数学杂志  2018, Vol. 38 Issue (3): 511-519   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
杜娟
王华
两个有界线性算子和的Drazin逆
杜娟, 王华    
内蒙古工业大学理学院, 内蒙古 呼和浩特 010021
摘要:本文研究了两个有界线性算子和的Drazin逆的问题.利用算子的预解式展开的方法, 得到了$(P+Q)^D$的具体表达式, 并将其应用到四分块算子矩阵$M=\begin{bmatrix} A & B \\ C & D \end{bmatrix}$的Drazin逆上, 推广了文献[14, 15]的结果.
关键词Drazin逆    算子矩阵    
DRAZIN INVERSE FOR THE SUM OF TWO BOUNDED LINEAR OPERATORS
DU Juan, WANG Hua    
College of Science, Inner Mongolia University of Technology, Hohhot 010021, China
Abstract: In this paper, we investigate the Drazin inverse for the sum of two bounded linear operators. Using the resolvent expansion of the operator, the explicit representation of the Drazin inverse $(P+Q)^D$ is given. Then, we apply our results to the Drazin inverse of the operator matrix $M=\begin{bmatrix} A & B \\ C & D \end{bmatrix}$, which generalize the results in [14, 15].
Key words: Drazin inverse     operator matrix    
1 引言与预备知识

广义逆理论是非常重要的研究领域之一, 它在求解奇异微分和差分方程、算子方程、马尔可夫链、迭代法数值分析等方面都有着非常广泛的应用.特别在求解微分方程组时, 矩阵的广义逆发挥着重要作用, 例如给定一类一阶奇异系统方程

$ AX^{'}(t)+BX(t)=0 (t\geq0), $

其中$A$为奇异矩阵.其通解可表示为

$ X(t)=e^{-{A_1}^DB_1}{A_1}^D{A_1}X(0), $

其中$ A_1=(\lambda{A}+B)^{-1}A, \ \ B_1=(\lambda{A}+B)^{-1}B $[1].

$20$世纪以来, 学者们对广义逆理论中的矩阵的Drazin逆的研究最为活跃. $1958$年, Drazin [2]在半群与结合环上引入Drazin逆, 并在两个矩阵$P$$Q$满足$PQ=QP=0$的条件下, 证明出了$(P+Q)^D=P^D+Q^D$.对于算子情形, $2009$年, Castro-González [3]等在$P^2Q= PQ^2=0 $条件下讨论了$P+Q$的Drazin可逆性, 并给出了$(P+Q)^D$的表达式; 同年, 邓春源[4]$P, Q$均为幂等算子, 且满足三个不同条件$PQP=0, PQP=PQ, PQP=P$时给出了$(P+Q)^D$的表达式; $2011$年, Cvetković [5]等在$ PQP=0, Q^2P=0 $的条件下, 给出了$(P+Q)^D$的表达式; $2014$年, 黄俊杰[6]等在$P^2Q+PQ^2=0, P^3Q=PQ^3=0$的条件下给出了$(P+Q)^D$的表达式.对于$P, Q$为矩阵情形,学者魏益民[7]、Hartwig [8]、卜长江[9]、刘喜富[10]等获得了很多好的结果.

对于矩阵而言, 其Drazin逆一定存在, 但对于算子并非如此.那么算子的Drazin逆在什么条件下存在, 如果存在, 其Drazin逆的表达式又是什么样的形式? 这是需要讨论的问题.本文讨论了$P^5Q=0, P^2Q+PQ^2=0, PQPQ=0$条件下, 两个有界线性算子和的Drazin逆的存在性, 并给出了$(P+Q)^D$的表达式.最后, 将所得结果应用到四分块算子矩阵$M=\begin{bmatrix} A & B \\ C & D \end{bmatrix}$的Drazin逆上.

为便于叙述, 文中通篇采用如下的假设及符号.设$X, Y$是复Banach空间, 记$\mathcal{B}(X, Y)$是从$X$$Y$的所有有界线性算子的集合; $\mathcal{B}(X)$是从$X$$X$的所有有界线性算子的集合; 对于$A\in \mathcal{B}(X, Y)$, $\rho(A), \sigma(A)$, $r(A)$分别表示其预解集, 谱集, 谱半径; $R(\lambda, A)$表示算子$A$的预解式$(\lambda I-A)^{-1}$.

下面给出本文用到的定义和引理.

定义1.1  设$A\in \mathcal{B}(X)$, 若存在$A^D \in \mathcal{B}(X)$, 使得算子方程组

$ \begin{array}{l} AA^D=A^DA, A^DAA^D=A^D, A^{k+1}A^D=A^k \end{array} $

对某个非负整数$k$成立, 则称$A$是Drazin可逆的.对于上述方程组, 若有解, 则解必定唯一, 这个唯一的解$A^D$称为$A$的Drazin逆, 并称使得方程组成立的最小非负整数$k$$A$的指标, 记为${\rm ind}(A). $${\rm ind}(A)=0$时, $A$是可逆的, 即$A^D=A^{-1}. $

引理1.2 [11]  设$A \in \mathcal{B}(X, Y), B \in \mathcal{B}(Y, X)$, 如果$BA$为Drazin可逆, 那么$AB$也为Drazin可逆, 且

$ \begin{eqnarray*} (AB)^D=A((BA)^D)^2B, \ \ \ {\rm ind}(AB)\leq {\rm ind}(BA)+1. \end{eqnarray*} $

引理1.3 [12]  设$A\in\mathcal{B}(X)$, 则$A^D$存在当且仅当$0\in \mathcal\rho(A)$$0\in \mathcal\sigma(A)$为预解式$R(\lambda, A)$的一个极点, 此时有

$ \begin{array}{l} R(\lambda, A)=\sum\limits_{n=1}^{{\rm ind}(A)}\lambda^{-n}A^{n-1}A^\pi-\sum\limits_{n=0}^\infty\lambda^n(A^D)^{n+1}, \end{array} $ (1.1)

其中$0 < |\lambda| < (r(A^D))^{-1}$, $A^\pi=I-AA^D$, $I$是单位算子.

  由引理1.3可知, $A$的Drazin逆$A^D$就是$R(\lambda, A)$的Laurent展开式中$-\lambda^0$的系数, 即

$ \begin{eqnarray*} A^D=-\frac{1}{2\pi{i}}\oint_\gamma\lambda^{-1}R(\lambda, A)d\lambda, \end{eqnarray*} $

其中$\gamma=\{\lambda:|\lambda|=\varepsilon\}$$\varepsilon$须足够小以使$\{\lambda:|\lambda|\leq\varepsilon\}\bigcap\sigma(A)=\{0\}.$

引理1.4 [13]  设算子矩阵$M=\begin{bmatrix} A & 0 \\ C & D \end{bmatrix}$, $A \in \mathcal{B}(X)$, $D \in \mathcal{B} (Y)$, $C \in \mathcal{B}(X, Y)$.若${\rm ind}(A)=r, {\rm ind}(D)=s, $那么$M$是Drazin可逆的, 且

$ M^D=\begin{bmatrix} A^D & 0 \\ X & D^D \end{bmatrix}, $

其中$X=\sum\limits_{i=0}^{r-1}(D^D)^{i+2}CA^iA^\pi+\sum\limits_{i=0}^{s-1}D^{\pi}D^iC(A^D)^{i+2}-D^DCA^D.$

2 主要结果

引理2.1  设$P, Q\in\mathcal{B}(X, Y)$均是Drazin可逆的, 且${\rm ind}(P)=t, $ ${\rm ind}(Q)=s$.若$P^5Q=0, P^2Q+PQ^2=0, PQPQ=0$, 则

$ \begin{array}{l} \Delta(\lambda)^{-1}=R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3), \end{array} $ (2.1)

其中$0 < |\lambda| < \min((r(P^D))^{-1}, (r(Q^D))^{-1}), \ \ \Delta(\lambda)=\lambda{I}-Q-R(\lambda, P)PQ. $

  由$P^5Q=0$, 知$P^DQ=0, $进而$P^\pi{PQ}=PQ.$再由$P^2Q+PQ^2=0$, 知

$ \begin{array}{l} P^kQ=-P^{k-1}Q^2=(-1)^{k+1} PQ^k, \ \ k \geq 2. \end{array} $ (2.2)

从而当$0 < |\lambda| < (r(P^D))^{-1}$时, 由(1.1)式, 知

$ \begin{array}{l} R(\lambda, P)PQ&=&\lambda^{-1}PQ+\lambda^{-2}P^2Q+\lambda^{-3}P^3Q+\lambda^{-4}P^4Q\\ &=&\lambda^{-1}PQ-\lambda^{-2}PQ^2+\lambda^{-3}P^3Q-\lambda^{-4}P^3Q^2\\ &=&(\lambda^{-2}PQ+\lambda^{-4}PQ^3)(\lambda{I}-Q). \end{array} $ (2.3)

注意到$PQ^3=P^3Q, PQPQ=0, $可知$PQ^3PQ=P^3QPQ=0.$由此

$ \begin{eqnarray*} \Delta(\lambda)&=&\lambda{I}-Q-R(\lambda, P)PQ\\ &=&(I-\lambda^{-2}PQ-\lambda^{-4}PQ^3)(\lambda{I}-Q)\\ &=&[\lambda^{-2}(\lambda^2-{PQ})-(\lambda^{-4}PQ^3-\lambda^{-6}PQ^3{PQ})](\lambda{I}-Q)\\ &=&\lambda^{-6}(\lambda^4-PQ^3)(\lambda^2-PQ)(\lambda{I}-Q). \end{eqnarray*} $

$(PQ)^2=0$以及$(PQ^3)^2=0, $$PQ, PQ^3$是Drazin可逆的, 且$(PQ)^D=0, (PQ^3)^D=0, $

$ R(\lambda^2, PQ)=\lambda^{-2}+\lambda^{-4}PQ, R(\lambda^4, PQ^3)=\lambda^{-4}+\lambda^{-8}PQ^3. $

于是, 当$0 < |\lambda| < \min((r(P^D))^{-1}, (r(Q^D))^{-1})$时,

$ \begin{eqnarray*} \Delta(\lambda)^{-1}&=&\lambda^{6}R(\lambda, Q)R(\lambda^2, PQ)R(\lambda^4, PQ^3)\\ &=&R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3). \end{eqnarray*} $

结论得证.

引理2.2  设$P, Q\in\mathcal{B}(X, Y)$均是Drazin可逆的, 且${\rm ind}(P)=t, {\rm ind}(Q)=s$.若$P^5Q=0, P^2Q+PQ^2=0, PQPQ=0$, 那么算子矩阵$M=\begin{bmatrix} P & PQ \\ I & Q \end{bmatrix}$的预解式可表示为

$ R(\lambda, M)=\\ \begin{bmatrix} (I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)R(\lambda, P) &\lambda^{-2}PQ+\lambda^{-4}PQ^3\\ R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)R(\lambda, P) &R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3) \end{bmatrix}, $

其中$0 < |\lambda| < \min((r(P^D))^{-1}, (r(Q^D))^{-1}).$

  令$\rho(\Delta)=\{ \lambda\in\mathcal{C}: \Delta(\lambda) \mbox{可逆} \}.$显然, $\rho(M)\bigcap\rho(P)=\rho(P)\bigcap\rho(\Delta).$则当$\lambda\in\rho(M)\cap\rho(P)$时,

$ \begin{array}{l} R(\lambda, M)= \begin{bmatrix} R(\lambda, P)+R(\lambda, P)PQ\Delta(\lambda)^{-1}R(\lambda, P) & R(\lambda, P)PQ\Delta(\lambda)^{-1} \\ \Delta(\lambda)^{-1}R(\lambda, P) & \Delta(\lambda)^{-1} \end{bmatrix}. \end{array} $ (2.4)

由(2.3), (2.1), (2.2)式, 以及$ PQPQ=0$, 有

$ \begin{eqnarray*} &&R(\lambda, P)PQ\Delta(\lambda)^{-1}\\ &=&(\lambda^{-2}PQ+\lambda^{-4}PQ^3)(\lambda{I}-Q)R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)\\ &=&\lambda^{-2}PQ+\lambda^{-4}PQ^3. \end{eqnarray*} $

于是, 便有

$ \begin{eqnarray*} R(\lambda, P)+R(\lambda, P)PQ\Delta(\lambda)^{-1}R(\lambda, P) &=&(I+R(\lambda, P)PQ\Delta(\lambda)^{-1})R(\lambda, P)\\ &=&(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)R(\lambda, P). \end{eqnarray*} $

引理得证.

引理2.3  设$P, Q\in\mathcal{B}(X, Y)$均是Drazin可逆的, 且${\rm ind}(P)=t, {\rm ind}(Q)=s$.若$P^5Q=0, P^2Q+PQ^2=0, PQPQ=0$, 则下列结论成立.

(ⅰ) $R(\lambda, Q)R(\lambda, P)$的Laurent展开式中$-\lambda^0$的系数为

$ \begin{eqnarray*} \sum\limits_{n=0}^{s-1}Q^nQ^\pi(P^D)^{n+2}-Q^DP^D +\sum\limits_{n=0}^{t-1}(Q^D)^{n+2}P^nP^\pi. \end{eqnarray*} $

(ⅱ) $\lambda^{-2}R(\lambda, Q)PQR(\lambda, P)$的Laurent展开式中$-\lambda^0$的系数为

$ \begin{eqnarray*} \sum\limits_{n=0}^{s-1}Q^nQ^\pi{PQ}(P^D)^{n+4}- \sum\limits_{n=1}^3(Q^D)^n{PQ}(P^D)^{4-n} +\sum\limits_{n=0}^{t-1}(Q^D)^{n+4}{PQ}P^nP^\pi. \end{eqnarray*} $

(ⅲ) $\lambda^{-4}R(\lambda, Q)PQ^3R(\lambda, P)$的Laurent展开式中$-\lambda^0$的系数为

$ \begin{eqnarray*} \sum\limits_{n=0}^{s-1}Q^nQ^\pi{PQ^3}(P^D)^{n+6}-\sum\limits_{n=1}^5(Q^D)^n{PQ^3}(P^D)^{6-n} +\sum\limits_{n=0}^{t-1}(Q^D)^{n+6}{PQ^3}P^nP^\pi. \end{eqnarray*} $

(ⅳ) $R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)R(\lambda, P)$的Laurent展开式中$-\lambda^0$的系数为$U-V+W, $其中

$ \begin{eqnarray*} &&U=Q^\pi\sum\limits_{n=0}^{s-1}Q^n(I+{PQ}(P^D)^{2}+{PQ^3}(P^D)^{4})(P^D)^{n+2}, \\ &&V=Q^DP^D+\sum\limits_{n=1}^3(Q^D)^n{PQ}(P^D)^{4-n}+\sum\limits_{n=1}^5(Q^D)^n{PQ^3}(P^D)^{6-n}, \\ &&W=\sum\limits_{n=0}^{t-1}(Q^D)^{n+2}(I+(Q^D)^{2}PQ+(Q^D)^{4}PQ^3)P^nP^\pi. \end{eqnarray*} $

  根据$R(\lambda, P)$$R(\lambda, Q)$的Laurent展开, 即可得到(ⅰ), (ⅱ), (ⅲ)中的结论.又注意到

$ \begin{eqnarray*} &&R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)R(\lambda, P)\\ &=&R(\lambda, Q)R(\lambda, P)+\lambda^{-2}R(\lambda, Q)PQR(\lambda, P)+\lambda^{-4}R(\lambda, Q)PQ^3R(\lambda, P), \end{eqnarray*} $

这样便由(ⅰ), (ⅱ), (ⅲ)可得结论(ⅳ).

定理2.4  设$P, Q\in\mathcal{B}(X, Y)$均是Drazin可逆的, 且${\rm ind}(P)=t, \ {\rm ind}(Q)=s$.若$P^5Q=0, \ P^2Q+PQ^2=0, \ PQPQ=0$, 则$P+Q$是Drazin可逆的, 且

$ \begin{array}{l} (P+Q)^D&=&P^D+PQ(P^D)^3+PQ^3(P^D)^5+\\&&Q(U-V)+((Q^D)^2PQ+(Q^D)^4PQ^3)P^D\\ &&+(Q^DPQ+(Q^D)^3PQ^3)(P^D)^2+\\&&(Q^D)^2PQ^3(P^D)^3+Q^DPQ^3(P^D)^4\\ &&+(W-V)P+Q^D+(Q^D)^3PQ+(Q^D)^5PQ^3, \end{array} $ (2.5)

其中$U, V, W$见引理2.3.

  令$A=\begin{bmatrix} I & Q \end{bmatrix} $, $B=\begin{bmatrix} P \\ I \end{bmatrix}$, 显然$P+Q=AB, $ $BA=M. $根据引理2.2, 知0为预解式$R(\lambda, BA)$的至多为$t+s+4$阶极点, 所以$BA$是Drazin可逆的.再由引理1.2, $AB$也是Drazin可逆的.

注意到$(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)R(\lambda, P)$$R(\lambda, Q)(I+\lambda^{-2}PQ+\lambda^{-4}PQ^3)$的Laurent展开式中$-\lambda^0$的系数分别为$P^D+PQ(P^D)^3+PQ^3(P^D)^5$$Q^D+(Q^D)^3{PQ}+(Q^D)^5{PQ^3}.$于是, 由引理1.3和引理2.2, 有

$ \begin{eqnarray*} (BA)^D&=&-\frac{1}{2\pi{i}}\oint_\gamma\lambda^{-1}R(\lambda, BA)d\lambda\\ &=& \begin{bmatrix} P^D+PQ(P^D)^3+PQ^3(P^D)^5 & 0 \\ U-V+W & Q^D+(Q^D)^3{PQ}+(Q^D)^5{PQ^3} \end{bmatrix}. \end{eqnarray*} $

$P^DQ=0, PQ^D=0$, 以及$PP^D=P^DP$

$ \begin{eqnarray*} &&(P^D+PQ(P^D)^3+PQ^3(P^D)^5)^2=(P^D)^2+PQ(P^D)^4+PQ^3(P^D)^6,\\ &&(U-V)(P^D+PQ(P^D)^3+PQ^3(P^D)^5)=(U-V)P^D, \\ &&W(P^D+PQ(P^D)^3+PQ^3(P^D)^5)\\ &=&\sum\limits_{n=0}^{t-1}(Q^D)^{n+2}P^nP^\pi(PQ(P^D)^3+PQ^3(P^D)^5)\\ &=&\sum\limits_{n=0}^3(Q^D)^{n+2}P^{n+1}Q(P^D)^3+\sum\limits_{n=0}^1(Q^D)^{n+2}P^{n+1}Q^3(P^D)^5,\\ &&(Q^D+(Q^D)^3{PQ}+(Q^D)^5{PQ^3})^2=(Q^D)^2+(Q^D)^4{PQ}+(Q^D)^6{PQ^3},\\ &&(Q^D+(Q^D)^3{PQ}+(Q^D)^5{PQ^3})(W-V)=Q^D(W-V),\\ &&(Q^D+(Q^D)^3{PQ}+(Q^D)^5{PQ^3})U\\ &=&((Q^D)^3{PQ}+(Q^D)^5{PQ^3})\sum\limits_{n=0}^{s-1}Q^nQ^\pi(P^D)^{n+2}\\ &=&\sum\limits_{n=0}^3(Q^D)^3PQ^{n+1}(P^D)^{n+2}+\sum\limits_{n=0}^1(Q^D)^5PQ^{n+3}(P^D)^{n+2}. \end{eqnarray*} $

$ \begin{eqnarray*} &&Z_1=\sum\limits_{n=0}^3(Q^D)^{n+2}P^{n+1}Q(P^D)^3+\sum\limits_{n=0}^1(Q^D)^{n+2}P^{n+1}Q^3(P^D)^5, \\ &&Z_2=\sum\limits_{n=0}^3(Q^D)^3PQ^{n+1}(P^D)^{n+2}+\sum\limits_{n=0}^1(Q^D)^5PQ^{n+3}(P^D)^{n+2}, \end{eqnarray*} $

$ \begin{array}{l}((BA)^D)^2=\\ \begin{bmatrix} (P^D)^2+PQ(P^D)^4+PQ^3(P^D)^6 & 0 \\ (U-V)P^D+Z_1+Z_2+Q^D(W-V) & (Q^D)^2+(Q^D)^4{PQ}+(Q^D)^6{PQ^3} \end{bmatrix}. \end{array} $ (2.6)

因此由引理1.2,

$ \begin{eqnarray*} (P+Q)^D&=&\begin{bmatrix} I & Q \end{bmatrix}((BA)^D)^2\begin{bmatrix} P \\ I \end{bmatrix}\\ &=&P^D +PQ(P^D)^3+PQ^3(P^D)^5+\\&&Q(U-V)+Q(Z_1+Z_2)P+(W-V)P \\ &&+Q^D+(Q^D)^3PQ+(Q^D)^5PQ^3\\ &=&P^D+PQ(P^D)^3+PQ^3(P^D)^5+Q(U-V)+\\&&((Q^D)^2PQ+(Q^D)^4PQ^3)P^D\\ &&+(Q^DPQ+(Q^D)^3PQ^3)(P^D)^2+\\&&(Q^D)^2PQ^3(P^D)^3+Q^DPQ^3(P^D)^4\\ &&+(W-V)P+Q^D+(Q^D)^3PQ+(Q^D)^5PQ^3. \end{eqnarray*} $

定理得证.

由定理2.4, 有如下推论.

推论2.5  设$P, Q \in \mathcal{B}(X, Y)$为Drazin可逆, 且${\rm ind}(P)=t, {\rm ind}(Q)=s. $$PQ=0$, 则

$ \begin{array}{l} (P+Q)^D=P^D+\sum\limits_{n=0}^{s-2}Q^{n+1}Q^\pi(P^D)^{n+2}-QQ^DP^D+\\\sum\limits_{n=0}^{t-2}(Q^D)^{n+2}P^{n+1}P^\pi-Q^DP^DP+Q^D. \end{array} $

推论2.6  设$P, Q \in \mathcal{B}(X, Y)$为Drazin可逆, 且${\rm ind}(P)=t, {\rm ind}(Q)=s. $$P^5Q=0, P^2Q+PQ^2=0, PQP=0$, 则

$ \begin{array}{l} (P+Q)^D&=&P^D+\sum\limits_{n=0}^{s-2}Q^{n+1}Q^\pi(P^D)^{n+2}-QQ^DP^D+\\&&\sum\limits_{n=0}^{t-2}(Q^D)^{n+2}P^{n+1}P^\pi-Q^DP^DP\\ &&+Q^D+(Q^D)^3PQ+(Q^D)^5PQ^3. \end{array} $

推论2.7  设$P, Q \in \mathcal{B}(X, Y)$为Drazin可逆, 且${\rm ind}(P)=t, {\rm ind}(Q)=s. $$P^5Q=0, P^2Q+PQ^2=0, QPQ=0$, 则

$ \begin{array}{l} (P+Q)^D&=&P^D+PQ(P^D)^3+PQ^3(P^D)^5+\sum\limits_{n=0}^{s-2}Q^{n+1}Q^\pi(P^D)^{n+2}-QQ^DP^D\\ &&+\sum\limits_{n=0}^{t-2}(Q^D)^{n+2}P^{n+1}P^\pi-Q^DP^DP+Q^D. \end{array} $

推论2.8  设$P, Q \in \mathcal{B}(X, Y)$为Drazin可逆, 且${\rm ind}(P)=t, \ {\rm ind}(Q)=s.$$P^3Q=0, P^2Q+PQ^2=0, PQPQ=0$, 则

$ \begin{array}{l} (P+Q)^D&=&P^D+PQ(P^D)^3+\sum\limits_{n=0}^{s-2}Q^{n+1}Q^\pi(P^D)^{n+2}+\\ &&\sum\limits_{n=0}^{s-2}Q^{n+1}Q^\pi{PQ}(P^D)^{n+4}-QQ^DP^D\\ &&-QQ^D{PQ}(P^D)^3+\sum\limits_{n=0}^{t-2}(Q^D)^{n+2}P^{n+1}P^\pi+\\ &&\sum\limits_{n=0}^{t-2}(Q^D)^{n+4}{PQ}P^{n+1}P^\pi-Q^DP^DP\\ &&-\sum\limits_{n=1}^3(Q^D)^{n}PQ(P^D)^{4-n}P+Q^D+(Q^D)^3PQ. \end{array} $
3 应用

本节将算子矩阵的Drazin逆应用到四分块算子矩阵$M=\begin{bmatrix} A & B \\ C & D \end{bmatrix}$的Drazin逆上, 并给出$M^D$的表达式, 其中$A\in\mathcal{B}(X), B\in\mathcal{B}(Y, X), C\in\mathcal{B}(X, Y), D\in\mathcal{B}(Y)$.关于四分块(算子)矩阵的Drazin逆请参见文献[3, 5, 9, 10, 13, 14, 16].下面假设$A, D$均是Drazin可逆的, 且${\rm ind}(A)=r, {\rm ind}(D)=s.$

定理3.1  若$A^2B=ABCB=CBCB=0, CAB+DCB=0, $则算子矩阵$M$是Drazin可逆的, 且

$ M^D=\left[\begin{smallmatrix} A^D+ABX(A^D)^2+(B+ABD^D)(XA^D+D^DX)+M_1 \ \ \ & B(D^D)^2+{AB}(D^D)^3+B{CB}(D^D)^4 \\ X+CBX(A^D)^2+CB{D^D}XA^D+CB(D^D)^2X \ \ \ & D^D+CB(D^D)^3 \end{smallmatrix}\right], $

其中

$ \begin{eqnarray*} &&M_1=BCB(X(A^D)^3+D^DX(A^D)^2 +(D^D)^2XA^D+(D^D)^3X), \\ &&X=\sum\limits_{i=0}^{r-1}(D^D)^{i+2}CA^iA^\pi+\sum\limits_{i=0}^{s-1}D^{\pi}D^iC(A^D)^{i+2}-D^DCA^D. \end{eqnarray*} $

  令$P=\begin{bmatrix} A & 0 \\ C & D \end{bmatrix}$, $Q=\begin{bmatrix} 0 & B \\ 0 & 0 \end{bmatrix}, $$M=P+Q.$由于

$ \begin{array}{l} ~~~~~ P^3Q=\begin{bmatrix} 0 & A^3B \\ 0 & CA^2B+DCAB+D^2CB \end{bmatrix}, \\ P^2Q+PQ^2=\begin{bmatrix} 0 & A^2B \\ 0 & CAB+DCB \end{bmatrix}, PQPQ=\begin{bmatrix} 0 & ABCB \\ 0 & CBCB \end{bmatrix}. \end{array} $

由此, 根据已知条件, 有$P^3Q=0, P^2Q+PQ^2=0, PQPQ=0.$又由引理1.4, 有

$ P^D=\begin{bmatrix} A^D & 0 \\ X & D^D \end{bmatrix}, $

其中$X=\sum\limits_{i=0}^{r-1}(D^D)^{i+2}CA^iA^\pi+\sum\limits_{i=0}^{s-1}D^{\pi}D^iC(A^D)^{i+2}-D^DCA^D.$从而, 根据推论2.8, 得到

$ \begin{array}{l} M^D&=&P^D+Q(P^D)^2+PQ(P^D)^3+Q{PQ}(P^D)^4\\ &=&\left[\begin{smallmatrix} A^D+ABX(A^D)^2+(B+ABD^D)(XA^D+D^DX)+M_1\ \ \ & B(D^D)^2+{AB}(D^D)^3+B{CB}(D^D)^4 \\ X+CBX(A^D)^2+CB{D^D}XA^D+CB(D^D)^2X \ \ \ & D^D+CB(D^D)^3 \end{smallmatrix}\right]. \end{array} $

定理得证.

推论3.2 [14]  若$A^2B=0, BCB=0, CAB=0, DCB=0, $

$ \begin{array}{l} M^D=\\\begin{bmatrix} A^D+ABX(A^D)^2+(B+ABD^D)(XA^D+D^DX) \ \ \ & B(D^D)^2+{AB}(D^D)^3 \\ X+CBX(A^D)^2+CB{D^D}XA^D+CB(D^D)^2X \ \ \ & D^D+CB(D^D)^3 \end{bmatrix}. \end{array} $

推论3.3  若$A^2B=CB=CAB=0, $

$ \begin{array}{l} M^D=\\\begin{bmatrix} A^D+ABX(A^D)^2+(B+ABD^D)(XA^D+D^DX) & B(D^D)^2+{AB}(D^D)^3 \\ X & D^D \end{bmatrix}. \end{array} $

推论3.4 [15]  若$AB=CB=0, $

$ \begin{array}{l} &&M^D=\begin{bmatrix} A^D+BXA^D+BD^DX & B(D^D)^2 \\ X & D^D \end{bmatrix}. \end{array} $
参考文献
[1] Campbell S L. The Drazin inverse and systems of second order linear differential equations[J]. Linear Multi-linear Alg., 1983, 14: 195–198. DOI:10.1080/03081088308817556
[2] Drazin M P. Pseudoinverses in associative rings and semigroups[J]. Amer. Math. Month., 1958, 65: 506–514. DOI:10.1080/00029890.1958.11991949
[3] Castro-González N, Dopazo E, Martlnez-Serrano M F. On the Drazin inverse of the sum of two operators and its application to operator matrices[J]. Math. Anal. Appl., 2009, 350: 207–215. DOI:10.1016/j.jmaa.2008.09.035
[4] Deng C Y. The Drazin inverse of the sum and difference of idempotents[J]. Linear Alg. Appl., 2009, 430: 1282–1291. DOI:10.1016/j.laa.2008.10.017
[5] Cvetković A S, Milovanović G V. On Drazin inverse of operator matrices[J]. Math. Anal. Appl., 2011, 375: 331–335. DOI:10.1016/j.jmaa.2010.08.080
[6] Huang J J, Shi Y F, Chen A. Additive results of the Drazin inverse for bounded linear operators[J]. Compl. Anal. Oper. The., 2014, 8: 349–358. DOI:10.1007/s11785-013-0307-5
[7] Hartwig R E, Wang G R, Wei Y M. Some additive results on Drazin inverse[J]. Linear Alg. Appl., 2001, 322: 207–217. DOI:10.1016/S0024-3795(00)00257-3
[8] Patricio P, Hartwig R E. Some additive results on Drazin inverses[J]. Appl. Math. Comput., 2009: 530–538.
[9] Bu C J, Feng C C, Bai S Y. Representations for the Drazin inverses of the sum of two matrices and some block matrices[J]. Appl. Math. Comput., 2012, 218: 10226–10237.
[10] Yang H, Liu X F. The Drazin inverse of the sum of two matrices and its applications[J]. Comput. Appl. Math., 2011, 235: 1412–1417. DOI:10.1016/j.cam.2010.08.027
[11] Wang G R, Wei Y M, Qiao S Z. Generalized inverses:theory and computations[M]. Beijing: Grad. Ser. Math. Sci. Press, 2003.
[12] Caradus S R. Operators theory of the generalized inverse[M]. Queen's Papers Pure Appl. Math., Vol. 38, Queen's Univ., Kingstom: Ontario, 1974.
[13] Meyer C D, Rose N J. The index and the Drazin inverse of block triangular matrices[J]. SIAM J. Appl. Math., 1977, 33: 1–7. DOI:10.1137/0133001
[14] Abdul Shakoor. 矩阵和、分块矩阵与修正矩阵的Drazin逆[D]. 重庆: 重庆大学, 2014. http://www.wanfangdata.com.cn/details/detail.do?_type=degree&id=D508710
[15] Cvetković-Ilić D S. A note on the representation for the Drazin inverse of 2×2 block matrices[J]. Linear Alg. Appl., 2008, 429: 242–248. DOI:10.1016/j.laa.2008.02.019
[16] Huang J J, Shi Y F, Alatancang. The representation of the Drazin inverse of anti-triangular operator matrices based on resolvent expansions[J]. Appl. Math. Comput., 2014, 242: 196–201.
[17] Cui R P, Li X L, Gao J L. The Drazin inverse of a modified matrix A-CB[J]. J. Math., 2014, 34(1): 12–16.