \xi_{i+1})$}, \end{array}\right. i = 0, 1, \cdots, 2n,

f_{j}(x) = \left\{\begin{array}{ll} |h_{n}(x)| , & \frac{j-1}{n}\pi\leq x<\frac{j}{n}\pi, \\ 0 , & x\in[0, 2\pi]\backslash [\frac{j-1}{n}\pi, \frac{j}{n}\pi), \end{array}\right.j = 1, \cdots, 2n,

g_{i}^{\sigma}(x) = \int_{0}^{2\pi}G_{r}(x-y)f_{i}^{\sigma}(y)\, {\rm d}y, i = 0, 1, \cdots, 2n,

g_{i}(x) = \int_{0}^{2\pi}G_{r}(x-y)f_{i}(y)\, {\rm d}y, i = , 1, \cdots, 2n.

引理3.2[4]  当 0<\sigma<\frac{\pi}{n} , 0<\beta<\frac{1}{4} , n\geq1 时, g_{0}^{\sigma}(x), g_{1}^{\sigma}(x), g_{2}^{\sigma}(x), \cdots, g_{2n}^{\sigma}(x) 为一组WT系.

引理3.3

\lim\limits_{\sigma\rightarrow0} \inf\limits_{C_{i}\neq0\atop i = 0, \cdots, 2n} \frac{\bigg\|\sum\limits_{i = 0}^{2n}C_{i}g_{i}^{\sigma}\bigg\|_{M}} {\bigg\|\sum\limits_{i = 0}^{2n}C_{i}f_{i}^{\sigma}\bigg\|_{M}} = \inf\limits_{C_{i}\neq0\atop i = 1, \cdots, 2n} \frac{\bigg\|\sum\limits_{i = 1}^{2n}C_{i}g_{i}\bigg\|_{M}} {\bigg\|\sum\limits_{i = 1}^{2n}C_{i}f_{i}\bigg\|_{M}}.

  因为 f_{i}^{\sigma}(x)\Rightarrow f_{i}(x), i = 1, \cdots, 2n , f_{0}^{\sigma}(x)\Rightarrow0 (\sigma\rightarrow0) .所以 g_{i}^{\sigma}(x) 一致收敛到 g_{i}(x), i = 1, \cdots, 2n , g_{0}^{\sigma}(x) 一致收敛到 0(\sigma\rightarrow0) .从而,得

\inf\limits_{C_{i}\neq0\atop i = 0, \cdots, 2n} \frac{\bigg\|\sum\limits_{i = 0}^{2n}C_{i}g_{i}^{\sigma}\bigg\|_{M}}{\bigg\|\sum\limits_{i = 0}^{2n}C_{i}f_{i}^{\sigma}\bigg\|_{M}} = \inf\limits_{C_{i}\neq0\atop i = 0, \cdots, 2n} \frac{\bigg\|\sum\limits_{i = 1}^{2n}C_{i}g_{i}\bigg\|_{M}}{\bigg\|\sum\limits_{i = 1}^{2n}C_{i}f_{i}\bigg\|_{M}}, \sigma\rightarrow0.

证毕.

引理3.4  当 0<\beta<\frac{1}{4} , n\geq1 时,有

\inf\limits_{C_{i}\neq0\atop i = 0, \cdots, 2n} \frac{\bigg\|\sum\limits_{i = 1}^{2n}C_{i}g_{i}\bigg\|_{M}}{\bigg\|\sum\limits_{i = 1}^{2n}C_{i}f_{i}\bigg\|_{M}}\geq\lambda_{n}.

  设 \inf\limits_{C_{i}\neq0\atop i = 0, \cdots, 2n} \frac{\big\|\sum\limits_{i = 1}^{2n}C_{i}g_{i}\big\|_{M}}{\big\|\sum\limits_{i = 1}^{2n}C_{i}f_{i}\big\|_{M}} = \frac{\big\|\sum\limits_{i = 1}^{2n}\bar{C}_{i}g_{i}\big\|_{M}}{\big\|\sum\limits_{i = 1}^{2n}\bar{C}_{i}f_{i}\big\|_{M}} = \bar{\mu}_{n} , \bar{g}(x) = \sum\limits_{i = 1}^{2n}\bar{C}_{i}g_{i}(x) , \bar{f}(x) = \sum\limits_{i = 1}^{2n}\bar{C}_{i}f_{i}(x) , |\bar{C}_{i}|\leq1, i = 1, \cdots, 2n .且存在 1\leq i_{0}\leq2n ,使 \bar{C}_{i_{0}} = (-1)^{i_{0}-1} .利用反证法,假设 \bar{\mu}_{n}<\lambda_{n} ,因为

G_{r}\ast h_{n}(x)-\bar{g}(x) = \sum\limits_{i = 1}^{2n}[(-1)^{i-1}-\bar{C}_{i}]g_{i}(x),

\bar{C}_{i_{0}} 作微小 \varepsilon -摄动,得 \bar{g}^{\varepsilon}(x) ,利用Rolle定理,有

S_{c}^{-}(G_{r}\ast h_{n}-\bar{g}^{\varepsilon})\leq2n-2,
(3.1)

由于

\begin{eqnarray*} &&\int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}(x)|}{\|\bar{g}\|_{(M)}}\bigg){\rm sgn}(\bar{g}(x))\bigg]g_{k}(x)\, {\rm d}x\\ & = &\bigg[\lambda_{n}(-1)^{k-1}- \bar{\mu}_{n}p\bigg(\frac{|\bar{C}_{k}|}{\|\bar{C}_{k}\|_{(M)}}\bigg){\rm sgn}(\bar{C}_{k})\bigg]\beta_{k} \end{eqnarray*}

严格交错变号,这里 k = 1, \cdots, 2n , 0<\beta_{k} = \int_{\frac{k-1}{n}\pi}^{\frac{k}{n}\pi} M(|h_{n}(x))|\, {\rm d}x .

\begin{eqnarray*} && \int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}}\bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg]g_{k}^{\sigma}(x)\, {\rm d}x\\ & = &\int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}(x)|}{\|\bar{g}\|_{(M)}}\bigg){\rm sgn}(\bar{g}(x))\bigg]g_{k}^{\sigma}(x)\, {\rm d}x\\ &&+\int_{0}^{2\pi} \bigg[p\bigg(\frac{|\bar{g}(x)|}{\|\bar{g}\|_{(M)}}\bigg){\rm sgn}(\bar{g}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}}\bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg]g_{k}^{\sigma}(x)\, {\rm d}x, \end{eqnarray*}

k = 1, \cdots, 2n .因为 g_{k}^{\sigma}(x) 一致收敛到 g_{k}(x) , \sigma\rightarrow0 , k = 1, \cdots, 2n , \bar{g}^{\varepsilon}(x) 一致收敛到 \bar{g}(x) , \varepsilon\rightarrow0 ,所以由 \sigma \varepsilon 的任意性,令它们充分小,有

\int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}}\bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg]g_{k}^{\sigma}(x)\, {\rm d}x, k = 1, \cdots, 2n

严格交错变号.如果

\int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}}\bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg]g_{0}^{\sigma}(x)\, {\rm d}x = 0,

则取 g_{0}^{\ast}(x) = g_{0}^{\sigma}(x)+g_{k}^{\ast}(x) , g_{k}^{\ast}(x) = g_{k}^{\sigma}(x) , k = 1, \cdots, 2n .如果

\int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}}\bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg]g_{0}^{\sigma}(x)\, {\rm d}x\neq0,

则取 g_{0}^{\ast}(x) = g_{0}^{\sigma}(x) -g_{0}^{\sigma}(x) , g_{k}^{\ast}(x) = g_{k}^{\sigma}(x) , k = 1, \cdots, 2n .使得

\int_{0}^{2\pi} \bigg[p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}}\bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg]g_{k}^{\ast}(x)\, {\rm d}x, k = 0, 1, \cdots, 2n

严格交错变号,其中 \{g_{0}^{\ast}(x), \cdots, g_{2n}^{\ast}(x)\} 为WT系.根据引理3.2和文献[9]中的性质3.6(2),有

\begin{eqnarray*} 2n&\leq&\bar{S}_{c}\bigg(p\bigg(\frac{|G_{r}\ast h_{n}(x)|}{\|G_{r}\ast h_{n}\|_{(M)}}\bigg){\rm sgn}(G_{r}\ast h_{n}(x))- p\bigg(\frac{|\bar{g}^{\varepsilon}(x)|}{\|\bar{g}^{\varepsilon}\|_{(M)}} \bigg){\rm sgn}(\bar{g}^{\varepsilon}(x))\bigg)\\ & = & \bar{S}_{c}(G_{r}\ast h_{n}(x)-\bar{g}^{\varepsilon}(x)), \end{eqnarray*}

与(3.1)式矛盾.即证得 \bar{\mu}_{n}\geq\lambda_{n} .

引理3.5  如果满足定理条件,则

b_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast})\geq\lambda_{n}.

  反证法,假设 b_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast})<\lambda_{n} ,存在 \tau>0 ,使得 b_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast})<\lambda_{n}-\tau ,根据引理3.3,引理3.4,存在 0<\sigma<\frac{\pi}{n} ,使得

\begin{matrix} \inf \\ \begin{matrix} {{C}_{i}}\ne 0 \\ i=0, \cdots , 2n \\\end{matrix} \\\end{matrix}\frac{\|\sum\limits_{i=0}^{2n}{{{C}_{i}}}g_{i}^{\sigma }{{\|}_{M}}}{\|\sum\limits_{i=0}^{2n}{{{C}_{i}}}f_{i}^{\sigma }{{\|}_{M}}}\ge {{\lambda }_{n}}-\tau .

从而 b_{2n}(\tilde{m}_{r, M}; L_{M}^{\ast})\geq\lambda_{n}-\tau ,出现矛盾.即证.

引理3.6[9]  设 \Lambda 是线性赋范空间 X 上的一个中心对称的闭凸子集,则

b_{n}(\Lambda; X)\leq d^{n}(\Lambda; X).

引理3.7[2]  对任意 \Lambda\subset X ,有

d_{n}(\Lambda; X)\leq d_{n}\prime(\Lambda; X);

d^{n}(\Lambda; X)\leq d_{n}\prime(\Lambda; X).

引理3.8  如果满足定理条件,则

{\rm sup}\{\|f-A_{2n}(f)\|_{M}|f\in\tilde{m}_{r, M}\}\leq\lambda_{n}.

  仿照文献[9],再结合文献[10]即可证.

综合引理2.1–2.3及引理3.2–3.8,即可完成定理的证明.关于极子空间的论断只需一些简单的计算便可得.

参考文献

Wu C X , Wang T F . Orlicz Space and Its Application. Haerbin: Heilongjiang Science and Technology Press, 1983: 1- 78

[本文引用: 3]

孙永生. 函数逼近论(上册). 北京: 北京师范大学出版社, 1989

[本文引用: 2]

Sun Y S . The Function Approximation Theory (Ⅰ). Beijing: Beijing Normal University Press, 1989

[本文引用: 2]

房艮孙.

广义周期样条类的极值问题和广义Bernoulli核的宽度

数学年刊, 1987, 8A (1): 75- 87

URL     [本文引用: 1]

Fang G S .

The extreme value problem of generalized periodic spline and the n-width of generalized bernoulli kernel

Chinese Annals of Mathematics, 1987, 8A (1): 75- 87

URL     [本文引用: 1]

毕宁.

关于 \tilde{m}_{r, p} 类宽度的精确估计

杭州师范学院学报, 1990, 11 (6): 7- 15

[本文引用: 4]

Bi N .

Exact estimates of n-widths for classes \tilde{m}_{r, p}

Journal of Hangzhou Normal College, 1990, 11 (6): 7- 15

[本文引用: 4]

房艮孙.

广义Bernoulli核的宽度和线性插值算子

科学通报, 1985, 30 (11): 806- 809

URL     [本文引用: 1]

Fang G S .

The n-width of the generalized bernoulli kernel and the linear interpolation operator

Science Bulletin, 1985, 30 (11): 806- 809

URL     [本文引用: 1]

Li C .

N-widths of \Omega_{p}^{r} in Lp

Approximation Theory and Its Applications, 1989, 5 (1): 47- 62

[本文引用: 1]

Wu G R D .

On N-widths of wome periodic convolution classes in Orlicz spaces

Approximation Theory and Its Applications, 1998, 14 (3): 25- 35

URL     [本文引用: 1]

孙永生, 黄达人.

关于广义Bernoulli核的宽度

科学通报, 1985, 30 (10): 728- 731

URL     [本文引用: 1]

Sun Y S , Huang D R .

On the N-width of the generalized Bernoulli kernel

Science Bulletin, 1985, 30 (10): 728- 731

URL     [本文引用: 1]

Allan P .

N-widths of sobolev spaces in Lp

Constructive Approximation, 1985, 1 (1): 15- 62

[本文引用: 4]

Wu G R D .

On approximation by polynomials in Orlicz spaces

Approximation Theory and Its Applications, 1991, 7 (3): 97- 110

URL     [本文引用: 1]

/