Miskolc Mathematical Notes HU e-ISSN 1787-2413 Vol. 19 (2018), No. 1, pp. 461–468 DOI: 10.18514/MMN.2018.2035
CHARACTERIZATIONS OF STRONG WELL-POSEDNESS FOR A CLASS OF MULTI-VALUED VARIATIONAL INEQUALITIES
M. OVEISIHA Received 03 June, 2016
Abstract. In this paper, by using the limiting subdifferential we consider the well-posedness for multi-valued variational inequalities and give some equivalence formulations for them. Moreover, we show that the strong well-posedness for a multi-valued variational inequality is equivalent to the existence and uniqueness of its solution.
2010Mathematics Subject Classification: 49K40; 49J52; 90C31
Keywords: limiting subdifferential, variational inequality, well-posedness, approximating se- quence
1. INTRODUCTION
The classical concept of well-posedness for a minimization problem, which has been known as the Tykhonov well-posedness, was introduced by Tykhonov [7] in 1966. A minimization problem is Tykhonov well-posed if it has a unique solution and every minimizing sequence of the problem converges to the unique solution. In the last decades, various concepts of well-posedness such as˛-well-posedness, Hadam- ard well-posedness, Levitin-Polyak well-posedness and well-posedness by perturb- ations have been presented and studied for optimization problems, see [3,4,10,12]
and references therein. The concept of well-posedness for hemivariational inequality was first introduced by Goeleven and Mentagui [2] to provide some conditions guar- anteing the existence and uniqueness of a solution for a hemivariational inequality.
Later, Xiao and Huang [10] considered a concept of well-posedness for a variational- hemivariational inequality and obtained the equivalence of well-posedness between the variational-hemivariational inequality and the corresponding inclusion problem.
Very recently, Xiao et al. [9] established two kinds of conditions under which the strong and weak well-posedness for the hemivariational inequality are equivalent to the existence and uniqueness of its solutions, respectively.
In this paper, by using the limiting subdifferential we extend the concept of well- posedness to a class of multi-valued variational inequality which include as a special
c 2018 Miskolc University Press
case the classical variational and hemivariational inequalities. Moreover, we estab- lish some equivalence results for them. The paper is organized as follows: Section 2 prepares briefly some preliminary notions and results used in sequel. In Section 3, we show that the strong well-posedness for a multi-valued variational inequality is equi- valent to the existence and uniqueness of its solution. Also, a metric characterization for the strong well-posedness of multi-valued variational inequality is obtained.
2. NOTATIONS AND PRELIMINARIES
LetX be a Banach space andXits topological dual space. The norm inX and Xwill be denoted byjj:jj:We denoteh:; :i; Œx; yandx; yŒthe dual pair betweenX andX, the line segment forx; y 2X, and the interior ofŒx; y, respectively. Now, we recall some concepts of subdifferentials that we need in the next section.
Definition 1([6]). LetX be a normed vector space,˝ be a nonempty subset of X,x2˝and"0. The set of"-normals to˝atxis
Nb".xI˝/WD fx2Xjlim sup
u!˝x
hx; u xi jju xjj "g: Assume thatx2˝, the limiting normal cone to˝atxis
N.xNI˝/WD lim sup
x! Nx;"#0
Nb".xI˝/:
LetJ WX ! NRbe finite atxN 2X; the limiting subdifferential ofJ atxN is defined as follows
@MJ.x/N WD fx2Xj.x; 1/2N..x; J.N x//N IepiJ /g:
Remark1 ([6]). The set-valued mappingx7!@MJ.x/has closed graph for locally Lipschitz functions.
Definition 2. LetT WX!2X be a set-valued mapping. T is said to be relaxed invariant monotone with respect to if there exists a constant ˛ such that for any x; y2X and anyu2T .x/; v2T .y/;one has
hv; .x; y/i C hu; .y; x/i ˛.jj.x; y/jj2C jj.y; x/jj2/:
Remark2. (1) WhenT WX !Xis a single-valued operator, we obtain the definition of a relaxed invariant monotone operator.
(2) If˛D0, then the Definition2reduces to the definition of an invariant mono- tone map.
Definition 3([11]). A mappingT WX !Xis said to be hemicontinuous if for any x1; x22X, the function t 7! hT .x1Ctx2/; x2ifrom Œ0; 1into 1;C1Œis continuous at0C.
ConditionC ([5]). LetWXX!X. Then, for anyx; y2X; 2Œ0; 1
.y; yC.x; y//D .x; y/; .x; yC.x; y//D.1 /.x; y/:
Remark3. By some computation, we can see that if ConditionC holds, then for anyx1; x22X and2Œ0; 1
.x1C.x2; x1/; x1/D.x2; x1/:
Now, suppose that J WX !R, WXX !X, AWX !X is a mapping and f 2Xis some given element. Consider the following multi-valued variational-like inequality associated with.A; f; J /:
M VLI.A; f; J /: FindxN2X such that for anyx2X, there exists2@MJ.x/;N that hAxN f C; .x;x/N i 0:
Definition 4. A sequencefxng X is said to be an approximating sequence for theM VLI.A; f; J /, if there existsfngwith n#0 such that for anyx2X there existsxn2@MJ.xn/, that
hAxn f Cxn; .x; xn/i njj.x; xn/jj:
Definition 5. The multi-valued variational-like inequalityM VLI.A; f; J /is said to be strongly well-posed if it has a unique solutionxN onX and for every approxim- ating sequencefxng,.x; xN n/converges strongly to0.
3. MAIN RESULTS
In this section, we establish some conditions under which the well-posedness for the multi-valued variational-like inequality is equivalent to the existence and unique- ness of its solution. Theorems in this section extend theorems in [8,9] from hemivari- ational inequalities with Clarke’s subdifferential which is convex to multi-valued variational-like inequalities with limiting subdifferential which is not necessarily con- vex.
Theorem 1. Assume that operator AWX !X is hemicontinuous and relaxed invariant monotone with constantcandJ is locally Lipschitz such that@MJ satisfies relaxed invariant monotonicity condition with constant ˛. Consider the following assertions:
(i) xN is a solution of theM VLI.A; f; J /.
(ii) xN is a solution of the following associated multi-valued variational-like in- equality:
AM VLI.A; f; J /: Find xN 2X such that for any x 2X there exists x2
@MJ.x/such thathAx f Cx; .x;x/N i 0:
IfcC˛0andis skew, then.i /).i i /. Ifsatisfies Condition C, then.i i /).i /.
Proof. .i /).i i /. Let xN 2X be a solution ofM VLI.A; f; J /. Hence, for any x2X there exists2@MJ.x/N such that
hAxN f C; .x;x/N i 0: (3.1)
By the relaxed invariant monotonicity of@MJ, for anyx2X andx2@MJ.x/, one has
hx; .x; x/N i C h; .x;x/N i ˛.jj.x;x/N jj2C jj.x; x/N jj2/: (3.2) It follows from relaxed invariant monotonicity of the operatorA, (3.1) and (3.2), that
hAx f Cx;.x;x/N i hAxCx; .x;x/N i hAxNC; .x;x/N i
D ŒhAx; .x; x/N i C hAx; .x;N x/N i C hx; .x; x/N i C h; .x;x/N i .cC˛/.jj.x;x/N jj2C jj.x; x/N jj2/;
which shows thatxN is a solution ofAM VLI.A; f; J /.
.i i /).i /. Conversely, letxN be a solution to theAM VLI.A; f; J /. Hence, for any x2X there existsx2@MJ.x/such that
hAx f Cx; .x;x/N i 0: (3.3) For any´2X andt2Œ0; 1, setx.t /D NxCt .´;x/N in inequality (3.3), we have
hA.xNCt .´;x//N f Cxt; .xNCt .´;x/;N x/N i 0;
thatxt 2@MJ.x.t //. It follows Condition C, that
hA.xNCt .´;x//N f Cxt; .´;x/N i 0:
SinceJ is locally Lipschitz, we deduce that@MJ is locally bounded (Corollary 1.81 in [6]). Hence, there exists a neighborhood ofxN and a constant` > 0such that for each´in this neighborhood andx2@MJ.´/, we havejjxjj `:Sincex.t /! Nx when t !0 fort to be sufficiently small jjxtjj `, without loss of generality we may assume thatxt !xin weak-topology. Now, hemicontinuity of the operator AonX implies that
hAxN f Cx; .´;x/N i 0:
By the arbitrariness of´2X, we conclude that xN is a solution ofM VLI.A; f; J /.
Proposition 1. LetCXbe nonempty, closed, convex and bounded,'WX ! Rbe proper, convex and lower semi-continuous andy2X be arbitrary. Assume that is continuous and affine with respect to the first argument and for eachx2X, there existsx.x/2Csuch that
hx.x/; .x; y/i '.y/ '.yC.x; y//:
Then, there existsy2Csuch that
hy; .x; y/i '.y/ '.yC.x; y//; 8x2X:
Proof. With some minor modification in the proof of Proposition 3.3 in [1], we
can deduce the proof.
Theorem 2. LetAWX !Xbe relaxed invariant monotone with constantcand J a l.s.c. function that its limiting subdifferential satisfies relaxed invariant mono- tonicity condition with constant˛. If˛Cc > 0andis continuous and affine with respect to the second argument and skew, thenM VLI.A; f; J /is strongly well-posed if and only if it has a unique solution onX.
Proof. The necessity is obvious. For the sufficiency, suppose thatM VLI.A; f; J / has a unique solutionx. SinceN xN is the unique solution ofM VLI.A; f; J /, for any x2X, there existsx2@MJ.x/N such that
hAxN f Cx; .x;x/N i 0: (3.4) Suppose thatfxngis an approximating sequence forM VLI.A; f; J /. It follows that there existsn#0such that for anyx2X, there existsn.x/2@MJ.xn/that
hAxn f Cn.x/; .x; xn/i njj.x; xn/jj:
Now, consider the nonempty, convex and bounded set cofAxn fCnjn2@MJ.xn/g. Hence, it follows from Proposition1 with '.x/Dnjjx xnjj that there exists n
which is independent onx, such that
hAxn f Cn; .x; xn/i njj.x; xn/jj; 8x2X:
By choosing n, we can set nDPm
iD1ini thatm2N, Pm
iD1i D1 and ni 2
@MJ.xn/. Hence,
m
X
iD1
ihAxn f Cni; .x; xn/i njj.x; xn/jj; 8x2X:
Now, setxD Nxin above inequality, yields
m
X
iD1
ihAxn f Cni; .x; xN n/i njj.x; xN n/jj:
Hence, it follows from relaxed invariant monotonicity of the operator A, relaxed invariant monotonicity of the@MJ, the skewness ofand above inequality that
njj.x; xN n/jj
m
X
iD1
ihAxn f Cni; .x; xN n/i
m
X
iD1
iŒhAxnCni; .x; xN n/i hAxNCni; .x; xN n/i
D
m
X
iD1
iŒ hAxN AxnCni ni; .x; xN n/i
m
X
iD1
2i.cC˛/jj.x; xN n/jj2D 2.cC˛/jj.x; xN n/jj2;
whereni 2@MJ.x/N is obtained from (3.4) by setting xDxn. Since cC˛ > 0, it follows that
jj.x; xN n/jj n
2.cC˛/:
Taking limit at both sides of the above inequality, implies that.x; xN n/ converges
strongly to0.
Example1. LetX DR; f D0,Abe the identity map andJ be defined as J.x/D
x2C2x ifx > 0;
x2 x ifx0:
The limiting subdifferential ofJ is
@MJ.x/D 8
<
:
2xC2 ifx > 0;
[-1,2] ifxD0;
2x 1 ifx < 0:
Letbe defined as.x; y/WDk.x y/, such that0 < k1. Then by some compu- tation we can see that@MJ; Aare relaxed invariant monotone with constants˛D1, cD12;respectively. Hence, all assumptions of Theorem2are fulfilled andxN D0is a unique solution of.M VLI /and therefore it is strongly well-posed.
For any > 0, consider the following two sets:
˝./D f NxW 8x2X; 9x2@MJ.x/N s.t. hAxN f Cx; .x;x/N i jj.x;x/N jjg; ./D f NxW 8x2X; 9x2@MJ.x/N s.t. h Ax; .x; x/N i C h f Cx; .x;x/N i
jj.x;x/N jjg: Lemma 1. Suppose thatAWX!Xis invariant monotone and hemicontinuous.
Then˝./D ./for all > 0.
Proof. Taking into account the invariant monotonicity of mappingA, it is easy to obtain that˝./ ./. For the other side suppose thatxN 2 ./. Then, for any x2X, there existsx2@MJ.x/N such that
h Ax; .x; x/N i C h f Cx; .x;x/N i jj.x;x/N jj: (3.5) SetxD NxCt .´;x/N in (3.5), that´2X andt2Œ0; 1, yields
h A.xNCt .´;x//; .N x;N xNCt .´;x//N i C h f Cx; .xNCt .´;x/;N x/N i jj.xNCt .´;x/;N x/N jj:
By using Condition C, we obtain
h A.xNCt .´;x//; .´;N x/N i C h f Cx; .´;x/N i jj.´;x/N jj: Now, it follows from the hemicontinuity of mappingAthat
hAxN f Cx; .´;x/N i jj.´;x/N jj;
which shows thatxN 2˝./. This completes the proof.
Lemma 2. Suppose thatAWX!Xis a hemicontinuous mapping. IfJ is locally Lipschitz and is continuous with respect to the second argument, then ˝./ is closed inX for all > 0.
Proof. Letfxng ˝./be a sequence such thatxn! NxinX. Then for anyx2X, there existsxn2@MJ.xn/such that
hAxn f Cxn; .x; xn/i jj.x; xn/jj: (3.6) Since J is locally Lipschitz, there exists a subsequence of xnthat convergent to a x2@MJ.x/N in weak-topology. Consider (3.6) with this subsequence, taking limit at both sides of it and using this fact that Ais hemicontinuous andis continuous with respect to the second argument, we obtain
hAxN f Cx; .x;x/N i jj.x;x/N jj;
which implies thatxN2˝./. This completes the proof.
Theorem 3. Suppose thatAWX!Xis hemicontinuous and invariant monotone with respect tothatis continuous with respect to the second argument, satisfies ConditionC and skew. ThenM VLI.A; f; J /is strongly well-posed if and only if
˝./¤¿;8 > 0 and diam.˝.//!0 as !0:
Proof. “Necessity” follows similarly from the first part of Theorem 3.1 in [8].
Hence, we prove the “Sufficiency”. Suppose that fxng X is an approximating sequence forM VLI.A; f; J /. Then there exist a nonnegative sequencen!0such that for anyx2X, there existsn2@MJ.xn/such that
hAxn f Cn; .x; xn/i njj.x; xn/jj; (3.7) it means thatxn2˝.n/. It follows from diam.˝.//!0, thatfxngis a cauchy sequence and so converges strongly to some pointxN2X. SinceJ is locally lipschitz, there exists a subsequence ofn.e.g. fnig/that is convergent to a2@MJ.x/N in weak-topology. Taking limit at the both side of (3.7) and using this fact thatAis monotone and is skew and continuous with respect to the second argument, we obtain
hAx f C; .x;x/N i D lim
i!1hAx f Cni; .x; xni/i lim
i!1hAxni f Cni; .x; xni/i lim
i!1 nijj.x; xni/jj D0:
Now, by using Theorem1,xN is a solution ofM VLI.A; f; J /. Since, diam.˝.//! 0when!0,M VLI.A; f; J /has a unique solution. This completes the proof.
REFERENCES
[1] F. Giannessi and A. Khan, “Regularization of non-coercive quasi variational inequalities,”Control Cyber., vol. 29, pp. 91 – 110, 2000.
[2] D. Goeleven and D. Mentagui, “Well-posed hemivariational inequalities,”Numer. Funct. Anal.
Optim., vol. 16, pp. 909 – 921, 1995.
[3] X. X. Huang, X. Q. Yang, and D. L. Zhu, “Levitin-polyak well-posedness of variational inequal- ity problems with functional constraints,”J. Global Optim., vol. 44, pp. 159 – 174, 2009, doi:
10.1007/s10898-008-9310-1.
[4] X. J. Long and N. J. Huang, “Metric characterizations of˛-well-posedness for symmetric quasi- equilibrium problems,”J. Global Optim., vol. 45, pp. 459 – 471, 2009, doi:10.1007/s10898-008- 9385-8.
[5] S. R. Mohan and S. K. Neogy, “On invex sets and preinvex functions,”J. Math. Anal. Appl., vol.
189, pp. 901 – 908, 1995, doi:10.1006/jmaa.1995.1057.
[6] B. S. Mordukhovich,Variational Analysis and Generalized Differential I, Basic theory, 1st ed., ser. Grundlehren. Berlin: Springer Berlin Heidelberg, 2006, vol. 330.
[7] A. N. Tikhonov, “On the stability of the functional optimization problem,”Comput. Math. Math.
Phys., vol. 6, pp. 28 – 33, 1966.
[8] Y. B. Xiao, N. J. Huang, and M. M. Wong, “Well-posedness of hemivariational inequalities and inclusion problems,”Taiwanese J. Math., vol. 15, pp. 1261 – 1276, 2011.
[9] Y. B. Xiao, X. Yang, and N. J. Huang, “Some equivalence results for well-posedness of hemivari- ational inequalities,”J. Global Optim., vol. 61, pp. 789 – 802, 2015, doi: 10.1007/s10898-014- 0198-7.
[10] Y. Xiao and N. Huang, “Well-posedness for a class of variational-hemivariational inequalities with perturbations,”J. Optim. Theory Appl., vol. 151, pp. 33 – 51, 2011, doi: 10.1007/s10957- 011-9872-9.
[11] E. Zeidler,Nonlinear Functional Analysis and its Applications. Berlin: Springer Berlin Heidel- berg, 1990, vol. II.
[12] J. Zeng, S. J. Li, W. Y. Zhang, and X. W. Xue, “Hadamard well-posedness for a set-valued optim- ization problem,”Optim. Lett., vol. 7, pp. 559 – 573, 2013, doi:10.1007/s11590-011-0439-3.
Author’s address
M. Oveisiha
Department of Mathematics, Faculty of Science, Imam Khomeini International University, P.O. Box 34149-16818, Qazvin, Iran
E-mail address:oveisiha@sci.ikiu.ac.ir