Hindawi Mathematical Problems in Engineering Volume 2017, Article ID 9737062, 9 pages https://doi.org/10.1155/2017/9737062
Research Article A Simultaneous Iteration Algorithm for Solving Extended Split Equality Fixed Point Problem Meixia Li,1 Xiping Kao,2 and Haitao Che1 1
School of Mathematics and Information Science, Weifang University, Weifang, Shandong 261061, China College of Mathematics and Systems Science, Shandong University of Science and Technology, Qingdao, Shandong 266590, China
2
Correspondence should be addressed to Meixia Li;
[email protected] Received 16 January 2017; Accepted 20 March 2017; Published 6 April 2017 Academic Editor: Roberto Fedele Copyright © 2017 Meixia Li et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. We study a kind of split equality fixed point problem which is an extension of split equality problem. We propose a kind of simultaneous iterative algorithm with a way of selecting the step length which does not need any a priori information about the operator norms and prove that the sequences generated by the iterative method converge weakly to the solution of this problem. Some numerical results are shown to confirm the feasibility and efficiency of the proposed methods.
1. Introduction Let 𝐶 and 𝑄 be nonempty closed and convex subsets of the real Hilbert spaces 𝐻1 and 𝐻2 , respectively. The split feasibility problem is to find 𝑥∈𝐶
such that 𝐴𝑥 ∈ 𝑄.
(1)
It can be used in various disciplines such as image restoration and radiation therapy treatment planning [1, 2]. These applications are in finite-dimensional Hilbert spaces [3, 4]. It also can be found in an infinite-dimensional real Hilbert space [5, 6]. Recently, Moudafi [7] introduced a new split equality feasibility problem. Let 𝐻1 , 𝐻2 , and 𝐻3 be real Hilbert spaces. Let 𝐴 : 𝐻1 → 𝐻3 and 𝐵 : 𝐻2 → 𝐻3 be two bounded linear operators. The split equality feasibility problem is to find 𝑥 ∈ 𝐶, 𝑦∈𝑄
(2) such that 𝐴𝑥 = 𝐵𝑦,
which allows asymmetric and partial relations between the variables 𝑥 and 𝑦. The interest is to cover many situations, for instance, applications in decomposition methods for PDEs, in game theory and in intensity-modulated radiation therapy (for short, IMRT).
Moudafi [8] introduced the simultaneous iterative method to solve the split equality feasibility problem. Furthermore, Moudafi studied the fixed point formulation to avoid using the projection. Assume Fix(𝑇) and Fix(𝑆) are the sets of fixed points of 𝑇 and 𝑆, respectively, where 𝑇 : 𝐻1 → 𝐻1 and 𝑆 : 𝐻2 → 𝐻2 are nonlinear operators such that Fix(𝑇) ≠ 0 and Fix(𝑆) ≠ 0. So the split equality fixed point problem is to find 𝑥 ∈ Fix (𝑇) , 𝑦 ∈ Fix (𝑆)
(3) such that 𝐴𝑥 = 𝐵𝑦,
where 𝑇 and 𝑆 are firmly quasi-nonexpansive mappings. In order to find the solution of the split equality problem, Che and Li [9] proposed the following iterative algorithm: 𝑢𝑛 = 𝑥𝑛 − 𝛾𝑛 𝐴∗ (𝐴𝑥𝑛 − 𝐵𝑦𝑛 ) , 𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 , V𝑛 = 𝑦𝑛 + 𝛾𝑛 𝐵∗ (𝐴𝑥𝑛 − 𝐵𝑦𝑛 ) ,
(4)
𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑆V𝑛 , and the weak convergence of the scheme (4) can also be established. Furthermore, Chang et al. [10] modified the
2
Mathematical Problems in Engineering
iterative scheme (4) and provided a unified framework for solving this problem without using the projection. The framework is as follows. 𝑢𝑛 = 𝑥𝑛 − 𝛾𝑛 𝐴∗ (𝐴𝑥𝑛 − 𝐵𝑦𝑛 ) ,
V𝑛 = 𝑦𝑛 + 𝛾𝑛 𝐵∗ (𝐴𝑥𝑛 − 𝐵𝑦𝑛 ) ,
(5)
+ (1 − 𝛼𝑛 ) [(1 − 𝜉) 𝐼 + 𝜉𝑆 ((1 − 𝜂) 𝐼 + 𝜂𝑆)] V𝑛 . They got the following conclusion that the sequence (𝑥𝑛 , 𝑦𝑛 ) generated by the above modification converges weakly to a solution of problem (3). Furthermore, some authors [11–14] studied the problems (1)–(3) in Banach space. They proposed effective algorithms and proved their convergence under some conditions. Recently, He and Sun [15] studied the problem of split convex feasibility and established a strongly convergent alternating algorithm. They proposed the following iterative algorithm to find 𝑥 ∈ 𝐶, (6)
2. Preliminaries In this paper, we recall some concepts, definitions, and conclusions, which are prepared for proving our main results. We write 𝑥𝑛 ⇀ 𝑥 to indicate that the sequence {𝑥𝑛 } converges weakly to 𝑥. 𝑥𝑛 → 𝑥 implies that {𝑥𝑛 } converges strongly to 𝑥. We denote by 𝐻1 a real Hilbert space with inner product ⟨⋅, ⋅⟩ and induced norm ‖ ⋅ ‖. A mapping 𝑇 : 𝐶 → 𝐶 is called
∗ ∗ 𝑇𝑥 − 𝑥 ≤ 𝑥 − 𝑥 ,
via the formula 𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
𝐾𝑛 = {𝑧 ∈ 𝐻3 : ℎ ( + ⟨𝜍𝑛 , 𝑧 − (
When 𝐾 = 𝐻3 , problem (8) is problem (3). Therefore, problem (8) is the extension of the split equality fixed point problem. We propose the simultaneous iterative algorithm for solving this problem, which avoids using the projection and the step length sequences do not depend on the operator norms ‖𝐴‖ and ‖𝐵‖. Furthermore, we prove the sequences generated by the algorithm weakly converge to a solution of the extended split equality fixed point problem. Numerical examples show the feasibility and efficiency of this algorithm.
(i) quasi-nonexpansive, if Fix(𝑇) ≠ 0,
such that 𝐴𝑥 = 𝐵𝑦 ∈ 𝐾,
𝜔𝑛 = 𝑃𝐾𝑛 (
(8) such that 𝐴𝑥 = 𝐵𝑦 ∈ 𝐾.
𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛
𝑦∈𝑄
𝑥 ∈ Fix (𝐺) , 𝑦 ∈ Fix (𝑀)
𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) [(1 − 𝜉) 𝐼 + 𝜉𝑇 ((1 − 𝜂) 𝐼 + 𝜂𝑇)] 𝑢𝑛 ,
In this article, we study the following problem of extended split equality fixed point, which is to find
(9)
(ii) quasi-pseudo-contractive, if Fix(𝑇) ≠ 0, ∗ ∗ 𝑇𝑥 − 𝑥 ≤ 𝑥 − 𝑥 + ‖𝑇𝑥 − 𝑥‖ ,
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ) 2
∀𝑥 ∈ 𝐶, 𝑥∗ ∈ Fix (𝑇) .
𝐴𝑥𝑛 + 𝐵𝑦𝑛 )⟩ ≤ 0} , 2
𝑢𝑛 = 𝑃𝐶 (𝑥𝑛 − 𝜉𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )) , V𝑛 = 𝑃𝑄 (𝑦𝑛 − 𝜉𝐵∗ (𝐵𝑥𝑛 − 𝜔𝑛 )) ,
∀𝑥 ∈ 𝐶, 𝑥∗ ∈ Fix (𝑇) ,
(7)
2 𝐶𝑛+1 × 𝑄𝑛+1 = {(𝑥, 𝑦) ∈ 𝐶𝑛 × 𝑄𝑛 : 𝑢𝑛 − 𝑥 2 2 2 + V𝑛 − 𝑦 ≤ 𝑥𝑛 − 𝑥 + 𝑦𝑛 − 𝑦 , 𝑓 (𝑥𝑛 ) + ⟨𝜇𝑛 , 𝑥 − 𝑥𝑛 ⟩ ≤ 0, 𝑔 (𝑦𝑛 ) + ⟨𝜂𝑛 , 𝑦 − 𝑦𝑛 ⟩ ≤ 0} ,
A mapping 𝑃𝐶 is said to be metric projection of 𝐻1 onto 𝐶 if, for every point 𝑥 ∈ 𝐻1 , there exists a unique nearest point in 𝐶 denoted by 𝑃𝐶𝑥 such that 𝑥 − 𝑃𝐶𝑥 ≤ 𝑥 − 𝑦 , ∀𝑦 ∈ 𝐶.
(11)
It is well known that 𝑃𝐶 is a nonexpansive mapping and is characterized by the following properties: 2 𝑃𝐶𝑥 − 𝑃𝐶𝑦 ≤ ⟨𝑥 − 𝑦, 𝑃𝐶𝑥 − 𝑃𝐶𝑦⟩ , ∀𝑥, 𝑦 ∈ 𝐻1 ,
𝑥𝑛+1 = 𝑃𝐶𝑛+1 (𝑥1 ) ,
⟨𝑥 − 𝑃𝐶𝑥, 𝑦 − 𝑃𝐶𝑥⟩ ≤ 0,
𝑦𝑛+1 = 𝑃𝑄𝑛+1 (𝑦1 ) , ∀𝑛 ∈ 𝑁,
2 2 2 𝑥 − 𝑦 ≥ 𝑥 − 𝑃𝐶𝑥 + 𝑦 − 𝑃𝐶𝑥 ,
where 𝜉 ∈ (0, min(1/‖𝐴‖2 , 1/‖𝐵‖2 )), 𝜇𝑛 ∈ 𝜕𝑓(𝑥𝑛 ), 𝜂𝑛 ∈ 𝜕𝑔(𝑦𝑛 ), and 𝜍𝑛 ∈ 𝜕ℎ((𝐴𝑥𝑛 + 𝐵𝑦𝑛 )/2). In some conditions such as Ω = {(𝑥, 𝑦) ∈ 𝐶 × 𝑄 : 𝐴𝑥 = 𝐵𝑦 ∈ 𝐾} ≠ 0, they proved that the sequences (𝑥𝑛 , 𝑦𝑛 ) → (𝑠, 𝑡), (𝑢𝑛 , V𝑛 ) → (𝑠, 𝑡), and 𝜔𝑛 → 𝑧∗ fl 𝐴𝑠 = 𝐵𝑡 ∈ 𝐾, where (𝑠, 𝑡) ∈ Ω.
(10)
∀𝑥 ∈ 𝐻1 , 𝑦 ∈ 𝐶,
∀𝑥 ∈ 𝐻1 , 𝑦 ∈ 𝐶, 2 (𝑥 − 𝑦) − (𝑃𝐶𝑥 − 𝑃𝐶𝑦) 2 2 ≥ 𝑥 − 𝑦 − 𝑃𝐶𝑥 − 𝑃𝐶𝑦 , ∀𝑥, 𝑦 ∈ 𝐻1 .
(12)
Mathematical Problems in Engineering
3
In the proof of our results, we need the following lemmas.
be sequences generated by
Lemma 1 (see [16]). Let 𝐻 be a real Hilbert space; then the following conclusions hold.
𝜔𝑛 = 𝑃𝐾𝑛 (
2 2 2 𝑥 − 𝑦 = ‖𝑥‖ + 𝑦 − 2 ⟨𝑥, 𝑦⟩ ,
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
𝐾𝑛 = {𝑧 ∈ 𝐻3 : ℎ (
∀𝑥, 𝑦 ∈ 𝐻, 2 2 2 𝛼𝑥 + (1 − 𝛼) 𝑦 = 𝛼 ‖𝑥‖ + (1 − 𝛼) 𝑦 − 𝛼 (1 − 𝛼) 𝑥 − 𝑦 ,
+ ⟨𝜍𝑛 , 𝑧 −
(13)
(14)
If 0 < 𝜉 < 𝜂 < 1/(1 + √1 + 𝐿2 ), then the following conclusions hold. (i) Fix(𝑇) = Fix(𝑇((1 − 𝜂)𝐼 + 𝜂𝑇)) = Fix(𝐺). (ii) If 𝑇 is demiclosed at 0, then 𝐺 is also demiclosed at 0. (iii) In addition, if 𝑇 : 𝐻 → H is quasi-pseudo-contractive, then the mapping 𝐺 is quasi-nonexpansive; that is,
∀𝑥 ∈ 𝐻, 𝑢∗ ∈ Fix (𝑇) = Fix (𝐺) .
(15)
3. Main Results In this section, we assume that (i) 𝐺 : 𝐻1 → 𝐻1 and 𝑀 : 𝐻2 → 𝐻2 are two 𝐿Lipschitzian and quasi-pseudo-contractive mappings with 𝐿 ≥ 1, Fix(𝐺) ≠ 0, and Fix(𝑀) ≠ 0. (ii) 𝐻1 , 𝐻2 , and 𝐻3 be real Hilbert spaces. 𝐴 : 𝐻1 → 𝐻3 and 𝐵 : 𝐻2 → 𝐻3 are two bounded linear operators; 𝐴∗ and 𝐵∗ are their adjoint operators, respectively. Let 𝐾 ⊂ 𝐻3 be nonempty closed convex set. We consider the extended split equality fixed point problem (8). Theorem 3. Let 𝐻1 , 𝐻2 , and 𝐻3 be real Hilbert spaces and 𝐾 be a closed convex level set 𝐾 = {𝑥 ∈ 𝐻3 : ℎ (𝑥) ≤ 0} ,
(17)
𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 , V𝑛 = 𝑦𝑛 − 𝜉𝑛 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) ,
Lemma 2 (see [16]). Let 𝐻 be a real Hilbert space and 𝑇 : 𝐻 → 𝐻 be a 𝐿-Lipschitzian mapping with 𝐿 ≥ 1. Denote
∗ ∗ 𝐺𝑥 − 𝑢 ≤ 𝑥 − 𝑢 ,
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ⟩ ≤ 0} , 2
𝑢𝑛 = 𝑥𝑛 − 𝜉𝑛 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) ,
∀𝑥, 𝑦 ∈ 𝐻, 𝛼 ∈ [0, 1] .
𝐺 fl (1 − 𝜉) 𝐼 + 𝜉𝑇 ((1 − 𝜂) 𝐼 + 𝜂𝑇) .
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ) 2
(16)
where ℎ : 𝐻3 → 𝑅 is convex function which is subdifferentiable on 𝐾 and its subdifferentials are bounded on bounded sets. 𝐴 : 𝐻1 → 𝐻3 and 𝐵 : 𝐻2 → 𝐻3 are two bounded linear operators with their adjoint operators 𝐴∗ and 𝐵∗ , respectively, 𝜎 ∈ (0, 1) is a parameter controlling step length, 0 < 𝛾 < 𝜂 < 1/(1 + √1 + 𝐿2 ), and {𝛼𝑛 } ⊂ (0, 1). Let {𝑥𝑛 }, {𝑦𝑛 }, {𝜔𝑛 }, {𝑢𝑛 }, and {V𝑛 }
𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑆V𝑛 , where 𝜍𝑛 ∈ 𝜕ℎ (
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
𝜉𝑛 ∈ (0, (1 − 𝜎) 2 2 𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 ⋅ ), 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )2 + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 )2
(18)
𝑇 fl (1 − 𝛾) 𝐼 + 𝛾𝐺 ((1 − 𝜂) 𝐼 + 𝜂𝐺) , 𝑆 fl (1 − 𝛾) 𝐼 + 𝛾𝑀 ((1 − 𝜂) 𝐼 + 𝜂𝑀) , where 𝐺 and 𝑀 are demiclosed at 0. Assume Ω = {(𝑥, 𝑦) ∈ Fix(𝐺) × Fix(𝑀) : 𝐴𝑥 = 𝐵𝑦 ∈ 𝐾} ≠ 0. Then (𝑥𝑛 , 𝑦𝑛 ) ⇀ (𝑡, 𝑠), (𝑢𝑛 , V𝑛 ) ⇀ (𝑡, 𝑠), and 𝜔𝑛 ⇀ 𝑧∗ fl 𝐴𝑡 = 𝐵𝑠, where (𝑡, 𝑠) ∈ Ω. Proof. It is obvious that 𝐾 ⊂ 𝐾𝑛 for any 𝑛 ∈ 𝑁. Let (𝑥∗ , 𝑦∗ ) ∈ Ω; namely, 𝑥∗ ∈ Fix(𝐺), 𝑦∗ ∈ Fix(𝑀), and 𝜔∗ = 𝐴𝑥∗ = 𝐵𝑦∗ ∈ 𝐾. By Lemma 1, we have 2 𝐴𝑥 + 𝐵𝑦𝑛 ∗ ) − 𝑃𝐾𝑛 (𝜔∗ ) 𝜔𝑛 − 𝜔 = 𝑃𝐾𝑛 ( 𝑛 2 2 1 1 ≤ (𝐴𝑥𝑛 − 𝜔∗ ) + (𝐵𝑦𝑛 − 𝜔∗ ) 2 2 1 2 1 2 = 𝐴𝑥𝑛 − 𝜔∗ + 𝐵𝑦𝑛 − 𝜔∗ 2 2 −
(19)
1 2 𝐴𝑥 − 𝐵𝑦𝑛 . 4 𝑛
2 ∗ 2 ∗ ∗ 𝑢𝑛 − 𝑥 = 𝑥𝑛 − 𝑥 − 𝜉𝑛 𝐴 (𝐴𝑥𝑛 − 𝜔𝑛 ) 2 2 = 𝑥𝑛 − 𝑥∗ + 𝜉𝑛2 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) − 2 ⟨𝑥𝑛 − 𝑥∗ , 𝜉𝑛 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )⟩ .
(20)
4
Mathematical Problems in Engineering
By (20), we have 2 ∗ 2 ∗ 2 2 ∗ 𝑢𝑛 − 𝑥 = 𝑥𝑛 − 𝑥 + 𝜉𝑛 𝐴 (𝐴𝑥𝑛 − 𝜔𝑛 ) 2 2 − 𝜉𝑛 𝐴𝑥𝑛 − 𝜔∗ − 𝜉𝑛 𝐴𝑥𝑛 − 𝜔𝑛 2 + 𝜉𝑛 𝜔𝑛 − 𝜔∗ .
(21)
2 2 ⋅ 𝑦𝑛 − 𝑦∗ + (1 − 𝛼𝑛 ) 𝜉𝑛2 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) 2 − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐵𝑦𝑛 − 𝜔∗ − 𝜉𝑛 (1 − 𝛼𝑛 )
Similarly, 2 ∗ 2 ∗ 2 2 ∗ V𝑛 − 𝑦 = 𝑦𝑛 − 𝑦 + 𝜉𝑛 𝐵 (𝐵𝑦𝑛 − 𝜔𝑛 ) 2 2 − 𝜉𝑛 𝐵𝑦𝑛 − 𝜔∗ − 𝜉𝑛 𝐵𝑦𝑛 − 𝜔𝑛 2 + 𝜉𝑛 𝜔𝑛 − 𝜔∗ .
2 2 1 ⋅ 𝐴𝑥𝑛 − 𝜔∗ + 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐵𝑦𝑛 − 𝜔∗ − 2 2 ⋅ 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐴𝑥𝑛 − 𝐵𝑦𝑛 − 𝛼𝑛 (1 − 𝛼𝑛 ) 2 2 ⋅ 𝑥𝑛 − 𝑇𝑢𝑛 + 𝛼𝑛 𝑦𝑛 − 𝑦∗ + (1 − 𝛼𝑛 )
2 2 ⋅ 𝐵𝑦𝑛 − 𝜔𝑛 − 𝛼𝑛 (1 − 𝛼𝑛 ) 𝑦𝑛 − 𝑆V𝑛 (22)
By (21), (22), (17), and (19), we have ∗ 2 ∗ 2 𝑥𝑛+1 − 𝑥 + 𝑦𝑛+1 − 𝑦 2 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 − 𝑥∗ 2 + 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑆V𝑛 − 𝑦∗ 2 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 − (1 − 𝛼𝑛 ) 𝑥∗ − 𝛼𝑛 𝑥∗ 2 + 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑆V𝑛 − (1 − 𝛼𝑛 ) 𝑦∗ − 𝛼𝑛 𝑦∗ 2 2 = 𝛼𝑛 𝑥𝑛 − 𝑥∗ + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 − 𝑥∗ 2 2 − 𝛼𝑛 (1 − 𝛼𝑛 ) 𝑥𝑛 − 𝑇𝑢𝑛 + 𝛼𝑛 𝑦𝑛 − 𝑦∗ 2 2 + (1 − 𝛼𝑛 ) 𝑆V𝑛 − 𝑦∗ − 𝛼𝑛 (1 − 𝛼𝑛 ) 𝑦𝑛 − 𝑆V𝑛 2 2 ≤ 𝛼𝑛 𝑥𝑛 − 𝑥∗ + (1 − 𝛼𝑛 ) 𝑢𝑛 − 𝑥∗ 2 2 − 𝛼𝑛 (1 − 𝛼𝑛 ) (1 − 𝛼𝑛 ) 𝑥𝑛 − 𝑇𝑢𝑛 + 𝛼𝑛 𝑦𝑛 − 𝑦∗ 2 2 + (1 − 𝛼𝑛 ) V𝑛 − 𝑦∗ − 𝛼𝑛 (1 − 𝛼𝑛 ) 𝑦𝑛 − 𝑆V𝑛 2 2 = 𝛼𝑛 𝑥𝑛 − 𝑥∗ + (1 − 𝛼𝑛 ) 𝑥𝑛 − 𝑥∗ + (1 − 𝛼𝑛 ) 2 2 ⋅ 𝜉𝑛2 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐴𝑥𝑛 − 𝜔∗ 2 − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐴𝑥𝑛 − 𝜔𝑛 + 2𝜉𝑛 (1 − 𝛼𝑛 ) 2 2 ⋅ 𝜔𝑛 − 𝜔∗ − 𝛼𝑛 (1 − 𝛼𝑛 ) 𝑥𝑛 − 𝑇𝑢𝑛 2 2 + 𝛼𝑛 𝑦𝑛 − 𝑦∗ + (1 − 𝛼𝑛 ) 𝑦𝑛 − 𝑦∗ + (1 − 𝛼𝑛 ) 2 2 ⋅ 𝜉𝑛2 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐵𝑦𝑛 − 𝜔∗ 2 2 − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐵𝑦𝑛 − 𝜔𝑛 − 𝛼𝑛 (1 − 𝛼𝑛 ) 𝑦𝑛 − 𝑆V𝑛 2 2 ≤ 𝛼𝑛 𝑥𝑛 − 𝑥∗ + (1 − 𝛼𝑛 ) 𝑥𝑛 − 𝑥∗ + (1 − 𝛼𝑛 ) 2 2 ⋅ 𝜉𝑛2 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐴𝑥𝑛 − 𝜔∗ 2 − 𝜉𝑛 (1 − 𝛼𝑛 ) 𝐴𝑥𝑛 − 𝜔𝑛 + 𝜉𝑛 (1 − 𝛼𝑛 )
2 2 = 𝑥𝑛 − 𝑥∗ + 𝑦𝑛 − 𝑦∗ + (1 − 𝛼𝑛 ) 2 2 ⋅ 𝜉𝑛2 (𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) ) 2 2 − (1 − 𝛼𝑛 ) 𝜉𝑛 (𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 ) 2 2 − 𝛼𝑛 (1 − 𝛼𝑛 ) (𝑥𝑛 − 𝑇𝑢𝑛 + 𝑦𝑛 − 𝑆V𝑛 ) −
1 2 (1 − 𝛼𝑛 ) 𝜉𝑛 𝐴𝑥𝑛 − 𝐵𝑦𝑛 . 2 (23)
Notice that ∗ 2 ∗ 2 ∗ 2 𝑥𝑛+1 − 𝑥 + 𝑦𝑛+1 − 𝑦 ≤ 𝑥𝑛 − 𝑥 + 𝑦𝑛 2 − 𝑦∗ + 𝜉𝑛 (1 − 𝛼𝑛 ) 2 2 ⋅ [𝜉𝑛 (𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) )
(24)
2 2 − (𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 )] . Thanks to 𝜉𝑛 ∈ (0, (1 − 𝜎) 2 2 𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 ), ⋅ 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )2 + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 )2
(25)
we obtain ∗ 2 ∗ 2 𝑥𝑛+1 − 𝑥 + 𝑦𝑛+1 − 𝑦 2 2 ≤ 𝑥𝑛 − 𝑥∗ + 𝑦𝑛 − 𝑦∗ .
(26)
2 2 𝑋𝑛 (𝑥∗ , 𝑦∗ ) = 𝑥𝑛 − 𝑥∗ + 𝑦𝑛 − 𝑦∗ .
(27)
𝑋𝑛+1 (𝑥∗ , 𝑦∗ ) ≤ 𝑋𝑛 (𝑥∗ , 𝑦∗ ) .
(28)
Let
We obtain
This implies that 𝑋𝑛 (𝑥∗ , 𝑦∗ ) is a nonincreasing sequence; hence lim𝑛→∞ 𝑋𝑛 (𝑥∗ , 𝑦∗ ) exists. As a result, {𝑥𝑛 } and {𝑦𝑛 } are
Mathematical Problems in Engineering
5 that is,
bounded sequences. Rewrite (23) as 2 2 (1 − 𝛼𝑛 ) 𝜉𝑛 [(𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 )
𝐴𝑡 = 𝐵𝑠. ∗
2 2 − 𝜉𝑛 (𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) )] 1 2 2 + 𝛼𝑛 (1 − 𝛼𝑛 ) (𝑥𝑛 − 𝑇𝑢𝑛 + 𝑦𝑛 − 𝑆V𝑛 ) + (1 2 (29) 2 2 2 ∗ ∗ − 𝛼𝑛 ) 𝜉𝑛 𝐴𝑥𝑛 − 𝐵𝑦𝑛 ≤ 𝑥𝑛 − 𝑥 + 𝑦𝑛 − 𝑦 2 2 − (𝑥𝑛+1 − 𝑥∗ + 𝑦𝑛+1 − 𝑦∗ ) = 𝑋𝑛 (𝑥∗ , 𝑦∗ ) − 𝑋𝑛+1 (𝑥∗ , 𝑦∗ ) . 𝐴𝑥𝑛 − 𝐵𝑦𝑛 → 0, 𝑇𝑢𝑛 − 𝑥𝑛 → 0, 𝑆V𝑛 − 𝑦𝑛 → 0, 𝐴𝑥𝑛 − 𝜔𝑛 → 0, 𝐵𝑦𝑛 − 𝜔𝑛 → 0.
By (31), we have 𝜔𝑛 ⇀ 𝑧 fl 𝐴𝑡 = 𝐵𝑠. Now, we prove 𝑧 ∈ 𝐾. We know that {𝜍𝑛 } is bounded. There exists 𝑀 > 0, such that ‖𝜍𝑛 ‖ ≤ 𝑀, where 𝑀 is a constant. Note that 𝜔𝑛 = 𝑃𝐾𝑛 ((𝐴𝑥𝑛 + 𝐵𝑦𝑛 )/2) ∈ 𝐾𝑛 , and we have ℎ(
𝐴𝑥𝑛 + 𝐵𝑦𝑛 𝐴𝑥𝑛 + 𝐵𝑦𝑛 ) + ⟨𝜍𝑛 , 𝜔𝑛 − ⟩ ≤ 0. 2 2
(37)
𝐴𝑥𝑛 + 𝐵𝑦𝑛 𝐴𝑥𝑛 + 𝐵𝑦𝑛 ) ≤ − ⟨𝜍𝑛 , 𝜔𝑛 − ⟩ 2 2 𝐴𝑥𝑛 + 𝐵𝑦𝑛 ≤ 𝑀 𝜔𝑛 − . 2
(38)
Hence, ℎ(
Letting 𝑛 → ∞ and taking the limit in (29), we have
(30)
(36) ∗
By the lower semicontinuity of ℎ, (38), and (31), we obtain 𝐴𝑥𝑛 + 𝐵𝑦𝑛 ) 2 𝐴𝑥𝑛 + 𝐵𝑦𝑛 ≤ 𝑀 lim inf 𝜔𝑛 − = 0. 𝑛→∞ 2
ℎ (𝑧∗ ) ≤ lim inf ℎ ( 𝑛→∞
(31)
(39)
Thus 𝑧∗ ∈ 𝐾. The proof is completed.
Then,
− 𝑥𝑛 = 0,
4. Consequent Results
− 𝑦𝑛 = 0,
In this section, we give some corollaries, which are easily obtained from Theorem 3. If 𝐺 = 𝑀, we have the following corollary.
lim 𝑢 𝑛→∞ 𝑛 lim V 𝑛→∞ 𝑛
lim 𝑥 − 𝑥𝑛 = lim 𝑇𝑢𝑛 − 𝑥𝑛 = 0, 𝑛→∞ 𝑛+1 𝑛→∞
(32)
lim 𝑦 − 𝑦𝑛 = lim 𝑆V𝑛 − 𝑦𝑛 = 0, 𝑛→∞ 𝑛+1 𝑛→∞
Corollary 4. Let 𝐻1 , 𝐻2 , and 𝐻3 be real Hilbert spaces and 𝐾 be a closed convex level set
which imply that {𝑥𝑛 } and {𝑦𝑛 } are asymptotically regular. Furthermore, we get
𝐾 = {𝑥 ∈ 𝐻3 : ℎ (𝑥) ≤ 0} ,
lim 𝑇𝑢𝑛 − 𝑢𝑛 = 0, 𝑛→∞ lim 𝑆V − V𝑛 = 0. 𝑛→∞ 𝑛
(33)
Since {𝑥𝑛 } and {𝑦𝑛 } are bounded sequences, there exist weakly convergent subsequences, say {𝑥𝑛𝑖 } ⊂ {𝑥𝑛 } such that 𝑥𝑛𝑖 ⇀ 𝑡; also {𝑦𝑛𝑖 } ⊂ {𝑦𝑛 } such that 𝑦𝑛𝑖 ⇀ 𝑠. The Opial property guarantees that the weakly subsequential limit of {(𝑥𝑛 , 𝑦𝑛 )} is unique. So we have 𝑥𝑛 ⇀ 𝑡, 𝑦𝑛 ⇀ 𝑠. Therefore 𝑢𝑛 ⇀ 𝑡, V𝑛 ⇀ 𝑠. Since 𝐺 and 𝑀 are demiclosed at 0, and from Lemma 2, by (33), we have 𝑇𝑡 = 𝑡, 𝑆𝑠 = 𝑠, which imply that 𝑡 ∈ Fix(𝐺), 𝑠 ∈ Fix(𝑀). Hence, (𝑡, 𝑠) ∈ Fix (𝐺) × Fix (𝑀) .
(34)
Furthermore, since 𝐴𝑥𝑛 −𝐵𝑦𝑛 ⇀ 𝐴𝑡−𝐵𝑠, by using the weakly lower semicontinuity of squared norm, we have inf 𝐴𝑥𝑛 − 𝐵𝑦𝑛 = lim 𝐴𝑥𝑛 − 𝐵𝑦𝑛 ‖𝐴𝑡 − 𝐵𝑠‖ ≤ lim 𝑛→∞ 𝑛→∞ = 0;
(35)
(40)
where ℎ : 𝐻3 → 𝑅 is convex function which is subdifferentiable on 𝐾 and its subdifferentials are bounded on bounded sets. 𝐴 : 𝐻1 → 𝐻3 and 𝐵 : 𝐻2 → 𝐻3 are two bounded linear operators with their adjoint operators 𝐴∗ and 𝐵∗ , respectively, 𝜎 ∈ (0, 1) is a parameter controlling step length, 0 < 𝛾 < 𝜂 < 1/(1 + √1 + 𝐿2 ), and {𝛼𝑛 } ⊂ (0, 1). Let {𝑥𝑛 }, {𝑦𝑛 }, {𝜔𝑛 }, {𝑢𝑛 }, and {V𝑛 } be sequences generated by 𝜔𝑛 = 𝑃𝐾𝑛 (
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
𝐾𝑛 = {𝑧 ∈ 𝐻3 : ℎ ( + ⟨𝜍𝑛 , 𝑧 −
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ) 2
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ⟩ ≤ 0} , 2
𝑢𝑛 = 𝑥𝑛 − 𝜉𝑛 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) , 𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 , V𝑛 = 𝑦𝑛 − 𝜉𝑛 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 ) , 𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑇V𝑛 ,
(41)
6
Mathematical Problems in Engineering where 𝐺 and 𝑀 are demiclosed at 0. Assume Ω = {(𝑥, 𝑦) ∈ Fix(𝐺) × Fix(𝑀) : 𝐴𝑥 = 𝐴𝑦 ∈ 𝐾} ≠ 0. Then (𝑥𝑛 , 𝑦𝑛 ) ⇀ (𝑠, 𝑡), (𝑢𝑛 , V𝑛 ) ⇀ (𝑠, 𝑡), and 𝜔𝑛 ⇀ 𝑧∗ fl 𝐴𝑠 = 𝐴𝑡, where (𝑠, 𝑡) ∈ Ω.
where 𝜍𝑛 ∈ 𝜕ℎ (
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
If 𝐺 = 𝑀 and 𝐴 = 𝐵, we have the following corollary.
𝜉𝑛 ∈ (0, (1 − 𝜎) 2 2 𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 ), ⋅ 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )2 + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 )2
(42)
Corollary 6. Let 𝐻1 and 𝐻3 be real Hilbert spaces and 𝐾 be a closed convex level set 𝐾 = {𝑥 ∈ 𝐻3 : ℎ (𝑥) ≤ 0} ,
𝑇 fl (1 − 𝛾) 𝐼 + 𝛾𝐺 ((1 − 𝜂) 𝐼 + 𝜂𝐺) , where 𝐺 is demiclosed at 0. Assume Ω = {(𝑥, 𝑦) ∈ Fix(𝐺) × Fix(𝐺) : 𝐴𝑥 = 𝐵𝑦 ∈ 𝐾} ≠ 0. Then (𝑥𝑛 , 𝑦𝑛 ) ⇀ (𝑡, 𝑠), (𝑢𝑛 , V𝑛 ) ⇀ (𝑡, 𝑠) and 𝜔𝑛 ⇀ 𝑧∗ fl 𝐴𝑡 = 𝐵𝑠, where (𝑡, 𝑠) ∈ Ω. If 𝐴 = 𝐵, we have the following corollary.
where ℎ : 𝐻3 → 𝑅 is convex function which is subdifferentiable on 𝐾 and its subdifferentials are bounded on bounded sets. 𝐴 : 𝐻1 → 𝐻3 is a bounded linear operator with its adjoint operator 𝐴∗ , and 𝜎 ∈ (0, 1) is a parameter controlling step length, 0 < 𝛾 < 𝜂 < 1/(1 + √1 + 𝐿2 ), and {𝛼𝑛 } ⊂ (0, 1). Let {𝑥𝑛 }, {𝑦𝑛 }, {𝜔𝑛 }, {𝑢𝑛 }, and {V𝑛 } be sequences generated by
Corollary 5. Let 𝐻1 and 𝐻3 be real Hilbert spaces and 𝐾 be a closed convex level set 𝐾 = {𝑥 ∈ 𝐻3 : ℎ (𝑥) ≤ 0} ,
𝜔𝑛 = 𝑃𝐾𝑛 (
𝐴𝑥𝑛 + 𝐴𝑦𝑛 ⟩ ≤ 0} , 2
𝑢𝑛 = 𝑥𝑛 − 𝜉𝑛 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 ) ,
∗
(44)
𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 , V𝑛 = 𝑦𝑛 − 𝜉𝑛 𝐴∗ (𝐴𝑦𝑛 − 𝜔𝑛 ) ,
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
𝜉𝑛 ∈ (0, (1 − 𝜎) 2 2 𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 ), ⋅ 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )2 + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 )2
𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑆V𝑛 ,
(48)
𝑇 fl (1 − 𝛾) 𝐼 + 𝛾𝐺 ((1 − 𝜂) 𝐼 + 𝜂𝐺) ,
where
where 𝐺 is demiclosed at 0. Assume Ω = {(𝑥, 𝑦) ∈ Fix(𝐺) × Fix(𝐺) : 𝐴𝑥 = 𝐴𝑦 ∈ 𝐾} ≠ 0. Then (𝑥𝑛 , 𝑦𝑛 ) ⇀ (𝑡, 𝑠), (𝑢𝑛 , V𝑛 ) ⇀ (𝑡, 𝑠), and 𝜔𝑛 ⇀ 𝑧∗ fl 𝐴𝑡 = 𝐴𝑠, where (𝑡, 𝑠) ∈ Ω.
𝐴𝑥𝑛 + 𝐵𝑦𝑛 ), 2
If 𝐺 = 𝑀 = 𝐴 = 𝐵 is an identity operator, we have the following corollary.
𝜉𝑛 ∈ (0, (1 − 𝜎)
𝑆 fl (1 − 𝛾) 𝐼 + 𝛾𝑀 ((1 − 𝜂) 𝐼 + 𝜂𝑀) ,
(47)
𝑢𝑛 = 𝑥𝑛 − 𝜉𝑛 𝐴 (𝐴𝑥𝑛 − 𝜔𝑛 ) ,
𝜍𝑛 ∈ 𝜕ℎ (
V𝑛 = 𝑦𝑛 − 𝜉𝑛 𝐴∗ (𝐴𝑦𝑛 − 𝜔𝑛 ) ,
𝑇 fl (1 − 𝛾) 𝐼 + 𝛾𝐺 ((1 − 𝜂) 𝐼 + 𝜂𝐺) ,
𝐴𝑥𝑛 + 𝐴𝑦𝑛 ⟩ ≤ 0} , 2
where
𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑇𝑢𝑛 ,
2 2 𝐴𝑥𝑛 − 𝜔𝑛 + 𝐵𝑦𝑛 − 𝜔𝑛 ⋅ ), 𝐴∗ (𝐴𝑥𝑛 − 𝜔𝑛 )2 + 𝐵∗ (𝐵𝑦𝑛 − 𝜔𝑛 )2
𝐴𝑥𝑛 + 𝐴𝑦𝑛 ) 2
𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) 𝑇V𝑛 ,
𝐴𝑥 + 𝐴𝑦𝑛 𝐾𝑛 = {𝑧 ∈ 𝐻3 : ℎ ( 𝑛 ) 2
𝜍𝑛 ∈ 𝜕ℎ (
+ ⟨𝜍𝑛 , 𝑧 −
𝐴𝑥𝑛 + 𝐴𝑦𝑛 ), 2
+ ⟨𝜍𝑛 , 𝑧 −
𝐴𝑥𝑛 + 𝐴𝑦𝑛 ), 2
𝐾𝑛 = {𝑧 ∈ 𝐻3 : ℎ (
(43)
where ℎ : 𝐻3 → 𝑅 is convex function which is subdifferentiable on 𝐾 and its subdifferentials are bounded on bounded sets. 𝐴 : 𝐻1 → 𝐻3 is a bounded linear operator with its adjoint operator 𝐴∗ , 𝜎 ∈ (0, 1) is a parameter controlling step length, 0 < 𝛾 < 𝜂 < 1/(1 + √1 + 𝐿2 ), and {𝛼𝑛 } ⊂ (0, 1). Let {𝑥𝑛 }, {𝑦𝑛 }, {𝜔𝑛 }, {𝑢𝑛 }, and {V𝑛 } be sequences generated by 𝜔𝑛 = 𝑃𝐾𝑛 (
(46)
(45)
Corollary 7. Let 𝐻1 and 𝐻3 be real Hilbert spaces and 𝐾 be a closed convex level set 𝐾 = {𝑥 ∈ 𝐻3 : ℎ (𝑥) ≤ 0} ,
(49)
where ℎ : 𝐻3 → 𝑅 is convex function which is subdifferentiable on 𝐾 and its subdifferentials are bounded on bounded sets.
Mathematical Problems in Engineering
7 ×104 16
25
14 20
12 10
15
8 10
6 4
5
2 0
0
5
10 15 20 Number of iterations
25
30
0
0
norm(Axn − Byn )
5
10
15 20 25 30 Number of iterations
35
40
45
35
40
45
norm(Axn − Byn )
×107 16000
3
14000
2.5
12000 2
10000 8000
1.5
6000
1
4000 0.5 0
2000 0
10
20 30 40 Number of iterations
50
60
norm(Axn − Byn )
0
0
5
10
15 20 25 30 Number of iterations
norm(Axn − Byn )
Figure 1: The behaviors of ‖𝐴𝑥𝑛 − 𝐵𝑦𝑛 ‖ for different initial points.
𝜎 ∈ (0, 1) is a parameter controlling step length, 0 < 𝛾 < 𝜂 < 1/(1 + √1 + 𝐿2 ), and {𝛼𝑛 } ⊂ (0, 1). Let {𝑥𝑛 }, {𝑦𝑛 }, {𝜔𝑛 }, {𝑢𝑛 }, and {V𝑛 } be sequences generated by 𝑥𝑛 + 𝑦𝑛 ), 2 𝑥 + 𝑦𝑛 𝑥 + 𝑦𝑛 ) + ⟨𝜍𝑛 , 𝑧 − 𝑛 ⟩ 𝐾𝑛 = {𝑧 ∈ 𝐻3 : ℎ ( 𝑛 2 2
5. Numerical Examples
𝜔𝑛 = 𝑃𝐾𝑛 (
≤ 0} , 𝑢𝑛 = (1 − 𝜉) 𝑥𝑛 + 𝜉𝜔𝑛 , 𝑥𝑛+1 = 𝛼𝑛 𝑥𝑛 + (1 − 𝛼𝑛 ) 𝑢𝑛 , V𝑛 = (1 − 𝜉) 𝑦𝑛 + 𝜉𝜔𝑛 , 𝑦𝑛+1 = 𝛼𝑛 𝑦𝑛 + (1 − 𝛼𝑛 ) V𝑛 ,
where 𝜍𝑛 ∈ 𝜕ℎ((𝑥𝑛 + 𝑦𝑛 )/2), 𝜉 ∈ (0, 1 − 𝜎). Assume Ω = {(𝑥, 𝑦) ∈ 𝐻1 × 𝐻1 : 𝑥 = 𝑦 ∈ 𝐾} ≠ 0. Then (𝑥𝑛 , 𝑦𝑛 ) ⇀ (𝑡, 𝑠), (𝑢𝑛 , V𝑛 ) ⇀ (𝑡, 𝑠), and 𝜔𝑛 ⇀ 𝑧∗ fl 𝑡 = 𝑠, where (𝑡, 𝑠) ∈ Ω.
(50)
In this section, we give an example to show some insight into the behavior of the algorithm presented in this paper. The whole codes are written in Matlab 7.0. All the numerical results are carried out on a personal Lenovo ThinkPad computer with Intel(R) Core(TM) i7-6500U CPU 2.50 GHz and RAM 8.00 GB. Let 𝐻1 = 𝑅2 , 𝐻2 = 𝑅3 . 𝐴 ∈ 𝑅3×2 and 𝐵 ∈ 𝑅3×3 are as follows. 0.8147 0.9134 𝐴 = (0.9058 0.6324) , 0.1270 0.0975
8
Mathematical Problems in Engineering Table 1: Number of iterations and the CPU time for different initial points. Initial point 𝑇
𝑥 = (1, 1) , 𝑥 = (5.2378, 2.6487)𝑇 , 𝑥 = (0.5221, 19.0936)𝑇 , 𝑥 = (0.0588, 5.4403)𝑇 ,
𝑦 = (1, 1, 1) 𝑦 = (6.8357, 43.6327, 17.3853)𝑇 𝑦 = (86.1193, 192.3197, 152.4829)𝑇 𝑦 = (14.1190, 12.9026, 11.0462)𝑇
36
38
34
37
32
36
30
35
28
34
26
33
24
32
22
31
20
30
18
0
100
200
300
400
500
29
60
83
59
82
58
81
57
80
56
79
55
78
54
77
53
76
52
75
51
0
100
Iter. 27 42 50 38
𝑇
200
300
400
500
74
Sec. 0.01 0.016 0.017 0.012
0
100
200
300
400
500
0
100
200
300
400
500
Figure 2: The behaviors of iterative numbers for different initial points.
0.2785 0.9649 0.9572 𝐵 = (0.5469 0.1576 0.4854) . 0.9575 0.9706 0.8003
𝑥 = (5.2378, 2.6487)𝑇 , 𝑦 = (6.8357, 43.6327, 17.3853)𝑇 , (51)
𝑛
𝑥13
𝑥22
Let 𝐾 = {𝑥 ∈ 𝑅 | ℎ(𝑥) = + + 𝑥3 − 2 ≤ 0}, where 𝑥 = (𝑥1 , 𝑥2 , 𝑥3 )𝑇 . For convenience, we take 𝜎 = 0.2, 𝛼𝑛 = 1/3 + 1/2𝑛 , and 𝐺 = 𝑀 = (1/2)𝐼. We choose ‖𝐴𝑥𝑛 − 𝐵𝑦𝑛 ‖ ≤ 10−6 as the stopping criterion. Figure 1 presents the behaviors of ‖𝐴𝑥𝑛 − 𝐵𝑦𝑛 ‖ in different initial points such as 𝑥 = (1, 1)𝑇 , 𝑦 = (1, 1, 1)𝑇 ,
𝑥 = (0.5221, 19.0936)𝑇 , 𝑦 = (86.1193, 192.3197, 152.4829)𝑇 , 𝑥 = (0.0588, 5.4403)𝑇 , 𝑦 = (14.1190, 12.9026, 11.0462)𝑇 . (52) It is easy to see that the presentation reveals that 𝐴𝑥 = 𝐵𝑦. Table 1 shows the number of iterations and the CPU time
Mathematical Problems in Engineering
9
for the above four initial points. We denote Iter. and Sec. as the number of iterations and the CPU time in seconds, respectively. Furthermore, for testing the stationary property of iterative numbers, we carry out 500 experiments for different initial points which are presented randomly, such as 𝑥 = rand (2, 1) , 𝑦 = rand (3, 1) , 𝑥 = 10 ∗ rand (2, 1) , 𝑦 = 10 ∗ rand (3, 1) , 𝑥 = 1000 ∗ rand (2, 1) ,
(53)
𝑦 = 600 ∗ rand (3, 1) , 𝑥 = 100000 ∗ rand (2, 1) , 𝑦 = 60000 ∗ rand (3, 1) , separately. In these cases, we take 𝜎 = 0.2, 𝛼𝑛 = 0.4 and choose ‖𝐴𝑥𝑛 − 𝑤𝑛 ‖ + ‖𝐵𝑦𝑛 − 𝑤𝑛 ‖ ≤ 10−6 as the stopping criterion. Figure 2 illustrates the behaviors of iterative numbers for different initial points, which reveals the stationary property of iterative numbers of the algorithm.
Conflicts of Interest The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments This project is supported by the Natural Science Foundation of China (Grant nos. 11401438 and 11571120) and Shandong Provincial Natural Science Foundation (Grant no. ZR2013FL032) and the Project of Shandong Province Higher Educational Science and Technology Program (Grant no. J14LI52).
References [1] Y. Censor, T. Bortfeld, B. Martin, and A. Trofimov, “A unified approach for inversion problems in intensity-modulated radiation therapy,” Physics in Medicine and Biology, vol. 51, no. 10, pp. 2353–2365, 2006. [2] Y. Censor, A. Motova, and A. Segal, “Perturbed projections and subgradient projections for the multiple-sets split feasibility problem,” Journal of Mathematical Analysis and Applications, vol. 327, no. 2, pp. 1244–1256, 2007. [3] Q. Yang, “The relaxed CQ algorithm solving the split feasibility problem,” Inverse Problems, vol. 20, no. 4, pp. 1261–1266, 2004. [4] B. Qu and N. Xiu, “A new halfspace-relaxation projection method for the split feasibility problem,” Linear Algebra and Its Applications, vol. 428, no. 5-6, pp. 1218–1229, 2008. [5] Y. Censor, T. Elfving, N. Kopf, and T. Bortfeld, “The multiplesets split feasibility problem and its applications for inverse problems,” Inverse Problems, vol. 21, no. 6, pp. 2071–2084, 2005.
[6] H.-K. Xu, “Iterative methods for the split feasibility problem in infinite-dimensional Hilbert spaces,” Inverse Problems, vol. 26, no. 10, Article ID 105018, 17 pages, 2010. [7] A. Moudafi, “Alternating CQ-algorithms for convex feasibility and split fixed-point problems,” Journal of Nonlinear and Convex Analysis, vol. 15, no. 4, pp. 809–818, 2014. [8] A. Moudafi, “A relaxed alternating CQ-algorithm for convex feasibility problems,” Nonlinear Analysis, Theory, Methods and Applications, vol. 79, no. 1, pp. 117–121, 2013. [9] H. Che and M. Li, “A simultaneous iterative method for split equality problems of two finite families of strictly pseudononspreading mappings without prior knowledge of operator norms,” Fixed Point Theory and Applications, vol. 2015, article 1, 2015. [10] S.-S. Chang, L. Wang, and L.-J. Qin, “Split equality fixed point problem for quasi-pseudo-contractive mappings with applications,” Fixed Point Theory and Applications, vol. 2015, article 208, 2015. [11] W. Takahashi, “The split feasibility problem and the shrinking projection method in banach spaces,” Journal of Nonlinear and Convex Analysis, vol. 16, no. 7, pp. 1449–1459, 2015. [12] F. Schopfer, T. Schuster, and A. K. Louis, “An iterative regularization method for the solution of the split feasibility problem in Banach spaces,” Inverse Problems, vol. 24, no. 5, Article ID 055008, 2008. [13] F. Wang, “A new algorithm for solving the multiple-sets split feasibility problem in Banach spaces,” Numerical Functional Analysis and Optimization, vol. 35, no. 1, pp. 99–110, 2014. [14] Y. Shehu, O. S. Iyiola, and C. D. Enyi, “An iterative algorithm for solving split feasibility problems and fixed point problems in Banach spaces,” Numerical Algorithms, vol. 72, no. 4, pp. 835– 864, 2016. [15] Z. H. He and J. T. Sun, “The problem of split convex feasibility and its alternating approximation algorithms,” Acta Mathematica Sinica, English Series, vol. 31, no. 12, pp. 1857–1871, 2015. [16] S.-S. Chang, “Some problems and results in the study of nonlinear analysis,” Nonlinear Analysis, Theory, Methods and Applications, vol. 30, no. 7, pp. 4197–4208, 1997.
Advances in
Operations Research Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Advances in
Decision Sciences Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Journal of
Applied Mathematics
Algebra
Hindawi Publishing Corporation http://www.hindawi.com
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Journal of
Probability and Statistics Volume 2014
The Scientific World Journal Hindawi Publishing Corporation http://www.hindawi.com
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
International Journal of
Differential Equations Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Volume 2014
Submit your manuscripts at https://www.hindawi.com International Journal of
Advances in
Combinatorics Hindawi Publishing Corporation http://www.hindawi.com
Mathematical Physics Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Journal of
Complex Analysis Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
International Journal of Mathematics and Mathematical Sciences
Mathematical Problems in Engineering
Journal of
Mathematics Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Volume 2014
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
#HRBQDSDĮ,@SGDL@SHBR
Journal of
Volume 201
Hindawi Publishing Corporation http://www.hindawi.com
Discrete Dynamics in Nature and Society
Journal of
Function Spaces Hindawi Publishing Corporation http://www.hindawi.com
Abstract and Applied Analysis
Volume 2014
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
International Journal of
Journal of
Stochastic Analysis
Optimization
Hindawi Publishing Corporation http://www.hindawi.com
Hindawi Publishing Corporation http://www.hindawi.com
Volume 2014
Volume 2014