- Research
- Open access
- Published:
Approximate solutions for set optimization with an order cone that has nonempty quasirelative interiors
Journal of Inequalities and Applications volume 2022, Article number: 101 (2022)
Abstract
In a real normed linear space, when the quasirelative interior is not empty, a class of order relation is introduced with Minkowski difference. Two classes of nonlinear functions are introduced, and their properties are discussed. A class of approximately efficient solutions and approximate weakly efficient solutions are introduced for set optimization. With nonlinear functions, optimality conditions are established for approximate solutions. Some examples are given to illustrate our main results.
1 Introduction
Set-valued optimization is the promotion of vector optimization that has applications in many aspects, such as engineering, cybernetics, military, finance, etc. The research on set-valued optimization has achieved fruitful results (see [6, 9, 13, 15]). The study of set optimization has been investigated by some authors (see [11, 12, 14, 21, 23]). Kuroiwa [14] considered the criteria of solutions of set optimization, showed some examples with respect to the criteria, introduced some type of semicontinuities for set-valued maps, and showed existence theorems of solutions. Karaman and Soyertem [11] used the Minkowski difference to define new order relations on the set family, and discussed the relations among these orders.
In the research of vector-optimization problems, different kinds of solutions defined by the ordering cone of the image space play an important role, mainly including efficient solutions, weakly efficient solutions, various properly efficient solutions, and corresponding approximate solutions. Among them, the weakly efficient solution defined by the nonempty interior based on an ordered cone has good properties. However, in many cases, the interior of the order cone is an empty set. For example, for any \(1< p<+\infty \), the normed space \(l^{p}\), partially ordered by the positive cone, is an important space in applications, however, the positive cone has an empty interior. Therefore, it is a very worthwhile topic to use the generalized interior such as the relative topological interior, quasiinterior, and quasirelative interior to define the corresponding weakly efficient solution for the vector-optimization problem. In this research, there have been many results, such as Adan and Novo [1] who used a relative topological interior to propose corresponding weakly efficient solutions, etc. Therefore, when the interior of the topology is an empty set, using a generalized interior to propose different approximate solutions and study some properties is also a very meaningful direction.
As is known, the efficient solution set may be empty, but approximate solutions always exist under weak assumptions. Therefore, it is very meaningful to introduce the concept of approximate solutions (see [5, 16, 22, 26]). Loridan [17] presented some properties of ϵ-solutions for vector-minimization problems where the function to be optimized takes its values in the Euclidean space . Qiu and Yang [19] studied the approximate solutions for the vector-optimization problem with set-valued functions, and discussed the relationships between approximate solutions and weak efficient solutions. Based on the set order relation introduced by Karaman et al. [2], Gupta et al. [7] defined a new concept of approximate weakly minimal solution for constrained set-optimization problems. Jahn and Ha [10] studied a new set-optimization problem, which is to minimize a set-valued map that takes a value in a real linear space; this set-valued map also has a preorder induced by a convex cone, and used the Minkowski difference to define a new order, and some of their properties were obtained. Zhao et al. [24] established a new nonlinear separation theorem to study the vector-optimization problem in which the topological interior or even the relative topological interior of the ordered cone may be empty.
Optimality conditions are important parts of vector-optimization problems and important foundations for establishing modern optimization algorithms [18, 20, 25]. Tung et al. [20] considered the set-optimization problem with mixed constraints, and investigated necessary and sufficient Karush–Kuhn–Tucker optimality conditions for strict minimal solutions. Zhao et al. [25] proposed a projected subgradient method for solving constrained nondifferentiable, quasiconvex, multiobjective optimization problems, and presented numerical results to illustrate their findings.
When the interior of the ordered cone is empty, but the quasirelative interior is nonempty, how do we introduce approximate solutions of set optimization? How do we establish the optimality conditions?
The paper is organized as follows. Section 2 gives some preliminaries. In Sect. 3, we give definitions of two kinds of nonlinear functions and investigate their properties. In Sect. 4, we give several kinds of approximate solutions for set optimization and study their properties. In Sect. 5, we establish optimality conditions for approximate solutions. Finally, Sect. 6 draws some conclusions from the paper.
2 Preliminaries
Let X be a linear space, Y be a real normed linear space, and K be a proper pointed convex cone in Y. Denote by \(\mathcal{P} ( Y )\) and \(\mathcal{B} ( Y )\) the families of nonempty subsets and nonempty bounded subsets of Y, respectively. Let A be a nonempty subset of Y, intA and clA denote the interior and closure of A, respectively. The generated cone of A is defined as
Suppose that A is a nonempty convex subset of Y, the quasirelative interior [24] of A is defined as
Remark 2.1
Let \(A \subset Y\) be a convex subset with a nonempty interior, then \(\mathrm{int}A=\mathrm{qri}A\).
Definition 2.1
([11])
Let \(A,B\in \mathcal{P} ( Y )\). The Minkowski difference of A and B is defined as
Lemma 2.1
([11])
Let \(A,B\in \mathcal{P} ( Y )\) and \(c\in Y\). Then,
-
(i)
\((c+A)\,\dot{-}\,B=c+(A\,\dot{-}\,B)\);
-
(ii)
\(A\,\dot{-}\,(c+B)=-c+(A\,\dot{-}\,B)\).
Lemma 2.2
([11])
If \(A\in \mathcal{B} ( Y )\), then \(A\,\dot{-}\,A=\{0_{Y}\}\).
Lemma 2.3
([4])
Let C be a nonempty convex subset of Y and . Then,
-
(i)
\(\mathrm{qri} ( \alpha C ) =\alpha \mathrm{qri}C\);
-
(ii)
\(t\mathrm{qri}C+ ( 1-t ) C\subset \mathrm{qri}C\), \(\forall t\in (0,1]\).
Remark 2.2
From Lemma 2.3, it follows that if C is a nonempty convex cone of Y, then \(C+\mathrm{qri}C\subset \mathrm{qri}C\).
In the rest of the paper, we assume that K and S are nonempty pointed convex cones of Y, \(\mathrm{int}K \ne \emptyset \) and \(\mathrm{qri}S \ne \emptyset \), respectively.
Definition 2.2
([7])
Let \(A,B\in \mathcal{P} ( Y )\). If \(( A \,\dot{-}\, B ) \cap ( -S ) \ne \emptyset \), then we note that A is less than or equal to B with respect to S, denoted by \(A\preceq _{S}^{m}B\).
Remark 2.3
The order relation that \(A\preceq _{S}^{m}B\) in [7] is equivalent to that \(A\preceq _{S}^{m_{2}}B\) in [11].
In the following, by using qriS, we introduce a new type of relationship.
Definition 2.3
Let \(A,B\in \mathcal{P} ( Y )\). If \(( A \,\dot{-}\, B ) \cap ( - \mathrm{qri}S ) \ne \emptyset \), then we note that A is less than B with respect to qriS, denoted by \(A\prec _{\mathrm{qri}S}^{m}B\).
Definition 2.4
([8])
Let \(A,B\in \mathcal{P} (Y )\). If \((\bigcap_{a\in A}{ ( a+K )} ) \cap B\ne \emptyset \), then we note that A is less than or equal to B with respect to K under the h relation, denoted by \(A\preceq _{K}^{h}B\).
Definition 2.5
([8])
Let \(A,B\in \mathcal{P} (Y )\). If \(A=B\), then \(A\preceq _{K}^{p}B\); if \(A\ne B \), then \(A\preceq _{K}^{p}B\Longleftrightarrow A\preceq _{K}^{h}B\), then we note that A is less than or equal to B with respect to K under the p relation.
Definition 2.6
([8])
Let \(A,B\in \mathcal{P} (Y )\). If \(( \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} ) \cap B\ne \emptyset \), then we note that A is weakly less than B with respect to K under the p relation, denoted by \(A\ll _{K}^{p}B\).
Proposition 2.1
Let \(A,B\in \mathcal{P} (Y )\). If \(A=B\), then \(A\not \ll _{K}^{p}B\).
Proof
Suppose to the contrary that \(A\ll _{K}^{p}B\), then \(( \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} ) \cap A\ne \emptyset \). There exists \(a_{1}\in A\) such that \(a_{1}\in \bigcap_{a\in A}{ ( a+K ) + \mathrm{int}K}\), therefore \(a_{1}\in a_{1}+K+\mathrm{int}K\subset a_{1}+\mathrm{int}K\). Then, we obtain \(0\in \mathrm{intK}\), which leads to a contradiction. Therefore, \(A\not \ll _{K}^{p}B\). □
Proposition 2.2
If \(A\preceq _{K}^{h}B\), then \(A\preceq _{K}^{p}B\).
Proof
It follows from the definition that the conclusion is true. □
Remark 2.4
The converse of Proposition 2.2 does not hold, as is shown in the following examples.
Example 1
Let \(A=B= \{ ( x,y ) | x^{2}+y^{2}\leqslant 1 \}\) and . It is clear that \(A\preceq _{K}^{p}B\). On the other hand, \(( \bigcap_{a\in A}{ ( a+K )} ) \cap B=\emptyset \). Hence, \(A\npreceq _{K}^{h}B\).
Example 2
Let \(Y=l^{2}\), and . It is clear that \(A\preceq _{K}^{p}B\). In the following, we prove \(( \bigcap_{a\in A}{ ( a+K )} ) \cap B = \emptyset \). Suppose to the contrary that \(( \bigcap_{a\in A}{ ( a+K )} ) \cap B \ne \emptyset \), then there exists \(a_{1}\in B=A\) such that \(a_{1}\in a+K\), \(\forall a\in A\), that is \(a_{1}-a\in K\) for any \(a\in A\), this obviously does not hold. Hence, \(A\npreceq _{K}^{h}B\).
Proposition 2.3
If \(A\ll _{K}^{p}B\), then \(A\preceq _{K}^{h}B\) and \(A\preceq _{K}^{p}B\).
Proof
Let \(A\ll _{K}^{p}\)B, then \(( \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} ) \cap B\ne \emptyset \). Since \(\bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} \subset \bigcap_{a\in A}{ ( a+K )}\), we obtain \(A\preceq _{K}^{h}B\). We obtain \(A\ne B\) from Proposition 2.1, therefore \(A\preceq _{K}^{p}B\). □
3 Nonlinear functional
Let \(\hat{e}\in \mathrm{qri}S\) and \(\eta \in Y\), we construct the function as
Let \(\hat{k}\in \mathrm{int}K\), \(\xi \in Y\), we construct the function as
Proposition 3.1
Let \(A,B\in \mathcal{P} ( Y )\). If \(A=r\hat{k}+B-\xi \), then \(I_{\hat{k}}^{p} ( A,B )=r\).
Proof
If \(A=r\hat{k}+B-\xi \), then it is clear that \(A\preceq _{K}^{p}r\hat{k}+B-\xi \), hence \(I_{\hat{k}}^{p} ( A,B )\leqslant r\). In the following, we prove \(A\npreceq _{K}^{p} ( r-\varepsilon ) \hat{k}+B-\xi \), \(\forall \varepsilon >0\). Otherwise, there exists \(\varepsilon _{1}>0\) such that \(A\preceq _{K}^{p} ( r-\varepsilon _{1} )\hat{k}+B-\xi \), hence \(A\preceq _{K}^{h} ( r-\varepsilon _{1} )\hat{k}+B-\xi \), that is \(\bigcap_{a\in A}{ ( a+K )}\cap ( ( r- \varepsilon _{1} )\hat{k}+B-\xi ) \ne \emptyset \). Therefore, \(\bigcap_{a\in A}{ ( a+K )}\cap ( A- \varepsilon _{1} \hat{k} )\ne \emptyset \). Hence, there exists \(a_{1}\in A\) such that \(a_{1}-\varepsilon _{1} \hat{k}\in a+K\), \(\forall a\in A\), thus \(-\varepsilon _{1} \hat{k}\in K\), hence \(\hat{k} \in -K\), which leads to a contradiction. Then, \(I_{\hat{k}}^{p} ( A,B ) \geqslant r\). Hence, we have \(I_{\hat{k}}^{p} ( A,B ) =r\). □
Proposition 3.2
Let \(A,B\in \mathcal{P} ( Y )\). If \(A\preceq _{K}^{h}r\hat{k}+B-\xi \), then \(A\preceq _{K}^{h} ( r+\varepsilon ) \hat{k}+B-\xi \), \(\forall \varepsilon >0\).
Proof
If \(A\preceq _{K}^{h}r \hat{k}+B-\xi \), then \(\bigcap_{a\in A}{ ( a+K )}\cap ( r\hat{k}+B- \xi ) \ne \emptyset \). Hence, there exists \(b_{1}\in B\) such that \(r\hat{k}+b_{1}-\xi \in \bigcap_{a\in A}{ ( a+K )}\). Hence, \(( r+\varepsilon ) \hat{k}+b_{1}-\xi \in \bigcap_{a\in A}{ ( a+K ) +\varepsilon \hat{k}}\subset \bigcap_{a\in A}{ ( a+K )}+\mathrm{int}K\subset \bigcap_{a\in A}{ ( a+K )}\), \(\forall \varepsilon >0\). Therefore, \(\bigcap_{a\in A}{ ( a+K )}\cap ( ( r+ \varepsilon ) \hat{k}+B-\xi ) \ne \emptyset \), that is \(A\preceq _{K}^{h} ( r+\varepsilon ) \hat{k}+B-\xi \). □
Remark 3.1
The following example indicates that the above proposition is not necessarily true for \(\preceq _{K}^{p}\).
Example 3
Let , , \(A=B= \{ ( x,y ) | x^{2}+y^{2}\leqslant 1 \}\), \(r=1\), \(\xi = ( 1,1 )\) and \(\hat{k}= ( 1,1 )\). It is clear that \(A\preceq _{K}^{p}r\hat{k}+B-\xi \). When \(0<\varepsilon <\sqrt{2}-1\), we obtain that \(\bigcap_{a\in A}{ ( a+K )}\cap ( ( r+ \varepsilon ) \hat{k}+B-\xi ) =\emptyset \). Hence, \(A\npreceq _{K}^{p} ( r+\varepsilon ) \hat{k}+B-\xi \).
Proposition 3.3
Let \(A,B\in \mathcal{P} ( Y )\), and \(\hat{e}\in \mathrm{qri}S\). Then, the following statements hold:
-
(i)
If \(I_{\hat{e}}^{m} ( A,B ) < r\), then \(A\prec _{\mathrm{qri}S}^{m}r\hat{e}+B-\eta \).
-
(ii)
Let \(A\,\dot{-}\,B\) be compact. If \(I_{\hat{e}}^{m} ( A,B ) =r\), then \(A\preceq _{\mathrm{cl}S}^{m}r\hat{e}+B-\eta \) and \(A\npreceq _{S}^{m} ( r-\lambda ) \hat{e}+B-\eta \), \(\forall \lambda >0\).
-
(iii)
If \(A\preceq _{S}^{m}r\hat{e}+B-\eta \) and \(A\npreceq _{S}^{m} ( r-\lambda ) \hat{e}+B-\eta \), \(\forall \lambda >0\), then \(I_{\hat{e}}^{m} ( A,B ) =r\).
Proof
(i) Assume that and let \(\mu =r-\alpha \). Then, there exists such that \(A\preceq _{S}^{m}t_{1}\hat{e}+B-\eta \) and \(\alpha \leqslant t_{1}<\alpha +\mu =r\), that is \(( A \,\dot{-}\, ( t_{1}\hat{e}+B-\eta ) ) \cap ( -S ) \ne \emptyset \). Hence, there exists \(s_{1}\in -S\) such that \(s_{1}\in A\,\dot{-}\,B-t_{1}\hat{e}+\eta \). Hence,
and
Hence, we obtain \(( A \,\dot{-}\, ( r\hat{e}+B-\eta ) ) \cap ( -\mathrm{qri}S ) \ne \emptyset \), that is \(A\prec _{\mathrm{qri}S}^{m}r\hat{e}+B-\eta \).
(ii) Let \(I_{\hat{e}}^{m} ( A,B ) =r\). Since \(I_{\hat{e}}^{m} ( A,B ) < r+\frac{1}{n}\) for all , we have \(A\prec _{\mathrm{qri}S}^{m} ( r+\frac{1}{n} ) \hat{e}+B- \eta \) from (i). Hence, we obtain \(A\preceq _{S}^{m} ( r+\frac{1}{n} ) \hat{e}+B-\eta \) for all , that is
Hence, there exists \(x_{n}\in A\,\dot{-}\,B\) such that
Since \(A\,\dot{-}\,B\) is compact, it follows that there exists a convergent subsequence of \(\{ x_{n} \} \). Without loss of generality, we let \(x_{n}\rightarrow x_{0}\in A\,\dot{-}\,B\). From (3.1), we obtain
Hence, we obtain \(( A \,\dot{-}\, ( r\hat{e}+B ) +\eta ) \cap ( -\mathrm{cl}S ) \ne \emptyset \), that is \(A\preceq _{\mathrm{cl}S}^{m}r\hat{e}+B-\eta \). Since \(I_{\hat{e}}^{m} ( A,B ) =r\), we obtain \(A\npreceq _{S}^{m} ( r-\lambda ) \hat{e}+B-\eta \) for any \(\lambda >0\) from the definition.
(iii) It follows that \(I_{\hat{e}}^{m} ( A,B ) =r\) from the definition. □
The converse of Proposition 3.3(i) is not true. The following example justifies this.
Example 4
Let \(Y=l^{2}\), \(\Vert y \Vert _{2}= ( \sum_{n=1} ^{ \infty}{y_{i}^{2}} ) ^{\frac{1}{2}}\), and . We obtain from [24]. Let \(a_{1}= ( -\frac{1}{4},-\frac{1}{4^{2}},-\frac{1}{4^{3}},\ldots , - \frac{1}{4^{n}},\ldots )\), \(a_{2}= ( -\frac{1}{5},-\frac{1}{5^{2}},-\frac{1}{5^{3}},\ldots - \frac{1}{5^{n}},\ldots )\), \(A= \{ a_{1},a_{2} \}\), \(\hat{e}= ( \frac{1}{3},\frac{1}{3^{2}},\frac{1}{3^{3}},\ldots , \frac{1}{3^{n}},\ldots )\), \(B= \{ 0_{Y} \}\), \(\eta = 0_{Y}\). We obtain \(( A \,\dot{-}\, B ) \cap ( - \mathrm{qri}S )\ne \emptyset \), hence, \(A\prec _{\mathrm{qri}S}^{m}B+r_{0}\hat{e}\) for \(r_{0}=0\).
Next, we verify \(I_{\hat{e}}^{m} ( A,B ) \geqslant r_{0}=0\). Suppose to the contrary that there exists \(t_{0}<0\) such that \(A\preceq _{S}^{m}t_{0}\hat{e}+B\), then \(A\preceq _{S}^{m}t_{0}\hat{e}\). We obtain \(a_{1}-t_{0}\hat{e}\in -S\) or \(a_{2}-t_{0}\hat{e}\in -S\). If \(a_{1}-t_{0}\hat{e}\in -S\), then \(-\frac{1}{4^{n}}-t_{0}\frac{1}{3^{n}}\leqslant 0\), for \(n\rightarrow \infty \) we obtain \(t_{0}\geqslant 0\); if \(a_{2}-t_{0}\hat{e}\in -S\), similarly, we obtain \(t_{0}\geqslant 0\).
Proposition 3.4
Let \(A,B\in \mathcal{P} ( Y )\), and \(\hat{k}\in \mathrm{int}K\). Then, the following statements are true:
-
(i)
If \(A\ll _{K}^{p}r\hat{k}+B-\xi \), then \(I_{\hat{k}}^{p} ( A,B ) < r\).
-
(ii)
Let B be compact, then \(I_{\hat{k}}^{p} ( A,B ) =r\) if and only if \(A\preceq _{\mathrm{cl}K}^{p}r\hat{k}+B-\xi \) and \(A\npreceq _{K}^{p} ( r-\varepsilon ) \hat{k}+B-\xi \), \(\forall \varepsilon >0\).
Proof
(i) If \(A\ll _{K}^{p}r\hat{k}+B-\xi \), then \(( \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} ) \cap ( r\hat{k}+B-\xi ) \ne \emptyset \), hence, there exists \(b_{1}\in B\) such that \(r\hat{k}+b_{1}-\xi \in \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K}\). Hence, there is \(y_{1}\in \bigcap_{a\in A}{ ( a+K )}\) such that \(r\hat{k}+b_{1}-\xi \in y_{1}+\mathrm{int}K\), then there exists \(\epsilon _{1}>0\) such that \(r\hat{k}+b_{1}-\xi -\epsilon _{1}\hat{k}\in y_{1}+\mathrm{int}K\), that is \(( \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} ) \cap ( ( r-\epsilon _{1} ) \hat{k}+B-\xi ) \ne \emptyset \). Then, we obtain \(( \bigcap_{a\in A}{ ( a+K )} ) \cap ( ( r-\epsilon _{1} ) \hat{k}+B-\xi ) \ne \emptyset \), i.e., \(A\preceq _{K}^{h} ( r-\epsilon _{1} ) \hat{k}+B-\xi \). We obtain \(A\preceq _{K}^{p} ( r-\epsilon _{1} ) \hat{k}+B-\xi \) from Proposition 2.2. Hence, \(I_{\hat{k}}^{p} ( A,B ) \leqslant r-\epsilon _{1}< r\).
(ii) Let \(I_{\hat{k}}^{p} ( A,B ) =r\). Case 1: if \(A=r\hat{k}+B-\xi \), then \(A\preceq _{\mathrm{cl}K}^{p}r\hat{k}+B-\xi \); Case 2: if \(A\ne r\hat{k}+B-\xi \), from the definition of infimum, it follows that there exists \(t_{n}\in [ r,r+\frac{1}{n} )\) such that \(A\preceq _{K}^{p}t_{n}\hat{k}+B-\xi \) for . We further assert that \(A\ne t_{n}\hat{k}+B-\xi \) for any . Otherwise, if there is such that \(A=t_{n_{1}}\hat{k}+B-\xi \), which together with \(A\ne r\hat{k}+B-\xi \) gives that \(t_{n_{1}}\ne r\). On the other hand, from Proposition 3.1, it follows that \(I_{\hat{k}}^{p} ( A,B ) =t_{n_{1}}\). This contradicts \(I_{\hat{k}}^{p} ( A,B ) =r\). Therefore, \(A\preceq _{K}^{h}t_{n}\hat{k}+B-\xi \). We obtain \(A\preceq _{K}^{h} ( r+\frac{1}{n} ) \hat{k}+B-\xi \) from Proposition 3.2, . Hence, there exists \(x_{n}\in B\) such that
Since B is compact, there exists a convergent subsequence of \(\{ x_{n} \} \). Without loss of generality, let \(x_{n}\rightarrow x_{0}\in B\). From (3.2), we have
Hence, we obtain \(\bigcap_{a\in A}{ ( a+\mathrm{cl}K ) \cap ( r\hat{k}+B-\xi )}\ne \emptyset \), hence, \(A\preceq _{\mathrm{cl}K}^{h}r\hat{k}+B-\xi \). Therefore, \(A\preceq _{\mathrm{cl}K}^{p}r\hat{k}+B-\xi \). We obtain \(A\npreceq _{K}^{p} ( r-\varepsilon ) \hat{k}+B-\xi \) for \(\varepsilon >0\) from the definition of \(I_{\hat{k}}^{p} ( A,B )=r\).
Conversely, let \(A\preceq _{\mathrm{cl}K}^{p}r\hat{k}+B-\xi \). Case 1: if \(A=r\hat{k}+B-\xi \), then \(A\preceq _{K}^{p}r\hat{k}+B-\xi \), then \(I_{\hat{k}}^{p} ( A,B ) \leqslant r\); Case 2: if \(A\ne r\hat{k}+B-\xi \), then \(A\preceq _{\mathrm{cl}K}^{h}r\hat{k}+B-\xi \), there exists \(y_{1}\in \bigcap_{a\in A}{ ( a+\mathrm{cl}K ) \cap ( r\hat{k}+B-\xi )}\). Then, \(y_{1}\in a+\mathrm{cl}K\) for all \(a\in A\) and \(y_{1}\in r\hat{k}+B-\xi \). Hence, \(y_{1}+\frac{1}{n}\hat{k}\in a+\mathrm{cl}K+\mathrm{int}K\subset a+ \mathrm{int}K\) for , then \(y_{1}+\frac{1}{n}\hat{k}\in \bigcap_{a\in A}{ ( a+ \mathrm{int}K )}\), so \(y_{1}+\frac{1}{n}\hat{k}\in \bigcap_{a\in A}{ ( a+ \mathrm{int}K )}\cap [ ( r+\frac{1}{n} ) \hat{k}+B-\xi ]\). Hence,
We obtain \(A\preceq _{\mathrm{int}K}^{h} ( r+\frac{1}{n} ) \hat{k}+B- \xi \), thus \(A\preceq _{K}^{h} ( r+\frac{1}{n} ) \hat{k}+B-\xi \), we obtain \(A\preceq _{K}^{p} ( r+\frac{1}{n} ) \hat{k}+B-\xi \) from Proposition 2.2. Hence, we have \(I_{\hat{k}}^{p} ( A,B ) \leqslant r+\frac{1}{n}\). Then, \(I_{\hat{k}}^{p} ( A,B ) \leqslant r\) for \(n\rightarrow \infty \). From the assumption we have \(I_{\hat{k}}^{p} ( A,B ) \geqslant r\). Hence, we have \(I_{\hat{k}}^{p} ( A,B ) =r\). □
The following example shows that the converse of Proposition 3.4(i) is false.
Example 5
Let \(A= \{ ( x,y ) | x^{2}+y^{2}\leqslant 1 \} \), \(B=A\), , \(\hat{k}= ( 1,1 )\) and \(\xi = ( 1,1 )\). We can obtain \(A=t\hat{k}+B-\xi \) for \(t=1\), and \(\bigcap_{a\in A}{ ( a+K )}\cap ( t\hat{k}+B- \xi ) =\emptyset \) for \(t<1\), respectively. Then, \(I_{\hat{k}}^{p} ( A,B ) =1\). However, when \(1< r\leqslant \sqrt{2}\), we have \(( \bigcap_{a\in A}{ ( a+K ) +\mathrm{int}K} ) \cap ( r\hat{k}+B-\xi ) =\emptyset \). Hence, \(A \not \ll _{K}^{p} r\hat{k}+B-\xi \).
Definition 3.1
A function is called m-increasing \(( p\text{-}\text{increasing} )\) on \(\mathcal{P} ( Y )\) if \(A,B\in \mathcal{P} ( Y )\) and \(A\preceq _{S}^{m}B ( A\preceq _{K}^{p}B )\) implies \(T ( A ) \leqslant T ( B )\).
Proposition 3.5
Let \(A\in \mathcal{P} ( Y )\). Then, \(I_{\hat{e}}^{m} ( \cdot ,A )\) is m-increasing on \(\mathcal{P} ( Y )\).
Proof
It can be proved similarly to [11, Proposition 15]. □
Corollary 3.1
Let \(A\in \mathcal{B} ( Y )\), then \(I_{\hat{e}}^{m} ( A-\eta ,A ) =0\).
Proof
. □
Proposition 3.6
Let \(A\in \mathcal{P} ( Y )\). Then, \(I_{\hat{k}}^{p} ( \cdot ,A )\) is p-increasing on \(\mathcal{P} ( Y )\).
Proof
Let \(D,E\in \mathcal{P} ( Y )\) and \(D\preceq _{K}^{p}E\). Case 1: if , then \(E\preceq _{K}^{p} ( \alpha +\epsilon ) \hat{k}+A-\xi \), \(\forall \epsilon >0\). Since \(D\preceq _{K}^{p}E\) and \(\preceq _{K}^{p}\) is transitive, we have \(D\preceq _{K}^{p} ( \alpha +\epsilon ) \hat{k}+A-\xi \). Hence, \(I_{\hat{k}}^{p} ( D,A ) \leqslant \alpha +\epsilon \). We obtain \(I_{\hat{k}}^{p} ( D,A ) \leqslant \alpha =I_{\hat{k}}^{p} ( E,A )\) for \(\epsilon \rightarrow 0^{+}\).
Case 2: If \(I_{\hat{k}}^{p} ( E,A ) =-\infty \), then . Since \(D\preceq _{K}^{p}E\) and \(\preceq _{K}^{p}\) is transitive, we have . Then, \(I_{\hat{k}}^{p} ( D,A ) =-\infty \).
Case 3: If \(I_{\hat{k}}^{p} ( E,A ) =+\infty \), then it is clear that \(I_{\hat{k}}^{p} ( D,A ) \leqslant I_{\hat{k}}^{p} ( E,A )\). □
Proposition 3.7
Let \(A,B\in P ( Y )\) and \(I_{\hat{e}}^{m} ( A,B )\) be finite. If \(A\,\dot{-}\,B\) is compact and S is closed, then \(A\preceq _{S}^{m}B-\eta \) if and only if \(I_{\hat{e}}^{m} ( A,B ) \leqslant 0\).
Proof
It can be proved similarly to [11, Proposition 19(i)]. □
Proposition 3.8
Let \(A\in \mathcal{B} ( Y )\), then \(I_{\hat{k}}^{p} ( A-\xi ,A ) =0\).
Proof
. When \(t=0\), \(A\preceq _{K}^{p}A\) is obviously true, that is \(I_{\hat{k}}^{p} ( A-\xi ,A ) \leqslant 0\). Now, we prove that \(A\npreceq _{K}^{p}A+t\hat{k}\), \(\forall t<0\). Suppose to the contrary that \(A\preceq _{K}^{p}A+t_{1}\hat{k}\) for some \(t_{1}<0\). Since \(A\ne A+t_{1}\hat{k}\), we obtain \(A\preceq _{K}^{h}A+t_{1}\hat{k}\), that is \(\bigcap_{a\in A}{ ( a+K )}\cap ( t_{1} \hat{k}+A ) \ne \emptyset \). Then, there is \(a_{1}\in A\) such that \(t_{1}\hat{k}+a_{1}\in \bigcap_{a\in A}{ ( a+K )}\), that is \(t_{1}\hat{k}+a_{1}\in a+K\), \(\forall a\in A\). Let \(a=a_{1}\), then \(t_{1}\hat{k}\in K\). Since \(t_{1}<0\), we have \(\hat{k}\in -K\). Therefore \(\mathrm{int}K\cap ( -K ) \ne \emptyset \), which contradicts that K is a pointed convex cone. Hence, \(A\npreceq _{K}^{p}A+t\hat{k}\) for \(\ t<0\), i.e., \(I_{\hat{k}}^{p} ( A-\xi ,A ) \geqslant 0\). Hence, \(I_{\hat{k}}^{p} ( A-\xi ,A ) =0\). □
Corollary 3.2
Let \(A,B\in \mathcal{P} ( Y )\). If B is compact, K is closed and \(I_{\hat{k}}^{p} ( A,B )\) is finite, then .
Proof
We have from Proposition 3.4(ii). Hence, . □
4 Approximately efficient solutions and approximate weakly efficient solutions for set-optimization problems
We consider a set-valued mapping \(F: X\rightrightarrows Y\) and a nonempty set \(T\subset X\) where \(F ( x ) \in \mathcal{P} ( Y )\) for each \(x\in X\). We deal with the constrained set-optimization problem \(( P )\) defined by
Definition 4.1
([7])
An element \(x_{0}\in T\) is called an m-efficient solution of \(( P )\), if \(F ( x ) \npreceq _{S}^{m}F ( x_{0} )\) or \(F ( x ) =F ( x_{0} )\) for any \(x\in T\). Let \(m\text{-}\text{E}_{S} ( F,T )\) denote the set of m-efficient solutions of \(( P )\).
In what follows, we introduce a new class of approximate solutions.
Definition 4.2
Let \(\eta \in Y\). An element \(x_{0}\in T\) is called an η-m-efficient solution of \(( P )\), if \(F ( x ) +\eta \npreceq _{S}^{m}F ( x_{0} )\) or \(F ( x ) +\eta =F ( x_{0} )\) for \(x\in T\). Let η-m-\(E_{S} ( F,T )\) denote the set of η-m-efficient solutions of \(( P )\).
In the following, by using qriS, we introduce another new class of weakly efficient solutions.
Definition 4.3
Let \(\eta \in Y\). An element \(x_{0}\in T\) is called
-
(i)
an m-weakly efficient solution of \(( P )\) with respect to qriS, if \(F ( x ) \nprec _{\mathrm{qri}S}^{m}F ( x_{0} )\) for \(x\in T\);
-
(ii)
an η-m-weakly efficient solution of \(( P )\) with respect to qriS, if \(F ( x ) +\eta \nprec _{\mathrm{qri}S}^{m}F ( x_{0} )\) for \(x\in T\).
Let m-\(W_{\mathrm{qri}S} ( F,T )\) and η-m-\(W_{ \mathrm{qri}S} ( F,T )\) denote the sets of m-weakly efficient solutions and η-m-weakly efficient solutions of \(( P )\) with respect to qriS, respectively.
Remark 4.1
If \(\eta =0\), then η-m-weakly efficient solutions reduce to m-weakly efficient solutions and η-m-efficient solutions reduce to m-efficient solutions of \(( P )\), respectively.
Theorem 4.1
Let \(\eta \in S\). Then, the following statements are true:
-
(i)
Let \(F ( x_{0} ) \in \mathcal{B} ( Y )\) and \(x_{0}\in m\)-\(E_{S} ( F,T ) \), then \(x_{0}\in \eta \)-m-\(E_{S} ( F,T )\).
-
(ii)
Let S be closed, \(F ( x_{0} )\in \mathcal{B} ( Y )\) and \(x_{0}\in \eta \)-m-\(E_{S} ( F,T )\), then \(x_{0}\in \eta \)-m-\(W_{\mathrm{qri}S} ( F,T )\).
-
(iii)
m-\(W_{\mathrm{qri}S} ( F,T ) \subset \bigcap_{\eta \in S\backslash \{ 0 \}} { \eta \textit{-} m \textit{-} W_{{\mathrm{qri}}{S}} ( F,T )}\).
-
(iv)
Let S be closed, then \(\bigcap_{\eta \in S\backslash \{ 0 \}} {\eta \textit{-} m \textit{-} W_{{\mathrm{qri}}{S}} ( F,T )} \subset \) m-\(W_{\mathrm{qri}S} ( F,T )\).
Proof
(i) ① If \(\eta =0\), then the conclusion holds obviously.
② If \(\eta \in S\backslash \{ 0 \}\), suppose that \(x_{0}\in m\)-\(E_{S} ( F,T ) \). \(\forall x\in T\), we consider two cases.
Case 1: when \(F ( x ) =F ( x_{0} )\), we prove \(( ( F ( x ) +\eta ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -S )=\emptyset \). In fact
Since \(\eta \in S\backslash \{ 0 \}\) and S is a proper pointed convex cone, then \(\eta \notin -S\). Hence, we obtain \(( ( F ( x ) +\eta ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -S )=\emptyset \), thus \(F ( x ) +\eta \npreceq _{S}^{m}F ( x_{0} )\);
Case 2: when \(F ( x ) \ne F ( x_{0} )\). Since \(x_{0}\in m\)-\(E_{S} ( F,T )\), we obtain \(F ( x ) \npreceq _{S}^{m}F ( x_{0} )\), that is
If \(F ( x )+\eta =F ( x_{0} )\), then \(x_{0}\in \eta \)-m-\(E_{S} ( F,T )\) holds obviously. If \(F ( x )+\eta \ne F ( x_{0} )\), then we prove \(( ( F ( x ) +\eta ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -S )=\emptyset \). Suppose to the contrary that there exists \(x_{1}\) such that \(( ( F ( x_{1} ) +\eta ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -S ) \ne \emptyset \), then there exist \(s_{1}\in -S\) and \(y_{0}\in F ( x_{1} ) \,\dot{-}\,F ( x_{0} )\) such that \(s_{1}=\eta +y_{0}\). Then, \(y_{0}=s_{1}-\eta \in -S-S\subset -S\). Thus, \(y_{0}\in ( F ( x_{1} ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -S )\), which contradicts (4.1).
Therefore, we obtain \(x_{0}\in \eta \)-m-\(E_{S} ( F,T )\).
(ii) Let \(x_{0}\in \eta \)-m-\(E_{S} ( F,T )\). \(\forall x\in T\), we consider two cases. Case 1: when \(F ( x ) +\eta \npreceq _{S}^{m}F ( x_{0} )\), that is \(( ( F ( x ) +\eta ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -S ) =\emptyset \). It follows from \(\mathrm{qri}S\subset S\) that
Case 2: when \(F ( x ) +\eta = F ( x_{0} )\), we obtain \(( F ( x ) + \eta ) \,\dot{-}\, F ( x_{0} ) = \{ 0 \}\) from \(F ( x_{0} ) \in \mathcal{B} ( Y )\). On the other hand, we obtain \(0\notin \mathrm{qri}S\) from [24, Lemma 4.1], then
Thus, \(F ( x ) +\eta \nprec _{\mathrm{qri}S}^{m}F ( x_{0} )\), \(\forall x\in T\). Hence, \(x_{0}\in \eta \)-m-\(W_{ \mathrm{qri}S} ( F,T )\).
(iii) Let \(x_{0}\in m\)-\(W_{\mathrm{qri}S} ( F,T )\), then \(F ( x ) \nprec _{\mathrm{qri}S}^{m}F ( x_{0} )\) for \(\forall x\in T\), that is
Suppose to the contrary that \(x_{0}\notin \bigcap_{\eta \in S\backslash \{ 0 \}} {\eta \text{-} m \text{-} W_{{\mathrm{qri}}{S}} ( F,T )}\), then \(x_{0}\notin {\eta}_{1} \text{-} m \text{-} W_{{\mathrm{qri}}{S}} ( F,T )\) for some \(\eta _{1} \in S\backslash \{ 0 \}\). Then, there exists \(x_{1}\in T\) such that \(( ( \eta _{1}+F ( x_{1} ) ) \,\dot{-}\,F ( x_{0} ) ) \cap ( -\mathrm{qri}S ) \ne \emptyset \). Then, there exists \(s_{1}\in -\mathrm{qri}S\) and \(y_{0}\in F ( x_{1} ) \,\dot{-}\,F ( x_{0} )\) such that \(s_{1}=\eta _{1}+y_{0}\). Hence, \(y_{0}=s_{1}-\eta _{1}\in -\mathrm{qri}S-S\subset -\mathrm{qri}S\), we obtain \(y_{0}\in ( F ( x_{1} ) \,\dot{-}\, F ( x_{0} ) ) \cap ( -\mathrm{qri}S )\), which contradicts (4.2).
(iv) Let \(x_{0}\in T\) and \(x_{0}\notin m\)-\(W_{\mathrm{qri}S} ( F,T )\). Then, there exists \(x_{2}\in T\) such that \(( F ( x_{2} ) \,\dot{-}\, F ( x_{0} ) ) \cap ( -\mathrm{qri}S ) \ne \emptyset \). Hence, there is \(s_{2}\in \mathrm{qri}S\) such that \(-s_{2}\in F ( x_{2} ) \,\dot{-}\,F ( x_{0} )\), and \(-\frac{s_{2}}{2}= -s_{2}+\frac{s_{2}}{2} \in F ( x_{2} ) { \,\dot{-}\,}F ( x_{0} )+\frac{s_{2}}{2}\). It follows that \(\frac {s_{2}}{2}\in \mathrm{qri}S\) from the fact that S is a convex cone. Therefore,
Hence, \(x_{0}\notin \frac{s_{2}}{2}\)-m-\(W_{\mathrm{qri}S} ( F,T )\). We obtain \(0 \notin \mathrm{qri}S\) from [24, Lemma 4.1], then \(s_{2} \ne 0\). Therefore, \(\frac{s_{2}}{2}\in S\backslash \{ 0 \}\). Hence, \(x_{0}\notin \bigcap_{\eta \in S\backslash \{ 0 \}}{}{\eta \text{-} m \text{-} W_{{\mathrm{qri}}{S}} ( F,T )}\). □
Remark 4.2
In the following, let \(e\in S\backslash \{ 0 \}\), we consider a special class of approximate weakly efficient solutions, where \(\eta =\epsilon e\), \(\epsilon > 0\). From Theorem 4.1(iii), we obtain m-\(W_{\mathrm{qri}S} ( F,T ) \subset \bigcap_{\epsilon >0} {\epsilon e\text{-}m\text{-}W_{\mathrm{qri}S} ( F,T )}\). However, the converse inclusion may not hold. The following example illustrates the case.
Example 6
Let , \(T= [ -1, 1 ] \), \(Y=l^{2}\), \(e= ( \frac{1}{2},\frac{1}{2^{2}},\frac{1}{2^{3}},\ldots , \frac{1}{2^{n}},\ldots )\), \(\Vert y \Vert _{2}= ( \sum_{i=1} ^{ \infty}{y_{i}^{2}} ) ^{\frac{1}{2}}\), , and . We obtain from [24]. Consider \(F: T\rightrightarrows Y\) defined as
A direct calculation gives that m-\(W_{\mathrm{qri}S} ( F,T ) = [ -1,0 )\), thus \([ -1,0 ) \subset \bigcap_{\epsilon >0} \epsilon e \)-m-\(W_{\mathrm{qri}S} ( F, T )\).
Let us verify \(0\in \epsilon e\)-m-\(W_{\mathrm{qri}S} ( F,T )\) for \(\forall \mathrm{\epsilon}>0\). In fact, \(F ( x ) \,\dot{-}\, F ( 0 ) +\epsilon e=F ( x ) +\epsilon e\). We consider two cases. Case 1: when \(x\in [ 0, 1 ] \), \(F ( x ) +\epsilon e= ( \frac{x+\epsilon}{2}, \frac{x+\epsilon}{2^{2}}, \ldots , \frac{x+\epsilon}{2^{n}}, \ldots )\notin -\mathrm{qri}K\); Case 2: when \(x\in [ -1,0 ) \), \(F ( x ) +\epsilon e= ( -\frac{1}{3}+ \frac{\epsilon}{2}, -\frac{1}{3^{2}}+\frac{\epsilon}{2^{2}}, \ldots , -\frac{1}{3^{n}}+\frac{\epsilon}{2^{n}}, \ldots )\). Since
therefore \(F ( x ) +\epsilon e\notin -\mathrm{qri}S\). It follows from the above two cases that \(( F ( x ) \,\dot{-}\,F ( 0 ) + \epsilon e ) \cap ( -\mathrm{qri}S )= \emptyset \) for \(\forall x\in T\). Thus, \(0\in \epsilon e\)-m-\(W_{ \mathrm{qri}S} ( F,T )\).
It follows from Theorem 4.1(iii) that
In the following, we prove
In fact, take \(x_{1}=-1\in T\), \(( F ( x_{1} ) + x_{0}e ) \,\dot{-}\,F ( x_{0} ) =F ( -1 ) \in - \mathrm{qri}S\). Hence, \(x_{0}\notin \epsilon _{0} e\)-m-\(W_{ \mathrm{qri}S} ( F, T )\), where \(\epsilon _{0}=x_{0}>0\). Hence, we obtain \(x_{0}\notin \cap_{\epsilon >0} \epsilon e\)-m-\(W_{ \mathrm{qri}S} ( F,T )\). From \(\bigcap_{\epsilon >0} \epsilon e\)-m-\(W_{\mathrm{qri}S} ( F,T ) \subset [ -1, 1 ]\) and (4.4) we obtain \(\bigcap_{\epsilon >0} \epsilon e\)-m-\(W_{\mathrm{qri}S} ( F,T ) \subset [ -1, 0 ]\), which together with (4.3) gives
Next, we introduce the efficient solution and approximately efficient solution under the \(\preceq _{K}^{p}\) and \(\ll _{K}^{p}\) order relation, respectively.
Definition 4.4
Let \(x_{0}\in T\) and \(\xi \in K\). An element \(x_{0}\in T\) is called
-
(i)
an efficient solution of \(( P )\) under order relation \(\preceq _{K}^{p}\), if \(F ( x ) \npreceq _{K}^{p}F ( x_{0} )\) or \(F ( x ) =F ( x_{0} )\) for \(\forall x\in T\);
-
(ii)
a ξ-efficient solution of \(( P )\) under order relation \(\preceq _{K}^{p}\), if \(F ( x ) +\xi \npreceq _{K}^{p}F ( x_{0} )\) or \(F ( x ) +\xi =F ( x_{0} )\) for \(\forall x\in T\);
-
(iii)
a weakly efficient solution of \(( P )\) under order relation \(\ll _{K}^{p}\), if \(F ( x )\) \(\not \ll _{K}^{p}\) \(F ( x_{0} )\) for \(\forall x\in T\);
-
(iv)
a ξ-weakly efficient solution of \(( P )\) under order relation \(\ll _{K}^{p}\), if \(F ( x ) +\xi \not \ll _{K}^{p} F ( x_{0} )\) for \(\forall x\in T\).
Let \(E_{K}^{p} ( F,T )\), ξ-\(E_{K}^{p} ( F,T )\), \(W_{K}^{p} ( F,T )\), and ξ-\(W_{K}^{p} ( F,T )\) denote the sets of efficient solutions, ξ-efficient solutions, weakly efficient solutions, and ξ-weakly efficient solutions of \(( P )\), respectively.
Theorem 4.2
Let \(\xi \in K\), then the following statements are true:
-
(i)
\(E_{K}^{p} ( F,T ) \subset W_{K}^{p} ( F,T )\);
-
(ii)
\(E_{K}^{p} ( F,T ) \subset \) ξ-\(E_{K}^{p} ( F,T )\);
-
(iii)
\(W_{K}^{p} ( F,T ) = \bigcap_{\xi \in K\backslash \{ 0 \}} {\xi \textit{-}W_{K}^{p} ( F,T )}\).
Proof
(i) Let \(x_{0}\in T\) and \(x_{0}\notin W_{K}^{p} ( F,T )\), then there exists \(x_{1}\in T\) such that \(F ( x_{1} ) \ll _{K}^{p}F ( x_{0} )\). It follows from Propositions 2.1 and 2.3 that \(F ( x_{1} ) \ne F ( x_{0} )\) and \(F ( x_{1} ) \preceq _{K}^{p}F ( x_{0} )\). Hence, \(x_{0}\notin E_{K}^{p} ( F,T )\).
(ii) If \(\xi =0\), then the conclusion is obviously true. If \(\xi \ne 0\), let \(x_{0}\in E_{K}^{p} ( F,T )\). \(\forall x\in T\), we consider two cases. Case 1: if \(F ( x ) +\xi =F ( x_{0} )\), then \(x_{0}\in \xi \)-\(E_{K}^{p} ( F,T )\) is obvious; Case 2: if \(F ( x ) +\xi \ne F ( x_{0} )\), in the following, we prove \(\bigcap_{y\in F ( x )}{ ( y+\xi +K ) \cap F ( x_{0} ) =\emptyset}\).
(1) When \(F ( x ) \ne F ( x_{0} )\), it follows from \(x_{0}\in E_{K}^{p} ( F,T )\) that \(F ( x ) \npreceq _{K}^{p}F ( x_{0} )\). Hence, \(\bigcap_{y\in F ( x )}{ ( y+K ) \cap F ( x_{0} ) =\emptyset}\). From \(y+\xi +K\subset y+K\) we obtain \(\bigcap_{y\in F ( x )}{ ( y+\xi +K )} \cap F ( x_{0} ) =\emptyset \). Hence, \(F ( x ) +\xi \npreceq _{K}^{p}F ( x_{0} )\);
(2) When \(F ( x ) =F ( x_{0} )\), in the following, we prove \(\bigcap_{y\in F ( x_{0} )}{ ( y+\xi +K ) \cap F ( x_{0} ) =\emptyset}\). Otherwise, there exists \(y_{0}\in F ( x_{0} )\) such that \(y_{0}\in \bigcap_{y\in F ( x_{0} )}{ ( y+ \xi +K )}\), then \(y_{0}\in y_{0}+\xi +K\). Therefore, \(-\xi \in K\). It follows that \(\xi \in K\cap ( -K ) = \{ 0 \}\). Hence \(\xi =0\), which contradicts \(\xi \ne 0\). Hence, \(F ( x ) +\xi \npreceq _{K}^{p}F ( x_{0} )\).
Summarizing the above discussion, we obtain \(x_{0}\in \xi \)-\(E_{K}^{p} ( F,T )\).
(iii) “⊂” Let \(x_{0}\in W_{K}^{p} ( F,T ) \), then \(( \bigcap_{y\in F ( x )}{ ( y+K )}+\mathrm{int}K ) \cap F ( x_{0} ) = \emptyset \), \(\forall x\in T\). For each \(\xi \in K\backslash \{ 0 \}\), \(\bigcap_{y\in F ( x )}{ ( y+\xi +K )}= \bigcap_{y\in F ( x )}{ ( y+K )}+ \xi \). Therefore, \(\bigcap_{y\in F ( x )}{ ( y+\xi +K )}+ \mathrm{int}K=\bigcap_{y\in F ( x )}{ ( y+K )}+\xi +\mathrm{int}K \subset \bigcap_{y\in F ( x )}{ ( y+K )}+\mathrm{int}K\). Hence, \(( \bigcap_{y\in F ( x )}{ ( y+\xi +K )}+\mathrm{int}K )\cap F ( x_{0} ) = \emptyset \), that is \(F ( x ) +\xi \not \ll _{K}^{p}F ( x_{0} )\). We obtain \(x_{0}\in \xi \)-\(W_{K}^{p} ( F,T )\).
“⊃” Let \(x_{0}\in T\) and \(x_{0}\notin W_{K}^{p} ( F,T )\). Then, there exists \(x_{1}\in T\) such that \(F ( x_{1} ) \ll _{K}^{p}F ( x_{0} )\), that is \(( \bigcap_{y\in F ( x_{1} )}{ ( y+K ) +\mathrm{int}K} ) \cap F ( x_{0} ) \ne \emptyset \). Hence, there is \(y_{0}\in F ( x_{0} )\) such that \(y_{0}\in \bigcap_{y\in F ( x_{1} )}{ ( y+K )}+\mathrm{int}K\), then there exists \(y_{1}\in \bigcap_{y\in F ( x_{1} )}{ ( y+K )}\) such that \(y_{0}\in y_{1}+\mathrm{int}K\), therefore \(y_{0}\in \mathrm{int} ( y_{1}+\mathrm{int}K )\). Therefore, there exists \(\xi _{1}\in K\backslash \{ 0 \}\) such that \(y_{0}-\xi _{1}\in y_{1}+\mathrm{int}K\), i.e., \(y_{0}\in y_{1}+\xi _{1}+\mathrm{int}K\subset \bigcap_{y\in F ( x_{1} )}{ ( y+K+\xi _{1} )}+\mathrm{int}K\). We obtain \(( \bigcap_{y\in F ( x_{1} )}{ ( y+K+ \xi _{1} )}+\mathrm{int}K ) \cap F ( x_{0} ) \ne \emptyset \), that is \(F ( x_{1} ) +\xi _{1}\ll _{K}^{p}F ( x_{0} )\). Therefore, \(x_{0}\notin {\xi _{1}}\)-\(W_{K}^{p} ( F,T )\). Hence, \(x_{0}\notin \bigcap_{\xi \in K\backslash \{ 0 \}}{\xi \text{-}}W_{K}^{p} ( F,T )\). □
Remark 4.3
Similar to the proof of Theorem 4.2(i), we can prove ξ-\(E_{K}^{p} ( F,T )\subset \xi \)-\(W_{K}^{p} ( F,T )\).
5 Optimality conditions
Consider the set-optimization problem
Theorem 5.1
Let \(F ( x )\) be compact for any \(x\in T\) and S be closed, \(x_{0}\in T\). Then, \(x_{0}\) is an η-m-efficient solution of \(( P )\) if and only if there exists a function that is m-increasing on \(\mathcal{P} ( Y )\) and satisfies the following statements:
-
(i)
\(\phi ( F ( x_{0} ) -\eta ) =0\),
-
(ii)
\(\phi ( F ( x ) ) >0\) for all \(x\in T\backslash \{ x\in T:F ( x ) +\eta =F ( x_{0} ) \}\).
Proof
Assume that \(x_{0}\in T\) is an η-m-efficient solution of \(( P )\). Define by \(\phi ( \cdot )=I_{\hat{e}}^{m} ( \cdot ,F ( x_{0} ) )\). Then, \(\phi ( \cdot )\) is m-increasing on \(\mathcal{P} ( Y )\) from Proposition 3.5. We have \(\phi ( F ( x_{0} ) -\eta ) =0\) from Corollary 3.1. Since \(x_{0}\) is an η-m-efficient solution, we obtain \(F ( x ) +\eta \npreceq _{S}^{m}F ( x_{0} )\) for all \(x\in T\backslash \{ x\in T:F ( x ) +\eta =F ( x_{0} ) \}\) from the definition. It follows from Proposition 3.7 that \(I_{\hat{e}}^{m} ( F ( x ) ,F ( x_{0} ) ) >0\). Hence, \(T ( F ( x ) ) >0\).
Conversely, let (i) and (ii) be true for some that is m-increasing on \(\mathcal{P} ( Y ) \). Suppose to the contrary that \(x_{0}\) is not an η-m-efficient solution of \(( P )\). Then, there exists \(x_{1}\in T\) such that \(F ( x_{1} ) +\eta \preceq _{S}^{m}F ( x_{0} )\) and \(F ( x_{1} ) +\eta \ne F ( x_{0} )\), we obtain \(I_{\hat{e}}^{m} ( F ( x_{1} ) ,F ( x_{0} ) ) \leqslant 0\) from Proposition 3.7. Hence, \(\phi ( F ( x_{1} ) ) \leqslant 0\). This contradicts (ii). Therefore, \(x_{0}\) is an η-m-efficient solution of \(( P )\). □
In what follows, we give an example to illustrate the necessity of the above theorem.
Example 7
Let , \(T= [ 0,1 ]\), \(Y=l^{2}\) and . Take \(\hat{e}= ( 1,1,\frac{1}{3},\frac{1}{4},\ldots ,\frac{1}{n}, \ldots )\), \(\eta = ( 1,2,\frac{1}{3},\frac{1}{4},\ldots ,\frac{1}{n}, \ldots )\). Let \(F:X\rightrightarrows Y\) be defined as
A direct calculation gives that \(F ( x ) \,\dot{-}\,F ( 0 ) = \{ ( 0,x,0,0, \ldots ) \}\) for \(x\in T\). Hence, we have \(( F ( x ) \,\dot{-}\,F ( 0 ) + \eta ) \cap ( -S ) =\emptyset \), thus \(F ( x ) +\eta \npreceq _{S}^{m}F ( 0 )\), and \(F ( x ) +\eta \ne F ( 0 )\) for \(\forall x\in T\). Hence, \(x_{0}=0\) is an η-m-efficient solution.
Define a function by
Then,
Thus, \(\phi ( F ( x ) ) >0\) for all \(x\in T\backslash \{ x\in T:F ( x ) +\eta =F ( 0 ) \}\).
Remark 5.1
According to Remark 2.1 and Theorem 4.1(i), Theorem 5.1 generalizes and improves [11, Theorem 2] in the following two aspects:
-
(i)
From a nonempty interior to a nonempty quasirelative interior of an order cone.
-
(ii)
From efficient solutions to approximately efficient solutions.
Theorem 5.2
Let \(T\subset X\), \(F:T\rightarrow 2^{Y}\) and K be closed, \(x_{0}\in T\) and \(F ( x_{0} ) \in \mathcal{P} ( Y )\). Suppose that there exists a functional satisfying
-
(i)
φ is p-increasing on \(\mathcal{P} ( Y )\);
-
(ii)
\(\varphi ( F ( x_{0} ) -\xi ) =0\);
-
(iii)
for any \(x\in T \backslash \{ x\in T:F ( x ) +\xi =F ( x_{0} ) \}\), \(\varphi ( F ( x ) ) >0\).
Then, \(x_{0}\) is a ξ-efficient solution of \(( P )\).
Proof
If \(x_{0}\) is not a ξ-efficient solution of \(( P )\), then there exists \(x_{1}\in T\) such that \(F ( x_{1} ) +\xi \preceq _{K}^{p}F ( x_{0} )\) and \(F ( x_{1} ) +\xi \ne F ( x_{0} )\). Then, \(\varphi ( F ( x_{1} ) ) \leqslant \varphi ( F ( x_{0} ) -\xi ) =0\), this is in contradiction with (iii). Hence, \(x_{0}\in T\) is a ξ-efficient solution of \(( P )\). □
Remark 5.2
We partly consider the converse proposition of Theorem 5.2. Let \(x_{0}\) be a ξ-efficient solution of \(( P )\), although we can construct a functional φ similar to the proof of necessity of Theorem 5.1, however, φ may not satisfy (i)–(iii) of Theorem 5.2. The following example illustrates the case.
Example 8
Let , \(T = [ 0,2 ]\) and . Define the set-valued map \(F\rightrightarrows Y\) as
Take \(x_{0}=2\), \(\hat{k}= ( 1,1 )\) and \(\xi = ( \frac{2-\sqrt{2}}{8},\frac{2-\sqrt{2}}{8} )\). Since \(\bigcap_{y\in F ( x )}{ ( y+K+\xi )} \cap F ( x_{0} ) =\emptyset \), we obtain \(F ( x ) +\xi \npreceq _{K}^{p}F ( x_{0} )\) for any \(x\in T\). Hence, \(x_{0}\) is a ξ-efficient solution. For the functional defined as \(\varphi ( A ) =I_{\hat{k}}^{p} ( A , F ( x_{0} ) )\), \(\forall A \in \mathcal{P} ( Y )\), we can verify that φ satisfies (i)–(ii). On the other hand, when \(x\in [ 0,1 )\), . If \(t=-\frac{2-\sqrt{2}}{8}\), then \(F ( x )+\xi =t\hat{k}+F ( x_{0} )\), hence \(F ( x ) +\xi \preceq _{K}^{p}t\hat{k}+F ( x_{0} )\); if \(t<-\frac{2-\sqrt{2}}{8}\), then \(F ( x ) +\xi \npreceq _{K}^{p}t\hat{k}+F ( x_{0} )\). Therefore, \(\varphi ( F ( x ) ) =-\frac{2-\sqrt{2}}{8}<0\), thus φ does not satisfy (iii).
6 Conclusions
In this paper, when the ordered cone S has nonempty quasirelative interiors or ordered cone K has nonempty interiors, we introduce several kinds of order relations \(\preceq _{K}^{p}\), \(\ll _{K}^{p}\), \(\preceq _{S}^{m}\), and \(\prec _{\mathrm{qri}K}^{m}\), respectively. We introduce two kinds of nonlinear functions, whose properties are discussed. With order relations \(\preceq _{K}^{p}\), \(\ll _{K}^{p}\), \(\preceq _{S}^{m}\), and \(\prec _{\mathrm{qri}K}^{m}\), we introduce several classes of approximately efficient solutions and approximate weakly efficient solutions, respectively, for set optimization, and study the relationship among them. The optimality conditions for approximate solutions of set optimization are established.
Availability of data and materials
Not applicable.
References
Adán, M., Novo, V.: Weak efficiency in vector optimization using a closure of algebraic type under cone-convexlikeness. Eur. J. Oper. Res. 149, 641–653 (2003)
Aubin, J.P., Frankowska, H.: Set-Valued Analysis. Birkhäuser, Basel (1990)
Borwein, J.M., Lewis, A.S.: Partially finite convex programming, Part I: Quasi relative interiors and duality theory. Math. Program. 57, 15–48 (1992)
Boţ, R.I., Csetnek, E.R., Wanka, G.: Regularity conditions via quasirelative interior in convex programming. SIAM J. Optimiz. 19, 217–233 (2008)
Gao, Y., Yang, X.M.: Properties of the nonlinear scalar functional and its applications to vector optimization problems. J. Global Optim. 73, 869–889 (2019)
Gong, X.H.: Optimality conditions for Henig and globally proper efficient solutions with ordering cone has empty interior. J. Math. Anal. Appl. 307, 12–31 (2005)
Gupta, M., Srivastava, M.: Approximate solutions and Levitin-Polyak well-posedness for set optimization using weak efficiency. J. Optim. Theory Appl. 186, 191–208 (2020)
Han, Y.: Connectedness of weak minimal solution set for set optimization problems. Oper. Res. Lett. 48, 820–826 (2020)
Han, Y., Huang, N.J.: Existence and connectivity of solutions for generalized vector quasiequilibrium problems. J. Optim. Theory Appl. 178, 131–152 (2018)
Jahn, J., Ha, T.: New order relations in set optimization. J. Optim. Theory Appl. 148, 209–236 (2011)
Karaman, E., Soyertem, M., et al.: Partial order relations on family of sets and scalarizations for set optimization. Positivity. 22, 783–802 (2018)
Khushboo, L.C.S.: A unified minimal solution in set optimization. J. Glob. Optim. 74, 195–211 (2019)
Küçük, M., Soyertem, M., et al.: Vectorization of set-valued maps with respect to total ordering cones and its applications to set-valued optimization problems. J. Math. Anal. Appl. 385, 285–292 (2012)
Kuroiwa, D.: Existence theorems of set optimization with set-valued maps. J. Inf. Optim. Sci. 24, 73–84 (2003)
Kuroiwa, D., Tanaka, T., Ha, T.X.D.: On cone convexity of set-valued maps. Nonlinear Anal., Theory Methods Appl. 30, 1487–1496 (1997)
Liu, C.P., Lee, H.W.J., Yang, X.M.: Optimality conditions and duality on approximate solutions in vector optimization with arcwise connectivity. Optim. Lett. 6, 1613–1626 (2012)
Loridan, P.: ε-Solutions in vector optimization. J. Optim. Theory Appl. 1984, 265–276 (1984)
Luu, D.V.: Second-order optimality conditions and duality for nonsmooth multiobjective optimization problems. Appl. Set-Valued Anal. Optim. 4, 41–54 (2022)
Qiu, Q.S., Yang, X.M.: Some properties of approximate solutions for vector optimization problem with set-valued functions. J. Glob. Optim. 47, 1–12 (2010)
Tung, L.T., Khai, T.T., et al.: Karush-Kuhn-Tucker optimality conditions and duality for set optimization problems with mixed constraints. J. Appl. Numer. Optim. 1, 277–291 (2019)
Vui, P.T., Anh, L.Q., Wangkeeree, R.: Levitin-Polyak well-posedness for set optimization problems involving set order relations. Positivisity. 23, 599–616 (2019)
White, D.J.: Epsilon efficiency. J. Optim. Theory Appl. 49, 319–337 (1986)
Zhang, C.L., Huang, N.J.: Set relations and weak minimal solutions for nonconvex set optimization problems with applications. J. Optim. Theory Appl. 190, 894–914 (2021)
Zhao, K.Q., Rong, W.D., Yang, X.M.: New nonlinear separation theorems and applications in vector optimization (in Chinese). Sci. Sin., Math. 47, 533–544 (2017)
Zhao, X.P., Yao, Y.H., et al.: A projected subgradient method for nondifferentiable quasiconvex multiobjective optimization problems. J. Optim. Theory Appl. 190, 82–107 (2021)
Zhou, Z.A., Peng, J.W.: Scalarization of set-valued optimization problems with generalized cone subconvex-likeness in real ordered linear spaces. J. Optim. Theory Appl. 154, 830–841 (2012)
Acknowledgements
The authors gratefully acknowledge the anonymous referees and editors for their constructive comments and advice on the earlier version of this paper.
Authors’ information
Yihong Xu (1969-), Professor, Doctor, major field of interest is in the area of set-valued optimization.
Funding
This research was supported by the National Natural Science Foundation of China Grant (11961047) and the Scientific Research Training Programs of Nanchang University (55002052, 55002136).
Author information
Authors and Affiliations
Contributions
PZ and YX together made the major analysis and the original draft preparation. BH analyzed all the results and made necessary improvements. PZ is the major contributor in writing the paper. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zhao, P., Xu, Y. & Huang, B. Approximate solutions for set optimization with an order cone that has nonempty quasirelative interiors. J Inequal Appl 2022, 101 (2022). https://doi.org/10.1186/s13660-022-02834-9
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-022-02834-9