Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Research Article
A Cyclic Method for Solutions of a Class of Split Variational
Inequality Problem in Banach Space
1
Bayero University, Department of Mathematical Sciences, Kano, Nigeria
2
African University of Science and Technology, Department of Mathematical Sciences, Abuja, Nigeria
3
Department of Mathematics, Sa’adatu Rimi College of Education, Kumbotso, Kano, Nigeria
Received 15 January 2020; Revised 26 February 2020; Accepted 10 March 2020; Published 10 April 2020
Copyright © 2020 Bashir Ali 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.
In this paper, a cyclic algorithm for approximating a class of split variational inequality problem is introduced and studied in some
Banach spaces. A strong convergence theorem is proved. Some applications of the theorem are presented. The results presented
here improve, unify, and generalize certain recent results in the literature.
1. Introduction
〈x − y, Tx − Ty〉 ≥ α‖Tx − Ty‖2 , for any x, y ∈ C.
Let C be a nonempty closed and convex subset of a real
Banach space E, with dual E∗ . Then, a mapping T : C ⟶ E (3)
is said to be
(3) Strongly monotone if
(1) Nonexpansive if ‖Tx − Ty‖ ≤‖x − y‖ for any x, y ∈ C.
It is easy to see from the definition of ϕ that, in a real NC (u) � w∗ ∈ E∗ : 〈u − z, w∗ 〉 ≥ 0, ∀z ∈ C. (18)
Hilbert space H, equation (14) reduces to
ϕ(x, y) � ‖x − y‖2 , ∀x, y ∈ H. Then, B is maximal monotone and B− 1 0 � VI(C, f).
Furthermore, given x, y, z ∈ E and τ ∈ (0, 1), we have
the following properties (see, for example, [18]):
Lemma 5 (see [23]). Let E be a uniformly convex and smooth
P1: (‖x‖ − ‖y‖)2 ≤ ϕ(x, y) ≤ (‖x‖ + ‖y‖)2 , real Banach space, and let xn and yn be two sequences of
P2: ϕ(x, y) � ϕ(x, z) + ϕ(z, y) + 2〈z − x, Jy − Jz〉, E. If either xn or yn is bounded and
limn⟶∞ ϕ(xn , yn ) � 0, then limn⟶∞ ‖xn − yn ‖ � 0.
P3: ϕ(τx + (1 − τ)y, z) ≤ τϕ(x, z) + (1 − τ)ϕ(y, z).
3. Main Results
Definition 1. Let E be a smooth, strictly convex, and re-
flexive real Banach space and let C be a nonempty, closed, Theorem 2. Let E1 be a uniformly smooth and 2-uniformly
and convex subset of E. The map ΠC : E ⟶ C defined by convex real Banach space and E2 a 2-uniformly smooth real
� ΠC (x) ∈ C such that ϕ(
x x, x) � inf y∈C ϕ(y, x) is called Banach space with smoothness constant d2 ∈ (0, 1). Let C be a
the generalized projection of E onto C. Clearly, in a real nonempty, closed, and convex subset of E1 . Let f : C ⟶ E∗1
Hilbert space H, the generalized projection ΠC coincides be a monotone and k-Lipschitz continuous map, and
with the metric projection PC from H onto C. A : E1 ⟶ E2 be a bounded linear operator with its adjoint
A∗ such that A ≠ 0. Let Ti : E2 ⟶ E2 , i � 1, 2, . . . , m be
Definition 2. Let E1 and E2 be two reflexive, strictly convex, nonexpansive mappings. Setting Γ � z ∈ VI(C, f) : Az ∈
and smooth Banach spaces. The collection of mappings ∩m
i�1 F(Ti )} and assuming Γ ≠∅. Let a sequence xn be
A : E1 ⟶ E2 is linear, and continuous is a normed linear generated by
space with norm defined by ‖A‖ � sup‖x‖≤1 ‖Ax‖. The dual
⎪
⎧ x1 � x ∈ C � C1 ,
operator A∗ :E∗2 ⟶E∗1 defined by 〈A∗ y∗ ,x〉�〈y∗ ,Ax〉∀x∈ ⎪
⎪
⎪
⎪ − 1 ∗
E1 , y∗ ∈E∗2 is called the adjoint operator of A. The adjoint ⎪ yn � J Jxn − cA J2 I − T[n] Axn , where [n] � n mod m,
⎪
⎪
⎪
⎪
operator A∗ has the property ‖A∗ ‖�‖A‖. ⎪
⎨ tn � ΠC J− 1 Jyn − λfyn ,
⎪ z � Π J− 1 Jy − λft ,
⎪
⎪
⎪ n C n n
Lemma 1 (see [19]). Let C be a nonempty closed and convex ⎪
⎪
⎪
⎪ ∈ ϕ
subset of a smooth, strictly convex, and reflexive real Banach ⎪
⎪
C n+1 � v C n : v, zn ≤ ϕ v, yn ≤ ϕ v, xn ,,
⎪
⎩
space E. Then, xn+1 � ΠCn+1 x1 ,
�� ��2
ϕ u, yn ≤ ‖u‖2 + ��Jxn �� + 2c〈xn , A∗ J2 T[n] − IAxn 〉
�� ��2
+ d∗2 c2 ‖A‖2 ���T[n] − IAxn ��� − 2〈u, Jxn 〉 − 2c〈Au, J2 T[n] − IAxn 〉
�� ��2
� ϕ u, xn + d∗2 c2 ‖A‖2 ���T[n] − IAxn ��� + 2c〈Axn − Au, J2 T[n] − IAxn 〉
�� ��2
� ϕ u, xn + d∗2 c2 ‖A‖2 ���T[n] − IAxn ��� (22)
Using the fact that E2 is 2-uniformly smooth and T[n] By the fact that u ∈ VI(C, f) and using property P2, we
being nonexpansive, we have have
2〈T[n] Axn − Au, J2 T[n] − IAxn 〉 ϕ u, zn ≤ ϕ u, yn − ϕ zn , yn + 2λ〈tn − zn , ftn 〉
�� ��2 �� ��2 � ϕ u, yn − ϕ zn , tn − ϕ tn , yn (26)
≤ d2 ��T[n] Axn − Au�� + ��T[n] Axn − Axn ��
�� ��2 (23) + 2〈zn − tn , Jyn − λftn − Jtn 〉.
− ���T[n] Axn − Au − T[n] Axn − Axn ���
�� ��2 �� ��2 Also from the fact that tn � ΠC J− 1 (Jyn − λfyn ), zn ∈ C,
≤ d2 − 1��Axn − Au�� + ���T[n] − IAxn ��� . the Lipschitz continuity of f, Lemma 1, and Lemma 3, we
obtain that
From (22) and (23), we get
�� ��2 〈zn − tn , Jyn − λftn − Jtn 〉 � 〈zn − tn , Jyn − λfyn − Jtn 〉
ϕ u, yn ≤ ϕ u, xn + c2 d∗2 ‖A‖2 ���T[n] − IAxn ���
+ λ〈zn − tn , fyn − ftn 〉
�� ��2 �� ��2
− 2c���T[n] − IAxn ��� + c d2 − 1��Axn − Au��
≤ λ〈zn − tn , fyn − ftn 〉
�� ��2
+ c���T[n] − IAxn ��� �� ���� ��
≤ kλ��zn − tn ����yn − tn ��
�� ��2
≤ ϕ u, xn − c1 − d∗2 c‖A‖2 ���T[n] − IAxn ��� (24)
kλ ��� ��2 �� ��2
�� ��2 ≤ �zn − tn �� + ��yn − tn ��
2
− c 1 − d2 ��Axn − Au��
�� ��2 kλ
≤ ϕ u, xn − c1 − d∗2 c‖A‖2 ���T[n] − IAxn ��� ≤
2α
ϕ zn , tn + ϕ tn , yn .
≤ ϕ u, xn . (27)
�� ��
Step 4. We show that lim ��zn − tn �� � 0. (40)
�� �� �� �� �� �� n⟶∞
lim ��xn − yn �� � lim ���T[n] − IAxn ��� � lim ��zn − tn ��
n⟶∞ n⟶∞ n⟶∞ Thus, (39) and (40) imply
�� �� �� ��
� lim ��zn − yn �� � 0. lim ��zn − yn �� � 0.
n⟶∞
n⟶∞
(41)
(30)
Since xn+1 ∈ Cn+1 ⊂ Cn ,
Step 5. We show that xn converges to an element of Γ
ϕ xn , yn � ϕ xn , xn+1 + ϕ xn+1 , yn + 2〈xn+1 − xn , Jyn − Jxn+1 〉
Since xn is a Cauchy sequence, we may assume that
≤ ϕ xn , xn+1 + ϕ xn+1 , xn + 2〈xn+1 − xn , Jyn − Jxn+1 〉 xn ⟶ x∗ .
From the fact that limn⟶∞ ‖xn − yn ‖ � limn⟶∞ ‖xn −
� 2〈xn − xn+1 , Jxn − Jxn+1 〉 − 2〈xn − xn+1 , Jyn − Jxn+1 〉.
zn ‖ � limn⟶∞ ‖tn − yn ‖ � 0, we obtain that yn ⟶ x∗ ,
(31) zn ⟶ x∗ , and tn ⟶ x∗ . Since A is a bounded linear
Taking limit as n ⟶ ∞, we have limn⟶∞ ϕ(xn , yn ) � 0. operator, we have that Axn ⟶ Ax∗ .
Similarly, From (35), we have
�� ��
ϕ xn , zn ≤ 2〈xn − xn+1 , Jxn − Jxn+1 〉 − 2〈xn − xn+1 , Jzn − Jxn+1 〉. lim ���T[n] − IAxn ��� � 0. (42)
n⟶∞
(32)
Thus, for i ∈ {1, 2, 3, . . . , m}, limn⟶∞ ‖(Ti − I)Axn ‖ � 0.
Taking limit as n ⟶ ∞, we have limn⟶∞ ϕ(xn , zn ) � 0. Since Ti is nonexpansive for each i ∈ {1, 2, 3, . . . , m}, we
Since xn is bounded, it follows from Lemma 5 that have that I − Ti is demiclosed at 0 for i ∈ {1, 2, 3, . . . , m}.
limn⟶∞ ‖xn − yn ‖ � 0 and limn⟶∞ ‖xn − zn ‖ � 0. And therefore, Ax∗ ∈ F(Ti )for each i ∈ {1, 2, 3, . . . , m}.
Now, Thus, Ax∗ ∈ ∩m i�1 F(Ti ). Next, we show that
�� ��2 �� ��2 x∗ ∈ VI(C, f)
ϕ u, xn − ϕ u, yn � ��xn �� − ��yn �� − 2〈u, Jxn − Jyn 〉 Define
�� �� �� �� �� �� �� ��
� ��xn �� − ��yn ����xn �� + ��yn �� + 2〈u, Jxn − Jyn 〉 fv + NC (v), if v ∈ C,
�� �� �� �� �� �� Bv � (43)
≤ ��xn − yn ����xn �� + ��yn �� + 2〈u, Jxn − Jyn 〉. ∅, if v ∉ C.
(33) By Lemma 4, B is a maximal monotone and 0 ∈ Bv if and
Taking limit as n ⟶ ∞, we have limn⟶∞ (ϕ(u, xn )− only if v ∈ VI(C, f).
ϕ(u, yn )) � 0. Let (v, w) ∈ G(B). (v, w) ∈ G(B) ⟹ w ∈ Bv � fv +
In a similar way, we also have limn⟶∞ (ϕ(u, xn )− NC (v) ⟹ w − fv ∈ NC (v).
ϕ(u, zn )) � 0. Thus, 〈v − p, w − fv〉 ≥ 0, ∀ p ∈ C. Since
From (24), we obtain zn � ΠC J− 1 (Jyn − λftn ) and v ∈ C, we have by Lemma 4
�� ��2 that 〈zn − v, Jyn − λftn − Jzn 〉 ≥ 0.
0 <c − d∗2 c2 ‖A‖2 ���T[n] − IAxn ��� ≤ ϕ u, xn − ϕ u, yn . Thus,
(34)
Jzn − Jyn
v − zn , + ftn ≥ 0, n≥0. (44)
Thus, λ
�� ��2
lim ���T[n] − IAxn ��� � 0. (35) Using the fact that zn ∈ C and w − fv ∈ NC (v), we have
n⟶∞
6 Abstract and Applied Analysis
〈v − zn , w〉 ≥ 〈v − zn , fv〉 ⎪
⎧ x1 � x ∈ C � C 1 ,
⎪
⎪
⎪
⎪ yn � xn − cA∗ I − T[n] Axn , where [n] � n mod m,
⎪
⎪
Jzn − Jyn ⎪
⎪
≥ 〈v − zn , fv〉 − 〈v − zn , + ftn 〉 ⎪
⎪
λ ⎨ tn � PC yn − λfyn ,
⎪ z � P y − λft ,
⎪
� 〈v − zn , fv − fzn 〉 +〈v − zn , fzn − ftn 〉 ⎪
⎪ n C n n
(45) ⎪
⎪ �� �� �� �� �� �� �� ��
⎪
⎪ Cn+1 � v ∈ Cn : �yn − v�� ≤ ��xn − v��, ��zn − v�� ≤ ��yn − v��,
�
⎪
⎪
Jzn − Jyn ⎪
⎩
− 〈v − zn , 〉 xn+1 � PCn+1 x1 ,
λ
(48)
Jz − Jyn
≥ 〈v − zn , fzn − ftn 〉 − 〈v − zn , n 〉. where c ⊂ [a, b], a, b ∈ (0, (1/‖A‖2 )), and λ ∈ (0, (1/k)).
λ
Then, the sequence xn converges strongly to a point z ∈ Γ.
Using the fact that J is uniformly continuous on
bounded sets and f is Lipschitz continuous, as n ⟶ ∞, we
have Proof. Setting E1 � H1 and E2 � H2 in Theorem 2, the result
∗ is as follows. □
〈v − x , w〉 ≥ 0. (46)
Since B is a maximal monotone, 0 ∈ Bx∗ , and hence Remark 2. Corollary 2 complements Theorem 1 (the result
x ∈ VI(C, f), therefore x∗ ∈ Γ.
∗
of Tian and Jiang [12]) in the sense that strong convergence
of the sequence generated is obtained to a solution of the
split variational inequality problem involving common fixed
Corollary 1. Let E1 be a uniformly smooth and 2-uniformly
points of finite family of nonexpansive mappings, while
convex real Banach space and E2 a 2-uniformly smooth real
weak convergence of the scheme is obtained in Theorem 1
Banach space with smoothness constant d2 ∈ (0, 1). Let C be a
involving only a single nonexpansive map. Though, addi-
nonempty, closed, and convex subset of E1 . Let f : C ⟶ E∗1
tional condition of projecting the iterates on the half spaces
be a monotone and k-Lipschitz continuous map and
Cn at each step is imposed.
A : E1 ⟶ E2 be a bounded linear operator with its adjoint
A∗ such that A ≠ 0. Let T : E2 ⟶ E2 be a nonexpansive
map. Let Γ � z ∈ VI(C, f) : Az ∈ F(T) ≠ ∅. Let a se- Lemma 6. Let H be a real Hilbert space and Q be a nonempty
quence xn be generated by closed convex subset of H. Let g : Q ⟶ H be an α- inverse
⎪
⎧ x1 � x ∈ C � C1 , strongly monotone mapping, that is,
⎪
⎪ 〈x − y, gx − gy〉 ≥ α‖gx − gy‖2 for any x, y ∈ Q. Let
⎪
⎪
⎪
⎪ yn � J− 1 Jxn − cA∗ J2 (I − T)Axn , μ ∈ (0, 2α). Then, the mapping PQ (I − μg) is nonexpansive.
⎪
⎪
⎪
⎪
⎨ tn � ΠC J− 1 Jyn − λfyn ,
⎪ z � Π J− 1 Jy − λft ,
⎪
⎪
⎪ n C n n Proof. Letx, y ∈ H:
⎪
⎪
⎪
⎪ Cn+1 � v ∈ Cn : ϕ v, yn ≤ ϕ v, xn , ϕ v, zn ≤ ϕ v, yn , �� �
⎪
⎪
⎪ ��PQ (x − μgx) − PQ (y − μgy)���2 ≤ ‖(x − μgx) − (y − μgy)‖2
⎩x �Π x ,
n+1 Cn+1 1
(47) � ‖x − y‖2 − 2μ〈x − y, gx − gy〉 + μ2 ‖gx − gy‖2 ≤ ‖x − y‖2
where c ∈ [a, b], a, b ∈ (0, (1/d∗2 ‖A‖2 )), d∗2 being the μ
smoothness constant of E∗1 as in Lemma 2, and λ ∈ (0, (α/k)), − μ2 − 〈x − y, gx − gy〉
α
α being a positive constant as in Lemma 3. Then, the sequence
xn converges to a point z ∈ Γ. ≤ ‖x − y‖2 .
(49)
Proof. The result followed from Theorem 2 by setting
Hence, PQ (I − μg) is nonexpansive. □
T[n] � T, ∀n ∈ N. □
Corollary 2. Let H1 and H2 be real Hilbert spaces. Let C be a Corollary 3. Let H1 and H2 be real Hilbert spaces. Let C and
nonempty, closed, and convex subset of H1 . A : H1 ⟶ H2 Q be two nonempty, closed, and convex subset of H1 and H2 ,
be a bounded linear operator such that A ≠ 0, f : C ⟶ H1 respectively. Let A : H1 ⟶ H2 be a bounded linear operator
be a monotone and k-Lipschitz continuous map, and such that A ≠ 0, f : C ⟶ H1 be a monotone and k-Lipschitz
Ti : H2 ⟶ H2 , i � 1, 2, . . . , m, be a nonexpansive map. Let continuous map, and g: H2 ⟶ H2 be an inverse strongly
Γ � z ∈ VI(C, f): Az ∈ F(T) ≠ ∅. Let a sequence xn be monotone mapping. Setting Γ � z ∈ VI(C, f) : Az ∈
generated by VI(Q, g)} ≠ ∅. Let a sequence xn be defined by
Abstract and Applied Analysis 7
⎪
⎧ yn � xn − cA∗I − PQ (I − μg)Axn , (1) Tr is single valued;
⎪
⎪
⎪
⎪ (2) Tr is a firmly nonexpansive type, i.e.,
⎪
⎪ tn � PC yn − λfyn ,
⎪
⎪
⎨
⎪ zn � PC yn − λftn ,
⎪
⎪ �� �� �� �� �� �� �� �� 〈Tr x − Tr y, JTr x − JTr y〉 ≤ 〈Tr x − Tr y, x − y〉, ∀x, y ∈ E;
⎪
⎪
⎪
⎪ Cn+1 � v ∈ Cn : ��yn − v�� ≤ ��xn − v��, ��zn − v�� ≤ ��yn − v��,
⎪
⎪ (54)
⎩x �P x ,
n+1 Cn+1 1
(50) (3) F(Tr ) � EP(F);
where c ⊂ [a, b], a, b ∈ (0, (1/‖A‖2 )), λ ∈ (0(1/k)), and (4) EP(F) is closed and convex.
μ ∈ (0, 2α) . Then, the sequence xn converges to a point
z ∈ Γ. Theorem 3. Let E1 be a uniformly smooth and 2-uniformly
convex real Banach space and E2 a 2-uniformly smooth real
Proof. Since PQ (I − μg) is nonexpansive for μ ∈ (0, 2α) and Banach space with smoothness constant d2 ∈ (0, 1). Let C and
z ∈ VI(Q, g) if and only if z � PQ (I − μg)z for μ > 0, putting Q be two nonempty, closed, and convex subsets of E1 and E2 ,
T � PQ (I − μg) in Corollary 1, we get the desired result. □ respectively. Let A : E1 ⟶ E2 be a bounded linear operator
with its adjoint A∗ such that A ≠ 0, f : C ⟶ E∗1 be a
4. Application to Equilibrium Problem monotone and k-Lipschitz continuous map, and
F : Q × Q ⟶ be a bifunction satisfying conditions
Let C be a nonempty closed convex subset of a real Banach (A1)–(A4). Let Γ � z ∈ VI(C, f) : Az ∈ EP(F) ≠ ∅. Let a
space E and let F: C × C ⟶ R be a bifunction. sequence xn be generated by
The equilibrium problem with respect to F and C is to ⎧
⎪ x1 � x ∈ C � C 1 ,
find z ∈ C such that ⎪
⎪
⎪
⎪ − 1 ∗
⎪
⎪ yn � J Jxn − cA J2 I − Tr Axn ,
F(z, y) ≥ 0, ∀y ∈ C. (51) ⎪
⎪
⎪
⎪
⎨ tn � ΠC J− 1 Jyn − λfyn ,
The set of solutions of the equilibrium problem men- ⎪
⎪ zn � ΠC J− 1 Jyn − λftn ,
⎪
⎪
tioned above is denoted by EP(F). For solving the equi- ⎪
⎪
⎪
⎪ Cn+1 � v ∈ Cn : ϕ v, yn ≤ ϕ v, xn , ϕ v, zn ≤ ϕ v, yn ,
librium problem, we assume that F satisfies the following ⎪
⎪
⎪
⎩x �Π x ,
conditions: n+1 Cn+1 1
(A1) F(x, x) � 0 for all x ∈ C; (55)
(A2) F is monotone, i.e., F(x, y) + F(y, x) ≤ 0 ∀x,
where c ∈ [a, b], a, b ∈ (0, (1/d∗2 ‖A‖2 )), d∗2 is the smoothness
y ∈ C;
constant of E∗1 as in Lemma 2, λ ∈ (0, (α/k)), α being a
(A3) for each x, y, z ∈ C, limt↓0 F(tz + (1 − t)x, y) ≤ positive constant as in Lemma 3, and Tr is the resolvent of F
F(x, y); for r > 0. Then, the sequence xn converges to a point z ∈ Γ.
(A4) for each x ∈ C, y↦F(x, y) is convex and lower
semicontinous. Proof. Putting T � Tr in Corollary 1, we get the desired
result. □
Lemma 7 (see [24]). Let E be a reflexive, strictly convex, and 5. Application to Maximal Monotone Operator
uniformly smooth Banach space and C be a nonempty closed
convex subset of E. Let F : C × C ⟶ R be a bifunction A set valued mapping B ⊂ E × E∗ with domain
satisfying conditions (A1)–(A4), then for any x ∈ E and r > 0, D(B) � {x ∈ E : Bx ≠ ∅} and range R(B) � ∪{Bx:
there exists a unique point z ∈ C such that x ∈ D(B)} is said to be monotone if 〈x − y, x∗ − y∗ 〉 ≥ 0
1 whenever (x, x∗ ), (y, y∗ ) ∈ B. A monotone operator
F(z, y) + 〈y − z, jz − jx〉 ≥ 0, ∀y ∈ C. (52) B ⊂ E × E∗ is said to be maximal monotone if its graph
r
G(B) � (x, y) : y ∈ Bx is not properly contained in the
graph of any other monotone mapping. We know that if B is
Lemma 8 (see [24]). Let E be a reflexive, strictly convex, and a maximal monotone, then the zero of B,
smooth Banach space and C be a nonempty closed convex B− 1 (0) � {x ∈ E : 0 ∈ Bx} is closed and convex. If E is a
subset of E. Let F: C × C ⟶ R be a bifunction satisfying smooth, strictly convex, and reflexive Banach space, then a
conditions (A1)–(A4), then for any x ∈ E and r > 0, define a monotone operator B : E ⟶ E∗ is maximal if and only if
mapping Tr : E ⟶ C by R(J + rB) � E∗ for each r > 0. Let E be a smooth, strictly
1 convex, and reflexive Banach space, C be a nonempty closed
Tr x � x ∈ C: F(z, y) + 〈y − z, jz − jx〉 ≥ 0 ∀y ∈ C. convex subset of E, and B ⊂ E × E∗ be a monotone operator
r
satisfying
(53)
D(B) ⊂ C ⊂ J− 1 (J + rB), (56)
Then, the following holds:
8 Abstract and Applied Analysis
for all r > 0. If B is a maximal monotone, then (56) holds for minimization problem. We denote the set of solution of the
C � D(B) and we can define the resolvent Sr : C ⟶ D(B) constraint convex minimization problem by argminx∈C ϕ(x).
by
Sr x � {x ∈ E: Jx ∈ Jz + rBz}, (57) Lemma 9 (see [12]). Let H be a real Hilbert space and C be a
− 1 nonempty closed convex subset of H. Let ϕ be a convex
for all x ∈ C, i.e., Sr � (J + rB) J. We know the following function of H into R. If ϕ is differentiable, then z is a solution
(see [23, 25–27]): of the constraint convex minimization problem if and only if
(1) Sr is single valued, z ∈ VI(C, ▽ϕ).
(2) F(Sr ) � B− 1 0, where F(Sr ) is the set of fixed points
of Sr , Theorem 5. Let H1 and H2 be real Hilbert spaces. Let C and
(3) Sr is a firmly nonexpansive type, i.e., Q be two nonempty, closed, and convex subset of H1 and H2 ,
respectively. Let A : H1 ⟶ H2 be a bounded linear operator
such that A ≠ 0, and f : C ⟶ H1 be a monotone and
〈Sr x − Sr y, JSr x − JSr y〉 ≤ 〈Sr x − Sr y, x − y〉, ∀x, y ∈ E. k-Lipschitz continuous. Let ϕ: H2 ⟶ R be a differentiable
(58) convex function and suppose ▽ϕ is α-inverse strongly
monotone mapping. Setting Γ � z ∈ VI(C, f) : Az ∈
argminy∈Q ϕ(y)}, assume that Γ ≠ ∅. Let a sequence
Theorem 4. Let E1 be a uniformly smooth and 2-uniformly
convex real Banach space and E2 a 2-uniformly smooth real ⎪
⎧
⎪ yn � xn − cA∗ I − PQ (I − μ▽ϕ)Axn ,
⎪
⎪
Banach space with smoothness constant d2 ∈ (0, 1). Let C and ⎪
⎪
⎪
⎪ tn � PC yn − λfyn ,
Q be two nonempty, closed, and convex subsets of E1 and E2 , ⎪
⎨
respectively; A : E1 ⟶ E2 be a bounded linear operator with ⎪ zn � PC yn − λftn ,
⎪
⎪ �� �� �� �� �� �� �� ��
its adjoint A∗ such that A ≠ 0; f : C ⟶ E∗1 be a monotone ⎪
⎪
⎪
⎪ Cn+1 � v ∈ Cn : ��yn − v�� ≤ ��xn − v��, ��zn − v�� ≤ ��yn − v��,
and k-Lipschitz continuous map; and B : E ⟶ E∗ be a ⎪
⎪
⎩x �P x ,
maximal monotone operator. Let Γ � z ∈ VI(C, f): Az ∈ n+1 Cn+1 1
B− 1 (0)} ≠ ∅. Let a sequence xn be generated by (61)
2
⎪
⎧ x1 � x ∈ C � C1 , where c ⊂ [a, b], a, b ∈ (0, (1/‖A‖ )), λ ∈ (0, (1/k)), and
⎪
⎪ μ ∈ (0, 2α) . Then, the sequence xn converges to a point
⎪
⎪ − 1 ∗
⎪
⎪ yn � J Jxn − cA J2 I − Sr Axn , z ∈ Γ.
⎪
⎪
⎪
⎪ tn � ΠC J− 1 Jyn − λfyn ,
⎨
⎪
⎪ zn � ΠC J− 1 Jyn − λftn , Proof. Putting T � ▽ϕ in Corollary 3, by Lemma 9 we get
⎪
⎪
⎪
⎪ the desired result. □
⎪
⎪ Cn+1 � v ∈ Cn : ϕ v, yn ≤ ϕ v, xn , ϕ v, zn ≤ ϕ v, yn ,
⎪
⎪
⎪
⎩ xn+1 � ΠCn+1 x1 ,
7. Application to Split Minimization Problem
(59)
Let H1 and H2 be real Hilbert spaces. Let C and Q be two
where c ∈ [a, b], a, b ∈ (0, (1/d∗2 ‖A‖2 )), d∗2 being the nonempty, closed, and convex subset of H1 and H2 , re-
smoothness constant of E∗1 as in Lemma 2, λ ∈ (0, (α/k)), α spectively, and ϕ1 : H1 ⟶ R, ϕ2 : H2 ⟶ R be two convex
being a positive constant as in Lemma 3, and Sr is the re- functions. Let AH1 ⟶ H2 be a bounded linear operator.
solvent of B for r > 0. Then, the sequence xn converges to a The problem of finding z satisfying the conditions
point z ∈ Γ.
z ∈ argminx∈C ϕ1 (x) such that Az ∈ argminy∈Q ϕ2 s(y),
6. Application to Constrained Convex Theorem 6. Let H1 and H2 be real Hilbert spaces. Let C and
Minimization Problem Q be two nonempty, closed, and convex subset of H1 and H2 ,
respectively. Let A: H1 ⟶ H2 be a bounded linear operator
The problem of finding y ∈ C such that such that A ≠ 0, and ϕ1 : H1 ⟶ R and ϕ2 : H2 ⟶ R be two
ϕ(y) � minx∈C ϕ(x), (60) differentiable convex functions. Suppose that ▽ϕ1 is k-Lip-
schitz continuous and ▽ϕ2 is α-inverse strongly monotone
where C is a nonempty closed convex subset of H and ϕ is a mapping, let Γ � z ∈ argminx∈C ϕ1 (x): Az ∈ argminy∈Q ϕ2
real-valued convex function is called constraint convex (y)} ≠ ∅. Let a sequence
Abstract and Applied Analysis 9
n X(n)
Convergence of the sequence x(n) generated by (67)
1 0.0059 1
2 0.0029
0.9
3 0.0015
0.8
4 0.0007
5 0.0004 0.7
9 0.0000 0.4
10 0.0000
0.3
11 0.0000
12 0.0000 0.2
13 0.0000
0.1
14 0.0000
15 0.0000 0
0 5 10 15 20 25
Number of iterations:n
Values of the sequence x(n) generated by (67)
Figure 1: The table and graph of sequence xn generated by (67) versus number of iterations n : � {0, 1, . . . , 15} with an initial choice of
x0 � 1.0000.
⎪
⎧ yn � xn − cA∗ I − PQ I − μ▽ϕ1 Axn , 1
⎪
⎪ fx � x, ∀x ∈ C. (65)
⎪
⎪ 3
⎪
⎪ tn � PC yn − λ▽ϕ2 yn ,
⎪
⎪
⎨ Then, f is monotone and VI(C, f) � 0.
⎪ zn � PC yn − λ▽ϕ2 tn ,
⎪
⎪ �� �� �� �� �� �� �� �� Let A : C ⟶ C be defined by
⎪
⎪
⎪
⎪ Cn+1 � v ∈ Cn : ��yn − v�� ≤ ��xn − v��, ��zn − v�� ≤ ��yn − v��, 1
⎪
⎪ Ax � x, ∀x ∈ C. (66)
⎩x �P x , 2
n+1 Cn+1 1
(63) Then, A is a bounded linear operator and ‖A‖2 � (1/4),
∗
A y � (1/2)y.
where c ⊂ [a, b], a, b ∈ (0, (1/‖A‖2 )), λ ∈ (0, (1/k)), and When z ∈ VI(C, f), Az � 0 ∈ F(T0 ) ∩ F(T1 ).
μ ∈ (0, 2α). Then, the sequence xn converges to a point So, Γ � z ∈ VI(C, f) : Az ∈ F(T0 ) ∩ F(T1 ) ≠ ∅.
z ∈ Γ. Clearly, Γ � {0}. Taking λ � (3/n2 ) and c � (4/n2 ). It
follows from Theorem 2 that a sequence xn is generated by
Proof. Since ϕ1 is convex, we have ▽ϕ1 is monotone. Putting the following algorithm:
f � ▽ϕ1 and g � ▽ϕ2 in Corollary 3 and by Lemma 9, we
get the desired result. □ ⎪
⎧
⎪ yn � xn − cA∗ I − T[n] Axn ,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
8. Numerical Example ⎪
⎪ tn � PC yn − λfyn ,
⎪
⎪
⎪
⎪
In this section, we present a numerical example to show the ⎪
⎪
⎪
⎪
convergence of a sequence generated by our algorithm. Let ⎪ zn � PC yn − λftn ,
⎨
E1 � E2 � R, C � [0, ∞). ⎪ (67)
⎪
⎪
⎪
⎪ 2 2
Set m � 2 and Let T0 , T1 : C ⟶ C be defined by ⎪ Cn+1 � v ∈ Cn : v ≤ xn − zn ,
⎪
⎪
2 ⎪
⎪ 2xn − 2zn
T0 x � x, ∀x ∈ C, ⎪
⎪
⎪
⎪
3 ⎪
⎪
(64) ⎪
⎪
⎪
⎪ x2 − z2n
1 ⎪
⎩ xn+1 � PCn+1 x1 � n ,
T1 x � x, ∀x ∈ C. 2xn − 2zn
2
Then, T0 and T1 are nonexpansive. converges strongly to 0 ∈ Γ (see Figure 1).
Let f : C ⟶ C be defined by
10 Abstract and Applied Analysis