On some properties of set-valued tensor complementarity problem

Let F(v)= \max_{\omega \in \Omega(v)} \mathcal{B}(\omega)v^{m-1} and R_i (\mathcal{B})(\omega) denote the ith row subtensor of \mathcal{B}(\omega). Hence F_i(v)= \max_{\omega \in \Omega(v)} R_i (\mathcal{B})(\omega)v^{m-1}. For a given v_0 \in \mathbb{R}^n then for \varepsilon > 0 \exists \; \omega_0 \in \Omega(v_0) such that R_i (\mathcal{B})(\omega_0)v_0^{m-1} > F_i(v_0) - \varepsilon . Since \Omega(v) is inner semicontinuous by Theorem 2.2 and Theorem 2.3, for any sequence \{v_n\} such that v_n \to v_0, \; \exists \; \{\omega_n\} \subset \Omega(v_n), \omega_n \to \omega_0. This implies F_i(v_n)= \max_{\omega \in \Omega(v_n)} R_i (\mathcal{B})(\omega)v_n^{m-1} \geq R_i (\mathcal{B})(\omega_n)v_n^{m-1}. This implies,

Here the equality follows from the continuity of R_i (\mathcal{B})(\omega) which is ensured by the continuity of \mathcal{B}(\omega). Since \varepsilon >0 and \{v_n\} is an arbitrary sequence converging to v_0, we obtain

This implies the lower semicontinuity of F_i. Again,

Therefore,

Let C^{\prime}\neq \emptyset. If \Bar{v} \in C^{\prime}, then

which is equivalent to

Again, \liminf_{t\to 0^+} F(\Bar{v} + t e) \geq F(\Bar{v}) >0, and \Bar{v} + t e>0 for t>0. For sufficiently small t>0 we have (\Bar{v} + t e) \in C. Hence C\neq \emptyset.

Let the mapping p(\omega) be bounded below. Therefore \exists \; \gamma \in \mathbb{R}, such that p(\omega) \geq \gamma e. Suppose v_0 and \omega_0 satisfy the given assumption. Hence we obtain

Then, for any \Bar{N} \in N_{\infty}, we have \omega_0 \in \cup_{n\in \Bar{N}} \Omega (n v_0). In particular, we observe the following:

(a) If \Bar{N}= \{ 1,2,... \}, then \exists \; n_1, such that \omega_0 \in \Omega(n_1 v_0);

(b) If \Bar{N}= \{ n_1 +1,... \} then \exists \; n_2 with n_2> n_1, such that \omega_0 \in \Omega(n_2 v_0).

By this process we find a sequence \{n_l\} such that \omega_0 \in \Omega(n_l v_0) and n_l \to \infty. Since \mathcal{B}(\omega_0) v_0^{m-1} >0, \exists \; \delta > 0, such that \mathcal{B}(\omega_0) v_0^{m-1} > \delta e. Taking l large enough to satisfy n_l > \max\{ -\frac{\gamma}{\delta}, 0 \} we write \delta n_l e > -\gamma e \geq -p(\omega). This implies that

Hence,

This implies n_l v_0 is a feasible point of SVTCP(\mathcal{B}(\omega), p(\omega), \Omega).

(a) We prove this part by showing that if \exists any non-zero solution then the set of tensors \{ \mathcal{B}(\omega) : \omega \in \Omega(v) \} is not strongly semi-positive. Note that, for any positive mapping p, \; v = 0 is a solution of SVTCP(\mathcal{B}(\omega),p(\omega), \Omega). We assume that there is another nonzero solution \Tilde{v}, i.e, \exists\; \Tilde{ \omega} \in \Omega(\Tilde{v}), such that

We want to prove that the set of tensors \{ \mathcal{B}(\omega) : \omega \in \Omega(v) \} is not strongly semi-positive. Let the set of tensors \{ \mathcal{B}(\omega) : \omega \in \Omega(v) \} be strongly semi-positive. Then there exists k\in [n] such that \Tilde{v}_k >0 and (\mathcal{B}(\Tilde{\omega})\Tilde{v}^{m-1})_k \geq 0, and hence (\mathcal{B}(\Tilde{\omega})\Tilde{v}^{m-1} + p(\Tilde{\omega}))_k > 0, which contradicts condition (Proof). Hence the set of tensors \{ \mathcal{B}(\omega) : \omega \in \Omega(v) \} is not strongly semi-positive.

(b) We prove this part by showing that if the set of tensors \{ \mathcal{B}(\omega) : \omega \in \Omega(v) \} is not weakly semi-positive then the SVTCP(\mathcal{B}(\omega),p(\omega), \Omega) with p(\omega) >0 does not have zero as its unique solution. We assume that \{ \mathcal{B}(\omega) : \omega \in \Omega(v) \} is not weakly semi-positive. Then, there exists a nonzero \Tilde{v} \geq 0, for all k\in I^+(\Tilde{v}) = \{ i: \Tilde{v}_i > 0, (\mathcal{B}(\omega)\Tilde{v}^{m-1})_k < 0 \} for all \omega \in \Omega (\Tilde{v}). Choose \tilde{\omega} \in \Omega (\Tilde{v}). Let p(\omega)= 1 for all \omega \neq \tilde{\omega} and

Therefore p(\omega) >0 for all \omega. According to the above construction, we have

This implies the nonzero vector \Tilde{v} is a solution of SVTCP(\mathcal{B}(\omega),p(\omega), \Omega).

Since set of tensor \{\mathcal{B}(\omega): \omega\in \Omega(v)\} is strongly semipositive, \forall\; p(\omega)>0 SVTCP(\mathcal{B}(\omega),p(\omega), \Omega) has zero as its unique solution. Also SOL\{\mathcal{B}(\omega), 0, \Omega\} = \{0\}. Thus for every \omega \in \Omega, \; \exists \; d>0 such that for TCP(\mathcal{B}(\omega), 0) and TCP(\mathcal{B}(\omega), d) zero is the only solution. Therefore by Corollary 3.2, Item (i) of [gowda2015z] we conclude that \mathcal{B}(\omega) is a Q-tensor for each \omega \in \Omega (v). This implies SOL(\mathcal{B}(\omega), p(\omega),\Omega) is nonempty.

Consider the SVTCP(\mathcal{B}(\omega), p(\omega), \Omega). Suppose \omega_0\in \Omega is such that \mathcal{B}(\omega_0) is a P-tensor then SOL(\mathcal{B}(\omega_0), p(\omega_0)) is nonempty and compact. Also suppose that \Omega^{-1}(\omega_0) \subseteq SOL(\mathcal{B}(\omega_0), p(\omega_0)). Now from the Theorem 17 of [zhou2013set], we know

Since \Omega^{-1}(\omega_0) \subseteq SOL(\mathcal{B}(\omega_0), p(\omega_0)) and SOL(\mathcal{B}(\omega_0), p(\omega_0)) is nonempty, we have

We establish the result by showing that if r(v) is not level bounded for the SVTCP(\mathcal{B}(\omega),p(\omega), \Omega) where \exists a bounded set \Omega, such that \Omega(v) \subset \Omega \; \; \forall\; v\in \mathbb{R}^n and \mathcal{B}(\omega) and p(\omega) are continuous on \Omega then the set of tensors \{ \mathcal{B}(\omega): \omega \in \Omega(v) \} does not have the limit-R_0 property. By this approach we first assume that r(v) is not level bounded. Then \exists a sequence \{v_n\}, such that \|v_n\| \to \infty as n \to \infty and \{r(v_n)\} is bounded. Then,

the minimizer is attained at \omega_n \in \Omega (v_n), whose existence is ensured by the compactness of \Omega(v_n), since \Omega(v) is closed and \Omega is bounded. Taking a subsequence if necessary, we can assume that \frac{v_n}{\|v_n\|^{m-1}} converges to v_0 and \{\omega_n\} converges to \omega_0. We have,

Since p(\omega) is continuous and \omega_n \in \Omega is bounded we have \lim_{n\to \infty} \frac{p(\omega_n)}{\|v_n\|^{m-1}}=0.

Taking limit on both sides of the equation (Proof) as n\to \infty we obtain,

Since v_0 is a nonzero vector, this implies that the set of tensors \{ \mathcal{B}(\omega): \omega \in \Omega(v) \} does not have limit-R_0 property.

We prove this by showing that if there exists v_0 \neq 0 and \omega_0 \in \cap_{\Bar{N} \in N_{\infty}} \cup_{n\in \Bar{N}} \Omega(nv_0) for the SVTCP(\mathcal{B}(\omega),p(\omega), \Omega) with p(\omega) and \mathcal{B}(\omega), continuous on the compact set \Omega and \Omega(v) \subset \Omega, \; \forall v\in \mathbb{R}^n such that v_0\geq 0,\; \; \mathcal{B}(\omega_0)v_0^{m-1} \geq 0,\;\; \mathcal{B}(\omega_0)v_0^m =0 then the merit function r(v) = \min_{\omega \in \Omega(v)} \| \min(v, \mathcal{B}(\omega)v^{m-1} + p(\omega))\| is not level bounded. Suppose \exists \; v_0 \neq 0 and \omega_0 \in \cap_{\Bar{N} \in N_{\infty}} \cup_{n\in \Bar{N}} \Omega(nv_0) such that

Then for any \Bar{N} \in N_{\infty}, we have \omega_0 \in \cup_{n\in \Bar{N}} \Omega (n v_0). In particular, we observe the following:

(a) If \Bar{N}= \{ 1,2,... \}, then \exists \; n_1, such that \omega_0 \in \Omega(n_1 v_0);

(b) If \Bar{N}= \{ n_1 +1,... \} then \exists \; n_2 with n_2> n_1, such that \omega_0 \in \Omega(n_2 v_0).

By this process we find a sequence \{n_l\} such that \omega_0 \in \Omega(n_l v_0) and n_l \to \infty. Therefore,

Now we consider the following cases.

Case-1: We consider the case when (v_0)_i>0. Then using (Proof) we conclude (\mathcal{B}(\omega_0) v_0^{m-1})_i = 0. Imposing the continuity of the function p(\omega) and the compactness of the set \Omega we conclude that \max_{\omega \in \Omega}p(\omega) is bounded. This implies that \exists sufficiently large k such that \max_{\omega \in \Omega}p_i(\omega) < n_l (v_0)_i. Hence,

Case-2: We consider the case when (v_0)_i=0. Here we consider the following two subcases.

Subcase-1: When n_l^{m-1} (\mathcal{B}(\omega_0)v_0^{m-1})_i + p_i(\omega_0) \geq 0, we have

Subcase-2: When n_l^{m-1} (\mathcal{B}(\omega_0)v_0^{m-1})_i + p_i(\omega_0) < 0, we obtain

using the fact p_i(\omega_0) \leq n_l^{m-1}(\mathcal{B}(\omega_0)v_0^{m-1})_i + p_i(\omega_0) < 0. Thus combining we have

Putting the facts (?), (?) and (?) together we obtain,

This implies that r(v) is not level bounded.