On the equivalence of Mann and Ishikawa iteration methods

Abstract

Authors

B. E. Rhoades
Department of Mathematics, Indiana University, Bloomington, IN47405-7106, USA

Stefan M. Soltuz:
“T. Popoviciu” Institute of Numerical Analysis, Romanian Academy

Keywords

?

Paper coordinates

B.E. Rhoades and Ş.M. Şoltuz, On the equivalence of Mann and Ishikawa iteration methods, Internat. J. Math. Math. Sci. 2003 (7), 451-459, https://doi.org/10.1155/S0161171203110198

PDF

About this paper

Journal

International Journal of Mathematics and Mathematical Sciences

 

Publisher Name

Wiley Online Library

Print ISSN
Online ISSN

google scholar link

[1] C. E. Chidume and S. A. Mutangadura, An example of the Mann iteration methodfor Lipschitz pseudocontractions, Proc. Amer. Math. Soc. 129 (2001), no. 8,2359–2363.
[2] C. E. Chidume and M. O. Osilike, Nonlinear accretive and pseudo-contractive operator equations in Banach spaces, Nonlinear Anal. 31 (1998), no. 7, 779–789.
[3] L. B. Ciric, A generalization of Banach’s contraction principle, Proc. Amer. Math.Soc. 45 (1974), 267–273.
[4] K. Deimling, Zeros of accretive operators, Manuscripta Math. 13 (1974), 365–374.
[5] S. Ishikawa, Fixed points by a new iteration method, Proc. Amer. Math. Soc. 44(1974), 147–150.
[6] T. Kato, Nonlinear semigroups and evolution equations, J. Math. Soc. Japan 19(1967), 508–520.
[7] W. R. Mann, Mean value methods in iteration, Proc. Amer. Math. Soc. 4 (1953),506–510.
[8] B. E. Rhoades, Fixed point iterations using infinite matrices, Trans. Amer. Math.Soc. 196 (1974), 161–176.
[9] ,____ Comments on two fixed point iteration methods, J. Math. Anal. Appl. 56(1976), no. 3, 741–750.
[10] ,_____ A comparison of various definitions of contractive mappings, Trans. Amer.Math. Soc. 226 (1977), 257–290.[11] X. Weng, Fixed point iteration for local strictly pseudo-contractive mapping, Proc.Amer. Math. Soc. 113 (1991), no. 3, 727–731.
[12],_____ H. K. Xu, A note on the Ishikawa iteration scheme, J. Math. Anal. Appl. 167 (1992),no. 2, 582–587

Paper (preprint) in HTML form

ON THE EQUIVALENCE OF MANN AND ISHIKAWA ITERATION METHODS

B. E. RHOADES and STEFAN M. SOLTUZ

Received 19 October 2001

We show that certain Mann and Ishikawa iteration schemes are equivalent for various classes of functions.

2000 Mathematics Subject Classification: 47H10.

The Mann iterative scheme was invented in 1953, see [7], and was used to obtain convergence to a fixed point for many functions for which the Banach principle fails. For example, the first author in [8] showed that, for any continuous selfmap of a closed and bounded interval, the Mann iteration converges to a fixed point of the function.

In 1974, Ishikawa [5] devised a new iteration scheme to establish convergence for a Lipschitzian pseudocontractive map in a situation where the Mann iteration process failed to converge.

Let XX be a Banach space. The Mann iteration is defined by

x0∈X,xn+1=(1−αn)​xn+αn​T​xn,n≥0,x_{0}\in X,\quad x_{n+1}=\left(1-\alpha_{n}\right)x_{n}+\alpha_{n}Tx_{n},\quad n\geq 0, (1)

where the αn∈(0,1)\alpha_{n}\in(0,1), for all n≥0n\geq 0.
The Ishikawa iteration scheme is defined by

u0∈X,un+1=(1−αn)​un+αn​T​vn,\displaystyle u_{0}\in X,\quad u_{n+1}=\left(1-\alpha_{n}\right)u_{n}+\alpha_{n}Tv_{n},
vn=(1−βn)​un+βn​T​un,n≥0,\displaystyle v_{n}=\left(1-\beta_{n}\right)u_{n}+\beta_{n}Tu_{n},\quad n\geq 0, (2)

where

0≤αn≤βn≤1,∀n≥00\leq\alpha_{n}\leq\beta_{n}\leq 1,\quad\forall n\geq 0 (3)

In specific situations, additional conditions are placed on (αn)n\left(\alpha_{n}\right)_{n} and (βn)n\left(\beta_{n}\right)_{n}. In [9], the first author modified the definition of Ishikawa by replacing (3) by

0≤αn,βn≤1,∀n≥00\leq\alpha_{n},\beta_{n}\leq 1,\quad\forall n\geq 0 (4)

During the past 25 years, a large literature has developed around the themes of establishing convergence of the Mann iteration for certain classes of functions, and then showing that the Ishikawa iteration, using (4), also converges.

Of course, having established the convergence of an Ishikawa method using (4), we obtain as a corollary the convergence of the corresponding Mann iteration method by setting each βn=0\beta_{n}=0.

A reasonable conjecture is that, whenever TT is a function for which Mann iteration converges, so does the Ishikawa iteration. Given the large variety of functions and spaces, such a global statement is, of course, not provable. However, in this paper, we do show that, for several classes of functions, Mann and Ishikawa iteration procedures are equivalent.

Picard iteration is defined by p0∈Xp_{0}\in X and pn+1=T​pnp_{n+1}=Tp_{n}, where TT is a selfmap of XX.

Theorem 1. Let XX be a normed space and let BB be a nonempty convex subset of X,T:B→BX,T:B\rightarrow B, with TT satisfying

‖T​x−T​y‖≤k​max⁡{‖x−y‖,‖x−T​x‖,‖y−T​y‖,‖x−T​y‖,‖y−T​x‖}\|Tx-Ty\|\leq k\max\{\|x-y\|,\|x-Tx\|,\|y-Ty\|,\|x-Ty\|,\|y-Tx\|\} (5)

for all x,y∈B,0≤k<1x,y\in B,0\leq k<1. Suppose that TT possesses a fixed point p∈Bp\in B. Then Picard iteration and certain Mann and Ishikawa iteration schemes converge strongly to pp.

Proof. Let p0∈Bp_{0}\in B and define pn+1=T​pn,n≥0p_{n+1}=Tp_{n},n\geq 0. From [3], it follows that (pn)n\left(p_{n}\right)_{n} is Cauchy in BB. Hence, it converges to a point x∗∈B¯x^{*}\in\bar{B}. From (5),

‖pn+1−p‖\displaystyle\left\|p_{n+1}-p\right\| =‖T​pn−T​p‖\displaystyle=\left\|Tp_{n}-Tp\right\|
≤kmax{∥pn−p∥,∥pn−Tpn∥\displaystyle\leq k\max\left\{\left\|p_{n}-p\right\|,\left\|p_{n}-Tp_{n}\right\|\right.
∥pn−Tp∥,∥Tpn−p∥}\displaystyle\left.\left\|p_{n}-Tp\right\|,\left\|Tp_{n}-p\right\|\right\} (6)
=kmax{∥pn−p∥,∥pn−pn+1∥\displaystyle=k\max\left\{\left\|p_{n}-p\right\|,\left\|p_{n}-p_{n+1}\right\|\right.
∥pn−p∥,∥pn+1−p∥}\displaystyle\left.\left\|p_{n}-p\right\|,\left\|p_{n+1}-p\right\|\right\}

Taking the limit as n→∞n\rightarrow\infty yields ‖p−x∗‖≤k​‖p−x∗‖\left\|p-x^{*}\right\|\leq k\left\|p-x^{*}\right\|, which implies that p=x∗p=x^{*}, that is, (pn)n\left(p_{n}\right)_{n} converges strongly to pp.

In [9], it was noted that [8, Theorem 6] could be extended to maps satisfying (5), that is, Mann iteration of a TT satisfying (5) with αn∈(0,1)\alpha_{n}\in(0,1) and bounded away from zero converges strongly to the unique fixed point pp of TT.

In [12], it was shown that, for TT satisfying (5) with αn∈(0,1)\alpha_{n}\in(0,1), the Ishikawa method, with each αn>0\alpha_{n}>0 and ∑αn=∞\sum\alpha_{n}=\infty, converges strongly to pp.

As shown in [10], inequality (5) is one of the most general contractive conditions for a single map.

We need the following lemma.
Lemma 2 (see [11]). Let (an)n\left(a_{n}\right)_{n} be a nonnegative sequence that satisfies the inequality

an+1≤(1−λn)​an+σn,a_{n+1}\leq\left(1-\lambda_{n}\right)a_{n}+\sigma_{n}, (7)

where λn∈(0,1)\lambda_{n}\in(0,1) for each n∈ℕ,∑n=1∞λn=∞n\in\mathbb{N},\sum_{n=1}^{\infty}\lambda_{n}=\infty, and σn=ϵn​λn,limϵn=0\sigma_{n}=\epsilon_{n}\lambda_{n},\lim\epsilon_{n}=0. Then liman=0\lim a_{n}=0.

We are able now to prove the following result.
Theorem 3. Let XX be a normed space, KK a nonempty closed convex subset of XX, and TT a Lipschitzian selfmap of KK with Lipschitz constant L≤1L\leq 1. Suppose that TT has a fixed point p∈Bp\in B. Let x0=u0∈Kx_{0}=u_{0}\in K, and define xnx_{n} and unu_{n} by (1) and (2), with αn,βn\alpha_{n},\beta_{n} satisfying (4), (i) limαn=limβn=0\lim\alpha_{n}=\lim\beta_{n}=0, and (ii) ∑αn=∞\sum\alpha_{n}=\infty. Then the following are equivalent:
(a) the Mann iteration converges strongly to pp,
(b) the Ishikawa iteration converges strongly to pp.

Proof. That (b) implies (a) is obvious setting βn=0\beta_{n}=0 in (2). We prove that (a) implies (b). From (1) and (2),

‖xn+1−un+1‖\displaystyle\left\|x_{n+1}-u_{n+1}\right\| =‖(1−αn)​(xn−un)+αn​(T​xn−T​vn)‖\displaystyle=\left\|\left(1-\alpha_{n}\right)\left(x_{n}-u_{n}\right)+\alpha_{n}\left(Tx_{n}-Tv_{n}\right)\right\|
≤(1−αn)​‖xn−un‖+αn​L​‖T​xn−T​vn‖\displaystyle\leq\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\alpha_{n}L\left\|Tx_{n}-Tv_{n}\right\|
=(1−αn)​‖xn−un‖+αn​L​‖(1−βn)​un+βn​T​un−xn‖\displaystyle=\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\alpha_{n}L\left\|\left(1-\beta_{n}\right)u_{n}+\beta_{n}Tu_{n}-x_{n}\right\|
=(1−αn)​‖xn−un‖+αn​L​‖(1−βn)​(un−xn)+βn​(T​un−xn)‖\displaystyle=\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\alpha_{n}L\left\|\left(1-\beta_{n}\right)\left(u_{n}-x_{n}\right)+\beta_{n}\left(Tu_{n}-x_{n}\right)\right\|
≤[1−αn​(1−L​(1−βn))]​‖xn−un‖+αn​βn​M,\displaystyle\leq\left[1-\alpha_{n}\left(1-L\left(1-\beta_{n}\right)\right)\right]\left\|x_{n}-u_{n}\right\|+\alpha_{n}\beta_{n}M, (8)

for some positive MM since (‖T​un−xn‖)n\left(\left\|Tu_{n}-x_{n}\right\|\right)_{n} is bounded. This fact is obvious if we prove that (‖un‖)n\left(\left\|u_{n}\right\|\right)_{n} is bounded. A simple induction lead us to

‖un+1‖\displaystyle\left\|u_{n+1}\right\| ≤(1−αn)​‖un‖+αn​‖T​vn‖\displaystyle\leq\left(1-\alpha_{n}\right)\left\|u_{n}\right\|+\alpha_{n}\left\|Tv_{n}\right\|
≤(1−αn)​‖un‖+αn​L​‖(1−βn)​un+βn​T​un‖\displaystyle\leq\left(1-\alpha_{n}\right)\left\|u_{n}\right\|+\alpha_{n}L\left\|\left(1-\beta_{n}\right)u_{n}+\beta_{n}Tu_{n}\right\|
≤(1−αn)​‖un‖+αn​L​(1−βn)​‖un‖+αn​βn​L​‖T​un‖\displaystyle\leq\left(1-\alpha_{n}\right)\left\|u_{n}\right\|+\alpha_{n}L\left(1-\beta_{n}\right)\left\|u_{n}\right\|+\alpha_{n}\beta_{n}L\left\|Tu_{n}\right\| (9)
=‖un‖≤⋯≤‖u0‖\displaystyle=\left\|u_{n}\right\|\leq\cdots\leq\left\|u_{0}\right\|

With an:=‖xn−un‖,λn:=αn​(1−L​(1−βn))∈(0,1)a_{n}:=\left\|x_{n}-u_{n}\right\|,\lambda_{n}:=\alpha_{n}\left(1-L\left(1-\beta_{n}\right)\right)\in(0,1), and σn:=αn​βn​M\sigma_{n}:=\alpha_{n}\beta_{n}M, for each n∈ℕn\in\mathbb{N}, the inequality of Lemma 2 is satisfied. Therefore,

lim‖xn−un‖=0\lim\left\|x_{n}-u_{n}\right\|=0 (10)

Since (a) is true, using (10),

‖un−p‖≤‖xn−p‖+‖xn−un‖,\left\|u_{n}-p\right\|\leq\left\|x_{n}-p\right\|+\left\|x_{n}-u_{n}\right\|, (11)

which implies that lim‖un−p‖=0\lim\left\|u_{n}-p\right\|=0.

Let XX be an arbitrary Banach space and let JJ be the normalized duality map from XX into 2X∗2^{X^{*}}. A map TT with domain D​(T)D(T) and range R​(T)R(T) is called strongly pseudocontractive (pseudocontractive) if, for each x,y∈D​(T)x,y\in D(T), there exists j​(x−y)∈J​(x−y)j(x-y)\in J(x-y) and a t>1​(t=1)t>1(t=1) such that

⟨T​x−T​y,j​(x−y)⟩≤1t​‖x−y‖2.\langle Tx-Ty,j(x-y)\rangle\leq\frac{1}{t}\|x-y\|^{2}. (12)

Equivalently, there exists a constant t>1t>1 such that

⟨(I−T)​x−(I−T)​y,j​(x−y)⟩≥t−1t​‖x−y‖2.\langle(I-T)x-(I-T)y,j(x-y)\rangle\geq\frac{t-1}{t}\|x-y\|^{2}. (13)

If we set k=(t−1)/tk=(t-1)/t, then the above inequality can be written in the form

⟨(I−T−k​I)​x−(I−T−k​I)​y,j​(x−y)⟩≥0\langle(I-T-kI)x-(I-T-kI)y,j(x-y)\rangle\geq 0 (14)

and, from a result of Kato [6],

‖x−y‖≤‖x−y+r​[(I−T−k​I)​x−(I−T−k​I)​y]‖,\|x-y\|\leq\|x-y+r[(I-T-kI)x-(I-T-kI)y]\|, (15)

for all x,y∈Xx,y\in X and r>0r>0.
Theorem 4. Let KK be a closed convex subset of an arbitrary Banach space XX and let TT be a Lipschitzian strongly pseudocontractive selfmap of KK. Let x0=u0∈Kx_{0}=u_{0}\in K, and xnx_{n} and unu_{n} be defined by (1) and (2), with αn,βn\alpha_{n},\beta_{n}, satisfying (4) and conditions (i) and (ii) of Theorem 3. Let pp be the fixed point of TT. Then the following are equivalent:
(a) the Mann iterative scheme converges to pp,
(b) the Ishikawa iteration scheme converges to pp.

Proof. The existence of a fixed point pp comes from [4, Corollary 1] which holds in an arbitrary Banach space. That (b) implies (a) is obvious setting βn=0\beta_{n}=0.

Without loss of generality, we may assume that the Lipschitz constant LL of TT is greater than or equal to 1 . If L∈(0,1]L\in(0,1], then the result follows from Theorem 3.

To prove that (a) implies (b), it is necessary to express ‖un+1−xn+1‖\left\|u_{n+1}-x_{n+1}\right\| in terms of (15). Using (1), (2), and the identity which appears as [2, formula (10), page 782], we obtain

‖un−xn‖=\displaystyle\left\|u_{n}-x_{n}\right\|= ‖un+1+αn​un−αn​T​vn−xn+1−αn​xn+αn​T​xn‖\displaystyle\left\|u_{n+1}+\alpha_{n}u_{n}-\alpha_{n}Tv_{n}-x_{n+1}-\alpha_{n}x_{n}+\alpha_{n}Tx_{n}\right\|
=\displaystyle= ∥(1+αn)un+1+αn(I−T−kI)un+1−(1−k)αnun\displaystyle\|\left(1+\alpha_{n}\right)u_{n+1}+\alpha_{n}(I-T-kI)u_{n+1}-(1-k)\alpha_{n}u_{n}
+(2−k)​αn2​(un−T​vn)+αn​(T​un+1−T​vn)\displaystyle+(2-k)\alpha_{n}^{2}\left(u_{n}-Tv_{n}\right)+\alpha_{n}\left(Tu_{n+1}-Tv_{n}\right)
−(1+αn)​xn+1−αn​(I−T−k​I)​xn+1+(1−k)​αn​xn\displaystyle-\left(1+\alpha_{n}\right)x_{n+1}-\alpha_{n}(I-T-kI)x_{n+1}+(1-k)\alpha_{n}x_{n}
−(2−k)αn2(xn−Txn)−αn(Txn+1−Txn)∥\displaystyle\quad-(2-k)\alpha_{n}^{2}\left(x_{n}-Tx_{n}\right)-\alpha_{n}\left(Tx_{n+1}-Tx_{n}\right)\|
=∥(1+αn)(un+1−xn+1)+αn[(I−T−kI)un+1−(I−T−kI)xn+1]\displaystyle=\|\left(1+\alpha_{n}\right)\left(u_{n+1}-x_{n+1}\right)+\alpha_{n}\left[(I-T-kI)u_{n+1}-(I-T-kI)x_{n+1}\right]
−(1−k)​αn​(un−xn)+(2−k)​αn2​(un−T​vn−xn+T​xn)\displaystyle\quad-(1-k)\alpha_{n}\left(u_{n}-x_{n}\right)+(2-k)\alpha_{n}^{2}\left(u_{n}-Tv_{n}-x_{n}+Tx_{n}\right)
+αn(Tun+1−Tvn−Txn+1+Txn)∥\displaystyle\quad+\alpha_{n}\left(Tu_{n+1}-Tv_{n}-Tx_{n+1}+Tx_{n}\right)\| (16)

Using the triangular inequality and (15),

∥un\displaystyle\|u_{n} −xn∥\displaystyle-x_{n}\|
≥\displaystyle\geq (1+αn)​‖(un+1−xn+1)+αn1+αn​[(I−T−k​I)​un+1−(I−T−k​I)​xn+1]‖\displaystyle\left(1+\alpha_{n}\right)\left\|\left(u_{n+1}-x_{n+1}\right)+\frac{\alpha_{n}}{1+\alpha_{n}}\left[(I-T-kI)u_{n+1}-(I-T-kI)x_{n+1}\right]\right\|
−(1−k)​αn​‖un−xn‖−(2−k)​αn2​‖un−T​vn−xn+T​xn‖\displaystyle-(1-k)\alpha_{n}\left\|u_{n}-x_{n}\right\|-(2-k)\alpha_{n}^{2}\left\|u_{n}-Tv_{n}-x_{n}+Tx_{n}\right\|
−αn​‖T​un+1−T​vn−T​xn+1+T​xn‖\displaystyle-\alpha_{n}\left\|Tu_{n+1}-Tv_{n}-Tx_{n+1}+Tx_{n}\right\|
≥\displaystyle\geq (1+αn)​‖un+1−xn+1‖−(1−k)​αn​‖un−xn‖\displaystyle\left(1+\alpha_{n}\right)\left\|u_{n+1}-x_{n+1}\right\|-(1-k)\alpha_{n}\left\|u_{n}-x_{n}\right\|
−(2−k)​αn2​‖un−T​vn−xn+T​xn‖−αn​‖T​un+1−T​vn−T​xn+1+T​xn‖\displaystyle-(2-k)\alpha_{n}^{2}\left\|u_{n}-Tv_{n}-x_{n}+Tx_{n}\right\|-\alpha_{n}\left\|Tu_{n+1}-Tv_{n}-Tx_{n+1}+Tx_{n}\right\| (17)

Solving the above inequality for ‖un+1−xn+1‖\left\|u_{n+1}-x_{n+1}\right\| gives

‖un+1−xn+1‖≤\displaystyle\left\|u_{n+1}-x_{n+1}\right\|\leq [1+(1−k)​αn]1+αn​‖un−xn‖\displaystyle\frac{\left[1+(1-k)\alpha_{n}\right]}{1+\alpha_{n}}\left\|u_{n}-x_{n}\right\|
+(2−k)​αn21+αn​‖un−T​vn−xn+T​xn‖\displaystyle+\frac{(2-k)\alpha_{n}^{2}}{1+\alpha_{n}}\left\|u_{n}-Tv_{n}-x_{n}+Tx_{n}\right\|
+αn1+αn​‖T​un+1−T​vn−T​xn+1+T​xn‖\displaystyle+\frac{\alpha_{n}}{1+\alpha_{n}}\left\|Tu_{n+1}-Tv_{n}-Tx_{n+1}+Tx_{n}\right\| (18)
≤\displaystyle\leq [1+(1−k)​αn]1+αn​‖un−xn‖+(2−k)​αn2​‖un−T​vn‖\displaystyle\frac{\left[1+(1-k)\alpha_{n}\right]}{1+\alpha_{n}}\left\|u_{n}-x_{n}\right\|+(2-k)\alpha_{n}^{2}\left\|u_{n}-Tv_{n}\right\|
+(2−k)​αn2​‖T​xn−xn‖+αn​‖T​un+1−T​vn‖\displaystyle+(2-k)\alpha_{n}^{2}\left\|Tx_{n}-x_{n}\right\|+\alpha_{n}\left\|Tu_{n+1}-Tv_{n}\right\|
+αn​‖T​xn+1−T​xn‖\displaystyle+\alpha_{n}\left\|Tx_{n+1}-Tx_{n}\right\|
‖un−T​vn‖≤\displaystyle\left\|u_{n}-Tv_{n}\right\|\leq ‖un−xn‖+‖xn−T​xn‖+‖T​xn−T​vn‖\displaystyle\left\|u_{n}-x_{n}\right\|+\left\|x_{n}-Tx_{n}\right\|+\left\|Tx_{n}-Tv_{n}\right\| (19)

Let LL denote the Lipschitz constant for TT. Then,

‖T​xn−T​vn‖\displaystyle\left\|Tx_{n}-Tv_{n}\right\| ≤L​‖xn−vn‖\displaystyle\leq L\left\|x_{n}-v_{n}\right\| (20)
‖vn−xn‖\displaystyle\left\|v_{n}-x_{n}\right\| =‖(1−βn)​un+βn​T​un−xn‖\displaystyle=\left\|\left(1-\beta_{n}\right)u_{n}+\beta_{n}Tu_{n}-x_{n}\right\|
≤(1−βn)​‖un−xn‖+βn​‖T​un−xn‖\displaystyle\leq\left(1-\beta_{n}\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}\left\|Tu_{n}-x_{n}\right\|
≤(1−βn)​‖un−xn‖+βn​[‖T​un−T​xn‖+‖T​xn−xn‖]\displaystyle\leq\left(1-\beta_{n}\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}\left[\left\|Tu_{n}-Tx_{n}\right\|+\left\|Tx_{n}-x_{n}\right\|\right] (21)
≤(1−βn)​‖un−xn‖+βn​L​‖un−xn‖+βn​‖T​xn−xn‖\displaystyle\leq\left(1-\beta_{n}\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}L\left\|u_{n}-x_{n}\right\|+\beta_{n}\left\|Tx_{n}-x_{n}\right\|
=(1−βn+βn​L)​‖un−xn‖+βn​‖T​xn−xn‖\displaystyle=\left(1-\beta_{n}+\beta_{n}L\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}\left\|Tx_{n}-x_{n}\right\|

Note that, for L≥1,1−βn+βn​L≤LL\geq 1,1-\beta_{n}+\beta_{n}L\leq L. Substituting (21) into (20) and then (20) into (19) gives

‖un−T​vn‖\displaystyle\left\|u_{n}-Tv_{n}\right\| ≤‖un−xn‖+‖xn−T​xn‖+L​[L​‖un−xn‖+βn​‖T​xn−xn‖]\displaystyle\leq\left\|u_{n}-x_{n}\right\|+\left\|x_{n}-Tx_{n}\right\|+L\left[L\left\|u_{n}-x_{n}\right\|+\beta_{n}\left\|Tx_{n}-x_{n}\right\|\right]
=(1+L2)​‖un−xn‖+(1+L​βn)​‖T​xn−xn‖\displaystyle=\left(1+L^{2}\right)\left\|u_{n}-x_{n}\right\|+\left(1+L\beta_{n}\right)\left\|Tx_{n}-x_{n}\right\| (22)
‖T​un+1−T​vn‖\displaystyle\left\|Tu_{n+1}-Tv_{n}\right\| ≤L​‖un+1−vn‖=L​‖(1−αn)​un+αn​T​vn−vn‖\displaystyle\leq L\left\|u_{n+1}-v_{n}\right\|=L\left\|\left(1-\alpha_{n}\right)u_{n}+\alpha_{n}Tv_{n}-v_{n}\right\|
≤L​[(1−αn)​‖un−vn‖+αn​‖T​vn−vn‖]\displaystyle\leq L\left[\left(1-\alpha_{n}\right)\left\|u_{n}-v_{n}\right\|+\alpha_{n}\left\|Tv_{n}-v_{n}\right\|\right] (23)

Using (21),

‖T​vn−vn‖\displaystyle\left\|Tv_{n}-v_{n}\right\| ≤‖T​vn−T​xn‖+‖T​xn−xn‖+‖xn−vn‖\displaystyle\leq\left\|Tv_{n}-Tx_{n}\right\|+\left\|Tx_{n}-x_{n}\right\|+\left\|x_{n}-v_{n}\right\|
≤(1+L)​‖xn−vn‖+‖T​xn−xn‖\displaystyle\leq(1+L)\left\|x_{n}-v_{n}\right\|+\left\|Tx_{n}-x_{n}\right\|
≤(1+L)​[L​‖xn−un‖+βn​‖T​xn−xn‖]+‖T​xn−xn‖\displaystyle\leq(1+L)\left[L\left\|x_{n}-u_{n}\right\|+\beta_{n}\left\|Tx_{n}-x_{n}\right\|\right]+\left\|Tx_{n}-x_{n}\right\| (24)
=(1+L)​L​‖xn−un‖+[(1+L)​βn+1]​‖T​xn−xn‖\displaystyle=(1+L)L\left\|x_{n}-u_{n}\right\|+\left[(1+L)\beta_{n}+1\right]\left\|Tx_{n}-x_{n}\right\|
‖un−vn‖\displaystyle\left\|u_{n}-v_{n}\right\| =‖un−(1−βn)​un−βn​T​un‖=βn​‖(un−T​un)‖\displaystyle=\left\|u_{n}-\left(1-\beta_{n}\right)u_{n}-\beta_{n}Tu_{n}\right\|=\beta_{n}\left\|\left(u_{n}-Tu_{n}\right)\right\|
≤βn​[‖un−xn‖+‖xn−T​xn‖+‖T​xn−T​un‖]\displaystyle\leq\beta_{n}\left[\left\|u_{n}-x_{n}\right\|+\left\|x_{n}-Tx_{n}\right\|+\left\|Tx_{n}-Tu_{n}\right\|\right] (25)
≤βn​[(1+L)​‖xn−un‖+‖xn−T​xn‖]\displaystyle\leq\beta_{n}\left[(1+L)\left\|x_{n}-u_{n}\right\|+\left\|x_{n}-Tx_{n}\right\|\right]

Substituting (25) and (24) into (23), we obtain

‖T​un+1−T​vn‖≤\displaystyle\left\|Tu_{n+1}-Tv_{n}\right\|\leq L​(1−αn)​[βn​(1+L)​‖un−xn‖+βn​‖T​xn−xn‖]\displaystyle L\left(1-\alpha_{n}\right)\left[\beta_{n}(1+L)\left\|u_{n}-x_{n}\right\|+\beta_{n}\left\|Tx_{n}-x_{n}\right\|\right]
+αn​L​[(1+L)​L​‖xn−un‖+[(1+L)​βn+1]​‖T​xn−xn‖]\displaystyle+\alpha_{n}L\left[(1+L)L\left\|x_{n}-u_{n}\right\|+\left[(1+L)\beta_{n}+1\right]\left\|Tx_{n}-x_{n}\right\|\right]
≤\displaystyle\leq [L​(1−αn)​βn​(1+L)+αn​L2​(1+L)]​‖xn−un‖\displaystyle{\left[L\left(1-\alpha_{n}\right)\beta_{n}(1+L)+\alpha_{n}L^{2}(1+L)\right]\left\|x_{n}-u_{n}\right\|} (26)
+{βn​L​(1−αn)+αn​L​[(1+L)​βn+1]}​‖xn−T​xn‖\displaystyle+\left\{\beta_{n}L\left(1-\alpha_{n}\right)+\alpha_{n}L\left[(1+L)\beta_{n}+1\right]\right\}\left\|x_{n}-Tx_{n}\right\|

Substituting (22) and (26) into (18) and using (1+αn)−1≤1−αn+αn2\left(1+\alpha_{n}\right)^{-1}\leq 1-\alpha_{n}+\alpha_{n}^{2}, yields

‖xn+1−un+1‖≤\displaystyle\left\|x_{n+1}-u_{n+1}\right\|\leq (1+(1−k)​αn)​(1−αn+αn2)​‖xn−un‖\displaystyle\left(1+(1-k)\alpha_{n}\right)\left(1-\alpha_{n}+\alpha_{n}^{2}\right)\left\|x_{n}-u_{n}\right\|
+(2−k)​αn2​[(1+L2)​‖xn−un‖+(1+L​βn)​‖T​xn−xn‖]\displaystyle+(2-k)\alpha_{n}^{2}\left[\left(1+L^{2}\right)\left\|x_{n}-u_{n}\right\|+\left(1+L\beta_{n}\right)\left\|Tx_{n}-x_{n}\right\|\right]
+(2−k)​αn2​‖xn−T​xn‖\displaystyle+(2-k)\alpha_{n}^{2}\left\|x_{n}-Tx_{n}\right\|
+αn​[L​(1−αn)​βn​(1+L)+αn​L2​(1+L)]​‖xn−un‖\displaystyle+\alpha_{n}\left[L\left(1-\alpha_{n}\right)\beta_{n}(1+L)+\alpha_{n}L^{2}(1+L)\right]\left\|x_{n}-u_{n}\right\| (27)
+αn​{βn​L​(1−αn)+αn​L​[(1+L)​βn+1]}​‖xn−T​xn‖\displaystyle+\alpha_{n}\left\{\beta_{n}L\left(1-\alpha_{n}\right)+\alpha_{n}L\left[(1+L)\beta_{n}+1\right]\right\}\left\|x_{n}-Tx_{n}\right\|
+αn​L​‖xn+1−xn‖\displaystyle+\alpha_{n}L\left\|x_{n+1}-x_{n}\right\|
≤\displaystyle\leq γn​‖xn−un‖+δn​‖xn−T​xn‖+αn​‖xn+1−xn‖\displaystyle\gamma_{n}\left\|x_{n}-u_{n}\right\|+\delta_{n}\left\|x_{n}-Tx_{n}\right\|+\alpha_{n}\left\|x_{n+1}-x_{n}\right\|

where

δn=\displaystyle\delta_{n}= αn​[(2−k)​(2+L​βn)​αn+[βn​L​(1−αn)+αn​L​[(1+L)​βn+1]]],\displaystyle\alpha_{n}\left[(2-k)\left(2+L\beta_{n}\right)\alpha_{n}+\left[\beta_{n}L\left(1-\alpha_{n}\right)+\alpha_{n}L\left[(1+L)\beta_{n}+1\right]\right]\right],
γn=\displaystyle\gamma_{n}= [1+(1−k)​αn]​(1−αn+αn2)+(2−k)​(1+L2)​αn2\displaystyle{\left[1+(1-k)\alpha_{n}\right]\left(1-\alpha_{n}+\alpha_{n}^{2}\right)+(2-k)\left(1+L^{2}\right)\alpha_{n}^{2}} (28)
+αn​L​(1+L)​[βn​(1−αn)+L​αn].\displaystyle+\alpha_{n}L(1+L)\left[\beta_{n}\left(1-\alpha_{n}\right)+L\alpha_{n}\right].

Note that

(1+(1−k)​αn)​(1−αn+αn2)\displaystyle\left(1+(1-k)\alpha_{n}\right)\left(1-\alpha_{n}+\alpha_{n}^{2}\right) =1−k​αn+k​αn2+(1−k)​αn3\displaystyle=1-k\alpha_{n}+k\alpha_{n}^{2}+(1-k)\alpha_{n}^{3}
≤1−k​αn+k​αn2+(1−k)​αn2\displaystyle\leq 1-k\alpha_{n}+k\alpha_{n}^{2}+(1-k)\alpha_{n}^{2} (29)
=1−k​αn+αn2\displaystyle=1-k\alpha_{n}+\alpha_{n}^{2}

Therefore,

γn≤1−kαn+αn{2αn+(2−k)(1+L2)αn\displaystyle\gamma_{n}\leq 1-k\alpha_{n}+\alpha_{n}\{2\alpha_{n}+(2-k)\left(1+L^{2}\right)\alpha_{n}
+L(1+L)[βn(1−αn)+Lαn]}\displaystyle\left.+L(1+L)\left[\beta_{n}\left(1-\alpha_{n}\right)+L\alpha_{n}\right]\right\}
=1−kαn+αn{[2+(2−k)(1+L2)+L2(1+L)]αn\displaystyle=1-k\alpha_{n}+\alpha_{n}\left\{\left[2+(2-k)\left(1+L^{2}\right)+L^{2}(1+L)\right]\alpha_{n}\right. (30)
+L(1+L)(1−αn)βn}\displaystyle\left.+L(1+L)\left(1-\alpha_{n}\right)\beta_{n}\right\}
≤1−k​αn+αn​M​(αn+βn)\displaystyle\leq 1-k\alpha_{n}+\alpha_{n}M\left(\alpha_{n}+\beta_{n}\right)

where M=2+(2−k)​(1+L2)+L2​(1+L)M=2+(2-k)\left(1+L^{2}\right)+L^{2}(1+L).
Since αn\alpha_{n} and βn\beta_{n} satisfy (i), there exists an integer NN such that M​(αn+βn)≤k​(1−k)M\left(\alpha_{n}+\beta_{n}\right)\leq k(1-k) for all n≥Nn\geq N.

Thus,

‖xn+1−un+1‖≤\displaystyle\left\|x_{n+1}-u_{n+1}\right\|\leq (1−k2​αn)​‖xn−un‖\displaystyle\left(1-k^{2}\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|
+αn{[(2−k)(2+Lβn)αn\displaystyle+\alpha_{n}\left\{\left[(2-k)\left(2+L\beta_{n}\right)\alpha_{n}\right.\right.
+[βnL(1−αn)+αnL((1+L)βn+1)]]∥xn−Txn∥\displaystyle\left.+\left[\beta_{n}L\left(1-\alpha_{n}\right)+\alpha_{n}L\left((1+L)\beta_{n}+1\right)\right]\right]\left\|x_{n}-Tx_{n}\right\| (31)
+L∥xn+1−xn∥}\displaystyle\left.+L\left\|x_{n+1}-x_{n}\right\|\right\}

With λn:=k2​αn,an=‖xn−un‖\lambda_{n}:=k^{2}\alpha_{n},a_{n}=\left\|x_{n}-u_{n}\right\|, and ϵn=\epsilon_{n}= the quantity in braces, we have

an+1≤(1−λn)​an+ϵn​λna_{n+1}\leq\left(1-\lambda_{n}\right)a_{n}+\epsilon_{n}\lambda_{n} (32)

Since xn→px_{n}\rightarrow p and TT is Lipschitzian, then is continuous. Therefore, limT​xn=T​p=p\lim Tx_{n}=Tp=p and ‖T​xn−xn‖→0\left\|Tx_{n}-x_{n}\right\|\rightarrow 0 as n→∞n\rightarrow\infty. Also lim‖xn+1−xn‖=0\lim\left\|x_{n+1}-x_{n}\right\|=0. Thus, the conditions of Lemma 2 are satisfied and lim‖xn−un‖=0\lim\left\|x_{n}-u_{n}\right\|=0.

Consequently,

‖un−p‖≤‖un−xn‖+‖xn−p‖⟶0 as ​n→∞,\left\|u_{n}-p\right\|\leq\left\|u_{n}-x_{n}\right\|+\left\|x_{n}-p\right\|\longrightarrow 0\quad\text{ as }n\rightarrow\infty, (33)

and the Ishikawa method converges.
Using the argument in [2], it follows that we also have corresponding theorem for Lipschitz strictly hemicontractive operators.

Let S:X→XS:X\rightarrow X be a Lipschitz operator with L>1L>1. It is well known that the operator S:X→XS:X\rightarrow X is strongly accretive if and only if ( I−SI-S ) is strongly pseudocontractive operator, and conversely. Consider the equation S​x=fSx=f, where f∈Xf\in X is given and SS is a strongly accretive operator. A fixed point for T​x=f+(I−S)​xTx=f+(I-S)x will be the solution of S​x=fSx=f, and conversely. If we consider in (1) and (2) the operator T​x=f+(I−S)​xTx=f+(I-S)x, then TT will be strongly pseudocontractive. Theorem 4 assures the equivalence between the convergencies of Mann and Ishikawa iteration. We consider equation x+S​x=fx+Sx=f, with SS an accretive operator, that is, ( I−SI-S ) is a pseudocontractive operator. A solution for x+S​x=fx+Sx=f is a fixed point for T​x=f−S​xTx=f-Sx, which is a strongly pseudocontractive operator. Replacing (1) and (2), we obtain the equivalence between the convergencies of Mann and Ishikawa iteration for an accretive operator. The solutions existences in the above two equations hold as in [2].

All our results hold for multivalued operators provided that this admit appropriate single-valued selections.

It has been shown in [1] that there exists a Lipschitzian pseudocontractive map defined on a compact subset of ℝ2\mathbb{R}^{2} for which an Ishikawa method, with αn≤βn\alpha_{n}\leq\beta_{n}, converges to a fixed point, but for which no Mann iterative method converges. Therefore, it is not possible to extend Theorem 4 to pseudocontractive maps.

References

[1] C. E. Chidume and S. A. Mutangadura, An example of the Mann iteration method for Lipschitz pseudocontractions, Proc. Amer. Math. Soc. 129 (2001), no. 8, 2359-2363.
[2] C. E. Chidume and M. O. Osilike, Nonlinear accretive and pseudo-contractive operator equations in Banach spaces, Nonlinear Anal. 31 (1998), no. 7, 779789.
[3] L. B. Ćirić, A generalization of Banach’s contraction principle, Proc. Amer. Math. Soc. 45 (1974), 267-273.
[4] K. Deimling, Zeros of accretive operators, Manuscripta Math. 13 (1974), 365-374.
[5] S. Ishikawa, Fixed points by a new iteration method, Proc. Amer. Math. Soc. 44 (1974), 147-150.
[6] T. Kato, Nonlinear semigroups and evolution equations, J. Math. Soc. Japan 19 (1967), 508-520.
[7] W. R. Mann, Mean value methods in iteration, Proc. Amer. Math. Soc. 4 (1953), 506-510.
[8] B. E. Rhoades, Fixed point iterations using infinite matrices, Trans. Amer. Math. Soc. 196 (1974), 161-176.
[9] _​_​_​_\_\_\_\_ , Comments on two fixed point iteration methods, J. Math. Anal. Appl. 56 (1976), no. 3, 741-750.
[10] _​_​_​_\_\_\_\_ , A comparison of various definitions of contractive mappings, Trans. Amer. Math. Soc. 226 (1977), 257-290.
[11] X. Weng, Fixed point iteration for local strictly pseudo-contractive mapping, Proc. Amer. Math. Soc. 113 (1991), no. 3, 727-731.
[12] H. K. Xu, A note on the Ishikawa iteration scheme, J. Math. Anal. Appl. 167 (1992), no. 2, 582-587.
B. E. Rhoades: Department of Mathematics, Indiana University, Bloomington, IN 47405-7106, USA

E-mail address: rhoades@indiana.edu
Stefan M. Soltuz: "T. Popoviciu" Institute of Numerical Analysis, P.O. Box 68-1, ClujNapoca 3400, Romania

E-mail address: ssoltuz@ictp-acad.math.ubbcluj.ro

2003

Related Posts