Synchronal algorithm and cyclic algorithm for fixed point problems and variational inequality problems in hilbert spaces

Abstract

We design synchronal algorithm and cyclic algorithm based on the general iterative algorithm proposed by Tian in 2010 for finding the common fixed point x* of finite family of strict pseudo-contractive mappings which is the solution of the variational inequality 〈(γ f - μF)x*, x - x*〉 ≤ 0, .

2000 Mathematics Subject Classification: 58E35; 47H09; 65J15.

1. Introduction

Let H be a real Hilbert space with the inner product 〈·,·〉 and the norm || · ||, respectively. Let C be nonempty closed subset of H.

Recall that a mapping T : C → H is said to be k-strict pseudo-contraction if there exists a constant k ∈ [0, 1) such that

(1.1)

These mappings are extensions of nonexpansive mappings which satisfy the inequality (1.1) with k = 0. That is, T : C → H is nonexpansive if

We denote by F(T) the set of fixed points of the mapping T, that is

We assume that F(T) ≠ ∅ it is well known that F(T) is closed convex.

Let F : C → H be a nonlinear operator, we consider the problem of finding a point x* ∈ C such that

We denote by V I(F, C) the set of solutions of this variational inequality problem.

Takahashi [1] introduced a classical CQ algorithm as follows:

where T is nonexpansive mapping, and {α_n} ⊂ [0, a] for some a ∈ [0, 1). Then they showed that {x _n } converged strongly to P_F(T)(x₀) by the hybrid method in the mathematical programming. But it is hard to compute by this algorithm, because projection has to be used in every process.

The hybrid steepest descent method of Yamada [2] conquered this deficiency and proposed the following algorithm for solving the variational inequality.

Take x₀ ∈ H arbitrarily and define {x _n } by

(1.2)

where T is a nonexpansive mapping on H, F is L-Lipschitzian and η-strongly monotone with k > 0, η > 0, 0 < μ < 2η/L². If {λ _n } is a sequence in (0, 1) satisfying the following conditions:

• (i) lim_n→∞ λ _n = 0;

• (ii)

• (iii) either or ,

then the sequence {x _n } converged strongly to the unique solution of the variational inequality

Besides, he also proposed cyclic algorithm:

where T_[n]= T _{n mod N}, he also got strong convergence theorems.

On the other hand, Marino and Xu [3] considered the following general iterative method: an initial x₀ is selected in H arbitrarily

(1.3)

where T is a nonexpansive mapping on H, f is a contraction, A is a linear bounded strongly positive operator, and {α _n } is a sequence in (0, 1) satisfying the following conditions:

(C1) lim_n→∞α _n = 0;

(C2)

(C3) either or .

They proved that the sequence {x _n } converged strongly to a fixed point of T which solves the variational inequality

Very recently, Tian [4] combined the iterative method (1.3) with the Yamada's method (1.2) and considered the following general iterative method

(1.4)

where T is a nonexpansive mapping on H, f is a contraction, and F is k- Lipschitzian and η-strongly monotone with k > 0, η > 0, 0 < μ < 2η/k².

He proved that if the sequence {α _n } of parameters satisfies (C1)-(C3), then the sequence {x _n } generated by (1.4) converged strongly to a fixed point of T which solves the variational inequality

(1.5)

In this paper we designed two algorithms for finding a common fixed point x* of finite strict pseudo-contractions which also solves the variational inequality

(1.6)

where N ≥ 1 is a positive integer and are N strict pseudo-contractions.

Let T be defined by

Where λ _i > 0 such that . We will show that the sequence {x _n } generated by the algorithm:

(1.7)

will converge strongly to a solution to the problem (1.6).

Another approach to the problem (1.6) is the cyclic algorithm. For each i = 1,..., N, let

where the constant β _i satisfies k _i < β _i < 1. Beginning with x₀ ∈ H, we define the sequence {x _n } cyclically by

Indeed, the algorithm above can be written as

(1.8)

where T_[n]= T _i , with i = n(modN ), 1 ≤ i ≤ N. We will show that this cyclic algorithm (1.8) is also strongly convergent if the sequences {α _n } and {β _n } are appropriately chosen.

We will use the notations:

• 1. ⇀ for weak convergence and → for strong convergence.

• 2. denotes the weak !-limit set of {x _n }.

2. Preliminaries

We need some facts and tools which are listed as below.

Definition 1 A mapping F : C → H is called η-strongly monotone if there exists a positive constant η > 0 such that

Definition 2 B is called to be strongly positive bounded linear operator on

H, if there is a constant with property

Lemma 2.1. (see[5]) Let C be a nonempty closed convex subset of a real Hilbert space H and T : C ! C is a nonexpansive mapping. If a sequence {x _n } in C such that x _n ⇀ z and (I - T)x _n → 0, then z = Tz.

Lemma 2.2. (see[6]) Let C be a nonempty closed convex subset of a real Hilbert space H. If T : C → C is a κ-strict pseudo-contraction, then the mapping I - T is demiclosed at 0. That is, if {x _n } is a sequence in C such thatand (I - T)x _n → 0, then.

Lemma 2.3. (see[7]) Assume {a _n } is a sequence of nonnegative real numbers such that

where {γ _n } is a sequence in (0, 1) and {δ _n } is a sequence in ℝ such that:

• (i) lim_n→∞ γ _n = 0 and ;

• (ii) lim_n→∞ δ _n /γ _n ≤ 0 or .

Then lim_n→∞a _n = 0.

Lemma 2.4. (see[4]) Let H be a real Hilbert space, f : H → H a contraction with coefficient 0 < α < 1, and F : H → H a k-Lipschitzian continuous operator and η-strongly monotone operator with k > 0, η > 0. Then for 0 < γ < μη/α,

That is, μF - γf is strongly monotone with coefficient μη - γα.

Lemma 2.5. (see[8]) Suppose S : C → H is a k-strict pseudo-contraction. Define T : C → H by Tx = λx + (1 - λ)Sx for each × ∈ C. Then, as λ ∈ [k, 1), T is a nonexpansive mapping such that F(T) = F(S).

Lemma 2.6. (see[6]) Assume C is a closed convex subset of a Hilbert space H. Given an integer N ≥ 1, assume for each 1 ≤ i ≤ N, T _i : C → C is a k _i -strict pseudo-contraction for some 0 ≤ k _i < 1. Assumeis a positive sequence such that. Suppose thatthen

Lemma 2.7. (see[9]) Assume T _i : H → H is a k _i -strict pseudo-contraction for some 0 ≤ k _i < 1 (1 ≤ i ≤ N ): Let, k _i < α _i < 1 (1 ≤ i ≤ N), if, then

Lemma 2.8. Let F : H → H be a η-strongly monotone and L-Lipschitzian operator with L > 0, η > 0. Assume that 0 < μ < 2η/L², and 0 < t < 1. Then ||(I - μtF)x - (I - μtF)y|| ≤ (1 - tτ) ||x - y||.

Proof. Put g = I μtF, then

Therefore,

that is,

□

3. Synchronal algorithm

Theorem 3.1. Let H be a real Hilbert space and let T _i : H → H be a k _i -strict pseudo-contraction for some k _i ∈ (0, 1) (i = 1,..., N ) such that, f be a contraction with coefficient β ∈ (0, 1) and λ _i be a positive constant such that. Let G : H → H be a η-strongly monotone and L-Lipschitzian operator with L > 0, η > 0. Assume that 0 < μ < 2η/L², . Given the initial guess x₀ ∈ H chosen arbitrarily and given sequences {α _n } and {β _n } in (0, 1), satisfying the following conditions:

(3.1a) lim_n→∞α _n = 0, ;

(3.1b) , ;

(3.1c) 0 ≤ max _i k _i ≤ β _n < a < 1 for all n ≥ 0;

let {x _n } be the sequences define d by the composite process (1.7), i.e.

Then {x _n } converges strongly to a common fixed point ofwhich solves the variational inequality (1.6).

Proof. Put , then by Lemma 2.6, we conclude that T is a k-strict pseudo-contraction with k = max {k _i : 1 ≤ i ≤ N} and .

We can rewrite the algorithm (1.7) as

Furthermore, by Lemma 2.5, we conclude that is a nonexpansive mapping and .

Step 1. {x _n } is bounded.

Take , from (1.7) and Lemma 2.9 we have

By simple induction, we have

Hence {x _n } is bounded.

From , we have v ∈ F (T ), hence

It follows that

So, we have

Therefore, {Tx _n } is bounded.

G is L-Lipschitzian, so

{Tx _n } is bounded, so is bounded.

f is a contraction, so f(x _n ) is bounded.

Step 2.

(3.1)

Observing that

we have

This in turn implies that

(3.2)

where M₁ is an appropriate constant such that . On the other hand, we note that

(3.3)

where M₂ is an appropriate constant such that M₂ ≥ sup_n≥1{||x _n - Tx _n ||}. Substituting (3.3) into (3.2) yields

where M₃ is an appropriate constant such that M₃ ≥ max{M₁, M₂}. By conditions (3.1a) and (3.1b) and Lemma 2.3, we obtain that lim_n→∞||x_n+1- x _n || = 0.

From (1.7), we observe that

It follows from the condition (3.1a) and the boundedness of {f(x _n )} and that

(3.4)

On the other hand,

Hence, by condition (3.1c), we have

From (3.1) and (3.4), we obtain

(3.5)

From the boundedness of {x _n }, we deduced that {x _n } converges weakly. Assume x _n ⇀ p, by Lemma 2.2 and (3.5), we obtain p = Tp. So, we have

(3.6)

Notice by Lemma 2.4, μG - γ f is strongly monotone, so the variational inequality (1.6) has a unique solution x* ∈ F(T).

Step 3.

(3.7)

Indeed, there exists a subsequence such that

Without loss of generality, we may further assume that . It follows from (3.6) that x ∈ F(T). Since x* is the unique solution of (1.6), we obtain

Step 4.

(3.8)

From Lemma 2.9, we have

This implies that

where and . , from (3.1a), we have lim_n→∞γ _n = 0; γ _n ≥ 2α _n (τ - γβ), from (3.1a), we have ; put M = sup {||x _n - x*|| : n ∈ N}, we have . So, lim_n→∞δ _n /γ _n ≤ 0. Hence, by Lemma 2.3, we conclude that x _n → x* as n → ∞. □

4. Cyclic algorithm

Theorem 4.1. Let H be a real Hilbert space and let T _i : H → H be a k _i -strict pseudo-contraction for some k _i ∈ (0, 1) (i = 1,..., N ) such thatand f be a contraction with coefficient β ∈ (0, 1). Let G : H → H be a η-strongly monotone and L-Lipschitzian operator with L > 0, η > 0. Assume that. Given the initial guess x₀ ∈ H chosen arbitrarily and given sequences {α _n } and {β _n } in (0, 1), satisfying the following conditions:

(4.1a) lim_n→∞α _n = 0,

(4.1b) ;

(4.1c) , or;

(4.1d) β_[n]∈ [k, 1), where k = max _i {k _i : 1 ≤ i ≤ N},

let {x _n } be the sequences define d by the composite process (1.8), i.e.

where T_[n]= T _i , with i = n(modN ), 1 ≤ i ≤ N, namely, T_[n]is one of T₁, T₂,..., T _N circularly. Then {x _n } converges strongly to a common fixed point ofwhich solves the variational inequality (1.6).

Proof. Step 1. {x _n } is bounded. Take , from (1.8) and Lemma 2.9 we have

By simple induction, we have

Hence {x _n } is bounded.

From the proof of Step 1 in Section 3, we know that {T_[n]x _n }, {f (x _n )}, {GA_[n]x _n } are bounded.

So, {A_[n]x _n } is bounded.

Step 2. lim_n→∞||x_n+N- x _n || = 0.

By (1.8) and Lemma 2.9, we have

(4.1)

where K₁ is an appropriate constant such that K₁ ≥ sup_n≥1{μ||GA_[n+1]x _n ||+ γ ||f(x _n )||}. By conditions (4.1a), (4.1b), (4.1c) and Lemma 2.3, we obtain ||x_n+N- x _n || → 0 as n → ∞.

Step 3. lim_{n→ ∞}||x _n - A_[n+N]··· A_[n+1]x _n || = 0.

From (1.8), we observe that

It follows from the condition (4.1a) and the boundedness of {f(x _n )} and {GA_[n+1]x _n } that

Recursively,

By condition (4.1d) and Lemma 2.5, we know that is nonexpansive, so we get

Proceeded accordingly, we have

Note that

From all the expressions above, we obtain

Since

we conclude ||x _n - A_[n+N]··· A_[n+1]x _n || → 0(n → ∞).

Step 4.

(4.2)

Take a subsequence , by step 3, we get

Notice that, for each n _j , is some permutation of the mappings A₁A₂ ··· A _N , since A₁, A₂,···, A _N are finite, all the finite permutation are N!, there must be some permutation appears infinite times.

Without loss of generality, suppose this permutation is A₁A₂···A _N , we can take a subsequence such that and

By Lemma 2.5, we conclude that A₁, A₂,···, A _N are all nonexpansive. It is easy to prove that is nonexpansive, so A₁A₂···A _N is.

By Lemma 2.2, we have q = A₁A₂ ··· A _N q. From Lemmas 2.5 and 2.7, we obtain

Step 5.

(4.3)

Indeed, there exists a subsequence such that

Without loss of generality, we may further assume that . It follows from (4.2) that x ∈ F(T). Since x* is the unique solution of (1.6), we obtain

Step 6. x _n → x*(n → ∞).

From Lemma 2.9, we have

This implies that

where and . , from (4.1a), we have lim_n→∞γ _n = 0; γ _n ≥ 2α _n (τ -γβ), from (4.1b), we have ; put M = sup {||x _n - x*||: n ∈ N}, we have . So, limsup_n→∞δ _n /γ _n ≤ 0. Hence, by Lemma 2.3, we conclude that x _n → x* as n → ∞. . □

Taking n = 1, β _n = 0 and T is nonexpansive mapping in Theorems 3.1 and 4.1, we get

Corollary 1 (see[4]) Let {x _n } be generated by the following algorithm

Assume the sequence {α _n } satisfies conditions:

(C1) lim_n→∞α _n = 0;

(C2) ;

(C3) either or

then {x _n } converged strongly to which solves the variational inequality

Taking n = 1, β _n = 0 and T is nonexpansive mapping, G = A, μ = 1 in Theorems 3.1 and 4.1, we get

Corollary 2 (see[3]) Let {x _n } be generated by the following algorithm:

Assume the sequence {α _n } satisfies conditions (C1)-(C3), then the sequence {x _n } converged strongly to a fixed point of T which solves the variational inequality

Taking n = 1, β _n = 0 and T is nonexpansive mapping, γ = 0 in Theorem 3.1 and Theorem 4.1, we get:

Corollary 3 (see[2]) Let {x _n } be generated by the following algorithm

where T is a nonexpansive mapping on H, F is L-Lipschitzian and η-strongly monotone with k > 0, η > 0, 0 < μ < 2η/L². If {λ _n } is a sequence in (0, 1) satisfies the following conditions:

• (i) lim_n→∞ λ _n = 0;

• (ii) ;

• (iii) either or

then the sequence {x _n } converged strongly to the unique solution of the variational inequality

Taking n = 1, β _n = 0 and T is nonexpansive mapping, γ = 0 in Theorem 4.1, we get

Corollary 4 (see[2]) Let {x _n } be generated by the following algorithm

where T_[n]= T _{n mod N}. Assume {λ _n } satisfies conditions (C1)-(C3) and C = F(T _N ··· T 1) = F (T₁T _N ··· T₃T₂) = ··· = F (T_{N - 1}T_{N - 2}··· T₁T _N ), then {x _n } converged strongly to the unique solution of the variational inequality