Nội dung

Yugoslav Journal on Operations Research 23(2013) Number 2, 299–310 DOI: 10.2298/YJOR130108025S ON SUFFICIENCY IN MULTIOBJECTIVE PROGRAMMING INVOLVING GENERALIZED (G,C, ρ )-TYPE I FUNCTIONS Yadvendra SINGH, Amod KUMAR, B. B. UPADHYAY and Vinay SINGH Department of Mathematics Faculty of Science Banaras Hindu University,Varanasi, India ysinghze@gmail.com Received: January, 2013 / Accepted: March, 2013 Abstract: In this paper, a new class of (G,C, ρ )-type I functions and their generalizations are introduced. We consider a class of differentiable multiobjective optimization problems and establish sufficient optimality conditions. The results of the paper are more general than those existing in the literature. Keywords: Multiobjective programming, (G,C, ρ )-convexity, efficient solution, type I functions, generalized convexity. MSC: 90C46; 52A01 1. INTRODUCTION Convexity plays an important role in optimization theory as it extends the validity of a local solution of a minimization problem to a global one. But in several real world problems, the notion of convexity is no longer sufficient, which motivated the introduction of various generalizations of convex functions. It has been found that only a few properties of convex functions are needed for establishing sufficiency and duality theorems. Hanson 299 300 Yadvendra Singh et al. / On Sufficiency In Multiobjective [11] introduced the concept of differentiable invexity, which is a generalization of the concept of convexity. After the work of Hanson, other classes of differentiable nonconvex functions have been introduced to generalize the class of invex functions from different points of view, see the in [7-9, 12, 13, 16, 21, 22 ]. Later, Kaul and Kaur [14] presented strictly pseudoinvex, pseudoinvex and quasiinvex functions. Hanson and Mond [12] defined two new classes of functions called type I and type II functions. Rueda and Hanson [23] have introduced pseudo type I and and quasi type I functions. Mishra [24] studied a multiple objective nonlinear programming problem by combining the concepts of type-I, pseudo-type-I, quasi-type-I, quasi-pseudo-type-I, pseudo-quasi-type-I and univex functions. More details on type-I functions can be found in Ye [35], Suneja and Srivastava [25], Mishra et al. [26-30]. Other classes of generalized type I functions have been introduced in [2, 15]. In [17] and [18], Liang et al. introduced (F, α , ρ , d)-convexity, which is uniformulation of generalized convexity and an extension of (F, ρ )-convexity [22] and generalized (F, ρ )-convexity [8]. They obtained optimality conditions and duality results for the single objective fractional problems. Yuan et al. [32] introduced (C, α , ρ , d)-convexity, which is a generalization of (F, α , ρ , d)-convexity. Chinchuluun et al. [10] and Long [19] later studied multiobjective fractional programming problems in the framework of (C, α , ρ , d)- convexity. Antczak[4] extended further Hanson’s invexity to G-invexity for scalar differentiable functions. In the natural way, Antczak’s definition of G-invexity was also extended to the case of differentiable vector-valued functions in [6]. Motivated by [4-6, 33], we consider a class of differentiable multiobjective optimization problems. We introduce some new generalizations of (G,C, ρ )-convex functions and establish sufficient optimality conditions for the optimization problem. The results of the paper extend and unify some earlier results from the literature to a more general class of functions. 2. DEFINITIONS AND PRELIMINARIES In this section, we provide some definitions and some results which will we be used in the sequel. The following convention for vector in Rn will be adopted. x < y if and only if xi < yi , for i = 1, ..., n; x 5 y if and only if xi ≤ yi , for i = 1, ...n; x 6 y if and only if xi ≤ yi , for i = 1, ..., n, but x 6= y; We consider the following nonlinear multiobjective programming problem: (MOP) minimize f (x) := ( f1 (x), ..., f p (x)), subject to g(x) := (g1 (x), ..., gq (x)) 5 0, x ∈ X, Yadvendra Singh et al. / On Sufficiency In Multiobjective 301 where X is a nonempty open subset of Rn . Let A denote the set of all feasible points of (MOP) and f : X → R p , g : X → Rq are differentiable functions at x0 ∈ A. The index set P = {1, 2, ..., p} and Q = {1, 2, ..., q}. For x0 ∈ A the index set J(x0 ) = { j ∈ Q : g j (x0 ) = 0} and gJ denote the vector for active constraints. In the sequel, we need the following vector minimization problem: (G-MOP) minimize G f f (x) := (G f1 f1 (x), ..., G f p f p (x)), subject to Gg g(x) := (Gg1 g1 (x), ..., Ggq gq (x)) 5 Gg (0), x ∈ X, where G f : R → R p and Gg : R → Rq are vector valued functions. Definition 2.1. We say that x0 ∈ A is an efficient solution for problem (MOP) if and only if there exists no x ∈ A such that f (x) 6 f (x0 ), that is, fi (x) ≤ fi (x0 ) for all i ∈ P with strict inequality for at least one i ∈ P. Definition 2.2. We say that x0 ∈ A is a weak efficient solution for problem (MOP) if and only if there exists no x ∈ A such that f (x) < f (x0 ), that is, fi (x) < fi (x0 ) for all i ∈ P. Let X be a subset of Rn . For our convenience, an element of (n + 1)-dimensional Euclidean space Rn+1 is represented, in the following, as the ordered pair (τ , ρ ) with τ ∈ Rn and ρ ∈ R. Definition 2.3. A function C : X × X × Rn+1 → R is convex on Rn+1 with respect to the third argument if and only if, for any fixed (x, x0 ) ∈ X × X, the inequality C(x,x0 ) (λ z1 + (1 − λ )z2 ) ≤ λ C(x,x0 ) (z1 ) + (1 − λ )C(x,x0 ) (z2 ), ∀ λ ∈ (0, 1), holds for all z1 = (τ1 , ρ1 ) ∈ Rn+1 and z2 = (τ2 , ρ2 ) ∈ Rn+1 , where τ1 , τ2 ∈ Rn and ρ1 , ρ2 ∈ R. Definition 2.4. Let f = ( f1 , .., f p ) : X → R p be a vector-valued function defined on a nonempty set X ⊂ Rn , I fi (x), be the range of fi , i ∈ P. If there exist a vector-valued function G f = (G f1 , ...., G f p ) : R → R p such that any of its component G fi : I fi (X) → R is a strictly increasing function on its domain and G fi ( fi ) is a differentiable function on X, and real numbers ρi (i ∈ P) such that for any x ∈ X(x 6= x0 ), the inequality G fi ( fi (x)) − G fi ( fi (x0 )) ≥ (>)C(x,x0 ) (5(G fi ( fi ))(x0 ), ρi ), (2.1) holds for each i ∈ P, then f is said to be (strictly) (G f ,C, ρ )-convex at x0 ∈ X, where ρ = (ρ1 , .....ρ p )T . The function f is said to be (strictly) (G f ,C, ρ )-convex over X if, for all x0 ∈ X, it is (strictly) (G f ,C, ρ )-convex. In particular, f is said to be strong (strictly)(G f ,C, ρ )-convex or strictly (G f ,C, ρ )-convex with respect to ρ > 0 or ρ = 0, respectively. 302 Yadvendra Singh et al. / On Sufficiency In Multiobjective In order to define an analogous class of (strictly)(G f ,C, ρ )-incave functions, the function G fi of inequality in the above definition should be replaced by the function −G fi . That is, the inequality −(G fi ( fi (x)) − G fi ( fi (x0 ))) ≥ (>)C(x,x0 ) (− 5 (G fi ( fi ))(x0 ), ρi ), holds for i ∈ P. Remark 2.1. From the above definition, (C, α , ρ , d)-convexity defined in [32] is a special case of (G f ,C, ρ )-convexity whenever G f (t) = t, t ∈ R. Therefore, (F, α , ρ )-convexity [17, 18] and (F, ρ )-convexity [22] are a special case of (G f ,C, ρ )-convexity since any sublinear functional is also convex. Theorem 2.1. Let G fi (i ∈ P) be strictly increasing function defined on I fi (X), Gg j ( j ∈ Q) be strictly increasing function defined on Ig j (X). Further, let 0 ∈ Ig j (X). Then x0 is an efficient (weak) solution for (MOP) if and only if x0 is an efficient (weak) solution for (G-MOP). 3. SUFFICIENT OPTIMALITY CONDITIONS We assume throughout the paper that G f is a vector objective function and that Gg is the constraint vector in (G-MOP). The definition of type I for single objective and constraint vector function [12] can be generalized easily to a multiple objective and constraint vector. Throughout this paper, the following notation will be used ρ = (ρ 1 , ρ 2 ), where ρ 1 = (ρ1 1 , ..., ρ p 1 ) ∈ R p and ρ 2 = (ρ1 2 , ..., ρq 2 ) ∈ Rq . C(x,x0 ) (∇G f ( f (x)), ρ 1 ) := (C(x,x0 ) (∇G f1 ( f1 (x)), ρ11 ), ...,C(x,x0 ) (∇G f p ( f p (x)), ρ p1 )). C(x,x0 ) (∇Gg (g(x)), ρ 2 ) := (C(x,x0 ) (∇Gg1 (g1 (x)), ρ12 ), ...,C(x,x0 ) (∇Ggq (gq (x)), ρq2 )). We are now ready to present the new classes of functions. Definition 3.5. ( f , g) is said to be (G,C, ρ )-type I at x0 , if for all x ∈ A we have G f ( f (x)) − G f ( f (x0 )) = C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ), (3.1) −Gg (g(x0 )) = C(x,x0 ) (5(Gg (g))(x0 ), ρ 2 ). (3.2) Remark 3.2. Let G f (t) = t, t ∈ R. Then, the above definition is a generalization of (G,C, ρ )-convexity defined in [33] and (C, α , ρ , d)-type I convexity defined in [34]. Yadvendra Singh et al. / On Sufficiency In Multiobjective 303 Definition 3.6. ( f , g) is said to be pseudoquasi (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) < G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) < 0, (3.3) −Gg (g(x0 )) 5 0 ⇒ C(x,x0 ) (5(Gg (g))(x0 ), ρ 2 ) 5 0. (3.4) If in the above definition, inequality (3.3) is satisfied as G f ( f (x)) 5 G f f (x0 ) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) < 0, (3.5) then, we say that ( f , g) is strictly pseudoquasi (G,C, ρ )-type I at x0 . Definition 3.7. ( f , g) is said to be weak strictly-pseudoquasi (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) 6 G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) < 0, 2 −Gg (g(x0 )) ≤ 0 ⇒ C(x,x0 ) (5(Gg (g))(x0 ), ρ ) ≤ 0. (3.6) (3.7) Definition 3.8. ( f , g) is said to be stong-pseudoquasi (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) 6 G f f (x0 ) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) 6 0, (3.8) −Gg (g(x0 )) 5 0 ⇒ C(x,x0 ) (5(Gg (g(x0 )), ρ 2 ) 5 0. (3.9) If in the above definition, inequality (3.8) is satisfied as G f ( f (x)) < G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) 6 0, (3.10) then we say that ( f , g) is weak pseudoquasi (G,C, ρ )-type I at x0 . Definition 3.9. ( f , g) is said to be sub-strictly-pseudoquasi (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) 5 G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) 6 0, (3.11) −Gg (g(x0 )) ≤ 0 ⇒ C(x,x0 ) (5(Gg (g))(x0 ), ρ 2 ) 5 0. (3.12) Definition 3.10. ( f , g) is said to be weak quasistrictly-pseudo (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) 6 G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) ≤ 0, (3.13) −Gg (g(x0 )) 5 0 ⇒ C(x,x0 ) (5(Gg (g))(x0 ), ρ 2 ) 6 0. (3.14) Definition 3.11. ( f , g) is said to be weak quasisemi-pseudo (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) 6 G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 )), ρ 1 ) ≤ 0, (3.15) −Gg (g(x0 )) ≤ 0 ⇒ C(x,x0 ) (5(Gg (g(x0 ), ρ 2 ) < 0. (3.16) 304 Yadvendra Singh et al. / On Sufficiency In Multiobjective Definition 3.12. ( f , g) is said to be weak strictly-pseudo (G,C, ρ )-type I at x0 , if for all x ∈ A, we have G f ( f (x)) 6 G f ( f (x0 )) ⇒ C(x,x0 ) (5(G f ( f ))(x0 ), ρ 1 ) < 0, (3.17) −Gg (g(x0 )) 5 0 ⇒ C(x,x0 ) (5(Gg (g))(x0 )), ρ 2 ) < 0. (3.18) Now, we establish a sufficient optimality condition for a feasible point to be an efficient solution for (G-MOP). Theorem 3.2. Let x0 be a feasible solution for (G-MOP), and let there exist vector u ∈ R p and vector v ∈ Rq such that p ∑ ui ∇(G fi ( fi ))(x0 ) + ∑ i=1 v j ∇(Gg j (g j ))(x0 ) = 0, (3.19) j∈J(x0 ) v j Gg j g j (x0 ) = 0, ∀ j ∈ Q, (3.20) v j ≥ 0, ∀ j ∈ Q, (3.21) ui > 0, ∀i ∈ P. (3.22) If ( f , gJ ) is strong pseudoquasi (G,C, ρ )-type I at x0 such that p ∑ ui ρi1 + ∑ i=1 v j ρ 2j ≥ 0, (3.23) j∈J(x0 ) C(x,x0 ) (0, r) < 0 ⇒ r < 0, ∀x ∈ X, (3.24) then, x0 is an efficient solution for (G-MOP). Proof : Suppose that x0 is not an efficient solution for (G-MOP). Then, there exists x ∈ A, such that G fi ( fi (x)) ≤ G fi ( fi (x0 )), ∀i ∈ P, (3.25) with strict inequality for at least one i ∈ P. Also, g j (x0 ) = 0, ∀ j ∈ J(x0 ). (3.26) Since ( f , gJ ) is strong pseudoquasi (G,C, ρ )-type I at x0 , from (3.25) and (3.26), it follows that C(x,x0 ) (5(G fi ( fi ))(x0 ), ρi1 ) ≤ 0, ∀i ∈ P, (3.27) with strict inequality for at least one i ∈ P, and C(x,x0 ) (5(Gg j (g j ))(x0 ), ρ j 2 ) ≤ 0, ∀ j ∈ J(x0 ). (3.28) Yadvendra Singh et al. / On Sufficiency In Multiobjective Let p τ = ∑ ui + i=1 ∑ 305 v j. j∈J(x0 ) Multiplying (3.27) and (3.28) with τ1 ui and τ1 v j , respectively, and then adding the inequalities, we have p 1 ∑ τ uiC(x,x0 ) (5(G fi ( fi ))(x0 ), ρi 1 ) + ∑ i=1 j∈J(x0 1 v jC(x,x0 ) (5(Gg j (g j ))(x0 ), ρ j 2 ) < 0. τ )) Using the convexity of C, we get p 1 p C(x,x0 ) ( ( ∑ ui 5 (G fi ( fi ))(x0 ) + ∑ v j 5 (Gg j (g j ))(x0 )), ∑ ui ρi1 + ∑ v j ρ 2j ) < 0. τ i=1 i=1 j∈J(x ) j∈J(x ) 0 0 From (3.19), it follows that p C(x,x0 ) (0, ∑ ui ρi1 + i=1 ∑ v j ρ 2j ) < 0. j∈J(x0 ) Therefore, from (3.24), it follows that p ∑ ui ρi1 + ∑ i=1 v j ρ 2j < 0. j∈J(x0 ) Which is a contradiction to (3.23). Hence, x0 is an efficient solution for (G-MOP). This complete the proof. We can weaken the strict inequality requirement that ui > 0 for all i ∈ P in the above theorem, but we require different convexity conditions on ( f , gJ ). This is given by the following two theorems. Theorem 3.3. Let x0 be a feasible solution for (G-MOP) and let there exist vector u ∈ R p and vector v ∈ Rq such that p ∑ ui ∇(G fi ( fi ))(x0 ) + ∑ i=1 v j ∇(Gg j (g j ))(x0 ) = 0, (3.29) j∈J(x0 ) v j Gg j g j (x0 ) = 0, ∀ j ∈ Q, (3.30) v j ≥ 0, ∀ j ∈ Q, (3.31) ui ≥ 0, ∀i ∈ P, (3.32) with strict inequality for at least one i ∈ P. If ( f , gJ ) is weak strictly pseudoquasi (G,C, ρ )-type I at x0 such that p ∑ ui ρi1 + ∑ i=1 j∈J(x0 ) v j ρ 2j ≥ 0, (3.33) 306 Yadvendra Singh et al. / On Sufficiency In Multiobjective C(x,x0 ) (0, r) < 0 ⇒ r < 0, ∀x ∈ X, (3.34) then, x0 is an efficient solution for (G-MOP). Proof: Suppose that x0 is not an efficient solution for (G-MOP). Then, there exists x ∈ A, such that G fi fi (x) ≤ G fi ( fi (x0 )), ∀i ∈ P, (3.35) with strict inequality for at least one i ∈ P. Also g j (x0 ) = 0, ∀ j ∈ J(x0 ). (3.36) Since ( f , gJ ) is weak strictly pseudoquasi (G,C, ρ )-type I at x0 , from (3.35) and (3.36), it follows that C(x,x0 ) (5(G fi ( fi ))(x0 ), ρi1 ) < 0, ∀i ∈ P, (3.37) C(x,x0 ) (5(Gg j (g j ))(x0 ), ρ j 2 ) ≤ 0, ∀ j ∈ J(x0 ). (3.38) Now, the proof is similar to that of Theorem 3.1. Theorem 3.4. Let x0 be a feasible solution for (G-MOP) and let there exist vector u ∈ R p and vector v ∈ Rq such that p ∑ ui ∇(G fi ( fi ))(x0 ) + ∑ i=1 v j ∇(Gg j (g j ))(x0 ) = 0, (3.39) j∈J(x0 ) v j Gg j (g j (x0 )) = 0, ∀ j ∈ Q, (3.40) v j ≥ 0, ∀ j ∈ J(x0 ), (3.41) (ui , v j ) ≥ 0, ∀i ∈ P, ∀ j ∈ Q. (3.42) If ( f , gJ ) is weak quasistrictly pseudo (G,C, ρ )-type I at x0 such that p ∑ ui ρi1 + ∑ i=1 v j ρ 2j ≥ 0, (3.43) j∈J(x0 ) C(x,x0 ) (0, r) < 0 ⇒ r < 0, ∀x ∈ X, (3.44) then, x0 is an efficient solution for (G-MOP). Proof: Suppose that x0 is not an efficient solution for (G-MOP). Then, there exists x ∈ A, such that G fi fi (x) ≤ G fi fi (x0 ), ∀i ∈ P, (3.45) Yadvendra Singh et al. / On Sufficiency In Multiobjective 307 with strict inequality for at least one i ∈ P. Also g j (x0 ) = 0, ∀ j ∈ J(x0 ) (3.46) Since ( f , gJ ) is weak quasistrictly pseudo (G,C, ρ )-type I at x0 , from (3.45) and (3.46), it follows that C(x,x0 ) (5(G fi ( fi ))(x0 ), ρi1 ) ≤ 0, ∀i ∈ P, (3.47) C(x,x0 ) (5(Gg j (g j ))(x0 ), ρ j 2 ) ≤ 0, ∀ j ∈ J(x0 ), (3.48) with strict inequality for at least one j ∈ J(x0 ). Now, the proof is similar to that of Theorem 3.1. Remark 3.3. Similarly, we can prove more results like Theorem 3.1-3.3 by varying the convexity condition on ( f , gJ ) and changing the sign of u and v. It is obvious that the Theorem 3.1 and 3.2 hold for weak efficient solutions too. However, it is important to know that convexity assumptions of Theorem 3.1 and 3.2 can be weakened for weak efficient solutions. Theorem 3.5. Let x0 be a feasible solution for (G-MOP) and let there exist vector u ∈ R p and vector v ∈ Rq such that the triple (x, u, v) satisfies system (3.19-3.22) of the Theorem 3.2. If ( f , gJ ) is weak pseudoquasi (G,C, ρ )-type I at x0 such that p ∑ ui ρi1 + ∑ i=1 v j ρ 2j ≥ 0, (3.49) j∈J(x0 ) C(x,x0 ) (0, r) < 0 ⇒ r < 0, ∀x ∈ X, (3.50) then, x0 is a weak efficient solution for (G-MOP). Proof: Suppose that x0 is not a weak efficient solution for (G-MOP). Then, there exists x ∈ A, such that G fi fi (x) < G fi fi (x0 ), ∀i ∈ P, (3.51) with strict inequality for at least one i ∈ P. Also, g j (x0 ) = 0, ∀ j ∈ J(x0 ). (3.52) Since ( f , gJ ) is weak pseudoquasi (G,C, ρ )-type I at x0 , from (3.51)and(3.52), it follows that C(x,x0 ) (5(G fi ( fi (x0 )), ρi1 ) ≤ 0, ∀i ∈ P, (3.53) with strict inequality for at least one i ∈ P, C(x,x0 ) (5(Gg j (g j (x0 )), ρ j 2 ) ≤ 0∀ j ∈ J(x0 ). Now, the proof is similar to that of Theorem 3.1. (3.54) 308 Yadvendra Singh et al. / On Sufficiency In Multiobjective Theorem 3.6. Let x0 be a feasible solution for (G-MOP) and let there exist vector u ∈ R p and vector v ∈ Rq such that the triple (x, u, v) satisfies system (3.19-3.22) of the Theorem 3.2. If ( f , gJ ) is pseudoquasi (G,C, ρ )-type I at x0 such that p ∑ ui ρi1 + ∑ i=1 v j ρ 2j ≥ 0, (3.55) j∈J(x0 ) C(x,x0 ) (0, r) < 0 ⇒ r < 0, ∀x ∈ X, (3.56) then, x0 is a weak efficient solution for (G-MOP). Proof: Suppose that x0 is not a weak efficient solution for (G-MOP). Then, there exists x ∈ A, such that G fi fi (x) < G fi fi (x0 ), ∀i ∈ P, (3.57) with strict inequality for at least one i ∈ P. Also, g j (x0 ) = 0, ∀ j ∈ J(x0 ). (3.58) Since ( f , gJ ) is pseudoquasi (G,C, ρ )-type I at x0 , from (3.57) and (3.58), it follows that C(x,x0 ) (5(G fi ( fi (x0 )), ρi1 ) < 0, ∀i ∈ P, (3.59) C(x,x0 ) (5(Gg j (g j (x0 )), ρ j 2 ) ≤ 0, ∀ j ∈ J(x0 ). (3.60) Now, the proof is similar to that of Theorem 3.2. Remark 3.4. The importance of the Theorems (3.4) and (3.5) lies in the fact that a similar result does not necessarily hold for efficient solutions. 4. CONCLUSION AND FUTURE DIRECTIONS In this paper, we have introduced a new class of (G,C, ρ )-type I function and their generalizations. For a class of differentiable multiobjective programming problems, we have established sufficient optimality conditions. The results of the paper may be utilized to formulate Mond-Weir and Wolfe type dual problems and establish duality theorems. Acknowledgement: The authors wish to express their deep gratitude to Dr. S. K. Mishra for his motivation and guidance throughout the preparation of the paper. The First author is supported by the Council of Scientific and Industrial Research, New Delhi, India, through grant no. 09/013(0474)/2012-EMR-1.