Bregman Asymptotic Pointwise Nonexpansive Mappings in Banach Spaces
Abstract
We first introduce a new class of mappings called Bregman asymptotic pointwise nonexpansive mappings and investigate the existence and the approximation of fixed points of such mappings defined on a nonempty, bounded, closed, and convex subset C of a real Banach space E. Without using the original Opial property of a Banach space E, we prove weak convergence theorems for the sequences produced by generalized Mann and Ishikawa iteration processes for Bregman asymptotic pointwise nonexpansive mappings in a reflexive Banach space E. Our results are applicable in the function spaces Lp, where 1 < p < ∞ is a real number.
1. Introduction
1.1. Some Facts about Gradients
1.2. Some Facts about Bregman Distances
1.3. Some Facts about Uniformly Convex Functions
Remark 1. Let E be a Banach space, let s > 0 be a constant, and let g : E → ℝ be a convex function which is uniformly convex on bounded subsets. Then
Definition 2. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E. A mapping T : C → E is said to be Bregman asymptotic pointwise nonexpansive if there exists a sequence of mappings θn : C → [0, ∞) such that
Imposing some restrictions on the behavior of an and bn, we can define the following subclass of Bregman asymptotic pointwise nonexpansive mappings.
Definition 3. Let C and ℬ𝒯(C) be as in Definition 2. We define ℬ𝒯r(C) as a class of all T ∈ ℬ𝒯(C) such that
In this paper, we first investigate the approximation of fixed points of a new class of Bregman asymptotic pointwise nonexpansive mappings defined on a nonempty, bounded, closed, and convex subset C of a real Banach space E. Without using the Opial property of a Banach space E, we prove weak convergence theorems for the sequences produced by generalized Mann and Ishikawa iteration processes. Our results improve and generalize many known results in the current literature; see, for example, [18, 21].
2. Preliminaries
In this section, we begin by recalling some preliminaries and lemmas which will be used in the sequel.
Definition 4 (see [10].)Let E be a Banach space. The function g : E → ℝ is said to be a Bregman function if the following conditions are satisfied:
- (1)
g is continuous, strictly convex, and Gâteaux differentiable;
- (2)
the set {y ∈ E : Dg(x, y) ≤ r} is bounded for all x ∈ E and r > 0.
Lemma 5 (see [9], [16].)Let E be a reflexive Banach space and g : E → ℝ a strongly coercive Bregman function. Then
- (1)
∇g : E → E* is one-to-one, onto, and norm-to-weak* continuous;
- (2)
〈x − y, ∇g(x)−∇g(y)〉 = 0 if and only if x = y;
- (3)
{x ∈ E : Dg(x, y) ≤ r} is bounded for all y ∈ E and r > 0;
- (4)
dom g* = E*, g* is Gâteaux differentiable and ∇g* = (∇g) −1.
We know the following two results; see [16, Proposition 3.6.4].
Theorem 6. Let E be a reflexive Banach space and g : E → ℝ a convex function which is bounded on bounded subsets of E. Then the following assertions are equivalent:
- (1)
g is strongly coercive and uniformly convex on bounded subsets of E;
- (2)
dom g* = E*, g* is bounded on bounded subsets and uniformly smooth on bounded subsets of E*;
- (3)
dom g* = E*, g* is Frchet differentiable and ∇g* is uniformly norm-to-norm continuous on bounded subsets of E*.
Theorem 7. Let E be a reflexive Banach space and g : E → ℝ a continuous convex function which is strongly coercive. Then the following assertions are equivalent:
- (1)
g is bounded on bounded subsets and uniformly smooth on bounded subsets of E;
- (2)
g* is Frchet differentiable and ∇g* is uniformly norm-to-norm continuous on bounded subsets of E*;
- (3)
dom g* = E*, g* is strongly coercive and uniformly convex on bounded subsets of E*.
Lemma 8 (see [23].)Let E be a Banach space and g : E → ℝ a Gâteaux differentiable function which is uniformly convex on bounded subsets of E. Let {xn} n∈ℕ and {yn} n∈ℕ be bounded sequences in E. Then the following assertions are equivalent:
- (1)
lim n→∞Dg(xn, yn) = 0;
- (2)
lim n→∞∥xn − yn∥ = 0.
Lemma 9 (see [10], [24].)Let E be a reflexive Banach space, g : E → ℝ a strongly coercive Bregman function, and V the function defined by
- (1)
Dg(x, ∇g*(x*)) = V(x, x*) for all x ∈ E and x* ∈ E*;
- (2)
V(x, x*)+〈∇g*(x*) − x, y*〉≤V(x, x* + y*) for all x ∈ E and x*, y* ∈ E*.
Lemma 10 (see [25].)Suppose {rk} k∈ℕ is a bounded sequence of real numbers and {dk,n} k,n∈ℕ is a doubly index sequence of real numbers which satisfy
The following Bregman Opial-like inequality has been proved in [26]. It is worth mentioning that the Bregman Opial-like inequality is different from the ordinary Opial inequality [6] and can be applied in uniformly convex Banach spaces.
Lemma 11 (see [26].)Let E be a Banach space and let g : E → (−∞, +∞] be a proper strictly convex function so that it is Gâteaux differentiable on int dom g. Suppose {xn} n∈ℕ is a sequence in dom g such that xn⇀v for some v ∈ int dom g. Then
Theorem 12 (see [16].)Let g : E → (−∞, +∞] be a function. Then the following assertions are equivalent:
- (1)
g is convex and lower semicontinuous;
- (2)
g is convex and weakly lower semicontinuous;
- (3)
epi (g) is convex and closed;
- (4)
epi (g) is convex and weakly closed,
where epi (g) = {(x, t) ∈ E × ℝ : g(x) ≤ t} denotes the epigraph of g.
3. Fixed Point Theorems and Demiclosedness Principle
Proposition 13. Let E be a reflexive Banach space and let g : E → ℝ be a strongly admissible function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E and let T ∈ ℬ𝒯r(C). Then T has a fixed point. Moreover, F(T) is closed and convex.
Proof. We first show that F(T) is nonempty. Let x in C be fixed. We define a function f : C → [0, ∞) by
Now, we show that F(T) is closed. Let {pn} n∈ℕ be a sequence in F(T) such that pn → p as n → ∞. Then we have that {pn} n∈ℕ is a bounded sequence in E. We claim that p ∈ F(T). Since g is continuous, we conclude that g(pn) → g(p) as n → ∞. This implies that
Let us show that F(T) is convex. For any p, q ∈ F(T), t ∈ (0,1), and n ∈ ℕ, we set x = tp + (1 − t)q and en(p, q) = max {an(p), an(q)}. We prove that x ∈ F(T). By the definition of Bregman distance (see (12)), we get
Lemma 14. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E and let T ∈ ℬ𝒯r(C). If {xn} n∈ℕ is a sequence of C such that lim n→∞∥Txn − xn∥ = 0, then, for any m ∈ ℕ, lim n→∞∥Tmxn − xn∥ = 0.
Proof. In view of (25), there exists a finite constant M1 > 0 such that
Theorem 15. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex and uniformly smooth on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E and let T ∈ ℬ𝒯r(C). If {xn} n∈ℕ converges weakly to z and lim n→∞∥Txn − xn∥ = 0, then Tz = z. That is, I − T is demiclosed at zero, where I is the identity mapping on E.
Proof. Let the function ϕ : E → [0, ∞) be defined by
4. Weak Convergence Theorems of Generalized Mann Iteration Process
In this section, we prove weak convergence theorems of generalized Mann iteration process in a reflexive Banach space.
Definition 16. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E and let T ∈ ℬ𝒯r(C). Let {nk} k∈ℕ be an increasing sequence in ℕ and let {γk} k∈ℕ ⊂ (0,1) be such that lim inf k→γk(1 − γk) > 0. The generalized Mann iteration process generated by the mapping T, the sequence {γk} k∈ℕ, and the sequence {nk} k∈ℕ, denoted by gM(T, {γk} k∈ℕ, {nk} k∈ℕ), is defined by the following iterative formula:
Definition 17. We say that a generalized Mann iteration process gM(T, {γk} k∈ℕ, {nk} k∈ℕ) is well defined if
Remark 18. Observe that by the definition of Bregman asymptotic pointwise nonexpansive, for every x ∈ C. Hence we can always select a subsequence of {an} n∈ℕ such that (67) holds. In other words, by a suitable choice of {nk} k∈ℕ we can always make gM(T, {γk} k∈ℕ, {nk} k∈ℕ) well defined.
We will prove a series of lemmas necessary for the proof of the generalized Mann process convergence theorem.
Lemma 19. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {γk} k∈ℕ ⊂ (0,1) such that lim inf k→∞γk(1 − γk) > 0. Let w ∈ F(T) and let gM(T, {γk} k∈ℕ, {nk} k∈ℕ) be a generalized Mann process. Then there exists λ ∈ ℝ such that lim k→∞D(xk, w) = λ.
Proof. Let w ∈ F(T) be arbitrary chosen. In view of (66), we obtain
Lemma 20. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {γk} k∈ℕ ⊂ (0,1) such that lim inf k→∞γk(1 − γk) > 0 and let gM(T, {γk} k∈ℕ, {nk} k∈ℕ) be a generalized Mann process. Then
Proof. In view of Proposition 13, we conclude that F(T) ≠ ∅. Let w ∈ F(T) be fixed. It follows from Lemma 19 that there exists λ ∈ ℝ such that lim k→∞Dg(xk, w) = λ. Let and let be the gauge of uniform convexity of g. By the definition of T, we obtain
In the next lemma, we prove that under suitable assumption the sequence {xk} k∈ℕ becomes an approximate fixed point sequence, which will provide an important step in the proof of the generalized Mann iteration process convergence. First, we need to recall the following notions.
Definition 21. A strictly increasing sequence {ni} i∈ℕ ⊂ ℕ is called quasiperiodic if the sequence {ni+1 − ni} i∈ℕ is bounded or equivalently if there exists a number p ∈ ℕ such that any block of consecutive natural numbers must contain a term of the sequence {ni} i∈ℕ. The smallest of such numbers will be called a quasi period of {ni} i∈ℕ.
Lemma 22. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {γk} k∈ℕ ⊂ (0,1) such that liminf k→∞γk(1 − γk) > 0. Let {nk} k∈ℕ ⊂ ℕ be such that the generalized Mann process gM(T, {γk} k∈ℕ, {nk} k∈ℕ) is well defined. If, in addition, the set of indices 𝒥 = {j ∈ ℕ : nj+1 = 1 + nj} is quasi-periodic, then {xk} k∈ℕ is an approximate fixed point sequence; that is,
Proof. In view of (66), we have
Theorem 23. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets. Let C be a nonempty, bounded, closed and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {γk} k∈ℕ ⊂ (0,1) such that lim inf k→∞γk(1 − γk) > 0. Let {nk} k∈ℕ ⊂ ℕ be such that the generalized Mann process gM(T, {γk} k∈ℕ, {nk} k∈ℕ) is well defined. If, in addition, the set of indices 𝒥 = {j ∈ ℕ : kj+1 = 1 + kj} is quasi-periodic, then the sequence {xk} k∈ℕ generated by gM(T, {γk} k∈ℕ, {nk} k∈ℕ) converges weakly to a fixed point of T.
Proof. In view of Lemma 22, we obtain
5. Weak Convergence of Generalized Ishikawa Iteration Process
The two-step Ishikawa iteration process is a generalization of the one-step Mann iteration process. The Ishikawa iteration process provides more flexibility in defining the algorithm parameters which is important from numerical implementation perspective.
Definition 24. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ be an increasing sequence. Let {βk} k∈ℕ ⊂ (0,1) and {γk} k∈ℕ ⊂ (0,1) be sequences of real numbers such that lim inf k→∞βk(1 − βk) > 0 and lim inf k→∞γk(1 − γk) > 0. The generalized Ishikawa iteration process generated by the mapping T, the sequences {βk} k∈ℕ ⊂ (0,1), and {γk} k∈ℕ ⊂ (0,1), and the sequence {nk} k∈ℕ denoted by gI(T, {βk} k∈ℕ, {γk} k∈ℕ, {nk} k∈ℕ) is defined by the following iterative scheme:
Definition 25. We say that a generalized Ishikawa iteration process gI(T, {βk} k∈ℕ, {γk} k∈ℕ, {nk} k∈ℕ) is well defined if
Lemma 26. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets of E. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {βk} k∈ℕ ⊂ (0,1) and {γk} k∈ℕ ⊂ (0,1) be sequences of real numbers such that lim inf k→∞βk(1 − βk) > 0 and lim inf k→∞γk(1 − γk) > 0. Let w ∈ F(T) and gI(T, {βk} k∈ℕ, {γk} k∈ℕ, {nk} k∈ℕ) be a generalized Ishikawa iteration process. Then there exists θ ∈ ℝ such that lim k→∞Dg(xk, w) = θ.
Proof. Let M7 > 1 be fixed. Since , there exists k0 ∈ ℕ such that for any k > k0, . Let and let be the gauge of uniform convexity of g. By the definition of T and in view of (93), we obtain
Lemma 27. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {βk} k∈ℕ ⊂ (0,1) and {γk} k∈ℕ ⊂ (0,1) be sequences of real numbers such that lim inf k→∞γk(1 − γk) > 0 and lim inf k→∞βk(1 − βk) > 0. Let w ∈ F(T) and gI(T, {γk} k∈ℕ, {βk} k∈ℕ, {nk} k∈ℕ) be a generalized Ishikawa iteration process. Then
Proof. In view of Proposition 13, F(T) ≠ ∅. Take any w ∈ F(T) arbitrarily chosen. Then, by Lemma 26, there exists θ ∈ ℝ such that lim k→∞Dg(xk, w) = θ. By the same arguments, as in the proof of Lemma 26, we conclude that
Lemma 28. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {γk} k∈ℕ ⊂ (0,1) such that lim inf k→∞γk(1 − γk) > 0. Let {nk} k∈ℕ ⊂ ℕ be such that the generalized Ishikawa process gI(T, {γk} k∈ℕ, {βk} k∈ℕ, {nk} k∈ℕ) is well defined. If, in addition, the set of indices 𝒥 = {j ∈ ℕ : nj+1 = 1 + nj} is quasi-periodic, then {xk} k∈ℕ is an approximate fixed point sequence; that is,
Theorem 29. Let E be a reflexive Banach space and let g : E → ℝ be a convex, continuous, strongly coercive, and Gâteaux differentiable function which is bounded on bounded subsets and uniformly convex on bounded subsets. Let C be a nonempty, bounded, closed, and convex subset of E. Let T ∈ ℬ𝒯r(C) and let {nk} k∈ℕ ⊂ ℕ. Let {γk} k∈ℕ ⊂ (0,1) such that lim inf k→∞γk(1 − γk) > 0. Let {nk} k∈ℕ ⊂ ℕ be such that the generalized Ishikawa process gI(T, {γk} k∈ℕ, {βk} k∈ℕ, {nk} k∈ℕ) is well defined. If, in addition, the set of indices 𝒥 = {j ∈ ℕ : kj+1 = 1 + kj} is quasi-periodic, then the sequence {nk} k∈ℕ generated by gI(T, {γk} k∈ℕ, {βk} k∈ℕ, {nk} k∈ℕ) converges weakly to a fixed point of T.
Remark 30. Theorem 29 improves Theorems 3.1 and 4.1 of [18] in the following aspects.
- (1)
For the structure of Banach spaces, we extend the duality mapping to more general case, that is, a convex, continuous, strongly coercive Bregman function which is bounded on bounded sets and uniformly convex and uniformly smooth on bounded sets.
- (2)
For the mappings, we extend the mapping from an asymptotic pointwise nonexpansive mapping to a Bregman asymptotic pointwise nonexpansive mapping.
- (3)
Since we do not need the weak sequential continuity of the duality mapping in Theorems 23 and 29 as was the case in [18], we can apply Theorem 29 in the Lebesgue space Lp where 1 < p < ∞ and p ≠ 2 while this space is not applicable for Theorems 3.1, 4.1, and 5.1 of [18].
Authors’ Contribution
All authors read and approved the final paper.
Conflict of Interests
The authors declare that they have no competing interests.
Acknowledgments
The authors would like to thank the editor and the referees for sincere evaluation and constructive comments which improved the paper considerably.