This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

11institutetext: Danyang Liu
[email protected]
Rong Hu
[email protected]
Yaping Fang, Corresponding author
[email protected]
1College of Mathematics, Sichuan University, Chengdu, Sichuan, P.R. China
2College of Applied Mathematics, Chengdu University of Information Technology, Chengdu, Sichuan, P.R. China

Solvability of a Regular Polynomial Vector Optimization Problem without Convexitythanks: This work was partially supported by the National Natural Science Foundation of China (11471230).

Danyang Liu1    Rong Hu2    Yaping Fang1
Abstract

In this paper we consider the solvability of a non-convex regular polynomial vector optimization problem on a nonempty closed set. We introduce regularity conditions for the polynomial vector optimization problem and study properties and characterizations of the regularity conditions. Under the regularity conditions, we study nonemptiness and boundedness of the solution sets of the problem. As a consequence, we establish two Frank-Wolfe type theorems for the non-convex polynomial vector optimization problem. Finally, we investigate the solution stability of the non-convex regular polynomial vector optimization problem.

Keywords:
Polynomial vector optimization problem Regularity condition Frank-Wolfe type theorem Non-convexity Stability

Mathematics Subject Classification (2020) 90C29; 90C23; 49K40.

1 Introduction

Throughout, 𝐑n\mathbf{R}^{n} denotes the nn-dimensional Euclidean space with the norm \|\cdot\| and the inner ,\langle\cdot,\cdot\rangle, and 𝐑+n:={x=(x1,,xn)𝐑n:xi0,i=1,,n}\mathbf{R}^{n}_{+}:=\{x=(x_{1},\cdots,x_{n})\in\mathbf{R}^{n}:x_{i}\geq 0,i=1,\cdots,n\}. In this paper we are interested in the following polynomial vector optimization problem on KK:

PVOP(K,f):𝐑+sMinxKf(x),\text{PVOP}(K,f):\qquad\mathbf{R}^{s}_{+}-Min_{x\in K}f(x),

where f=(f1,,fs):𝐑n𝐑sf=(f_{1},\dots,f_{s}):\mathbf{R}^{n}\mapsto\mathbf{R}^{s} is a vector polynomial such that each component function fif_{i} is a polynomial with its degree  deg fi=di\mbox{ deg }f_{i}=d_{i}, and K𝐑nK\subseteq\mathbf{R}^{n} is a nonempty closed set (not necessarily convex set or semi-algebraic set HHV ; BR ).

Recall that a point xKx^{*}\in K is a Pareto efficient solution of PVOP(K,f)\text{PVOP}(K,f) if

f(x)f(x)𝐑+s\{0},xKf(x)-f(x^{*})\notin-\mathbf{R}^{s}_{+}\backslash\{0\},\quad\forall x\in K

and xKx^{*}\in K is a weak Pareto efficient solution of PVOP(K,f)(K,f) if

f(x)f(x) int 𝐑+s,xK.f(x)-f(x^{*})\notin-\mbox{ int }\mathbf{R}^{s}_{+},\quad\forall x\in K.

The Pareto efficient solution set and the weak Pareto efficient solution set of PVOP(K,f)\text{PVOP}(K,f) are denoted by SOLs(K,f)SOL^{s}(K,f) and SOLw(K,f)SOL^{w}(K,f) respectively. Clearly, SOLs(K,f)SOLw(K,f)SOL^{s}(K,f)\subseteq SOL^{w}(K,f). When s=1s=1, PVOP(K,f)\text{PVOP}(K,f) collapses to the polynomial scalar optimization problem:

PSOP(K,f):MinxKf(x),\mbox{PSOP}(K,f):\qquad\emph{Min}_{x\in K}f(x),

whose solution set is denoted by SOL(K,f)SOL(K,f).

For the polynomial scalar optimization problem, regularity condition has been used in AQ to investigate the nonemptiness of the solution set and the continuity of the solution mapping for a quadratic programming problem. Hieu HV1 established a Frank-Wolfe type theorem for a polynomial scalar optimization problem on a nonempty closed set by proving the nonemptiness and boundedness of its solution set when the objective function is bounded from below on the constraint set and the regularity condition holds. Hieu et al. HV2 proved that the solution set of an optimization problem corresponding to a polynomial complementarity problem is nonempty and compact by using the regularity condition of the polynomial complementarity problem. Kim et al. DTN proved the existence of Pareto efficient solutions of an unconstrained polynomial vector optimization problem when the Palais-Smale-type condition holds and the image of the objective function has a bounded section. When KK is a convex semi-algebraic set and ff is convex, Jiao et al. LGJ proved that PVOP(K,f)\text{PVOP}(K,f) has a Pareto efficient solution if and only if the image f(K)f(K) of ff has a nonempty bounded section. Inspired by the above works, in this paper, we study the nonemptiness and boundedness of the solution sets of PVOP(K,f)\text{PVOP}(K,f) without assuming any convexity of the objective function.

The rest of this paper is organized as follows: In Section 2, we present some notations and preliminary results. In Section 3, we give some sufficient and necessary conditions for regularity conditions. In Section 4, we discuss local properties of the regularity conditions. Section 5 is devoted to the study of solvability of PVOP(K,f)\text{PVOP}(K,f) under regularity conditions. In Section 6, we discuss the solution stability under the regularity conditions. Finally, we makes a concluding remark in Section 7.

2 Preliminaries

In this section, we recall some concepts and results that will be used in this paper. A nonempty subset CC of 𝐑n\mathbf{R}^{n} is called a cone if txCtx\in C for any xCx\in C and any t>0t>0. Given a nonempty closed set K𝐑nK\subset\mathbf{R}^{n}, the asymptotic cone KK_{\infty} of KK is defined by

K={v𝐑n:thereexisttk+andxkKsuchthatlimk+xktk=v}.K_{\infty}=\{v\in\mathbf{R}^{n}:\mbox{there}\ \mbox{exist}\ t_{k}\rightarrow+\infty\ \mbox{and}\ x_{k}\in K\ \mbox{such}\ \mbox{that}\ \lim_{k\rightarrow+\infty}\frac{x_{k}}{t_{k}}=v\}.

As known, KK_{\infty} is a closed cone and (K)=K(K_{\infty})_{\infty}=K_{\infty}, and KK is bounded if and only if K={0}K_{\infty}=\{0\}. If KK is a convex set, then KK_{\infty} is also a convex cone and K=RecKK_{\infty}=\mbox{Rec}\ K, where RecK\mbox{Rec}\ K is the recession cone of KK defined by

RecK={v𝐑n:x+tvK,xK,t0}.\mbox{Rec}\ K=\{v\in\mathbf{R}^{n}:x+tv\in K,\,\forall x\in K,\forall t\geq 0\}.

It is known that K+K=KK+K_{\infty}=K. The above results can been found in RC ; AA .

Definition 1

Let p=(p1,,ps):𝐑n𝐑sp=(p_{1},\dots,p_{s}):\mathbf{R}^{n}\mapsto\mathbf{R}^{s} be a vector polynomial with  deg pi=di\mbox{ deg }p_{i}=d_{i}, i=1,,si=1,\dots,s. We say that pp^{\infty} is the vector recession polynomial (or the vector leading term) of pp, where

p(x)=(p1(x),p2(x),,ps(x)) and pi(x)=limλ+pi(λx)λdi,x𝐑n.p^{\infty}(x)=(p^{\infty}_{1}(x),p^{\infty}_{2}(x),\dots,p^{\infty}_{s}(x))\emph{ and }p^{\infty}_{i}(x)=\lim_{\lambda\to+\infty}\frac{p_{i}(\lambda x)}{\lambda^{d_{i}}},\quad\forall x\in\mathbf{R}^{n}.
Remark 1

When s=1s=1, pp^{\infty} is a recession polynomial of pp (see HV1 ).

Definition 2

(See e.g. BT1 ; BT2 ; DTN ) A function h:𝐑n𝐑sh:\mathbf{R}^{n}\mapsto\mathbf{R}^{s} is said to be bounded from below on KK if there exists r𝐑sr\in\mathbf{R}^{s} such that h(x)r+𝐑+sh(x)\in r+\mathbf{R}^{s}_{+} for all xKx\in K. Clearly, hh is bounded from below on KK if and only if its component hih_{i} is bounded from below on KK for all ii.

Definition 3

We say that  PVOP (K,f)\text{ PVOP }(K,f) is weakly regular (resp. strongly regular) if SOLs(K,f)SOL^{s}(K_{\infty},f^{\infty}) (resp. SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty})) is bounded.

Remark 2

Clearly, strong regularity implies weak regularity. When s=1s=1, both weak regularity and strong regularity coincide with the regularity in (HV1, , Definition 2.1).

The following result plays an important role in establishing the existence of the Pareto efficient solutions of PVOP(K,F)\text{PVOP}(K,F).

Lemma 1

(MG, , Proposition 13) Given λint𝐑+s\lambda\in\mathrm{int}\ \mathbf{R}^{s}_{+} and x0Kx_{0}\in K, define g(x)=j=1sλjfj(x)g(x)=\sum^{s}_{j=1}\lambda_{j}f_{j}(x) and Gx0={xK:fi(x)fi(x0),i=1,2,,s}G_{x_{0}}=\{x\in K:f_{i}(x)\leq f_{i}(x_{0}),i=1,2,\dots,s\}. If xSOL(Gx0,g)x^{*}\in SOL(G_{x_{0}},g), then xSOLs(K,f)x^{*}\in SOL^{s}(K,f).

In what follows we always assume the each component polynomial fif_{i} of the objective function ff has a degree di1d_{i}\geq 1.

3 Regularity of PVOP(K,f)\text{PVOP}(K,f)

In this section, we shall discuss properties and characterizations of regularity of PVOP(K,f)\text{PVOP}(K,f).

3.1 Conditions for regularity

In this subsection we shall show that the regularity of PVOP(K,f)\text{PVOP}(K,f) is closely related to the regularity of PSOP(K,fi)\text{PSOP}(K,f_{i}). To do so, we first give a characterization of SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset.

Proposition 1

SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset if and only if 0SOLw(K,f)0\notin SOL^{w}(K_{\infty},f^{\infty}).

Proof We only need to prove the sufficiency. Suppose that 0SOLw(K,f)0\notin SOL^{w}(K_{\infty},f^{\infty}). Then there exists v0K\{0}v_{0}\in K_{\infty}\backslash\{0\} such that

f(v0)f(0)=f(v0) int 𝐑+s,f^{\infty}(v_{0})-f^{\infty}(0)=f^{\infty}(v_{0})\in-\mbox{ int }\mathbf{R}^{s}_{+},

which yields

fi(v0)<0,i=1,,s.f^{\infty}_{i}(v_{0})<0,\quad i=1,\dots,s.

Let vKv\in K_{\infty}. It follows that

fi(tv0)fi(v)=tdifi(v0)fi(v)<0,i=1,,sf^{\infty}_{i}(tv_{0})-f^{\infty}_{i}(v)=t^{d_{i}}f^{\infty}_{i}(v_{0})-f^{\infty}_{i}(v)<0,\quad i=1,\cdots,s

for all sufficiently large t>0t>0. Since vKv\in K_{\infty} is arbitrary, SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset. ∎

Example 1

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=x13x12x23x1+2x2+1,f2(x1,x2)=x22x1x2+x11f_{1}(x_{1},x_{2})=x^{3}_{1}-x^{2}_{1}x_{2}-3x_{1}+2x_{2}+1,f_{2}(x_{1},x_{2})=-x^{2}_{2}-x_{1}x_{2}+x_{1}-1

and

K={(x1,x2)𝐑2:x10,x2x10}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}\geq 0,x_{2}-x_{1}\geq 0\}.

It is easy to verify that K=KK=K_{\infty}, f1(x1,x2)=x13x12x2f^{\infty}_{1}(x_{1},x_{2})=x^{3}_{1}-x^{2}_{1}x_{2}, and f2(x1,x2)=x22x1x2f^{\infty}_{2}(x_{1},x_{2})=-x^{2}_{2}-x_{1}x_{2}. Then 0SOLw(K,f)0\notin SOL^{w}(K_{\infty},f^{\infty}) since f1(1,2)=1<0=f1(0,0)f^{\infty}_{1}(1,2)=-1<0=f^{\infty}_{1}(0,0) and f2(1,2)=6<0=f2(0,0)f^{\infty}_{2}(1,2)=-6<0=f^{\infty}_{2}(0,0). By Proposition 1, SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset.

Proposition 2

If SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty})\neq\emptyset, then SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty}) is a nonempty cone.

Proof Since SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty})\neq\emptyset, by Proposition 1, 0SOLw(K,f)0\in SOL^{w}(K_{\infty},f^{\infty}). Suppose on the contrary that there exist v0SOLw(K,f)v_{0}\in SOL^{w}(K_{\infty},f^{\infty}) and t0>0t_{0}>0 such that t0v0SOLw(K,f)t_{0}v_{0}\notin SOL^{w}(K_{\infty},f^{\infty}). Then there exists v1Kv_{1}\in K_{\infty} such that

fi(v1)fi(t0v0)<0,i=1,,s.f^{\infty}_{i}(v_{1})-f^{\infty}_{i}(t_{0}v_{0})<0,\quad i=1,\dots,s. (1)

Dividing the both sides of the inequality (1) by t0dit^{d_{i}}_{0}, we obtain fi(v1t0)fi(v0)<0f^{\infty}_{i}(\frac{v_{1}}{t_{0}})-f^{\infty}_{i}(v_{0})<0 for all i{1,,s}i\in\{1,\dots,s\}. This reaches a contradiction to v0SOLw(K,f)v_{0}\in SOL^{w}(K_{\infty},f^{\infty}) since v1t0K\frac{v_{1}}{t_{0}}\in K_{\infty}. Thus, SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty}) is a nonempty cone. ∎

Remark 3

By Proposition 2, PVOP(K,f)(K,f) is strongly regular if and only if SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty}) is empty or SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}.

Proposition 3

If SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, then fif_{i} is unbounded from below on KK for all i{1,,s}i\in\{1,\dots,s\}.

Proof Suppose on the contrary that there exists i0{1,2,,s}i_{0}\in\{1,2,\dots,s\} such that fi0f_{i_{0}} is bounded from below on KK. Then SOL(K,fi0)=SOL(K_{\infty},f^{\infty}_{i_{0}})=\emptyset since SOL(K,fi0)SOLw(K,f)SOL(K_{\infty},f^{\infty}_{i_{0}})\subseteq SOL^{w}(K_{\infty},f^{\infty}). By Lemma 1, there exists v0K\{0}v_{0}\in K_{\infty}\backslash\{0\} such that fi0(v0)<fi0(0)=0f^{\infty}_{i_{0}}(v_{0})<f^{\infty}_{i_{0}}(0)=0. Since v0K\{0}v_{0}\in K_{\infty}\backslash\{0\}, there exist tk>0t_{k}>0 with tk+t_{k}\rightarrow+\infty and xkKx_{k}\in K such that tk1xkv0t^{-1}_{k}x_{k}\rightarrow v_{0} as k+k\rightarrow+\infty. Since fi0f_{i_{0}} is bounded from below on KK, there exists a constant ll such that

fi0(xk)tkdi0ltkdi0.\frac{f_{i_{0}}(x_{k})}{t^{d_{i_{0}}}_{k}}\geq\frac{l}{t^{d_{i_{0}}}_{k}}.

Letting k+k\rightarrow+\infty, we obtain fi0(v0)0f^{\infty}_{i_{0}}(v_{0})\geq 0 which is a contradiction to fi0(v0)<0f^{\infty}_{i_{0}}(v_{0})<0. ∎

When fif_{i} is bounded from below on KK for some i{1,,s}i\in\{1,\cdots,s\}, by Proposition 3 and Proposition 1, we know that 0SOLw(K,f)0\in SOL^{w}(K_{\infty},f^{\infty}). In the following proposition, we further show 0SOLs(K,f)0\in SOL^{s}(K_{\infty},f^{\infty}) when each component fif_{i} is bounded from below on KK.

Proposition 4

If ff is bounded from below on KK, then 0SOLs(K,f)0\in SOL^{s}(K_{\infty},f^{\infty}).

Proof Suppose that ff is bounded from below on KK. Then fif_{i} is bounded from below on KK for all ii. By Proposition 3 and Proposition 1, fi(x)fi(0)=0f^{\infty}_{i}(x)\geq f^{\infty}_{i}(0)=0 for all xKx\in K_{\infty} and for all i{1,,s}i\in\{1,\dots,s\}. This means 0SOLs(K,f)0\in SOL^{s}(K_{\infty},f^{\infty}). ∎

The following result shows that the weak (strong) regularity of  PVOP (K,f)\text{ PVOP }(K,f) is closely related to the regularity of  PVOP (K,fi)\text{ PVOP }(K,f_{i}).

Theorem 3.1

(i) SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\} if and only if SOL(K,fi)={0}SOL(K_{\infty},f^{\infty}_{i})=\{0\} for all i{1,,s}i\in\{1,\dots,s\}.

(ii) If SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}, then SOL(K,fi)SOL(K_{\infty},f^{\infty}_{i})\neq\emptyset for all i{1,,s}i\in\{1,\dots,s\}.

Proof (i) Assume that SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}. Since

SOL(K,fi)SOLw(K,f),SOL(K_{\infty},f^{\infty}_{i})\subseteq SOL^{w}(K_{\infty},f^{\infty}),

it suffices to show SOL(K,fi)SOL(K_{\infty},f^{\infty}_{i})\neq\emptyset for all i{1,,s}i\in\{1,\dots,s\}. Suppose on the contrary that there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that SOL(K,fi0)=SOL(K_{\infty},f^{\infty}_{i_{0}})=\emptyset which is equivalent to 0SOL(K,fi0)0\notin SOL(K_{\infty},f^{\infty}_{i_{0}}) (by Lemma 1). Then there exists v0K\{0}v_{0}\in K_{\infty}\backslash\{0\} such that fi0(v0)<0f^{\infty}_{i_{0}}(v_{0})<0. Let λ int 𝐑+s\lambda\in\mbox{ int }\mathbf{R}^{s}_{+}. Consider the function

Fλ(x)=i=1sλifi(x) and Sv0={xK:f(x)f(v0)}.F_{\lambda}(x)=\sum^{s}_{i=1}\lambda_{i}f^{\infty}_{i}(x)\;\text{ and }\;S_{v_{0}}=\{x\in K_{\infty}:f^{\infty}(x)\leq f^{\infty}(v_{0})\}. (2)

Then Sv0S_{v_{0}} is nonempty and closed. Next we shall show that Sv0S_{v_{0}} is bounded. If not, then there exists the sequence {xk}Sv0\{x_{k}\}\subseteq S_{v_{0}} such that xk+\|x_{k}\|\to+\infty as k+k\to+\infty. Without loss of generality, we assume that xk0\|x_{k}\|\neq 0 and xkxkxK\{0}\frac{x_{k}}{\|x_{k}\|}\to x^{*}\in K_{\infty}\backslash\{0\}. Since xkSv0x_{k}\in S_{v_{0}}, we have

fi(xk)fi(v0),i=1,,s.f^{\infty}_{i}(x_{k})\leq f^{\infty}_{i}(v_{0}),\quad\forall i=1,\dots,s.

Dividing the both sides of the above inequality by xkdi\|x_{k}\|^{d_{i}} and letting k+k\rightarrow+\infty, we get

fi(x)0=fi(0),i=1,,s,f^{\infty}_{i}(x^{*})\leq 0=f^{\infty}_{i}(0),\quad\forall i=1,\dots,s,

which together with 0SOLw(K,f)0\in SOL^{w}(K_{\infty},f^{\infty}) yields xSOLw(K,f)x^{*}\in SOL^{w}(K_{\infty},f^{\infty}), a contradiction. Thus, Sv0S_{v_{0}} is bounded. By the known Weierstrass’ theorem, we have SOL(Sv0,Fλ)SOL(S_{v_{0}},F_{\lambda})\neq\emptyset. It follows from Lemma 1 that

SOL(Sv0,Fλ)SOLs(K,f)SOLw(K,f)={0},\emptyset\neq SOL(S_{v_{0}},F_{\lambda})\subseteq SOL^{s}(K_{\infty},f^{\infty})\subseteq SOL^{w}(K_{\infty},f^{\infty})=\{0\},

which implies SOL(Sv0,Fλ)={0}SOL(S_{v_{0}},F_{\lambda})=\{0\}, and so 0Sv00\in S_{v_{0}}. By the definition of Sv0S_{v_{0}}, we get fi0(v0)0f^{\infty}_{i_{0}}(v_{0})\geq 0, a contradiction to fi0(v0)<0f^{\infty}_{i_{0}}(v_{0})<0. Therefore, SOL(K,fi)SOL(K_{\infty},f^{\infty}_{i})\neq\emptyset for all i{1,,s}i\in\{1,\dots,s\}.

For the converse, assume that SOL(K,fi)={0}SOL(K_{\infty},f^{\infty}_{i})=\{0\} for all i{1,,s}i\in\{1,\dots,s\} and there exists v0SOLw(K,f)\{0}v_{0}\in SOL^{w}(K_{\infty},f^{\infty})\backslash\{0\}. Then

f(0)f(v0) int 𝐑+s,f^{\infty}(0)-f^{\infty}(v_{0})\notin-\mbox{ int }\mathbf{R}^{s}_{+},

which implies that fi0(v0)fi0(0)f^{\infty}_{i_{0}}(v_{0})\leq f^{\infty}_{i_{0}}(0) for some i0{1,,s}i_{0}\in\{1,\dots,s\}. This reaches a contradiction to SOL(K,fi0)={0}SOL(K_{\infty},f^{\infty}_{i_{0}})=\{0\}.

(ii) Suppose on the contrary that there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that SOL(K,fi0)=SOL(K_{\infty},f^{\infty}_{i_{0}})=\emptyset. By Lemma 1, 0SOL(K,fi0)0\notin SOL(K_{\infty},f^{\infty}_{i_{0}}). Then there exists v0K\{0}v_{0}\in K_{\infty}\backslash\{0\} such that fi0(v0)<0f^{\infty}_{i_{0}}(v_{0})<0. Let λ int 𝐑+s\lambda\in\mbox{ int }\mathbf{R}^{s}_{+}. By similar arguments as in the proof of (i), we have

SOL(Sv0,Fλ)SOLs(K,f)={0},\emptyset\neq SOL(S_{v_{0}},F_{\lambda})\subseteq SOL^{s}(K_{\infty},f^{\infty})=\{0\},

where Sv0S_{v_{0}} and FλF_{\lambda} are defined as in (2). The rest is same as the one of (i), and so we omit it. ∎

Remark 4

The following example shows that SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\} does not imply SOL(K,fi)={0}SOL(K_{\infty},f^{\infty}_{i})=\{0\} for each i{1,,s}i\in\{1,\dots,s\}.

Example 2

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=x13x2+1,f2(x1,x2)=x23x11f_{1}(x_{1},x_{2})=x^{3}_{1}-x_{2}+1,f_{2}(x_{1},x_{2})=x^{3}_{2}-x_{1}-1

and the constraint set

K={(x1,x2)𝐑2:x10,x20}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}\geq 0,x_{2}\geq 0\}.

It is easy to verify that K=KK=K_{\infty}, f1(x1,x2)=x13f^{\infty}_{1}(x_{1},x_{2})=x^{3}_{1} and f2(x1,x2)=x23f^{\infty}_{2}(x_{1},x_{2})=x^{3}_{2}. Clearly, SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}. However, SOL(K,f1)={(x1,x2)𝐑2:x1=0,x20}{0}SOL(K_{\infty},f^{\infty}_{1})=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}=0,x_{2}\geq 0\}\neq\{0\} and SOL(K,f2)={(x1,x2)𝐑2:x10,x2=0}{0}SOL(K_{\infty},f^{\infty}_{2})=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}\geq 0,x_{2}=0\}\neq\{0\}.

Proposition 5

PVOP(K,f)\mathrm{PVOP}(K,f) is strongly regular and ff is bounded from below on KK if and only if SOL(K,fi)={0}SOL(K_{\infty},f^{\infty}_{i})=\{0\} for all i{1,,s}i\in\{1,\dots,s\}.

Proof The sufficiency follows immediately from Proposition 4 and Theorem 3.1 (i).

Next we prove the sufficiency. By Theorem 3.1(i),  PVOP (K,f)\text{ PVOP }(K,f) is strongly regular. Suppose on the contrary that there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that fi0f_{i_{0}} is not bounded from below on KK. Let xKx\in K. Then there exists {xk}k=1K\{x_{k}\}^{\infty}_{k=1}\subseteq K such that

fi0(xk)kfi0(x)f_{i_{0}}(x_{k})\leq-k\leq f_{i_{0}}(x) (3)

for all sufficiently large kk. We claim that {xk}k=1K\{x_{k}\}^{\infty}_{k=1}\subseteq K is unbounded. Indeed, if not, then we may assume that xk+\|x_{k}\|\to+\infty and xkxkvK\{0}\frac{x_{k}}{\|x_{k}\|}\to v\in K_{\infty}\backslash\{0\} as k+k\to+\infty. Dividing the both sides of (3) by xkdi\|x_{k}\|^{d_{i}} and letting k+k\rightarrow+\infty, we get fi(v)0f^{\infty}_{i}(v)\leq 0, a contradiction to SOL(K,fi0)={0}SOL(K_{\infty},f^{\infty}_{i_{0}})=\{0\}. Hence, {xk}k=1\{x_{k}\}^{\infty}_{k=1} is bounded. Without loss of generality, we may assume that xkxKx_{k}\to x^{*}\in K as k+k\to+\infty. It follows from (3) that

fi0(x)=limk+fi0(xk)limk+k=,f_{i_{0}}(x^{*})=\lim_{k\to+\infty}f_{i_{0}}(x_{k})\leq\lim_{k\to+\infty}-k=-\infty,

a contradiction. Thus, ff is bounded from below on KK. ∎

3.2 Regularity and R0R_{0}-property

It has been shown in Oett ; FYP ; FangH ; Gowda ; Lop that R0R_{0}-property plays an important role in studying the compactness of the solution sets of complementarity problems as well as the upper continuity of the solution mappings. In this subsection, we shall show that the weak (strong) regularity of  PVOP (K,f)\mbox{ PVOP }(K,f) is closely related to the R0R_{0}-property of the following weak vector complementarity problem CGY ; Yangxq

WVCP(K,f):FindxKsuch thatf(x)(K)𝐑+sw+andf(x),x int 𝐑+s,\mbox{WVCP}(K_{\infty},\nabla f^{\infty}):\ \mbox{Find}\ x\in K_{\infty}\,\mbox{such that}\,\nabla f^{\infty}(x)\in(K_{\infty})^{w+}_{\mathbf{R}^{s}_{+}}\,\mbox{and}\,\langle\nabla f^{\infty}(x),x\rangle\notin\mbox{ int }\mathbf{R}^{s}_{+},

where f=(f1,f2,,fs)\nabla f^{\infty}=(\nabla f^{\infty}_{1},\nabla f^{\infty}_{2},\dots,\nabla f^{\infty}_{s}) is the gradient of ff^{\infty},

(K)𝐑+sw+={vL(𝐑n,𝐑s):v,x int 𝐑+s,xK}(K_{\infty})^{w+}_{\mathbf{R}^{s}_{+}}=\{v\in L(\mathbf{R}^{n},\mathbf{R}^{s}):\langle v,x\rangle\notin-\mbox{ int }\mathbf{R}^{s}_{+},\,\forall x\in K_{\infty}\}

is the weak dual cone of KK_{\infty} with respect to 𝐑+s\mathbf{R}^{s}_{+}, and L(𝐑n,𝐑s)L(\mathbf{R}^{n},\mathbf{R}^{s}) is the space of all bounded linear operators from 𝐑n\mathbf{R}^{n} to 𝐑s\mathbf{R}^{s}. We denote the solution set of WVCP(K,f)\mbox{WVCP}(K_{\infty},\nabla f^{\infty}) by SOLWVCP(K,f)SOL_{WVCP}(K_{\infty},\nabla f^{\infty}). Recall that WVCP(K,f)\mbox{WVCP}(K_{\infty},\nabla f^{\infty}) is of type R0R_{0} FYP if SOLWVCP(K,f)={0}SOL_{WVCP}(K_{\infty},\nabla f^{\infty})=\{0\}.

Theorem 3.2

Assume that KK is convex. If SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, then WVCP(K,f)\mbox{WVCP}(K_{\infty},\nabla f^{\infty}) is of type R0R_{0}. Conversely, if WVCP(K,f)\mbox{WVCP}(K_{\infty},\nabla f^{\infty}) is of type R0R_{0}, then  PVOP (K,f)\mbox{ PVOP }(K,f) is strongly regular.

Proof It is clear that 0SOLWVCP(K,f)0\in SOL_{WVCP}(K_{\infty},\nabla f^{\infty}).

Suppose that SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}. By Theorem 3.1(i), we have

SOL(K,fi)={0},i=1,,s.SOL(K_{\infty},f_{i}^{\infty})=\{0\},i=1,\cdots,s.

This implies fi(v)>0f^{\infty}_{i}(v)>0 for all vK\{0},i=1,,sv\in K_{\infty}\backslash\{0\},i=1,\dots,s. This together with the Euler’s Homogeneous Function Theorem yields

f(v),v=(f1(v),v,,fs(v),v)=(d1f1(v),,dsfs(v)) int 𝐑+s\langle\nabla f^{\infty}(v),v\rangle=(\langle\nabla f^{\infty}_{1}(v),v\rangle,\dots,\langle\nabla f^{\infty}_{s}(v),v\rangle)=(d_{1}f^{\infty}_{1}(v),\dots,d_{s}f^{\infty}_{s}(v))\in\mbox{ int }\mathbf{R}^{s}_{+}

for all vK\{0}v\in K_{\infty}\backslash\{0\}, where did_{i} is the degree of fi,i=1,,sf_{i},i=1,\cdots,s. As a consequence, SOLWVCP(K,f)={0}SOL_{WVCP}(K_{\infty},\nabla f^{\infty})=\{0\} and so WVCP(K,f)\text{WVCP}(K_{\infty},\nabla f^{\infty}) is of type R0R_{0}.

For the converse, suppose WVCP(K,f)\mbox{WVCP}(K_{\infty},\nabla f^{\infty}) is of type R0R_{0}. To complete the proof, it suffices to show SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\} (by Remark 3). Suppose on the contrary that there exists vSOLw(K,f)\{0}v^{*}\in SOL^{w}(K_{\infty},f^{\infty})\backslash\{0\}. By (LKY, , Theorem 4), we get

f(v),xv int 𝐑+s,xK.\langle\nabla f^{\infty}(v^{*}),x-v^{*}\rangle\notin-\mbox{ int }\mathbf{R}^{s}_{+},\quad\forall x\in K_{\infty}.

This implies vSOLWVCP(K,f)v\in SOL_{WVCP}(K_{\infty},\nabla f^{\infty}) since KK_{\infty} is a closed convex cone. This reaches a contradiction. ∎

The following example illustrates the conclusion of Theorem 3.2.

Example 3

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with f1(x1,x2)=x12+x22,f2(x1,x2)=x22f_{1}(x_{1},x_{2})=x^{2}_{1}+x^{2}_{2},f_{2}(x_{1},x_{2})=x^{2}_{2} and

K={(x1,x2)𝐑2:x2x10}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{2}\geq x_{1}\geq 0\}.

Then K=KK=K_{\infty}, f1(x1,x2)=(2x1,2x2)\nabla f^{\infty}_{1}(x_{1},x_{2})=(2x_{1},2x_{2}), and f2(x1,x2)=(0,2x2)\nabla f^{\infty}_{2}(x_{1},x_{2})=(0,2x_{2}). It is easy to verify that 0SOLw(K,f)0\in SOL^{w}(K_{\infty},f^{\infty}). Let x=(x1,x2)Kx=(x_{1},x_{2})\in K_{\infty} be such that f(x),x int 𝐑+2\langle\nabla f^{\infty}(x),x\rangle\notin\mbox{ int }\mathbf{R}^{2}_{+}. It follows that (2x12+2x22,2x22) int 𝐑+2(2x^{2}_{1}+2x^{2}_{2},2x^{2}_{2})\notin\mbox{ int }\mathbf{R}^{2}_{+} This implies x=(0,0)x=(0,0). As a consequence, WVCP(K,f)\mbox{WVCP}(K_{\infty},\nabla f^{\infty}) is of type R0R_{0}. By Theorem 3.2, PVOP(K,f)\text{PVOP}(K,f) is strongly regular.

Remark 5

Assume that KK is convex and ff is bounded from below on KK. Then by Theorem 3.2 and Proposition 4, SOLWVCP(f,K)={0}SOLs(K,f)={0}SOL_{WVCP}(\nabla f^{\infty},K_{\infty})=\{0\}\Rightarrow SOL^{s}(K_{\infty},f^{\infty})=\{0\}. The following example shows that the converse is not true in general.

Example 4

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with f1(x1,x2)=x12,f2(x1,x2)=x22f_{1}(x_{1},x_{2})=x^{2}_{1},f_{2}(x_{1},x_{2})=x^{2}_{2} and

K={(x1,x2)𝐑2:x20,x10}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{2}\geq 0,x_{1}\geq 0\}.

Clearly, KK is convex, K=KK=K_{\infty}, and ff is bounded from below on KK. It is easy to verify that f1(x1,x2)=f1(x1,x2),f2(x1,x2)=f2(x1,x2)f_{1}(x_{1},x_{2})=f^{\infty}_{1}(x_{1},x_{2}),f_{2}(x_{1},x_{2})=f^{\infty}_{2}(x_{1},x_{2}), f1(x1,x2)=(2x1,0)\nabla f^{\infty}_{1}(x_{1},x_{2})=(2x_{1},0), f2(x1,x2)=(0,2x2)\nabla f^{\infty}_{2}(x_{1},x_{2})=(0,2x_{2}), and SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}. Let x0=(0,1)x_{0}=(0,1). Then f(x0),x0=(0,2) int 𝐑+s\langle\nabla f^{\infty}(x_{0}),x_{0}\rangle=(0,2)\notin\mbox{ int }\mathbf{R}^{s}_{+} and f(x0),w=(0,2w2)int𝐑+2\langle\nabla f^{\infty}(x_{0}),w\rangle=(0,2w_{2})\notin-\mbox{int}\mathbf{R}^{2}_{+} for all w=(w1,w2)Kw=(w_{1},w_{2})\in K_{\infty}. This means x0SOLWVCP(K,f)\{0}x_{0}\in SOL_{WVCP}(K_{\infty},\nabla f^{\infty})\backslash\{0\}, and so WVCP(K,f)\text{WVCP}(K_{\infty},\nabla f^{\infty}) is not of type R0R_{0}.

4 Local Properties of Regularity Conditions

In this section, we investigate local properties of (strong) weak regularity of  PVOP (K,f)\text{ PVOP }(K,f). Given an integer dd, in what follows, we always let 𝐏d\mathbf{P}_{d} denote the family of all polynomials of degree at most dd, and

Xdn(x):=(1,x1,,xn,x12,,xn2,,x1d,x1d1x2,x1d1x3,,x2d,x2d1x3,x2d1x3,,xnd),X^{n}_{d}(x):=(1,x_{1},\dots,x_{n},x^{2}_{1},\dots,x^{2}_{n},\dots,x^{d}_{1},x^{d-1}_{1}x_{2},x^{d-1}_{1}x_{3},\dots,x^{d}_{2},x^{d-1}_{2}x_{3},x^{d-1}_{2}x_{3},\dots,x^{d}_{n}),

whose components are listed by the lexicographic ordering. The dimension of 𝐏d\mathbf{P}_{d} is denoted by κd\kappa_{d}. Then, for each polynomial p𝐏dp\in\mathbf{P}_{d}, there exists a unique α𝐑κd\alpha\in\mathbf{R}^{\kappa_{d}} such that p(x)=α,Xdn(x)p(x)=\langle\alpha,X^{n}_{d}(x)\rangle. 𝐏d\mathbf{P}_{d} can be endowed with a norm p:=α=α12++ακd2\|p\|:=\|\alpha\|=\sqrt{\alpha^{2}_{1}+\dots+\alpha^{2}_{\kappa_{d}}}. Let pk𝐏dp^{k}\in\mathbf{P}_{d} with pkp𝐏dp^{k}\to p\in\mathbf{P}_{d} and xk𝐑nx^{k}\in\mathbf{R}^{n} with xkx𝐑nx^{k}\to x\in\mathbf{R}^{n}. It is easy to verify that (pk)p(p^{k})^{\infty}\to p^{\infty} and pk(xk)p(x)p^{k}(x^{k})\to p(x) as k+k\to+\infty.

Given 𝐝=(d1,,ds)𝐑s\mathbf{d}=(d_{1},\dots,d_{s})\in\mathbf{R}^{s} with did_{i} being an integer, i=1,,si=1,\cdots,s, let 𝐏𝐝=𝐏d1××𝐏ds\mathbf{P}_{\mathbf{d}}=\mathbf{P}_{d_{1}}\times\dots\times\mathbf{P}_{d_{s}}. Denoted by 𝐆𝐑w𝐝\mathbf{GR}^{\mathbf{d}}_{w} (resp. 𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s}) the family of all vector polynomials pp with  deg pi=di,i=1,,s\text{ deg }p_{i}=d_{i},i=1,\dots,s, such that  PVOP (K,p)\text{ PVOP }(K,p) is strongly (resp. weakly) regular. Then we have the following results.

Proposition 6

𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s} and 𝐆𝐑w𝐝\mathbf{GR}^{\mathbf{d}}_{w} are nonempty.

Proof We only need to prove that 𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s} is nonempty since 𝐆𝐑s𝐝𝐆𝐑w𝐝\mathbf{GR}^{\mathbf{d}}_{s}\subseteq\mathbf{GR}^{\mathbf{d}}_{w}. If KK is bounded, then K={0}K_{\infty}=\{0\}. In this case SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, and so PVOP(K,f)(K,f) is strongly regular for every vector polynomial ff. Suppose that KK is unbounded. Then there exists x=(x1,,xn)K\{0}x^{*}=(x^{*}_{1},\dots,x^{*}_{n})\in K_{\infty}\backslash\{0\}. Without loss of generality, we suppose that xi00x^{*}_{i_{0}}\neq 0. Consider the vector polynomial f=(f1,,fs):𝐑n𝐑sf=(f_{1},\dots,f_{s}):\mathbf{R}^{n}\mapsto\mathbf{R}^{s} with fi(x)=(xi0xi0)di,i=1,,sf_{i}(x)=-(x^{*}_{i_{0}}x_{i_{0}})^{d_{i}},i=1,\cdots,s. Then fi(x)f_{i}(x) is a polynomials ff of degree did_{i} and fi(tx)=(xi0)2ditdif_{i}(tx^{*})=-(x^{*}_{i_{0}})^{2d_{i}}t^{d_{i}}\to-\infty as t+t\to+\infty. As a consequence, SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, and so f𝐆𝐑s𝐝f\in\mathbf{GR}^{\mathbf{d}}_{s}. ∎

Proposition 7

𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s} is open in 𝐏𝐝\mathbf{P}_{\mathbf{d}}.

Proof We shall prove that 𝐏𝐝\𝐆𝐑s𝐝\mathbf{P}_{\mathbf{d}}\backslash\mathbf{GR}^{\mathbf{d}}_{s} is closed in 𝐏𝐝\mathbf{P}_{\mathbf{d}}. Let {fk}𝐏𝐝\𝐆𝐑s𝐝\{f^{k}\}\subseteq\mathbf{P}_{\mathbf{d}}\backslash\mathbf{GR}^{\mathbf{d}}_{s} with fk=(f1k,,fsk)f^{k}=(f^{k}_{1},\dots,f^{k}_{s}) such that fk=(f1k,,fsk)f=(f1,,fs)f^{k}=(f^{k}_{1},\dots,f^{k}_{s})\rightarrow f=(f_{1},\dots,f_{s}) as k+k\to+\infty. We can suppose that deg fi=di\mbox{deg }f_{i}=d_{i} for all i{1,2,,s}i\in\{1,2,\dots,s\} since f𝐆𝐑s𝐝f\notin\mathbf{GR}^{\mathbf{d}}_{s} when  deg fi0<di0\mbox{ deg }f_{i_{0}}<d_{i_{0}} for some i0{1,,s}i_{0}\in\{1,\dots,s\}, where did_{i} is the ii-th component of 𝐝\mathbf{d}. Since SOLw(K,(fk))SOL^{w}(K_{\infty},(f_{k})^{\infty}) is unbounded for all kk, there exists xkSOLw(K,(fk))x_{k}\in SOL^{w}(K_{\infty},(f_{k})^{\infty}) such that xk+\|x_{k}\|\to+\infty. Without loss of generality, we assume that xkxkxK\{0}\frac{x_{k}}{\|x_{k}\|}\to x^{*}\in K_{\infty}\backslash\{0\}. We claim that xSOLw(K,f)x^{*}\in SOL^{w}(K_{\infty},f^{\infty}). Indeed, if not, then there exists vKv\in K_{\infty} such that

fi(v)<fi(x),i=1,,s.f^{\infty}_{i}(v)<f^{\infty}_{i}(x^{*}),\quad i=1,\cdots,s. (4)

Since xkSOLw(K,(fk))x_{k}\in SOL^{w}(K_{\infty},(f_{k})^{\infty}), we have

(fk)(xkv)(fk)(xk) int 𝐑+s.(f_{k})^{\infty}(\|x_{k}\|v)-(f_{k})^{\infty}(x_{k})\notin-\mbox{ int }\mathbf{R}^{s}_{+}.

Then for each kk, there exists ik,v{1,,s}i_{k,v}\in\{1,\dots,s\} such that

(fik,vk)(xkv)(fik,vk)(xk)0.(f^{k}_{i_{k,v}})^{\infty}(\|x_{k}\|v)-(f^{k}_{i_{k,v}})^{\infty}(x_{k})\geq 0.

Since the set {1,2,,s}\{1,2,\dots,s\} is finite, without loss of generality, we suppose that there exists i0,v{1,,s}i_{0,v}\in\{1,\dots,s\} such that

(fi0,vk)(xkv)(fi0,vk)(xk)0,k.(f^{k}_{i_{0,v}})^{\infty}(\|x_{k}\|v)-(f^{k}_{i_{0,v}})^{\infty}(x_{k})\geq 0,\quad\forall k.

Since (fik)fi(f^{k}_{i})^{\infty}\to f^{\infty}_{i} as k+k\to+\infty, dividing the both sides of the above inequality by xkdi0,v\|x_{k}\|^{d_{i_{0,v}}} and letting k+k\to+\infty, we get

fi0,v(v)fi0,v(x).f^{\infty}_{i_{0,v}}(v)\geq f^{\infty}_{i_{0,v}}(x^{*}).

This reaches a contradiction to (4)(\ref{fypa1}), and so xSOLw(K,f)\{0}x^{*}\in SOL^{w}(K_{\infty},f^{\infty})\backslash\{0\}. By Proposition 2, SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty}) is unbounded. Hence, 𝐏𝐝\𝐆𝐑s𝐝\mathbf{P}_{\mathbf{d}}\backslash\mathbf{GR}^{\mathbf{d}}_{s} is closed. ∎

Remark 6

When s=1s=1, Proposition 7 reduces to (HV1, , Lemma 4.1).

In the following result, we shall show that strong regularity of a polynomial vector optimization problem remains stable under a small perturbation.

Theorem 4.1

The following conclusions hold:

(i) If SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, then there exists ϵ>0\epsilon>0 such that SOLw(K,g)={0}SOL^{w}(K_{\infty},g^{\infty})=\{0\} for all g𝐏𝐝g\in\mathbf{P}_{\mathbf{d}} satisfying gf<ϵ\|g-f\|<\epsilon;

(ii) If SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, then there exists ϵ>0\epsilon>0 such that SOLw(K,g)=SOL^{w}(K_{\infty},g^{\infty})=\emptyset for all g𝐏𝐝g\in\mathbf{P}_{\mathbf{d}} satisfying gf<ϵ\|g-f\|<\epsilon.

Proof Since 𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s} is open in 𝐏𝐝\mathbf{P}_{\mathbf{d}} (by Proposition 7) and f𝐆𝐑s𝐝f\in\mathbf{GR}^{\mathbf{d}}_{s}, there exists an open ball 𝐁(f,δ)𝐆𝐑s𝐝\mathbf{B}(f,\delta)\subseteq\mathbf{GR}^{\mathbf{d}}_{s} such that either SOL(K,g)={0}SOL(K_{\infty},g^{\infty})=\{0\} or SOL(K,g)=SOL(K_{\infty},g^{\infty})=\emptyset for all g𝐁(f,δ)g\in\mathbf{B}(f,\delta).

(i) It suffices to show that there exists ϵ(0,δ)\epsilon\in(0,\delta) such that SOL(K,g)={0}SOL(K_{\infty},g^{\infty})=\{0\} for all g𝐁(f,ϵ)g\in\mathbf{B}(f,\epsilon) when SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}. Suppose on the contrary that for any ϵ(0,δ)\epsilon\in(0,\delta), there exists gϵ𝐏𝐝g^{\epsilon}\in\mathbf{P}_{\mathbf{d}} with gϵf<ϵ\|g^{\epsilon}-f\|<\epsilon such that SOLw(K,(gϵ))=SOL^{w}(K_{\infty},(g^{\epsilon})^{\infty})=\emptyset. By Lemma 1, there exists xϵK\{0}x_{\epsilon}\in K_{\infty}\backslash\{0\} such that

(giϵ)(xϵ)<(giϵ)(0)=0,i=1,,s.(g_{i}^{\epsilon})^{\infty}(x_{\epsilon})<(g_{i}^{\epsilon})^{\infty}(0)=0,\quad i=1,\cdots,s. (5)

Since gϵfg^{\epsilon}\to f as ϵ0\epsilon\to 0, we have (gϵ)f(g^{\epsilon})^{\infty}\to f^{\infty} as ϵ0\epsilon\to 0. Without loss of generality, we assume that xϵxϵxK\{0}\frac{x_{\epsilon}}{\|x_{\epsilon}\|}\to x^{*}\in K_{\infty}\backslash\{0\} as ϵ0\epsilon\to 0. Since gϵ𝐁(f,ϵ)𝐆𝐑s𝐝g^{\epsilon}\in\mathbf{B}(f,\epsilon)\subset\mathbf{GR}^{\mathbf{d}}_{s}, we get  deg (giϵ)=di\mbox{ deg }(g_{i}^{\epsilon})^{\infty}=d_{i}. Dividing the both sides of (5) by xϵdi\|x_{\epsilon}\|^{d_{i}} and letting ϵ0\epsilon\rightarrow 0, we get

fi(x)0,i=1,,s,f_{i}^{\infty}(x^{*})\leq 0,\quad i=1,\cdots,s,

which reaches a contradiction to SOL(K,f)={0}SOL(K_{\infty},f^{\infty})=\{0\}.

(ii) It suffices to show that there exists ϵ(0,δ)\epsilon\in(0,\delta) such that SOL(K,g)=SOL(K_{\infty},g^{\infty})=\emptyset for all g𝐁(f,ϵ)g\in\mathbf{B}(f,\epsilon) when SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset. Suppose on the contrary that for any ϵ(0,δ)\epsilon\in(0,\delta), there exists gϵ𝐏𝐝g^{\epsilon}\in\mathbf{P}_{\mathbf{d}} with gϵf<ϵ\|g^{\epsilon}-f\|<\epsilon such that SOLw(K,(gϵ))={0}SOL^{w}(K_{\infty},(g^{\epsilon})^{\infty})=\{0\}. By (i) of Theorem 3.1, we have

0=(giϵ)(0)(giϵ)(x),i=1,,s0=(g_{i}^{\epsilon})^{\infty}(0)\leq(g_{i}^{\epsilon})^{\infty}(x),\quad i=1,\cdots,s

for any xKx\in K_{\infty}. Since gϵfg^{\epsilon}\to f as ϵ0\epsilon\to 0, we have (gϵ)f(g^{\epsilon})^{\infty}\to f^{\infty} as ϵ0\epsilon\to 0. Letting ϵ0\epsilon\rightarrow 0 in the above inequality, we get

fi(0)fi(x),i=1,,s.f_{i}^{\infty}(0)\leq f_{i}^{\infty}(x),\quad i=1,\cdots,s.

Since xKx\in K_{\infty} is arbitrary, again from (i) of Theorem 3.1 we get 0SOL(K,f)0\in SOL(K_{\infty},f^{\infty}), a contradiction. ∎

Observe that f=(f+g)f^{\infty}=(f+g)^{\infty} for all g=(g1,,gs)𝐏𝐝g=(g_{1},\dots,g_{s})\in\mathbf{P}_{\mathbf{d}} with  deg gi< deg fi\text{ deg }g_{i}<\text{ deg }f_{i}, i=1,,si=1,\dots,s. As a consequence, we have the following result.

Proposition 8

Let f=(f1,,fs):𝐑n𝐑sf=(f_{1},\cdots,f_{s}):\mathbf{R}^{n}\mapsto\mathbf{R}^{s} be a vector polynomial. Then for any vector polynomial g=(g1,,gs)g=(g_{1},\dots,g_{s}) with deggi<degfi,i=1,,s\mathrm{deg}\ g_{i}<\mathrm{deg}\ f_{i},i=1,\cdots,s, the following conclusions hold:

  • (i)

    If SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}, then SOLs(K,(f+g))={0}SOL^{s}(K_{\infty},(f+g)^{\infty})=\{0\}.

  • (ii)

    If SOLs(K,f)=SOL^{s}(K_{\infty},f^{\infty})=\emptyset, then SOLs(K,(f+g))=SOL^{s}(K_{\infty},(f+g)^{\infty})=\emptyset.

  • (iii)

    If SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, then SOLw(K,(f+g))={0}SOL^{w}(K_{\infty},(f+g)^{\infty})=\{0\}.

  • (iv)

    If SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, then SOLw(K,(f+g))=SOL^{w}(K_{\infty},(f+g)^{\infty})=\emptyset.

The following result is a direct consequence of Theorem 4.1 and Proposition 8.

Corollary 1

Let f=(f1,,fs):𝐑n𝐑sf=(f_{1},\cdots,f_{s}):\mathbf{R}^{n}\mapsto\mathbf{R}^{s} be a vector polynomial. Then for any vector polynomial g=(g1,,gs)g=(g_{1},\dots,g_{s}) with deggi<degfi,i=1,,s\mathrm{deg}\ g_{i}<\mathrm{deg}\ f_{i},i=1,\cdots,s, the following conclusions hold:

  • (i)

    If PVOP(K,f)\mathrm{PVOP}(K,f) is weakly regular, then PVOP(K,f+g)\mathrm{PVOP}(K,f+g) is weakly regular.

  • (ii)

    If PVOP(K,f)\mathrm{PVOP}(K,f) is strongly regular, then PVOP(K,f+g)\mathrm{PVOP}(K,f+g) is strongly regular.

5 Existence Results for PVOP(K,f)\text{PVOP}(K,f) with Regularity

In this section we shall study emptiness and boundedness of the solution sets of PVOP(K,f)\text{PVOP}(K,f) under the regularity condition.

5.1 The weak regularity case

First, we give a necessary condition for the existence of Pareto efficient solutions of PVOP(K,f)\text{PVOP}(K,f).

Theorem 5.1

Assume that one of the following conditions hold:

  • (i)

    SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}.

  • (ii)

    there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that SOL(K,fi0)={0}SOL(K_{\infty},f^{\infty}_{i_{0}})=\{0\}.

Then SOLs(K,f)SOL^{s}(K,f) is nonempty.

Proof Let λ int 𝐑+s\lambda\in\mbox{ int }\mathbf{R}^{s}_{+} and x0Kx_{0}\in K. Define gλ(x)=i=1sλifi(x)g_{\lambda}(x)=\sum^{s}_{i=1}\lambda_{i}f_{i}(x) and Gx0={xK:f(x)f(x0)}G_{x_{0}}=\{x\in K:f(x)\leq f(x_{0})\}. Clearly, Gx0G_{x_{0}} is nonempty and closed. We assert that Gx0G_{x_{0}} is bounded. If not, then there exists {xk}Gx0\{x_{k}\}\subset G_{x_{0}} such that xk+\|x_{k}\|\to+\infty and xkxkx¯K\{0}\frac{x_{k}}{\|x_{k}\|}\to\bar{x}\in K_{\infty}\backslash\{0\} as k+k\to+\infty. It follows from xkGx0x_{k}\in G_{x_{0}} that

fi(xk)fi(x0),i=1,,s.f_{i}(x_{k})\leq f_{i}(x_{0}),\quad i=1,\cdots,s.

Dividing the both sides of the above inequality by xkdi\|x_{k}\|^{d_{i}} and letting k+k\rightarrow+\infty, we get

fi(x¯)fi(0)=0,i=1,,s,f^{\infty}_{i}(\bar{x})\leq f^{\infty}_{i}(0)=0,\quad i=1,\cdots,s,

a contradiction to SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\} as well as SOL(K,fi0)={0}SOL(K_{\infty},f_{i_{0}}^{\infty})=\{0\}. Hence, Gx0G_{x_{0}} is compact. By the Weierstrass\emph{Weierstrass}^{\prime} Theorem, SOL(Gx0,gλ)SOL(G_{x_{0}},g_{\lambda})\neq\emptyset. Since SOL(Gx0,gλ)SOLs(K,f)SOL(G_{x_{0}},g_{\lambda})\subseteq SOL^{s}(K,f) (by Lemma 1), we have SOLs(K,f)SOL^{s}(K,f)\neq\emptyset. ∎

The following example shows that SOLs(K,f)SOL^{s}(K,f) may be unbounded when SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}.

Example 5

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with f1(x1,x2)=x12x21,f2(x1,x2)=x23+1f_{1}(x_{1},x_{2})=x^{2}_{1}-x_{2}-1,f_{2}(x_{1},x_{2})=x^{3}_{2}+1 and

K={(x1,x2)𝐑2:x2x10}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{2}\geq x_{1}\geq 0\}.

It is easy to see that K=KK=K_{\infty}, f1(x1,x2)=x12f^{\infty}_{1}(x_{1},x_{2})=x^{2}_{1}, and f2(x1,x2)=x23f^{\infty}_{2}(x_{1},x_{2})=x^{3}_{2}. Clearly, SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}. On the other hand, SOLs(K,f)SOL^{s}(K,f) is unbounded since

{(x1,x2)K:x1=0,x20}SOLs(K,f).\{(x_{1},x_{2})\in K:x_{1}=0,x_{2}\geq 0\}\subseteq SOL^{s}(K,f).

The following result gives a Frank-Wolf type theorem for textPVOP(K,f)text{PVOP}(K,f) under the weak regularity condition.

Corollary 2

If PVOP(K,f)\mathrm{PVOP}(K,f) is weakly regular and f:𝐑n𝐑𝐬f:\mathbf{R}^{n}\mapsto\mathbf{R^{s}} is bounded from below on KK, then SOLs(K,f)SOL^{s}(K,f) is nonempty.

Proof It follows directly from Proposition 4, Remark 3 and Theorem 5.1. ∎

The following result shows that the existence of the Pareto efficient solutions of a polynomial vector optimization problem is preserved when the vector objective function is perturbed by a lower degree polynomial.

Theorem 5.2

Assume that one of the following conditions hold:

  • (i)

    SOLs(K,f)={0}SOL^{s}(K_{\infty},f^{\infty})=\{0\}.

  • (ii)

    there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that SOL(K,fi0)={0}SOL(K_{\infty},f^{\infty}_{i_{0}})=\{0\}.

Then SOLs(K,f+g)SOL^{s}(K,f+g) is nonempty for all g=(g1,,gs)𝐏𝐝g=(g_{1},\dots,g_{s})\in\mathbf{P}_{\mathbf{d}} with deggi<degfi\mathrm{deg}\ g_{i}<\mathrm{deg}\ f_{i}, i=1,,si=1,\dots,s.

Proof It follows directly from Proposition 8(i) and Theorem 5.1. ∎

5.2 The strong regularity case

In this subsection, we study nonemptiness and boundedness of the solution sets of  PVOP (K,f)\text{ PVOP }(K,f) under the strong regularity condition. Next, we give a necessary condition for the existence of the weak Pareto efficient solutions of  PVOP (K,f)\text{ PVOP }(K,f).

Theorem 5.3

If SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, then SOLw(K,f)=SOL^{w}(K,f)=\emptyset.

Proof Suppose that there exists x0SOLw(K,f)x_{0}\in SOL^{w}(K,f). By Proposition 1, 0SOLw(K,f)0\notin SOL^{w}(K_{\infty},f^{\infty}). Then there exists v1K\{0}v_{1}\in K_{\infty}\backslash\{0\} such that

f(v1)f(0)=f(v1) int 𝐑+s.f^{\infty}(v_{1})-f^{\infty}(0)=f^{\infty}(v_{1})\in-\mbox{ int }\mathbf{R}^{s}_{+}.

This means that

fi(v1)<fi(0)=0,i=1,,s.f^{\infty}_{i}(v_{1})<f^{\infty}_{i}(0)=0,\quad i=1,\dots,s. (6)

Since v1Kv_{1}\in K_{\infty}, there exist tk>0t_{k}>0 with tk+t_{k}\rightarrow+\infty and xkKx_{k}\in K such that tk1xkv1t^{-1}_{k}x_{k}\rightarrow v_{1} as k+k\rightarrow+\infty. Since x0SOLw(K,f)x_{0}\in SOL^{w}(K,f) and xkKx_{k}\in K, there exists ixk{1,,s}i_{x_{k}}\in\{1,\dots,s\} such that

fixk(xk)fixk(x0)0,k.f_{i_{x_{k}}}(x_{k})-f_{i_{x_{k}}}(x_{0})\geq 0,\quad\forall k.

Without loss of generality, we can assume that there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that

fi0(xk)fi0(x0)0.f_{i_{0}}(x_{k})-f_{i_{0}}(x_{0})\geq 0.

Dividing the both sides of the above inequality by tkdi0t_{k}^{d_{i_{0}}} and letting k+k\rightarrow+\infty, we have

fi0(v1)0,f^{\infty}_{i_{0}}(v_{1})\geq 0,

a contradiction to (6). ∎

Remark 7

By Theorem 5.3, SOLw(K,f)SOL^{w}(K,f)\neq\emptyset implies that SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty})\neq\emptyset. The following example shows that the converse does not hold in general.

Example 6

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=(x14x241)2+x14,f2(x1,x2)=(x12x221)2+x12f_{1}(x_{1},x_{2})=(x^{4}_{1}x^{4}_{2}-1)^{2}+x^{4}_{1},f_{2}(x_{1},x_{2})=(x^{2}_{1}x^{2}_{2}-1)^{2}+x^{2}_{1}

and K=𝐑nK=\mathbf{R}^{n}. It is easy to verify that 0SOLw(K,f)0\in SOL^{w}(K_{\infty},f^{\infty}). On the other hand, f1>0f_{1}>0 and f2>0f_{2}>0, but f(1n,n)=(1n4,1n2)(0,0)f(\frac{1}{n},n)=(\frac{1}{n^{4}},\frac{1}{n^{2}})\to(0,0) as n+n\to+\infty. This implies SOLw(K,f)=SOL^{w}(K,f)=\emptyset.

As a direct consequence of Proposition 1 and Theorem 5.3, we have the following result.

Theorem 5.4

If 0SOLw(K,f)0\notin SOL^{w}(K_{\infty},f^{\infty}), then SOLw(K,f)=SOL^{w}(K,f)=\emptyset.

Remark 8

By Theorem 5.3 (or Theorem 5.4), SOLw(K,f)SOL^{w}(K,f)\neq\emptyset implies SOLw(K,f)SOL^{w}(K_{\infty},f^{\infty})\neq\emptyset. The following example shows that SOLs(K,f)SOL^{s}(K_{\infty},f^{\infty}) may be empty when SOLs(K,f)SOL^{s}(K,f)\neq\emptyset.

Example 7

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=x221,f2(x1,x2)=x13+x2+1f_{1}(x_{1},x_{2})=-x^{2}_{2}-1,f_{2}(x_{1},x_{2})=-x^{3}_{1}+x_{2}+1

and

K={(x1,x2)𝐑2:0x11,x22}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:0\leq x_{1}\leq 1,x_{2}\geq 2\}.

Then K={(x1,x2)𝐑2:x1=0,x20}K_{\infty}=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}=0,x_{2}\geq 0\} and f(x1,x2)=(f1(x1,x2),f2(x1,x2))=(x22,x13)f^{\infty}(x_{1},x_{2})=(f^{\infty}_{1}(x_{1},x_{2}),f^{\infty}_{2}(x_{1},x_{2}))=(-x^{2}_{2},-x^{3}_{1}). It is easy to verify that SOLs(K,f)=SOL^{s}(K_{\infty},f^{\infty})=\emptyset and (1,2)SOLs(K,f)(1,2)\in SOL^{s}(K,f).

The following result shows that SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\} is sufficient for the existence of the Pareto efficient solutions as well as boundedness of the weak Pareto efficient solution set.

Theorem 5.5

If SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, then SOLs(K,f)SOL^{s}(K,f) is nonempty and SOLw(K,f)SOL^{w}(K,f) is compact.

Proof SOLs(K,f)SOL^{s}(K,f)\neq\emptyset follows from Theorem 3.1(i) and Theorem 5.1(ii). The closedness of SOLw(K,f)SOL^{w}(K,f) is clear. Next we prove that SOLw(K,f)SOL^{w}(K,f) is bounded. If not, then there exists ykSOLw(K,f)y_{k}\in SOL^{w}(K,f) such that yk+\|y_{k}\|\rightarrow+\infty as k+k\rightarrow+\infty. Without loss of generality, assume that yk0\|y_{k}\|\neq 0 and ykykvK\{0}\frac{y_{k}}{\|y_{k}\|}\rightarrow v\in K_{\infty}\backslash\{0\}. Let yKy\in K. Since ykSOLw(K,f)y_{k}\in SOL^{w}(K,f), there exists iyk{1,,s}i_{y_{k}}\in\{1,\dots,s\} such that

fiyk(y)fiyk(yk)0.f_{i_{y_{k}}}(y)-f_{i_{y_{k}}}(y_{k})\geq 0.

Since the set {1,,s}\{1,\dots,s\} is finite, without loss of generality, we can assume that there exists i0{1,,s}i_{0}\in\{1,\dots,s\} such that

fi0(y)fi0(yk)0,k.f_{i_{0}}(y)-f_{i_{0}}(y_{k})\geq 0,\quad\forall k.

Dividing the both sides of the above inequality by ykdi0\|y_{k}\|^{d_{i_{0}}} and letting k+k\to+\infty, we have fi0(v)fi0(0)=0f^{\infty}_{i_{0}}(v)\leq f^{\infty}_{i_{0}}(0)=0. This implies vSOL(K,fi0)\{0}v\in SOL(K_{\infty},f^{\infty}_{i_{0}})\backslash\{0\}, a contradiction. ∎

As a consequence of Theorem 5.5 and Theorem 3.1(i), we have the following result.

Corollary 3

If SOL(K,fi)={0}SOL(K_{\infty},f^{\infty}_{i})=\{0\} for each i{1,2,,s}i\in\{1,2,\dots,s\}, then SOLs(K,f)SOL^{s}(K,f) is nonempty and SOLw(K,f)SOL^{w}(K,f) is bounded.

Now we give an example to illustrate the conclusion of Corollary 3.

Example 8

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=x23x12x1x2+1,f2(x1,x2)=x22x11f_{1}(x_{1},x_{2})=x^{3}_{2}-x^{2}_{1}-x_{1}x_{2}+1,f_{2}(x_{1},x_{2})=x^{2}_{2}-x_{1}-1

and

K={(x1,x2)𝐑2:x110,x2x1+10,ex11x20}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}-1\geq 0,x_{2}-x_{1}+1\geq 0,e^{x_{1}-1}-x_{2}\geq 0\}.

It is easy to verify that f1(x1,x2)=x23,f2(x1,x2)=x22f^{\infty}_{1}(x_{1},x_{2})=x^{3}_{2},f^{\infty}_{2}(x_{1},x_{2})=x^{2}_{2},

K={(x1,x2)𝐑2:x10,x2x10},K_{\infty}=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}\geq 0,x_{2}-x_{1}\geq 0\},

SOL(K,f1)={(0,0)}SOL(K_{\infty},f^{\infty}_{1})=\{(0,0)\}, and SOL(K,f2)={(0,0)}SOL(K_{\infty},f^{\infty}_{2})=\{(0,0)\}. By Corollary 3, SOLs(K,f)SOL^{s}(K,f) is nonempty and SOLw(K,f)SOL^{w}(K,f) is compact. It is worth mentioning that (DTN, , Theorem 4.1) and (LGJ, , Theorem 3.1) cannot be applied in this example since ff is non-convex on 𝐑2\mathbf{R}^{2} and KK is neither convex nor semi-algebraic set.

Remark 9

By Theorem 5.1(ii), SOLs(K,f)SOL^{s}(K,f) is nonempty when there exists some i0{1,,s}i_{0}\in\{1,\cdots,s\} such that SOL(K,fi0)={0}SOL(K_{\infty},f^{\infty}_{i_{0}})=\{0\}. If SOL(K,fi)={0}SOL(K_{\infty},f^{\infty}_{i})=\{0\} for all i{1,,s}i\in\{1,\cdots,s\}, then SOLs(K,f)SOL^{s}(K,f) is nonempty and bounded(by Corollary 3). The following example shows that SOLs(K,f)SOL^{s}(K,f) may be unbounded when there exists {i0,j0}{1,,s}\{i_{0},j_{0}\}\subset\{1,\cdots,s\} such that SOL(K,fi0)={0}SOL(K_{\infty},f^{\infty}_{i_{0}})=\{0\} and SOL(K,fj0){0}SOL(K_{\infty},f^{\infty}_{j_{0}})\neq\{0\}.

Example 9

Consider the vector polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=x13,f2(x1,x2)=x12+x2f_{1}(x_{1},x_{2})=x^{3}_{1},\quad f_{2}(x_{1},x_{2})=-x^{2}_{1}+x_{2}

and

K={(x1,x2)𝐑2:x20,x1x20}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{2}\geq 0,x_{1}-x_{2}\geq 0\}.

It is easy to verify that K=KK=K_{\infty}, f1(x1,x2)=x13,f2(x1,x2)=x12f^{\infty}_{1}(x_{1},x_{2})=x^{3}_{1},f^{\infty}_{2}(x_{1},x_{2})=-x^{2}_{1}, SOL(K,f1)={0}SOL(K_{\infty},f^{\infty}_{1})=\{0\}, SOL(K,f2)=SOL(K_{\infty},f^{\infty}_{2})=\emptyset. On the other hand, SOLs(K,f)SOL^{s}(K,f) is unbounded since

{(x1,x2)K:x10,x2=0}SOLs(K,f).\{(x_{1},x_{2})\in K:x_{1}\geq 0,x_{2}=0\}\subseteq SOL^{s}(K,f).

It has been shown in Corollary 2 that  PVOP (K,f)\text{ PVOP }(K,f) with weak regularity admits a weak Pareto efficient solution provided that ff is bounded from below on KK. In the following Frank-Wolf type theorem, we further prove the existence of Pareto efficient solutions and compactness of the weak Pareto efficient solution set if we strengthen weak regularity by strong regularity.

Corollary 4

If  PVOP(K,f)\mathrm{PVOP}(K,f) is strongly regular and f:𝐑n𝐑𝐬f:\mathbf{R}^{n}\mapsto\mathbf{R^{s}} is bounded from below on KK, then SOLs(K,f)SOL^{s}(K,f) is nonempty and SOLw(K,f)SOL^{w}(K,f) is compact.

Proof It follows directly from Proposition 5 and Theorem 5.5. ∎

Remark 10

Corollary 4 extends (HV1, , Theorem 3.1) to the vector case.

The following example shows that the converse of Theorem 5.5, Corollary 3 and Corollary 4 does not hold in general.

Example 10

Consider the polynomial f=(f1,f2)f=(f_{1},f_{2}) with

f1(x1,x2)=x1x2+1,f2(x1,x2)=x1x2+x11f_{1}(x_{1},x_{2})=x_{1}x_{2}+1,\quad f_{2}(x_{1},x_{2})=x_{1}x_{2}+x_{1}-1

and

K={(x1,x2)𝐑2:x11,x21}.K=\{(x_{1},x_{2})\in\mathbf{R}^{2}:x_{1}\geq 1,x_{2}\geq 1\}.

Then f1(x1,x2)=f2(x1,x2)=x1x2f^{\infty}_{1}(x_{1},x_{2})=f^{\infty}_{2}(x_{1},x_{2})=x_{1}x_{2} and K=𝐑+2K_{\infty}=\mathbf{R}^{2}_{+}. It is easy to verify that SOLs(K,f)=SOLw(K,f)={(1,1)}SOL^{s}(K,f)=SOL^{w}(K,f)=\{(1,1)\}. However, SOL(K,f1)=SOL(K,f2)={(x1,x2)𝐑+2:x1x2=0}SOL(K_{\infty},f^{\infty}_{1})=SOL(K_{\infty},f^{\infty}_{2})=\{(x_{1},x_{2})\in\mathbf{R}^{2}_{+}:x_{1}x_{2}=0\} are unbounded.

The following result shows that some properties of the solution sets of a strongly regular polynomial vector optimization problem are preserved when its objective polynomial is perturbed by a lower degree polynomial.

Theorem 5.6

Let g=(g1,,gs):𝐑n𝐑sg=(g_{1},\cdots,g_{s}):\mathbf{R}^{n}\mapsto\mathbf{R}^{s} be a vector polynomial with deg\mathrm{deg} gi<degg_{i}<\mathrm{deg} fi,i=1,,sf_{i},i=1,\cdots,s. Then the following conclusions hold:

  • (i)

    If SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, then SOLw(K,f+g)=SOL^{w}(K,f+g)=\emptyset.

  • (ii)

    If SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, then SOLs(K,f+g)SOL^{s}(K,f+g) is nonempty and SOLw(K,f+g)SOL^{w}(K,f+g) is compact.

Proof (i) follows from Proposition 8(iv) and Theorem 5.3, and (ii) follows from Proposition 8(ii) and Theorem 5.5. ∎

6 Stability Analysis

In this section we investigate the solution stability of a regular polynomial vector optimization problem. First, we shows that some properties of the solution sets of a strongly regular polynomial vector optimization problem are stable under a small perturbation.

Theorem 6.1

The following conclusions hold:

  • (i)

    If SOLw(K,f)=SOL^{w}(K_{\infty},f^{\infty})=\emptyset, then there exists ϵ>0\epsilon>0 such that SOLw(K,f+g)=SOL^{w}(K,f+g)=\emptyset for all g𝐏𝐝g\in\mathbf{P}_{\mathbf{d}} satisfying g<ϵ\|g\|<\epsilon.

  • (ii)

    If SOLw(K,f)={0}SOL^{w}(K_{\infty},f^{\infty})=\{0\}, then there exists ϵ>0\epsilon>0 such that SOLs(K,f+g)SOL^{s}(K,f+g) is nonempty and SOLw(K,f+g)SOL^{w}(K,f+g) is compact for all g𝐏𝐝g\in\mathbf{P}_{\mathbf{d}} satisfying g<ϵ\|g\|<\epsilon.

Proof (i) follows from Theorem 4.1 and Theorem 5.3 and (ii) follows from Theorem 4.1 and Theorem 5.5. ∎

In the sequel we shall investigate the local boundedness and upper semicontinuity of the weak Pareto efficient solution mapping. Recall that a set-valued mapping T:𝐑κ2𝐑nT:\mathbf{R}^{\kappa}\rightrightarrows 2^{\mathbf{R}^{n}} is said to be upper semi-continuous at xx if for any neighborhood UU of T(x)T(x), there exists a neighborhood VV of xx such that T(y)UT(y)\subseteq U for any yVy\in V. A set-valued mapping T:𝐑κ2𝐑nT:\mathbf{R}^{\kappa}\rightrightarrows 2^{\mathbf{R}^{n}} is locally bounded at xx if there exists an open neighborhood VV of xx such that yVT(y)\cup_{y\in V}T(y) is bounded.

Lemma 2

(RC, , Theorem 5.7 and Theorem 5.19) If T:𝐑κ2𝐑nT:\mathbf{R}^{\kappa}\rightrightarrows 2^{\mathbf{R}^{n}} is locally bounded at xx and TT has a closed graph Gph(T)\mathrm{Gph}(T), where Gph(T)={(x,y)𝐑κ×𝐑n:yT(x)}\mathrm{Gph}(T)=\{(x,y)\in\mathbf{R}^{\kappa}\times\mathbf{R}^{n}:y\in T(x)\}, then TT is upper semi-continuous at xx.

Theorem 6.2

Assume that KK is convex and PVOP(K,f)\mathrm{PVOP}(K,f) is strongly regular. Then the following conclusions hold:

  • (i)

    SOLw(K,)SOL^{w}(K,\cdot) is locally bounded at ff, i.e., there exists δ>0\delta>0 such that

    Qδ=h𝐁(0,δ)SOLw(K,f+h)Q_{\delta}=\cup_{h\in\mathbf{B}(0,\delta)}SOL^{w}(K,f+h)

    is bounded, where the 𝐁(0,δ)\mathbf{B}(0,\delta) is an open ball in 𝐏𝐝\mathbf{P}_{\mathbf{d}} with center at 0 and radius δ>0\delta>0.

  • (ii)

    SOLw(K,)SOL^{w}(K,\cdot) is upper semi-continuous on 𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s}.

Proof (i) Since 𝐆𝐑s𝐝\mathbf{GR}^{\mathbf{d}}_{s} is open in 𝐏𝐝\mathbf{P}_{\mathbf{d}} (by Proposition 7) and f𝐆𝐑s𝐝f\in\mathbf{GR}^{\mathbf{d}}_{s}, there exists δ>0\delta>0 such that f+𝐁¯(0,δ)𝐆𝐑s𝐝f+\bar{\mathbf{B}}(0,\delta)\subseteq\mathbf{GR}^{\mathbf{d}}_{s}, where 𝐁¯(0,δ)\bar{\mathbf{B}}(0,\delta) denotes the closure of 𝐁(0,δ)\mathbf{B}(0,\delta). Suppose on the contrary that Qδ=SOLw(K,f+𝐁(0,δ))Q_{\delta}=SOL^{w}(K,f+\mathbf{B}(0,\delta)) is unbounded. Then there exist {hk}𝐁(0,δ)\{h^{k}\}\subseteq\mathbf{B}(0,\delta) and xkSOLw(K,f+hk)x_{k}\in SOL^{w}(K,f+h^{k}) such that xk+\|x_{k}\|\to+\infty as k+k\to+\infty. Without loss of generality, we may assume that xkxkxK\{0}\frac{x_{k}}{\|x_{k}\|}\to x^{*}\in K_{\infty}\backslash\{0\} and hkh𝐁¯(0,δ)h^{k}\to h\in\bar{\mathbf{B}}(0,\delta). Let xKx\in K and vKv\in K_{\infty}. Then x+xkvKx+\|x_{k}\|v\in K for all kk. Since xkSOLw(K,f+hk)x_{k}\in SOL^{w}(K,f+h^{k}), we have

(f+hk)(x+xkv)(f+hk)(xk) int 𝐑+s.(f+h^{k})(x+\|x_{k}\|v)-(f+h^{k})(x_{k})\notin-\mbox{ int }\mathbf{R}^{s}_{+}.

Then for each kk, there exists ik,v{1,,s}i_{k,v}\in\{1,\dots,s\} such that

(fik,v+hik,vk)(x+xkv)(fik,v+hik,vk)(xk)0.(f_{i_{k,v}}+h^{k}_{i_{k,v}})(x+\|x_{k}\|v)-(f_{i_{k,v}}+h^{k}_{i_{k,v}})(x_{k})\geq 0.

Since {1,,s}\{1,\dots,s\} is finite, we can assume that there exists i0,v{1,,s}i_{0,v}\in\{1,\dots,s\} such that

(fi0,v+hi0,vk)(x+xkv)(fi0,v+hi0,vk)(xk)0,k.(f_{i_{0,v}}+h^{k}_{i_{0,v}})(x+\|x_{k}\|v)-(f_{i_{0,v}}+h^{k}_{i_{0,v}})(x_{k})\geq 0,\quad\forall k.

Dividing the both sides of the above inequality by xkdi0,v\|x_{k}\|^{d_{i_{0,v}}} and letting k+k\to+\infty, we get

(fi0,v+hi0,v)(v)(fi0,v+hi0,v)(x).(f_{i_{0,v}}+h_{i_{0,v}})^{\infty}(v)\geq(f_{i_{0,v}}+h_{i_{0,v}})^{\infty}(x^{*}).

Since vKv\in K_{\infty} is arbitrary, we have xSOLw(K,(f+h))\{0}x^{*}\in SOL^{w}(K_{\infty},(f+h)^{\infty})\backslash\{0\}. On the other hand, since f+hf+𝐁¯(0,δ)𝐆𝐑s𝐝f+h\in f+\bar{\mathbf{B}}(0,\delta)\subseteq\mathbf{GR}^{\mathbf{d}}_{s}, by Proposition 2 we have xSOLw(K,(f+h))x^{*}\notin SOL^{w}(K_{\infty},(f+h)^{\infty}), a contradiction. Thus, SOLw(K,)SOL^{w}(K,\cdot) is locally bounded at ff.

(ii) Let f𝐆𝐑s𝐝f\in\mathbf{GR}^{\mathbf{d}}_{s}. By (i) and Lemma 2, we only need to prove that the graph Gph(SOLw(K,))\mbox{Gph}(SOL^{w}(K,\cdot)) of SOLw(K,)SOL^{w}(K,\cdot) is closed in 𝐏𝐝×𝐑n\mathbf{P}_{\mathbf{d}}\times\mathbf{R}^{n}. Let (fk,yk)(f,y)(f^{k},y_{k})\to(f,y) as k+k\to+\infty with ykSOLw(K,fk)y_{k}\in SOL^{w}(K,f^{k}). Then for any zKz\in K, we have

fk(z)fk(yk) int 𝐑+s.f^{k}(z)-f^{k}(y_{k})\notin-\mbox{ int }\mathbf{R}^{s}_{+}.

Letting k+k\to+\infty, we have

f(z)f(y) int 𝐑+s,zK,f(z)-f(y)\not\in-\mbox{ int }\mathbf{R}^{s}_{+},\quad\forall z\in K,

which means ySOLw(K,f)y\in SOL^{w}(K,f). Thus, Gph(SOLw(K,))\mbox{Gph}(SOL^{w}(K,\cdot)) is closed in 𝐏𝐝×𝐑n\mathbf{P}_{\mathbf{d}}\times\mathbf{R}^{n}.. ∎

7 Conclusion

In this paper we extend the concept of regularity due to Hieu HV1 to the polynomial vector optimization problem. Under regularity conditions, we investigate nonemptiness and boundedness of the solution sets of a non-convex polynomial vector optimization problem on a nonempty closed set (not necessarily semi-algebraic set). As a consequence, we derive two Frank-Wolfe type theorems for a non-convex polynomial vector optimization problem. We also discuss the solution stability. Our results extend and improve the corresponding results of DTN ; BT1 ; BT2 ; HV1 ; LGJ .

References

  • (1) Ha, H.V., Pham, T.S.(eds.): Genericity in Polynomial Optimization. World Science Publishing, Singapore (2017)
  • (2) Benedetti, R., Risler, J.J.(eds.): Real Algebraic and Semi-Algebraic Sets. Hermann, Paris (1990)
  • (3) Ansari, Q.H., Lalitha, C.S., Mehta, M.(eds.): Generalized Convexity, Nonsmooth Variational Inequalities, and Nonsmooth Optimization. Chapman and Hall/CRC, New York (2013)
  • (4) Hieu, V.T.: A regularity condition in polynomial optimization. arXiv:1808.06100
  • (5) Hieu, V.T., Wei, Y.M., Yao, J.C.: Notes on the optimization problems corresponding to polynomial complementarity problems. J. Optim. Theory Appl. 184, 687–695 (2020)
  • (6) Kim, D.S., Pham, T.S., Tuyen, V.T.: On the existence of Pareto solutions for polynomial vector optimization problems. Math. Program. 177, 321–341 (2019)
  • (7) Jiao, L.G., Lee, J.H., Sisarat, N.: Multi-objective convex polynomial optimization and semidefinite programming relaxations. arXiv:1903.10137
  • (8) Rockafellar, R.T., Wets, R.J-B.(eds.): Variational Analysis. Springer, Berlin (2009)
  • (9) Auslender, A., Teboulle, M.(eds.): Asymptotic Cones and Functions in Optimization and Variational Inequalities. Springer, New York (2003)
  • (10) Bao, B.T., Mordukhovich, B.S.: Variational principles for set-valued mappings with applications to multiobjective optimization. Control Cybern. 36, 531–562 (2007)
  • (11) Bao, B.T., Mordukhovich, B.S.: Relative Pareto minimizers for multiobjective problems: existence and optimality conditions. Math. Program. 122, 301–347 (2010)
  • (12) Giannessi, F., Mastroeni, G., Pellegrini, L.: On the theory of vector optimization and variational inequalities. Image space analysis and separation. In: Giannessi, F. (ed.): Vector Variational Inequalities and Vector Equilibria. Nonconvex Optimization and Its Applications, vol. 38, Springer, Boston, MA. (2000)
  • (13) Huang, N.J. and Fang, Y.P.: The upper semicontinuity of the solution maps in vector implicit quasicomplementarity problems of type R0R_{0}. Appl. Math. Lett. 16, 1151–1156 (2003)
  • (14) Fang, Y.P. and Huang, N.J.: On the upper semi-continuity of the solution map to the vertical implicit homogeneous complementarity problem of type R0R_{0}. Positivity 10, 95–104 (2006)
  • (15) Oettli, W., Yen, N.D.: Quasicomplementarity problems of type R0R_{0}, J. Optim. Theory Appl. 89, 467–474 (1996)
  • (16) R. López: Stability results for polyhedral complementarity problems. Comput. Math. Appl. 58. 1475–1486 (2009)
  • (17) Gowda, M. S., Sossa, D.: Weakly homogeneous variational inequalities and solvability of nonlinear equations over cones. Math. Program., Ser. A, 177, 149–171 (2019)
  • (18) Yang, X. Q.: Vector complementarity and minimal element problems. J. Optim. Theory Appl. 77, 483–495 (1993)
  • (19) Chen, G.Y., Yang, X.Q.: The vector complementary problem and its equivalence with weak minimal elements in ordered spaces. J. Math. Anal. Appl. 153, 136–158 (1990)
  • (20) Lee, G.M., Kim, D.S., Lee, B.S., Yen N.D.: Vector variational inequality as a tool for studying vector optimization problems. In: Giannessi, F.(ed.) Vector Variational Inequalities and Vector Equilibria. Nonconvex Optimization and Its Applications, vol. 38, Springer, Boston, MA. (2000)