Seminar on Fixed Point Theory Cluj-Napoca, Volume 3, 2002, 209-216 http://www.math.ubbcluj.ro/ nodeacj/journal.

htm

ITERATIVE APPROXIMATION OF FIXED POINTS FOR PSEUDO-CONTRACTIVE OPERATORS

VASILE BERINDE Department of Mathematics and Computer Science Faculty of Sciences North University of Baia Mare Abstract. For a nonexpansive (Liptschitzian) operator F which is not a strict contraction, the Picard iteration does not converge generally to a xed point. By adding some pseudo-contractive type assumptions, it is possible to show that several other iterations (Krasnoselski-Schaefer, Mann, Ishikawa) converge to a xed point of F . The main aim of the paper is to survey some old and recent results especially related to the convergence of Krasnoselski-Schaefer iteration in the class of generalized pseudo-contractive and lipschitzian operators.

1. Introduction Many of the most important nonlinear problems of applied mathematics reduce to nding solutions of nonlinear functional equations (integral equations, dierential equations etc.) which can be formulated in terms of nding the xed points of a given nonlinear operator of an innite dimensional function space X into itself: (1) x = Tx There is a classical general existence theory of xed points for mappings satisfying compactness conditions associated with the names of Brower, Schander, Leray etc. as well as an abundant literature of metrical xed point theorems, which establish the existence (and uniqueness) of xed points for maps satisfying a variety of contractive conditions ([16]). The rst basic result is the classical Picard-Banach-Caccioppoli principle THEOREM 0. Let (X, d) be a complete metric space and T : X X a strict contraction, that is, there exists , 0 < 1 such that (2) by (3) xn = T (xn1 ) = T n (x0 ), n = 1, 2, ...
209

d(T x, T y ) d(x, y ) for all x, y X

Then the Picard iteration (the sequence of successive approximation) (xn ), given

210

VASILE BERINDE

converges to the unique xed point x of T, xn x (as n ). The convergence order of the Picard iteration in Theorem 0 is given by both a priori and a posteriori estimates: (4) (4) d(xn , x ) d(xn , x ) n d(x0 , x1 ), n 1; 1

d(xn , xn1 ), n 1; 1 This fundamental result in the xed point theory has been extended to some larger classes of contractive and generalized contractive operators, see[16], for example, by replacing the strict contractive condition (2) by a weaker condition of the following type (2) d(T x, T y ) (d(x, y )), x, y X

where : R+ R+ is a certain compassion function or by a more general one d(T x, T y ) (d(x, y )), d(x, T x), d(x, T y ), d(y, T x), d(y, T y )), x, y X, where : R5 + R+ stands for a certain 5-dimensional comparison function (see [2], [21]). Important results as Kannan or Ciric theorems belong to the second class of generalized contractions [16]. If the generalized contractive condition in a such xed point theorem is strong enough, that is, the comparison function satises some essential conditions, then the (possible unique) xed point of T can be obtained by means of the sequence of successive approximations (Picard iteration). But, if the contractive condition is slightly weaker, as for example the case whenT is only nonexpansive (2) d(T x, T y ) d(x, y ), x, y X,

then the Picard iteration (3) need no longer converge to a xed point of T (if any). In fact, in general, a nonexpansive operator T need not have a xed point and even if T possesses a xed point, the Picard iteration need not converge to this point, see Example 1 in Section 2. Even if the xed point can be obtained by the Picard iteration, it is of interest to determine other iteration procedures that could converge faster, in a certain sense, to the xed point. 2. Other fixed point iteration procedures The next example illustrates the case of a nonexpansive operator possessing an unique xed point, for which the Picard iteration does not converge to that xed point, except for the case when the initial approximation coincides to the xed point itself.

ITERATIVE APPROXIMATION OF FIXED POINTS

211

EXAMPLE 1. ([9]) Let X = [0, 1] and T : [0, 1] [0, 1] be the linear function T (x) = 1 x, 0 x 1. a) T is nonexpansive; b) T has an unique xed point

Then :

1 FT = {x [0, 1]/T (x) = x} = { }. 2 c) The Picard iteration xn = T (xn1 ), n = 1, 2, ..., yields the oscilla1 tory sequence a, 1 a, a, 1 a, a, ...for any x0 = a(a = 2 ). Taking into account the fact that the class of nonexpansive operators is very important in applications, we have to impose certain additional conditions on the ambient space or on the operator itself, in order to ensure the existence of a xed point or to guarantee the convergence of the Picard iteration to a xed point of the operator. In what concern the existence problem, the following result was obtained independently by Browder, Gohde and Kirk (see [3], for example): THEOREM 1. Let C be a closed, bounded, and convex subset of a uniformly convex Banach space, T : C C a nonexpansive map. Then T has a xed point. The proofs of Theorem 1 are, unfortunately, all not constructive. As shown by Example 1, the Picard iteration does not converge (to the xed point) of such a nonexpansive operator. A similar situation is encountered for the class of lipschitzian (and pseudocontractive, in some sense) operators, when, even if the xed point is unique, the Picard iteration does not converge (see [1]). To remove these diculties we need to consider some other sequential procedures to be used for approximating xed points. We present here a few chronological reference points. In 1953 W. R. Mann [17] introduced an iteration procedure which can be represented in the following form (5) xn+1 = (1 an )xn + an T xn , n = 0, 1, 2, ...

where (an ) is sequence satisfying: (i) an [0, 1] and (ii) an = . Mann showed that, if T is an continuous selfmap of a closed interval [a, b] with at most one xed point, then the iteration scheme (5), with an = 1/(n + 1),converges to the xed point of T. In 1955, Krasnoselski [18] showed that, if X is a uniformly convex Banach space, and T : X X is nonexpansive, then the iteration (xn ), 1 (6) xn+1 = (xn + T xn ), n = 0, 1, 2, ... . 2 converges to a xed point of T. Later, in 1957, Schaefer considered the extension of (6), by replacing 1/2 by a constant [0, 1], that is, he introduced the iteration procedure (xn ), (6) xn+1 = (1 )xn + T xn , n = 0, 1, 2, ... and proved similar results [19].

212

VASILE BERINDE

It is easy to see that (xn ) given by (6) is in fact the Picard iteration for the associated operator U = (1 ) I + T, where I is the identity. If we take = 1 2 , from (6) we obtain (6), and, by putting = 1, from (6) we obtain the Picard iteration. Iteration (6), which will be called in the sequel KrasnoselskiSchaefer iteration, is a particular case of the Mann iteration (5), obtained from the last one for an = (const). It was proven for continuous mappings that, if the Mann iterative process converges, then it must converge to a xed point of T . But if T is not continuous, there is no guarantee that, even if the Mann process converges, it will converge to a xed point of T , as shown by the next example. EXAMPLE 2. Let X = [0, 1] and T : X X the map dened by T 0 = T 1 = 0 and T x = 1, for x (0, 1). Then T is a selfmap of [0, 1],having the unique xed point x = 0. However, the Mann iteration, with an = 1/(n + 1) and x0 (0, 1) converges to 1, which is not a xed point of T . By adding some pseudocontractive hypotheses to a nonexpansive map, Ishikawa [7] proved the following result. THEOREM 3. Let E be a convex, compact subset of a Hilbert space H, T : E E a lipschitzian and pseudocontractive map. Then the sequence (xn ), dened by (7) xn+1 = (1 n ) xn + n T [(1 n )xn + n T xn ], n = 0, 1, 2, ...

where (n ), (n ) are sequences of positive numbers satisfying the conditions 0 n n 1, lim n = 0 and n n = , converges strongly to a xed point of T. The next diagram represents the most important iteration procedures considered here xn+1 = T xn , n 0 =1 1890 Picard xn+1 = 1 2 (xn + T xn ), n 0 1955, Krasnoselski =
1 2

xn+1 = (1 )xn + T xn , n 0, 0 1 an = (const)

1957 (Krasnoselski-)Schaeer

xn+1 = (1 an ) xn + an T xn , n 0, an [0, 1] bn = 0

1953

Mann

xn+1 = (1 an ) xn + an T [(1 bn )xn + bn T xn ], n 0, 0 an , bn 1 1974 Ishikawa

ITERATIVE APPROXIMATION OF FIXED POINTS

213

3. Pseudo-contractive operators The interest in pseudocontractive mappings is due mainly to a) their usefulness as an additional assumption to Lipschitz type conditions in proving convergence of xed point iterative procedures; b) their connection with the important class of nonlinear accretive operators. Let H be a Hilbert space and T : H H a selfmap. Denition 1. T is said to be pseudocontractive on C H if (10) Tx Ty x y
2

+ T x T y (x y )

x, y C

or equivalently (10) T x T y, x y x y
2

x, y C.

Denition 2. T is called generalized pseudocontractive, if r > 0 such that (11) Tx Ty

2

r xy

+ T x T y r(x y )

x, y C,

or equivalently (11) T x T y, x y r x y
2

x, y C

Remark. For r = 1, from Denition 2 we obtain Denition 1. Denition 3. The operator T is said to be strictly pseudocontractive on C if there exists k < 1 such that (12) Tx Ty
2

xy

+ k T x T y (x y )

x, y C,

Dention 4. The operator T is called strongly pseudocontractive on C if there exist t > 1 and r > 0 such that xy
2

(1 + r)(x y ) rt(T x T y ) ,

x, y C.

Remarks. 1) For t = 1, from Denition 4 we obtain Denition 1; 2) T is pseudo-conttractive if and only if I T is accretive ([21]). The class of pseudo-contractive operators usually associated with lipschitzian properties has been studied extensively by various authors, see [22] and references therein. The following result has been proved by Verma [12]. THEOREM 4. Let K be a non-empty closed convex subset of H and T : K K a lipschitzian and generalized pseudocontractive operator (with constant s and r, respectively, r < 1). r ) Then, for any , 0 < < 12(1 2r +s2 the iteration (xn ),given by xn+1 = (1 )xn + T xn ,

n0

converges strongly to the unique xed point x of T : xn x (n ).

214

VASILE BERINDE

In our paper [1] we completed the result of Verma by inserting both a priori and a posteriori error estimates (Theorem 3.1) and by nding the fastest iteration amongst all Krasnoleski-Schafer iterations. Moreover, when the Picard iteration and Krasnoselski-Schaefer iteration converge simultaneously, it is possible to compare the fastest one, in a certain sense of rate of convergence. Generally, it is possible to nd the fastest Krasnoselski-Schaefer iteration in the family, as shown by THEOREM 5 ([1]). Let K be a non-empty closed convex subset of a Hilbert space and T : K K a generalized pseudocontractive (with constant r, 0 < r < 1) and r ) lipschitzian (with constant s > 0) operator and (0, 1) such that 0 < < 12(1 2r +s2 . Then (i) T has an unique xed point x K ; (ii) The Krasnoselski-Schaefer iteration (xn ) converges strongly to x , for each x0 K ; n (iii) xn x 1 x1 x0 , n 1 xn x 1 xn xn1 , n 1 where = (1 )2 + 2(1 )r + 2 s2 . (iv) The fastest Krasnoselski-Schaefer iteration is obtained for 1r 0 = . 1 2r + s2 EXAMPLE 3.([1]) Let K = 1 2 , 2 and T : K K the function dened by 1 T (x) = x ,x 1 , 2 . Then 2 1) T is Lipschitzian with constant s = 4; 2) T is generalized pseudocontractive with constant r > 0 arbitrary; 3) The Picard iteration, with x0 = a = 1yields the oscillatory sequence 1 1 a, , a, , .... a a 4) The Krasnoselki-Schaefer iteration converges to x = 1, for any r ) 0, 2(1 172r , r < 1; 5) For r = 0, 5, the fastest Krasnoselski-Schaefer iteration is obtained taking 0 = 1 , 32 i.e. xn+1 = 1 32 31xn + 1 xn , n0
63 8

which converges to x = 1, (slowly, because the contraction coecient 0 = 0, 992 is very close to 1). Final remarks

The class of pseudo-contractive operators has been intensively studied in the last decade (see, for example, [4], [5]). We considered a more general class of pseudocontractive operators in [20].

ITERATIVE APPROXIMATION OF FIXED POINTS

215

References
[1] BERINDE, V., The fastest iteration for approximating xed points of lipschitzian pseudocontractions, Proceed. 2nd Palestinian Int. Math. Conference (in print). [2] BERINDE, V., Generalized contractions and applications (in Romanian), Editura CUB PRESS 22, Baia Mare, 1997. [3] BROWDER, F.E., PETRYSHIN, W.V., Construction of xed points of nonlinear mappings in Hilbert spaces, J. Math. Anal. Appl. 20(1967), 197-228. [4] CHIDUME, C.E., Global iteration schemes for strongly pseudo-contractive maps, Proc. Amer. Mat. Soc. 126(1998), 9, 2641-2649. [5] CHIDUME, C.E., Approximation of xed points of strongly pseudocontractive mappings, Proc. Amer. Mat. Soc. 120(1994), 2, 515551. [6] GHOSH, M.K., DEBNATH, L., Convergence of Ishikawa iterates of quasi-nonexpansive mappings, J. Math. Anal. Appl. 207(1997), 96-103. [7] ISHIKAWA, S., Fixed points by a new iteration method, Proc. Amer. Mat. Soc. 44(1974), 1, 147150. [8] PETRYSHYN, W.V., WILIAMSON, T.E., Strong and weak Convergence of the Sequence of Successive Approximations for Quasi-Nonexpansive Mappings, J. Math. Anal. Appl. 43(1973), 459-497. [9] RHOADES, B.E., Some xed point iteration procedure, Internat. J. Math.&Math.Sci., 14(1991), 1, 1-16. [10] RHOADES, B.E., A general principle for Mann iterations, Indian. J. Pure Appl. Math. 26(8), 1995, 751-762. [11] RHOADES, B.E., Convergence of an Ishikawa-Type Iteration Scheme for a Generalized Contraction, J. Math.Anal. Appl., 185(1994), No. 2, 350-355. [12] VERMA, R.U., A xed point theorem involving lipschitzian generalized pseudo-contractions, Proceed. Royal Irish., 97A(1997), No.1, 83-86. [13] VERMA, R.U., Strongly nonlinear variational inequalities and generalized pseudocontractions, Rend. Circ. Mat. Palermo, XLVIII(1999), 201-208. [14] BERINDE, V., On some stability results for xed point iterative procedures, (submitted). M [15] PICARD, E., emoire sur la th eorie des equations aux deriv ee partielles et la metode des approximations successives, J. Math. Pures. Appl., 6(1890), 145-210. [16] RUS, A.I., Generalized contractions and applications, Cluj University Press, Cluj-Napoca, 2001. [17] MANN, W.R., Mean value methods in iteration, Proc. Amer. Mat. Soc. 4(1953), 506510. [18] KRASNOSELSKI, M.A., Two remarks on the method of successive approximations, Uspechi Mat. Nauk. 10(1955), No. 1(63), 123-127. [19] SCHAEFER, H., Uber die Methode Sukzessiver Approximation, Jber. Deutsch Math.-Verein, 59(1957), 131-140. [20] BERINDE, V., On the convergence of Krasnoselski-Schaefer iteration for pseudo-contractions (in preparation). [21] BROWDER, F.E., Nonlinear mappings of nonexpansive and accretive type in Banach spaces, Bull. Amer. Math. Soc. 73(1967), 875-882. [22] CHIDUME, C.E., Global iteration schemes for strongly pseudocontractive maps, Proc. Math. Soc. 126(1998), 2641-2649.