Professional Documents
Culture Documents
DOI: 10.3934/math.2023364
Received: 29 August 2022
Revised: 02 December 2022
Accepted: 28 December 2022
http://www.aimspress.com/journal/Math Published: 12 January 2023
Research article
Projection methods for quasi-nonexpansive multivalued mappings in
Hilbert spaces
Abstract: This paper proposes a modified D-iteration to approximate the solutions of three quasi-
nonexpansive multivalued mappings in a real Hilbert space. Due to the incorporation of an inertial
step in the iteration, the sequence generated by the modified method converges faster to the common
fixed point of the mappings. Furthermore, the generated sequence strongly converges to the required
solution using a shrinking technique. Numerical results obtained indicate that the proposed iteration is
computationally efficient and outperforms the standard forward-backward with inertial step.
Keywords: quasi-nonexpansive multivalued mappings; weak and strong convergence; inertial
technical term; D-iteration; projection methods
Mathematics Subject Classification: 47H09, 47H10
1. Introduction
Let K be a nonempty closed and convex subset of real Hilbert space H. Define S : K → K to
be a continuous mapping. A point ū ∈ K is said to be a fixed point of S if S (ū) = ū. Also, the F(S )
represents the set of all fixed points of S . Several authors have investigated the existence of fixed points
for theorems of single-valued nonexpansive mappings (for example, [1–5]).
Mann [6] proposed the following method in 1953 for approximating the fixed point of a
nonexpansive mapping S in a Hilbert space H :
un+1 = an un + (1 − an )S un , ∀ n ≥ 1, (1.1)
where {an } is a sequence in [0, 1].
7243
Ishikawa [7] generalized Mann’s iterative algorithm (1.1) in 1974 by introducing the iteration:
u0 ∈ K chosen arbitrary,
vn = (1 − an )un + an S un ,
(1.2)
un+1 = (1 − bn )un + bn S vn , n ≥ 0,
where {a jn } be a sequence in [, 1 − ] for some ∈ (0, 1), for each j ∈ {1, 2, . . . , r}.
Sainuan [10] developed a new iteration called P-iteration in 2015. The P-iteration is defined as:
u1 ∈ K chosen arbitrary,
vn = (1 − an )un + an S un ,
(1.5)
ρ = + ,
n (1 − b n )vn b n S v n
u = (1 − c )S v + c S ρ , n ≥ 1,
n+1 n n n n
where I is the identity mapping. It was proved that if {γn } is non-decreasing and {λn } ⊂ [0, 1) with
X∞
λn kun − un−1 k2 < ∞, (1.8)
n=1
then algorithm (1.7) converges weakly to a zero of B.
Nakajo and Takahashi [18] proposed modifying Mann’s iteration method (1.1) to obtain a strong
convergence theorem in Hilbert spaces H :
u0 ∈ K, chosen arbitrary,
vn = (1 − an )un + an S un ,
Kn = {x ∈ K : kvn − xk ≤ kun − xk},
(1.9)
Rn = {x ∈ K : hu0 − un , un − xi},
un+1 = PKn ∩Rn u0 , ∀ n ≥ 0,
where {an } ⊆ [0, a] for some a ∈ [0, 1). They proved that the sequence {un } converges strongly to
PF(S ) u0 .
In 2021, Chaolamjiak et al. [14] proposed modifying SP iteration method (1.4) to obtain a strong
convergence theorem in Hilbert spaces H :
u0 , u1 ∈ K, R1 = K,
tn = un + λn (un − un−1 ),
vn ∈ (1 − an )tn + an S 1 tn ,
ρn ∈ (1 − bn )vn + bn S 2 vn ,
(1.10)
wn ∈ (1 − cn )ρn + cn S 3 ρn ,
Kn = {x ∈ K : kwn − xk2 ≤ kun − xk2 + 2λ2n kun − un−1 k2 − 2λn hun − x, un−1 − un i},
Rn = {x ∈ Rn−1 : hu1 − un , un − xi ≥ 0},
un+1 = PKn ∩Rn u1 ,
for all n ≥ 1, where {an }, {bn } and {cn } ⊂ (0, 1). They proved that the sequence {un } converges strongly
to a common fixed point of S 1 , S 2 and S 3 .
The results [11,18,19,21] provide incentive. In order to locate a common fixed point of three quasi-
nonexpansive multivalued mappings, we introduce the D-iterative approach with the inertial technical
term. We can prove strong convergence theorems by combining shrinking projection methods with
inertial D-iteration. Finally, we compare our inertial projection method to the traditional projection
method and conduct numerical tests to support our major findings with different choices of the initial
values x0 and x1 in 4 case.
2. Preliminaries
Let CB(K) and K(K) denote the families of nonempty closed bounded, and compact, respectively.
The Hausdorff metric on CB(K) is defined by:
( )
H(C, Q) = max sup d(u, Q), sup d(v, C) , ∀ C, Q ∈ CB(K),
u∈C v∈Q
Lemma 2.5. [18] Let K be a nonempty closed convex subset of a real Hilbert space H and PK : H → K
be the metric projection from H onto K. Then
for all n ≥ 1, ∞ n=1 γn < ∞, and there exists a real number β with 0 ≤ βn ≤ β < 1 for all n ≥ 1. Then,
P
the followings hold
(a) n≥1 [αn − αn−1 ]+ < ∞, where [t]+ = max{t, 0};
P
(b) there exists α∗ ∈ [0, ∞) such that limn→∞ αn = α∗ .
Lemma 2.7. [20] Let H be a real Hilbert space. Then, for each u, v ∈ H and t ∈ [0, 1]
(a) ku − vk2 ≤ kuk2 + kvk2 − 2hu, vi;
(b) ktu − (1 − t)vk2 = tkuk2 + (1 − t)kvk2 − t(1 − t)ku − vk2 ;
(c) If {un } is a sequence in H such that un * u, then
lim sup kun − vk2 = lim sup kun − uk2 + ku − vk2 .
n→∞ n→∞
3. Main results
Theorem 3.1. Let K be a closed convex subset of a real Hilbert space H and S 1 , S 2 , S 3 : H → CB(K)
be quasi-nonexpansive multivalued mappings with Υ := F(S 1 ) ∩ F(S 2 ) ∩ F(S 3 ) , ∅ and I − S i is
demiclosed at 0 for all i ∈ {1, 2, 3}. Let {un } be a sequence generated by
u0 , u1 ∈ K chosen arbitrary,
tn = un + λn (un − un−1 ),
vn ∈ (1 − an )tn + an S 1 tn ,
(3.1)
ρn ∈ (1 − bn )S 1 tn + bn S 2 vn ,
un+1 ∈ (1 − cn )S 2 vn + cn S 3 ρn ,
for all n ≥ 1, where {an }, {bn } and {cn } ⊂ (0, 1). Assume that the following conditions hold
(a) ∞ n=1 λn kun − un−1 k < ∞;
P
(b) 0 < lim inf n→∞ an < lim supn→∞ an < 1;
(c) 0 < lim inf n→∞ bn < lim supn→∞ bn < 1;
(d) 0 < lim inf n→∞ cn < lim supn→∞ cn < 1.
If S 1 , S 2 and S 3 satisfy Condition (A), then the sequence {un } converges weakly to a common fixed
point of S 1 , S 2 and S 3 .
and
kun+1 − ūk = k(1 − cn )yn + cn zn − ūk
= k(1 − cn )(yn − ū) + cn (zn − ū)k
≤ (1 − cn )kyn − ūk + cn kzn − ūk
= (1 − cn )d(yn , S 2 ū) + cn d(zn , S 3 ū)
≤ (1 − cn )H(S 2 vn , S 2 ū) + cn H(S 3 ρn , S 3 ū)
(3.5)
≤ (1 − cn )kvn − ūk + cn kρn − ūk
≤ (1 − cn )ktn − ūk + cn ktn − ūk
= ktn − ūk − cn ktn − ūk + cn ktn − ūk
= ktn − ūk
≤ kun − ūk + λn kun − un−1 k.
Using Lemma 2.6, (3.5) and the assumption (a), we have limn→∞ kun − ūk exists. Thus, {un } is bounded
and also {ρn }, {vn } and {tn }. From Lemma 2.7(b), we get
and
kun+1 − ūk2 = k(1 − cn )yn + cn zn − ūk2
= k(1 − cn )(yn − ū) + cn (zn − ū)k2
= (1 − cn )kyn − ūk2 + cn kzn − ūk2 − cn (1 − cn )kyn − zn k2
(3.8)
= (1 − cn )d(yn , S 2 ū)2 + cn d(zn , S 3 ū)2 − cn (1 − cn )kyn − zn k2
≤ (1 − cn )H(S 2 vn , S 2 ū)2 + cn H(S 3 ρn , S 3 ū)2 − cn (1 − cn )kyn − zn k2
≤ (1 − cn )kvn − ūk2 + cn kρn − ūk2 − cn (1 − cn )kyn − zn k2 .
Using conditions (a)–(d), limn→∞ kun − ūk exists and (3.10), we obtain
Because {un } is bounded, there exists a subsequence {unk } of {un } such that unk * ū some ū ∈ K.
From (3.12), we have tnk * ū. Because I − S 1 is demiclosed at 0 and (3.11), we obtain ū ∈ S 1 ū.
From (3.13), we have vnk * ū. Because I − S 2 is demiclosed at 0 and (3.11), we obtain ū ∈ S 2 ū. It
follows from (3.14) that ρnk * ū. Again, because I −S 3 is demiclosed at 0 and (3.11), we have ū ∈ S 3 ū.
This implies that ū ∈ Υ. Now, we show that {un } converges weakly to ū. We take another subsequence
{umk } of {un } converging weakly to some u∗ ∈ Υ. Because limn→∞ kun − ūk exists and Lemma 2.3.Thus,
we have ū = u∗ .
Theorem 3.2. Let K be a nonempty closed convex subset of a real Hilbert space H and S 1 , S 2 , S 3 :
K → CB(K) be quasi-nonexpansive multivalued mappings with Υ := F(S 1 ) ∩ F(S 2 ) ∩ F(S 3 ) , ∅ and
I − S i is demiclosed at 0 for all i ∈ {1, 2, 3}. Let {un } be a sequence generated by
u0 , u1 ∈ K, K1 = K,
tn = un + λn (un − un−1 ),
vn ∈ (1 − an )tn + an S 1 tn ,
ρn ∈ (1 − bn )S 1 tn + bn S 2 vn ,
(3.15)
wn ∈ (1 − cn )S 2 vn + cn S 3 ρn ,
Kn+1 = {x ∈ Kn : kwn − xk2 ≤ kun − xk2 + 2λ2n kun − un−1 k2 − 2λn hun − x, un−1 − un i},
un+1 = PK u1 ,
n+1
for all n ≥ 1, where {an }, {bn } and {cn } ⊂ (0, 1). Assume that the following conditions hold
(a) ∞ n=1 λn kun − un−1 k < ∞;
P
(b) 0 < lim inf n→∞ an < lim supn→∞ an < 1;
(c) 0 < lim inf n→∞ bn < lim supn→∞ bn < 1;
(d) 0 < lim inf n→∞ cn < lim supn→∞ cn < 1.
If S 1 , S 2 and S 3 satisfy Condition (A), then the sequence {un } converges strongly to a common fixed
point of S 1 , S 2 and S 3 .
Proof. Step I. Show that {un } is well defined. Using S 1 , S 2 and S 3 satisfy Condition (A), Lemma 2.1,
Υ is closed and convex. Firstly, we show that Kn is closed and convex for all n ≥ 1. Since induction
on n that Kn is closed and convex. For n = 1, K1 = K is closed and convex. Suppose that Kn is closed
and convex for some n ≥ 1. Using the definition Kn+1 and Lemma 2.4, we have that Kn+1 is closed and
convex. Thus, Kn is closed and convex for all n ≥ 1. Next, we show that Υ ⊆ Kn for each n ≥ 1. From
Lemma 2.6(b) and S 1 , S 2 and S 3 satisfy Condition (A), let ū ∈ Υ for xn ∈ S 1 tn , yn ∈ S 2 vn , zn ∈ S 3 ρn
and using (3.15), we obtain
Therefore, from (3.18), ū ∈ Kn , n ≥ 1. This implies that Υ ⊆ Kn for each n ≥ 1, and hence, Kn , ∅.
Thus, {un } is well defined.
Step II. Show that un → u ∈ K as n → ∞. Since un ∈ PKn u1 , Kn+1 ⊆ Kn , and un+1 ∈ Kn , we obtain
Since Υ ⊆ Kn , we obtain
kun − u1 k ≤ kx − u1 k, ∀ n ≥ 1, (3.20)
for all x ∈ Υ. The inequalities (3.19) and (3.20) imply that the sequence {un − u1 } is bounded and
non-decreasing. Therefore, limn→∞ kun − u1 k exists.
For m > n, from the definition of Kn , we obtain um ∈ PKm u1 ∈ Km ⊆ Kn . Using Lemma 2.5, we have
From limn→∞ kun − u1 k exists and follows (3.21), we have that limn→∞ kun − um k = 0. Therefore, {un } is
a Cauchy sequence in K, and so un → u ∈ K as n → ∞.
Step III. Show that limn→∞ ktn − xn k = limn→∞ kxn −yn k = limn→∞ kyn −zn k = 0, where xn ∈ S 1 tn , yn ∈
S 2 vn and zn ∈ S 3 ρn . From Step II, we obtain limn→∞ kun+1 − un k = 0. Because un+1 ∈ Kn , we have that
From Step II, we know that un → u ∈ K. It follows (3.27), we obtain that tn → u. Because I − S 1 is
demiclosed at 0, we have u ∈ F(S 1 ). In the same way, we have that u ∈ F(S 2 ) and u ∈ F(S 3 ). This
implies that u ∈ Υ.
Step IV. Show that u = PΥ u1 . From u ∈ Υ and (3.19), we obtain
ku − u1 k ≤ kx − u1 k, ∀ x ∈ Υ.
Using the definition of the projection operator, we can conclude that u = PΥ u1 .
Theorem 3.3. Let K be a nonempty closed convex subset of a real Hilbert space H and S 1 , S 2 , S 3 :
K → CB(K) be quasi-nonexpansive multivalued mappings with Υ := F(S 1 ) ∩ F(S 2 ) ∩ F(S 3 ) , ∅ and
I − S i is demiclosed at 0 for all i ∈ {1, 2, 3}. Let {un } be a sequence generated by
u0 , u1 ∈ K, R1 = K,
tn = un + λn (un − un−1 ),
vn ∈ (1 − an )tn + an S 1 tn ,
ρn ∈ (1 − bn )S 1 tn + bn S 2 vn ,
(3.28)
+ ρ ,
w n ∈ (1 − c n )S 2 vn cn S 3 n
Kn = {x ∈ K : kwn − xk2 ≤ kun − xk2 + 2λ2n kun − un−1 k2 − 2λn hun − x, un−1 − un i},
Rn = {x ∈ Rn−1 : hu1 − un , un − xi ≥ 0},
u = P
Kn ∩Rn u1 ,
n+1
for all n ≥ 1, where {an }, {bn } and {cn } ⊂ (0, 1). Assume that the following conditions hold
(a) ∞ n=1 λn kun − un−1 k < ∞;
P
(b) 0 < lim inf n→∞ an < lim supn→∞ an < 1;
(c) 0 < lim inf n→∞ bn < lim supn→∞ bn < 1;
(d) 0 < lim inf n→∞ cn < lim supn→∞ cn < 1.
If S 1 , S 2 and S 3 satisfy Condition (A), then the sequence {un } converges strongly to a common fixed
point of S 1 , S 2 and S 3 .
Proof. From the same method of Theorem 3.2 step by step, we can conclude the proof by replacing
Kn+1 by Kn , expect in Step 1. Showing that Υ ⊆ Kn for each n ≥ 1. Next, we show that Υ ⊆ Rn for all
n ≥ 1. Indeed, by mathematical induction, for n = 1, we obtain Υ ⊆ K = R1 . Suppose that Υ ⊆ Rn for
all n ≥ 1. Because un+1 is the projection of u1 onto Kn ∩ Rn , we obtain
hu1 − un+1 , un+1 − xi ≥ 0, ∀ x ∈ Kn ∩ Rn .
kun − u1 k ≤ kq − u1 k, ∀ q ∈ Υ. (3.29)
From (3.19) and (3.29) we have the sequence {un − u1 } is bounded and non-decreasing, and so
limn→∞ kun − u1 k exists. For m > n, by definition of Rn , we have um = PRm u1 ∈ Rm ⊆ Rn . Using
Lemma 2.5, we have (3.21). Because limn→∞ kun − u1 k exists, it follows (3.21), we obtain limn→∞ kum −
un k = 0. Therefore, {un } is a Cauchy sequence in K, and hence un → q ∈ K as n → ∞. In fact, we have
limn→∞ kun+1 −un k = 0. Using the same proof of Steps 3 and 4 in Theorem 3.2, we obtain q = PΥ u1 .
4. Numerical results
It is commonly known that computing the projection of a point on an intersection is quite difficult.
However, this can also be stated as the following optimization problem for computing purposes
where K∗ = Kn ∩Rn . See [22] for a list of several more approaches to handle projection onto intersection
of sets computationally.
The set Kn+1 can be found by Kn ∩ Rn , where
Rn = {x ∈ H : kwn − xk2 ≤ kun − xk2 + 2λ2n kun − un−1 k2 − 2λn hun − x, un−1 − un i}. (4.2)
The projection can be thought of as the following optimization problem by point (4.2):
where Kn+1 = Kn ∩ Rn .
R3 → CB(R3 ) as:
if u ∈ K1 ,
{(5, 5, 5)}
S 1u =
1
{v = (v1 , v2 , v3 ) ∈ K : (v1 − 5)2 + (v2 − 5)2 + (v3 − 5)2 ≤
p
} otherwise,
kuk1
{(5, 5, 5)}
if u ∈ K1 ,
S 2u =
arctan(19u 2 − 65)
{v = (5, v2 , 5) ∈ K : v2 ∈ [(u2 + 5)(
) + u2 , 5]} otherwise,
2
and
{(5, 5, 5)}
if u ∈ K1 ,
S 3u =
sin(19u2 − 10)
{v = (5, 5, v3 ) ∈ K : v3 ∈ [(u2 − 5)( ) + u2 , 5]} otherwise.
5
We see that S 1 , S 2 and S 3 are quasi-nonexpansive and F(S 1 ) ∩ F(S 2 ) ∩ F(S 3 ) = {(5, 5, 5)}. Let an =
n+4
5n+5
, bn = 7n+2
n+2
, cn = 7n+7
9n+9
and
( )
1
, 0.035 if un , un−1 ,
min
λn = (n + 1)2 kun − un−1 k
0.035
otherwise.
We compare a numerical test between our inertial method defined in Theorem 3.3 and method (1.10).
The stopping criterion is defined by kun+1 − un k < 10−10 . We make different choices of the initial values
x0 and x1 as follows (see Figure 1 and Table 1)
Case 1 : x0 = (2.6816, 2.4389, 2.8891) and x1 = (2.7733, 2.2146, 2.2555).
Case 2 : x0 = (3.9587, 4.6121, 2.9779) and x1 = (3.7123, 3.4894, 4.8867).
Case 3 : x0 = (4.9401, 3.9274, 2.8475) and x1 = (3.8675, 3.7185, 4.7133).
Case 4 : x0 = (3.9933, 4.9899, 4.2187) and x1 = (3.7722, 3.6190, 2.9152).
Case 1
Case 2
Case 3
Case 4
Figure 1. Valued of kun+1 − un k in Cases 1–4.
5. Conclusions
In this paper, we proved convergence theorems in Hilbert spaces using a modified D-iteration.We
proved the weak and strong convergence of the iterative algorithms to the common fixed point under
some suitable assumptions.
Acknowledgments
Firstly, Anantachai Padcharoen (anantachai.p@rbru.ac.th) would like to thank the Research and
Development Institute of Rambhai Barni Rajabhat University. Finally, Kritsana Sokhuma would like
to thank Phranakhon Rajabhat University.
Conflict of interest
References
1. F. E. Browder, Nonexpansive nonlinear operators in a Banach space, Proc. Nat. Acad. Sci. USA.,
54 (1965), 1041–1044. https://doi.org/10.1073/pnas.54.4.1041
2. D. Göhde, Zum prinzip der kontraktiven abbildung, Math. Nachr., 30 (1965), 251–258.
https://doi.org/10.1002/mana.19650300312
3. S. Reich, Approximate selections, best approximations, fixed points, and invariant sets, J. Math.
Anal. Appl., 62 (1978), 104–113. https://doi.org/10.1016/0022-247X(78)90222-6
4. K. Goebel, W. A. Kirk, Topics in metric fixed point theory, Cambridge Studies in Advanced
Mathematics, 28, Cambridge: Cambridge University Press, 1990.
5. W. A. Kirk, A fixed point theorem for mappings which do not increase distances, Am. Math. Mon.,
72 (1965), 1004–1006. https://doi.org/10.2307/2313345
6. W. R. Mann, Mean value methods in iteration, Proc. Am. Math. Soc., 4 (1953), 506–510.
https://doi.org/10.2307/2032162
7. S. Ishikawa, Fixed points by a new iteration method, Proc. Am. Math. Soc., 44 (1974), 147–150.
https://doi.org/10.2307/2039245
8. M. A. Noor, New approximation schemes for general variational inequalities, J. Math. Anal. Appl.,
251 (2000), 217–229. https://doi.org/10.1006/jmaa.2000.7042
9. I. Yildirim, M. Özdemir, A new iterative process for common fixed points of finite families of
non-self-asymptotically non-expansive mappings, Nonlinear Anal.-Theor., 71 (2009), 991–999.
https://doi.org/10.1016/j.na.2008.11.017
10. P. Sainuan, Rate of convergence of P-iteration and S-iteration for continuous functions on closed
intervals, Thai J. Math., 13 (2015), 449–457.
11. J. Daengsaen, A. Khemphet, On the rate of convergence of P-iteration, SP-iteration, and D-iteration
methods for continuous nondecreasing functions on closed intervals, Abstr. Appl. Anal., 2018
(2018), 7345401. https://doi.org/10.1155/2018/7345401