Regularization Methods to Approximate Solutions of Variational Inequalities ()
1. Introduction
In this paper, we study the regularization methods to approximate the solutions of the following variational inequalities
(1)
and
(2)
where A is a hemi-continuous monotone operator form real Banach reflexive X into its dual space
and D is a nonempty convex closed subset of X and
is the value of the linear functional
at
, i.e.
. We suppose that (1) or (2) has the exact solutions in D. For the variational inequality (2), the real function f defined on X is assumed to be lower semi-continuous, convex and is not identical to infinity.
These variational inequalities are used in the study of nonlinear problems in physics and engineering, especially in finding solutions to problems in non-linear partial differential equations [1] .
The problem to approximate solutions of variational inequalities (1) or (2) has been posed for a long time, and has been studied by many famous mathematicians in the world, including mathematicians Al’ber Ya. I., Abramov A., Ramm A. G., Riyazantseva P., Browder F. E., Liskovets O. A., who have contributed many foundational works [2] - [19] .
In the 1960s, Browder F. E. was the first to study the problem of finding approximate solutions of (1), (2) and to study the stability of the solutions of these problems by a method that he called the monotone operator method [9] .
After Tikhonov A. N. introduced the variational method, which was later called the Tikhonov regularization method to solve the ill-posed problems in works [20] [21] [22] [23] , Albert Ia. I. was the first to study the stability of the solution of variational inequality (1) according to the Tikhonov regularization method by adding the conjugate operator [4] [5] [6] . Actually, this idea of Browder F. E. has been used in works [9] but has not been prominent in the method.
Liskovevtz O. A. [15] [16] [17] [18] , Nguyen Van Kinh [24] [25] [26] [27] [28] , have achieved many results in this field, especially for variational inequality (1), where the noise operator of the exact operator is non-monotone. However, there are still many unresolved open issues. This paper will address some of these open problems. Detail, we shall study regularization methods to approximate solutions of variational inequalities (1) and (2) with operator A being monotone hemi-continuous form real Banach reflexive X into its dual space
, but instead of knowing the exact data
, we only know its approximate data
satisfying certain specified conditions and D is a nonempty convex closed subset of X.
The paper structure consists of 3 sections: Section 1 the introduction briefly summarizes the recent research results and comes up with problems that need to be studied; Section 2 presents regularization method for variational inequality (1) and Section 3 presents regularization for general variational inequality (2).
2. Regularization Method to Approximate Variational Inequality
First, we give the definition and some properties of topology on non-empty subsets of a topological space, some properties of the variational inequality with the hemi-continuous, monotone operator and dual mapping.
Suppose that X is a topology space. We denote
as the set of all non-empty subsets of X. We will define a topology on
as follows:
Definition 2.1 [29] Topology on the set
that has a basis topology, which is of the form
, where U is an arbitrary open set of X. This topology is called B-topology on
. If topology space X is topology defined by metric then B-topology is called β-topology.
The above topological definition can be found in [29] . The following are some characteristics of this topology.
Proposition 2.1 [29] If the sequence of the sets
converges to the set
, denoted by
and if the sequence of the sets
, with
and
then
.
Proposition 2.2 [29] If the sequence of the sets
B-converges to
,
, then arbitrary subsequence of this sequence also B-converges to the set M.
It is easy to see that the sequence of sets
B-converges to the set M, which means that for any open set U containing M, then for
“large enough” then
. If X is a metric space, then what has just been said is expressed by the following proposition:
Proposition 2.3 [29] Suppose that X is a metric space. Then we have
where
, which is called the deviation of the set
with respect to the set M.
Proposition 2.4 [29] Suppose that the sequence of sets
in Hausdorff topology space X, has the following property:
Taking arbitrarily
, we get a sequence of points
in X. Any subsequence of this sequence has one of the following properties:
1)
, where
is the set of limit points of the sequence
.
2)
, where M is a some subset of X.
Then,
.
Conversely, if M is a compact set and
, then both properties (1) and (2) are true if we replace the set M by the set
.
The operator
is called monotone if it satisfies the following condition
(3)
The operator A is called strictly monotone if (3) with an equals sign occurs if and only if
.
The operator A is called hemi-continuous at
if for any x such that:
A is called hemi-continuous on
if it is continuous at every point
.
An extension of Minty’s Lemma is that the following variational inequalities are equivalent:
Lemma 2.1 [29] If
is a monotone and hemi-continuous, then the following variational inequalities are equivalent, i.e., they have the same of the solution sets:
(4)
(5)
where
and D is convex closed subset of
.
Lemma 2.2 Let
be a non-negative real, continuous, and
be a hemi-continuous at
,
be a convex subset of X and
. If
(6)
then
(7)
Proof. Suppose that
satisfies (6). Let
, then
, with
, because
is convex and
.
By replacing x in the variational inequality (6) by m, we obtain
or
Therefore
(8)
Since
is a continuous function, then
as
and since B is hemi-continuous at z, then the left hand side of (8) converges to
as
. Therefore, we have
Lemma is proved.
The mapping
is called dual mapping of the norm space X if it satisfies the following conditions
This concept was first studied by Browder F. E. [8] and Vaiberg M. M. [30] .
The norm space X is called strictly convex if it has the following property
Next, we state some properties of duality mapping, which are used in the following sections.
Proposition 2.5 [30] The dual mapping
is single if dual space
is strictly convex.
The mapping
is called coercive if
where
is a non-negative real function and
as
.
Proposition 2.6 [30] The dual mapping
is coercive, monotone. Moreover, if X is a strictly convex space then U is strictly monotone.
Proposition 2.7 [30] If X is a reflexive real space and its dual space is strictly convex then the dual mapping
is hemi-continuous.
From the definition of the dual mapping, it follows that
Now we begin to present the regularization method to approximate solution of the variational inequality (1).
In the section, we assume that is a real reflective Banach space X with the conjugate space
being strictly convex. Then, the dual mapping
is single, coercive, monotone and hemi-continuous (Proposition 2.5, Proposition 2.6, Proposition 2.7).
As we already know the problem of finding the solution of the variational inequalities (1) is generally an ill-posed problem. Therefore, it is necessary to study some variational methods solving such problems to ensure that the solutions are stable, i.e. the solutions depend continuously on small changes to the exact data. We will present the regularization method with small parameter to approximate the variational inequality (1).
Suppose with exact data
, the solution set of the variational inequality (1) is
. Since A is a hemi-continuous monotone operator from real Banach reflexive X into its dual space
and D is a nonempty convex closed subset of X, then
is a convex, closed set [30] . Instead of knowing the exact data
, we only know its approximate data
satisfying the following conditions
(9)
and
(10)
where the operator
,
,
is an arbitrary operator, not necessarily monotone or hemi-continuous, and the real function
is non-negative, continuous and increasing no faster than some linear function, i.e., there exist
such that
(11)
In this work, we construct an approximate solution of (1), according to the regularization method, is the solutions of the variational inequality
(12)
where
is a small parameter.
Remark 2.1
1) If the operator
is non-monotone, non-hemicontinuous then in [17] , Liskovets O. A. studied the Tikhonov regularization solution of (1) which is the solution of (12) with
, i.e., the solution of the variational inequality:
(13)
2) If the operator
is non-monotone, non-hemicontinuous, then in [18] , Liskovets O. A. studied the approximate solution of (1) which is the solution of the variational inequality:
(14)
Since the operator
is arbitrary, the two sets of solutions for (12) and (14) are different, so the author’s and [18] ’s methods are different. The following example demonstrates it:
Let
be real normal space, therefore, its conjugate space also be
and its dual mapping U be identity mapping. Let
,
.
Suppose that the exact operator
, is defined by
Clearly the operator A is monotone and hemi-continuous on D.
The approximate operator
of the operator A is defined by
Therefore, the operator
is non-monotone, non-hemicontinuous on D and
Let
. Therefore, the variational inequality (12) has the form
Consequently,
is the solution of the variational inequality (12).
On the other hand, the following inequality is not true
So
is not the solution of the variational inequality (14).
Therefore, the solution sets of (12) and (14) are different.
In the following, we study the existence of an approximate solution and the stability of the solution of the regualaiation method (12).
We denote the solution set of (12) as
, where
. The following theorem asserts the existence of a solution of (12), that is, the existence of the Tikhonov regularization operator.
Theorem 2.1 If
, then
.
Proof. We consider the following variational inequality
(15)
By assumption, the operators A and U are single, monotone, hemicontinuous, so
is also single, monotone, hemicontinuous.
+ If D is non-bounded, then the operator
is coercive on D. Indeed, we have
Therefore, if
then
So, the operator is coercive on D.
Since
is monotone, hemi-continuous, coercive, according to [8] [31] the variational inequality (15) has a solution in D.
+ If D is bounded, according to [32] , the variational inequality (15) has a solution in D.
We will show that the solutions of (15) are all solutions of (12). Indeed, taking
as an arbitrary solution of (15), we have
Since
is the solution (15) then
Consequently,
So,
.
Theorem is proved.
Theorem 2.2 If
and
,
, where
are constants,
is the set of the solution of (12) then the sequence of the sets
is uniformly bounded in X.
Proof. By assumption
and by Theorem 2.1. then
. Let
, we have
or
Consequently
(16)
Since the operator is monotone, hemi-continuous and Lemma 2.2, from (16) it follows that
or
So
(17)
In (17) we take
, with
is the solution of (1), we have
Consequently
(18)
Since
, then from (18) it follows that
(19)
By assumption
,
, and (19) is a quadratic inequality in terms of
, then the set of
in (19) is bounded. Since
are constants which do not depend on
, then the sequence of the sets
is uniformly bounded, i.e., there exists a constant k which do not depend on
such that
Theorem is proved.
Remark 2.2
1) If
is uniformly bounded then the sequence of sets
, with
being the set of the solutions of (13), is uniformly bounded [18] .
2) The conditions for the sequence of sets
, with
being the set of the solutions of (14), is uniformly bounded , which is the same as theorem above [17] .
3) Instead of regularization for the solution of the variational inequality (1), we regulate for the solution of the equation
, where A is given exactly and
is an approximation of
satisfying (9), in [3] [4] the condition is uniformly bounded the sequence of solution sets of (13) with
replaced by A, is
, with k constant.
The following theorems talk about the stability of the solution of the variational inequality (1) according to the Tikhonov regularization method.
Theorem 2.3 With the same assumptions as in Theorem 2.2, the sequence of the sets
, B-weakly converges to exactly solution set
of the variational inequality (1) as
.
Proof. Let arbitrary
, from Theorem 2.2 it follows that the sequence
is uniformly bounded. Since X is a reflexive space and
, there exists a subsequence
of the sequence such that it weakly converges to
(since
and D is weakly closed). We shall denote
.
Since
is the solution of (12), we have
(20)
The argument is the same as in the proof of Theorem 2.2, using Lemma 2.2, from (20) it follows that
(21)
Since the operator
is monotone, from (21) it follows that
(22)
Since the function
is continuous and the sequence
is bounded, the sequence
is bounded. Therefore, in (22) taking
, therefore,
, we have
Consequently,
.
From this, according to Proposition 2.4, it follows that the sequence of the sets
B-weakly converges to exactly solution set
of the variational inequality (1) as
.
Theorem is proved.
We denote
Since X is a reflexive real Banach space and
is a closed convex subset in X, then
.
Theorem 2.4 The sequence of the sets
B-weakly converges to
as
,
. Moreover, let arbitrary
and
then
Proof. + We first prove
as
,
.
Since
,
, then we take
is small enough such that
(23)
Therefore, let arbitrary
, then the sequence
is uniformly bounded (Theorem 2.2). Since X is a reflexive space and
, there exists a subsequence
of the sequence such that
(since
and D is weakly closed) as
,
. The argument is the same as in the proof of Theorem 2.3, we have
.
Since
is a solution of (12), we have
From the above inequalities, it follows that
From the above inequality, according to Lemma 2.2, it follows that
(24)
Since the operator
is monotone, from (24) it follows that
or
(25)
In (25), taking
, we have
Hence, from (25) it follows that
(26)
Since sequence
is bounded and
is continuous, in (26) taking
,
, we obtain
(27)
From (27), according to Lemma of Minty, it follows that
(28)
Due to the property of the dual mapping U, from (28) it follows that
Hence
Consequently,
Therefore, the sequence of sets
B- weakly converges to
as
,
.
+ Next, we prove
, as
,
.
Since the operator A is monotone, from (24) it follows that
The argument is the same as in the above proof and from the above inequality, we have
or
(29)
In (29), replacing x by
, we have
(30)
According to the property of the dual mapping U, from (30) it follows that
(31)
In (31), taking
,
, we obtain
The above happens for every subsequence
of the sequence
, so the sequence itself also converges to
as
,
.
Theorem is proved.
Corollary 2.1 If X is a E-space [33] , then
B-converges to
as
,
.
Proof. It follows from Theorem 2.4.
Corollary 2.2 Suppose that X is a E-space and the set
has only one element (this happens, for example, A is a stricly monotone operator and X is a strictly convex space). Therefore, if
arbitrarily in
, then
, as
,
.
Proof. It follows from Theorem 2.4.
If we don’t know if (1) has a solution before, then the following theorem tells us that.
Theorem 2.5 The necessary and sufficient condition for the variational inequality (1) to have a solution is that the sequence of sets
B-weakly converges to some unique bounded subset of X as,.
Proof. Suppose that the sequence of sets B-weakly converges to bounded subset as,. Since X is a reflexive Banach space and by Proposition 2.1, without loss of generality, we can assume that Z is a weakly compact set.
Let arbitrary, then there exists the subsequence of the sequence such that as, (by Proposition 2.4). The argument is the same as in the proof of Theorem 2.3, it follows that satisfies
Hence, is a solution of (1).
Conversely, if, then from Theorem 2.4, it follows that the sequence of sets B- weakly converges to bounded subset as.
Theorem is proved.
We now present an example where the above regularization method can be applied. This example is slightly modified in [18] [30] .
As usual, the notation, , is the normed space of real functions p-integrable in the Lebesgue sense on the interval [0,1]. As we all know, is a reflecxive Banach space, and its conjugate space is, with, is strictly convex.
Let be a non-negative real function, continuous on the variable on the unit square and is also a definite and continuous real function on the unit square, without decreasing according to the variable t and satisfies the condition
where is a continuous, non-negative real function on, and b is a positive constant.
The operator A is defined as:
(32)
According to [30] , this operator acts from to, with the domain
Operator A is monotone. Indeed, let, we have
Since the function is not negative and is not decrease respect to the variable t, the right hand side of the above inequality is non-negative. So,
According to [30] the operator hemi-continuous.
The perturbed operator of A is defined as follows
(33)
where is a measurable real function on the unit square as follows
(34)
and is also a measurable real function on the unit square as follows
(35)
with is a non negative real function in.
According to [30] , from (34), (35) it follows that acts from to, with the domain.
Clearly, in general, is not monotone and not continuous. Moreover, after calculating and estimating, we have
(36)
where
Since, then.
By using the inequality
we have an estimate of the function g as follows:
where, and.
We have g is continuous and
Therefore, we can apply the above regularization method to find the approximate solution of the variational inequality (1) with the operator A is given by (32) and its perturbed operator is given by (33).
3. Regularizer of General Variational Inequality
3.1. Regularization Method Solving the General Variational Inequality-Strong Stability
In this subsection, we consider X to be a real reflexive space and satisfies the following conditions:
1) There exists a mapping U from X to the conjugate space, which maps the bounded subsets of X to the bounded subset of;
2) For every fixed that belongs to X, then, as (this condition is often called U as satisfying the coercive condition);
3) For every bounded sequence and such that:
Remark 3.1 If X is a uniformly convex space and the conjugate space is a strictly convex space, then the dual mapping satisfies the above conditions [30] .
We consider the following variational inequality
(37)
where A is a monotone, hemi-continuous, acting from X into, with domain, , , is lower semi-continuous and D is a convex closed subset of.
As we know the solution of (37) is which satisfies (37). The class of problems of the form (37) is often called the general variational inequalities. This class of problems contains special cases: the first kind operator equations, where A is a monotonous operator; variational inequality problems on convex sets with monotone operators on convex sets.
If then we have studied in the Section 2.
The existence of solutions of (37) has been studied by Browder F. E. and many other mathematicians in the world, some typical works can be mentioned as [10] [11] [34] .
For the first time, the author studies the approximate solution of the variational inequality (37) by the Tikhonov regularization method [24] [25] [26] [27] [28] , as follows:
Assume that for exact data the exact solution set of (37) is. In fact, instead of knowing this exact data, we only know its approximate data that satisfies the conditions
(38)
and
(39)
where, is not necessarily a monotone operator or hemi-continuous.
The author has studied the approximate solution of (37) with noisy data according to Tikhonov regularization method with small parameter, which is the solutions of the following variational inequality:
(40)
The solutions of (40) are the elements that satisfy (40). Let’s also denote the solution set of (40) is, with.
The following lemma is necessary for proving the existence of a solution of (5.2).
Lemma 3.1 [12] Suppose T is a operator with defined on the convex subset of X into, hemi-continuous on and f is a below semi-continuous function from X into with f is not identical to infinity. If z is a solution of the variational inequality
then it is also a solution of the variational inequality
Theorem 3.1 If, then.
Proof. We consider the following variational inequality
(41)
We will show that the variational inequality (41) has solutions and that these solutions are also solutions of the variational inequality (37).
From and the property (2) of the operator U it follows that there exists some non-negative real with, as such that
(42)
Since f is a below semi-continuous, convex function, there exist a constant such that
(43)
Indeed, suppose that (43) is not true, therefore there exists some sequence, with and, as, such that
Consequently
or
(44)
From f is a convex function and (44), it follows that
(45)
The inequality (45) is contract with the lower semi-continuous property of f which is mapping any bounded subset of X to bounded set.
Using the monotony of the operator A, we have
(46)
From (41), (43) and (46), it follows that
(47)
From (47), it follows that
(48)
+ If D is not bounded, from the monotone, hemi-continuous of the operator, (48), according to [31] it follows that the variational inequality (41) has the solutions belong to D.
+ If D is bounded, according to [32] , it follows that the variational inequality (41) has the solutions belong to D.
Let z be arbitrary solution of (41), we have
(49)
So, z is a solution of the variational inequality (40), i.e.,.
In case D is an unbounded set, then the following theorem states that the sequence of sets is uniformly bounded.
Theorem 3.2 If and, k is a constant, then the sequence of the sets is uniformly bounded.
Proof. Since, then (Theorem 3.1). Let, then we have
Consequently,
(50)
In inequality (50) we take then
Therefore, in inequality (50) we replace x by, we get
Since, from the above inequality implies
(51)
Since k is not dependent on and since the mapping U turns bounded set into bounded set, the sequence is uniformly bounded. From this, deduce the sequence of sets is also uniformly bounded.
Theorems 3.3 and 3.4 below talk about the stability of the solution of inequality (41) by the Tikhonov regularization method.
Theorem 3.3 With the assumptions given in Theorem 3.2, then the sequence of sets B-weakly converges to the set, as.
Proof. Let, then is uniformly bounded (Theorem 3.2). Since X is a reflexive Banach space, there exists a subsequence of this sequence such that, as. We prove that.
The argument is the same as in the proof of Theorem 3.2, we get the inequality (50) and in this inequality we replace by, we get
(52)
Due to the lower semi-continuity of f and, in (52) given, we get
Since A is hemi-continuous, according to Lemma 3.1, from the above variational inequality, it implies that
So, , that means B-weakly converges to, as.
We enter the following notation
Due to the coercive property of the mapping U on the closed convex set, the set. The following theorem gives a stronger result than Theorem 3.3.
Theorem 3.4 If, then B-converges to, as,.
Proof. Since, , then is small enough, then the sequence of sets is unformly bounded (Theorem 3.2). Let, then is bounded. Since X is a reflexive Banach space, there exists a subsequence of the sequence such that, as,.
The same argument as in the proof of Theorem 3.3, we obtain.
From (52) it follows that
(53)
In (53) given, , we get
According to Minty’s Lemma [30] , from the above variational inequality, we get
So,.
In inequality (53) we replace x by, we get
(54)
We have
(55)
Due to property (3) of the mapping U, from (55) it follows that
This means the sequence B- converges strongly to, as,.
Theorem is proved.
Corollary 3.1 With the same assumptions as in Theorem 3.4, the sequence of sets B-converges strongly to, as,.
Proof. It is easy to deduce from Theorem 3.4 and Proposition 2.1
3.2. Regularization Method Solving the General Variational Inequality with Small Parameter-Weakly Stability
In this subsection, we consider X as a real reflexive Banach space with its conjugate space being strictly convex. Therefore, dual mapping is single, monotone and hemi-continuous [34] .
For the first time, the author studies the approximate solution of the variational inequality (37) by Tikhonov regularization method with small parameter [24] [25] [26] [27] [28] as follows:
Assume that for exact data the exact solution set of (37) is. In fact, instead of knowing this exact data, we only know its approximate data that satisfies the conditions (38) and (39) and, is not necessarily a monotone operator or hemi-continuous.
The author has studied the approximate solution of (37) with noisy data according to Tikhonov regularization method Tikhonov with small parameter, which is the solutions of the following variational inequality:
(56)
The solutions of (56) are the elements that satisfy (56). Let’s also denote the solution set of (56) is, with.
Theorem 3.5. If then the variational inequality (56) has a solution.
Proof. We consider the following variational inequality
(57)
The same argument as in the proof of Theorem 3.1, the variational inequality (57) has a solution in D. Let z be a solution of (57). Due to the monotony of, from (57) it follows that
(58)
We will prove that every solution of (58) is a solution of (57). Indeed, let z be a solution of (58), we have
So,.
Theorem is proved.
In the case that the set D is not bounded, then the following theorem states that the uniformly boundedness of the sequence of sets.
Theorem 3.6 If, , (: constant), then the sequence of the sets of (55) is uniformly bounded in X.
Proof. Since, then (Theorem 3.5). Let, we have
From the variational inequality above, it follows that
(59)
Due to Lemma 2.2, Theorem 3.1, from (59) it follows that
(60)
Due to the monotony of A, from (60) it follows that
(61)
From (58), (59) it follows that
(62)
Due to, from (60) it follows that
(63)
Since, , from (63) it follows that the sequence is uniformly bounded.
Theorem is proved.
The following theorem states B-weakly stability of the solution of the variational inequality (37) according to the Tikhonov regularization method with small parameters.
Theorem 3.7 With the same assumptions as in Theorem 3.6, then the sequence B-weakly converges to, as.
Proof. Since, , , then the sequence is uniformly bounded. Let, then the sequence is bounded in the reflexive Banach space X. Therefore, there exists a subsequence of this sequence such that (since D is the weakly closed set). Since is a solution of (56), we have
Consequently,
Due to Lemma 2.2 and Lemma 3.1, the monotony of A and U, from the above inequality, it follows that
(64)
In (64), taking, we obtain
Due to Lemma 3.1, from the above inequality, it follows that
So,. It means that the sequence B-weakly converges to, as.
Theorem is proved.
We denote, then (since is the convex closed subset in reflexive Banach space X.
The following theorem gives us a stronger result than the above theorem.
Theorem 3.8 If, then the sequence B-strongly converges to, as,. Moreover, if let, , then, as,.
Proof. Since, , then we consider such that
(65)
With satisfying (65), then the sequence is uniformly bounded (Theorem 3.6). Let, then sequence is bounded in the reflexive Banach space X. Therefore, there exists a subsequence of this sequence such that as,. The same argument as in the proof of Theorem 3.7, we get the variational inequality same (64), that is
In this variational inequality, we take, then
Consequently,
(66)
or
Since the sequence is bouned and the function is continuous, then the sequence is bounded. Therefore, in (66) taking, , we obtain
Due to Lemma of Minty, from the above variational inequality, it follows that
Due to the property of dual mapping U, from the above variational inequality, it follows that
So, , that is, the sequence B-weakly converges to, as,.
On the other hand, we have
(67)
From (67) it follows that
In (67) taking, , we obtain
This proves that the sequence has only one limit point, so the sequence itself converges to, as,.
Corollary 3.2 Necessary and sufficient condition for the sequence B-weakly converges to some bounded subset of X as, the variational inequality (40) has a solution.
Proof. Similar to the proof of theorem 2.5.
Remark 3.2: If is monotone operator and, the above results are the same with results in [18] .
We now study the approximate solution of the variational inequality (37) according to Tikhonov regularization method with small parameters, which is the solutions of the following variational inequality:
(68)
where the same assumption of the first of this subset.
The solution of the variational inequality (68) is the elements that satisfy (68). We still denote the solution set of (68) as the set.
Theorem 3.9 If then the variational inequality (68) has a solution.
Proof. The same argument as in the proof of Theorem 3.5, we have the following inequality which has a solution:
(69)
It is easy to show that every solution of (69) is a solution of (68). So,.
Theorem 3.10 If, , (: constant), then the sequence of the sets of (56) is uniformly bounded in X.
Proof. Since, then (Theorem 3.9). Let, we have
Consequently,
Since A is the monotone operator, from the above inequality it follows that
(70)
We have
Consequently,
(71)
Since, from (71) it follows that
(72)
Since, from (72) the sequence is uniformly bounded.
Theorem is proved.
The following theorem talks about B-weakly stability of the solution of the variational inequality (37) by the Tikhonov regularization method with small parameter (68).
Theorem 3.11 With the assumption as in Theorem 3.10. Then, the sequence B-weakly converges to, as.
Proof. From the assumption of Theorem 3.10, then. Let, then the sequence is bounded in the reflexive Banach X, therefore, there exists a subsequence of this sequence such that, as. Since is a solution of (65), we have
Consequently,
(73)
Since the operators are monotone, from (73) it follows that
(74)
Since the function is continuous and lower semi continuous and the sequence is bounded, then in (74) taking, we obtain
So, , that is, the sequence B-weakly converges to, as.
Theorem is proved.
Theorem 3.12 If, , , then the sequence B-weakly converges to. Moreover, let and, then, as,.
Proof. Since, , we consider being small such that
(75)
With satisfying (75), then and is uniformly bounded. Let, then the sequence is bounded in X is a reflexive Banach space, there exists a subsequence of this sequence such that, as,.
The same argument as in the proof of Theorem 3.10, we obtain
(76)
In (76), taking, , due to the continulty of the function and the boundedness of the sequence, we obtain
(77)
Due to Lemma of Minty, from (74) it follows that
Consequently,
So,.
On the other hand,
(78)
In (78) taking, , we obtain
The above argument shows that the sequence has only one limit point, so it is this sequence that converges to, as,.
Theorem is proved.
Corollary 3.2 Necessary and sufficient condition for the sequence B-weakly converges to some bounded subset of X as, the variational inequality (37) has a solution.
Proof. Similar to the proof of Theorem 2.5.
Remark 3.3: We can use Tikhonov regularization methods in Subsections 2.2, 2.3 to find the approximate solution in the given example above.