An application of Separation Theorem

246 Views Asked by At

Let $X$ be a Hausdorff locally convex topological vector space. Suppose $X_0 \subset X$ is nonempty convex set, $g:\; X\to \mathbb R^m$ is a convex vector function (each component $g_i(x): X\to \mathbb R$ is convex function, $i=1,2,\ldots,m$). Prove that, if there is no $x\in X_0$ such that $g(x)<0$ then there exists $\lambda\in \mathbb R_{+}^m\setminus\{0\}$ such that $\langle \lambda, g(x)\rangle \geq 0$ for all $x\in X_0$.

Here is my attempt: Suppose that there is no $x\in X_0$ such that $g(x)<0$. For each $x_0\in X_0$, define

$$V(x)=\{\beta \in \mathbb R^m:\; g(x)<\beta\}\; \mbox{ and }\; V=\cup_{x\in X_0}V(x).$$ where $g(x)<\beta$ means $g_i(x)<\beta_i$ for each $i=1, \ldots, m$. Assume that $\beta^{(1)}\in V(x_1)$, $\beta^{(2)}\in V(x_2)$ then for $\lambda \in (0,1)$ we have $$g(\lambda x_1 + (1-\lambda)x_2)\leq \lambda g(x_1)+(1-\lambda)g(x_2)\leq \lambda \beta^{(1)}+(1-\lambda)\beta^{(2)}.$$ Thus $\lambda \beta^{(1)}+(1-\lambda)\beta^{(2)} \in V(\lambda x_1 + (1-\lambda)x_2)$ and since $\lambda x_1 + (1-\lambda)x_2 \in X_0$, we have $V$ is convex. $V\neq \emptyset$ because of the nonemptyness of $X_0$ [Take $x_0\in X_0$, and $\beta^{(0)}=g(x_0)+(1, \ldots, 1) \in V$]. By the hypothesis, $0\notin V$. Applying Separation Theorem, there exists $\lambda\in \mathbb R^m\setminus\{0\}$ such that

$$\langle \lambda, \beta\rangle\leq 0, \mbox{ for all } \beta\in V.$$ Noting that if $\beta\in V$ then $\beta +(p,\ldots,p) \in V$ for any $p>0$, from above inequality we can imply $\lambda \geq 0$, i.e., $\lambda_i \geq 0$ for all $i=1, \ldots, m$. On the other hand for each $x\in X_0$, let $\beta=g(x)+(1/n,1/n, \ldots, 1/n) \in V$, we have $$\langle \lambda, g(x)\rangle + \sum_{i=1}^m \frac{\lambda_i}{n} \geq 0.$$ Let $n\to \infty$ we conclude that $\langle \lambda, g(x)\rangle \geq 0$, for each $x\in X_0$. QED.

Is my proof right? Do we need any assumption on interior in above proof when using Separation Theorem?

Thanks in advance.

1

There are 1 best solutions below

3
On BEST ANSWER

The proof looks good. You do not need any assumption on the interior as you are separating a point from a convex set. You would need a non-emptyness condition on the interior if you would like to separate two convex sets in infinite-dimensional spaces.