Skip to main content

Advertisement

Nonexpansive semigroups in CAT(κ) spaces

Article metrics

  • 901 Accesses

Abstract

The main purpose of this paper is to study Browder type convergence theorems for a nonexpansive semigroup with geometric approaches in a CAT(κ) space. Besides, we determine a necessary and sufficient condition for convergence of a Browder type iteration associated to a uniformly asymptotically regular nonexpansive semigroup on the unit sphere in an infinite-dimensional Hilbert space.

MSC:47H20, 47H10.

1 Introduction

Let (X,d) be a metric space, C a closed convex subset of X and T:CC a mapping. Recall that T is nonexpansive on C if d(Tx,Ty)d(x,y), for all x,yC. We denote by F(T) the fixed point set of the mapping T. A one-parameter family S={T(t):t0} of self-mappings of C is called a strongly continuous nonexpansive semigroup on C if the following conditions are satisfied:

  1. (i)

    for each t0, T(t) is a nonexpansive mapping on C;

  2. (ii)

    T(0)x=x, for all xC;

  3. (iii)

    T(s+t)=T(s)T(t), for all s,t0;

  4. (iv)

    for each xC, the mapping tT(t)x from [0,) into C is continuous.

Let F ( S ) denote the common fixed point set of all mappings in S.

There have been considerably many interesting results of iterative methods for approximating fixed points of nonexpansive mappings, nonexpansive semigroups, and their generalizations which solve some variational inequalities problems due to their various applications in several physical problems, such as in operations research, economics, and engineering design; see, e.g., [14] and the references therein.

Suppose that X is a real Hilbert space and u is an arbitrary point of X. If T is nonexpansive on C, then for each α(0,1) there exists a unique x α C such that x α =αu+(1α)T x α because the mapping xαu+(1α)Tx is a contraction. In 1967, Browder [5] was the pioneer to consider an implicit scheme and prove the following strong convergence theorem of this algorithm in a Hilbert space.

Theorem 1.1 Let C be a bounded closed convex subset of a Hilbert space and T a nonexpansive mapping on C. Let u be an arbitrary point of C and define x α C by

x α =αu+(1α)T x α ,for α(0,1).

Then as α0, { x α } converges strongly to a point of F(T) nearest to u.

The extension work of Browder’s type convergence theorems has been tremendously studied for not only one single nonexpansive mapping, but, most significantly, a semigroup of nonexpansive mappings; see, e.g., [6, 7] and the references therein.

This paper is devoted to studying Browder’s type iterations for a nonexpansive semigroup in a CAT(κ) space, where κR, which is a specific type of metric space. Intuitively, triangles in a CAT(κ) space are ‘slimmer’ than corresponding ‘model triangles’ in a standard space of constant curvature κ (see Section 2). Complete CAT(0) spaces are often called Hadamard spaces. A few recent new convergence results of classical iterations on CAT(κ) spaces with κ>0 are obtained; see, e.g., [810] and the references therein.

In [11], Dhompongsa et al. extended Suzuki’s result [7], Theorem 3] on common fixed points of a nonexpansive semigroup in a Hilbert space to a complete CAT(0) space.

Theorem 1.2 ([11], Theorem 4.4])

Let C be a bounded closed convex subset of a CAT(0) space, S={T(t):t0} a strongly continuous nonexpansive semigroup on C, and two sequences { α n }(0,1), { t n }(0,) such that lim n t n = lim n α n / t n =0. Choose arbitrarily a point x 0 C and for each nN let x n be the fixed point of the mapping x α n x 0 (1 α n )T( t n )x. Then F ( S ) and { x n } converges strongly to a point of F ( S ) nearest to x 0 .

In 2010, Acedo and Suzuki [6] proved a Browder type convergence theorem for uniformly asymptotically regular (UAR for short) nonexpansive semigroups (see Section 2) in Hilbert spaces under a weaker condition on { α n } and { t n } than that in Theorem 1.2.

Theorem 1.3 ([6], Theorem 2.3])

Let C be a closed convex subset of a Hilbert space, S={T(t):t0} a UAR and strongly continuous nonexpansive semigroup on C such that F ( S ) , and two sequences { α n }(0,1), { t n }[0,) such that lim n α n = lim n α n / t n =0. Fix x 0 C and define a sequence { x n } in C by

x n = α n x 0 +(1 α n )T( t n ) x n .

Then { x n } converges strongly to a point of F ( S ) nearest to x 0 .

The preceding two theorems lead naturally to the question of whether or not they can be extended to a CAT(κ) space with κ>0. The purpose of this article is to investigate this question with the geometric approaches in CAT(κ) spaces.

This paper is organized as follows. In Section 2 we recall the definition of geodesic metric spaces and summarize some useful lemmas and the main properties of CAT(κ) spaces. In Section 3 we present some technical results about Δ-convergence of a sequence in a complete CAT(1) space. In Section 4 we establish our main results (Theorems 4.2, 4.3, 4.4) of Browder’s iterations for nonexpansive semigroups in CAT(κ) spaces and conclude that Theorems 1.2 and 1.3 can be generalized to CAT(κ) spaces under the same respective conditions on the coefficients { α n } and { t n }. It is noteworthy that, however, without the UAR assumption, the sequence { x n } established in Theorem 1.2 is not necessarily convergent to the nearest point of F ( S ) to x 0 if lim n t n = t ˆ (0,] even when lim n α n =0 (and therefore lim n α n / t n =0); see Example 4.5. Furthermore, we determine a necessary and sufficient condition for a Browder type convergence theorem associated to a nonexpansive semigroup on the unit sphere in an infinite-dimensional Hilbert space. We also propose an open problem in Section 5.

2 Preliminaries

Let (X,d) be a metric space. For any subset E of X and xX, the diameter of E and the distance from x to E are defined, respectively, by

diam E = sup { d ( x , y ) : x , y E } , d ( x , E ) = inf { d ( x , y ) : y E } .

We always denote the open ball and the closed ball centered at x with radius r>0 by B(x,r) and B ¯ (x,r), respectively.

Let C be a closed convex subset of X and let S={T(t):t0} be a family of self-mappings of C. Then S is called

  1. (i)

    asymptotically regular on C if for any h0 and any xC,

    lim t d ( T ( h ) T ( t ) x , T ( t ) x ) =0;
  2. (ii)

    uniformly asymptotically regular (in short UAR) on C if for any h0 and any bounded subset D of C,

    lim t sup x D d ( T ( h ) T ( t ) x , T ( t ) x ) =0.

For x,yX, a geodesic path joining x to y (or a geodesic from x to y) is an isometric mapping γ:[a,b]RX such that γ(a)=x, γ(b)=y, i.e., d(γ(t),γ( t ))=|t t |, for all t, t [a,b]. Therefore d(x,y)=ba. The image of γ is called a geodesic (segment) from x to y and we shall denote a definite choice of this geodesic segment by [x,y]. We remark that composing γ with a translation (this is still an isometry), one can always choose the interval [a,b] to be [0,], where =ba. A point z=γ(t) in the geodesic [x,y] will be written as z=(1λ)xλy, where λ=(ta)/(ba), and so d(z,x)=λd(x,y) and d(z,y)=(1λ)d(x,y).

Let r>0. The metric space (X,d) is said to be

  1. (i)

    a geodesic (metric) space if any two points in X are joined by a geodesic;

  2. (ii)

    uniquely geodesic if there is exactly one geodesic joining x to y for all x,yX;

  3. (iii)

    r-geodesic space if any two points x,yX with d(x,y)<r are joined by a geodesic;

  4. (iv)

    r-uniquely geodesic if any two points x,yX with d(x,y)<r are joined by a unique geodesic in X.

A subset C of X is convex if every pair of points x,yC can be joined by a geodesic in X and the image of every such geodesic is contained in C. If this condition holds for all points x,yC with d(x,y)<r, then C is said to be r-convex.

The n-dimensional sphere S n is the set {x=( x 1 ,, x n + 1 ) R n + 1 :(xx)=1}, where () denote the Euclidean scalar product. It is endowed with the following metric: Let d S n : S n × S n R be the function that assigns to each pair (A,B) S n × S n the unique number d(A,B)[0,π] such that

cos d S n (A,B)=(AB).

Then d S n is a metric [12], I.2.1].

Definition 2.1 Given a real number κ, we denote by M κ n the following metric spaces:

  1. (i)

    if κ=0, then M 0 n is the Euclidean space R n ;

  2. (ii)

    if κ>0, then M κ n is obtained from the sphere S n by multiplying the distance function by 1/ κ ;

  3. (iii)

    if κ<0, then M κ n is obtained from the sphere H n by multiplying the distance function by 1/ κ , where H n is the hyperbolic n-space.

It is well known that M κ n is a geodesic metric space. If κ0, then M κ n is uniquely geodesic. If κ>0, then M κ n is π/ κ -uniquely geodesic, and any open ball (respectively, closed) ball of radius π/(2 κ ) (respectively, <π/(2 κ )) in X is convex [12], I.2.11]. The diameter of M κ n will be denoted D κ and thus D κ is π/ κ if κ>0, and ∞ otherwise.

Given two distinct points A,B S n with d(A,B)=<π there is a natural way to parameterize a unique geodesic joining A to B: consider the path c(t)=(cost)A+(sint)u, t[0,], where the initial vector u R n + 1 is the unit vector in the direction of B(AB)A. We shall refer to the image of c as a minimal great arc joining A to B.

The spherical angle between two minimal great arcs issuing from a point of S n , with the initial vectors u and v, say, is the unique number α[0,π] such that cosα=(uv). A spherical triangle in S n consists of a choice of three distinct points (its vertices) A,B,C S n , and three minimal great arcs (its sides), one joining each other of vertices. The vertex angle at C is defined to be the spherical angle between the sides of joining C to A and C to B.

Proposition 2.2 (The Spherical Law of Cosines [12], I.2.13])

Let a geodesic triangle in M κ n (κ>0) have sides a, b, c and angles α, β, γ at the vertices opposite to the sides of length a, b, c, respectively. Then

cos( κ c)=cos( κ a)cos( κ b)+sin( κ a)sin( κ b)cosγ.

In particular, fixing a, b and κ, c is a strictly increasing function of γ, varying from |ab| to a+b as γ varies from 0 to π.

A geodesic triangle in a metric space X consists of three points p,q,rX, its vertices, and a choice of three geodesic segments [p,q], [q,r], [r,p] joining them, its sides. Such a geodesic triangle will be denoted ([p,q],[q,r],[r,p]) or (less accurately if X is not uniquely geodesic) (p,q,r). If a point xX lies in the union of [p,q], [q,r] and [r,p], then we write x.

A triangle ¯ =( p ¯ , q ¯ , r ¯ ) in M κ 2 is called a comparison triangle for =([p,q],[q,r],[r,p]) in X if d M κ 2 ( p ¯ , q ¯ )=d(p,q), d M κ 2 ( q ¯ , r ¯ )=d(q,r) and d M κ 2 ( r ¯ , p ¯ )=d(r,p). Such a triangle ¯ M κ 2 always exists if the perimeter d(p,q)+d(q,r)+d(r,p) of is less than 2 D κ ; it is unique up to an isometry of M κ 2 [12], I.2.14]. A point x ¯ [ q ¯ , r ¯ ] is called a comparison point for x[q,r] if d M κ 2 ( q ¯ , x ¯ )=d(q,x).

A geodesic triangle in X is said to satisfy the CAT(κ) inequality if, given a comparison triangle ¯ M κ 2 for , for all x,y,

d(x,y) d M κ 2 ( x ¯ , y ¯ ),

where x ¯ , y ¯ ¯ are respective comparison points of x, y.

Definition 2.3 If κ0, then X is called a CAT(κ) space if X is a geodesic space all of whose geodesic triangles satisfy the CAT(κ) inequality.

If κ>0, then X is called a CAT(κ) space if X is D κ -geodesic and all geodesic triangles in X of perimeter less than 2 D κ satisfy the CAT(κ) inequality.

In particular, Hilbert spaces are CAT(0). A CAT(κ) space is a CAT( κ ) space for every κ κ. A CAT(κ) space X is ( D κ -)uniquely geodesic (if κ>0) and any open (respectively, closed) ball of radius D κ /2 (respectively, < D κ /2) in X is convex [12], II.1.4].

Lemma 2.4 Let (X,d) be a CAT(κ) space and let α,β[0,1]. Then

  1. (i)

    [12], II.2. Exercise 2.3(1)] for pX and x,yB(p, D κ /2), we have

    d ( α x ( 1 α ) y , p ) αd(x,p)+(1α)d(y,p);
  2. (ii)

    for x,yX, we have

    d ( α x ( 1 α ) y , β x ( 1 β ) y ) =|αβ|d(x,y);
  3. (iii)

    [10], Lemma 3.3] if κ>0, for x,y,zX with max{d(x,y),d(y,z),d(x,z)}<M D κ /2, we have

    d ( α x ( 1 α ) y , α x ( 1 α ) z ) sin [ ( 1 α ) M ] sin M d(y,z).

Let p, q, r be three distinct points of X with d(p,q)+d(q,r)+d(r,p)<2 D κ . The κ-comparison angle between q and r at p, denoted p ( κ ) (q,r), is the angle at p ¯ in a comparison triangle ( p ¯ , q ¯ , r ¯ ) M κ 2 for (p,q,r).

Let γ:[0,a]X and let γ :[0, a ]X be two geodesic paths with γ(0)= γ (0). Given t(0,a] and t (0, a ], we consider the comparison triangle ¯ (γ(0),γ(t), γ ( t )) and the κ-comparison angle γ ( 0 ) ( κ ) (γ(t), γ ( t )). The (Alexandrov) angle or the upper angle between the geodesic paths γ and γ is the number (γ, γ )[0,π] defined by

( γ , γ ) = lim sup t , t 0 γ ( 0 ) ( κ ) ( γ ( t ) , γ ( t ) ) = lim ϵ 0 sup 0 < t , t < ϵ γ ( 0 ) ( κ ) ( γ ( t ) , γ ( t ) ) .

If X is uniquely geodesic, pq and pr, the angle of (p,q,r) in X at p is the (Alexandrov) angle between the geodesic segments [p,q] and [p,r] issuing from p and is denoted p (q,r).

Proposition 2.5 ([12], I.1.13, 1.14 and II.1.8])

Let X be a metric space and let γ, γ , γ be three geodesics issuing from a common point. Then

  1. (i)

    (γ, γ )=( γ ,γ);

  2. (ii)

    (γ, γ )(γ, γ )+( γ , γ );

  3. (iii)

    if γ:[a,a]X is a geodesic with a>0, and if γ , γ :[0,a]X are defined by γ (t)=γ(t) and γ (t)=γ(t), then ( γ , γ )=π.

3 Basic properties and Δ-convergence

This section contains a number of primary results in [8] which are crucial to the study of our problem. The following proposition states very useful properties of the metric projection in a complete CAT(1) space.

Proposition 3.1 ([8], Proposition 3.5])

Let X be a complete CAT(1) space and let CX be nonempty closed and π-convex. Suppose that xX such that d(x,C)<π/2. Then the following are satisfied:

  1. (i)

    There exists a unique point P C xC such that d(x, P C x)=d(x,C).

  2. (ii)

    If xC and yC with y P C x, then P C x (x,y)π/2.

  3. (iii)

    If diam(X)π, then for any yC,

    d( P C x, P C y)=d( P C x,y)d(x,y).

The mapping P C of X onto C in Proposition 3.1 is called the metric projection.

The next result shows the existence property of fixed points for a nonexpansive mapping.

Proposition 3.2 ([8], Theorem 3.9])

Let X be a complete CAT(1) space such that diamX<π/2. Then every nonexpansive mapping T:XX has at least one fixed point.

The rest of this section is devoted to presenting several closely related characterizations of Δ-convergence. For this purpose, we start with some basic definitions of an asymptotic radius and an asymptotic center. Let { x n } be a bounded sequence in a complete CAT(1) space X. For xX and CX, let r(x,{ x n })= lim sup n d(x, x n ). The asymptotic radius r({ x n }) of { x n } is given by

r ( { x n } ) =inf { r ( x , { x n } ) : x X } ;

the asymptotic radius r C ({ x n }) with respective to C of { x n } is given by

r C ( { x n } ) =inf { r ( x , { x n } ) : x C } ;

the asymptotic center A({ x n }) of { x n } is given by the set

A ( { x n } ) = { x X : r ( x , { x n } ) = r ( { x n } ) } ;

the asymptotic center A C ({ x n }) with respective to C of { x n } is given by the set

A C ( { x n } ) = { x C : r ( x , { x n } ) = r C ( { x n } ) } .

Proposition 3.3 ([8], Proposition 4.1])

Let X be a complete CAT(1) space and CX nonempty closed and π-convex. If { x n } be a sequence in X such that r C ({ x n })<π/2, then A C ({ x n }) consists of exactly one point.

Definition 3.4 A sequence { x n } in X is said to Δ-converge to xX if x is the unique asymptotic center of every subsequence of { x n }. In this case, we write Δ- lim n x n =x and x is called the Δ-limit of { x n }.

The next result is an immediate consequence of the preceding proposition.

Proposition 3.5 ([8], Corollary 4.4])

Let X be a complete CAT(1) space and { x n } a sequence in X. If r({ x n })<π/2, then { x n } has a Δ-convergent subsequence.

4 Main results

Since the validity of all our results, including the proofs as well, on CAT(1) spaces can be restored on any CAT(κ) space with κ>0 by rescaling without major changes, we will pay our attention to CAT(1) spaces. In addition, when we deal with a CAT(κ) space, the hypothesis d( x 0 ,F(T))<π/4 for each theorem in this section is replaced by d( x 0 ,F(T))< D κ /4 and so can be dropped if κ0 (in this case, D κ =; refer to Section 2 for the definition).

To verify our result, the following basic property of asymptotical regularity is required; also cf. [6], Proposition 2.1] in a topological vector space.

Lemma 4.1 Let C be a subset of a metric space (X,d) and let S={T(t):t0} be a family of self-mappings of C such that T(s+t)=T(s)T(t), for all s,t0. If S is asymptotically regular, then

F ( T ( t ) ) = F ( S ) ,for all t>0.

Proof Fix t>0. Then F ( S ) F(T(t)). To prove F(T(t)) F ( S ) , let x be a fixed point of T(t). For h0, we obtain

d ( T ( h ) x , x ) = lim k d ( T ( h ) T ( t ) k x , T ( t ) k x ) = lim k d ( T ( h + t k ) x , T ( t k ) x ) = lim s d ( T ( h + s ) x , T ( s ) x ) = 0

and therefore x F ( S ) . □

Let (X,d) be a complete CAT(1) space, C a closed π-convex subset of X and T:CC a nonexpansive mapping. Then F(T) is closed. Also it is seen that F(T) is π-convex. Indeed, for x,yF(T) with d(x,y)<π, we have

d ( x , T ( λ x ( 1 λ ) y ) ) d ( x , λ x ( 1 λ ) y ) =(1λ)d(x,y)

and similarly d(y,T(λx(1λ)y))λd(x,y). Thus both equalities must hold and hence λx(1λ)y and T(λx(1λ)y) belong to the unique geodesic [x,y]. This implies that T(λx(1λ)y)=λx(1λ)y.

Now consider a family S={T(t):t0} of nonexpansive self-mappings of C such that F ( S ) . From the previous remark, each fixed point set F(T(t)) is closed and convex, hence so is F ( S ) . Fix x 0 C with 0<d( x 0 , F ( S ) )<π/4. Let p= P F ( S ) x 0 (the uniqueness follows from Proposition 3.1(i)) and ϵ=d( x 0 ,p). Then for each t0, T(t) maps the closed π-convex set C B ¯ (p,ϵ) into itself. For any α(0,1] and t0, define the mapping S α , t :C B ¯ (p,ϵ)C B ¯ (p,ϵ) by S α , t x=α x 0 (1α)T(t)x. Observe that S α , t is a contraction. In fact, for x,yC B ¯ (p,ϵ), we apply Lemma 2.4(iii) to get

d( S α , t x, S α , t y)sin [ ( 1 α ) π 2 ] d ( T ( t ) x , T ( t ) y ) sin [ ( 1 α ) π 2 ] d(x,y).

Then S α , t has a unique fixed point in C B ¯ (p,ϵ).

Let [t] denote the maximum integer no greater than t. We now extend Theorem 1.3 in Hilbert spaces to CAT(κ) spaces as the following result.

Theorem 4.2 Let X be a complete CAT(1) space, C a closed π-convex subset of X, S={T(t):t0} a UAR and strongly continuous nonexpansive semigroup on C with F ( S ) , and two sequences { α n }(0,1], { t n }(0,) such that lim n α n = lim n α n / t n =0. Choose arbitrarily a point x 0 C such that d( x 0 , F ( S ) )<π/4. Let p= P F ( S ) x 0 and ϵ=d( x 0 ,p). Define a sequence { x n } in C B ¯ (p,ϵ) by the implicit iteration

x n = α n x 0 (1 α n )T( t n ) x n .
(4.1)

Then { x n } converges strongly to the point p of F ( S ) nearest to x 0 .

Proof If x 0 F ( S ) , then { x n } reduces to the constant sequence { x 0 , x 0 ,}. Suppose that x 0 F ( S ) . We shall prove that any subsequence of { x n } contains a subsequence converging strongly to p from which it follows that { x n } also converges strongly to the point p. Let { y n } be any subsequence of { x n }. Proposition 3.5 asserts that { y n } has a Δ-convergent subsequence. By passing to a subsequence we may assume that { y n } Δ-converges to a point y. Let { β n } and { s n } be the respective corresponding subsequences of { α n } and { t n }.

Observe that y F ( S ) . To see this, take s ˆ = lim sup n s n . There are three cases:

  1. (i)

    s ˆ =0,

  2. (ii)

    0< s ˆ <,

  3. (iii)

    s ˆ =.

By passing to a subsequence if necessary it may be assumed that s ˆ = lim n s n . We discuss each case as follows.

Suppose that s ˆ =0. Fix t0 and we obtain

d ( T ( t ) y , y n ) d ( T ( t ) y , T ( [ t s n ] s n ) y ) + d ( T ( [ t s n ] s n ) y , T ( [ t s n ] s n ) y n ) + j = 1 [ t / s n ] d ( T ( ( j 1 ) s n ) y n , T ( j s n ) y n ) d ( T ( t [ t s n ] s n ) y , y ) + d ( y , y n ) + [ t s n ] d ( y n , T ( s n ) y n ) max 0 s s n d ( T ( s ) y , y ) + d ( y , y n ) + t β n s n d ( x 0 , T ( s n ) y n ) .

This implies that

lim sup n d ( T ( t ) y , y n ) lim sup n d(y, y n );

hence T(t)y=y for all t0, that is, y F ( S ) .

If 0< s ˆ <, then

d ( T ( s ˆ ) y , y n ) d ( T ( s ˆ ) y , T ( s n ) y ) + d ( T ( s n ) y , T ( s n ) y n ) + d ( T ( s n ) y n , y n ) d ( y , T ( | s n s ˆ | ) y ) + d ( y , y n ) + d ( T ( s n ) y n , y n ) d ( y , T ( | s n s ˆ | ) y ) + d ( y , y n ) + β n d ( x 0 , T ( s n ) y n )

and thus

lim sup n d ( T ( s ˆ ) y , y n ) lim sup n d(y, y n ).

Hence T( s ˆ )y=y from which it follows that y F ( S ) by Lemma 4.1.

If s ˆ =, fix t0. For all sufficiently large n we get

d ( T ( t ) y , y n ) d ( T ( t ) y , T ( t ) y n ) + d ( T ( t ) y n , y n ) d ( y , y n ) + β n d ( T ( t ) y n , x 0 ) + ( 1 β n ) d ( T ( t ) y n , T ( s n ) y n ) d ( y , y n ) + β n d ( T ( t ) y n , x 0 ) + ( 1 β n ) d ( y n , T ( s n t ) y n ) d ( y , y n ) + β n d ( T ( t ) y n , x 0 ) + β n ( 1 β n ) d ( x 0 , T ( s n t ) y n ) + ( 1 β n ) 2 d ( T ( s n ) y n , T ( s n t ) y n ) d ( y , y n ) + β n d ( T ( t ) y n , x 0 ) + β n ( 1 β n ) d ( x 0 , T ( s n t ) y n ) + ( 1 β n ) 2 d ( T ( s n t + t ) y n , T ( s n t ) y n )

which yields

lim sup n d ( T ( t ) y , y n ) lim sup n d(y, y n )

since

lim sup n d ( T ( s n t + t ) y n , T ( s n t ) y n ) lim n sup x C B ¯ ( p , ϵ ) d ( T ( h n + t ) x , T ( h n ) x ) =0,

where h n = s n t. It follows that T(t)y=y and therefore y F ( S ) . This finishes the proof that y is a common fixed point of all mappings in S.

Next, we claim that { y n } converges strongly to y. We suppose on the contrary that

r ( { y n } ) =σ>0.

From y=T( s n )y, we obtain

d ( y , y n ) β n d ( y , x 0 ) + ( 1 β n ) d ( y , T ( s n ) y n ) β n d ( y , x 0 ) + ( 1 β n ) d ( y , y n ) ,

and hence taking the limit superior as n yields

lim sup n d ( y , T ( s n ) y n ) = lim sup n d(y, y n )=σ,
(4.2)

since lim n β n =0. Recall that

d( x 0 , y n )=(1 β n )d ( x 0 , T ( s n ) y n ) <d ( x 0 , T ( s n ) y n ) ,for all n.

Then

d ( y n , T ( s n ) y n ) =d ( x 0 , T ( s n ) y n ) d( x 0 , y n )>0,for all n.

Recall that x 0 F ( S ) . We have y x 0 and hence

lim sup n d( x 0 , y n )>σ.
(4.3)

According to (4.2) and (4.3), by passing to a subsequence again we may assume that

d( x 0 , y n )>0,d(y, y n )>0,d ( y , T ( s n ) y n ) >0,for all n.

Let ( x ¯ 0 , y ¯ , T ( s n ) y n ¯ ) be a comparison triangle for ( x 0 ,y,T( s n ) y n ) in S 2 . Observe that

y ¯ n ( x ¯ 0 , y ¯ )π/2.
(4.4)

For, contrarily, if y ¯ n ( x ¯ 0 , y ¯ )<π/2, since y ¯ n ( x ¯ 0 , T ( s n ) y n ¯ )=π, then y ¯ n ( y ¯ , T ( s n ) y n ¯ )>π/2; see Proposition 2.5. By the law of cosines in S 2 (Proposition 2.2), since d( y n ,T( s n ) y n )>0, it follows that

cos d S 2 ( y ¯ , T ( s n ) y n ¯ ) <cos d S 2 ( y ¯ , y ¯ n ),

which is a contradiction because T ( s n ) ¯ is nonexpansive. Notice that

0<d(y, y n ) d S 2 ( y ¯ , y ¯ n ).

Hence (4.4) implies that

cos d S 2 ( x ¯ 0 , y ¯ )cos d S 2 ( x ¯ 0 , y ¯ n )cos d S 2 ( y ¯ , y ¯ n )<cos d S 2 ( x ¯ 0 , y ¯ n ),

or equivalently,

d S 2 ( x ¯ 0 , y ¯ n )< d S 2 ( x ¯ 0 , y ¯ ).

Similarly,

d S 2 ( y ¯ , y ¯ n )< d S 2 ( x ¯ 0 , y ¯ ).

The previous two inequalities, together with the CAT(κ) inequality, yield

d( x 0 , y n )<d( x 0 ,y),d(y, y n )<d( x 0 ,y),for all n.
(4.5)

Let ( x ¯ 0 , y ¯ , y ¯ n ) be a comparison triangle for ( x 0 ,y, y n ) in S 2 . Since d S 2 ( y ¯ n ,[ x ¯ 0 , y ¯ ])<π/2, Proposition 3.1(i) assures that there is a unique point u ¯ n [ x ¯ 0 , y ¯ ] nearest to y ¯ n . Let u n [ x 0 ,y] such that d( x 0 , u n )= d S 2 ( x ¯ 0 , u ¯ n ) and d( u n ,y)= d S 2 ( u ¯ n , y ¯ ). By passing to a subsequence again it may be assumed that { u ¯ n } and { u n } converge, respectively, to u ¯ [ x ¯ 0 , y ¯ ] and u[ x 0 ,y]. Since

σ = lim sup n d ( y , y n ) lim sup n d ( u , y n ) lim sup n d S 2 ( u ¯ , u ¯ n ) + lim sup n d S 2 ( u ¯ n , y ¯ n ) = lim sup n d S 2 ( u ¯ n , y ¯ n ) lim sup n d S 2 ( y ¯ , y ¯ n ) = lim sup n d ( y , y n ) ,
(4.6)

this guarantees that u=y.

On the other hand, according to (4.6), by passing to a subsequence we may suppose that

d S 2 ( u ¯ n , y ¯ n )> σ 2 ,for all n.

Let a n = d S 2 ( x ¯ 0 , u ¯ n ), b n = d S 2 ( u ¯ n , y ¯ n ), c n = d S 2 ( x ¯ 0 , y ¯ n ) and γ n = u ¯ n ( x ¯ 0 , y ¯ n ). Then by (4.5), b n >σ/2 and c n < d S 2 ( x ¯ 0 , y ¯ ). Observe that u n x 0 . Otherwise, since y x 0 , we have x ¯ 0 ( y ¯ , y ¯ n )= u ¯ n ( y ¯ , y ¯ n )π/2 which implies that

d(y, y n )= d S 2 ( y ¯ , y ¯ n ) d S 2 ( x ¯ 0 , y ¯ )=d( x 0 ,y).

But this contradicts to (4.5). Therefore u n x 0 and so γ n π/2 by Proposition 3.1(ii). The law of cosines in S 2 yields

cos c n = cos a n cos b n + sin a n sin b n cos γ n cos a n cos b n .

Since σ/2< b n c n < d S 2 ( x ¯ 0 , y ¯ )<π/2, this implies that

cos a n cos c n cos b n > cos d S 2 ( x ¯ 0 , y ¯ ) cos ( σ / 2 ) >cos d S 2 ( x ¯ 0 , y ¯ ).

Hence

d S 2 ( x ¯ 0 , u ¯ n )= a n < cos 1 ( cos d S 2 ( x ¯ 0 , y ¯ ) cos ( σ / 2 ) ) < d S 2 ( x ¯ 0 , y ¯ ).

It follows that

d(y, u n )= d S 2 ( y ¯ , u ¯ n )= d S 2 ( x ¯ 0 , y ¯ ) d S 2 ( x ¯ 0 , u ¯ n )> d S 2 ( x ¯ 0 , y ¯ )δ>0,

where δ= cos 1 [cos d S 2 ( x ¯ 0 , y ¯ )/cos(σ/2)]. Taking the limit as n yields uy, which is a contradiction. Thus σ=0 and so { y n } converges strongly to y, as claimed.

It remains to prove that y= P F ( S ) x 0 . Let q F ( S ) and consider a comparison triangle ( x ¯ 0 , q ¯ , T ( s n ) y n ¯ ) for ( x 0 ,q,T( s n ) y n ). From

d S 2 ( q ¯ , T ( s n ) y n ¯ ) d S 2 ( q ¯ , y ¯ n ),

it is seen that y ¯ n ( q ¯ , T ( s n ) y n ¯ )π/2. Hence y ¯ n ( x ¯ 0 , q ¯ )π/2 and so

d( x 0 , y n )= d S 2 ( x ¯ 0 , y ¯ n ) d S 2 ( x ¯ 0 , q ¯ )=d( x 0 ,q).

We then take the limit as n and obtain

d( x 0 ,y)d( x 0 ,q),

which shows that y is the nearest point in F ( S ) to x 0 . Consequently, we conclude that { x n } converges strongly to P F ( S ) x 0 , which completes the proof. □

It is worthy emphasizing that a uniformly asymptotical regularity hypothesis of Theorem 4.2 is superfluous when lim n t n = lim n α n / t n =0; see case (i) in the proof of Theorem 4.2. Thus, we state the result as follows.

Theorem 4.3 Let X be a complete CAT(1) space, C a closed π-convex subset of X, S={T(t):t0} a strongly continuous nonexpansive semigroup on C with F ( S ) , and two sequences { α n }(0,1], { t n }(0,) such that lim n t n = lim n α n / t n =0. Choose arbitrarily a point x 0 C such that d( x 0 , F ( S ) )<π/4. Let p= P F ( S ) x 0 and ϵ=d( x 0 ,p). Then the sequence { x n } in C B ¯ (p,ϵ) defined by (4.1) converges strongly to the point p.

Furthermore, if X is a CAT(0) space (recall that D 0 =) in Theorem 4.3, then the corresponding assumption d( x 0 , F ( S ) )<π/4 is d( x 0 , F ( S ) )<, which can be dropped. Moreover, the sequence { x n } defined by (4.1) is bounded. In fact, since

d ( p , x n ) α n d ( p , x 0 ) + ( 1 α n ) d ( p , T ( t n ) x n ) α n d ( p , x 0 ) + ( 1 α n ) d ( p , x n ) ,

this shows that d(p, x n )d(p, x 0 ). Then { x n } is bounded and so is {T(t) x n :t0,nN}. We restate Theorem 1.2 without assuming boundedness of C as follows.

Theorem 4.4 Let X be a complete CAT(0) space, C a closed convex subset of X, S={T(t):t0} a strongly continuous nonexpansive semigroup on C with F ( S ) , and two sequences { α n }(0,1], { t n }(0,) such that lim n t n = lim n α n / t n =0. Choose arbitrarily a point x 0 C. Then the sequence { x n } in C defined by (4.1) converges strongly to a point of F ( S ) nearest to x 0 .

Nonetheless, Theorem 4.2 has a uniformly asymptotical regularity hypothesis that cannot in general be removed when lim n t n = t ˆ (0,] as illustrated in the following example.

Example 4.5 Consider the complex plane (C,||), where || is the absolute value, so that it is a complete CAT(0) space. For t0, define a self-mapping T(t) of by T(t)(z)=z e i t , zC. Then S={T(t):t0} is a strongly continuous nonexpansive semigroup with F(S)={0}. The family S is not UAR on because

lim t sup | z | = 1 | T ( π ) T ( t ) ( z ) T ( t ) ( z ) | = lim t sup | z | = 1 |z e i ( t + π ) z e i t |=2.

Let α n =1/n, t n =nπ, nN, so that lim n t n = and lim n α n / t n =0. Choose z 0 =1. Define a sequence { z n } in by (4.1), that is,

z n = { 1 2 n 1 , n  is odd , 1 , n  is even .

Hence the sequence { z n } is divergent.

Next if we take α n =1/n, t n =(1/n)+2π, nN, then lim n t n =2π and lim n α n / t n =0. Choose z 0 =1 and define a sequence { z n } in by (4.1) to get

z n = 1 / n 1 [ 1 ( 1 / n ) ] e i / n .

However, the sequence { z n } converges to (1+i)/2F(S).

These two examples explain that strong convergence of the sequence defined by (4.1) in Theorem 4.2 ceases to be true without the UAR assumption if lim n t n = t ˆ (0,].

The next example shows that the condition lim n α n = lim n α n / t n =0 in Theorem 4.2 is also necessary when X is the unit sphere in 2 , the infinite-dimensional Hilbert space of square-summable sequences.

Example 4.6 Let S 2 be the unit sphere in 2 endowed with the following intrinsic metric d: S 2 × S 2 R: for x,y S 2 , d(x,y)[0,π] such that

cosd(x,y)=(xy)= k = 1 x(k)y(k).

This space ( S 2 ,d) is a CAT(1) space.

Take one element v=(1,0,,0,) of the canonical basis of S 2 . Let C={x S 2 :d(x,v)r} be a closed ball centered at v in S 2 , where 0<r<π/4. For t0, define a mapping T(t):CC by

T(t)(x)= ( 1 e t ) v e t x,

that is,

( T ( t ) ( x ) ) (1)=cos [ e t cos 1 x ( 1 ) ] ,
(4.7)

and for k=2,3, ,

( T ( t ) ( x ) ) (k)= { 0 , if  x ( 1 ) = 1 , x ( k ) sin [ e t cos 1 x ( 1 ) ] 1 x ( 1 ) 2 , if  x ( 1 ) 1 .
(4.8)

For x,yC, from Lemma 2.4(iii), we obtain

d ( T ( t ) x , T ( t ) y ) = d ( ( 1 e t ) v e t x , ( 1 e t ) v e t y ) sin ( e t π / 2 ) sin π / 2 d ( x , y ) d ( x , y ) .

Then T(t) is nonexpansive. In fact, v is the unique fixed point of T(t) for t>0. Furthermore, it is seen that S={T(t):t0} is a UAR and strongly continuous nonexpansive semigroup on C.

Choose u=( u k )C with 0< u 1 <1 so that 0< cos 1 u 1 =d(u,v)r. For α(0,1) and t(0,), define x(α,t) by

x ( α , t ) = α u ( 1 α ) T ( t ) x ( α , t ) = cos [ ( 1 α ) d ( u , T ( t ) x ( α , t ) ) ] u + sin [ ( 1 α ) d ( u , T ( t ) x ( α , t ) ) ] A ( α , t ) A ( α , t ) ,
(4.9)

where

A(α,t)=T(t)x(α,t) ( u T ( t ) x ( α , t ) ) u.

We remark that A ( α , t ) 2 =1 ( u T ( t ) x ( α , t ) ) 2 . Let { α n }(0,1], { t n }(0,) be two sequences and define a sequence { x n } in B ¯ (v, cos 1 u 1 )C by

x n =x( α n , t n )= α n u(1 α n )T( t n ) x n .

We shall prove that if the sequence { x n } converges to the common fixed point v of S, then lim n α n = lim n α n / t n =0.

First, assume that β= lim sup n α n . Since lim n x n (1)=1 and | x n (k)| 1 x n ( 1 ) 2 for k2, we obtain from (4.7) and (4.8) that

lim n ( T ( t n ) ( x n ) ) (1)=1, lim n ( T ( t n ) ( x n ) ) (k)=0.

Therefore lim n T( t n )( x n )=v, which implies that

lim n ( u T ( t n ) ( x n ) ) = u 1 , lim n d ( u , T ( t n ) x n ) = cos 1 u 1 ;
(4.10)

hence

lim n A( α n , t n )(1)=1 u 1 2 , lim n A ( α n , t n ) = 1 u 1 2 .
(4.11)

Let θ=d(u,v)= cos 1 u 1 so that sinθ= 1 u 1 2 . Using (4.9), (4.10), and (4.11), we have

1 = lim n x n ( 1 ) = lim inf n { u 1 cos [ ( 1 α n ) d ( u , T ( t n ) x n ) ] + A ( α n , t n ) ( 1 ) A ( α n , t n ) sin [ ( 1 α n ) d ( u , T ( t n ) x n ) ] } lim sup n { u 1 cos [ ( 1 α n ) d ( u , T ( t n ) x n ) ] } + lim inf n { A ( α n , t n ) ( 1 ) A ( α n , t n ) sin [ ( 1 α n ) d ( u , T ( t n ) x n ) ] } = cos θ cos [ ( 1 β ) θ ] + sin θ sin [ ( 1 β ) θ ] = cos ( β θ ) .

This shows that β=0 because θ>0 and so lim n α n =0.

To prove that lim n α n / t n =0, we recall the inequality t1 e t for all t0. Since

t n d(v, x n ) ( 1 e t n ) d(v, x n )=d ( x n , T ( t n ) ( x n ) ) = α n d ( u , T ( t n ) ( x n ) ) ,

it follows that

0 = lim n d ( v , x n ) lim sup n [ α n t n d ( u , T ( t n ) ( x n ) ) ] = θ lim sup n α n t n .

Consequently, lim sup n α n / t n =0 and therefore lim n α n / t n =0 as desired.

The following result is an immediate consequence of Theorem 4.2 and Example 4.6.

Theorem 4.7 Let X be the unit sphere in an infinite-dimensional Hilbert space, and let { α n }(0,1], { t n }(0,) be two sequences. Then the following conditions are equivalent:

  1. (i)

    lim n α n = lim n α n / t n =0.

  2. (ii)

    Let C be a closed convex subset of X with diamC<π/2, S={T(t):t0} a UAR and strongly continuous nonexpansive semigroup on C with F ( S ) , x 0 C such that d( x 0 , F ( S ) )<π/4, p= P F ( S ) x 0 , ϵ=d( x 0 ,p) and { x n } is a sequence in C B ¯ (p,ϵ) defined by (4.1).

Then { x n } converges strongly to p.

5 Remark

A semitopological semigroup S is a semigroup equipped with a Hausdorff topology such that for each tS the mappings sts and sst from S into S are continuous. A semitopological semigroup S is left (respectively, right) reversible if any two closed right (respectively, left) ideals of S have nonvoid intersection, i.e., a S ¯ b S ¯ (respectively, S a ¯ S b ¯ ), for a,bS, where E ¯ denotes the closure of a set E in a topological space. The class S of all left reversible semitopological semigroups includes all groups, all commuting semigroups, and all normal left amenable semitopological semigroups, i.e., the space CB(S) of bounded continuous functions on S has a left invariant mean; see [2, 3, 1315]. If S is left (respectively, right) reversible, (S,) is a directed system when the binary relation on S is defined by ab if and only if {a} a S ¯ {b} b S ¯ (respectively, {a} S a ¯ {b} S b ¯ ), a,bS.

Let S be a semitopological semigroup and C a closed convex subset of a metric space (X,d). A family S={T(s):sS} is called a representation of S on C if for each sS, T(s) is a mapping from C into C and T(st)=T(s)T(t), for all s,tS. The representation is called a strongly continuous nonexpansive semigroup on C (or a continuous representation of S as nonexpansive mappings on C) if the following conditions are satisfied (see [3]):

  1. (i)

    for each tS, T(t) is a nonexpansive mapping on C;

  2. (ii)

    for each xC, the mapping tT(t)x from S into C is continuous.

The representation S of S is called

  1. (i)

    asymptotically regular on C if for any hS and any xC,

    lim t S d ( T ( h ) T ( t ) x , T ( t ) x ) =0;
  2. (ii)

    uniformly asymptotically regular (in short UAR) on C if for any hS and any bounded subset D of C,

    lim t S sup x D d ( T ( h ) T ( t ) x , T ( t ) x ) =0.

Problem 5.1 Let X be a complete CAT(1) space, C a closed π-convex subset of X, S a commutative (or left reversible) semitopological semigroup, and S={T(t):tS} a continuous representation of S as nonexpansive mappings on C. Can we obtain the result analogous to Theorem 4.2 corresponding to a continuous representation S={T(t):tS} of S as nonexpansive mappings on C?

To answer this problem, the following conjecture is required and hence needs to be verified.

Conjecture 5.2 Let S be a commutative (or left reversible) semitopological semigroup, C a subset of a metric space (X,d) and S={T(t):tS} a representation of S on C. If S is asymptotically regular, then

F ( T ( t ) ) = F ( S ) ,for all tS.

References

  1. 1.

    Lau AT-M, Miyake H, Takahashi W: Approximation of fixed points for amenable semigroups of nonexpansive mappings in Banach spaces. Nonlinear Anal. 2007, 67: 1211–1225. 10.1016/j.na.2006.07.008

  2. 2.

    Lau AT-M: Normal structure and common fixed point properties for semigroups of nonexpansive mappings in Banach spaces. Fixed Point Theory Appl. 2010., 2010: Article ID 580956

  3. 3.

    Lau AT-M, Zhang Y: Fixed point properties for semigroups of nonlinear mappings and amenability. J. Funct. Anal. 2012, 263(10):2949–2977. 10.1016/j.jfa.2012.07.013

  4. 4.

    Suzuki T: Strong convergence of Krasnoselskii and Mann’s type sequences for one parameter nonexpansive semigroups without Bochner integrals. J. Math. Anal. Appl. 2005, 305: 227–239. 10.1016/j.jmaa.2004.11.017

  5. 5.

    Browder FE: Convergence of approximants to fixed points of nonexpansive nonlinear mappings in Banach spaces. Arch. Ration. Mech. Anal. 1967, 24: 82–90.

  6. 6.

    Acedo GL, Suzuki T: Browder’s convergence for uniformly asymptotically regular nonexpansive semigroups in Hilbert spaces. Fixed Point Theory Appl. 2010., 2010: Article ID 418030

  7. 7.

    Suzuki T: On strong convergence to common fixed points of nonexpansive semigroups in Hilbert spaces. Proc. Am. Math. Soc. 2003, 131(7):2133–2136. 10.1090/S0002-9939-02-06844-2

  8. 8.

    Espínola R, Fernández-León A: CAT(κ)-spaces, weak convergence and fixed points. J. Math. Anal. Appl. 2009, 353: 410–427. 10.1016/j.jmaa.2008.12.015

  9. 9.

    Huang S: The Δ-convergence of iterations for nonexpansive mappings in CAT(κ) spaces. J. Nonlinear Convex Anal. 2012, 13(3):541–554.

  10. 10.

    Pia̧tek B: Halpern iteration in CAT(κ) spaces. Acta Math. Sin. Engl. Ser. 2011, 27: 635–646. 10.1007/s10114-011-9312-7

  11. 11.

    Dhompongsa S, Fupinwong W, Kaewkhaob A: Common fixed points of a nonexpansive semigroup and a convergence theorem for Mann iterations in geodesic metric spaces. Nonlinear Anal. 2009, 70: 4268–4273. 10.1016/j.na.2008.09.012

  12. 12.

    Bridson M, Haefliger A: Metric Spaces of Non-Positive Curvature. Springer, Berlin; 1999.

  13. 13.

    Holmes RD, Lau AT-M: Asymptotically non-expansive actions of topological semigroups and fixed points. Bull. Lond. Math. Soc. 1971, 3: 343–347. 10.1112/blms/3.3.343

  14. 14.

    Lau AT-M: Semigroup of nonexpansive mappings on a Hilbert space. J. Math. Anal. Appl. 1985, 105(2):514–522. 10.1016/0022-247X(85)90066-6

  15. 15.

    Lau AT-M, Takahashi W: Weak convergence and nonlinear ergodic theorems for reversible semigroups of nonexpansive mappings. Pac. J. Math. 1987, 126: 277–294. 10.2140/pjm.1987.126.277

Download references

Acknowledgements

This research was supported by a grant NSC 102-2115-M-259-004 from the National Science Council of Taiwan. The author, therefore, thanks the NSC financial support. The author would like to express the most sincere thanks to the referees for their careful reading of the manuscript, important comments and the citation of Ref. [13]. The author is also very grateful to Professor Anthony Lau for giving valuable remarks and Refs. [14] and [3], and for suggesting presenting Problem 5.1 for future studies.

Author information

Correspondence to Shuechin Huang.

Additional information

Competing interests

The author declares that she has no competing interests.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Reprints and Permissions

About this article

Cite this article

Huang, S. Nonexpansive semigroups in CAT(κ) spaces. Fixed Point Theory Appl 2014, 44 (2014) doi:10.1186/1687-1812-2014-44

Download citation

Keywords

  • nonexpansive mapping
  • asymptotically regular
  • uniformly asymptotically regular
  • Δ-convergence
  • CAT(κ) space