Author's personal copy

J. Math. Anal. Appl. 415 (2014) 513–524

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and Applications

www.elsevier.com/locate/jmaa

On split common fixed point problems

Rapeepan Kraikaew, Satit Saejung ∗
Department of Mathematics, Faculty of Science, Khon Kaen University, Khon Kaen 40002, Thailand

a r t i c l e i n f o a b s t r a c t

Article history: Based on the convergence theorem recently proved by the second author, we modify
Received 8 February 2013 the iterative scheme studied by Moudafi for quasi-nonexpansive operators to obtain
Available online 29 January 2014 strong convergence to a solution of the split common fixed point problem. It is
Submitted by B. Sims
noted that Moudafi’s original scheme can conclude only weak convergence. As
Keywords: a consequence, we obtain strong convergence theorems for split variational inequality
Split common fixed point problems problems for Lipschitz continuous and monotone operators, split common null point
Split feasibility problem problems for maximal monotone operators, and Moudafi’s split feasibility problem.
Split variational inequality problem © 2014 Elsevier Inc. All rights reserved.
Split null point problem

1. Introduction

Let C and Q be closed convex subsets of Hilbert spaces H1 and H2 , respectively and A : H1 → H2 be
a bounded linear operator. The split feasibility problem (SFP) which was first introduced by Censor and
Elfving [4] is to find

‚∈C
x such that A‚
x ∈ Q. (1)

Suppose that PC and PQ are the (orthogonal) projections onto the sets C and Q, respectively. Assuming
‚ ∈ H1 solves (1) if and only if
that SFP is consistent (i.e., (1) has a solution), it is not difficult to see that x
it solves the fixed-point equation
! "
‚ = PC I + γA∗ (PQ − I)A x
x ‚,

where γ > 0 is any positive constant, I is the identity operator and A∗ denotes the adjoint of A. To solve (1),
in the setting of the finite dimensional case, Byrne [2] proposed the following so-called CQ algorithm:
! "
xn+1 = PC xn + γAt (PQ − I)Axn , n ∈ N,

* Corresponding author.
E-mail addresses: [email protected] (R. Kraikaew), [email protected] (S. Saejung).

0022-247X/$ – see front matter © 2014 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jmaa.2014.01.068
 Author's personal copy

514 R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524

where γ ∈ ]0, L2 [, with L being the largest eigenvalue of the matrix At A (t stands for matrix transposition).
SFP is important and has been widely studied because it plays a prominent role in the signal processing
and image reconstruction problem. Initiated by SFP, several “split type” problems have been investigated
and studied, for example, the split variational inequality problem (SVIP) and the split null point problem
(SCNP). We will consolidate these problems. Let U : H1 → H1 and T : H2 → H2 be two operators with
nonempty fixed point sets Fix(U ) := {x ∈ H1 : x = U x} and Fix(T ), respectively. The split common fixed
point problem (SCFP) is to find

‚ ∈ Fix(U ) such that

x x ∈ Fix(T ).
A‚

If U := PC and T := PQ , then Fix(U ) = C and Fix(T ) = Q and hence SCFP immediately reduces to
SFP. In the case that U and T are directed operators, Censor and Segal [5] proposed and proved, still in
finite-dimensional spaces, the convergence of the following algorithm:
! "
xn+1 = U xn + γAt (T − I)Axn , n ∈ N,

where γ and L are as mentioned before. Note that a class of directed operators includes the metric projec-
tions. Hence the result of Censor et al. recovers Byrne’s CQ algorithm.
Moudafi [9] recently studied the convergence properties of a relaxed algorithm for SCFP for a class of
quasi-nonexpansive operators T such that I − T is demiclosed at zero. He also proved a weak convergence
theorem as shown below.

Theorem 1.1. Given a bounded linear operator A : H1 → H2 , let U : H1 → H1 and T : H2 → H2 be two

quasi-nonexpansive operators with nonempty sets Fix(U ) = C and Fix(T ) = Q. Assume that I − U and
I − T are demiclosed at zero. Suppose Γ := {x ∈ C: Ax ∈ Q} ̸= ? and define an iterative sequence {xn } by
Y
x ∈ H1 ,
] 0
_
un = xn + γβA∗ (T − I)Axn ,
_
[
xn+1 = (1 − αn )un + αn U (un ),
1
where β ∈ ]0, 1[, αn ∈ ]0, 1[ and γ ∈ ]0, λβ [ with λ = ∥A∗ A∥. Then {xn } converges weakly to x
‚ ∈ Γ provided
that αn ∈ ]δ, 1 − δ[ for a small enough δ > 0.

Note that, in the setting of finite dimensional spaces, weak and strong convergences are equivalent.
Differently, in infinite dimensional cases, they are not the same. Furthermore, Moudafi’s result [9] can
guarantee only weak convergence. In most cases, strong convergence is more desirable than weak convergence.
In this paper, we slightly modify the algorithm to obtain a strong convergence.

2. Definitions and preliminaries

Throughout, let H be a real Hilbert space with inner product ⟨·,·⟩ and the induced norm ∥ · ∥. We denote
the strong and weak convergence of a sequence {xn } in H to an element x ∈ H by xn → x and xn ⇀ x,
respectively. For a closed convex subset C of H, the (metric) projection PC : H → C is defined for each
x ∈ H as the unique element PC x ∈ C such that
) *
∥x − PC x∥ = inf ∥x − z∥: z ∈ C .

For x ∈ H and y ∈ C, it is known that

y = PC x ⇐⇒ ⟨y − x, z − y⟩ > 0 for all z ∈ C.

 Author's personal copy

R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524 515

In this paper, the fixed-point set of an operator T : H → H is denoted by Fix(T ), that is, Fix(T ) =
{x ∈ H: x = T x}.
Let us recall some definitions of operators involved in our study.

Definition 2.1. An operator T : H → H is called:

• L-Lipschitzian if

∥T x − T y∥ 6 L∥x − y∥ for all x, y ∈ H;

• a contraction if it is α-Lipschitzian with α ∈ [0, 1[, and in this case, we also say that T is a contraction
with the coefficient α;
• nonexpansive if T is 1-Lipschitzian;
• quasi-nonexpansive if Fix(T ) ̸= ? and

∥T x − p∥ 6 ∥x − p∥ for all x ∈ H, p ∈ Fix(T );

equivalently, for all x ∈ H and p ∈ Fix(T ),

1
⟨x − T x, p − x⟩ 6 − ∥x − T x∥2 ;
2

• strongly quasi-nonexpansive if T is quasi-nonexpansive and

xn − T x n → 0

whenever {xn } is a bounded sequence in H and ∥xn − p∥ − ∥T xn − p∥ → 0 for some p ∈ Fix(T );

• monotone if

⟨T x − T y, x − y⟩ > 0 for all x, y ∈ H.

Proposition 2.2. If T : H → H is a nonexpansive operator, then the following inequality holds for all x, y ∈ H

+ , 1. .2
x − y, (I − T )x − (I − T )y > .(I − T )x − (I − T )y . .
2

Proof. Since T is nonexpansive, we have

∥x − y∥2 > ∥T x − T y∥2

. .2
= .(I − T )x − (I − T )y − (x − y).
. .2 + ,
= .(I − T )x − (I − T )y . − 2 x − y, (I − T )x − (I − T )y + ∥x − y∥2 .

Therefore we get

+ , 1. .2
x − y, (I − T )x − (I − T )y > .(I − T )x − (I − T )y . . ✷
2

Corollary 2.3. Let S : H → H be a quasi-nonexpansive operator and

T := (1 − α)I + αS,
 Author's personal copy

516 R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524

for some α ∈ ]0, 1]. Then, for all x ∈ H and p ∈ Fix(T ), we have the following inequality

1
⟨x − T x, p − x⟩ 6 − ∥x − T x∥2 .
2α

Proof. Obviously, Fix(T ) = Fix(S). It follows from Proposition 2.2 that

α 1
⟨x − T x, p − x⟩ = α⟨x − Sx, p − x⟩ 6 − ∥x − Sx∥2 = − ∥x − T x∥2 .
2 2α

The proof is finished. ✷

3. Main results

Let us recall first the result proved by the second author.

Theorem 3.1. (See [13].) Let C be a closed and convex subset of a Hilbert space H and let T : C → C be a
strongly quasi-nonexpansive operator such that I − T is demiclosed at zero. Suppose that x0 ∈ C and {xn }
is a sequence generated iteratively by x1 ∈ C and

xn+1 = αn x0 + (1 − αn )T xn ,
q∞
where {αn } is a sequence in ]0, 1[ such that limn→∞ αn = 0 and n=1 αn = ∞. Then {xn } converges
strongly to a fixed point PFix(T ) x0 of T .

Recall that an operator T is demiclosed at zero [15] if

T x = 0 whenever xn ⇀ x and T xn → 0.

3.1. The split common fixed point problem

Throughout this paper, let Γ := {x ∈ Fix(U ): Ax ∈ Fix(T )}. It is clear that Γ is closed and convex.

Theorem 3.2. Let U : H1 → H1 be a strongly quasi-nonexpansive operator and T : H2 → H2 be a quasi-

nonexpansive operator such that both I − U and I − T are demiclosed at zero. Let A : H1 → H2 be a bounded
linear operator with L = ∥A∗ A∥. Suppose that Γ ̸= ?. Let {xn } ⊂ H1 be a sequence generated by
I
x0 ∈ H1 ,
! "
xn+1 = αn x0 + (1 − αn )U xn + γA∗ (T − I)Axn ,

where the parameter γ and the sequence {αn } satisfy the following conditions:

(a) γ ∈ ]0, L1 [,
q∞
(b) {αn } ⊂ ]0, 1[, limn→∞ αn = 0 and n=0 αn = ∞.

Then xn → PΓ x0 .

The following lemma is extracted from Lemma 6.2 of [6] which is needed for proving our main result.
 Author's personal copy

R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524 517

Lemma 3.3. Let H1 and H2 be Hilbert spaces. Let T : H2 → H2 be a nonexpansive operator and A : H1 → H2
be a bounded linear operator with L = ∥A∗ A∥. For a positive real number γ, define the operator W : H1 → H1
by

W := I + γA∗ (T − I)A.

Then the following hold:

• For all x, y ∈ H1 ,
. .2
∥W x − W y∥2 6 ∥x − y∥2 + γ(γL − 1).(T − I)Ax − (T − I)Ay . .

In addition, if T := (1 − α)I + αS where S : H2 → H2 is a nonexpansive operator, then

3 4
2 2 1 . .
.(T − I)Ax − (T − I)Ay .2 .
∥W x − W y∥ 6 ∥x − y∥ + γ γL −
α

• If Ax ∈ Fix(T ), then x ∈ Fix(W ) and the converse holds provided that γ ∈ ]0, L1 [.

Proof. • Let x, y ∈ H1 . Then we have

.! " ! ".2
∥W x − W y∥2 = . x + γA∗ (T − I)Ax − y + γA∗ (T − I)Ay .
. ! ".2
= .(x − y) + γA∗ (T − I)Ax − (T − I)Ay .
+ ! ", . ! ".2
= ∥x − y∥2 + 2γ x − y, A∗ (T − I)Ax − (T − I)Ay + γ 2 .A∗ (T − I)Ax − (T − I)Ay .
+ ,
= ∥x − y∥2 + 2γ Ax − Ay, (T − I)Ax − (T − I)Ay
+ ! " ! ",
+ γ 2 A∗ (T − I)Ax − (T − I)Ay , A∗ (T − I)Ax − (T − I)Ay
+ ,
= ∥x − y∥2 + 2γ Ax − Ay, (T − I)Ax − (T − I)Ay
+ ! " ,
+ γ 2 AA∗ (T − I)Ax − (T − I)Ay , (T − I)Ax − (T − I)Ay
+ ,
6 ∥x − y∥2 + 2γ Ax − Ay, (T − I)Ax − (T − I)Ay
. .. .2
+ γ 2 .AA∗ ..(T − I)Ax − (T − I)Ay . .

Therefore we have
+ ,
∥W x − W y∥2 6 ∥x − y∥2 + 2γ Ax − Ay, (T − I)Ax − (T − I)Ay
. .2
+ γ 2 L.(T − I)Ax − (T − I)Ay . . (2)

It follows from Proposition 2.2 that

. .2
∥W x − W y∥2 6 ∥x − y∥2 + γ(γL − 1).(T − I)Ax − (T − I)Ay . .

Furthermore, if T := (1 − α)I + αS where S is a nonexpansive operator, then

+ , + ,
Ax − Ay, (T − I)Ax − (T − I)Ay = α Ax − Ay, (S − I)Ax − (S − I)Ay
−α . .
6 .(I − S)Ay − (I − S)Ax.2
2
1 . .2
= − .(I − T )Ay − (I − T )Ax. .
2α
 Author's personal copy

518 R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524

Hence from (2) and Proposition 2.2, we obtain

3 4
2 2 1 . .
.(T − I)Ax − (T − I)Ay .2 .
∥W x − W y∥ 6 ∥x − y∥ + γ γL −
α

• It is obvious that Ax ∈ Fix(T ) implies x ∈ Fix(W ). To see the converse, let γ ∈ ]0, L1 [. Let x ∈ Fix(W )
and z ∈ H1 be such that Az ∈ Fix(T ). It follows that z ∈ Fix(W ) and hence we get
. .2
∥x − z∥2 = ∥W x − W z∥2 6 ∥x − z∥2 + γ(γL − 1).(T − I)Ax. .

Since γL < 1, we have (T − I)Ax = 0, that is, Ax ∈ Fix(T ). ✷

Corollary 3.4. Let T : H2 → H2 be a quasi-nonexpansive operator and A, W be operators defined as in

Lemma 3.3. Then
. .2
∥W x − z∥2 6 ∥x − z∥2 + γ(γL − 1).(T − I)Ax. ,

for all x ∈ H1 and z ∈ H1 such that Az ∈ Fix(T ).

Lemma 3.5. Let U : H1 → H1 be a strongly quasi-nonexpansive operator and T : H2 → H2 be a quasi-

nonexpansive operator. Let A : H1 → H2 be a bounded linear operator with L = ∥A∗ A∥. Define the operator
W : H1 → H1 as in Lemma 3.3 where γL < 1. Suppose that Fix(U ) ∩ Fix(W ) ̸= ? and {xn } is a bounded
sequence in H1 . Then the following are equivalent:

(a) U W xn − W xn → 0 and W xn − xn → 0;
(b) U W xn − xn → 0;
(c) ∥xn − p∥ − ∥U W xn − p∥ → 0 for some p ∈ Fix(U ) ∩ Fix(W ).

Proof. It is obvious that (a) ⇒ (b) ⇒ (c). We now show that (c) ⇒ (a). Suppose that ∥xn − p∥ −
∥U W xn − p∥ → 0 for some p ∈ Fix(U ) ∩ Fix(W ). By using Corollary 3.4 and the quasi-nonexpansiveness
of U , we get

∥U W xn − p∥ 6 ∥W xn − p∥ 6 ∥xn − p∥.

Therefore we have ∥W xn − p∥ − ∥U W xn − p∥ → 0. Since U is strongly quasi-nonexpansive, we have

U W xn − W xn → 0. Notice that ∥xn − p∥2 − ∥U W xn − p∥2 → 0. Using Corollary 3.4 again gives
. .2
γ(1 − γL).(T − I)Axn . 6 ∥xn − p∥2 − ∥U W xn − p∥2 → 0.

Since γL < 1, we get W xn − xn = γA∗ (T − I)Axn → 0. Then (a) is satisfied and the proof is finished. ✷

Proof of Theorem 3.2. To conclude the result, by using Theorem 3.1, it suffices to show that:

(♠) the operator U W is strongly quasi-nonexpansive, where W := I + γA∗ (T − I)A;

(♥) I − U W is demiclosed at zero.

We first note that Γ = Fix(U ) ∩ Fix(W ) = Fix(U W ). Indeed, it follows from Lemma 3.3 that
) *
Γ = x ∈ H1 : x ∈ Fix(U ) and Ax ∈ Fix(T )
) *
= x ∈ H1 : x ∈ Fix(U ) and x ∈ Fix(W )
= Fix(U ) ∩ Fix(W ).
 Author's personal copy

R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524 519

Then Fix(U ) ∩ Fix(W ) ̸= ?. We next show that Fix(U ) ∩ Fix(W ) = Fix(U W ). To see this, it suffices to
show Fix(U W ) ⊂ Fix(U ) ∩ Fix(W ). Then let p ∈ Fix(U ) ∩ Fix(W ) and x ∈ Fix(U W ). By using Lemma 3.5
with xn ≡ x, we get that W x = x and U W x = W x, that is, x ∈ Fix(U ) ∩ Fix(W ). So our assertion is
obtained. Combining this fact with Lemma 3.5, we have (♠). To prove (♥), let {xn } be a sequence such
that xn − U W xn → 0 and xn ⇀ x for some x ∈ H1 . It follows from Lemma 3.5 that xn − W xn → 0 and
yn − U yn → 0 where yn ≡ W xn . Notice that yn ⇀ x. Since I − U and I − T are demiclosed at zero, we
have x ∈ Fix(W ) ∩ Fix(U ) = Fix(U W ). ✷

4. Another split problems deduced from SCFP

4.1. The split variational inequality problem

Let H1 and H2 be two real Hilbert spaces. Given operators f : H1 → H1 and g : H2 → H2 , a bounded
linear operator A : H1 → H2 and nonempty closed convex subsets C ⊂ H1 and Q ⊂ H2 , the split variational
inequality problem (SVIP) is the problem of finding a point x‚ ∈ VIP(C, f ) such that A‚
x ∈ VIP(Q, g), that is,
I + ,
x‚ ∈ C such that f (‚ x), x − x
‚ > 0 for all x ∈ C,
+ ,
y‚ := A‚
x ∈ Q such that g(‚ y ), y − y‚ > 0 for all y ∈ Q.

This is equivalent to the problem of finding x

‚ ∈ Fix(PC (I − λf )) such that A‚x ∈ Fix(PQ (I − λg)) where
λ > 0. We denote the set of solutions by SVIP(A, C, Q, f, g). Therefore SVIP can be viewed as SCFP. Under
appropriate conditions of the operators f and g, we can apply our result for SVIP.
In the work of Censor et al. [6], the operators f and g are assumed to be α-inverse strongly monotone
where α > 0, that is,
+ , . .2 + , . .2
x − y, f (x) − f (y) > α.f (x) − f (y). and u − v, g(u) − g(v) > α.g(u) − g(v). ,

for all x, y ∈ H1 and u, v ∈ H2 . It is known that if f is α-inverse strongly monotone and λ ∈ ]0, 2α[ then
PC (I − λf ) is strongly quasi-nonexpansive and I − PC (I − λf ) is demiclosed at zero. Hence their result
becomes a special case of ours. However, since every α-inverse strongly monotone operator is monotone and
Lipschitz continuous, the latter class of operators is then more general. It is worth noting that there exists a
monotone Lipschitz continuous operator f such that PC (I − λf ) fails to be quasi-nonexpansive [7]. Thanks
to the extragradient method introduced by Korpelevi [8], we obtain a slight modification of such operators
and prove a strong convergence theorem for SVIP in the case when f and g are monotone and Lipschitz
continuous. More precisely, the following corollary is established.

Corollary 4.1. Let C and Q be nonempty closed convex subsets of Hilbert spaces H1 and H2 , respectively. Let
f : H1 → H1 and g : H2 → H2 be monotone and κ-Lipschitz continuous operators on C and Q, respectively
and A : H1 → H2 a bounded linear operator with ∥A∗ A∥ = L. Suppose that SVIP(A, C, Q, f, g) ̸= ?. Define
an iterative sequence {xn } ⊂ H1 by
I
x0 ∈ H1 ,
! "
xn+1 = αn x0 + (1 − αn )U xn + γA∗ (T − I)Axn ,

where γ ∈ ]0, L1 [,
! "
U := PC I − λf PC (I − λf ) ,
! "
T := PQ I − λgPQ (I − λg) , (3)

and λ ∈ ]0, κ1 [. Then the sequence {xn } converges strongly to x

‚ ∈ SVIP(A, C, Q, f, g).
 Author's personal copy

520 R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524

Before giving a proof, we present the following two lemmas.

Lemma 4.2. Let f : H → H be monotone and κ-Lipschitz continuous on C. Let S := PC (I − λf ) where

‚ and xn − Sxn → 0, then x
λ > 0. If {xn } is a sequence in C satisfying xn ⇀ x ‚ ∈ VIP(C, f ).

Proof. Since f is monotone and continuous, we have (see e.g., [14])

+ ,
‚ ∈ VIP(C, f )
x ⇐⇒ f (x), x − x
‚ >0 for all x ∈ C.

Let x ∈ C. Note that

+ ,
xn − λf (xn ) − Sxn , Sxn − x > 0 for all n ∈ N.

Next, we consider
+ , + ,
λf (x), xn − x 6 λf (xn ), xn − x
+ , + ,
= λf (xn ), xn − Sxn + λf (xn ), Sxn − x
+ , + ,
= λf (xn ), xn − Sxn − xn − λf (xn ) − Sxn , Sxn − x + ⟨xn − Sxn , Sxn − x⟩
+ ,
6 λf (xn ), xn − Sxn + ⟨xn − Sxn , Sxn − x⟩
. .
6 λ.f (xn ).∥xn − Sxn ∥ + ∥xn − Sxn ∥∥Sxn − x∥.

Hence
+ , . . 1
f (x), xn − x 6 .f (xn ).∥xn − Sxn ∥ + ∥xn − Sxn ∥∥Sxn − x∥.
λ

Since {f (xn )} is bounded, xn − Sxn → 0 and xn ⇀ x

‚, we have
+ , + ,
f (x), x
‚ − x = lim f (x), xn − x 6 0.
n→∞

The proof is finished. ✷

The following lemma is extracted from [12].

Lemma 4.3. Let f : H → H be a monotone and κ-Lipschitz operator on C and λ be a positive number. Let
V := PC (I − λf ) and S := PC (I − λf V ). Then, for all q ∈ VIP(C, f ), we have
! "
∥Sx − q∥2 6 ∥x − q∥2 − 1 − λ2 κ2 ∥x − V x∥2 .

In particular, if κλ < 1, then S is a strongly quasi-nonexpansive operator and Fix(S) = Fix(V ) = VIP(C, f ).

Proof. Let q ∈ VIP(C, f ). Note that

.! " .2 .! " .2
∥Sx − q∥2 6 . x − λf (V x) − q . − . x − λf (V x) − Sx.
+ ,
= ∥x − q∥2 + 2λ q − Sx, f (V x) − ∥x − Sx∥2
+ ,
= ∥x − q∥2 + 2λ q − V x, f (V x) − f (q)
+ , + ,
+ 2λ q − V x, f (q) + 2λ V x − Sx, f (V x) − ∥x − Sx∥2
+ ,
6 ∥x − q∥2 + 2λ V x − Sx, f (V x) − ∥x − Sx∥2
 Author's personal copy

R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524 521

+ ,
= ∥x − q∥2 + 2λ V x − Sx, f (V x) − ∥x − V x∥2 − 2⟨x − V x, V x − Sx⟩ − ∥V x − Sx∥2
+ ,
= ∥x − q∥2 − ∥x − V x∥2 − ∥V x − Sx∥2 + 2 x − λf (V x) − V x, Sx − V x .

Now we estimate the last term of the preceding expression

+ , + , + ,
x − λf (V x) − V x, Sx − V x = x − λf (x) − V x, Sx − V x + λf (x) − λf (V x), Sx − V x
+ ,
6 λf (x) − λf (V x), Sx − V x
6 λκ∥x − V x∥∥Sx − V x∥.

So we have

∥Sx − q∥2 6 ∥x − q∥2 − ∥x − V x∥2 − ∥V x − Sx∥2 + 2λκ∥x − V x∥∥Sx − V x∥

6 ∥x − q∥2 − ∥x − V x∥2 − ∥V x − Sx∥2 + λ2 κ2 ∥x − V x∥2 + ∥Sx − V x∥2
! "
= ∥x − q∥2 − 1 − λ2 κ2 ∥x − V x∥2 .

Assume further that κλ < 1 and let {xn } be a sequence in H such that ∥Sxn − q∥ − ∥xn − q∥ → 0 for
some q ∈ Fix(S). It follows from the above inequality that xn − V xn → 0 which can be easily deduced
to xn − Sxn → 0. Therefore S is strongly quasi-nonexpansive and it is not difficult to see that Fix(S) =
Fix(V ) = VIP(C, f ). ✷

Proof of Corollary 4.1. It follows from Lemma 4.3 that both operators U and T defined in (3) are strongly
quasi-nonexpansive. We next show that I − U is demiclosed at zero. Let {xn } be a sequence in H1 such
that xn − U xn → 0 and xn ⇀ x. Notice that ∥xn − q∥2 − ∥U xn − q∥2 → 0 for some q ∈ VIP(C, f ). Using
Lemma 4.3, we get
! ". .2
1 − λ2 κ2 .xn − PC (I − λf )xn . 6 ∥xn − q∥2 − ∥U xn − q∥2 → 0.

Thus xn − PC (I − λf )xn → 0. Therefore, by Lemma 4.2, we get x ∈ VIP(C, f ) = Fix(U ). Similarly, I − T

is also demiclosed at zero. Then the result follows from Theorem 3.2. ✷

4.2. The split common null point problem

Given two set-valued operators B1 ⊂ H1 × H1 and B2 ⊂ H2 × H2 and a bounded linear operator

A : H1 → H2 , the split common null point problem (SCNP) is the problem of finding

‚ ∈ H1
x such that 0 ∈ B1 (‚
x) and 0 ∈ B2 (A‚
x). (4)

Recently, Byrne et al. [3] proposed a strong convergence theorem for finding such a solution x
‚ when B1
and B2 are maximal monotone. Recall that B ⊂ H × H is:

• monotone if ⟨x − y, u − v⟩ > 0 for all (x, u) ∈ B1 and (y, v) ∈ B2 ;

• maximal monotone if it is monotone and its graph is not properly contained in the graph of any other
monotone operator.

For a maximal monotone operator B ⊂ H × H and λ > 0, we can define a single-valued operator

JλB =: (I + λB)−1 : H → H.
 Author's personal copy

522 R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524

It is known that JλB is firmly nonexpansive, that is, for all x, y ∈ H,

+ , . .2
x − y, JλB x − JλB y > .JλB x − JλB y . ,

and
! "
0 ∈ B(‚
x) ⇐⇒ ‚ ∈ Fix JλB .
x

Therefore, the problem (4) is equivalent to the problem of finding

! " ! "
‚ ∈ H1
x such that x
‚ ∈ Fix JλB1 and A‚
x ∈ Fix JλB2 ,

where λ is a positive real number, that is, the SCNP reduces to the SCFP.
The result of Byrne et al. [3] is a consequence of our Theorem 3.2.

Corollary 4.4. (See [3].) Let H1 and H2 be two Hilbert spaces. Given two set-valued maximal monotone
operators B1 : H1 → 2H1 and B2 : H2 → 2H2 and a bounded linear operator A : H1 → H2 with L = ∥A∗ A∥,
we define an iterative sequence {xn } by
I
x0 ∈ H1 ,
! ! " " (5)
xn+1 = αn x0 + (1 − αn )JλB1 xn + γA∗ JλB2 − I Axn ,

where the parameters λ, γ and the sequence {αn } satisfy the following conditions:

(a) λ > 0, γ ∈ ]0, L2 [,

q∞
(b) {αn } ⊂ ]0, 1[, limn→∞ αn = 0 and n=0 αn = ∞.

Suppose that the solution set of (4), says Γ , is nonempty. Then xn → x

‚ ∈ Γ.

Remark 4.5.

(1) Notice that Corollary 4.4 can be viewed as a corollary of our Theorem 3.2 for the following reasons.
(a) For a maximal monotone B and λ > 0, it is known that JλB is firmly nonexpansive and hence
nonexpansive. Moreover, I − JλB is demiclosed at zero [1] and

1 1
JλB = I + S,
2 2

for some nonexpansive operator S : H → H.

1
(b) For B2 and A defined as in Corollary 4.4, it follows from Lemma 3.3 with α = 2 that

.! " .2
∥W x − y∥2 6 ∥x − y∥2 + γ(γL − 2). JλB2 − I Ax. ,

for all x ∈ H1 and y ∈ H2 such that Ay ∈ Fix(JλB2 ) where W := I + γA∗ (JλB2 − I)A. So, in this
case, the parameter γ can be relaxed, that is, γ ∈ ]0, L2 [ instead of ]0, L1 [.
(2) Our Theorem 3.2 allows the parameter λ for JλB1 and JλB2 in Corollary 4.4 to be chosen differently.
(3) The strong limit x‚ of the sequence {xn } generated by (5) is indeed the nearest point projection of x0
onto the solution set Γ .
 Author's personal copy

R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524 523

4.3. Moudafi’s split feasibility problem

Let H1 , H2 and H3 be Hilbert spaces and C ⊂ H1 , Q ⊂ H2 be nonempty closed convex sets. Let
A : H1 → H3 , B : H2 → H3 be bounded linear operators. Moudafi’s split feasibility problem [10,11] is the
problem of finding

x ∈ C and y ∈ Q such that Ax = By. (6)

We will transform this problem into the original SFP. Let us denote

H1 := H1 × H2 ,
H2 := H3 × H3 ,
C := C × Q ⊂ H1 ,
) *
Q := (z, w) ∈ H2 : z = w .

Define a linear operator A : H1 → H2 by

A(x, y) = (Ax, By) for all (x, y) ∈ H1 .

If the set Γ := {(x, y) ∈ C: A(x, y) ∈ Q} is nonempty, then (x, y) ∈ H1 solves (6) if and only if

! "
(x, y) = PC I + γA∗ (PQ − I)A (x, y).

Note that:

• PC (x, y) = (PC x, PQ y) for all (x, y) ∈ H1 ;

2 , 2 ) for all (z, w) ∈ H2 ;
• PQ (z, w) = ( z+w z+w

• A (z, w) = (A z, B ∗ w) for all (z, w) ∈ H2 .

∗ ∗

As a consequence of our Theorem 3.2, the following iterative sequence {(xn , yn )} defined by

Y
_ x0 ∈ H1 ,
_
_
_
_ y0 ∈ H2 ,
_
_ 3 4
]
γ ∗
xn+1 = αn x0 + (1 − αn )PC xn + A (Byn − Axn ) ,
_
_ 2
_
_ 3 4
_
_ γ
_
[ yn+1 = αn y0 + (1 − αn )PQ yn + B ∗ (Axn − Byn ) ,
2

converges strongly to (‚x, y‚) which simultaneously solves Moudafi’s split feasibility problem (6) and is nearest
to the initial guess (x0 , y0 ).

Acknowledgments

The corresponding author is supported by the Thailand Research Fund and Khon Kaen University under
the TRF Research Career Development Grant RSA 5680002.
 Author's personal copy

524 R. Kraikaew, S. Saejung / J. Math. Anal. Appl. 415 (2014) 513–524

References

[1] F.E. Browder, Semicontractive and semiaccretive nonlinear mappings in Banach spaces, Bull. Amer. Math. Soc. 74 (1968)
660–665.
[2] C. Byrne, Iterative oblique projection onto convex sets and the split feasibility problem, Inverse Problems 18 (2002)
441–453.
[3] C. Byrne, Y. Censor, A. Gibali, S. Reich, The split common null point problem, J. Nonlinear Convex Anal. 13 (2012)
759–775.
[4] Y. Censor, T. Elfving, A multiprojection algorithm using Bregman projections in a product space, Numer. Algorithms 8
(1994) 221–239.
[5] Y. Censor, A. Segal, The split common fixed point problem for directed operators, J. Convex Anal. 16 (2009) 587–600.
[6] Y. Censor, A. Gibali, S. Reich, Algorithms for the split variational inequality problem, Numer. Algorithms 59 (2012)
301–323.
[7] C.E. Chidume, S.A. Mutangadura, An example of the Mann iteration method for Lipschitz pseudocontractions, Proc.
Amer. Math. Soc. 129 (2001) 2359–2363.
[8] G.M. Korpelevi , An extragradient method for finding saddle points and for other problems, Ekonom. i Mat. Metody 12
(1976) 747–756 (in Russian).
[9] A. Moudafi, A note on the split common fixed-point problem for quasi-nonexpansive operators, Nonlinear Anal. 74 (2011)
4083–4087.
[10] A. Moudafi, A relaxed alternating CQ-algorithm for convex feasibility problems, Nonlinear Anal. 79 (2013) 117–121.
[11] A. Moudafi, Alternating CQ-algorithm for convex feasibility and split fixed-point problems, J. Nonlinear Convex Anal.
(2014), in press.
[12] N. Nadezhkina, W. Takahashi, Weak convergence theorem by an extragradient method for nonexpansive mappings and
monotone mappings, J. Optim. Theory Appl. 128 (2006) 191–201.
[13] S. Saejung, Halpern’s iteration in Banach spaces, Nonlinear Anal. 73 (2010) 3431–3439.
[14] W. Takahashi, Nonlinear Functional Analysis, Yokohama Publishers, Yokohama, 2000.
[15] W. Takahashi, Introduction to Nonlinear and Convex Analysis, Yokohama Publishers, Yokohama, 2009.