数学物理学报  2017, Vol. 37 Issue (2): 228-238   PDF    
扩展功能
加入收藏夹
复制引文信息
加入引用管理器
Email Alert
RSS
本文作者相关文章
王维
朱玉灿
Fusion框架系统的局部框架扰动的稳定性
王维, 朱玉灿     
福州大学数学与计算机科学学院 福州 350116
摘要:该文在Hilbert空间中一般的框架序列扰动形式下,利用正交投影的性质和对偶框架的性质研究了原序列张成的闭子空间与扰动序列张成的闭子空间的关系,并探讨了局部框架的一般扰动对fusion框架系统稳定性的影响.这些结果推广和改进了由Casazza,Kutyniok和Li等得到的著名结果.
关键词框架序列    扰动    局部框架    fusion框架系统    稳定性    
Stability of Fusion Frame Systems under Perturbation of the Local Frame Vectors
Wang Wei, Zhu Yucan     
College of Mathematics and Computer Science, Fuzhou University, Fuzhou 350116
Abstract: This paper studies the relation between the two subspaces spanned by a frame sequence and its general perturbed version in Hilbert spaces by using the properties of orthogonal projection and the properties of dual frames. This paper also studies the stability of fusion frame systems under general perturbation of the local frame vectors. Our results generalize and improve the remarkable results which had been obtained by Casazza, Kutyniok and Li.
Key words: Frame sequence     Perturbation     Local frame     Fusion frame systems     Stability    
1 引言

从二十世纪八十年代开始, 随着小波分析理论与应用研究的发展, 框架理论及其应用的研究已经成为了国内外众多学者研究的热点, 并已取得了一系列重要的研究成果[1-2].目前, 框架理论不仅成为数学研究领域中的重要内容之一, 并且已被广泛应用于图像处理[3]、编码和信号传输[4]、无线通讯[5]、数据量化[6]等领域.然而普通框架的理论仍满足不了一些工程应用中的要求, 传感器就是一个典型例子, 它要求把要处理的任务进行分布式处理, 这恰恰是普通框架所实现不了的.换句话说, 若常把大量的数据分配给单一的框架系统, 就会导致数据量太大而使得框架系统难以有效的处理数据, 这种情况就要求将一个大的框架系统分成一系列小的框架子系统, 每个子系统能够有效的处理好局部的数据.由此Casazza等人在研究用局部框架重构全局框架时提出了子空间框架 (frame ofsubspaces)[7]的概念, 现在我们通常称之为fusion框架[8], 而这种思想也符合计算机对大型的数据处理进行分布式计算的算法.为了满足更多工程应用中的需要, 许多学者开始对fusion框架进行了研究, Casazza, Kutyniok和Li等人在fusion框架的理论及其应用领域都取得了不少的研究成果 (参见文献[7-10]).

对fusion框架的研究具有很大的实用价值, 尤其对fusion框架系统稳定性的研究有着深远的意义.在实际应用中存在着很多有关fusion框架系统稳定性的问题, 比如某些检波器组为了调整传输条件需要移动到稍微偏移原来位置的地方[11], 那么这种移动会对整个检波器系统带来什么样的影响?又比如在无线传感网络中, 由于受到外部自然环境的影响, 某个传感器或者某个子传感网络的位置可能会发生偏移[12], 那么这种偏移是否会影响整个无线传感网络的稳定性?为了清楚地描述诸如受影响的传感器网络的性能, 我们需要对fusion框架系统在扰动下的稳定性进行深入研究.另外, 在研究fusion框架的构造时, fusion框架在扰动下的稳定性决定能否从一个已知的fusion框架构造新的fusion框架.

Casazza等人于文献[8]中在fusion框架系统下, 分别探讨了子空间序列的扰动与包含在子空间中局部框架的扰动对fusion框架系统稳定性的影响.本文将主要研究在局部框架扰动下的fusion框架系统的稳定性, 文中采用不同的方法, 研究了在Hilbert空间中一般的框架序列扰动形式下, 原序列张成的闭子空间与扰动序列张成的闭子空间的关系, 并探讨了局部框架的一般扰动对fusion框架系统稳定性的影响, 这些结果推广和改进了由Casazza, Kutyniok和Li等得到的著名结果[8].

本文采用如下记号:设${\cal H}$为可分的复Hilbert空间, $W$是的${\cal H}$闭子空间, 用$\pi_{W}$表示从${\cal H}$$W$的正交投影.设$I$, $J_{i}$是自然数集$N$的子集; $\ell^{2}(I)$表示满足$\sum\limits_{i\in I}|a_{i}|^{2} < +\infty$的复数列$\{a_i\}_{i\in I}$全体所成的线性空间.设$X$, $Y$为Banach空间, 记$L (X, Y)$表示从$X$$Y$的有界线性算子的集合, 若$X=Y$, 则记$L (X, Y)=L (X)$; 若$Q\in L (X, Y)$, 记$N_{Q}$为算子$Q$的核.

首先介绍Hilbert空间中的框架、框架序列以及fusion框架等概念.

定义1.1  设序列$\{f_i\}_{i\in I}\subset {\cal H}$, 如果存在正数$A, B$, 使对任意$f\in {\cal H}$

$A{{\left\| f \right\|}^{2}}\le \sum\limits_{i\in I}{|\langle f,{{f}_{i}}\rangle {{|}^{2}}\le B{{\left\| f \right\|}^{2}}}$ (1.1)

成立, 那么称$\{f_i\}_{i\in I}$${\cal H}$的框架. $A, B$分别称为框架的下界, 上界.

如果只有不等式 (1.1) 的右边成立, 则称$\{f_i\}_{i\in I}$${\cal H}$界为$B$的Bessel序列.

如果$A=B$, 则称$\{f_i\}_{i\in I}$${\cal H}$的紧框架.如果$A=B=1$, 则称$\{f_i\}_{i\in I}$${\cal H}$的Parseval框架.

定义1.2  设$\{f_i\}_{i\in I}$${\cal H}$的框架, 定义有界线性算子$T, T^{*}$如下

$\begin{gathered} T:{\ell ^2}(I) \to H, \;T(c) = \sum\limits_{i \in I} {{c_i}{f_i}, } \;c = {\{ {c_i}\} _{i \in I}} \in {\ell ^2}(I), \hfill \\ {T^*}:H \to {\ell ^2}(I), \;{T^*}(f) = {\{ \langle f, {f_i}\rangle \} _{i \in I}}, \;f \in H, \hfill \\ \end{gathered} $ (1.2)

则称$T$$\{{f_i} \}_{i\in I}$的预框架算子, 而称有界线性算子$S=TT^{*}$$\{{f_i} \}_{i\in I}$的框架算子

$ S: {\cal H}\rightarrow{\cal H}, \ Sf=\sum\limits_{i\in I}{}\langle f, {f_i} \rangle {f_i}, \ f\in {\cal H}. $

定义1.3  如果序列$\{f_i\}_{i\in I}\subset {\cal H}$$\overline{span}\{{f_i} \}_{i\in I}$的框架, 且框架界为$A, B$, 那么序列$\{{f_i} \}_{i\in I}$称为${\cal H}$的框架序列, 且框架界为$A, B$.

定义1.4[7]  设$\{W_{i}\}_{i\in I}$${\cal H}$中的一个闭子空间序列, $\{v_{i}\}_{i\in I}$是一个正数集, 若存在正数$C, D$, 使得对任意的$f\in {\cal H}$, 有

$ C\|f\|^2\leq\sum\limits_{i\in I}{}v^2_i\|\pi_{W_i}f\|^2\leq D\|f\|^2, $

那么我们称$\{(W_{i}, v_{i})\}_{i\in I}$${\cal H}$的fusion框架, $C, D$为fusion框架的下界和上界.特别地, 如果$C=D=1$, 那么我们称$\{(W_{i}, v_{i})\}_{i\in I}$${\cal H}$的Parseval fusion框架.

定义1.5[8]  设$\{(W_{i}, v_{i})\}_{i\in I}$${\cal H}$的fusion框架, 且对于每个$i\in I$, $\{f_{ij}\}_{j\in J_{i}}$$W_{i}$的一个框架, 那么我们称$\{(W_{i}, v_{i}, \{f_{ij}\}_{j\in J_{i}})\}_{i\in I}$${\cal H}$的一个fusion框架系统.如果$C$$D$$\{(W_{i}, v_{i})\}_{i\in I}$的fusion框架界, 那么就称$C$$D$为fusion框架系统对应的fusion框架界; 如果对于每个$i\in I$, 局部框架$\{f_{ij}\}_{j\in J_{i}}$都有公共的框架界$A$$B$, 那么就称$A$$B$为该fusion框架系统的局部框架界.如果对于固定的$i\in I$, $\{g_{ij}\}_{j\in J_{i}}$也为$W_{i}$的一个框架且满足

$ h=\sum\limits_{j\in J_{i}}{}\langle h, f_{ij}\rangle g_{ij}=\sum\limits_{j\in J_{i}}{}\langle h, g_{ij}\rangle f_{ij}, \ \forall h\in W_{i}, $

则称$\{g_{ij}\}_{j\in J_{i}}$为局部框架$\{f_{ij}\}_{j\in J_{i}}$的对偶框架.

定义1.6[8]  设$\{{f_i} \}_{i\in I}\subset{\cal H}$为一个序列, $0\leq\lambda_{1}, \lambda_{2} < 1$, $\mu\geq0$.如果序列$\{g_{i}\}_{i\in I}\subset{\cal H}$满足:对任意的有限子集$J\subset I$和任意复数$c_{i}, i\in J $, 都有

$\bigg\|\sum\limits_{i\in J}{}c_{i}({f_i} -g_{i})\bigg\|\leq\lambda_{1}\bigg\|\sum\limits_{i\in J}{}c_{i}{f_i} \bigg\|+\lambda_{2}\bigg\|\sum\limits_{i\in J}{}c_{i}g_{i}\bigg\|+\mu\bigg(\sum\limits_{i\in J}{}|c_{i}|^{2}\bigg)^{1/2}, $ (1.3)

那么称序列$\{g_{i}\}_{i\in I}$是序列$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动.

定义1.6考虑的是序列的一般扰动情况, 当$\mu=0$时, 便是文献[8]中的$(\lambda_{1}, \lambda_{2})$-扰动.

定义1.7[8]  设$\{W_{i}\}_{i\in I}$$\{\widetilde{W}_{i}\}_{i\in I}$${\cal H}$的两个闭子空间序列, 设$\{v_{i}\}_{i\in I}$是一正数序列, 令$0\leq\beta_{1}, \beta_{2} < 1$, $\varepsilon>0$.如果对于任意的$f\in {\cal H}$以及每个$i\in I$, 都有

$ \|(\pi_{W_{i}}-\pi_{\widetilde{W}_{i}})f\|\leq\beta_{1}\|\pi_{W_{i}}f\|+\beta_{2}\|\pi_{\widetilde{W}_{i}}f\|+\varepsilon\|f\|, $

那么称$\{(\widetilde{W}_{i}, v_{i})\}_{i\in I}$$\{(W_{i}, v_{i})\}_{i\in I}$的一个$(\beta_{1}, \beta_{2}, \varepsilon)$-扰动.

定义1.8[13]  设$V, W$${\cal H}$的闭子空间, 则$V$$W$的间隙 (gap) 定义如下

$ \delta(V, W)=\sup\limits_{x\in V, \|x\|=1}\inf\limits_{y\in W}\|x-y\|. $
2 主要定理及其证明

在文献[8]中的命题5.4(ⅱ) 中, Casazza和Kutyniok等人得到了“如果设序列$\{g_{i}\}_{i\in I}\subset{\cal H}$是框架序列$\{{f_i} \}_{i\in I}\subset{\cal H}$的一个$(\lambda_{1}, \lambda_{2}, 0)$-扰动, $W=\overline{span}\{{f_i} \}_{i\in I}$, $\widetilde{W}=\overline{span}\{g_{i}\}_{i\in I}$.那么对于任意的$f\in {\cal H}$, 都有$\|\pi_{W}(\pi_{\widetilde{W}}(f))\|\geq (\frac{1-\lambda_{2}}{1+\lambda_{1}}-\lambda_{1}\frac{1+\lambda_{2}}{1-\lambda_{1}}-\lambda_{2})\|\pi_{\widetilde{W}}(f)\|$”的结论, 并以此结论来分析了fusion框架系统在局部框架的$(\lambda_{1}, \lambda_{2}, 0)$-扰动下的稳定性.在上述结论中, 通过简单计算可以验证, 当参数$\lambda_{1}, \lambda_{2}$都小于1/5时, 必有$\frac{1-\lambda_{2}}{1+\lambda_{1}}-\lambda_{1}\frac{1+\lambda_{2}}{1-\lambda_{1}}-\lambda_{2}>0$, 然而这对于序列$\{g_{i}\}_{i\in I}$的扰动要求过于苛刻, 给实际应用带来了很大的限制.下面我们考虑更为一般的扰动情况, 即在序列$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动的情况下, $W$$\widetilde{W}$具有什么样的关系.

引理2.1[1]  设序列$\{{f_i} \}_{i\in I}\subset{\cal H}$, 则$\{{f_i} \}_{i\in I}$${\cal H}$的Bessel序列且其Bessel界为$B$当且仅当由式子 (1.2) 定义的有界线性算子$T$满足$\|T\|\leq\sqrt{B}$.此时有

$ \bigg\|\sum\limits_{i\in I}{}c_{i}{f_i} \bigg\|\leq\sqrt{B}\bigg(\sum\limits_{i\in I}{}|c_{i}|^{2}\bigg)^{1/2}, \ \forall c=\{c_{i}\}_{i\in I}\in\ell^{2}(I). $

引理2.2[1]  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的框架序列, 则$\{{f_i} \}_{i\in I}$${\cal H}$的Bessel序列.

引理2.3  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的框架序列.如果序列$\{g_{i}\}_{i\in I}\subset{\cal H}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动, 那么级数$\sum\limits_{i\in I}c_{i}g_{i}$${\cal H}$中收敛, 且对任意的$\{c_{i}\}_{i\in I}\in \ell^{2}(I)$, 有

$\begin{align} & \frac{1-{{\lambda }_{1}}}{1+{{\lambda }_{2}}}\left\| \sum\limits_{i\in I}{{}}{{c}_{i}}{{f}_{i}} \right\|-\frac{\mu }{1+{{\lambda }_{2}}}{{(\sum\limits_{i\in I}{{}}|{{c}_{i}}{{|}^{2}})}^{1/2}} \\ & \le \left\| \sum\limits_{i\in I}{{}}{{c}_{i}}{{g}_{i}} \right\|\frac{1+{{\lambda }_{1}}}{1-{{\lambda }_{2}}}\left\| \sum\limits_{i\in I}{{}}{{c}_{i}}{{f}_{i}} \right\|+\frac{\mu }{1-{{\lambda }_{2}}}{{(\sum\limits_{i\in I}{{}}|{{c}_{i}}{{|}^{2}})}^{1/2}}, \\ \end{align}$ (2.1)
$\begin{align} & \frac{1-{{\lambda }_{2}}}{1+{{\lambda }_{1}}}\left\| \sum\limits_{i\in I}{{}}{{c}_{i}}{{g}_{i}} \right\|-\frac{\mu }{1+{{\lambda }_{1}}}{{(\sum\limits_{i\in I}{{}}|{{c}_{i}}{{|}^{2}})}^{1/2}} \\ & \le \left\| \sum\limits_{i\in I}{{}}{{c}_{i}}{{f}_{i}} \right\|\frac{1+{{\lambda }_{2}}}{1-{{\lambda }_{1}}}\left\| \sum\limits_{i\in I}{{}}{{c}_{i}}{{g}_{i}}+ \right\|\frac{\mu }{1-{{\lambda }_{1}}}{{(\sum\limits_{i\in I}{{}}|{{c}_{i}}{{|}^{2}})}^{1/2}}. \\ \end{align}$ (2.2)

  由引理2.1和2.2知, $\{{f_i} \}_{i\in I}$为Bessel序列且级数$\sum\limits_{i\in I}c_{i}{f_i} $${\cal H}$中收敛.如果$I$是有限集, 直接验证结论成立.现在我们来讨论$I$是无限集的情况, 记$I=\{n_{1}, n_{2}, \cdots, n_{k}, \cdots\}$.设$\{{f_i} \}_{i\in I}$的Bessel界为$B$, 则由式子 (1.3) 可得:当$k\geq l>0$时, 有

$\begin{align} & \left\| \sum\limits_{i={{n}_{l}}}{{{n}_{k}}}{{c}_{i}}{{g}_{i}} \right\|\frac{1+{{\lambda }_{1}}}{1-{{\lambda }_{2}}}\left\| \sum\limits_{i={{n}_{l}}}{{{n}_{k}}}{{c}_{i}}{{f}_{i}} \right\|+\frac{\mu }{1-{{\lambda }_{2}}}{{(\sum\limits_{i={{n}_{l}}}{{{n}_{k}}}|{{c}_{i}}{{|}^{2}})}^{1/2}} \\ & \le \frac{(1+{{\lambda }_{1}})\sqrt{B}+\mu }{1-{{\lambda }_{2}}}{{(\sum\limits_{i={{n}_{l}}}{{{n}_{k}}}|{{c}_{i}}{{|}^{2}})}^{1/2}}. \\ \end{align}$ (2.3)

由于正项级数$\sum\limits_{i\in I}|c_{i}|^{2}$是收敛的, 因此序列$\{\sum\limits_{i=n_{1}}^{n_{k}}c_{i}g_{i}\}_{k=1}^{\infty}$${\cal H}$中的Cauchy列, 因而级数$\sum\limits_{i\in I}c_{i}g_{i} $${\cal H}$中 (按${\cal H}$中的范数) 也收敛.由式子 (1.3) 知:对任意的$\{c_{i}\}_{i\in I}\in \ell^{2}(I)$, 有

$ \begin{align} & \left\| \sum\limits_{i={{n}_{1}}}{{{n}_{k}}}{{c}_{i}}{{g}_{i}} \right\|\ge \left\| \sum\limits_{i={{n}_{1}}}{{{n}_{k}}}{{c}_{i}}{{f}_{i}} \right\|-\left\| \sum\limits_{i={{n}_{1}}}{{{n}_{k}}}{{c}_{i}}({{f}_{i}}-{{g}_{i}}) \right\| \\ & \ge \left\| \sum\limits_{i={{n}_{1}}}{{{n}_{k}}}{{c}_{i}}{{f}_{i}} \right\|-{{\lambda }_{1}}\left\| \sum\limits_{i={{n}_{1}}}{{{n}_{k}}}{{c}_{i}}{{f}_{i}} \right\|-{{\lambda }_{2}}\left\| \sum\limits_{i={{n}_{1}}}{{{n}_{k}}}{{c}_{i}}{{g}_{i}} \right\|-\mu {{(\sum\limits_{i={{n}_{1}}}{{{n}_{k}}}|{{c}_{i}}{{|}^{2}})}^{1/2}}, \\ \end{align} $

也即对任意的$\{c_{i}\}_{i\in I}\in \ell^{2}(I)$, 有

$\bigg\|\sum\limits_{i=n_{1}}{n_{k}}c_{i}g_{i}\bigg\|\geq\frac{1-\lambda_{1}}{1+\lambda_{2}}\bigg\|\sum\limits_{i=n_{1}}{n_{k}}c_{i}{f_i} \bigg\|-\frac{\mu}{1+\lambda_{2}} \bigg(\sum\limits_{i=n_{1}}{n_{k}}|c_{i}|^{2}\bigg)^{1/2}.$ (2.4)

在 (2.3) 式和 (2.4) 式中, 令$l=1, k\rightarrow+\infty$, 可得: (2.1) 式和 (2.2) 式成立.

引理2.4[14]  设$\{{f_i} \}_{i\in I}\subset{\cal H}$为一个框架序列, 框架界为$A, B$, 且$0\leq\lambda_{1}, \lambda_{2} < 1$, $\mu\geq0$.如果序列$\{g_{i}\}_{i\in I}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动, 且$\delta_{N} < 1$, $\lambda_{1}+\frac{\mu}{\sqrt{A}(1-\delta_{N}^{2})^{1/2}} < 1$, 其中$\delta_{N}=\delta (N_{T}, N_{U})$, $T$$U$为分别由如下定义的有界线性算子

$T: \ell^{2}(I)\rightarrow{\cal H}, \ T(c)=\sum\limits_{i\in I}{}c_{i}{f_i}, \ c=\{c_{i}\}_{i\in I}\in \ell^{2}(I), $ (2.5)
$U: \ell^{2}(I)\rightarrow{\cal H}, \ U(c)=\sum\limits_{i\in I}{}c_{i}g_{i}, \ c=\{c_{i}\}_{i\in I}\in \ell^{2}(I), $ (2.6)

$\{g_{i}\}_{i\in I}$${\cal H}$的一个框架序列, 且其框架界为

$ \frac{[(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}-\mu]^{2}}{(1+\lambda_{2})^{2}}, \ \ \frac{[(1+\lambda_{1})\sqrt{B}+\mu]^{2}}{(1-\lambda_{2})^{2}}. $

引理2.5[1]  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的框架, 框架界为$A, B$.设$S$$\{{f_i} \}_{i\in I}$的框架算子, 那么$\{S^{-1}{f_i} \}_{i\in I}$${\cal H}$的框架, 且框架界为$B^{-1}, A^{-1}$.并且$\{{f_i} \}_{i\in I}$$\{S^{-1}{f_i} \}_{i\in I}$${\cal H}$的互为对偶框架.

注2.1  如果$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的框架序列, 由定义1.3以及引理2.5, $\{S^{-1}{f_i} \}_{i\in I}$$\overline{span}\{{f_i} \}_{i\in I}$的框架, 且框架下界和上界分别为$B^{-1}, A^{-1}$.并且$\{{f_i} \}_{i\in I}$$\{S^{-1}{f_i} \}_{i\in I}$$\overline{span}\{{f_i} \}_{i\in I}$的互为对偶框架.

定理2.1  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的一个框架序列且框架界为$A, B$, 且$0\leq\lambda_{1}, \lambda_{2} < 1$.如果序列$\{g_{i}\}_{i\in I}\subset {\cal H}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动, 并且$\delta_{N} < 1$满足条件:

(ⅰ) $1-\lambda_{1}-2\lambda_{2}>0;$

(ⅱ) $0\leq\mu < \frac{(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}(1-\lambda_{1}-2\lambda_{2})}{3-\lambda_{1}}$.

其中$T$$U$分别由式子 (2.5) 和 (2.6) 定义的有界线性算子, $\delta_{N}=\delta (N_{T}, N_{U})$.

$W=\overline{span}\{{f_i} \}_{i\in I}$, $\widetilde{W}=\overline{span}\{g_{i}\}_{i\in I}$, 那么对于任意的$f\in {\cal H}$, 都有

$ \|\pi_{W}(\pi_{\widetilde{W}}(f))\|\geq\frac{(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}(1-\lambda_{1}-2\lambda_{2})-\mu(3-\lambda_{1})} {(1+\lambda_{1})[(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}-\mu]}\|\pi_{\widetilde{W}}(f)\|. $

  由$0\leq\lambda_{1}, \lambda_{2} < 1$以及条件 (i) 可知$0 < \frac{1-\lambda_{1}-2\lambda_{2}}{3-\lambda_{1}} < 1$, 因此再结合条件 (ⅱ) 有

$ 0\leq\mu<\frac{(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}(1-\lambda_{1}-2\lambda_{2})}{3-\lambda_{1}}<(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}. $

$ \widetilde{A}=\frac{[(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}-\mu]^{2}}{(1+\lambda_{2})^{2}}, \, \, \widetilde{B}=\frac{[(1+\lambda_{1})\sqrt{B}+\mu]^{2}}{(1-\lambda_{2})^{2}}, $

则由引理2.4可知, $\{g_{i}\}_{i\in I}$${\cal H}$的框架序列, 即$\{g_{i}\}_{i\in I}$$\widetilde{W}$的框架, 且其框架下界和上界分别为$\widetilde{A}, \ \widetilde{B}$.设$S_{G}$$G=\{g_{i}\}_{i\in I}$的框架算子, 根据注2.1可知, $\{S_{G}^{-1}g_{i}\}_{i\in I}$也为$\widetilde{W}$的框架, 并且其框架下界和框架上界分别为$\widetilde{B}^{-1}, \ \widetilde{A}^{-1}$, 因此对于任意的$f\in {\cal H}$, 有

$\bigg(\sum\limits_{i\in I}{}|\langle f, S_{G}^{-1}g_{i}\rangle|^{2}\bigg)^{1/2}=\bigg(\sum\limits_{i\in I}{}|\langle \pi_{\widetilde{W}}(f), S_{G}^{-1}g_{i}\rangle|^{2}\bigg)^{1/2} \leq\sqrt{\widetilde{A}^{-1}}\| \pi_{\widetilde{W}}(f)\|.$ (2.7)

下面, 我们来证明这个主要结论.对于任意的$f\in {\cal H}$, 有

$ \begin{align} & \left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{f}_{i}}-{{\pi }_{W}}({{\pi }_{{\tilde{W}}}}(f)) \right\|=\left\| {{\pi }_{W}}(\sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{f}_{i}}-{{\pi }_{{\tilde{W}}}}(f)) \right\| \\ & =\left\| {{\pi }_{W}}(\sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{f}_{i}}-\sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{g}_{i}}) \right\| \\ & =\left\| {{\pi }_{W}}(\sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle ({{f}_{i}}-{{g}_{i}})) \right\| \\ & \le \left\| {{\pi }_{W}} \right\|\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle ({{f}_{i}}-{{g}_{i}}) \right\| \\ & \le \left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle ({{f}_{i}}-{{g}_{i}}) \right\|, \\ \end{align} $

因为$\{g_{i}\}_{i\in I}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动, 由式子 (1.3) 可以得到

$ \begin{align} & \left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle ({{f}_{i}}-{{g}_{i}}) \right\| \\ & \le {{\lambda }_{1}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{f}_{i}} \right\|+{{\lambda }_{2}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{g}_{i}} \right\|+\mu {{(\sum\limits_{i\in I}{{}}|\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{|}^{2}})}^{1/2}}. \\ \end{align} $

又因为

$ \bigg\|\sum\limits_{i\in I}{}\langle f, S_{G}^{-1}g_{i}\rangle {f_i} -\pi_{W}(\pi_{\widetilde{W}}(f))\bigg\|\geq\bigg\|\sum\limits_{i\in I}{}\langle f, S_{G}^{-1}g_{i}\rangle {f_i} \bigg\|-\|\pi_{W}(\pi_{\widetilde{W}}(f))\|, $

因而, 由式子 (2.2) 和式子 (2.7) 可得:对于任意的$f\in {\cal H}$, 有

$ \begin{align} & \left\| {{\pi }_{W}}({{\pi }_{{\tilde{W}}}}(f)) \right\| \\ & \ge (1-{{\lambda }_{1}})\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{f}_{i}} \right\|-{{\lambda }_{2}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{g}_{i}} \right\|-\mu {{(\sum\limits_{i\in I}{{}}|\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{|}^{2}})}^{1/2}} \\ & (1-{{\lambda }_{1}})[\frac{1-{{\lambda }_{2}}}{1+{{\lambda }_{1}}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{g}_{i}} \right\|-\frac{\mu }{1+{{\lambda }_{1}}}{{(\sum\limits_{i\in I}{{}}|\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{|}^{2}})}^{1/2}}] \\ & -{{\lambda }_{2}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{g}_{i}} \right\|-\mu {{(\sum\limits_{i\in I}{{}}|\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{|}^{2}})}^{1/2}} \\ & =[\frac{(1-{{\lambda }_{1}})(1-{{\lambda }_{2}})}{1+{{\lambda }_{1}}}-{{\lambda }_{2}}]\left\| \sum\limits_{i\in I}{{}}\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{g}_{i}} \right\|-[\frac{(1-{{\lambda }_{1}})\mu }{1+{{\lambda }_{1}}}+\mu ]{{(\sum\limits_{i\in I}{{}}|\langle f,S_{G}^{-1}{{g}_{i}}\rangle {{|}^{2}})}^{1/2}} \\ & [\frac{(1-{{\lambda }_{1}})(1-{{\lambda }_{2}})}{1+{{\lambda }_{1}}}-{{\lambda }_{2}}]\left\| {{\pi }_{{\tilde{W}}}}(f) \right\|-\frac{2\mu }{1+{{\lambda }_{1}}}\cdot \sqrt{{{{\tilde{A}}}^{-1}}}\left\| {{\pi }_{{\tilde{W}}}}(f) \right\| \\ & =\frac{(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{N}^{2})}^{1/2}}(1-{{\lambda }_{1}}-2{{\lambda }_{2}})-\mu (3-{{\lambda }_{1}})}{(1+{{\lambda }_{1}})[(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{N}^{2})}^{1/2}}-\mu ]}\left\| {{\pi }_{{\tilde{W}}}}(f). \right\| \\ \end{align} $

证毕.

当定义1.6中, 取$\mu=0$时, 有$N_{T}=N_{U}$, 因而$\delta_{N}=0$, 则有下面的推论.

推论2.1  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的一个框架序列, 且$0\leq\lambda_{1} < 1, 0\leq\lambda_{2} < \frac{1-\lambda_{1}}{2}$.如果$\{g_{i}\}_{i\in I}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, 0)$-扰动, 令$W=\overline{span}\{{f_i} \}_{i\in I}$, $\widetilde{W}=\overline{span}\{g_{i}\}_{i\in I}$, 那么对于任意的$f\in{\cal H}$, 有

$\|\pi_{W}(\pi_{\widetilde{W}}(f))\|\geq\frac{1-\lambda_{1}-2\lambda_{2}}{1+\lambda_{1}}\|\pi_{\widetilde{W}}(f)\|.$ (2.8)

注2.2  文献[8]中命题5.4(ⅱ) 的结果可以简化写为

$ \|\pi_{W}(\pi_{\widetilde{W}}(f))\|\geq\bigg[1-\frac{2(\lambda_{1}+\lambda_{2})}{1-\lambda_{1}^{2}}\bigg]\|\pi_{\widetilde{W}}(f)\|, $

且当$\lambda_{1}, \lambda_{2}\leq1/5$时, 有$1-\frac{2(\lambda_{1}+\lambda_{2})}{1-\lambda_{1}^{2}}>0$.对比推论2.1的式子 (2.8) 有

$ \begin{gathered} \frac{{1 - {\lambda _1} - 2{\lambda _2}}}{{1 + {\lambda _1}}} - [1-\frac{{2({\lambda _1} + {\lambda _2})}}{{1-\lambda _1^2}}] \hfill \\ = \frac{{(1 - {\lambda _1} - 2{\lambda _2})(1 - {\lambda _1}) - 1 + \lambda _1^2 + 2({\lambda _1} + {\lambda _2})}}{{1 - \lambda _1^2}} \hfill \\ = \frac{{2(\lambda _1^2 + {\lambda _1}{\lambda _2})}}{{1 - \lambda _1^2}} \geqslant 0. \hfill \\ \end{gathered} $

因此, 推论2.1个给出了一个比文献[8]中命题5.4(ⅱ) 更大的下界.在推论2.1中, 可以看到当$\lambda_{1}, \lambda_{2} < 1/3$, 就有$\frac{1-\lambda_{1}-2\lambda_{2}}{1+\lambda_{1}}>0$.我们以不同的方法对文献[8]中命题5.4(ⅱ) 的结果的改进, 得到一个更优的下界, 使得在框架序列的$(\lambda_{1}, \lambda_{2}, 0)$-扰动中可选择的$\lambda_{1}, \lambda_{2}$更多, 适用范围更广, 在实际应用中也更加灵活.

类似的, 我们有下面的结论.

定理2.2  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的一个框架序列且框架界为$A, B$, 且$0\leq\lambda_{1}, \lambda_{2} < 1$, 如果序列$\{g_{i}\}_{i\in I}\subset{\cal H}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动, 且满足下面的条件:

(ⅰ) $1-2\lambda_{1}-\lambda_{2}>0;$

(ⅱ) $0\leq\mu < \frac{(1-2\lambda_{1}-\lambda_{2})\sqrt{A}}{2}.$

$W=\overline{span}\{{f_i} \}_{i\in I}$, $\widetilde{W}=\overline{span}\{g_{i}\}_{i\in I}$, 那么对于任意的$f\in {\cal H}$, 都有

$ \|\pi_{\widetilde{W}}(\pi_{W}(f))\|\geq\frac{(1-2\lambda_{1}-\lambda_{2})\sqrt{A}-2\mu}{(1+\lambda_{2})\sqrt{A}}\|\pi_{W}(f)\|. $

  设$S_{F}$$F=\{{f_i} \}_{i\in I}$的框架算子.根据注2.1, 对于任意的$f\in {\cal H}$, 有

$\begin{align} & \sum\limits_{i\in I}{{}}|\langle f, S_{F}^{-1}{{f}_{i}}\rangle {{|}^{2}}=\sum\limits_{i\in I}{{}}|\langle f, {{\pi }_{W}}(S_{F}^{-1}{{f}_{i}})\rangle {{|}^{2}} \\ & =\sum\limits_{i\in I}{{}}|\langle {{\pi }_{W}}(f), S_{F}^{-1}{{f}_{i}}\rangle {{|}^{2}} \\ & \le \frac{1}{A}\|{{\pi }_{W}}(f){{\|}^{2}}. \\ \end{align}$ (2.9)

对于任意的$f\in {\cal H}$, 有

$ \begin{align} & \left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}} \right\|-\left\| {{\pi }_{\widetilde{W}}}({{\pi }_{W}}(f)) \right\| \\ & \le \left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}}-{{\pi }_{\widetilde{W}}}({{\pi }_{W}}(f)) \right\| \\ & =\left\| {{\pi }_{\widetilde{W}}}(\sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}}-{{\pi }_{W}}(f)) \right\| \\ & \le \left\| {{\pi }_{\widetilde{W}}} \right\|\left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}}-\sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{f}_{i}} \right\| \\ & \le \left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle ({{f}_{i}}-{{g}_{i}}) \right\|, \\ \end{align} $

由式子 (1.3) 和式子 (2.7), 我们可以得到

$ \begin{align} & \left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle ({{f}_{i}}-{{g}_{i}}) \right\| \\ & {{\lambda }_{1}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{f}_{i}} \right\|+{{\lambda }_{2}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}} \right\|+\mu {{(\sum\limits_{i\in I}{{}}|\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{|}^{2}})}^{1/2}} \\ & ({{\lambda }_{1}}+\frac{\mu }{\sqrt{A}})\left\| {{\pi }_{W}}(f) \right\|+{{\lambda }_{2}}\left\| \sum\limits_{i\in I}{{}}\langle f,S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}} \right\|. \\ \end{align} $

移项并合并同类项后, 根据式子 (2.1) 和式子 (2.7), 则有

$ \begin{align} & \|{{\pi }_{\widetilde{W}}}({{\pi }_{W}}(f))\|\ge (1-{{\lambda }_{2}})\|\sum\limits_{i\in I}{{}}\langle f, S_{F}^{-1}{{f}_{i}}\rangle {{g}_{i}}\|-\frac{{{\lambda }_{1}}\sqrt{A}+\mu }{\sqrt{A}}\|{{\pi }_{W}}(f)\| \\ & \ge (1-{{\lambda }_{2}})[\frac{1-{{\lambda }_{1}}}{1+{{\lambda }_{2}}}\|\sum\limits_{i\in I}{{}}\langle f, S_{F}^{-1}{{f}_{i}}\rangle {{f}_{i}}\|-\frac{\mu }{1+{{\lambda }_{2}}}{{(\sum\limits_{i\in I}{{}}|\langle f, S_{F}^{-1}{{f}_{i}}\rangle {{|}^{2}})}^{1/2}}] \\ & -\frac{{{\lambda }_{1}}\sqrt{A}+\mu }{\sqrt{A}}\|{{\pi }_{W}}(f)\| \\ & \ge (1-{{\lambda }_{2}})(\frac{1-{{\lambda }_{1}}}{1+{{\lambda }_{2}}}\|{{\pi }_{W}}(f)\|-\frac{\mu }{1+{{\lambda }_{2}}}\cdot \frac{1}{\sqrt{A}}\|{{\pi }_{W}}(f)\|) \\ & -\frac{{{\lambda }_{1}}\sqrt{A}+\mu }{\sqrt{A}}\|{{\pi }_{W}}(f)\| \\ & =\frac{(1-2{{\lambda }_{1}}-{{\lambda }_{2}})\sqrt{A}-2\mu }{(1+{{\lambda }_{2}})\sqrt{A}}\|{{\pi }_{W}}(f)\|. \\ \end{align} $

证毕.

当定义1.6中的$\mu=0$时, 有下面的推论.

推论2.2  设$\{{f_i} \}_{i\in I}\subset{\cal H}$${\cal H}$的一个框架序列, 且$0\leq\lambda_{2} < 1, 0\leq\lambda_{1} < \frac{1-\lambda_{2}}{2}$.如果序列$\{g_{i}\}_{i\in I}\subset{\cal H}$$\{{f_i} \}_{i\in I}$的一个$(\lambda_{1}, \lambda_{2}, 0)$-扰动, 令$W=\overline{span}\{{f_i} \}_{i\in I}$, $\widetilde{W}=\overline{span}\{g_{i}\}_{i\in I}$, 那么对于任意的$f\in{\cal H}$, 有

$ \|\pi_{\widetilde{W}}(\pi_{W}(f))\|\geq\frac{1-2\lambda_{1}-\lambda_{2}}{1+\lambda_{2}}\|\pi_{W}(f)\|. $

推论2.3  我们注意到, 在推论2.1中的结果中只涉及到参数$\lambda_{1}$$\lambda_{2}$, 并没有涉及到框架序列$\{{f_i} \}_{i\in I}$的框架界.因此在参数$\mu=0$的扰动情况下, 我们关于正交投影算子$\pi_{W}$$\pi_{\widetilde{W}}$的讨论是具有对称性的.利用对称性, 在推论2.1的结果中, 将不等式的左边$\|\pi_{W}(\pi_{\widetilde{W}}(f))\|$$\pi_{W}$$\pi_{\widetilde{W}}$进行置换后$\|\pi_{\widetilde{W}}(\pi_{W}(f))\|$, 只需将不等号右边的$\lambda_{1}$$\lambda_{2}$进行置换也可推出推论2.2.

下面给出一个fusion框架系统中局部框架扰动的更为一般的定理.先给出一个引理.

引理2.6[8]  设$\{(W_{i}, v_{i})\}_{i\in I}$${\cal H}$的fusion框架, fusion框架界为$C, D$.选择$0\leq\beta_{1}, $ $\beta_{2} < 1$以及$\varepsilon>0$, 使得$(1-\beta_{1})\sqrt{C}-\varepsilon (\sum\limits_{i\in I}v_{i}^{2})^{1/2}>0$.如果$\{(\widetilde{W}_{i}, v_{i})\}_{i\in I}$$\{(W_{i}, v_{i})\}_{i\in I}$的一个$(\beta_{1}, $ $ \beta_{2}, \varepsilon)$-扰动, 则$\{(\widetilde{W}_{i}, v_{i})\}_{i\in I}$${\cal H}$的fusion框架, 且fusion框架界为

$ \bigg[\frac{(1-\beta_{1})\sqrt{C}-\varepsilon(\sum\limits_{i\in I}v_{i}^{2})^{1/2}}{1+\beta_{2}}\bigg]^{2}, \ \ \ \bigg[\frac{(1+\beta_{1})\sqrt{D}+\varepsilon(\sum\limits_{i\in I}v_{i}^{2})^{1/2}}{1-\beta_{2}}\bigg]^{2}. $

定理2.3  设$\{(W_{i}, v_{i}, \{f_{ij}\}_{j\in J_{i}})\}_{i\in I}$${\cal H}$的一个fusion框架系统, 且其fusion框架界为$C, D$.对于每个$i\in I$, $\{f_{ij}\}_{j\in J_{i}}$$W_{i}$的框架, 框架下界为$A_{i}$$0 < A=\inf\limits_{i\in I}A_{i}$.假设对每个$i\in I$, $\{g_{ij}\}_{j\in J_{i}}$$\{f_{ij}\}_{j\in J_{i}}$的一个$(\lambda_{1}, \lambda_{2}, \mu)$-扰动, 令$\widetilde{W}_{i}=\overline{span}\{g_{ij}\}_{j\in J_{i}}$, 且$0\leq\lambda_{1}, \lambda_{2} < 1$满足下面的条件:

(ⅰ) $ 1-\lambda_{1}-2\lambda_{2}>0;$

(ⅱ) $1-2\lambda_{1}-\lambda_{2}>0;$

(ⅲ) $ 0\leq\mu < \min\Big\{\frac{(1-\lambda_{1})\sqrt{A}(1-\delta_{N}^{2})^{1/2}(1-\lambda_{1}-2\lambda_{2})}{3-\lambda_{1}}, \ \frac{(1-2\lambda_{1}-\lambda_{2})\sqrt{A}}{2} \Big\}, $

其中$\delta_{N_{i}}=\delta (N_{T_{i}}, N_{U_{i}})$, 且$1>\delta_{N}=\sup\limits_{i\in I}\delta_{N_{i}}$, $T_{i}$$U_{i}$为分别由如下定义的有界线性算子:

$ \begin{gathered} {T_i}:{\ell ^2} \to H,\;{T_i}(c) = \sum\limits_{j \in {J_i}} {} {c_j}{f_{ij}},\;c = {\{ {c_j}\} _{j \in {J_i}}} \in {\ell ^2}({J_i}), \hfill \\ {U_i}:{\ell ^2} \to H,\;{U_i}(c) = \sum\limits_{j \in {J_i}} {} {c_j}{g_{ij}},\;c = {\{ {c_j}\} _{j \in {J_i}}} \in {\ell ^2}({J_i}). \hfill \\ \end{gathered} $

如果选取$\varepsilon_{1}, \varepsilon_{2}>0$使得

$ \begin{align} & 1-\varepsilon _{1}^{2}={{[\frac{(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{N}^{2})}^{1/2}}(1-{{\lambda }_{1}}-2{{\lambda }_{2}})-\mu (3-{{\lambda }_{1}})}{(1+{{\lambda }_{1}})[(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{N}^{2})}^{1/2}}-\mu ]}]}^{2}}, \\ & 1-\varepsilon _{2}^{2}={{[\frac{(1-2{{\lambda }_{1}}-{{\lambda }_{2}})\sqrt{A}-2\mu }{(1+{{\lambda }_{2}})\sqrt{A}}]}^{2}}, \\ \end{align} $

$\sqrt{C}-\sqrt{\varepsilon_{1}+\varepsilon_{2}}(\sum\limits_{i\in I}v_{i}^{2})^{1/2}>0$, 那么$\{(\widetilde{W}_{i}, v_{i})\}_{i\in I}$${\cal H}$的一个fusion框架, 且其fusion框架界为

$ \bigg[\sqrt{C}-\sqrt{\varepsilon_{1}+\varepsilon_{2}}\bigg(\sum\limits_{i\in I}v_{i}^{2}\bigg)^{1/2}\bigg]^{2}, \ \bigg[\sqrt{D}+\sqrt{\varepsilon_{1}+\varepsilon_{2}}\bigg(\sum\limits_{i\in I}v_{i}^{2}\bigg)^{1/2}\bigg]^{2}. $

  由$0\leq\lambda_{1}, \lambda_{2} < 1$以及条件 (i) 可知$0 < \frac{1-\lambda_{1}-2\lambda_{2}}{3-\lambda_{1}} < 1$, 那么对于每个$i\in I$, 由条件 (ⅲ) 可知

$ \begin{align} & \mu <\frac{(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{N}^{2})}^{1/2}}(1-{{\lambda }_{1}}-2{{\lambda }_{2}})}{3-{{\lambda }_{1}}} \\ & <\frac{(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{{{N}_{i}}}^{2})}^{1/2}}(1-{{\lambda }_{1}}-2{{\lambda }_{2}})}{3-{{\lambda }_{1}}} \\ & <(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{{{N}_{i}}}^{2})}^{1/2}}. \\ \end{align} $

再由定理2.1, 对于任意的$f\in {\cal H}$, 有

$ \begin{align} & \|{{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}}=\|{{\pi }_{{{W}_{i}}}}{{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{W}_{i}}}}){{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}} \\ & \ge {{[\frac{(1-{{\lambda }_{1}})\sqrt{{{A}_{i}}}{{(1-\delta _{{{N}_{i}}}^{2})}^{1/2}}(1-{{\lambda }_{1}}-2{{\lambda }_{2}})-\mu (3-{{\lambda }_{1}})}{(1+{{\lambda }_{1}})[(1-{{\lambda }_{1}})\sqrt{{{A}_{i}}}{{(1-\delta _{{{N}_{i}}}^{2})}^{1/2}}-\mu]}]}^{2}}\|{{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}} \\ & +\|(I-{{\pi }_{{{W}_{i}}}}){{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}} \\ & ={{[\frac{1-{{\lambda }_{1}}-2{{\lambda }_{2}}}{1+{{\lambda }_{1}}}-\frac{2\mu (1+{{\lambda }_{2}})}{(1+{{\lambda }_{1}})[(1-{{\lambda }_{1}})\sqrt{{{A}_{i}}}{{(1-\delta _{{{N}_{i}}}^{2})}^{1/2}}-\mu]}]}^{2}}\|{{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}} \\ & +\|(I-{{\pi }_{{{W}_{i}}}}){{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}} \\ & \ge {{[\frac{1-{{\lambda }_{1}}-2{{\lambda }_{2}}}{1+{{\lambda }_{1}}}-\frac{2\mu (1+{{\lambda }_{2}})}{(1+{{\lambda }_{1}})[(1-{{\lambda }_{1}})\sqrt{A}{{(1-\delta _{N}^{2})}^{1/2}}-\mu]}]}^{2}}\|{{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}} \\ & +\|(I-{{\pi }_{{{W}_{i}}}}){{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}}=(1-\varepsilon _{1}^{2})\|{{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{W}_{i}}}}){{\pi }_{{{\widetilde{W}}_{i}}}}(f){{\|}^{2}}, \\ \end{align} $

因此对于每个$i\in I$, 不等式$\|(I-\pi_{W_{i}})\pi_{\widetilde{W}_{i}}(f)\|^{2}\leq\varepsilon_{1}^{2}\|\pi_{\widetilde{W}_{i}}(f)\|^{2}$成立.

同样对于每个$i\in I$, 由定理2.2, 对于任意的$f\in {\cal H}$, 有

$ \begin{align} & \|{{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}=\|{{\pi }_{{{\widetilde{W}}_{i}}}}{{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{\widetilde{W}}_{i}}}}){{\pi }_{{{W}_{i}}}}(f){{\|}^{2}} \\ & \ge {{[\frac{(1-2{{\lambda }_{1}}-{{\lambda }_{2}})\sqrt{{{A}_{i}}}-2\mu }{(1+{{\lambda }_{2}})\sqrt{{{A}_{i}}}}]}^{2}}\|{{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{\widetilde{W}}_{i}}}}){{\pi }_{{{W}_{i}}}}(f){{\|}^{2}} \\ & ={{[\frac{1-2{{\lambda }_{1}}-{{\lambda }_{2}}}{1+{{\lambda }_{2}}}-\frac{2\mu }{(1+{{\lambda }_{2}})\sqrt{{{A}_{i}}}}]}^{2}}\|{{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{\widetilde{W}}_{i}}}}){{\pi }_{{{W}_{i}}}}(f){{\|}^{2}} \\ & \ge {{[\frac{1-2{{\lambda }_{1}}-{{\lambda }_{2}}}{1+{{\lambda }_{2}}}-\frac{2\mu }{(1+{{\lambda }_{2}})\sqrt{A}}]}^{2}}\|{{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{\widetilde{W}}_{i}}}}){{\pi }_{{{W}_{i}}}}(f){{\|}^{2}} \\ & =(1-\varepsilon _{2}^{2})\|{{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}+\|(I-{{\pi }_{{{\widetilde{W}}_{i}}}}){{\pi }_{{{W}_{i}}}}(f){{\|}^{2}}, \\ \end{align} $

因此对于每个$i\in I$, 不等式$\|(I-\pi_{\widetilde{W}_{i}})\pi_{W_{i}}(f)\|^{2}\leq\varepsilon_{2}^{2}\|\pi_{W_{i}}(f)\|^{2}$成立.

所以对于任意的$f\in {\cal H}$, 都有

$ \begin{align} & \|({{\pi }_{{{W}_{i}}}}-{{\pi }_{{{\widetilde{W}}_{i}}}})(f){{\|}^{2}}=\langle ({{\pi }_{{{W}_{i}}}}-{{\pi }_{{{\widetilde{W}}_{i}}}})(f),({{\pi }_{{{W}_{i}}}}-{{\pi }_{{{\widetilde{W}}_{i}}}})(f)\rangle \\ & =\langle {{({{\pi }_{{{W}_{i}}}}-{{\pi }_{{{\widetilde{W}}_{i}}}})}^{2}}(f),f\rangle \\ & =\langle ({{\pi }_{{{W}_{i}}}}-{{\pi }_{{{\widetilde{W}}_{i}}}}{{\pi }_{{{W}_{i}}}}+{{\pi }_{{{\widetilde{W}}_{i}}}}-{{\pi }_{{{W}_{i}}}}{{\pi }_{{{\widetilde{W}}_{i}}}})(f),f\rangle \\ & \le \|(I-{{\pi }_{{{\widetilde{W}}_{i}}}})({{\pi }_{{{W}_{i}}}}(f))+(I-{{\pi }_{{{W}_{i}}}})({{\pi }_{{{\widetilde{W}}_{i}}}}(f))\|\|f\| \\ & \le \|(I-{{\pi }_{{{\widetilde{W}}_{i}}}}){{\pi }_{{{W}_{i}}}}(f)\|\|f\|+\|(I-{{\pi }_{{{W}_{i}}}}){{\pi }_{{{\widetilde{W}}_{i}}}}(f)\|\|f\| \\ & \le {{\varepsilon }_{1}}\|{{\pi }_{{{\widetilde{W}}_{i}}}}(f)\|\|f\|+{{\varepsilon }_{2}}\|{{\pi }_{{{W}_{i}}}}(f)\|\|f\| \\ & \le {{\varepsilon }_{1}}\|{{\pi }_{{{\widetilde{W}}_{i}}}}\|\|f{{\|}^{2}}+{{\varepsilon }_{2}}\|{{\pi }_{{{W}_{i}}}}\|\|f{{\|}^{2}} \\ & \le ({{\varepsilon }_{1}}+{{\varepsilon }_{2}})\|f{{\|}^{2}}. \\ \end{align} $

再由引理2.6可得$\{(\widetilde{W}_{i}, v_{i})\}_{i\in I}$${\cal H}$的一个fusion框架, 且其fusion框架界为

$ \bigg[\sqrt{C}-\sqrt{\varepsilon_{1}+\varepsilon_{2}}\bigg(\sum\limits_{i\in I}v_{i}^{2}\bigg)^{1/2}\bigg]^{2}, \ \bigg[\sqrt{D}+\sqrt{\varepsilon_{1}+\varepsilon_{2}}\bigg(\sum\limits_{i\in I}v_{i}^{2}\bigg)^{1/2}\bigg]^{2}. $

证毕.

注2.4  定理2.3将文献[8]中定理5.6的局部框架的$(\lambda_{1}, \lambda_{2})$-扰动形式推广到更为一般的$(\lambda_{1}, \lambda_{2}, \mu)$-扰动形式的情况, 给出了局部框架的$(\lambda_{1}, \lambda_{2}, \mu)$-扰动的fusion框架系统的稳定性定理, 结果更具一般性和代表性.同时, 当$\mu=0$时, 定理2.3给出了比文献[8]中的定理5.6更优的fusion框架界, 改进了由Casazza, Kutyniok和Li等得到的著名结果.

参考文献
[1] Christensen O.Frames and Bases:An Introductory Course.Boston:Birkhäuser, 2008
[2] Casazza P G.Finite frames:Theory and Applications.Boston:Birkhäuser, 2012
[3] Chan R H, Riemenschneider S D, Shen L, et al. Tight frame:an efficient way for high-resolution image reconstruction. Appl Comput Harmon Anal, 2004, 17: 91–115. DOI:10.1016/j.acha.2004.02.003
[4] Strohmer T, Heath J R. Grassmanian frames with applications to coding and communications. Appl Comput Harmon Anal, 2003, 14: 257–275. DOI:10.1016/S1063-5203(03)00023-X
[5] Strohmer T. Approximation of dual Gabor frames, window decay, and wireless communications. Appl Comput Harmon Anal, 2001, 11(2): 243–262. DOI:10.1006/acha.2001.0357
[6] Benedetto J J, Powell A M, Yilmaz O. Second order sigma-delta quantization of finite frame expansions. Appl Comput Harmon Anal, 2006, 20: 126–148. DOI:10.1016/j.acha.2005.04.003
[7] Casazza P G, Kutyniok G. Frames of Subspaces, in Wavelets, Frames and Operators Theory. Amer Math Soc, 2004, 35: 87–113.
[8] Casazza P G, Kutyniok G, Li Shidong. Fusion frames and distributed processing. Appl Comput Harmon Anal, 2008, 25(1): 114–132. DOI:10.1016/j.acha.2007.10.001
[9] Găvruta P. On the duality of fusion frames. J Math Anal Appl, 2007, 333: 871–879. DOI:10.1016/j.jmaa.2006.11.052
[10] Asgari M S. New characterizations of fusion frames (frames of subspaces). Proc Indian Acad Sci, 2009, 119(3): 369–382.
[11] Craig M S, Genter R L. Geophone array formation and semblance evaluation. Geophysics, 2006, 71: 1–8.
[12] Iyengar S S, Brooks R R.Distributed Sensor Network.London:Chapman Hall, 2005
[13] Kato T.Perturbation Theory for Linear Operator.New York:Springer-Verlag, 1984
[14] Christensen O. Operators with closed range, pseudo-inverses, and perturbation of frames for a subspace. Canad Math Bull, 1999, 42(1): 37–45. DOI:10.4153/CMB-1999-004-5