R E S E A R C H
Open Access
Solutions for a variational inclusion problem
with applications to multiple sets split
feasibility problems
Lai-Jiu Lin
1*, Yew-Dann Chen
1,2and Chih-Sheng Chuang
3*Correspondence:
maljlin@cc.ncue.edu.tw
1Department of Mathematics,
National Changhua University of Education, Changhua, 50058, Taiwan
Full list of author information is available at the end of the article
Abstract
In this paper, we first study the set of common solutions for two variational inclusion problems in a real Hilbert space and establish a strong convergence theorem of this problem. As applications, we study unique minimum norm solutions of the following problems: multiple sets split feasibility problems, system of convex constrained linear inverse problems, convex constrained linear inverse problems, split feasibility problems, convex feasibility problems. We establish iteration processes of these problems and show the strong convergence theorems of these iteration processes. MSC: 47J20; 47J25; 47H05; 47H09
Keywords: firmly nonexpansive mapping; proximal-contraction; multiple sets split feasibility problem; maximum monotone; convex feasibility problem
1 Introduction
LetC,C, . . . ,Cmbe nonempty closed convex subsets of a real Hilbert spaceH. The well-known convex feasibility problem (CFP) is to findx∗∈Hsuch that
x∗∈C∩C∩ · · · ∩Cm.
The convex feasibility problem has received a lot of attention due to its diverse applica-tions in mathematics, approximation theory, communicaapplica-tions, geophysics, control theory, biomedical engineering. One can refer to [, ].
IfCandCare closed vector spaces of a real Hilbert spaceH, in , von Neumann showed that any sequence{xn}is generated from the method of alternating projections:
x∈H,x=PCx,x=PCx,x=PCx, . . . . Then{xn}converges strongly to somex¯∈ C∩C. IfCandCare nonempty closed convex subsets ofH, Bregman [] showed that the sequence{xn}generated from the method of alternating projection converges weakly to a point inC∩C. Hundal [] showed that the strong convergence fails ifCandCare nonempty closed convex subsets ofH. Recently, Boikanyoet al.[] proposed the following process:
xn+=JβGn
αnu+ ( –αn)xn+en
, n= , , . . . . (.)
and
xn=JρGn
λnu+ ( –λn)xn–+en
, n= , , . . . , (.)
whereu,x∈Hare given arbitrarily,GandGare two set-valued maximal monotone op-erators withJG
β = (I+βG)–, and{αn},{λn},{βn},{ρn},{en},{en}are sequences. Boikanyo
et al.[] proved that the sequence{xn}converges strongly to a pointx¯∈G–()∩G–() under suitable conditions.
The split feasibility problem (SFP) is to find a point
x∗∈Csuch thatAx∗∈Q,
whereC,Qare nonempty closed convex subsets of real Hilbert spacesH,H, respec-tively.A:H→H is a bounded linear operator. The split feasibility problem (SFP) in finite dimensional real Hilbert spaces was first introduced by Censor and Elfving [] for modeling inverse problems which arise from medical image reconstruction. Since then, the split feasibility problem (SFP) has received much attention due to its applications in signal processing, image reconstruction, approximation theory, control theory, biomed-ical engineering, communications, and geophysics. For examples, one can refer to [, , –] and related literature. A special case of problem (SFP) is the convexly constrained linear inverse problem in a finite dimensional real Hilbert space []:
(CLIP) Findx¯∈Csuch thatAx¯=b,
whereCis a nonempty closed convex subset of a real Hilbert spaceHandbis a given element of a real Hilbert spaceH, which has extensively been investigated by using the Landweber iterative method []:
xn+:=xn+γAT(b–Axn), n∈N.
Let C,C, . . . ,Cm be nonempty closed convex subsets of H, let Q,Q, . . . ,Qm be nonempty closed convex subsets ofH, and letA,A, . . . ,Am:H→Hbe bounded lin-ear operators. The well-known multiple sets split feasibility problem (MSSFP) is to find
x∗∈Hsuch that
x∗∈Cisuch thatAix∗∈Qi for alli= , , . . . ,m.
The multiple sets split feasibility problem (MSSFP) contains convex feasibility problem (CFP) and split feasibility problem (SFP) as special cases [, , ]. Indeed, Censoret al.
[] first studied this type of problem. Xu [] and Lopezet al.[] also studied this type of problem. In , Boikanyo and Moroşanu [] gave the following algorithm:
vn+:=anu+bnvn+cnJαGnvn, n∈N∪ {},
vn:=fnu+gnvn–+hnJβGnvn–, n∈N,
(.)
whereuis given inH, andG,G are two set-valued maximal monotone mappings on
proved that{vn}in (.) converges strongly to somex¯∈G–()∩G–() under suitable conditions.
Motivated by the above works, we consider the following algorithm:
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnJδGn(I–δnB)vn, n∈N∪ {},
vn:=fnu+gnvn–+hnJγGn(I–γnB)vn–, n∈N,
(.)
where G,G are two set-valued maximal monotone mappings on a real Hilbert space
H,B,B:C→H are two mappings,{an},{bn},{cn},{fn},{gn}, and{hn}are sequences in [, ]. We show that the sequence{vn}generated by (.) converges strongly to some
¯
x∈(B+G)–()∩(B+G)–() under suitable conditions.
Algorithm (.) contains as special cases the inexact proximal point algorithm (e.g., [, ]) and the generalized contraction proximal point algorithm (e.g., []). Our conclusion extends and unifies Boikanyo and Moroşanu’s result in [], Wang and Cui’s result in [] becomes a special case. For details, one can see Section .
In this paper, we first study the set of common solutions for two variational inclusion problems in a Hilbert space and establish a strong convergence theorem of this problem. As applications, we study unique minimum norm solutions of the following problems: multiple sets split feasibility problems, system of convex constrained linear inverse prob-lems, convex constrained linear inverse probprob-lems, split feasibility probprob-lems, convex feasi-bility problems. We establish iteration processes of these problems and show strong con-vergence theorems of these iteration processes.
2 Preliminaries
Throughout this paper, letH,H,H,Hdenote real Hilbert spaces with the inner product
·,· and the norm · ; letNbe the set of all natural numbers andR+be the set of all pos-itive real numbers. A set-valued mappingAwith domainD(A) onHis called monotone ifu–v,x–y ≥ for anyu∈Ax,v∈Ayand for allx,y∈D(A). A monotone operatorAis called maximal monotone if its graph{(x,y) :x∈D(A),y∈Ax}is not properly contained in the graph of any other monotone mapping. The set of all zero points ofAis denoted by
A–(),i.e.,A–() ={x∈H: ∈Ax}. In what follows, we denote the strong convergence and the weak convergence of{xn}tox∈Hbyxn→xandxnx, respectively. In order to facilitate our discussion, in the next section, we recall some facts. The following equality is easy to check:
αx+βy+γz
=αx+βy+γz–αβx–y–αγx–z–βγy–z (.) for eachx,y,z∈Handα,β,γ∈[, ] withα+β+γ = . Besides, we also have
x+y≤ x+ y,x+y (.)
for each x,y∈H. Let C be a nonempty closed convex subset of H, and a mapping
is said to be quasi-nonexpansive if Fix(T)=∅andTx–y ≤ x–yfor allx∈Cand
y∈Fix(T). A mappingT:C→His said to be firmly nonexpansive if
Tx–Ty≤ x–y– (I–T)x– (I–T)y
for everyx,y∈C. Besides, it is easy to see thatFix(T) is a closed convex subset ofCifT:
C→His a quasi-nonexpansive mapping. A mappingT:C→His said to beα -inverse-strongly monotone (α-ism) if
x–y,Tx–Ty ≥αTx–Ty
for allx,y∈Handα> .
The following lemmas are needed in this paper.
Lemma .[] Let A:H→Hbe a bounded linear operator,and let A∗be the adjoint
of A.Suppose that C is a nonempty closed convex subset ofH,and that G:C→His a
firmly nonexpansive mapping.Then A∗(I–G)A isA-ism,that is,
A A
∗(I–G)Ax–A∗(I–G)Ay
≤x–y,A∗(I–G)Ax–A∗(I–G)Ay
for all x,y∈H.
Lemma . Let C be a nonempty closed convex subset ofH,and let G:C→Hbe a firmly nonexpansive mapping.Suppose thatFix(G)is nonempty.Thenx–Gx,Gx–w ≥for each x∈Hand each w∈Fix(G).
LetCbe a nonempty closed convex subset ofH. Then, for eachx∈H, there is a unique elementx¯∈Csuch thatx–x¯=miny∈Cx–y. Here, we setPCx=x¯andPCis said to be the metric projection fromHontoC.
Lemma .[] Let C be a nonempty closed convex subset ofH,and let PCbe the metric
projection fromHonto C.Thenx–PCx,PCx–y ≥for each x∈Hand each y∈C.
For a set-valued maximal monotone operatorGonHandr> , we may define an oper-atorJG
r :H→HwithJrG= (I+rG)–which is called the resolvent mapping ofGforr.
Lemma .[, ] Let G:H→Hbe a set-valued maximal monotone mapping.Then we have that
(i) for eachα> ,JG
α is a single-valued and firmly nonexpansive mapping.
(ii) D(JG
α) =HandFix(JαG) =G–().
(iii) x–JαGx ≤x–JβGxfor eachx∈Hand allα,β∈(,∞)with <α≤β. Lemma .[] Let G be a set-valued maximal monotone operator onH.For a> ,we define the resolvent JaG= (I+aG)–.Then the following holds:
JαGx–JβGx
≤α–β
α
for allα,β> and x∈H.In fact,
JαGx–JβGx ≤ |α–β|
α J
G
αx–x
for allα,β> and x∈H.
Lemma .[] Let{Sn}be a sequence of real numbers that does not decrease at
infin-ity,in the sense that there exists a subsequence{Sni}i≥of{Sn}such that Sni<Sni+for all
i∈N.Consider the sequence{τ(n)}n≥n defined byτ(n)=max{n≤k≤n:τk<τk+}for some sufficiently large number n∈N.Then{τ(n)}n≥nis a nondecreasing sequence with
τ(n)→ ∞as n→ ∞,and for all n≥n,
Sτ(n)≤Sτ(n)+ and Sn≤Sτ(n)+.
In fact,max{Sτ(n),Sn} ≤Sτ(n)+.
Lemma .[] Let{an}and{bn}be two sequences in[, ].Let{en}be a sequence of
non-negative real numbers.Let{tn}and{kn}be two sequences of real numbers.Let{Sn}n∈Nbe
a sequence of nonnegative real numbers with
Sn+≤( –an)( –bn)Sn+antn+bnkn+en
for each n∈N.Assume that:
∞
n=
an=∞, lim sup
n→∞ tn≤, lim supn→∞ kn≤, and ∞
n=
en=∞.
Thenlimn→∞Sn= .
A mappingT:H→His said to be averaged ifT= ( –α)I+αS, whereS:H→His a nonexpansive mapping andα∈(, ).
Lemma .[] Let C be a nonempty closed convex subset of Hand T:C→Hbe a mapping.Then the following hold:
(i) Tis a nonexpansive mapping if and only ifI–T is -inverse-strongly monotone
(-ism).
(ii) IfSisν-ism,thenγS isγν-ism.
(iii) Sis averaged if and only ifI–Sisν-ismfor someν> .
Indeed,Sisα-averaged if and only ifI–Sis (α)-ismforα∈(, ). (iv) If S and T are averaged,then the compositionST is also averaged.
(v) If the mappings {Ti}ni=are averaged and have a common fixed point,then n
i=Fix(Ti) =Fix(T· · ·Tn)for eachn∈N.
Lemma .[] Let T be a nonexpansive self-mapping on a nonempty closed convex sub-set C ofH,and let{xn}be a sequence in C.If xnw andlimn→∞xn–Txn= ,then
LetCbe a nonempty closed convex subset ofH. The indicator functionιCdefined by
ιCx=
⎧ ⎨ ⎩
, x∈C,
∅, x∈/C
is a proper lower semicontinuous convex function and its subdifferential∂ιCdefined by
∂ιCx=
z∈H:y–x,z ≤ιC(y) –ιC(x),∀y∈H
is a maximal monotone operator []. Furthermore, we also define the normal coneNCu ofCatuas follows:
NCu=
z∈H:z,v–u ≤,∀v∈C.
We can define the resolventJλ∂iCof∂iCforλ> ,i.e.,
Jλ∂iCx= (I+λ∂iC)–x
for allx∈H. Since
∂iCx=
z∈H:iCx+z,y–x ≤iCy,∀y∈H
=z∈H:z,y–x ≤,∀y∈C=NCx
for allx∈C, we have that
u=Jλ∂iCx ⇔ x∈u+λ∂iCu
⇔ x–u∈λNCu
⇔ x–u,y–u ≤, ∀y∈C ⇔ u=PCx.
3 Main results
LetC,QandQbe nonempty closed convex subsets ofH,HandH, respectively. For eachi= , andκi> , letBibe aκi-inverse-strongly monotone mapping ofCintoH, and letGibe a set-valued maximal monotone mapping onHsuch that the domain ofGi is included inCfor eachi= , . LetFbe a firmly nonexpansive mapping ofHintoH andFbe a firmly nonexpansive mapping ofHintoH. NoteJλG= (I+λG)–andJrG= (I+rG)–for eachλ> andr> . LetA:H→Hbe a bounded linear operator,A:
H→Hbe a bounded linear operator, andA∗i be the adjoint ofAifori= , . Throughout this paper, we use these notations unless specified otherwise.
Theorem . Suppose that(B+G)–()∩(B+G)–()is nonempty,and{an},{bn},{cn},
{fn},{gn},{hn}are sequences in[, ]such that an+bn+cn= ,fn+gn+hn= , <an< ,
and <fn< for each n∈N.For an arbitrarily fixed u∈H,define a sequence{vn}by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnJδGn(I–δnB)vn, n∈N∪ {},
Thenlimn→∞vn=P(B+G)–()∩(B+G)–()u provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a≤δn≤b< κand <f≤γn≤g< κfor eachn∈Nand for some
a,b,f,g∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof Given any fixed¯v:=P(B+G)–()∩(B+G)–()u. Thenv¯=J
G
δn(I–δnB)v¯andv¯=J G γn(I–
γnB)v¯. SinceBisκ-ism and κδn>, it follows from Lemma . thatI–δnBis averaged. HenceI–δnBis nonexpansive. By Lemma .,JδGn(I–δnG) andJ
G
γn(I–γnG) are averaged. HenceJG
δn(I–δnG) andJ G
γn(I–γnG) are nonexpansive. By condition (iv), we may assume that there exist positive real numberscandhsuch thatcn≥> andhn≥h> for each
n∈N. For eachn∈N, we have that
vn–v¯
= fnu+gnvn–+hnJγGn(I–γnB)vn––v¯
≤fnu–v¯+gnvn––¯v+hn JγGn(I–γnB)vn––v¯
≤fnu–v¯+gnvn––¯v+hnvn––¯v
=fnu–v¯+ (gn+hn)vn––v¯
=fnu–v¯+ ( –fn)vn––v¯ (.)
and
vn+–v¯
= anu+bnvn+cnJδGn(I–δnB)vn–v¯
≤anu–v¯+bnvn–v¯+cn JδGn(I–δnB)vn–v¯
≤anu–v¯+bnvn–v¯+cnvn–v¯
=anu–v¯+ (bn+cn)vn–v¯
≤( –an)fnu–v¯+ ( –fn)vn––v¯
+anu–v¯ by (.)
=fn( –an)u–v¯+( –an)( –fn)vn––v¯+anu–v¯
= – ( –an)( –fn)u–v¯+ ( –an)( –fn)vn––v¯
≤maxu–v¯,vn––v¯
≤maxu–v¯,v–v¯
.
By the mathematical induction method, we know that{vn–},{vn}and{vn}are bounded sequences. By Lemma ., we have that
JG
δn(I–δnB)vn–v¯
= JG
δn(I–δnB)vn–J G
δn(I–δnB)v¯
≤ (I–δnB)vn– (I–δnB)¯v
– I–JG
δn(I–δnB)
vn–
I–JG
δn(I–δnB)
≤ vn–v¯– I–JδGn(I–δnB)
vn–
I–JG
δn(I–δnB)
¯ v
=vn–¯v– vn–JδGn(I–δnB)vn
for eachn∈N. Hence,
vn+–v¯
= an(u–v¯) +bn(vn–v¯) +cn
JG
δn(I–δnB)vn–v¯
≤ bn(vn–v¯) +cn
JG
δn(I–δnB)vn–v¯
+ anu–v¯,vn+–v¯
= ( –an) bn(vn–v¯) +cn
JG
δn(I–δnB)vn–v¯
+ anu–v¯,vn+–v¯
= ( –an)bnvn–v¯+cn J G
δn(I–δnB)vn–v¯
–bncn vn–JδGn(I–δnB)vn
+ anu–v¯,vn+–v¯
≤bnvn–v¯+cn JδGn(I–δnB)vn–v¯
+ anu–v¯,vn+–v¯
≤bnvn–v¯+cn
vn–v¯– vn–JδGn(I–δnB)vn
+ anu–v¯,vn+–v¯
= (bn+cn)vn–v¯+ anu–v¯,vn+–v¯ –cn vn–JδGn(I–δnB)vn
, (.)
where
bn:= bn
bn+cn
, cn:= cn
bn+cn .
Similarly, we have
vn–v¯≤(gn+hn)vn––v¯+ fnu–v¯,vn–v¯
–hn JγGn(I–γnB)vn––vn–
. (.)
Consequently, it follows from (.) and (.) that
vn+–v¯
≤(bn+cn)vn–v¯+ anu–v¯,vn+–v¯ –cn JδGn(I–δnB)vn–vn
≤( –an)( –fn)vn––¯v+ fnu–v¯,vn–v¯
–hn JγGn(I–γnB)vn––vn–
+ anu–¯v,vn+–¯v
–cn JδGn(I–δnB)vn–vn
= ( –an)( –fn)vn––v¯+ fn( –an)u–v¯,vn–v¯
–hn( –an) JG
γn(I–γnB)vn––vn–
+ anu–v¯,vn+–v¯
–cn JδGn(I–δnB)vn–vn
For eachn∈N, setSn:=vn––v¯. ThenSn+=vn+–v¯and (.) become
Sn+≤( –an)( –fn)Sn+ anu–v¯,vn+–v¯ + fn( –an)u–v¯,vn–¯v
–cn JδGn(I–δnB)vn–vn
–hn( –an) JG
γn(I–γnB)vn––vn–
≤( –an)( –fn)Sn+ anu–v¯,vn+–v¯ + fn( –an)u–v¯,vn–¯v. (.)
Case : {Sn} is eventually decreasing,i.e., there exists a natural numberN such that
vn+–v¯ ≤ vn––v¯for eachn≥N. So,{Sn}is convergent andlimn→∞vn––v¯ exists. For alln∈N,cn≥c,hn≥hand (.), we have that
≤c JG
δn(I–δnB)vn–vn
+h( –an) JG
γn(I–γnB)vn––vn–
≤( –an)( –fn)Sn–Sn++ anu–v¯,vn+–v¯ + fn( –an)u–¯v,vn–v¯ . (.)
Noting via condition (i) and the fact that{vn}is bounded that
lim n→∞
( –an)( –fn)Sn–Sn+
= ,
lim
n→∞anu–v¯,vn+–v¯ =n→∞lim fn( –an)u–v¯,vn–¯v = ,
we conclude from (.) that
lim n→∞
c JG
δn(I–δnB)vn–vn
+h( –an) JG
γn(I–γnB)vn––vn–
= .
Therefore,
lim n→∞ J
G
δn(I–δnB)vn–vn =n→∞lim J G
γn(I–γnB)vn––vn– = . (.) Since{vn}is a bounded sequence inH, there is a subsequence{v(n)k}of{vn}such that
v(n)k x¯∈Hand lim sup
n→∞ u–v¯,vn–v¯ =k→∞limu–v¯,v(n)k–v¯ =u–v¯,x¯–v¯ .
On the other hand, <a≤δn≤b< κ, there exists a subsequence{δnkj}of{δn}such that
{δnkj}converges to a numberδ¯∈[a,b]. By Lemma ., we have
v(n)kj –JδG¯(I–δ¯B)v(n)kj
≤ v(n)kj–JδGnkj (I–δnkjB)v(n)kj
+ JG
δnkj(I–δ¯B)v(n)kj–JδG¯(I–δ¯B)v(n)kj + JG
δnkj(I–δnkjB)v(n)kj –J
G
≤ v(n)kj–JδG
nkj(I–δnkjB)v(n)kj +|δnkj–δ¯|Bunkj +|
δnkj–δ¯|
¯
δ J
G
¯
δ (I–δ¯B)v(n)kj– (I–δ¯B)v(n)kj →. (.) By (.) and Lemma .,x¯∈Fix(JG
¯
δ (I–δ¯B)) = (B+G)
–(). Since <c≤γ
n≤d< κ, there exists a subsequence{γnkj}of{γn}such that{γnkj}converges to a numberγ¯∈[c,d]. We have that
vn+–vn= anu+bnvn+cnJδGn(I–δnB)vn–vn
≤anu–vn+cn JδGn(I–δnB)vn–vn (.) and
vn–vn–= fnu+gnvn–+hnJγGn(I–γnB)vn––vn–
≤fnu–vn–+hn JγGn(I–γnB)vn––vn– . (.) Since{vn}is bounded, we conclude from (.), (.), (.), and conditions (i), (ii) that
lim
n→∞vn+–vn= . (.)
By Lemma ., we have
v(n)kj+–Jγ¯(I–γ¯B)v(n)kj+
≤ v(n)kj+–Jγnkj(I–γnkjB)v(n)kj+ + Jγ
nkj(I–γnkjB)v(n)kj+–Jγnkj(I–γ¯B)v(n)kj+ + Jγnkj(I–γ¯B)v(n)kj+–Jr¯(I–rB¯ )v(n)kj+
≤ v(n)kj+–Jγnkj(I–γnkjB)v(n)kj+ +|γnkj –γ¯|Bv(n)kj+
+|
γnkj–γ¯|
¯
γ Jγ¯(I–γ¯B)v(n)kj+– (I–γ¯B)v(n)kj+ →. (.)
Sincelimn→∞vn+–vn= , we know thatv(n)kj+x¯. By (.) and Lemma ., we know thatJG
¯
γ (I–γ¯B)x¯=x¯. So,x¯∈(B+G)–(). This shows thatx¯∈(B+G)–()∩ (B+G)–(). It follows fromv¯:=P(B+G)–()∩(B+G)–()uand Lemma . that
lim sup
n→∞ u–v¯,vn–v¯ =u–v¯,x¯–v¯ ≤. (.)
By (.) and (.),
lim sup
n→∞ u–v¯,vn+–v¯ =lim sup
n→∞
u–¯v,vn+–vn +u–v¯,vn–v¯
≤lim sup
Applying Lemma . to inequality (.) withtn= u–v¯,vn+–¯v andkn= ( –an)u–
¯
v,vn–¯v, we obtain from (.) and (.) and conditions (i), (ii) thatlimn→∞Sn= . That is, limn→∞vn–=¯v. And then it follows from (.) that limn→∞vn=v¯. Thus, limn→∞vn=v¯.
Case : Suppose that{Sn}is not an eventually decreasing sequence. Let{Sni}be a sub-sequence of{Sn}such thatSni≤Sni+for alli≥, also consider the sequence of integers
{τ(n)}n≥n, defined byτ(n) =max{k≤n,Sk<Sk+}, for somen(nis a sufficiently large
number). Then{τ(n)}n≥n is a nondecreasing sequence aslimn→∞τ(n) =∞, and for all n≥n, one has that
Sτ(n)≤Sτ(n)+ and Sn≤Sτ(n)+. (.)
That is,max{Sτ(n),Sn} ≤Sτ(n)+. For suchn≥n, it follows from (.) that
vτ(n)––v¯ ≤ vτ(n)+–v¯ (.)
and
vn––v¯ ≤ vτ(n)+–¯v. (.)
From (.) and (.), we obtain
Sτ(n)≤Sτ(n)+
≤( –aτ(n))( –fτ(n))Sτ(n)+ aτ(n)u–¯v,vτ(n)+–v¯
+ fτ(n)( –aτ(n))u–v¯,vτ(n)–v¯
–cτ(n) JδGτ(n)(I–δτ(n)B)vτ(n)–vτ(n)
–hτ(n) JγGτ(n)(I–γτ(n)B)vτ(n)––vτ(n)–
≤( –aτ(n))( –fτ(n))Sτ(n)+ aτ(n)u–¯v,vτ(n)+–v¯
+ fτ(n)( –aτ(n))u–v¯,vτ(n)–v¯ . (.)
Just as the argument of Case , we have
⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩
limn→∞vτ(n)–JδGτ(n)(I–δτ(n)B)vτ(n)= ,
limn→∞vτ(n)––JγGτ(n)(I–γτ(n)B)vτ(n)–= ,
lim supn→∞u–v¯,vτ(n)+–v¯ ≤, lim supn→∞u–v¯,vτ(n)–v¯ ≤, limn→∞vτ(n)+–vτ(n)= .
(.)
By (.) and (.), we have
Sτ(n)+≤( –aτ(n))( –fτ(n))Sτ(n)++ aτ(n)u–v¯,vτ(n)+–v¯
for alln≥n. This implies that
Sτ(n)+≤( –aτ(n))( –fτ(n))Sτ(n)++aτ(n)Kvτ(n)+–vτ(n)
+ fτ(n)( –aτ(n)) +aτ(n)
u–¯v,vτ(n)–¯v
for alln≥n, whereK= u–v¯. Furthermore, we have
Sτ(n)+≤
aτ(n)Kvτ(n)+–vτ(n)
aτ(n)+fτ(n)( –aτ(n))
+ u–v¯,vτ(n)–v¯
≤Kvτ(n)+–vτ(n)+ u–v¯,vτ(n)–v¯ . (.)
Hence, it follows from (.) and (.) that
lim
n→∞Sτ(n)+= . (.)
By (.) and (.), we know thatlimn→∞Sn= . Then, just as the argument in the proof of Case , we obtainlimn→∞vn=v¯. Therefore, the proof is completed.
Remark .
(i) If we putB=B= in Theorem ., then Theorem . is reduced to Theorem in
[].
(ii) Boikanyoet al.[] showed that strong convergence theorem of a proximal point algorithm with error can be obtained from strong convergence of a proximal point algorithm without errors. Therefore, in Theorem ., we study strong convergence of variational inclusion problems without error.
As a simple consequence of Theorem ., we have the following theorem.
Theorem . Suppose that(B+G)–()is nonempty,and{an},{bn},{cn}are sequences
in[, ]such that an+bn+cn= , <an< for each n∈N.For an arbitrary fixed u∈H,
define a sequence{xn}by
x∈His chosen arbitrarily,
xn+:=anu+bnxn+cnJδGn(I–δnB)xn, n∈N∪ {}.
Thenlimn→∞vn=P(B+G)–()u if the following conditions are satisfied:
(i) limn→∞an= ,∞n=an=∞;
(ii) <a≤δn≤b< κfor eachn∈Nand for somea,b∈R+;
(iii) lim infn→∞cn> .
Proof Setfn= ,B= ,gn+hn= ,{gn}and{hn}are sequences in [, ], andG=∂ιH.
Define a sequencevnby
and
vn=vn–=xn–.
Then
vn+=anu+bnvn+cnJδGn(I–δnB)vn, n∈N∪ {},
and
vn=fnu+gnvn–+hnvn–=fnu+gnvn–+hnJ
∂ιH
γn vn–, n∈N∪ {}.
Since
G– () = (∂ιH)– =FixJ∂ιH γn
=Fix(PH) =H,
it is easy to see that
(B+G)–() =
(B+G)–()∩H
=(B+G)–()∩(B+G)–()
=∅.
Then Theorem . follows from Theorem ..
Remark . Following the same argument as in Remark in [], we see that Theo-rems . and . contain [, Theorem .], [, Theorem ], [, TheoTheo-rems -] and many other recent results as special cases.
4 Applications
Now, we recall the following multiple sets split feasibility problem (MSSFP-A):
Findx¯∈Hsuch thatx¯∈G– ()∩G– (),Ax¯∈Fix(F) andAx¯∈Fix(F).
Theorem .[] Given anyx¯∈H,{λn}and{γn}are sequences in(,∞).
(i) Ifx¯is a solution of(MSSFP-A),then
JG
λn(I–ρnA ∗
(I–F)A)J G
rn (I–σnA∗(I–F)A)x¯=x¯for eachn∈N.
(ii) Suppose thatJG λn(I–ρnA
∗
(I–F)A)J G
rn (I–σnA∗(I–F)A)x¯=x¯with
<ρn<A+, <σn<A+ for eachn∈Nand the solution set of(MSSFP-A)is
nonempty.Thenx¯is a solution of(MSSFP-A).
In order to study the convergence theorems for the solution set of multiple split feasi-bility problem (MSSFP-A), we need the following problems and the following essential tool which is a special case of Theorem . in []:
(SFP-) Findx¯∈Hsuch thatx¯∈Fix
JG
ρn
andAx¯∈Fix(F).
Lemma . Given anyx¯∈H.
(i) Ifx¯is a solution of(SFP-),thenJG
ρn(I–ρnA ∗
(ii) Suppose thatJG ρn(I–ρnA
∗
(I–F)A)x¯=x¯with <ρn<A
+for eachn∈N,and
the solution set of(SFP-)is nonempty.Then(I–ρnA∗(I–F)A)and
JG
ρn(I–ρnA ∗
(I–F)A)are averaged andx¯is a solution of(SFP-).
Proof (i) Suppose thatx¯∈His a solution of (SFP-). Thenx¯∈Fix(JρGn),Ax¯∈Fix(F) for eachn∈N. It is easy to see that
JG ρn
I–ρnA∗(I–F)A
¯
x=x¯, n∈N.
(ii) Since the solution set of (SFP-) is nonempty, there exists w¯ ∈H such thatw¯ ∈ Fix(JG
ρn),Aw¯ ∈Fix(F). Thenw¯ ∈G –
(). If we putG=GandF=F, we get that the solution set of (MSSFP-A) is nonempty. By Lemma . we have that
A∗(I–F)Ais
A
-ism. (.)
By (.), <ρn<A
+, and Lemma .(ii), (iii), we know that
I–ρnA∗(I–F)Ais averaged for eachn∈N. (.) On the other hand, for eachn∈N,JG
ρn is a firmly nonexpansive mappings, it is easy to see that
JG ρn is
-averaged. (.)
Hence, by (.), (.) and Lemma .(iv) and (v), we see that
JG ρn
I–ρnA∗(I–F)A
is averaged.
Since
JG ρn
I–ρnA∗(I–F)A
¯ x=x¯,
so
JG ρn
I–ρnA∗(I–F)A
JG
ρn
I–ρnA∗(I–F)A
¯ x=x¯.
Then Lemma . follows from Theorem . by takingG=G,F=Fandρn=rn.
Remark . From the following result, we know that Lemma . is more useful than The-orem ..
Theorem . Suppose that the solution setAof(MSSFP-A)is nonempty and{an},{bn},
{cn},{fn},{gn},{hn}are sequences in[, ]such that an+bn+cn= ,fn+gn+hn= , <an< ,
and <fn< for each n∈N.For an arbitrarily fixed u∈H,a sequence{vn}is defined by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnJρGn(I–ρnA ∗
(I–F)A)vn, n∈N∪ {},
vn:=fnu+gnvn–+hnJσGn(I–σnA ∗
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a≤ρn≤b<A
+, <c≤σn≤d<
A+ for eachn∈N,and for some a,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof SinceFiis firmly nonexpansive, it follows from Lemma . thatA∗i(I–Fi)AiisAi
-ism for eachi= , . For eachi= , , putBi=A∗i(I–Fi)Aiin Lemma .. Then algorithm in Theorem . follows immediately from algorithm in Theorem .. SinceAis nonempty, by Lemma ., we have that
¯
w∈FixJG ρn
I–ρnA∗(I–F)A
∩FixJG σn
I–σnA∗(I–F)A
=∅ (.)
for eachn∈N. This implies that
¯
w∈FixJG
ρn(I–ρnB)
∩FixJG
σn(I–σnB)
=∅ (.)
for eachn∈N. Hence,
¯
w∈(B+G)–()∩(B+G)–()
=∅.
By Theorem .,limn→∞vn=x¯, wherex¯=P(B+G)–()∩(B+G)–()u. That is,
¯
x∈(B+G)–()∩(B+G)–()
and
¯x–u ≤ q–u
for allq∈(B+G)–()∩(B+G)–(). Since
¯
x∈(B+G)–()∩(B+G)–(),
we know that
¯
x∈FixJG
ρn(I–ρnB)
∩FixJG
σn(I–σnB)
.
That is,
¯
x∈FixJG ρn
I–ρnA∗(I–F)A
and
¯
x∈FixJG σn
I–σnA∗(I–F)A
.
By Lemma ., we get thatx¯∈A. Similarly, ifq∈(B+G)–()∩(B+G)–(), then
q∈A. Thereforex¯=PAu. This shows thatlimn→∞vnis a unique solution of the
opti-mization problem
min q∈A
q–u.
Therefore, the proof is completed.
In the following theorem, we study the following multiple sets split feasibility problem (MSSMVIP-A):
Findx¯∈Hsuch thatx¯∈C,Ax¯∈QandAx¯∈Q.
LetAdenote the solution set of (MSSMVIP-A). The following theorem is a special case of Theorem .. Hence, it is also a special case of Theorem ..
Theorem . Suppose thatAis nonempty,and that{an},{bn},{cn},{fn},{gn},{hn}are
sequences in[, ]with an+bn+cn= ,fn+gn+hn= , <an< ,and <fn< for each
n∈N.For an arbitrary fixed u∈H,a sequence{vn}is defined by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnPC(I–ρnA∗(I–PQ)A)vn, n∈N∪ {},
vn:=fnu+gnvn–+hnPC(I–σnA∗(I–PQ)A)vn–, n∈N.
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a≤ρn≤b<A+, <c≤σn≤d<A+ for eachn∈Nand for some a,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof PutG=G=∂ιC,F=PQ, andF=PQ. ThenG,Gare two set-valued maximum monotone mappings,F andFare firmly nonexpansive mappings. SinceJρ∂ιnC=PC and
J∂ιC
σn =PC, we haveFix(F) =Fix(PQ) =Q,Fix(F) =Fix(PQ) =Q
,Fix(J∂ιC
ρn ) =Fix(PC) =C andFix(J∂ιC
In the following theorem, we study the following split feasibility problem (MSSMVIP-A):
Findx¯∈Hsuch thatx¯∈C∩Q,Ax¯∈QwhereQis a nonempty closed
subset ofH.
LetAdenote the solution set of problem (MSSMVIP-A). The following is also a spe-cial case of Theorem ..
Theorem . Suppose that Qis a nonempty closed convex subset ofH,Ais nonempty,
and{an},{bn},{cn},{fn},{gn},{hn}are sequences in[, ]with an+bn+cn= ,fn+gn+hn= , <an< ,and <fn< for each n∈N.For an arbitrary fixed u∈H,a sequence{vn}is
defined by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnPC(I–ρnA∗(I–PQ)A)vn, n∈N∪ {},
vn:=fnu+gnvn–+hnPC(I–σn(I–PQ))vn–, n∈N.
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a≤ρn≤b<A+, <c<σn<d< for eachn∈Nand for some
a,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof PutA=IandH=H in Theorem .. Then Theorem . follows from
Theo-rem ..
In the following theorem, we study the following convex feasibility problem (MSSMVIP-A):
Findx¯∈Hsuch thatx¯∈C∩Q∩Q, whereQ,Qare nonempty closed
subsets ofH.
LetAdenote the solution set of (MSSMVIP-A). The following is a special case of Theorem ..
Theorem . Suppose that Q and Qare nonempty closed convex subsets ofH,Ais
nonempty,and{an},{bn},{cn},{fn},{gn},{hn}are sequences in[, ]with an+bn+cn= ,
a sequence{vn}is defined by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnPC(I–ρn(I–PQ))vn, n∈N∪ {},
vn:=fnu+gnvn–+hnPC(I–σn(I–PQ))vn–, n∈N.
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a<ρn<b<, <c<σn<d< for eachn∈Nand for somea,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof PutA=A=IandH=H=Hin Theorem .. Then Theorem . follows from
Theorem ..
In the following theorem, we study the following convex feasibility problem (MSSMVIP-A):
Findx¯∈Hsuch thatx¯∈Q∩Q, whereQandQare nonempty closed
convex subsets ofH.
LetAdenote the solution set of (MSSMVIP-A).
The following existent theorem of a convex feasibility problem follows immediately from Theorem ..
Theorem . Suppose that Q and Qare nonempty closed convex subsets ofH,Ais
nonempty,and{an},{bn},{cn},{fn},{gn},{hn}are sequences in[, ]with an+bn+cn= ,
fn+gn+hn= , <an< ,and <fn< for each n∈N.For an arbitrary fixed u∈H.Define
a sequence{vn}by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cn(I–ρn(I–PQ))vn, n∈N∪ {},
vn:=fnu+gnvn–+hn(I–σn(I–PQ))vn–, n∈N.
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
(iii) <a<ρn<b<, <c<σn<d< for eachn∈Nand for somea,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof PutC=H, thenPC=PH. Then Theorem . follows from Theorem ..
In the following theorem, we study the following system of convexly constrained linear inverse problem (SCCLIP):
Findx¯∈Hsuch thatx¯∈C,Ax¯=bandAx¯=b, whereb∈Handb∈H.
LetAdenote the solution set of (SCCLIP).
Theorem . Suppose thatAis nonempty,and b∈H,b∈H.Let{an},{bn},{cn},{fn},
{gn},and{hn}be sequences in[, ]with an+bn+cn= ,fn+gn+hn= , <an< ,and <fn< for each n∈N.For an arbitrary fixed u∈H,a sequence{vn}is defined by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnPC(vn–ρnA∗(Avn–b)), n∈N∪ {},
vn:=fnu+gnvn–+hnPC(vn––σnA∗(Avn––b)), n∈N.
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a≤ρn≤b<A
+, <c≤σn≤d<
A+ for eachn∈Nand for some a,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Proof PutQ={b}andQ={b}. Then Theorem . follows from Theorem ..
In the following theorem, we study the following convexly constrained linear inverse problem (CCLIP):
Findx¯∈Hsuch thatx¯∈C∩QandAx¯=b, whereb∈HandQis a nonempty
closed convex subset ofH.
LetAdenote the solution set of (CCLIP).
Theorem . Suppose that Qis a nonempty closed convex subset ofH.Ais nonempty,
b∈H,and{an},{bn},{cn}, {fn},{gn},{hn} are sequences in[, ]with an+bn+cn= ,
fn+gn+hn= , <an< ,and <fn< for each n∈N.For an arbitrary fixed u∈H,
a sequence{vn}is defined by
⎧ ⎪ ⎨ ⎪ ⎩
v∈His chosen arbitrarily,
vn+:=anu+bnvn+cnPC(vn–ρn(Avn–b)), n∈N∪ {},
Thenlimn→∞vnis a unique solution of the following optimization problem:
min q∈A
q–u
provided the following conditions are satisfied:
(i) limn→∞an=limn→∞fn= ;
(ii) either∞n=an=∞or
∞
n=fn=∞;
(iii) <a≤ρn≤b<A
+, <c<σn<d<
for eachn∈N,and for some
a,b,c,d∈R+;
(iv) lim infn→∞cn> ,lim infn→∞hn> .
Remark . The iteration in Theorem . is different from the Landweber iterative method []:
xn+:=xn+γAT(b–Axn), n∈N.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
L-JL designed and coordinated this research project and revised the paper. Y-DC carried out the project, drafted and revised the manuscript. C-SC coordinated the project and revised the paper.
Author details
1Department of Mathematics, National Changhua University of Education, Changhua, 50058, Taiwan.2Department of
Chemical and Materials Engineering, Southern Taiwan University of Science and Technology, Tainan, 71005, Taiwan.
3Department of Applied Mathematics, National Sun Yat-sen University, Kaohsiung, Taiwan.
Acknowledgements
Prof. C-S Chuang was supported by the National Science Council of Republic of China while he work on the publish, and Y-D Chen was supported by Southern Taiwan University of Science and Technology.
Received: 3 April 2013 Accepted: 7 November 2013 Published:06 Dec 2013
References
1. Combettes, PL: The convex feasible problem in image recovery. In: Hawkes, P (ed.) Advanced in Image and Electron Physics, vol. 95, pp. 155-270. Academic Press, New York (1996)
2. Stark, H: Image Recovery: Theory and Applications. Academic Press, New York (1987)
3. Bregman, LM: The relaxation method of finding the common point of convex sets and its application to the solution of problems in convex programming. USSR Comput. Math. Math. Phys.7, 200-217 (1967)
4. Hundal, H: An alternating projection that does not converge in norm. Nonlinear Anal.57, 35-61 (2004)
5. Boikanyo, OA, Morosanu, G: Strong convergence of the method of alternating resolvents. J. Nonlinear Convex Anal.
14, 221-229 (2013)
6. Censor, Y, Elfving, T: A multiprojection algorithm using Bregman projection in a product space. Numer. Algorithms8, 221-239 (1994)
7. Bauschke, HH, Borwein, JM: On projection algorithm for solving convex feasible problems. SIAM Rev.38, 376-426 (1996)
8. Byrne, C: Iterative oblique projection onto convex sets and the split feasibility problem. Inverse Probl.18, 441-453 (2002)
9. Byrne, C: A unified treatment of some iterative algorithms in signal processing and image reconstruction. Inverse Probl.20, 103-120 (2004)
10. Censor, Y, Bortfeld, T, Martin, B, Trofimov, A: A unified approach for inversion problems in intensity modulated radiation therapy. Phys. Med. Biol.51, 2353-2365 (2003)
11. Censor, Y, Elfving, T, Kopf, N, Bortfeld, T: The multiple-sets split feasibility problem and its applications for inverse problems. Inverse Probl.21, 2071-2084 (2005)
12. López, G, Martín-Márquez, V, Xu, HK: Iterative algorithms for the multiple-sets split feasibility problem. In: Censor, Y, Jiang, M, Wang, G (eds.) Biomedical Mathematics: Promising Directions in Imaging, Therapy Planning and Inverse Problems, pp. 243-279. Medical Physics Publishing, Madison (2010)
13. Lopez, G, Martinez-Marquez, V, Wang, F, Xu, HK: Solving the split feasibility problem without prior knowledge of matrixn. Inverse Probl.28, Article ID 085004 (2012)
15. Xu, HK: Iterative methods for the split feasibility problem in infinite-dimensional Hilbert spaces. Inverse Probl.26, Article ID 105018 (2010)
16. Xu, HK: Iterative methods for the split feasibility problem in infinite-dimensional Hilbert spaces. Inverse Probl.26, Article ID 105018 (2010)
17. Yang, Q: The relaxed CQ algorithm solving the split feasible problem. Inverse Probl.20, 1261-1266 (2004) 18. Eicke, B: Iteration methods for convexly constrained ill-posed problems in Hilbert space. Numer. Funct. Anal. Optim.
13, 413-429 (1992)
19. Landweber, L: An iteration formula for Fredholm integral equations of the first kind. Am. J. Math.73, 615-624 (1951) 20. Boikanyo, OA, Morosanu, G: A contraction proximal point algorithm with two monotone operators. Nonlinear Anal.
75, 5686-5692 (2012)
21. Kamimura, S, Takahashi, W: Approximating solutions of maximal monotone operators in Hilbert spaces. J. Approx. Theory106, 226-240 (2000)
22. Xu, HK: Iterative algorithm for nonlinear operators. J. Lond. Math. Soc.2, 240-256 (2002)
23. Yao, Y, Noor, MA: On convergence criteria of generalized proximal point algorithms. J. Comput. Appl. Math.217, 46-55 (2008)
24. Wang, F, Cui, H: On the contraction-proximal point algorithms with multi-parameters. J. Glob. Optim.54, 485-491 (2012)
25. Yu, ZT, Lin, LJ, Chuang, CS: A Unified Study of The Split Feasible Problems With Applications. J. Nonlinear Convex Anal. (to appear)
26. Takahashi, W: Nonlinear Functional Analysis-Fixed Point Theory and Its Applications. Yokohama Publishers, Yokohama (2000)
27. Marino, G, Xu, HK: Convergence of generalized proximal point algorithm. Commun. Pure Appl. Anal.3, 791-808 (2004)
28. Maingé, PE: Strong convergence of projected subgradient methods for nonsmooth and nonstrictly convex minimization. Set-Valued Anal.16, 899-912 (2008)
29. Combettes, PL: Solving monotone inclusions via compositions of nonexpansive averaged operators. Optimization
53, 475-504 (2004)
30. Goebel, A, Kirk, WA: Topics in Metric Fixed Point Theory. Cambridge University Press, Cambridge (1990) 31. Rockafellar, RT: Monotone operators and the proximal point algorithm. SIAM J. Control Optim.14, 877-898 (1976) 32. Boikanyo, OA, Morosanu, G: Inexact Halpern-type proximal point algorithms. J. Glob. Optim.57, 11-26 (2011) 33. Yu, ZT, Lin, LZ: Hierarchical problem with applications to mathematical programming with multiple sets split feasibility constraints. Fixed Point Theory Appl.2013, Article ID 283 (2013). doi:10.1186/1687-1812-2013-283
10.1186/1687-1812-2013-333