数学物理学报, 2024, 44(1): 12-25

$2\times2$分块对角算子矩阵拟谱的精细刻画

申润拴,, 侯国林,*

内蒙古大学数学科学学院 呼和浩特 010021

The Fine Pseudo-spectra of$2 \times 2$Diagonal Block Operator Matrices

Shen Runshuan,, Hou Guolin,*

School of Mathematical Sciences, Inner Mongolia University, Hohhot 010021

通讯作者: 侯国林, E-mail:smshgl@imu.edu.cn

收稿日期: 2022-11-30   修回日期: 2023-05-15  

基金资助: 国家自然科学基金(12261064)
国家自然科学基金(11861048)
内蒙古自然科学基金(2021MS01004)

Received: 2022-11-30   Revised: 2023-05-15  

Fund supported: National Natural Science Foundation of China(12261064)
National Natural Science Foundation of China(11861048)
Natural Science Foundation of Inner Mongolia(2021MS01004)

作者简介 About authors

申润拴,E-mail:65240147@qq.com

摘要

设$A$,$B$为可分 Hilbert 空间$X$中的稠定闭线性算子,$M_{0}=\left(\begin{array}{cc}{A} & {0}\\{0}& {B}\end{array} \right)$表示$2\times2$分块算子矩阵. 文中精细刻画算子矩阵$M_{0}$在对角扰动情形下的拟点谱、拟剩余谱与拟连续谱, 所得结论与点谱、剩余谱和连续谱的结果进行了比较, 并用例子进行了辅证. 最后, 采用空间分解技巧, 用主对角元的信息刻画$M_{0}$在上三角扰动情形下的拟点谱分布.

关键词: 算子矩阵; 拟点谱; 拟剩余谱; 拟连续谱; 拟零子空间

Abstract

Let$A$,$B$be densely closed linear operators in a separable Hilbert space$X$and$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$be the corresponding$2\times2$block operator matrices. In this paper, we establish the fine pseudo-spectra of$M_{0}$including the pseudo-point spectrum, the pseudo-residual spectrum, and the pseudo-continuous spectrum under diagonal perturbation, which are, respectively, compared with its point spectrum, residual spectrum, and continuous spectrum. And a concrete example is constructed to justify the proved result. Finally, we obtain the pseudo-point spectrum of$M_{0}$under the upper-triangular perturbation by using the technology of space decomposition.

Keywords: Operator matrices; Pseudo-point spectrum; Pseudo-residual spectrum; Pseudo-continuous spectrum; Pseudo-null space

PDF (557KB) 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

申润拴, 侯国林. $2\times2$分块对角算子矩阵拟谱的精细刻画[J]. 数学物理学报, 2024, 44(1): 12-25

Shen Runshuan, Hou Guolin. The Fine Pseudo-spectra of$2 \times 2$Diagonal Block Operator Matrices[J]. Acta Mathematica Scientia, 2024, 44(1): 12-25

1 引言

线性算子谱理论是泛函分析的重要组成部分, 它作为算子特征的一项衡量指标, 在航空学、量子力学、生态学等许多学科中有重要应用. 在相关应用中, 线性算子谱仍存在一些亟需解决的问题, 比如, 线性算子的谱不易精确计算, 往往通过误差分析对其进行近似估计; 线性算子谱对扰动非常敏感等. 为了解决这些问题, 人们提出了拟谱的概念, 拟谱能够更好地刻画非正规系统的稳定性[1-5], 学者们从不同方面对拟谱进行了刻画, 并得到较好的结论. 文献[6]研究了线性算子的和、差与乘积之拟谱的保持问题. 文献[7]研究了非自伴算子在广义强预解意义下拟谱的局部收敛性. 文献[8-11]论述了线性算子的本质拟谱、本质近似拟谱与拟亏谱. 文献[12]探讨了带有最小拟谱的算子与算子正规性之间的关系. 针对双结构分块矩阵, 文献[13]研究了拟谱与结构拟谱的等价关系, 将[14]中结构矩阵的一些结论推广到了双结构情形. 文献[15]讨论了分块算子矩阵对角扰动下的本质拟谱在一定条件下与其对角元本质拟谱的包含关系. 文献[16,17]利用分块矩阵对角元的拟谱刻画了整个分块矩阵拟谱的上下界. 对于拟谱的更多结论请参阅[1,18,19]及其引用文献.

算子矩阵是以算子为元素的矩阵, 在数学物理、应用力学、最优控制等领域有重要应用. 近二十年来,$2\times2$上三角算子矩阵和$2\times2$算子矩阵的谱分析和谱补问题取得了丰富的成果, 参见[20-24]和塞尔维亚学者 Dragana S. 最近出版的专著[25]. 鉴于拟谱良好的稳定性, 类比于线性算子谱分析的研究方法, 文献[26]对拟谱作了细化, 讨论了闭线性算子与其共轭算子的拟点谱、拟剩余谱、拟连续谱之间的关系, 进而给出了 Hamilton 算子的拟谱结构, 将 Hamilton 矩阵的谱对称性推广到拟谱情形, 并在一定条件下用对角元的拟谱刻画了整体算子的拟谱分布. 受文献[15,17,26]的启发, 本文从分块算子矩阵角度出发, 尝试利用对角元的拟点谱、拟剩余谱、拟连续谱刻画整个分块算子矩阵的拟点谱、拟剩余谱、拟连续谱. 算子矩阵的扰动算子形式较复杂,$2\times 2$对角算子矩阵的扰动算子可能是对角的, 也可能是上 (下) 三角的, 也可以为一般$2\times 2$形式, 这给拟谱的精细研究带来了困难, 本文剖析$2\times 2$对角算子矩阵在对角扰动下的拟点谱、拟剩余谱和拟连续谱的分布, 所得结论与谱分析的结论进行了对比, 进一步还得到$2\times2$分块对角算子矩阵在上三角扰动下的拟点谱分布. 文中构造的一些实例辅证了结论的有效性.

2 预备知识

在下文中, 始终用$X$表示无穷维复可分的 Hilbert 空间. 符号$\mathcal{B}(X), \mathcal{C}(X)$分别表示$X$上的有界线性算子全体构成的集合和$X$中稠定闭线性算子全体构成的集合. 设$T$是$X$中的稠定线性算子, 用$T^{\ast}$,$\mathcal{D}(T)$,$\mathcal{N}(T)$,$\mathcal{R}(T)$和$\mathcal{R}(T)^{\bot}$分别表示$T$的共轭算子、 定义域、 零子空间、 值域和值域的正交补. 用$T\mid_{\mathcal{M}}$表示$T$在空间$\mathcal{M}$上的限制. 统一用$I$表示给定空间上的单位算子. 若$T\in \mathcal{C}(X)$, 我们用$\sigma(T)$,$\rho(T)$,$\sigma_{p}(T)$,$\sigma_{r}(T)$和$\sigma_{c}(T)$分别表示$T$的谱集、预解集、点谱、剩余谱和连续谱.

下面给出稠定闭线性算子的拟谱、拟点谱、拟剩余谱与拟连续谱定义.

定义 2.1[1] 令$T\in \mathcal{C}(X)$. 对于任意的$\varepsilon>0$,算子$T$的拟谱$\sigma_{\varepsilon}(T)$定义如下

$\sigma_{\varepsilon}(T)=\sigma(T)\cup\{\lambda\in\rho(T): \|(T-\lambda I)^{-1}\|>\frac{1}{\varepsilon}\}.$

利用扰动原理, 等价地可得到[2],

$\begin{align*}\sigma_{\varepsilon}(T)=\bigcup_{\| E\|<\varepsilon}{\sigma(T+E)} =\{\lambda\in\mathbb{C}:\text{存在} E\in\mathcal{B}(X), \| E\|<\varepsilon, \text{使得} \lambda\in\sigma(T+E)\}. \end{align*}$

定义 2.2[26] 令$T\in \mathcal{C}(X)$. 对于任意的$\varepsilon>0$, 定义

$\begin{align*} \sigma_{\varepsilon, p}(T):&=\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}{\sigma_p}(T+E)\\ & =\{\lambda\in\mathbb{C}: \text{存在} E\in\mathcal{B}(X), \| E\|<\varepsilon, \text{使得} \lambda\in\sigma_{p}(T+E)\};\end{align*}$
$\begin{align*} \sigma_{\varepsilon, r}(T):&=\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}{\sigma_r}(T+E)\\ & =\{\lambda\in\mathbb{C}: \text{存在} E\in\mathcal{B}(X), \| E\|<\varepsilon, \text{使得} \lambda\in\sigma_{r}(T+E)\};\\\end{align*}$
$\begin{align*} \sigma_{\varepsilon, c}(T):&=\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}{\sigma_c}(T+E)\\ & =\{\lambda\in\mathbb{C}: \text{存在} E\in\mathcal{B}(X), \| E\|<\varepsilon, \text{使得} \lambda\in\sigma_{c}(T+E)\},\end{align*}$

称$\sigma_{\varepsilon, p}(T)$,$\sigma_{\varepsilon, r}(T)$和$\sigma_{\varepsilon, c}(T)$分别为$T$的拟点谱、 拟剩余谱和拟连续谱.

注 2.1 容易看出

$\begin{align*} &\sigma(T)\subset\sigma_{\varepsilon}(T), \sigma_{\alpha}(T)\subset\sigma_{\varepsilon,\alpha}(T), \alpha\in\{p, r, c\},\\ & \sigma_{\varepsilon}(T)=\sigma_{\varepsilon, p}(T)\cup\sigma_{\varepsilon, r}(T)\cup\sigma_{\varepsilon, c}(T). \end{align*}$

当$\varepsilon\rightarrow 0$时, 拟点谱、拟剩余谱和拟连续谱分别退化为点谱、剩余谱和连续谱; 但对较大的正数$\varepsilon$, 拟点谱、拟剩余谱、拟连续谱之间可能是有交集的 (参见文献[26]).

定义 2.3 设$T\in\mathcal{C}(X)$, 对于任意的$\varepsilon>0$,$T$的拟预解集$\rho_{\varepsilon}(T)$定义为

$\begin{align*} \rho_{\varepsilon}(T)&=\bigcap\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}{\rho(T+E)} =\{\lambda\in\mathbb{C}: \text{对于任意的} E\in\mathcal{B}(X), \| E\|<\varepsilon, \text{均有} \lambda\in\rho(T+E)\}. \end{align*}$

注 2.2 设$\varepsilon>0$, 则$\rho_{\varepsilon}(T)=\mathbb{C}\setminus \sigma_{\varepsilon}(T)$且$\rho_{\varepsilon}(T)\subset\rho(T)$.

注 2.3 文中考虑的算子矩阵$M_{0}$为对角算子矩阵, 在研究其精细拟谱时, 扰动算子$E$应为一般的$2\times 2$分块算子矩阵. 本文采用从简单到复杂的思路, 首先探讨$\sigma^{D}_{\varepsilon, p}(M_{0})$,$\sigma^{D}_{\varepsilon, r}(M_{0})$和$\sigma^{D}_{\varepsilon, c}(M_{0})$的分布, 接着给出了$\sigma^{T}_{\varepsilon, p}(M_{0})$的刻画, 这为精细拟谱$\sigma_{\varepsilon, p}(M_{0})$,$\sigma_{\varepsilon, r}(M_{0})$和$\sigma_{\varepsilon, c}(M_{0})$的最终描述奠定了基础. 这里的记号$\sigma^{D}_{\varepsilon, \ast}(M_{0})$和$\sigma^{T}_{\varepsilon, \ast}(M_{0}) (\ast\in\{p, r, c\})$分别表示$M_{0}$在对角扰动和上三角扰动情形下的精细拟谱.

定义 2.4 假设$T\in\mathcal{C}(X)$且$\varepsilon>0$. 定义$T$的$\varepsilon$-零子空间$\mathcal{N}_{\varepsilon}(T)$和$\varepsilon$-值域$\mathcal{R}_{\varepsilon}(T)$分别为

$\begin{align*} \mathcal{N}_{\varepsilon}(T)& =\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\ \| E\|<\varepsilon}}\mathcal{N}(T+E) =\{x\in\mathcal{D}(T): \text{使得} (T+E)x=0, \text{其中} E\in\mathcal{B}(X), \| E\|<\varepsilon\};\\ \mathcal{R}_{\varepsilon}(T)&=\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\ \| E\|<\varepsilon}}\mathcal{R}(T+E) =\{(T+E)x: x\in \mathcal{D}(T), E\in\mathcal{B}(X) \text{且} \| E\|<\varepsilon\}. \end{align*}$

并称$\alpha_{\varepsilon}(T)=\dim\mathcal{N}_{\varepsilon}(T)$为$T$的$\varepsilon$-零度,$\beta_{\varepsilon}(T)=\dim\mathcal{R}_{\varepsilon}(T)^{\perp}$为$T$的$\varepsilon$-余维数.

3 主要结论及其证明

本节主要讨论$2\times 2$分块算子矩阵在不同扰动情形下的拟点谱、拟剩余谱和拟连续谱, 用对角元的信息给出它们的精细刻画. 首先, 我们考虑扰动算子为对角情形, 即

$E=\left( \begin{array} {cc}{E_{11}} & {0}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X), $

其中$E_{ii}\in\mathcal{B}(X), \|E_{ii}\|<\varepsilon, i=1, 2$, 并且$\|E\|:=\max\limits_{i=1, 2}\{\|E_{ii}\|\}<\varepsilon$.

对于拟点谱而言, 我们可得到类似于点谱的性质.

定理 3.1 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} \sigma_{\varepsilon, p}^{D}(M_{0})=\sigma_{\varepsilon, p}(A)\cup\sigma_{\varepsilon, p}(B). \end{equation*}$

若$\lambda\in\sigma_{\varepsilon, p}(A)$, 则存在$E_{1}\in\mathcal{B}(X)$,$\| E_{1}\|<\varepsilon$, 使得$\lambda\in\sigma_{p}(A+E_{1})$, 即存在非零元素$x _{0}\in \mathcal{D}(A)$, 使得$(A+E_{1}-\lambda I)x_{0}=0$. 令$x=\left( \begin{array} {c}{x_{0}} \\ {0} \end{array} \right)\neq0$且$E=\left( \begin{array} {cc}{E_{1}} & {0}\\ {0}& {0} \end{array} \right).$显然,$E\in\mathcal{B}(X\oplus X), \| E\|=\| E_{1}\|<\varepsilon$, 并且有

$\begin{equation*}(M_{0}+E-\lambda I)x=\left( \begin{array} {cc}{A+E_{1}-\lambda I} & {0}\\ {0}& {B-\lambda I} \end{array} \right)\left( \begin{array} {c}{x_{0}} \\ {0} \end{array} \right)=0. \end{equation*}$

因此,$\lambda\in\sigma_{\varepsilon, p}^{D}(M_{0}).$

同理,$\sigma_{\varepsilon, p}(B)\subset\sigma_{\varepsilon, p}^{D}(M_{0}).$

若$\lambda\in\sigma_{\varepsilon, p}^{D}(M_{0}),$则存在$E=\left( \begin{array} {cc}{E_{11}} & {0}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X), \|E\|<\varepsilon,$使得$\lambda\in\sigma_{p}(M_{0}+E).$故存在不全为零的元素$f, g$, 使得

$\begin{equation*} (M_{0}+E-\lambda I)\left( \begin{array} {c}{f} \\ {g} \end{array} \right)=\left( \begin{array} {cc}{A+E_{11}-\lambda I} & {0}\\ {0}& {B+E_{22}-\lambda I} \end{array} \right)\left( \begin{array} {c}{f} \\ {g} \end{array} \right)=0. \end{equation*}$

若$f=0$, 则$g\neq0$, 进而得到$\lambda\in\sigma_{\varepsilon, p}(B).$若$f\neq0$, 则$\lambda\in\sigma_{\varepsilon, p}(A).$因此, 我们得到$\lambda\in\sigma_{\varepsilon, p}(A)\cup\sigma_{\varepsilon, p}(B)$. 证毕.

关于对角算子矩阵的剩余谱, 我们引用如下结论.

命题 3.1[20] 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} \sigma_{r}(M_{0})=(\sigma_{r}(A)\setminus\sigma_{p}(B))\cup(\sigma_{r}(B)\setminus\sigma_{p}(A)). \end{equation*}$

由于拟剩余谱和拟点谱在一定条件下是有交集的, 因此上述关于剩余谱的结论不能完全类推到拟剩余谱情形, 我们得到如下包含关系.

命题 3.2 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} (\sigma_{\varepsilon, r}(A)\setminus\sigma_{\varepsilon, p}(B))\cup(\sigma_{\varepsilon, r}(B)\setminus\sigma_{\varepsilon, p}(A))\subseteq\sigma_{\varepsilon, r}^{D}(M_{0}) \subseteq\sigma_{\varepsilon, r}(A)\cup\sigma_{\varepsilon, r}(B). \end{equation*}$

首先证明$\sigma_{\varepsilon, r}(A)\setminus\sigma_{\varepsilon, p}(B)\subseteq\sigma_{\varepsilon, r}^{D}(M_{0}).$假设$\lambda\in\sigma_{\varepsilon, r}(A)\setminus\sigma_{\varepsilon, p}(B)$, 则存在$E_{1}\in\mathcal{B}(X),$$\|E_{1}\|<\varepsilon,$使得$\mathcal{N}(A+E_{1}-\lambda I)=\{0\},$$\overline{\mathcal{R}(A+E_{1}-\lambda I)}\neq X$且对于任意的$E_{2}\in\mathcal{B}(X),$$\|E_{2}\|<\varepsilon,$使得$\mathcal{N}(B+E_{2}-\lambda I)=\{0\}.$令

$\begin{equation*} E=\left( \begin{array} {cc}{E_{1}} & {0}\\ {0}& {E_{2}} \end{array} \right)\in\mathcal{B}(X\oplus X), \end{equation*}$

则$\|E\|<\varepsilon$且

$M_{0}+E-\lambda I=\left( \begin{array} {cc}{A+E_{1}-\lambda I} & {0}\\ {0}& {B+E_{2}-\lambda I} \end{array} \right), \overline{\mathcal{R}(M_{0}+E-\lambda I)}\neq X\oplus X.$

这表明$\lambda\in\sigma_{\varepsilon, r}^{D}(M_{0}).$

同理可证$(\sigma_{\varepsilon, r}(B)\setminus\sigma_{\varepsilon, p}(A))\subseteq\sigma_{\varepsilon, r}^{D}(M_{0}).$

下证$\sigma_{\varepsilon, r}^{D}(M_{0})\subseteq\sigma_{\varepsilon, r}(A)\cup\sigma_{\varepsilon, r}(B).$若$\lambda\in\sigma_{\varepsilon, r}^{D}(M_{0}),$根据定义 2.2 可知, 存在$E=\left( \begin{array} {cc}{E_{11}} & {0}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X)$, 使得

$\mathcal{N}(M_{0}+E-\lambda I)=\{0\} \text{且} \overline{\mathcal{R}(M_{0}+E-\lambda I)}\neq X\oplus X.$

于是,

$\mathcal{N}(A+E_{11}-\lambda I)=\{0\} \text{且} \mathcal{N}(B+E_{22}-\lambda I)=\{0\}.$

又由$\overline{\mathcal{R}(M_{0}+E-\lambda I)}\neq X\oplus X$知,$\overline{\mathcal{R}(A+E_{11}-\lambda I)}\neq X$或$\overline{\mathcal{R}(B+E_{22}-\lambda I)}\neq X.$所以, 当$\overline{\mathcal{R}(A+E_{11}-\lambda I)}\neq X$时, 有$\lambda\in\sigma_{\varepsilon, r}(A);$当$\overline{\mathcal{R}(B+E_{22}-\lambda I)}\neq X$时, 我们得到$\lambda\in\sigma_{\varepsilon, r}(B).$证毕.

然而,$\sigma_{\varepsilon, r}^{D}(M_{0})\subseteq(\sigma_{\varepsilon, r}(A)\setminus\sigma_{\varepsilon, p}(B))\cup(\sigma_{\varepsilon, r}(B)\setminus\sigma_{\varepsilon, p}(A))$是不一定成立的, 见下例.

例 1 设$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$表示$2\times2$分块对角算子矩阵, 其中$A, B$分别是 Hilbert 空间$X=l^{2}$中给定的有界线性算子,

$\begin{gathered} Bx=(0, x_{1}, x_{2}, x_{3}, \cdots ), \\ Ax=(x_{1}, x_{1}+x_{2}, x_{2}+x_{3}, x_{3}+x_{4}, \cdots ), \end{gathered}$

对任意的$x=(x_{1}, x_{2}, x_{3}, x_{4}, \cdots )\in X.$

根据文献[26,例 2.3], 我们得到: 当$\varepsilon>\frac{\sqrt{5}}{2}$时,

$\begin{equation*} \frac{1}{2}\in\sigma_{\varepsilon, p}(A)\cap\sigma_{\varepsilon, r}(A). \end{equation*}$

下面验证当$\varepsilon>\frac{\sqrt{5}}{2}$时, 同样有$\frac{1}{2}\in\sigma_{\varepsilon, p}(B)\cap\sigma_{\varepsilon, r}(B)$. 为此, 对于任意的$x=(x_{1}, x_{2}, \cdots )\in X$, 选取扰动算子$E_{1}$如下,

$E_{1}x=(0, \frac{\varepsilon}{2}x_{1}, 0, 0, \cdots ).$

经计算得,$\|E_{1}\|\leq\frac{\varepsilon}{2}<\varepsilon.$

$\begin{gathered} (B+E_{1})x=(0, (\frac{\varepsilon}{2}+1)x_{1}, x_{2}, x_{3}, \cdots ), \\ (\frac {1}{2}I-B-E_{1})x=(\frac {1}{2}x_{1}, \frac {1}{2}x_{2}-(\frac{\varepsilon}{2}+1)x_{1}, \frac {1}{2}x_{3}-x_{2}, \cdots ). \end{gathered}$

显然,$\frac {1}{2}I-B-E_{1}$是单射. 为了说明$\frac {1}{2}\in\sigma_{\varepsilon, r}(B)$, 只需验证$\overline{\mathcal{R}(\frac {1}{2}I-B-E_{1})}\neq X.$

对于任意的$x=(x_{1}, x_{2}, x_{3}, x_{4}, \cdots )\in X$, 经计算得到

$\begin{gathered} B^{\ast}x=(x_{2}, x_{3}, x_{4}, \cdots ), \\ E^{\ast}_{1}x=(\frac{\varepsilon}{2}x_{2}, 0, 0, \cdots ). \end{gathered}$

进而

$\begin{gathered}(B^{\ast}+E^{\ast}_{1})x=((\frac{\varepsilon}{2}+1)x_{2}, x_{3}, x_{4}, \cdots ),\\ (\frac {1}{2}I-B^{\ast}-E^{\ast}_{1})x=(\frac {1}{2}x_{1}-(\frac{\varepsilon}{2}+1)x_{2}, \frac {1}{2}x_{2}-x_{3}, \frac {1}{2}x_{3}-x_{4}, \cdots ). \end{gathered}$

令$(\frac {1}{2}I-B^{\ast}-E^{\ast}_{1})x=0,$我们有

$\begin{equation*} x=(c, \frac {1}{(\varepsilon+2)}c, \frac {1}{2(\varepsilon+2)}c, \frac {1}{4(\varepsilon+2)}c, \cdots ), \end{equation*}$

其中$c$为非零常数. 经验证$x\in X$, 因此$\mathcal{N}(\frac {1}{2}I-B^{\ast}-E^{\ast}_{1})\neq \{0\},$即$\overline{\mathcal{R}(\frac {1}{2}I-B-E_{1})}\neq X.$故$\frac {1}{2}\in\sigma_{r}(B+E_{1}),$亦即$\frac {1}{2}\in\sigma_{\varepsilon, r}(B).$

对任意的$x\in X$, 取另一个有界扰动算子$E_{2}$, 使得$E_{2}x=(\frac {1}{2}x_{1}, -x_{1}, 0, \cdots )$. 经计算

$\begin{gathered} (B+E_{2})x=((\frac {1}{2}x_{1}, 0, x_{2}, x_{3} \cdots ), \\ (\frac {1}{2}I-B-E_{2})x=(0, \frac {1}{2}x_{2}, \frac {1}{2}x_{3}-x_{2}, \frac {1}{2}x_{4}-x_{3}, \cdots ), \end{gathered}$
$\begin{equation*} \|E_{2}\|=\sup\limits_{x\in X\setminus \{0\}}\frac{\|E_{2}x\|}{\|x\|}=\sup\limits_{x\in X\setminus \{0\}}\frac{\sqrt{\frac{1}{4}x^{2}_{1}+x^{2}_{1}}}{\|x\|}\leq\frac{\sqrt{5}}{2}. \end{equation*}$

这意味着$\frac{1}{2}\in\sigma_{p}(B+E_{2}),$故当$\varepsilon>\frac{\sqrt{5}}{2}$时,$\frac{1}{2}\in\sigma_{\varepsilon, p}(B),$进而$\frac{1}{2}\in\sigma_{\varepsilon, p}(B)\cap\sigma_{\varepsilon, r}(B)$.

对于任意的$x=(x_{1}, x_{2}, \cdots )\in X$, 取扰动算子$E_{3}$为

$E_{3}x=(0, 0, \frac{\varepsilon}{2}x_{2}, 0, \cdots ).$

$E=\left( \begin{array} {cc}{E_{3}} & {0}\\ {0}& {E_{1}} \end{array} \right), $

可以验证$\frac{1}{2}I-M_{0}-E$是单射, 且$\overline{\mathcal{R}(\frac{1}{2}I-M_{0}-E)}\neq X\oplus X$. 由此可得$\frac{1}{2}\in\sigma_{\varepsilon, r}^{D}(M_{0}).$

综上所述,$\frac{1}{2}\not\in(\sigma_{\varepsilon, r}(A)\setminus\sigma_{\varepsilon, p}(B))\cup(\sigma_{\varepsilon, r}(B)\setminus\sigma_{\varepsilon, p}(A))$且$\frac{1}{2}\in\sigma_{\varepsilon, r}^{D}(M_{0}).$

通过对拟剩余谱定义的进一步分析, 我们得到了如下结论.

定理 3.2 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} \sigma_{\varepsilon, r}^{D}(M_{0})\setminus\sigma_{p}(M_{0})=(\sigma_{\varepsilon, r}(A)\cup\sigma_{\varepsilon, r}(B))\backslash(\sigma_{p}(A)\cup\sigma_{p}(B)). \end{equation*}$

设$\lambda\in\sigma_{\varepsilon, r}^{D}(M_{0})\setminus\sigma_{p}(M_{0}),$根据定义 2.2 可知, 存在算子$E=\left( \begin{array} {cc}{E_{11}} & {0}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X)$, 使得

$\mathcal{N}(M_{0}+E-\lambda I)=\{0\} \text{且} \overline{\mathcal{R}(M_{0}+E-\lambda I)}\neq X\oplus X.$

从而, 有

$\mathcal{N}(A+E_{11}-\lambda I)=\{0\} \text{且} \mathcal{N}(B+E_{22}-\lambda I)=\{0\}.$

由$\overline{\mathcal{R}(M_{0}+E-\lambda I)}\neq X\oplus X$知,$\overline{\mathcal{R}(A+E_{11}-\lambda I)}\neq X$或$\overline{\mathcal{R}(B+E_{22}-\lambda I)}\neq X.$因为$\lambda\not\in\sigma_{p}(M_{0})$, 故$\mathcal{N}(A-\lambda I)=\{0\}$且$\mathcal{N}(B-\lambda I)=\{0\}.$所以,当$\overline{\mathcal{R}(A+E_{11}-\lambda I)}\neq X$时, 有

$\begin{equation*} \lambda\in\sigma_{\varepsilon, r}(A)\backslash(\sigma_{p}(A)\cup\sigma_{p}(B)); \end{equation*}$

当$\overline{\mathcal{R}(B+E_{22}-\lambda I)}\neq X$时, 我们得到

$\begin{equation*} \lambda\in\sigma_{\varepsilon, r}(B)\backslash(\sigma_{p}(A)\cup\sigma_{p}(B)). \end{equation*}$

因此,$\lambda\in(\sigma_{\varepsilon, r}(A)\cup\sigma_{\varepsilon, r}(B))\backslash(\sigma_{p}(A)\cup\sigma_{p}(B)).$

下面我们证明当$\lambda\in(\sigma_{\varepsilon, r}(A)\cup\sigma_{\varepsilon, r}(B))\backslash(\sigma_{p}(A)\cup\sigma_{p}(B))$时,$\lambda\in\sigma_{\varepsilon, r}^{D}(M_{0})\setminus\sigma_{p}(M_{0}).$

事实上, 当$\lambda\in\sigma_{\varepsilon, r}(A)\backslash(\sigma_{p}(B)\cup\sigma_{p}(A))$时, 存在$E_{1}\in\mathcal{B}(X), \|E_{1}\|<\varepsilon,$使得

$\mathcal{N}(A+E_{1}-\lambda I)=\{0\}, \overline{\mathcal{R}(A+E_{1}-\lambda I)}\neq X$

$\mathcal{N}(A-\lambda I)=\{0\}, \mathcal{N}(B-\lambda I)=\{0\}.$

令$E=\left( \begin{array} {cc}{E_{1}} & {0}\\ {0}& {0} \end{array} \right)\in\mathcal{B}(X\oplus X)$, 可得

$M_{0}+E-\lambda I=\left( \begin{array} {cc}{A+E_{1}-\lambda I} & {0}\\ {0}& {B-\lambda I} \end{array} \right)$

是单射且$\overline{\mathcal{R}(M_{0}+E-\lambda I)}\neq X\oplus X.$因此$\lambda\in\sigma_{\varepsilon, r}^{D}(M_{0})\backslash\sigma_{p}(M_{0}).$

同理, 当$\lambda\in\sigma_{\varepsilon, r}(B)\backslash(\sigma_{p}(B)\cup\sigma_{p}(A))$时,$\lambda\in\sigma_{\varepsilon, r}^{D}(M_{0})\setminus\sigma_{p}(M_{0}).$

综上所述,

$\begin{equation*} \sigma_{\varepsilon, r}^{D}(M_{0})\setminus\sigma_{p}(M_{0})=(\sigma_{\varepsilon, r}(A)\cup\sigma_{\varepsilon, r}(B))\backslash(\sigma_{p}(A)\cup\sigma_{p}(B)). \end{equation*}$

证毕.

同样, 对于分块对角算子矩阵而言, 我们也不能将连续谱的性质直接推广到拟连续谱上. 为了说明这个问题, 先陈述一个结论.

命题 3.3[20] 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} \sigma_{c}(M_{0})=(\sigma_{c}(A)\cap\sigma_{c}(B)) \cup(\sigma_{c}(A)\cap\rho(B))\cup(\sigma_{c}(B)\cap\rho(A)). \end{equation*}$

对于分块对角算子矩阵的拟连续谱, 我们可得到如下类似于连续谱的包含关系.

命题 3.4 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} (\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B)) \cup(\sigma_{\varepsilon, c}(A)\cap\rho_{\varepsilon}(B))\cup(\sigma_{\varepsilon, c}(B)\cap\rho_{\varepsilon}(A))\subseteq\sigma_{\varepsilon, c}^{D}(M_{0})\subseteq(\sigma_{\varepsilon, c}(A)\cup\sigma_{\varepsilon, c}(B)). \end{equation*}$

如果$\lambda\in\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B)$, 那么存在$E_{1}, E_{2}\in\mathcal{B}(X), \|E_{1}\|<\varepsilon, \|E_{2}\|<\varepsilon,$使得$A+E_{1}-\lambda I$和$B+E_{2}-\lambda I$是单射,$\overline{\mathcal{R}(A+E_{1}-\lambda I)}=X, \overline{\mathcal{R}(B+E_{2}-\lambda I)}=X$且$\mathcal{R}(A+E_{1}-\lambda I)\neq X, \mathcal{R}(B+E_{2}-\lambda I)\neq X.$令

$\begin{equation*} E=\left( \begin{array} {cc}{E_{1}} & {0}\\ {0}& {E_{2}} \end{array} \right)\in\mathcal{B}(X\oplus X), \|E\|=\max_{i=1, 2}\{\|E_{i}\|\}<\varepsilon, \end{equation*}$

可得

$M_{0}+E-\lambda I \text{是单射} \overline{\mathcal{R}(M_{0}+E-\lambda I)}=X\oplus X, \text{且} \mathcal{R}(M_{0}+E-\lambda I)\neq X\oplus X,$

即$\lambda\in\sigma_{\varepsilon, c}^{D}(M_{0})$.

下面证明$(\sigma_{\varepsilon, c}(A)\cap\rho_{\varepsilon}(B))\subseteq\sigma_{\varepsilon, c}^{D}(M_{0})$.

设$\lambda\in\sigma_{\varepsilon, c}(A)\cap\rho_{\varepsilon}(B)$, 因为$\lambda\in\sigma_{\varepsilon, c}(A)$, 即存在$E_{3}\in\mathcal{B}(X), \|E_{3}\|<\varepsilon,$使得

$A+E_{3}-\lambda I \text{是单射}, \overline{\mathcal{R}(A+E_{3}-\lambda I)}=X \text{且} \mathcal{R}(A+E_{3}-\lambda I)\neq X.$

又由于$\lambda\in\rho_{\varepsilon}(B)$, 故对于任意的$E_{4}\in\mathcal{B}(X), \|E_{4}\|<\varepsilon,$都有

$B+E_{4}-\lambda I \text{是单射} \text{且} \mathcal{R}(B+E_{4}-\lambda I)=X.$

$\begin{equation*} E_{0}=\left( \begin{array} {cc}{E_{3}} & {0}\\ {0}& {E_{4}} \end{array} \right)\in\mathcal{B}(X\oplus X), \end{equation*}$

可得$\|E\|=\max\limits_{i=3, 4}\{\|E_{i}\|\}<\varepsilon,$并且

$M_{0}+E_{0}-\lambda I \text{是单射}, \overline{\mathcal{R}(M_{0}+E_{0}-\lambda I)}=X\oplus X, \text{且} \mathcal{R}(M_{0}+E_{0}-\lambda I)\neq X\oplus X,$

这表明$\lambda\in\sigma_{\varepsilon, c}^{D}(M_{0}).$

同理可证$(\sigma_{\varepsilon, c}(B)\cap\rho_{\varepsilon}(A))\subseteq\sigma_{\varepsilon, c}^{D}(M_{0}).$

通过以上的分析,$\sigma_{\varepsilon, c}^{D}(M_{0})\subseteq(\sigma_{\varepsilon, c}(A)\cup\sigma_{\varepsilon, c}(B))$是显然的. 证毕.

经过对拟连续谱定义的进一步剖析, 我们得到如下结论.

定理 3.3 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{equation*} \sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B))=((\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B))\cap(\rho(A)\cap\rho(B))) \cup(\sigma_{\varepsilon, c}(A)\cap\rho(B))\cup(\sigma_{\varepsilon, c}(B)\cap\rho(A)). \end{equation*}$

如果$\lambda\in\sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B))$, 那么存在算子$E=\left( \begin{array} {cc}{E_{11}} & {0}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X)$, 使得

$\mathcal{N}(M_{0}+E-\lambda I)=0, \overline{\mathcal{R}(M_{0}+E-\lambda I)}=X\oplus X \text{且} \mathcal{R}(M_{0}+E-\lambda I)\neq X\oplus X.$

于是$A+E_{11}-\lambda I$和$B+E_{22}-\lambda I$是单射,$\overline{\mathcal{R}(A+E_{11}-\lambda I)}=X$且$\overline{\mathcal{R}(B+E_{22}-\lambda I)}=X$,$\mathcal{R}(A+E_{11}-\lambda I)\neq X$或$\mathcal{R}(B+E_{22}-\lambda I)\neq X.$以下三种情形分别进行讨论.

(i) 当$\mathcal{R}(A+E_{11}-\lambda I)\neq X$且$\mathcal{R}(B+E_{22}-\lambda I)\neq X$时, 由于$\lambda\in\rho(A)\cap\rho(B),$有

$\begin{equation*} \lambda\in(\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B))\cap(\rho(A)\cap\rho(B)). \end{equation*}$

(ii) 当$\mathcal{R}(A+E_{11}-\lambda I)\neq X, \mathcal{R}(B+E_{22}-\lambda I)=X$时, 由于$\lambda\in\rho(A)\cap\rho(B),$我们有

$\begin{equation*} \lambda\in\sigma_{\varepsilon, c}(A)\cap\rho(B). \end{equation*}$

(iii) 当$\mathcal{R}(A+E_{11}-\lambda I)=X, \mathcal{R}(B+E_{22}-\lambda I)\neq X$时, 由于$\lambda\in\rho(A)\cap\rho(B),$我们有

$\begin{equation*} \lambda\in\sigma_{\varepsilon, c}(B)\cap\rho(A). \end{equation*}$

结合情形 (i)-(iii), 可以断言

$\begin{equation*} \sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B))\subseteq((\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B))\cap(\rho(A)\cap\rho(B))) \cup(\sigma_{\varepsilon, c}(A)\cap\rho(B))\cup(\sigma_{\varepsilon, c}(B)\cap\rho(A)). \end{equation*}$

下证反包含关系成立, 即证明

$((\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B))\cap(\rho(A)\cap\rho(B))) \cup(\sigma_{\varepsilon, c}(A)\cap\rho(B))\cup(\sigma_{\varepsilon, c}(B)\cap\rho(A))\subseteq\sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B)).$

事实上, 若$\lambda\in(\sigma_{\varepsilon, c}(A)\cap\sigma_{\varepsilon, c}(B))\cap(\rho(A)\cap\rho(B))$, 那么存在$E_{1}, E_{2}\in\mathcal{B}(X),$$\|E_{1}\|<\varepsilon,$$\|E_{2}\|<\varepsilon,$使得$A+E_{1}-\lambda I$和$B+E_{2}-\lambda I$是单射,$\overline{\mathcal{R}(A+E_{1}-\lambda I)}=X, \overline{\mathcal{R}(B+E_{2}-\lambda I)}=X,$且$\mathcal{R}(A+E_{1}-\lambda I)\neq X, \mathcal{R}(B+E_{2}-\lambda I)\neq X.$令

$\begin{equation*} E_{0}=\left( \begin{array} {cc}{E_{1}} & {0}\\ {0}& {E_{2}} \end{array} \right)\in\mathcal{B}(X\oplus X), \end{equation*}$

则$\|E\|=\max\limits_{i=1, 2}\{\|E_{i}\|\}<\varepsilon,$$M_{0}+E_{0}-\lambda I$是单射, 且

$ \overline{\mathcal{R}(M_{0}+E_{0}-\lambda I)}=X\oplus X, \mathcal{R}(M_{0}+E_{0}-\lambda I)\neq X\oplus X.$

进而, 得

$\begin{equation*} \lambda\in\sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B)). \end{equation*}$

如果$\lambda\in\sigma_{\varepsilon, c}(A)\cap\rho(B),$那么存在$E_{3}\in\mathcal{B}(X), \|E_{3}\|<\varepsilon,$使得

$\begin{equation*} \mathcal{N}(A+E_{3}-\lambda I)=\{0\}, \overline{\mathcal{R}(A+E_{3}-\lambda I)}=X, \mathcal{R}(A+E_{3}-\lambda I)\neq X. \end{equation*}$

而$\lambda\in\rho(B)$意味着$\mathcal{N}(B-\lambda I)=\{0\}$且$\overline{\mathcal{R}(B-\lambda I)}=X.$令$\hat{E}=\left( \begin{array} {cc}{E_{3}} & {0}\\ {0}& {0} \end{array} \right)\in\mathcal{B}(X\oplus X), \|\hat{E}\|=\|E_{3}\|<\varepsilon,$故

$M_{0}+\hat{E}-\lambda I \text{是单射}, \overline{\mathcal{R}(M_ {0}+\hat{E}-\lambda I)}=X\oplus X \text{且} \mathcal{R}(M_{0}+\hat{E}-\lambda I)\neq X\oplus X.$

因此, 有$\lambda\in\sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B)).$

同理可得$\sigma_{\varepsilon, c}(B)\cap\rho(A)\subseteq\sigma_{\varepsilon, c}^{D}(M_{0})\cap(\rho(A)\cap\rho(B)).$证毕.

接下来考虑扰动算子为上三角情形, 即

$E=\left( \begin{array} {cc}{E_{11}} & {E_{12}}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X),$

其中$E_{ij}\in\mathcal{B}(X), \|E_{ij}\|<\varepsilon, i=1, j=1, 2.$下三角扰动情形类似, 本文不再讨论. 在下文中出现的常数$\gamma$满足$0<\gamma<\varepsilon,$并且$\gamma J$表示有界线性算子$J$的数乘. 为了表述的方便, 记

$\begin{eqnarray*} \Delta_{0}&=&\{\lambda\in\mathbb{C}:\alpha_{\varepsilon}(B-\lambda I)>\beta_{\varepsilon}(A-\lambda I)\};\\ \Delta_{1}&=&\{\lambda\in\mathbb{C}:1\leq\alpha_{\varepsilon}(B-\lambda I)=\beta_{\varepsilon}(A-\lambda I)\};\\ \Delta_{2}&=&\{\lambda\in\mathbb{C}:0<\alpha_{\varepsilon}(B-\lambda I)<\beta_{\varepsilon}(A-\lambda I)\}. \end{eqnarray*}$

首先我们陈述下列重要结论.

定理 3.4 设$T\in\mathcal{C}(X)$, 则$\mathcal{N}_{\varepsilon}(T)$是闭的.

先证明$\mathcal{N}_{\varepsilon}(T)=(\bigcap\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast}))^{\perp}.$

任取$x\in\mathcal{N}_{\varepsilon}(T)$, 则存在$E\in\mathcal{B}(X), \|E\|<\varepsilon,$有$(T+E)x=0.$从而对于任意的$y\in \mathcal{D}(T^{\ast}),$有

$\begin{equation*} (x, (T^{\ast}+E^{\ast})y)=((T+E)x, y)=0. \end{equation*}$

这说明$x\in\mathcal{R}(T^{\ast}+E^{\ast})^{\perp},$即$x\in\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast})^{\perp},$反之也成立. 于是, 我们首先得到

$\begin{equation*} \mathcal{N}_{\varepsilon}(T)=\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast})^{\perp}. \end{equation*}$

若$x_{0}\in\bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\|E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast})^{\perp}$,则存在$E_{1}\in\mathcal{B}(X), \| E_{1}\|<\varepsilon,$使得$x_{0}\in\mathcal{R}(T^{\ast}+E_{1}^{\ast})^{\perp},$从而$x_{0}\bot\mathcal{R}(T^{\ast}+E_{1}^{\ast})$, 进而, 得$x_{0}\bot\bigcap\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast}).$因此,$x_{0}\in(\bigcap\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast}))^{\perp},$反之亦然.故

$\begin{equation} \bigcup\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast})^{\perp}=(\bigcap\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast}))^{\perp}. \end{equation}$

综上所述,$\mathcal{N}_{\varepsilon}(T)=(\bigcap\limits_{\substack{E\in\mathcal{B}(X)\\\| E\|<\varepsilon}}\mathcal{R}(T^{\ast}+E^{\ast}))^{\perp}$, 进而$\mathcal{N}_{\varepsilon}(T)$是闭的.证毕.

结合空间分解技巧, 我们可得到如下结论.

定理 3.5 设$A, B\in\mathcal{C}(X)$,$M_{0}=\left( \begin{array} {cc}{A} & {0}\\ {0}& {B} \end{array} \right)$是$2\times2$分块对角算子矩阵, 则

$\begin{eqnarray*} \sigma_{\varepsilon, p}^{T}(M_{0})=\sigma_{\varepsilon, p}(A)\cup\Delta_{0}\cup\Delta_{1}\cup\Delta_{2}.\end{eqnarray*}$

先证明$\sigma_{\varepsilon, p}(A)\cup\Delta_{0}\cup\Delta_{1}\cup\Delta_{2}\subseteq\sigma_{\varepsilon, p}^{T}(M_{0}).$类似于定理 3.1 的证明, 可得$\sigma_{\varepsilon, p}(A)\subseteq\sigma_{\varepsilon, p}^{T}(M_{0})$.下证$\Delta_{0}\setminus\sigma_{\varepsilon, p}(A)\subseteq\sigma_{\varepsilon, p}^{T}(M_{0}).$

设$\lambda\in\Delta_{0}\setminus\sigma_{\varepsilon, p}(A),$由于注意到$\beta_{\varepsilon}(A-\lambda I)<\infty$, 其隐含着$\mathcal{R}_{\varepsilon}(A-\lambda I)$是闭的. 现将空间$X\oplus X$分解为:$X\oplus X=X\oplus\mathcal{N}_{\varepsilon}(B-\lambda I)\oplus\mathcal{N}_{\varepsilon}(B-\lambda I)^{\perp}=\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}\oplus\mathcal{R}_{\varepsilon}(A-\lambda I)\oplus X.$下面分两种情形进行讨论.

(i) 如果存在$E_{12}, E_{22}\in\mathcal{B}(X), \|E_{12}\|<\varepsilon, \|E_{22}\|<\varepsilon,$使得$E_{12}, E_{22}$中的每一个扰动算子, 都有$\mathcal{N}(E_{12})\cap\mathcal{N}(B+E_{22}-\lambda I)\neq\{0\},$那么必存在非零元素$x\in \mathcal{D}(B),$满足$E_{12}x=0$且$(B+E_{22}-\lambda I)x=0$. 故取$0\neq y_{0}=\left( \begin{array} {c}{0} \\ {x} \end{array} \right)\in \mathcal{D}(A)\oplus \mathcal{D}(B),$对于任意的$E_{11}\in\mathcal{B}(X), \|E_{11}\|<\varepsilon,$取

$\begin{equation*} E=\left( \begin{array} {cc}{E_{11}} & {E_{12}}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X), \end{equation*}$

故$\|E\|=\max\limits_{i, j}\{\|E_{ij}\|\}<\varepsilon$, 并且

$\begin{eqnarray*} (M_{0}+E-\lambda I)y_{0}=\left( \begin{array} {cc}{A+E_{11}-\lambda I} & {E_{12}}\\ {0}& {B+E_{22}-\lambda I} \end{array} \right)\left( \begin{array} {c}{0} \\ {x} \end{array} \right)=0. \end{eqnarray*}$

因此, 我们有$\lambda\in\sigma_{\varepsilon, p}^{T}(M_{0}).$

(ii) 如果对任意两个扰动算子$E_{12}, E_{22}$,都有$\mathcal{N}(E_{12})\cap\mathcal{N}(B+E_{22}-\lambda I)=\{0\},$那么, 很容易得出$E_{12}\mid_{\mathcal{N}_{\varepsilon}(B-\lambda I)}$是单射. 这样,

$\begin{equation*} \dim E_{12}\mathcal{N}_{\varepsilon}(B-\lambda I)=\alpha_{\varepsilon}(B-\lambda I)>\beta_{\varepsilon}(A-\lambda I). \end{equation*}$

定义可逆算子

$\begin{equation*} J_{1}:Z_{1}\rightarrow\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}, \|J_{1}\|=1 \end{equation*}$

与下方有界算子

$\begin{equation*} J_{2}:Z_{2}\rightarrow\mathcal{R}_{\varepsilon}(A-\lambda I), \|J_{2}\|=1, \end{equation*}$

使得$Z_{1}\oplus Z_{2}=\mathcal{N}_{\varepsilon}(B-\lambda I), \dim Z_{2}=\alpha_{\varepsilon}(B-\lambda I)-\beta_{\varepsilon}(A-\lambda I)$.取定

$\begin{equation*} E_{12}=\left( \begin{array} {ccc} {0}& {\gamma J_{2}} &{0}\\ {\gamma J_{1}} & {0}&{0} \end{array} \right):\left(\begin{array} {c}{Z_{1}} \\ {Z_{2}}\\ {\mathcal{N}_{\varepsilon}(B-\lambda I)^{\perp}} \end{array} \right)\rightarrow\left(\begin{array} {c}{\mathcal{R}_{\varepsilon}(A-\lambda I)} \\ {\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}} \end{array} \right), \end{equation*}$

显然,$\|E_{12}\|=\max\{\|\gamma J_{1}\|, \|\gamma J_{2}\|\}=\gamma<\varepsilon$. 根据$\mathcal{R}_{\varepsilon}(A-\lambda I)$的闭性, 我们很容易得到, 存在$E_{11}, E_{22}\in\mathcal{B}(X), \|E_{ii}\|<\varepsilon, i=1, 2$, 满足

$\begin{equation*} E_{12}\mathcal{N}(B+E_{22}-\lambda I) \bigcap\mathcal{R}(A+E_{11}-\lambda I)\neq\{0\}. \end{equation*}$

$\begin{equation*}0\neq y\in E_{12}\mathcal{N}(B+E_{22}-\lambda I) \bigcap\mathcal{R}(A+E_{11}-\lambda I)\subset X, \end{equation*}$

故存在$x_{1}\in \mathcal{D}(A), x_{2}\in Z_{2}\subseteq \mathcal{D}(B)$使得

$\begin{equation*} y=(A_{1}+(E_{11})_{1}-\lambda I)x_{1}=-\gamma J_{2}x_{2}. \end{equation*}$

在这里,$A_{1}+(E_{11})_{1}-\lambda I=(A+E_{11}-\lambda I)\mid_{\mathcal{N}_{\varepsilon}(A-\lambda I)^{\perp}}$. 由于$y\neq 0$, 所以$x_{1}\neq0, x_{2}\neq0.$这样

$\begin{eqnarray*} (M_{0}+E-\lambda I)\left(\begin{array} {c}{x_{1}} \\ {0}\\ {x_{2}}\\ {0} \end{array} \right)&=&\left( \begin{array} {cccc} {A_{1}+(E_{11})_{1}-\lambda I}&{0}& {\gamma }J_{2}& {0}\\ {0}& {\gamma }J_{1}& {0}& {0}\\ {0} & {0}& {0}& {B_{1}+(E_{22})_{1}-\lambda I} \end{array}\right)\left( \begin{array} {c}x_{1}\\ 0\\ x_{2}\\ 0 \end{array}\right)\\&=&(A_{1}+(E_{11})_{1}-\lambda I)x_{1}+\delta J_{2}x_{2}\\&=&0. \end{eqnarray*}$

在这里,$B_{1}+(E_{22})_{1}-\lambda I=(B+E_{22}-\lambda I)\mid_{\mathcal{N}_{\varepsilon}(B-\lambda I)^{\perp}}.$因此, 我们有$\lambda\in\sigma_{\varepsilon, p}^{T}(M_{0}).$

下证$\Delta_{1}\cup\Delta_{2}\subseteq\sigma_{\varepsilon, p}^{T}(M_{0}).$现将空间$X\oplus X$分解为

$X\oplus X=X\oplus\mathcal{N}_{\varepsilon}(B-\lambda I)\oplus\mathcal{N}_{\varepsilon}(B-\lambda I)^{\perp}=\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}\oplus\overline{\mathcal{R}_{\varepsilon}(A-\lambda I)}\oplus X.$

考虑以下两种情形.

(a)$\lambda\in\{\lambda\in\mathbb{C}:1=\alpha_{\varepsilon}(B-\lambda I)=\beta_{\varepsilon}(A-\lambda I)\}$.

因为$1=\beta_{\varepsilon}(A-\lambda I)<\infty,$所以$\mathcal{R}_{\varepsilon}(A-\lambda I)$闭的. 现定义一下方有界算子

$\begin{equation*} J_{3}:\mathcal{N}_{\varepsilon}(B-\lambda I)\rightarrow\mathcal{R}_{\varepsilon}(A-\lambda I), \|J_{3}\|=1. \end{equation*}$

$\begin{equation*} \widehat{E_{12}}=\left( \begin{array} {cc}{0} & {0} \\ {\gamma J_{3}}& {0} \end{array} \right):\left(\begin{array} {c}{\mathcal{N}_{\varepsilon}(B-\lambda I)} \\ {\mathcal{N}_{\varepsilon}(B-\lambda I)^{\perp}}\end{array} \right)\rightarrow\left(\begin{array} {c}{\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}} \\ {\mathcal{R}_{\varepsilon}(A-\lambda I)} \end{array} \right). \end{equation*}$

显然, 存在$\widehat{E_{11}}, \widehat{E_{22}}\in\mathcal{B}(X), \|\widehat{E_{ii}}\|<\varepsilon, i=1, 2$, 满足

$\begin{equation*} \widehat{E_{12}}\mathcal{N}(B+\widehat{E_{22}}-\lambda I) \bigcap\mathcal{R}(A+\widehat{E_{11}}-\lambda I)\neq\{0\}. \end{equation*}$

这样, 如同$\lambda\in\Delta_{0}$的情形(ii), 容易得到$\lambda\in\sigma_{\varepsilon, p}^{T}(M_{0}),$这里从略.

(b)$\lambda\in(\Delta_{1}\cup\Delta_{2})\setminus\{\lambda\in\mathbb{C}:1=\alpha_{\varepsilon}(B-\lambda I)=\beta_{\varepsilon}(A-\lambda I)\}$.

定义一非下方有界算子

$\begin{equation*} J_{4}:\mathcal{N}_{\varepsilon}(B-\lambda I)\rightarrow\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}, \|J_{4}\|=1. \end{equation*}$

$\begin{equation*} E' _{12}=\left( \begin{array} {cc}{\gamma J_{4}} & {0} \\ {0}& {0} \end{array} \right):\left(\begin{array} {c}{\mathcal{N}_{\varepsilon}(B-\lambda I)} \\ {\mathcal{N}_{\varepsilon}(B-\lambda I)^{\perp}}\end{array} \right)\rightarrow\left(\begin{array} {c}{\mathcal{R}_{\varepsilon}(A-\lambda I)^{\perp}} \\ {\overline{\mathcal{R}_{\varepsilon}(A-\lambda I)}} \end{array} \right), \end{equation*}$

使得$E' _{12}\mathcal{N}_{\varepsilon}(B-\lambda I)=\{0\}$. 故对于任意的$0\neq y_{1}\in\mathcal{N}_{\varepsilon}(B-\lambda I),$必存在$E' _{22}\in\mathcal{B}(X), \|E' _{22}\|<\varepsilon,$使得$(B+E' _{22}-\lambda I)y_{1}=0.$进而,$E' _{12}y_{1}=0$. 对于任意的$E' _{11}\in\mathcal{B}(X), \|E' _{11}\|<\varepsilon,$令

$E' =\left( \begin{array} {cc}{E' _{11}} & {E' _{12}}\\ {0}& {E' _{22}} \end{array} \right)\in\mathcal{B}(X\oplus X), \text{且} y=\left(\begin{array} {c}{0} \\ {y_{1}} \end{array} \right)\neq0,$

我们有$\|E' \|=\max\limits_{i, j}\{\|E' _{ij}\|\}<\varepsilon,$

$\begin{equation*} (M_{0}+E' -\lambda I)y= \left( \begin{array} {cc}{A+E' _{11}-\lambda I} & {E' _{12}}\\ {0}& {B+E' _{22}-\lambda I} \end{array} \right)\left(\begin{array} {c}{0} \\ {y_{1}} \end{array} \right)=0. \end{equation*}$

因此,$\lambda\in\sigma_{\varepsilon, p}^{T}(M_{0}).$

假设$\lambda\not\in\sigma_{\varepsilon, p}(A)\cup\Delta_{0}\cup\Delta_{1}\cup\Delta_{2},$以下分两种情形来讨论.

i)$\lambda\not\in\sigma_{\varepsilon, p}(A)$和$\lambda\in\{\lambda\in\mathbb{C}:\alpha_{\varepsilon}(B-\lambda I)=\beta_{\varepsilon}(A-\lambda I)=0\}$.

对于任意的

$\begin{equation*} E=\left( \begin{array} {cc}{E_{11}} & {E_{12}}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X), \|E\|=\max\limits_{i, j}\{\|E_{ij}\|\}<\varepsilon, \end{equation*}$

假设$\left(\begin{array} {c}{y_{2}} \\ {y_{3}} \end{array} \right)\in \mathcal{D}(A)\oplus \mathcal{D}(B)$, 满足

$\begin{equation} (M_{0}+E-\lambda I)\left(\begin{array} {c}{y_{2}} \\ {y_{3}} \end{array} \right)=\left( \begin{array} {cc}{A+E_{11}-\lambda I} & {E_{12}}\\ {0}& {B+E_{22}-\lambda I} \end{array} \right)\left(\begin{array} {c}{y_{2}} \\ {y_{3}} \end{array} \right)=0. \end{equation}$

由于$\alpha_{\varepsilon}(B-\lambda I)=\beta_{\varepsilon}(A-\lambda I)=0$, 所以$y_{3}=0.$又由于$\lambda\not\in\sigma_{\varepsilon, p}(A)$, 因而对于任意的扰动算子$E_{11}\in\mathcal{B}(X), \|E_{11}\|<\varepsilon, A+E_{11}-\lambda I$是单射, 故$y_{2}=0.$这表明$M_{0}+E-\lambda I$是单射, 再结合$E$的任意性, 可得$\lambda\not\in\sigma_{\varepsilon, p}^{T}(M_{0}).$

ii)$\lambda\not\in\sigma_{\varepsilon, p}(A)$和$\lambda\in\{\lambda\in\mathbb{C}: \alpha_{\varepsilon}(B-\lambda I)=0$且$0<\beta_{\varepsilon}(A-\lambda I)\}$.

$\alpha_{\varepsilon}(B-\lambda I)=0$意味着$\mathcal{N}_{\varepsilon}(B-\lambda I)=\{0\}$. 即对于任意的$E_{22}\in\mathcal{B}(X),$$\|E_{22}\|<\varepsilon,$$B+E_{22}-\lambda I$是单射. 又因为$\lambda\not\in\sigma_{\varepsilon, p}(A)$, 所以对于任意的$E_{11}\in\mathcal{B}(X), \|E_{11}\|<\varepsilon, A+E_{11}-\lambda I$是单射.这样, 对于任意的

$E=\left( \begin{array} {cc}{E_{11}} & {E_{12}}\\ {0}& {E_{22}} \end{array} \right)\in\mathcal{B}(X\oplus X), \|E_{12}\|<\varepsilon,$

都有$M_{0}+E-\lambda I$是单射, 因此$\lambda\not\in\sigma_{\varepsilon, p}^{T}(M_{0}).$证毕.

注 3.1 设$\varepsilon>0$,

$E=\left( \begin{array} {cc}{E_{11}} & {E_{12}}\\ {0}& {E_{22}} \end{array} \right), \|E_{ij}\|<\varepsilon, i=1, j=1, 2,$

为上三角扰动. 当$\lambda\in\sigma_{\varepsilon, p}^{T}(M_{0})$,$\lambda\notin\sigma_{\varepsilon, p}(A)$且$E_{12}=0$时, 可得$\alpha_ {\varepsilon}(B-\lambda I)>0$且$\Delta_{0}\cup\Delta_{1}\cup\Delta_{2}\subseteq \sigma_{\varepsilon, p}(B)$, 由定理 3.5 可导出$\lambda\in\sigma_{\varepsilon, p}(B)$. 另一方面, 当$\lambda\in\sigma_{\varepsilon, p}(B)\setminus\sigma_{\varepsilon, p}(A)$且$E_{12}=0$, 我们有$\lambda\in\Delta_{0}\cup\Delta_{1}\cup\Delta_{2},$再结合由定理 3.5 得到$\lambda\in\sigma_{\varepsilon, p}^{D}(M_{0})$. 因此, 定理 3.5 是定理 3.1 的进一步推广.

参考文献

Trefethen L N, Embree M. Spectra and Pseudospectra:The Behavior of Nonnormal Matrices and Operators. Princeton: Princeton University Press, 2005

[本文引用: 3]

Davies E B. Linear Operators and Their Spectra. Cambridge: Cambridge University Press, 2007

[本文引用: 2]

Lui S H.

A pseudospectral mapping theorem

Math Comp, 2003, 244(72): 1841-1854

[本文引用: 1]

Lancaster P, Psarrakos P.

On the pseudospectra of matrix polynomials

SIAM J Matrix Anal Appl, 2005, 27(1): 115-129

DOI:10.1137/S0895479804441420      URL     [本文引用: 1]

Shargorodsky E.

Pseudospectra of semigroup generators

Bull Lond Math Soc, 2010, 42(6): 1031-1034

DOI:10.1112/blms/bdq062      URL     [本文引用: 1]

Cui J L, Forstall V, Li C K, et al.

Properties and preservers of the pseudospectrum

Linear Algebra Appl, 2012, 436(2): 316-325

DOI:10.1016/j.laa.2011.03.044      URL     [本文引用: 1]

Bögli S.

Local convergence of spectra and pseudospectra

J Spectr Theor, 2018, 8(3): 1051-1098

DOI:10.4171/jst      URL     [本文引用: 1]

Ammar A, Jeribi A.

The essential pseudo-spectra of a sequence of linear operators

Complex Anal Oper Theory, 2018, 12(3): 835-848

DOI:10.1007/s11785-018-0776-7      [本文引用: 1]

Ammar A, Jeribi A, Mahfoudhi K.

A characterization of the essential approximation pseudospectrum on a Banach space

Filomat, 2017, 31(11): 3599-3610

DOI:10.2298/FIL1711599A      URL     [本文引用: 1]

One impetus for writing this paper is the issue of approximation\n pseudospectrum introduced by M. P. H.Wolff in the journal of approximation\n theory (2001). The latter study motivates us to investigate the essential\n approximation pseudospectrum of closed, densely defined linear operators on a\n Banach space. We begin by defining it and then we focus on the\n characterization, the stability and some properties of these pseudospectra.

Ammar A, Jeribi A, Mahfoudhi K.

Some description of essential structured approximate and defect pseudospectrum

Korean J Math, 2020, 28(4): 673-697

[本文引用: 1]

Hagger R, Lindner M, Seidel M.

Essential pseudospectra and essential norms of band-dominated operators

J Math Anal Appl, 2016, 437(1): 255-291

DOI:10.1016/j.jmaa.2015.11.060      URL     [本文引用: 1]

Raouafi S.

Operators with minimal pseudospectra and connections to normality

Oper Matrices, 2020, 14(1): 91-103

[本文引用: 1]

Ferro R, Virtanen J A.

A note on structured pseudospectra of block matrices

J Comput Appl Math, 2017, 322: 18-24

DOI:10.1016/j.cam.2017.03.020      URL     [本文引用: 1]

Rump S M.

Eigenvalues, pseudospectrum and structured perturbations

Linear Algebra Appl, 2006, 413 (2-3): 567-593

DOI:10.1016/j.laa.2005.06.009      URL     [本文引用: 1]

Ammar A, Boukettaya B, Jeribi A.

A note on the essential pseudospectra and application

Linear Multilinear Algebra, 2016, 64(8): 1474-1483

DOI:10.1080/03081087.2015.1091436      URL     [本文引用: 2]

Roy N, Karow M, Bora S, et al.

Approximation of pseudospectra of block triangular matrices

Linear Algebra Appl, 2021, 623: 398-419

DOI:10.1016/j.laa.2020.08.008      URL     [本文引用: 1]

Kostić V R, Cvetković L, Šanca E.

From pseudospectra of diagonal blocks to pseudospectrum of a full matrix

J Comput Appl Math, 2021, 386: 113265

[本文引用: 2]

Pal A, Yakubovich D V.

Infinite-Dimentional features of matrices and pseudospectra

J Math Anal Appl, 2017, 447: 109-127

DOI:10.1016/j.jmaa.2016.09.064      URL     [本文引用: 1]

Frommer A, Jacob B, Vorberg L, et al.

Pseudospectrum enclosures by discretization

Integr Equ Oper Theory, 2021, 93(1): 1-31

DOI:10.1007/s00020-020-02616-2      [本文引用: 1]

The abstract issue of ‘Krylov solvability’ is extensively discussed for the inverse problem $$Af=g$$\n \n A\n f\n =\n g\n \n where A is a (possibly unbounded) linear operator on an infinite-dimensional Hilbert space, and g is a datum in the range of A. The question consists of whether the solution f can be approximated in the Hilbert norm by finite linear combinations of $$g,Ag,A^2g,\\dots $$\n \n g\n,\n A\n g\n,\n \n A\n 2\n \n g\n,\n ⋯\n \n, and whether solutions of this sort exist and are unique. After revisiting the known picture when A is bounded, we study the general case of a densely defined and closed A. Intrinsic operator-theoretic mechanisms are identified that guarantee or prevent Krylov solvability, with new features arising due to the unboundedness. Such mechanisms are checked in the self-adjoint case, where Krylov solvability is also proved by conjugate-gradient-based techniques.

吴德玉, 阿拉坦仓. 分块算子矩阵谱理论及其应用. 北京: 科学出版社, 2013

[本文引用: 3]

Wu D Y, Alatancang. Spectral Theory of the Block Operator Matrix and Application. Beijing: Science Press, 2013

[本文引用: 3]

Zhang S F, Wu Z Y, Zhong H J.

Continuous spectrum, point spectrum and residual spectrum of operator matrices

Linear Algebra Appl, 2010, 433(3): 653-661

DOI:10.1016/j.laa.2010.03.036      URL     [本文引用: 1]

Zhang H Y, Du H K.

Browder spectra of upper-triangular operator matrices

J Math Anal Appl, 2006, 323: 700-707

DOI:10.1016/j.jmaa.2005.10.073      URL     [本文引用: 1]

Li Y, Du H K.

The intersection of essential approximate point spectra of operator matrices

J Math Anal Appl, 2007, 323: 1171-1183

DOI:10.1016/j.jmaa.2005.11.032      URL     [本文引用: 1]

Yang L L, Cao X H.

Perturbations of spectra and property (R) for upper triangular operator matrices

J Math Anal Appl, 2023, 519(1): 126797

DOI:10.1016/j.jmaa.2022.126797      URL     [本文引用: 1]

Cvetković-Ilić Dragana S. Completion Problems on Operator Matrices. Rhode Island: American Mathematical Society, 2022

[本文引用: 1]

Shen R S, Hou G L, and Chen A.

On the pseudo-spectra and the related properties of infinite-dimensional Hamiltonian operators

Linear Multilinear Algebra, DOI:10.1080/03081087.2021.2006127

[本文引用: 5]

/