数学物理学报  2017, Vol. 37 Issue (2): 248-256   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
王学武
φ-半伪压缩算子对的Ishikawa型迭代程序的收敛性和稳定性
王学武     
山东工商学院 数学与信息科学学院 山东烟台 264005
摘要:该文引入算子对的稳定性概念,在φ(x)/x非减的条件下,利用与以往文献不同的方法,研究了φ-半伪压缩算子对的Ishikawa型迭代序列的收敛性和稳定性.这些结果推广,改进和完善了迭代逼近理论的相关结果.
关键词φ-半伪压缩算子    Ishikawa型迭代序列    收敛性    稳定性    不动点    
Convergence and Stability of Ishikawa Type Iterative Sequences for the φ-Hemi Pseudocontractive Operator Pair
Wang Xuewu     
School of Mathematics and Information Science, Shandong Institute of Business and Technology, Shandong Yantai 264005
Abstract: In this paper, the stability conceptions for operator pair is introduced. Under the condition that φ(x)/x is nondecreasing, using the different method, we study the convergence and stability of Ishikawa type iterative sequences for the φ-hemi pseudocontrative operator pair. This results presented in this paper extend, improve and perfect the well-known corresponding results in iterative approximation theory.
Key words: φ-Hemi pseudocontrative operator     Ishikawa type iterative sequence     Convergence     Stability     Fixed point    
1 预备知识

在全文中, 设$X$是实Banach空间, $X^{\ast}$$X$的对偶空间, $C$$X$的非空闭凸子集, $\langle\cdot, \cdot\rangle$$X$$X^{\ast}$的对偶对.称集值映象$J:X\rightarrow 2^{X^{\ast}}$:

$ J(x)=\{f\in X^{\ast}:\langle x, f\rangle=\|x\|^{2}=\|f\|^{2}\}, \forall x\in X $

是正规对偶映象.

定义1.1  设$T$$C$上的一个自映象, $F(T)$$T$的不动点集.

(1) 称$T$$L$-Lipschitz的, 若$\exists L>0$, $\forall x, y\in C$, 有$\|Tx-Ty\|\leq L\|x-y\|$;

(2) 称$T$$\phi$-半伪压缩的, 如果存在严格增函数$\phi:[0, \infty)\rightarrow [0, \infty)$$\phi(0)=0$, 使得对任意的$x\in C, p\in F(T)$, 都存在$j(x-p)\in J(x-p)$

$ \langle Tx-p, j(x-p)\rangle\leq \|x-p\| ^{2}-\phi(\|x-p\|)\|x-p\|. $

定义1.2  设$\{\alpha_{n}\}, \{\beta_{n}\}$是[0, 1]上的数列, $T, S$$C$上的自映象, $\forall x_{0}\in C$,在$C$上定义迭代程序: $x_{n+1}=f(T, S, \alpha_{n}, \beta_{n}, x_{n})$.设$F(T)\bigcap F(S)\neq \emptyset$, $\{x_{n}\}$强收敛于$T$$S$的公共不动点$p$. $\{y_{n}\}$$C$上任意序列, 定义数列$\{\varepsilon_{n}\}$:

$ \varepsilon_{n}=\|y_{n+1}-f(T, S, \alpha_{n}, \beta_{n}, y_{n})\|. $

称迭代程序$x_{n+1}=f(T, S, \alpha_{n}, \beta_{n}, x_{n})$关于$T, S$

(1) 稳定的, 若$\lim\limits_{n\rightarrow \infty}\varepsilon_{n}=0$, 则$\lim\limits_{n\rightarrow\infty}y_{n}=p$;

(2) 几乎稳定的, 若$\sum\limits_{n=1}^{\infty}\varepsilon_{n}<\infty$, 则$\lim\limits_{n\rightarrow\infty}y_{n}=p$;

(3) 伪稳定的, 若$\varepsilon_{n}=o(\alpha_{n})$$\varepsilon_{n}=o(\beta_{n})$, 则$\lim\limits_{n\rightarrow\infty}y_{n}=p$;

(4) 弱稳定的, 若$\varepsilon_{n}=\varepsilon_{n}'+\varepsilon_{n}"$, 且$ \sum\limits_{n=1}^{\infty}\varepsilon_{n}'<\infty, \varepsilon_{n}"=o(\alpha_{n})$$\varepsilon"_{n}=o(\beta_{n})$, 则$\lim\limits_{n\rightarrow\infty}y_{n}=p$.

引理1.3[1]  设$X$是实的Banach空间, $J:X\rightarrow 2^{X^{\ast}}$是正规对偶映象,则对任意的$x, y\in X$$j(x+y)\in J(x+y)$都有$\|x+y\|^{2}\leq \|x\|^{2}+2\langle y, j(x+y)\rangle.$

引理1.4[1]  设$\{a_{n}\}$, $\{b_{n}\}$, $\{c_{n}\}$$\{d_{n}\}$是四个非负实数数列,且存在$n_{0}\in {\Bbb N}$

$ a_{n+1}\leq (1-b_{n})a_{n}+c_{n}+d_{n}, n>n_{0}. $

如果$ \sum\limits_{n=1}^{\infty}b_{n}=\infty, \sum\limits_{n=1}^{\infty}d_{n}<\infty $$ c_{n}=o(b_{n}) $, 则$\lim\limits_{n\rightarrow \infty}a_{n}=0$.

在一致光滑的Banach空间的框架下, Chidume[2]和Hirano[3]分别研究了$\phi$-半伪压缩算子和有界$\phi$半伪压缩算子的具误差的Mann迭代序列收敛性.

在任意实Banach空间的框架下, Xu[4], Chang[5, p248, 定理7.2.1]和Gu[6]分别在有界集, 值域有界和迭代序列有界的条件下, 研究了$\phi$-半伪压缩算子的具误差的Ishikawa迭代序列收敛性.

1998年, Osilike[7]在实Banach空间的框架下, 研究了Ishikawa迭代程序的稳定性,得到下面结果:

定理O[7]  设$X$是任意实Banach空间, $C$$X$非空闭凸子集, $T$$C$上的Lipschitz的$\phi$-强伪压缩算子, $F(T)\neq \emptyset$.若$\{\alpha_{n}\}$$\{\beta_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $\sum\limits_{n=1}^{\infty}\alpha_{n}=\infty$;

(ⅱ) $\sum\limits_{n = 1}^\infty {\alpha _n^2} < \infty ,\sum\limits_{n = 1}^\infty {{\alpha _n}} {\beta _n} < \infty $.则对任意的$x_{0}\in C$, 由Ishikawa迭代程序

$\left\{ \begin{array}{ll} x_{n+1}=(1-\alpha_{n})x_{n}+\alpha_{n}Ty_{n}, \\ y_{n}=(1-\beta_{n})x_{n}+\beta_{n}Tx_{n}, n\in {\Bbb N} \end{array}\right.$ (1.1)

生成的序列$\{x_{n}\}$关于$T$是几乎稳定的.

2005年, Chidume[8]在实赋范线性空间上, 在没有有界性的情况下, 研究了$\phi$-半伪压缩算子的迭代序列的收敛性, 得到下面结果:

定理C[8]  设$X$是任意实的赋范线性空间, $C$$X$非空子集, $T$$X$$C$上的一致连续映象, $F(T)\neq \emptyset$.对任意的$x_{0}\in X$, 迭代序列

$ x_{n+1}=a_{n}x_{n}+b_{n}Tx_{n}+c_{n}u_{n} $

中的误差项$\{u_{n}\}$$X$的有界序列, $\{a_{n}\}, \{b_{n}\}$$\{c_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $a_{n}+b_{n}+c_{n}=1, n\in{\Bbb N}$;

(ⅱ) $\sum\limits_{n=1}^{\infty}(b_{n}+c_{n})=\infty$;

(ⅲ) $\sum\limits_{n=1}^{\infty}(b_{n}+c_{n})^{2}<\infty$;

(ⅳ) $\sum\limits_{n=1}^{\infty}c_{n}<\infty$.

如果存在严格增函数$\phi:[0, \infty)\rightarrow [0, \infty)$$\phi(0)=0$, 对任意的$x\in X, x^{\ast}\in F(T)$都存在$j(x-x^{\ast})\in J(x-x^{\ast})$使得

$ \langle Tx-x^{\ast}, j(x-x^{\ast})\rangle\leq \|x-x^{\ast}\|^{2}-\phi(\|x-x^{\ast}\|), $

则迭代序列$\{x_{n}\}$强收敛于$T$的不动点$x^{\ast}$.

纵观$\phi$-伪压缩算子的迭代逼近问题研究的发展轨迹, 主要表现在三个方面:结论建立在Banach空间或赋范空间上; 去掉有界性 (值域有界, 迭代序列有界) 条件; 消弱对迭代参数数列的限制.

本文的主要工作是通过引入算子对的稳定性概念, 研究算子对的Ishikawa型迭代程序

$\left\{ \begin{array}{ll} \forall \, \, x_{0}\in C, \\ x_{n+1}=(1-\alpha_{n})x_{n}+\alpha_{n}Ty_{n}, \\ y_{n}=(1-\beta_{n})x_{n}+\beta_{n}Sx_{n}, n\in {\Bbb N}\\ \end{array}\right.$ (1.2)

的收敛性和稳定性.在Banach空间框架下, 利用与以往文献不同的方法, 没有有界性条件限制, 去掉了常见的定理O的条件 (ⅱ) (或定理C的条件 (ⅲ)), 在$\phi(x)/x$非减条件下, 建立并证明了收敛性定理和稳定性定理.这些结果推广, 改进和完善了文献[2-13]以及相关文献的主要结果.

2 迭代序列的收敛性

定理2.1  设$C$是实Banach空间$X$的非空闭凸子集, $T, S:C\rightarrow C$$L$-Lipschitz的, $T$$\phi$-半伪压缩映象, $\phi(x)/x$是非减函数, $F(T)\bigcap F(S)\neq\emptyset.$$\{\alpha_{n}\}$$\{\beta_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $\lim\limits_{n\rightarrow\infty}\alpha_{n}=0, \lim\limits_{n\rightarrow\infty}\beta_{n}=0$;

(ⅱ) $ \sum\limits_{n=1}^{\infty}\alpha_{n}=\infty.$

则由 (1.2) 式定义的迭代序列$\{x_{n}\}$强收敛于$T$$S$的唯一公共不动点.

  根据算子$T$的性质和假设, $F(T)$是单点集, 不妨设$p\in F(T)\bigcap F(S)$.根据引理1.3和迭代程序 (1.2) 有

$\begin{align} & \|{{x}_{n+1}}-p{{\|}^{2}}\le {{(1-{{\alpha }_{n}})}^{2}}\|{{x}_{n}}-p{{\|}^{2}}+2{{\alpha }_{n}}\langle T{{y}_{n}}-p, j({{x}_{n+1}}-p)\rangle \\ & \le {{(1-{{\alpha }_{n}})}^{2}}\|{{x}_{n}}-p{{\|}^{2}}+2{{\alpha }_{n}}\langle T{{x}_{n+1}}-p, j({{x}_{n+1}}-p)\rangle \\ & +2{{\alpha }_{n}}\langle T{{y}_{n}}-T{{x}_{n+1}}, j({{x}_{n+1}}-p)\rangle \\ & \le {{(1-{{\alpha }_{n}})}^{2}}\|{{x}_{n}}-p{{\|}^{2}}+2{{\alpha }_{n}}[\|{{x}_{n+1}}-p{{\|}^{2}}-\phi (\|{{x}_{n+1}}-p\|)\|{{x}_{n+1}}-p\|] \\ & +2{{\alpha }_{n}}L\|{{y}_{n}}-{{x}_{n+1}}\|\|{{x}_{n+1}}-p\|. \\ \end{align}$ (2.1)

根据 (1.2) 式, 又得到

$\begin{align} & \|{{y}_{n}}-{{x}_{n+1}}\|=\|(1-{{\alpha }_{n}})({{x}_{n}}-{{y}_{n}})+{{\alpha }_{n}}(T{{y}_{n}}-{{y}_{n}})\| \\ & \le (1-{{\alpha }_{n}})\|{{x}_{n}}-{{y}_{n}}\|+{{\alpha }_{n}}\|T{{y}_{n}}-p+p-{{y}_{n}}\| \\ & \le (1-{{\alpha }_{n}})\|{{x}_{n}}-{{y}_{n}}\|+{{\alpha }_{n}}(1+L)(\|{{y}_{n}}-{{x}_{n}}\|+\|{{x}_{n}}-p\|) \\ & \le (1+{{\alpha }_{n}}L)\|{{x}_{n}}-{{y}_{n}}\|+{{\alpha }_{n}}(1+L)\|{{x}_{n}}-p\| \\ & \le (1+{{\alpha }_{n}}L){{\beta }_{n}}\|S{{x}_{n}}-{{x}_{n}}\|+{{\alpha }_{n}}(1+L)\|{{x}_{n}}-p\| \\ & \le (1+{{\alpha }_{n}}L){{\beta }_{n}}\|S{{x}_{n}}-p+p-{{x}_{n}}\|+{{\alpha }_{n}}(1+L)\|{{x}_{n}}-p\| \\ & \le ({{\alpha }_{n}}+{{\beta }_{n}}){{M}^{2}}\|{{x}_{n}}-p\|, \\ \end{align}$ (2.2)

其中$M=L+1$.于是

$\begin{align} & 2{{\alpha }_{n}}L\|{{y}_{n}}-{{x}_{n+1}}\|\|{{x}_{n+1}}-p\|=2{{\alpha }_{n}}L({{\alpha }_{n}}+{{\beta }_{n}}){{M}^{2}}\|{{x}_{n}}-p\|\|{{x}_{n+1}}-p\| \\ & \le {{\alpha }_{n}}({{\alpha }_{n}}+{{\beta }_{n}}){{M}^{3}}(\|{{x}_{n}}-p{{\|}^{2}}+\|{{x}_{n+1}}-p{{\|}^{2}}). \\ \end{align}$ (2.3)

将 (2.3) 式代入 (2.1) 式得到

$\begin{align} & \|{{x}_{n+1}}-p{{\|}^{2}}\le {{(1-{{\alpha }_{n}})}^{2}}\|{{x}_{n}}-p{{\|}^{2}}+2{{\alpha }_{n}}[\|{{x}_{n+1}}-p{{\|}^{2}}-\phi (\|{{x}_{n+1}}-p\|)\|{{x}_{n+1}}-p\|] \\ & +{{\alpha }_{n}}({{\alpha }_{n}}+{{\beta }_{n}}){{M}^{3}}(\|{{x}_{n}}-p{{\|}^{2}}+\|{{x}_{n+1}}-p{{\|}^{2}}) \\ & \le [{{(1-{{\alpha }_{n}})}^{2}}+{{\alpha }_{n}}({{\alpha }_{n}}+{{\beta }_{n}}){{M}^{3}}]\|{{x}_{n}}-p{{\|}^{2}}-2{{\alpha }_{n}}\phi (\|{{x}_{n+1}}-p\|)\|{{x}_{n+1}}-p\| \\ & +[{{\alpha }_{n}}({{\alpha }_{n}}+{{\beta }_{n}}){{M}^{3}}+2{{\alpha }_{n}}]\|{{x}_{n+1}}-p{{\|}^{2}}. \\ \end{align}$ (2.4)

$\tau_{n}=\alpha_{n}(\alpha_{n}+\beta_{n})M^{3}$.化简不等式 (2.4) 得

$\begin{align} & (1-{{\tau }_{n}}-2{{\alpha }_{n}})\|{{x}_{n+1}}-p{{\|}^{2}} \\ & \le [{{(1-{{\alpha }_{n}})}^{2}}+{{\tau }_{n}}]\|{{x}_{n}}-p{{\|}^{2}}-2{{\alpha }_{n}}\phi (\|{{x}_{n+1}}-p\|)\|{{x}_{n+1}}-p\|. \\ \end{align}$ (2.5)

显然, $\tau_{n}\rightarrow 0(n\rightarrow\infty)$.于是$\exists N_{1}\in {\Bbb N}$, $\forall n>N_{1}$$1-\tau_{n}-2\alpha_{n}>0$.

(A) 证明$\mathop {\lim \inf }\limits_{n \to \infty } \left\| {{x_n} - p} \right\| = 0$.若$\liminf_{n\rightarrow\infty}\|x_{n}-p\|=2\rho>0$, 则$\exists N_{2}\in {\Bbb N}(N_{2}>N_{1})$, $\forall n>N_{2}$$\|x_{n}-p\|\geq \rho$.根据$\phi(x)/x$非减性, 则有

$\phi(\|x_{n}-p\|)\geq \lambda \|x_{n}-p\|, n>N_{2}, $ (2.6)

其中$\lambda=\phi(\rho)/\rho>0$.于是根据 (2.5) 和 (2.6) 式有

$\|x_{n+1}-p\|^{2}\leq\frac{(1-\alpha_{n})^{2}+\tau_{n}}{1-\tau_{n}-2\alpha_{n}+2\alpha_{n}\lambda}\|x_{n}-p\|^{2}, \forall n> N_{2}.$ (2.7)

由于$\tau_{n}=o(\alpha_{n})$, 于是$\exists N_{3}(N_{3}>N_{2})$使得

$\frac{(1-\alpha_{n})^{2}+\tau_{n}}{1-\tau_{n}-2\alpha_{n}+2\alpha_{n}\lambda}=1- \frac{2\alpha_{n}\lambda-2\tau_{n}-\alpha_{n}^{2} }{1-\tau_{n}-2\alpha_{n}+2\alpha_{n}\lambda}<1-\frac{\lambda}{2}\alpha_{n}, n>N_{3}.$ (2.8)

根据 (2.7) 和 (2.8) 式得

$\|x_{n+1}-p\|^{2}\leq(1-\frac{\lambda}{2}\alpha_{n})\|x_{n}-p\|^{2}, n>N_{3}.$ (2.9)

根据$\sum\limits_{n=1}^{\infty}\alpha_{n}=\infty$和引理1.4, 得到$\lim\limits_{n\rightarrow\infty}\|x_{n}-p\|=0$, 这与假设矛盾.

(B) 证明$\lim\limits_{n\rightarrow\infty}\|x_{n}-p\|=0$.根据条件 (i) 和$\tau_{n}=o(\alpha_{n})$, 对任意$\varepsilon >0 (\phi(\varepsilon)>0)$, 当$n$充分大时

$ (2\tau_{n}+\alpha_{n}^{2})\varepsilon^{2}-2\alpha_{n}\varepsilon\phi(\varepsilon)<0. $

根据 (A): $\liminf\limits_{n\rightarrow\infty}\|x_{n}-p\|=0$, 则存在$ \{x_{n}\}$的子序列$\{x_{n_{j}}\}$使得$\lim\limits_{j\rightarrow\infty}\|x_{n_{j}}-p\|=0.$因此对任意的$\varepsilon\in (0, 1)$, $\exists n_{j_{0}}\in {\Bbb N}$

$\begin{align} & \|{{x}_{{{n}_{j}}}}-p\|\le \varepsilon, \forall j>{{j}_{0}}, \underset{n\to \infty }{\mathop{\lim \inf }}\, \|{{x}_{n}}-p\|=0 \\ & (2{{\tau }_{n}}+\alpha _{n}^{2}){{\varepsilon }^{2}}-2{{\alpha }_{n}}\varepsilon \phi (\varepsilon )<0, \forall n>{{n}_{{{j}_{0}}}}. \\ \end{align}$ (2.10)

接下来证明:对任意的$n_{j}\geq \max\{n_{j_{0}}, N_{3}\}$

$\|x_{n_{j}+i}-p\|\leq \varepsilon, \forall \, i\in {\Bbb N}.$ (2.11)

事实上, 当$i=0$时, (2.11) 式显然成立.假设当$i=k$时, (2.11) 式成立, 即$\|x_{n_{j}+k}-p\|\leq \varepsilon$.如果

$\|x_{n_{j}+k+1}-p\|> \varepsilon, $ (2.12)

根据$\phi(x)/x$的非减性, 可知函数$\phi(x)\cdot x$也是非减的, 于是有

$\phi(\|x_{n_{j}+k+1}-p\|)\|x_{n_{j}+k+1}-p\|\geq \phi(\varepsilon)\varepsilon >0.$ (2.13)

根据 (2.5) 和 (2.10) 式有

$ \begin{align} & \|{{x}_{{{n}_{j}}+k+1}}-p{{\|}^{2}}\le \frac{{{(1-{{\alpha }_{{{n}_{j}}+k}})}^{2}}+2{{\tau }_{{{n}_{j}}+k}}}{1-{{\tau }_{{{n}_{j}}+k}}-2{{\alpha }_{{{n}_{j}}+k}}}\|{{x}_{{{n}_{j}}+k}}-p{{\|}^{2}}-\frac{2{{\alpha }_{{{n}_{j}}+k}}\varepsilon \phi (\varepsilon )}{1-{{\tau }_{{{n}_{j}}+k}}-2{{\alpha }_{{{n}_{j}}+k}}} \\ & \le \|{{x}_{{{n}_{j}}+k}}-p{{\|}^{2}}+\frac{(2{{\tau }_{{{n}_{j}}+k}}+\alpha _{{{n}_{j}}+k}^{2}){{\varepsilon }^{2}}-2{{\alpha }_{{{n}_{j}}+k}}\varepsilon \phi (\varepsilon )}{1-{{\tau }_{{{n}_{j}}+k}}-2{{\alpha }_{{{n}_{j}}+k}}} \\ & <\|{{x}_{{{n}_{j}}+k}}-p{{\|}^{2}}\le {{\varepsilon }^{2}}. \\ \end{align} $

所以$\|x_{n_{j}+k+1}-p\|< \varepsilon$, 这与 (2.12) 式矛盾, 因此 (2.11) 式成立.这表明$\lim\limits_{n\rightarrow\infty}\|x_{n}-p\|=0$.因此序列$\{x_{n}\}$强收敛于$T$$S$唯一公共不动点.

在定理2.1中, 取$T=S$, 则得到下面结果:

定理2.2  设$C$是实Banach空间$X$的非空闭凸子集, $T:C\rightarrow C$$L$-Lipschitz的$\phi$-半伪压缩映象, $\phi(x)/x$是非减函数, $F(T)\neq\emptyset.$$\{\alpha_{n}\}$$\{\beta_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $\lim\limits_{n\rightarrow\infty}\alpha_{n}=0, \lim\limits_{n\rightarrow\infty}\beta_{n}=0$;

(ⅱ) $ \sum\limits_{n=1}^{\infty}\alpha_{n}=\infty.$

则由 (1.1) 式定义的迭代序列$\{x_{n}\}$强收敛于$T$的唯一不动点.

3 迭代程序的稳定性

定理3.1  设$C$是实Banach空间$X$的非空闭凸子集, $T, S:C\rightarrow C$$L$-Lipschitz的, $T$$\phi$-半伪压缩映象, $\phi(x)/x$是非减函数, $F(T)\bigcap F(S)\neq\emptyset.$ $\{\alpha_{n}\}$$\{\beta_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $\lim\limits_{n\rightarrow\infty}\alpha_{n}=0, \lim\limits_{n\rightarrow\infty}\beta_{n}=0$;

(ⅱ) $ \sum\limits_{n=1}^{\infty}\alpha_{n}=\infty.$

$\{y_{n}\}$$C$上的任意序列, 定义数列$\{\varepsilon_{n}\}\in {\Bbb R}^{+}$如下:

$\left\{ \begin{array}{ll} \mu_{n}=(1-\beta_{n})y_{n}+\beta_{n}Sy_{n}, \\ \varepsilon_{n}=\|y_{n+1}-(1-\alpha_{n})y_{n}-\alpha_{n}T\mu_{n}\|, n\in {\Bbb N}.\\ \end{array}\right.$ (3.1)

如果$\varepsilon_{n}=\varepsilon_{n}^{\prime}+\varepsilon_{n}^{\prime\prime}$, $\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}<\infty$$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$, 则序列$\{y_{n}\}$是有界的.

  设$v_{n}=y_{n+1}-(1-\alpha_{n})y_{n}-\alpha_{n}T\mu_{n}$, 则有

$y_{n+1}=(1-\alpha_{n})y_{n}+\alpha_{n}T\mu_{n}+v_{n}, n\in {\Bbb N}.$ (3.2)

显然$\|v_{n}\|=\varepsilon_{n}=\varepsilon_{n}^{\prime}+\varepsilon_{n}^{\prime\prime}$, $\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}<\infty$$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$.设$p\in F(T)\bigcap F(S)$, 根据 (3.1) 和 (3.2) 式得

$ \begin{align} & \|{{\mu }_{n}}-{{y}_{n+1}}\|\le {{\beta }_{n}}\|S{{y}_{n}}-{{y}_{n}}\|+{{\alpha }_{n}}\|T{{\mu }_{n}}-{{y}_{n}}\|+\|{{v}_{n}}\| \\ & \le {{\beta }_{n}}(L+1)\|{{y}_{n}}-p\|+{{\alpha }_{n}}(\|{{\mu }_{n}}-p\|+\|{{y}_{n}}-p\|)+\|{{v}_{n}}\| \\ & \le {{\beta }_{n}}(L+1)\|{{y}_{n}}-p\|+{{\alpha }_{n}}[(1-{{\beta }_{n}})+{{\beta }_{n}}L+1]\|{{y}_{n}}-p\|+\|{{v}_{n}}\| \\ & \le (L+2)({{\alpha }_{n}}+{{\beta }_{n}})\|{{y}_{n}}-p\|+\|{{v}_{n}}\|. \\ \end{align} $

根据 (3.2) 式有

$y_{n+1}-p=(1-\alpha_{n})(y_{n}-p)+\alpha_{n}(Ty_{n+1}-p)+ \alpha_{n}(T\mu_{n}-Ty_{n+1})+v_{n}, $ (3.3)

于是根据 (3.3) 式和引理1.3得

$\begin{align} & \|{{y}_{n+1}}-p{{\|}^{2}}=(1-{{\alpha }_{n}})\langle {{y}_{n}}-p, j({{y}_{n+1}}-p)\rangle +{{\alpha }_{n}}\langle T{{y}_{n+1}}-p, j({{y}_{n+1}}-p)\rangle \\ & +{{\alpha }_{n}}\langle T{{\mu }_{n}}-T{{y}_{n+1}}, j({{x}_{n+1}}-p)\rangle +\langle {{v}_{n}}, j({{y}_{n+1}}-p) \\ & \le (1-{{\alpha }_{n}})\|{{y}_{n}}-p\|\|{{y}_{n+1}}-p\|+{{\alpha }_{n}}[\|{{y}_{n+1}}-p{{\|}^{2}}-\phi (\|{{y}_{n+1}}-p\|)\|{{y}_{n+1}}-p\|] \\ & +{{\alpha }_{n}}L\|{{\mu }_{n}}-{{y}_{n+1}}\|\|{{y}_{n+1}}-p\|+\|{{v}_{n}}\|\|{{y}_{n+1}}-p\| \\ & \le (1-{{\alpha }_{n}})\|{{y}_{n}}-p\|\|{{y}_{n+1}}-p\|+{{\alpha }_{n}}[\|{{y}_{n+1}}-p{{\|}^{2}}-\phi (\|{{y}_{n+1}}-p\|)\|{{y}_{n+1}}-p\|] \\ & +{{\alpha }_{n}}{{L}_{n}}\|{{y}_{n}}-p\|\|{{y}_{n+1}}-p\|+(1+L)\|{{v}_{n}}\|\|{{y}_{n+1}}-p\|, \\ \end{align}$ (3.4)

其中$L_{n}=L(L+2)(\alpha_{n}+\beta_{n})$.不失一般性, 假设对任意的$n\in {\Bbb N}$, $\|y_{n}-p\|>0$.

(C) 证明$\lim\limits_{n\rightarrow\infty}\|y_{n}-p\|\neq\infty$.若$\lim\limits_{n\rightarrow\infty}\|y_{n}-p\|=\infty$, 则$\exists N_{4}\in {\Bbb N}$使得$\|y_{n}-p\|>1$ $(n>N_{4})$.根据$\phi(x)/x$的非减性, 则有$\phi(\|y_{n+1}-p\|)\geq \phi(1)\|y_{n+1}-p\|$.消去 (3.4) 式两端因子$\|y_{n+1}-p\|$

$ (1-\alpha_{n})\|y_{n+1}-p\|\leq [1-(1-L_{n})\alpha_{n}]\|y_{n}-p\| -\alpha_{n}\phi(1)\|y_{n+1}-p\|+(1+L)\|v_{n}\|. $

由于$\lim\limits_{n\rightarrow\infty}\alpha_{n}=0$, 则$\exists N_{5}$, 使得

$ 1-\alpha_{n}(1-\phi(1))>\frac{1}{2}, \forall n>N_{5} $

$ \frac{1-(1-L_{n})\alpha_{n}}{1-\alpha_{n}(1-\phi(1))}=1-\frac{\alpha_{n} (\phi(1)-L_{n})}{1-\alpha_{n}(1-\phi(1))}\leq 1-\frac{\phi(1)}{2}\alpha_{n}, \forall n>N_{5}. $

于是$\forall n>\max\{N_{4}, N_{5}\}$

$\|y_{n+1}-p\|\leq \Big[1-\frac{\phi(1)}{2}\alpha_{n}\Big]\|y_{n}-p\| +2(1+L)\varepsilon_{n}^{\prime}+2(1+L)\varepsilon_{n}^{\prime\prime}.$ (3.5)

因为$\sum\limits_{n=1}^{\infty}\varepsilon_{n}'<\infty$$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$, 根据 (3.5) 式和引理1.4有$\lim\limits_{n\rightarrow\infty}\|y_{n}-p\|= 0$.这与假设矛盾.

(D) 证明$\{\|y_{n}-p\|\}$是有界的.根据 (C):$\lim\limits_{n\rightarrow\infty}\|y_{n}-p\|\neq \infty$, 则$\exists M(M>1)$$\{y_{n}\}$的子序列$\{y_{n_{j}}\}$使得$\|y_{n_{j}}-p\|\leq M$.由于$\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}< \infty$, 不妨设$\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}=Q$.又由于$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$, 则存在$N_{6}$, 使得当$n_{j}>N_{6}$时有

$2(1+L)\varepsilon_{n_{j}}^{\prime\prime}-\frac{\phi(1)}{2}M\alpha_{n_{j}}<0, \, \, \, 2(1+L)\varepsilon_{n_{j}}"\leq 1.$ (3.6)

因为对任意的$j\in {\Bbb N}$, $\|y_{n_{j}}-p\|\leq M$, 于是对于任意的$n_{j}>\max\{N_{4}, N_{5}, N_{6}\}$, 显然有

$ \|y_{n_{j}}-p\|\leq M, \, \, \, \, \|y_{n_{j+1}}-p\|\leq M. $

若子列$\{\|y_{n_{j}}-p\|\}$中的相邻两项$\|y_{n_{j}}-p\|$$\|y_{n_{j+1}}-p\|$之间存在某一项$\|y_{n_{j}+k}-p\|\in \{ \|y_{n}-p\|\}$

$ \|y_{n_{j}+k}-p\|>M>1, $

其中$k\in \{1, 2, \cdots, n_{j+1}-n_{j}-1\}$, 则可以证明

$ \|y_{n_{j}+k}-p\|\leq M+2(L+1)Q+1. $

事实上, 由于$\forall j\in {\Bbb N}$, $\|y_{n_{j}}-p\|\leq M$, 若数列$\{\|y_{n}-p\|\}$中的第$n_{j}+k$$\|y_{n_{j}+k}-p\|>M>1$, 则第$n_{j}+k$项前一定有连续$m$项大于$M$, 第$n_{j}+k-m-1$项小于或等于$M$, 即

$\|y_{n_{j}+k-1}-p\|> M, \|y_{n_{j}+k-2}-p\|> M, \cdots, \|y_{n_{j}+k-m}-p\|>M, $ (3.7)

$\|y_{n_{j}+k-m-1}-p\|\leq M, $ (3.8)

其中$m\in \{0, 1, 2, \cdots, k-1\}$.于是, 对任意$n_{j}>\max\{N_{4}, N_{5}, N_{6}\}$, 根据 (3.5)-(3.8) 式有

$ \begin{align} & \|{{y}_{{{n}_{j}}+k}}-p\|\le \|{{y}_{{{n}_{j}}+k-m-1}}-p\|+2(1+L)\sum\limits_{i=1}^{m+1}{\varepsilon _{{{n}_{j}}+k-i}^{\prime }}+2(1+L)\varepsilon {{"}_{{{n}_{j}}+k-m-1}} \\ & +\sum\limits_{i=1}^{m}{\{}2(1+L)\varepsilon _{{{n}_{j}}+k-i}^{\prime \prime }-\frac{\phi (1)}{2}M{{\alpha }_{{{n}_{j}}+k-i}}\} \\ & \le M+2(1+L)Q+1. \\ \end{align} $

所以$\{\|y_{n}-p\|\}$是有界的, 这说明序列$\{y_{n}\}$是有界的.

定理3.2  设$C$是实Banach空间$X$的非空闭凸子集, $T, S:C\rightarrow C$$L$-Lipschitz的, $T$$\phi$-半伪压缩映象, $\phi(x)/x$是非减函数, $F(T)\bigcap F(S)\neq\emptyset.$又设$\{\alpha_{n}\}$$\{\beta_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $\lim\limits_{n\rightarrow\infty}\alpha_{n}=0, \lim\limits_{n\rightarrow\infty}\beta_{n}=0$;

(ⅱ) $ \sum\limits_{n=1}^{\infty}\alpha_{n}=\infty.$

对任意的$x_{0}\in C$, Ishikiawa型迭代序列$\{x_{n}\}$如 (1.2) 式所定义.设$\{y_{n}\}$$C$上的任意序列, 定义数列$\{\varepsilon_{n}\}\in {\Bbb R}^{+}$如下:

$\left\{ \begin{array}{ll} \mu_{n}=(1-\beta_{n})y_{n}+\beta_{n}Sy_{n}, \\ \varepsilon_{n}=\|y_{n+1}-(1-\alpha_{n})y_{n}-\alpha_{n}T\mu_{n}\|, n\in {\Bbb N}, \\ \end{array}\right.$ (3.9)

其中$\varepsilon_{n}=\varepsilon_{n}^{\prime}+\varepsilon_{n}^{\prime\prime}$, $\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}<\infty$$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$.则

(1) 由 (1.2) 式定义的迭代序列$\{x_{n}\}$强收敛于$T$$S$唯一公共不动点;

(2) 由 (1.2) 式定义的迭代程序$\{x_{n}\}$关于$T, S$是弱稳定的.

  根据定理2.1, 结论 (1) 成立.下面证明结论 (2).

$v_{n}=y_{n+1}-(1-\alpha_{n})y_{n}-\alpha_{n}T\mu_{n}$, 则有

$y_{n+1}=(1-\alpha_{n})y_{n}+\alpha_{n}T\mu_{n}+v_{n}, n\in {\Bbb N}.$ (3.10)

根据 (3.9) 和 (3.10) 式得到

$\left\{ \begin{array}{ll} y_{n+1}=(1-\alpha_{n})y_{n}+\alpha_{n}T^{n}\mu_{n}+v_{n}, n\in {\Bbb N}, \\ \mu_{n}=(1-\beta_{n})y_{n}+\beta_{n}S^{n}y_{n}, n\in {\Bbb N}.\\ \end{array}\right.$ (3.11)

显然, $\|v_{n}\|=\varepsilon_{n}=\varepsilon_{n}^{\prime}+\varepsilon_{n}^{\prime\prime}$, $\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}<\infty$$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$.

$p\in F(T)\bigcap F(S)$, 根据定理3.1, 存在$M_{0}$, $\forall n\in {\Bbb N}$$\|y_{n}-p\|\leq M_{0}$.与证明定理2.1的 (2.5) 式类似, 根据 (3.11) 式得到

$\begin{align} & (1-{{\tau }_{n}}-2{{\alpha }_{n}})\|{{y}_{n+1}}-p{{\|}^{2}} \\ & \le [{{(1-{{\alpha }_{n}})}^{2}}+{{\tau }_{n}}]\|{{y}_{n}}-p{{\|}^{2}}-2{{\alpha }_{n}}\phi (\|{{y}_{n+1}}-p\|)\|{{y}_{n+1}}-p\|+2\|{{v}_{n}}\|\|{{y}_{n+1}}-p\| \\ & \le [{{(1-{{\alpha }_{n}})}^{2}}+{{\tau }_{n}}]\|{{y}_{n}}-p{{\|}^{2}}-2{{\alpha }_{n}}\phi (\|{{y}_{n+1}}-p\|)\|{{y}_{n+1}}-p\|+2{{M}_{0}}(\varepsilon _{n}^{\prime }+\varepsilon _{n}^{\prime \prime }), \\ \end{align}$ (3.12)

其中$\tau_{n}=\alpha_{n}(\alpha_{n}+\beta_{n})M^{3}$$M=L+1$.根据 (3.12) 式, 利用与定理2.1的证明 (A) 的类似方法, 可以证明

$ \liminf\limits_{n\rightarrow\infty} \|y_{n}-p\|=0, $

于是存在$\{\|y_{n}-p\|\}$的子列$\{\|y_{n_{j}}-p\|\}$使得

$ \lim\limits_{j\rightarrow\infty}\|y_{n_{j}}-p\|=0. $

因此对任意的$\varepsilon>0$, 利用$\sum\limits_{n=1}^{\infty}\varepsilon_{n}^{\prime}<\infty$$\varepsilon_{n}^{\prime\prime}=o(\alpha_{n})$, 应用定理3.1的证明 (D) 的类似方法, 可以证明, 存在$N\in{\Bbb N}$, 使得当$ n>N$时有

$ \|y_{n}-p\|\leq C\varepsilon, $

其中$C$是某个非负常数, 因此$\lim\limits_{n\rightarrow\infty}y_{n}=p$.所以由 (1.2) 式定义的迭代程序$\{x_{n}\}$关于$T, S$是弱稳定的.

在定理3.2中, 取$T=S$, 得到下面结果:

定理3.3  设$C$是实Banach空间$X$的非空闭凸子集, $T:C\rightarrow C$$L$-Lipschitz的$\phi$-半伪压缩映象, $\phi(x)/x$是非减函数, $F(T)\neq\emptyset.$$\{\alpha_{n}\}$$\{\beta_{n}\}$$[0,1]$上的数列且满足:

(ⅰ) $\lim\limits_{n\rightarrow\infty}\alpha_{n}=0, \lim\limits_{n\rightarrow\infty}\beta_{n}=0$;

(ⅱ) $ \sum\limits_{n=1}^{\infty}\alpha_{n}=\infty.$

(1) 由 (1.1) 式定义的迭代序列$\{x_{n}\}$强收敛于$T$的唯一公共不动点;

(2) 由 (1.1) 式定义的迭代程序$\{x_{n}\}$关于$T$是弱稳定的.

参考文献
[1] Chang S S, Joseph Lee H W, Chan C K. On Reich's strong convergence theorem for asymptotically nonexpansive mappings in Banach spaces. Nonlinear Analysis, 2007, 66: 2364–2374. DOI:10.1016/j.na.2006.03.025
[2] Chidume C E, Chidume C O. Convergence theorem for zeros of generalized Lipschitz generalized φ-quasi-accretive operators. Proc Amer Math Soc, 2005, 134: 243–251.
[3] Hirano H, Huang Z. Convergence theorems for multi-valued φ-hemicontractive operators and phi-strongly accretive operators. Comput Math Appl, 2003, 46: 1461–1471. DOI:10.1016/S0898-1221(03)90183-0
[4] Xu Y G. Iterative processes with random errors for fixed point of φ-pseudocontractive operator. Rostock Math Kolloq, 2005, 59: 87–97.
[5] Chang S S, Cho Y J, Zhou H.Iterative Methods for Nonlinear Operator Equations in Banach Spaces.Huntington, NY:Nova Science Publishers, 2002
[6] Gu F. Convergence theorem for φ-pseudo contractive type mappings in normed linear spaces. Northeast Math J, 2001, 17: 340–346.
[7] Osilike M O. Stability of the Mann and Ishikawa iteration procedures for φ-strong pseudocontractions and nonlinear equations of the φ-strong accretive type. J Math Anal Appl, 1998, 227: 319–334. DOI:10.1006/jmaa.1998.6075
[8] Chidume C E, Chidume C O. Convergence theorem for fixed points of uniformly continuous generalized φ-hemicontactive mappings. J Math Anal Appl, 2005, 303: 545–554. DOI:10.1016/j.jmaa.2004.08.060
[9] Zhou H Y, Chang S S, Cho Y J. Weak stability of the Ishikawa iteration procedures for φ-hemicontractions operators. Appl Math Lett, 2001, 14: 949–954. DOI:10.1016/S0893-9659(01)00070-2
[10] Gu F. Necessary and sufficient condition of the strong convergence for two finite families of uniformly L-Lipschitzain mappings. Acta Math Sci, 2011, 31B: 2058–2066.
[11] Kim J K, Cho S Y, Qin X. Some results on generalized equilibrium problems involving strictly pseudocon-tractive mappings. Acta Math Sci, 2011, 31B: 2041–2057.
[12] Cho S Y, Kang S M. Approximation of common solutions of variational inequalities via strict pseudocon-tractions. Acta Math Sci, 2012, 32B: 1607–1618.
[13] Shehu Y. Approximation of fixed points and variational solutions for pseudo-contractive mappings in Banach spaces. Acta Math Sci, 2014, 34B: 409–423.