Abstract
In this paper, we consider the generalized polynomial variational inequality, which is a subclass of generalized variational inequalities; and it covers several classes of variational inequalities with polynomial functions studied recently in the literature. A well-known existence and uniqueness theorem for the generalized variational inequality was established by Pang and Yao (SIAM J Control Optim 33:168–184, 1995). It is not difficult to show that the conditions of this theorem do not hold for generalized variational inequalities with general polynomial functions. In this paper, in terms of properties of the involved polynomial and by making use of the theory related to exceptional family of elements, we establish an existence and uniqueness theorem for the generalized polynomial variational inequality. A specific example is given to confirm our theoretical findings.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(f,g: \mathbb {R}^{n}\rightarrow \mathbb {R}^{n}\) be two continuous functions, and let K be a nonempty, closed and convex subset of \(\mathbb {R}^n\). The so-called generalized variational inequality, denoted by GVI(g, f, K), is to find an \(\mathbf {x} \in \mathbb {R}^n\) such that
which was introduced by Noor [20] in 1988. When \(g(\mathbf {x})=\mathbf {x}\), GVI(g, f, K) reduces to the standard finite-dimensional variational inequality [5, 9]. Furthermore, when \(K=\mathbb {R}^n_+\), GVI(g, f, K) further reduces to the standard complementarity problem [7].
The existence and uniqueness of solutions to the variational inequality is an important issue in the studies of the theory, algorithms and applications for the variational inequality. It is known that a vector \(\mathbf {x}\in \mathbb {R}^n\) solves GVI(g, f, K) if and only if it is a solution of the generalized normal equation:
where \(P_K(\mathbf {u})\) denotes the projection of the vector \(\mathbf {u}\in \mathbb {R}^n\) onto the set K. A well-known result on the existence and uniqueness for GVI(g, f, K), achieved by Pang and Yao in [21, Proposition 3.9], is described as follows.
Theorem 1
Let K be a nonempty, closed and convex subset of \(\mathbb {R}^n\), and let f and g be two continuous functions from \(\mathbb {R}^n\) into itself with g being injective. Suppose
-
(a)
there exists a vector \(\mathbf {u}\in g^{-1}(K)\) and positive scalars \(\alpha \) and L such that
$$\begin{aligned} \Vert g(\mathbf {x})-g(\mathbf {u})\Vert \le L\Vert \mathbf {x}-\mathbf {u}\Vert \end{aligned}$$holds for all \(\mathbf {x}\in g^{-1}(K)\) with \(\Vert \mathbf {x}\Vert \ge \alpha \);
-
(b)
f is strongly monotone with respect to g on K, i.e., there is a scalar \(c > 0\) such that
$$\begin{aligned} {[}f(\mathbf {x})-f(\mathbf {y}){]}^\top [g(\mathbf {x})-g(\mathbf {y})]\ge c\Vert \mathbf {x}-\mathbf {y}\Vert ^2 \end{aligned}$$(1)holds for all \(g(\mathbf {x}),g(\mathbf {y})\in K\) with \(\mathbf {x}\ne \mathbf {y}\).
Then, there exists a unique vector \(\bar{\mathbf {x}}\in \mathbb {R}^n\) satisfying \(g(\mathbf {x})=P_K[g(\mathbf {x})-f(\mathbf {x})]\).
When f and g are both polynomials, the corresponding GVI(g, f, K) is called the generalized polynomial variational inequality in this paper, and we denote it by the GPVI. In recent years, several subclasses of the GPVI have been studied extensively. These subclasses include tensor complementarity problems [1, 3, 11, 14, 19, 23,24,25] (also see survey papers [12, 13, 22]), polynomial complementarity problems [6, 15], generalized polynomial complementarity problems [16], tensor variational inequalities [26], and polynomial variational inequalities [10]. Many interesting results for these several subclasses were achieved by using the special properties of polynomials. It is natural to ask how to study the GPVI by using special properties of polynomials? In particular, we find that for general polynomials f and g, it is impossible that the function f is strongly monotone with respect to g on the set K (see Proposition 1 in the next section). This implies that the result of Theorem 1 cannot be generally applied to the GPVI. A natural question is that how to investigate the existence and uniqueness of solutions to the GPVI?
In this paper, our main purpose is to establish an existence and uniqueness theorem for the GPVI by making use of properties of the involved polynomials. In the next section, we introduce some symbols and concepts, and discuss the existence of solutions to the GPVI by using the concept of the exceptional family of elements and degree theory. In Sect. 3, we develop an existence and uniqueness theorem for the GPVI, and derive several corollaries for subclasses of the GPVI. Moreover, we give an example in support of our theoretical findings. Conclusions are given in the last section.
2 Preliminaries
We first give an introduction of tensors, which plays an important role in our analysis. A tensor is a natural extension of a matrix. For any given positive integers m and n with \(m,n\ge 2\), we call \(\mathcal {A}=(a_{i_1i_2\cdots i_m})\), where \(a_{i_1i_2\cdots i_m}\in \mathbb {R}\) for \(i_j\in \{1,2,\ldots ,n\}\) and \(j\in \{1,2,\ldots ,m\}\), an m-th order n-dimensional real square tensor; and denote the space of m-th order n-dimensional real square tensors by \(\mathbb {R}^{[m,n]}\).
For any \(\mathcal {A}=(a_{i_1i_2\cdots i_m})\in \mathbb {R}^{[m,n]}\) and \(\mathbf {x}=(x_1,x_2,\ldots ,x_n)^\top \in \mathbb {R}^n\), \(\mathcal {A} \mathbf {x}^{m-1}\) is an n-dimensional vector whose ith component is given by
and \(\mathcal {A} \mathbf {x}^{m}\) is a homogeneous polynomial of degree m, defined by
\(\mathcal {A}=(a_{i_1i_2\cdots i_m})\in \mathbb {R}^{[m,n]}\) is said to be a positive definite tensor if and only if \(\mathcal {A}\mathbf {x}^m > 0\) for all \(\mathbf {x}\in \mathbb {R}^n{\setminus }\{\mathbf {0}\}\).
Let
where
then the corresponding GVI(g, f, K) is a generalized polynomial variational inequality, denoted by \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\).
We have the following observation.
Proposition 1
Let K be a nonempty, closed and convex subset of \(\mathbb {R}^n\), and let \(f,g:\mathbb {R}^n\rightarrow \mathbb {R}^n\) be two polynomials defined by
for some \(m^\prime \in \{1,2,\ldots ,m-1\}\) and \(l^\prime \in \{1,2,\ldots ,l-1\}\), where
If \(g(\mathbf {0})\in K\) and \(m-m^\prime +l-l^\prime >2\), then the function f is not strongly monotone with respect to g on K.
Proof
Suppose that f is strongly monotone with respect to g on K, i.e., there exists a constant \(c >0\) such that (1) holds for any \(g(\mathbf {x}), g(\mathbf {y})\in K\) with \(\mathbf {x}\ne \mathbf {y}\). Take \(\mathbf {y}=\mathbf {0}\), then we can get from (1) that
where \(\mathcal {C}^{(q-1)}\in \mathbb {R}^{[m+l-q,n]}\) for all \(q\in \{2,3,\ldots ,m^\prime +l^\prime \}\). By dividing both sides of (5) by \(\Vert \mathbf {x}\Vert ^{m+l-m^\prime -l^\prime }\), we get
Let \(\Vert \mathbf {x}\Vert \rightarrow \mathbf {0}\), it follows by the condition \(m-m^\prime +l-l^\prime >2\) that the left-hand side of the inequality (6) is bounded; but the right-hand side of (6) tends to \(\infty \), which leads to a contradiction. \(\square \)
Proposition 1 demonstrates that if \(g(\mathbf {0})\in K\), the degrees of both f and g defined by (2) are greater or equal to 1, and at least one of f and g has no term of degree 1, then f is not strongly monotone with respect to g on K. This implies that the result of Theorem 1 for GVI(g, f, K) cannot be generally applied to \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\). In order to study the existence and uniqueness of solutions to \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\), we need to use the concept of exceptional family of elements. In fact, the concept of exceptional family of elements has been extensively used to investigate the existence of the standard variational inequalities and complementarity problems. The following definition is similar to the those in [8, 27, 28].
Definition 1
Let f, g be two continuous functions. A set of points \(\{\mathbf {x}^r\}\subseteq \mathbb {R}^n\) is called an exceptional family of elements for the pair (f, g) with respect to \(\hat{\mathbf {x}}\in \mathbb {R}^n\), if \(\Vert \mathbf {x}^r\Vert \rightarrow \infty \) as \(r\rightarrow \infty \); and for each \(\mathbf {x}^r\), there exists a positive scalar \(\alpha ^r\) such that
where \(\mathcal{N}_K(\pi _r)\) denotes the normal cone of the convex set K at the point the point \(\pi _r\).
In order to establish the existence result for \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) by using the concept of exceptional family of elements, we need to use the degree theory. We now review some basic concepts and results of degree theory [18]. Suppose \(\Omega \) is a bounded open set in a finite-dimensional Hilbert space H, we use \(\bar{\Omega }\) and \(\partial \Omega \) to denote its closure and boundary in H, respectively. Let \(\phi : \bar{\Omega }\rightarrow H\) be continuous, and \(\mathbf {p}\in H\) with \(\mathbf {p} \notin \phi (\partial \Omega )\). Then, the topological degree of \(\phi \) over \(\Omega \) with respect to \(\mathbf {p}\) can be defined, which is denoted by deg\((\phi ,\Omega , \mathbf {p})\). Let \(\mathcal {H}(\mathbf {x},t): H \times [0,1]\rightarrow H\) be continuous. Suppose that for some bounded open set \(\Omega \) in H, \(\mathbf {0}\notin H(\partial \Omega ,t)\) for all \(t\in \) [0,1]. Then, the homotopy invariance property of degree says that deg(\(\mathcal {H}(\cdot ,\)t\(),\Omega ,\mathbf {0})\) is independent of t. In particular, if the set
is bounded, then for any bounded open set \(\Omega \) in H that contains the set \(\Psi \), it follows that
The following result is very useful for us to prove the main result in this paper.
Lemma 1
For two continuous mappings \(f,g: \mathbb {R}^n\rightarrow \mathbb {R}^n\) and a nonempty, closed and convex set K in \(\mathbb {R}^n\), there exists either a solution of GVI(g, f, K) or an exceptional family of elements with respect to any given \(\hat{\mathbf {x}}\in \mathbb {R}^n\) for the pair (f, g).
Proof
The proof is similar to the one in [27]. For the sake of completeness, we give it here. Suppose that GVI(g, f, K) has no solution, we need to show there exists an exceptional family of elements with respect to any given \(\hat{\mathbf {x}}\in \mathbb {R}^n\) for the pair (f, g). Let \(\Phi : \mathbb {R}^n\rightarrow \mathbb {R}^n\) be defined by
Then, \(\Phi \) is a continuous function; and \(\mathbf {x}\in \mathbb {R}^n\) solves GVI(g, f, K) if and only if it is a solution of \(\Phi (\mathbf {x})=\mathbf {0}\). For any \(\mathbf {x},~\hat{\mathbf {x}}\in \mathbb {R}^n\), we define
We claim that the set \(\mathcal {F}:=\{\mathbf {x}\in \mathbb {R}^n: H(\mathbf {x},t)=\mathbf {0}\;\text{ for } \text{ some }\; t\in [0,1]\}\) is unbounded. Suppose by the way of contradiction that \(\mathcal {F}\) is bounded. Then there exists a bounded open set \(\Omega \) in \(\mathcal {F}\) such that \(\mathbf {x}\notin \mathcal {F}\) for all \(\mathbf {x}\in \partial \Omega \). Thus, by the homotopy invariance theorem of degree, it follows that
which implies that \(\Phi (\mathbf {x})=\mathbf {0}\) has a solution in \(\Omega \). This is a contradiction with the assumption at the beginning of the proof. So, the set \(\mathcal {F}\) is unbounded. Thus, there exists an unbounded sequence \(\{\mathbf {x}^r\}\subseteq \mathcal {F}\). Without loss of generality, we may assume that \(\Vert \mathbf {x}^r\Vert > \Vert {\hat{\mathbf {x}}}\Vert \) for all r. By the definition of \(\mathcal {F}\), for each \(\mathbf {x}^r\) there is a scalar \(t^r\in [0,1]\) such that
Since GVI(g, f, K) has no solution, we have \(\Phi (\mathbf {x}^r)\ne \mathbf {0}\). We deduce from (8) that \(t^r\ne 0,1\). Thus, in the rest of the proof, it is sufficient to consider the case of \(t^r\in (0,1)\). In this case, we show that GVI(g, f, K) has an exceptional family with respect to \(\hat{\mathbf {x}}\).
From (8) it follows that
Denote \(\alpha _r:=\frac{t^r}{1-t^r}\) and \(\pi _r:=\alpha _r(\mathbf {x}^r-\hat{\mathbf {x}})+g(\mathbf {x}^r)\). Then, \(\pi _r=P_K[g(\mathbf {x}^r)-f(\mathbf {x}^r)]\), which implies that \(\pi _r\in K\), and that \(\pi _r\) is the unique solution to the following convex program:
Obviously, the function h is locally Lipschitz continuous, and hence, by the corollary of Proposition 2.4.3 of [4] we have
Therefore, \(\{\mathbf {x}^r\}\) is an exceptional family with respect to \(\hat{\mathbf {x}}\) for GVI(g, f, K).
The proof is complete. \(\square \)
3 Main result
In this section, by using Lemma 1 we establish an existence and uniqueness result for \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\); and then, we give an example in support of our theoretical findings.
Theorem 2
Let f and g be defined by (2) and (3), and let \(K\subseteq \mathbb {R}^n\) be a nonempty, closed and convex set with \(g(\mathbf {0})\in K\). Suppose that
-
(a)
f is strictly monotone with respect to g on K, i.e.,
$$\begin{aligned}{}[f(\mathbf {x})-f(\mathbf {y})]^\top [g(\mathbf {x})-g(\mathbf {y})]>0\quad \text{ for } \text{ all }\; g(\mathbf {x}),g(\mathbf {y})\in K; \end{aligned}$$ -
(b)
for any nonzero vector \(\mathbf {x}\in \mathbb {R}^n\), \(\langle \mathcal {A}^{(1)}\mathbf {x}^{m-1},\mathcal {B}^{(1)}\mathbf {x}^{l-1}\rangle \ne 0\);
-
(c)
one of \(\mathcal {A}^{(1)}\) and \(\mathcal {B}^{(1)}\), whose order is even and is larger than or equal to another, is positive definite,
then for any given vectors \(\mathbf {a}\in \mathbb {R}^n\) and \(\mathbf {b}\in \) K, \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) has a unique solution.
Proof
Without loss of generality, we assume that \(m\ge l\). In this case, it follows from the condition (c) that \(\mathcal {A}^{(1)}\) is positive definite.
We first prove the nonemptyness of \(SOL(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\). Suppose, on the contrary, that \(SOL(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)=\emptyset \). Then, it follows from Lemma 1 that there exists an exceptional family of elements for the pair (f, g) with respect to \(\mathbf {0}\in \mathbb {R}^n\), that is to say, there exists a sequence \(\{\mathbf {x}^r\}_{r=1}^\infty \subseteq \mathbb {R}^n\) satisfying \(\Vert \mathbf {x}^r\Vert \rightarrow \infty ~as~r\rightarrow \infty \), and scalars \(\alpha _r>0\) such that (7) holds, and hence, from the definition of the normal cone, we have
By dividing both sides of (9) by \(\Vert \mathbf {x}^r\Vert ^{m+l-2}\) , we get
that is,
For any r, we let \( \bar{\mathbf {x}}^r=\frac{\mathbf {x}^r}{\Vert \mathbf {x}^r\Vert }\). Without loss of generality, we may assume \(\bar{\mathbf {x}}^r\rightarrow \bar{\mathbf {x}}\). Since \(\mathcal {A}^{(1)}\) is positive definite and \(\bar{\mathbf {x}}\in \mathbb {R}^n{\setminus }\{\mathbf {0}\}\), it holds from (10) that
Moreover, since f is strictly monotone with respect to g on K and \(g(\mathbf {0})\in K\), by taking \(\mathbf {y}=\mathbf {0}\), it follows that
By dividing both sides by \(\Vert \mathbf {x}^r\Vert ^{m+l-2}\) and let \(r\rightarrow \infty \), we have
which, together with the condition (b), implies that
Furthermore, we have
Thus, for all sufficiently large r, we deduce that
that is, for all sufficiently large r,
Now, we consider the following two cases.
-
(i)
If \(\{u^r\}\) is unbounded, then, without loss of generality, we assume that \(u^r\rightarrow -\infty \) as \(r\rightarrow \infty \). Thus, it follows from (10) and (11) that
$$\begin{aligned} \begin{array}{rcl} 0&{}\le &{}\left\langle \frac{\mathbf {x}^r}{\Vert \mathbf {x}^r\Vert },\frac{\mathbf {y}-\sum \nolimits _{k=1}^{m-1}\mathcal {A}^{(k)}({\mathbf {x}^r})^{m-k}-\sum \nolimits _{p=1}^{l-1}\mathcal {B}^{(p)}({\mathbf {x}^r})^{l-p}-\mathbf {a}-\mathbf {b}}{\Vert \mathbf {x}^r\Vert ^{m-1}}\right\rangle +u^r\\ &{}\rightarrow &{} -\infty , \end{array} \end{aligned}$$which is a contradiction.
-
(ii)
If \(\{u^r\}\) is bounded, then there exists a convergent subsequence of \(\{u^r\}\), and we denote it as \(\{u^r\}\). We assume that \(u^r\rightarrow -\beta \) as \(r\rightarrow \infty \), then \(\beta \ge 0\). So, we have
$$\begin{aligned} \begin{array}{rcl} 0&{}\le &{}\lim _{r\rightarrow \infty }\left\langle \frac{\mathbf {x}^r}{\Vert \mathbf {x}^r\Vert },\frac{\mathbf {y}-\sum \nolimits _{k=1}^{m-1}\mathcal {A}^{(k)}({\mathbf {x}^r})^{m-k}-\sum \nolimits _{p=1}^{l-1}\mathcal {B}^{(p)}({\mathbf {x}^r})^{l-p}-\mathbf {a}-\mathbf {b}}{\Vert \mathbf {x}^r\Vert ^{m-1}}\right\rangle +\lim \limits _{r\rightarrow \infty }u_r\\ &{}=&{}-\langle (\bar{\mathbf {x}}),\mathcal {A}^{(1)}(\bar{\mathbf {x}})^{m-1}\rangle -\beta <0, \end{array} \end{aligned}$$which is also a contradiction.
Therefore, \(SOL(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) is nonempty.
Now, we show the uniqueness of solutions to \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\). Suppose that \(\mathbf {x}^*\) and \(\bar{\mathbf {x}}\) are two different solutions to \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\). Then,
which yields that
This is a contradiction with the condition (a). Therefore, \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) has a unique solution.\(\square \)
In the following, we give an example to show Theorem 2 cannot be covered by Theorem 1.
Example 1
Consider GVI(g, f, K), where
We show that f and g satisfy all conditions in Theorem 2, but do not satisfy the conditions of Theorem 1. We also give the unique solution of GPVI(g, f, K).
(i) Consider the condition (a) of Theorem 2. We show that function f is strictly monotone with respect to g on the set K. It is easy to see that
For any \(g(\mathbf {x}),g(\mathbf {y})\in K\), it follows that \(x_1\ge 0\), \(y_1\ge 0\), \(x_1\ne y_1\), and hence,
Therefore, the function f is strictly monotone with respect to g on the set K.
(ii) Consider the condition (b) of Theorem 2. In this example, the leading tensors of f and g are the same, i.e., \(\mathcal {A}^{(1)}=\mathcal {B}^{(1)}=(d_{i_1i_2i_3i_4})\in \mathbb {R}^{[4,2]}\) with \(d_{1111}=1\), \(d_{2222}=1\), and other entries being zero. Thus, it is obvious that
for any nonzero vector \(\mathbf {x}\in \mathbb {R}^2\), i.e., the condition (b) of Theorem 2 holds.
(iii) Consider the condition (c) of Theorem 2. For any \(\mathbf {x}\in \mathbb {R}^2{\setminus }\{\mathbf {0}\}\), it is obvious that
which implies that both the leading tensors of f and g are positive definite, i.e., the condition (c) of Theorem 2 holds.
(iv) Consider the condition (a) of Theorem 1. By Proposition 1, it is easy to see that f is not strongly monotone with respect to g on K, i.e., the condition (a) of Theorem 1 is not satisfied.
(v) Consider the condition (b) of Theorem 1. Suppose the condition (b) of Theorem 1 holds, i.e., there exists a vector \(\mathbf {u}\in g^{-1}(K)\) and positive scalars \(\alpha \) and L such that
Then, \(x_1\ge 0\), \(u_1\ge 0\), and
and hence,
Since \(u_1\ge 0\), it follows when \(x_1\rightarrow +\infty \) that the left-hand side of inequality (12) tends to +\(\infty \), which leads to a contradiction.
(vi) We give the unique solution of GPVI(g, f, K). In this example, we need to find \(x_1\in \mathbb {R}\) such that
Obviously, (13) has a unique solution \(x_1^*=0\). Thus, \(\mathbf {x}^*=(0,0)^\top \) is the unique solution of GPVI(g, f, K).
When \(K=\mathbb {R}^n_+\), \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) becomes the generalized polynomial complementarity problem investigated in [16], denoted by \(GPCP(\Lambda ,\mathbf {a},\Theta ,\mathbf {b})\). The following result is an immediate consequence of Theorem 2, which gives an existence and uniqueness result for \(GPCP(\Lambda ,\mathbf {a},\Theta ,\mathbf {b})\).
Corollary 1
Let f and g be defined by (2) and (3). Suppose that f is strictly monotone with respect to g on \(\mathbb {R}^n_+\), \(\langle \mathcal {A}^{(1)}(\mathbf {x})^{m-1},\mathcal {B}^{(1)}(\mathbf {x})^{l-1}\rangle \ne 0\) for any nonzero vector \(\mathbf {x}\in \mathbb {R}^n\), and one of \(\mathcal {A}^{(1)}\) and \(\mathcal {B}^{(1)}\), whose order is even and is lager than or equal to another, is positive definite, then for any given vectors \(\mathbf {a}\in \mathbb {R}^n\) and \(\mathbf {b}\in \mathbb {R}^n_+\), \(GPCP(\Lambda ,\mathbf {a},\Theta ,\mathbf {b})\) has a unique solution.
When \(g(\mathbf {x})=\mathbf {x}\), \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) becomes the polynomial variational inequality investigated in [10], denoted by \(PVI(K,\Lambda ,\mathbf {a})\). The following result is an immediate consequence of Theorem 2, which gives an existence and uniqueness result for \(PVI(K,\Lambda ,\mathbf {a})\).
Corollary 2
Let \(K\subseteq \mathbb {R}^n\) be a nonempty, closed and convex set with \(\mathbf {0} \in K\). Suppose that the function \(\sum _{k=1}^{m-1}\mathcal {A}^{(k)}\mathbf {x}^{m-k}\) is strictly monotone on K where \(\Lambda :=\Big (\mathcal {A}^{(1)},\ldots ,\mathcal {A}^{(m-1)}\Big )\in \mathcal {F}_{m,n}\) defined by (3), and the tensor \(\mathcal {A}^{(1)}\) is positive definite on K. Then for any given \(\mathbf {a} \in \mathbb {R}^n\), \(PVI(K,\Lambda ,\mathbf {a})\) has a unique solution.
We would like to emphasize that the results in Corollaries 1 and 2 are new ones on the existence and uniqueness for the generalized polynomial complementarity problem and the polynomial variational inequality, respectively.
When \(g(\mathbf {x})=\mathbf {x}\) and \(f(\mathbf {x})=\mathcal {A}\mathbf {x}^{m-1}+\mathbf {a}\) with \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) and \(\mathbf {a}\in \mathbb {R}^n\), \(GPVI(\Lambda ,\mathbf {a},\Theta ,\mathbf {b},K)\) reduces to the tensor variational inequality investigated in [26]. In this case, Corollary 2 reduces to [26, Theorem 4.1], since \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) is positive definite on K under the assumptions that \(\mathbf {0}\in K\) and f is strictly monotone on K. Furthermore, when \(K=\mathbb {R}^n_+\), this problem reduces to the tensor complementarity problem, and in this case, Corollary 2 can further reduce to [2, Corollary 1(b)] if the involved tensor is a Gram tensor. Moreover, for the tensor complementarity problem, some existence and uniqueness results were established in [1, 17].
4 Conclusions
The generalized polynomial variational inequality is a generalized variational inequality with the involved functions being polynomials, however, the known existence and uniqueness theorem for the generalized variational inequality cannot be generally applied to the generalized polynomial variational inequality, since the hypotheses of Theorem 1 do not hold for the general polynomial functions. In this paper, in terms of a characterization theorem related to exceptional family of elements and by using properties of polynomial functions, we established an existence and uniqueness theorem for the generalized polynomial variational inequality, by which we also obtained new existence and uniqueness theorems for the generalized polynomial complementarity problem and the polynomial variational inequality. They can be seen as extensions of the corresponding results for the tensor variational inequality and the tensor complementarity problem. By exploiting properties of polynomials, more theoretical results for the generalized polynomial variational inequality can be further developed.
References
Bai, X.L., Huang, Z.H., Wang, Y.: Global uniqueness and solvability for tensor complementarity problems. J. Optim. Theory Appl. 170(1), 72–84 (2016)
Balaji, R., Palpandi, K.: Positive definite and Gram tensor complementarity problems. Optim. Lett. 12, 639–648 (2018)
Che, M., Qi, L., Wei, Y.: Positive-definite tensors to nonlinear complementarity problems. J. Optim. Theory Appl. 168(2), 475–487 (2016)
Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, New York (1983)
Facchinei, F., Pang, J.S.: Finite Dimensionl Variatioal Inequalities and Complementarity Problems. Springer, New York (2003)
Gowda, M.S.: Polynomial complementarity problems. Pac. J. Optim. 13(2), 227–241 (2017)
Han, J., Xiu, N., Qi, H.D.: Nonlinear Complementarity Theory and Algorithms. Shanghai Science and Technology Press, Shanghai (2006)
Han, J., Huang, Z.H., Fang, S.C.: Solvability of variational inequality problems. J. Optim. Theory Appl. 122(3), 501–520 (2004)
Harker, P.T., Pang, J.S.: Finite-dimensional variational inequality and nonlinear complementarity problems: a survey of theory, algorithms and applications. Math. Program. 48(1), 161–220 (1990)
Hieu, Vu T.: Solution maps of polynomial variational inequalities. arXiv:1807.00321 (2018)
Huang, Z.H., Qi, L.: Formulating an \(n\)-person noncooperative game as a tensor complementarity problem. Comput. Optim. Appl. 66(3), 557–576 (2017)
Huang, Z.H., Qi, L.: Tensor complementarity problems—part I: basic theory. J. Optim. Theory Appl. 183(1), 1–23 (2019). https://doi.org/10.1007/s10957-019-01566-z
Huang, Z.H., Qi, L.: Tensor complementarity problems – part III: applications. J. Optim. Theory Appl. 183(3) (2019). https://doi.org/10.1007/s10957-019-01573-0
Huang, Z.H., Suo, Y.Y., Wang, J.: On \(Q\)-tensors. arXiv:1509.03088. To appear in Pac. J. Optim. (2016)
Ling, L., He, H., Ling, C.: On error bounds of polynomial complementarity problems with structured tensors. Optimization 67(2), 341–358 (2018)
Ling, L., He, H., Ling, C.: Properties of the solution set of generalized polynomial complementarity problems. arXiv:1905.00670v1. To appear in Pac. J. Optim. (2019)
Liu, D.D., Li, W., Vong, S.W.: Tensor complementarity problems: the GUS-property and an algorithm. Linear Multilinear Algebra 66(9), 1726–1749 (2018)
Lloyd, N.G.: Degree Theory. Cambridge University Press, London (1978)
Luo, Z., Qi, L., Xiu, N.: The sparsest solutions to \(Z\)-tensor complementarity problems. Optim. Lett. 11(3), 471–482 (2017)
Noor, M.A.: Quasi variational inequalities. Appl. Math. Lett. 1(4), 367–370 (1988)
Pang, J.S., Yao, J.C.: On a generalization of a normal map and equation. SIAM J. Control Optim. 33, 168–184 (1995)
Qi, L., Huang, Z.H.: Tensor complementarity problems—part II: solution methods. J. Optim. Theory Appl. 183(2) (2019). https://doi.org/10.1007/s10957-019-01568-x
Song, Y., Qi, L.: Tensor complementarity problem and semi-positive tensors. J. Optim. Theory Appl. 169(3), 1069–1078 (2016)
Song, Y., Yu, G.: Properties of solution set of tensor complementarity problem. J. Optim. Theory Appl. 170(1), 85–96 (2016)
Wang, Y., Huang, Z.H., Bai, X.L.: Exceptionally regular tensors and tensor complementarity problems. Optim. Method Softw. 31(4), 815–828 (2016)
Wang, Y., Huang, Z.H., Qi, L.: Global uniqueness and solvability of tensor variational inequalities. J. Optim. Theory Appl. 177(1), 137–152 (2018)
Zhao, Y.B.: Existence of a solution to nonlinear variational inequality under generalized positive homogeneity. Oper. Res. Lett. 25(5), 231–239 (1999)
Zhao, Y.B., Han, J., Qi, H.D.: Exceptional families and existence theorems for variational inequality problems. J. Optim. Theory Appl. 101(2), 475–495 (1999)
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This author’s work was supported by the National Natural Science Foundation of China (Grant Nos. 11431002 and 11871051).
Rights and permissions
About this article
Cite this article
Wang, J., Huang, ZH. & Xu, Y. Existence and uniqueness of solutions of the generalized polynomial variational inequality. Optim Lett 14, 1571–1582 (2020). https://doi.org/10.1007/s11590-019-01461-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11590-019-01461-6