I have been studying the applications of geometric Hahn-Banach Theorems. I came across this book Functional Analysis by S. Kesavan where the geometric Hahn-Banach Theorem is applied on Vector-Valued Integration(here, vector-valued integration does not mean the integration of vector-valued function. It means that the integral of a function must be a vector in a Banach space.). But I have a very hard time understanding exactly how it is applied.

Theorem Let $\varphi:[0,1]\rightarrow V$ be a continuous mapping into a real Banach space $X$. Then, the integral of $\varphi$ over $[0,1]$ exists.
Since $[0,1]$ is compact, the set $\overline{P}$ which is the closure (in $X$) of the set $P$ which is the convex hull of $\varphi([0,1])$,(i.e., the smallest convex set containing $\varphi([0,1])$)), is compact, by completeness of $X$.
Let $L$ be an arbitrary finite collection of continuous linear functionals on $V$. Define $$E_L=\{y\in \overline{P}:f(y)=\int_{0}^{1}f(\varphi(t))\mathrm{d}t\;\mathrm{for\;all}\; f\in L\}.$$ It is immediate to see that $_L$ is a closed set.
Step 1: For any such finite collection $L$ of continuous linear functionals, $E_L\ne \phi$. To see this, let $L=\{f_1,...,f_k\}$. Define $\mathcal{A}:X\rightarrow \mathbb{R}^{k}$ by $$\mathcal{A}(x)=(f_1(x),...,f_k(x)).$$ Then, $\mathcal{A}$ is a continuous linear transformation and so $K=\mathcal{A}(\overline{P})$ is a compact and convex set. If $(t_1,...,t_k)\notin K$, then, by the Hahn-Banach Theorem, we can find constants $c_1,...,c_k$ such that $$\sum_{i=1}^{k}c_iu_i<\sum_{i=1}^{k}c_it_i$$ for all $(u_1,...,u_k)\in K$. In particular, for all $t\in [0,1]$, we have $$\sum_{i=1}^{k}c_if_i(\varphi(t))< \sum_{i=1}^{k}c_it_i.$$ Integrating this inequality over $[0,1]$, we get $$\sum_{i=1}^{k}c_im_i < \sum_{i=1}^{k}c_it_i$$ where $$m_i=\int_{0}^{1}f_i(\varphi(t))\mathrm{d}t.$$ In other words, if $(t_1,...,t_k)\notin K$, then $(t_1,...,t_k)\ne (m_1,...,m_k)\in K$. Thus, there exists $y\in \overline{P}$ such that, for $1\le i\le k$, we have $$m_i=f_i(y).$$ This means that $y\in E_L$, i.e., $E_L$ is non-empty.
Step 2: Let $I$ be a finite indexing set and $L_i$ be finite collections of elements of $X^*$ for each $i\in I$. Then, $L=\underset{i \in I}{\bigcup} L_i$ is still finite and further, it is easy to see that, $$\underset{i\in I}{\bigcap} E_{L_i}= E_L.$$
It now follows from the previous step that the class of closed sets, $$\{E_L: L \;\mathrm{is\; a\; finite\; subset\; of}\; X^*\},$$ has finite intersection property. Since $\overline{P}$ is compact, it now follows that $$\underset{L,\;\mathrm{finite\; subset\; of}\; X^*}{\bigcap} E_L \ne \phi.$$
In particular, there exists $y$ such that $y\in E_{(f)}$ for every $f\in X^*$, i.e., $y$ satisfies $$f(y)=\int_{0}^{1}f(\varphi(t))\mathrm{d}t,$$ for every $f\in V^*$. Thus, $y=\int_{0}^{1}\varphi(t)\mathrm{d}t$. This completes the proof.

According to the article, the Hahn-Banach Theorem applied here is:

Let $A$ and $B$ be non-empty and disjoint convex sets in a real normed linear space $X$. Assume that $A$ is closed and that $B$ is compact. Then, $A$ and $B$ can be separated strictly by a closed hyperplane, i.e. there exists $f\in X^*,\; a\in \mathbb{R}$ and $\epsilon >0$ such that $f(x)\le \alpha -\epsilon$ and $f(y)\ge \alpha -\epsilon $ for all $x\in A$ and $y\in B$.

The problem I have is this:
We have, $K=\mathcal{A}(\varphi(Q))$ as a compact and convex set.
But for the theorem, mentioned above, to be applied, we need another closed convex set and that must be disjoint from $K$ and constant $\alpha$. The article does not specify the other set. Why? Can somebody elaborate on it, please?


The other set is the singleton $\{(t_1, \dots, t_k)\}$. There is then a linear functional $g \in (\mathbb{R}^k)^*$ such that $g((t_1, \dots, t_k)) > g(v)$ for all $v \in K$. But a linear functional $g$ on $\mathbb{R}^k$ is necessarily of the form $g((v_1, \dots, v_k)) = \sum_{i=1}^k c_i v_i$ for some scalars $c_1, \dots, c_k$, so we get the result.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.