238 lines
11 KiB
TeX
238 lines
11 KiB
TeX
\section{Convex Sets and Seminorms}
|
|
\label{section:seminorms}
|
|
|
|
|
|
\begin{definition}[Convex]
|
|
\label{definition:convex}
|
|
Let $E$ be a vector space over $K \in \RC$, then $A \subset E$ is \textbf{convex} if for any $x, y \in A$, $\bracs{\lambda x + (1 - \lambda) y| \lambda \in [0, 1]} \subset A$.
|
|
\end{definition}
|
|
|
|
\begin{definition}[Convex Hull]
|
|
\label{definition:convex-hull}
|
|
Let $E$ be a vector space over $K \in \RC$ and $A \subset E$, then the set
|
|
\[
|
|
\text{Conv}(A) = \bracs{\sum_{j = 1}^n t_j x_j \bigg | \seqf{t_j} \subset [0, 1], \seqf{x_j} \subset E, \sum_{j = 1}^n t_j = 1 }
|
|
\]
|
|
|
|
is the \textbf{convex hull} of $A$.
|
|
\end{definition}
|
|
|
|
\begin{definition}[Convex Circled Hull]
|
|
\label{definition:convex-circled-hull}
|
|
Let $E$ be a vector space over $K \in \RC$ and $A \subset E$, then the set
|
|
\[
|
|
\aconv(A) = \bracs{\sum_{j = 1}^n t_j x_j \bigg | \seqf{t_j} \subset K, \seqf{x_j} \subset E, \sum_{j = 1}^n |t_j| \le 1 }
|
|
\]
|
|
|
|
is the \textbf{convex circled hull} of $A$.
|
|
\end{definition}
|
|
|
|
|
|
|
|
\begin{lemma}
|
|
\label{lemma:convex-interior}
|
|
Let $E$ be a TVS over $K \in \RC$, $A \subset E$ be convex, $x \in A^o$, and $y \in \ol{A}$, then
|
|
\[
|
|
\bracs{tx + (1 - t)y|t \in (0, 1)} \subset A^o
|
|
\]
|
|
|
|
\end{lemma}
|
|
\begin{proof}[Proof, {{\cite[II.1.1]{SchaeferWolff}}}. ]
|
|
Fix $t \in (0, 1)$. Using translation, assume without loss of generality that $tx + (1 - t)y = 0$. In which case, $x = \alpha y$ where $\alpha = (1 - t)/t$. By (TVS2), $\alpha A^o \in \cn^o(y)$. Since $y \in \ol{A}$, $\alpha A^o \cap A \ne \emptyset$, so there exists $z \in A^o$ such that $\alpha z \in A$.
|
|
|
|
Let $\mu = \alpha/(\alpha - 1)$, then since $\alpha < 0$, $\mu \in (0, 1)$ and
|
|
\[
|
|
\mu z + (1 - \mu)\alpha z = \frac{\alpha z}{\alpha - 1} + \frac{(\alpha - 1 - \alpha)\alpha z}{\alpha - 1} = 0
|
|
\]
|
|
|
|
By (TVS1) and (TVS2),
|
|
\[
|
|
U = \underbrace{\bracs{\mu w + (1 - \mu)\alpha z|w \in A^o}}_{\subset A} \in \cn^o(0)
|
|
\]
|
|
|
|
so $0 \in A^o$.
|
|
\end{proof}
|
|
|
|
\begin{lemma}
|
|
\label{lemma:convex-gymnastics}
|
|
Let $E$ be a TVS over $K \in \RC$, $A, B \subset E$ be convex, then the following sets are convex:
|
|
\begin{enumerate}
|
|
\item $A^o$.
|
|
\item $\ol{A}$.
|
|
\item $A + B$.
|
|
\item For any $\lambda \in K$, $\lambda A$.
|
|
\end{enumerate}
|
|
\end{lemma}
|
|
\begin{proof}
|
|
(1): By \autoref{lemma:convex-interior}.
|
|
|
|
(2): Let $x, y \in \ol{A}$. By \autoref{definition:closure}, there exists filters $\fF, \mathfrak{G} \subset 2^A$ such that $\fF$ converges to $x$ and $\mathfrak{G}$ converges to $y$. In which case,
|
|
\[
|
|
\fU = \bracs{tE + (1 - t)F|t \in [0, 1], E \in \fF, F \in \mathfrak{G}} \subset 2^A
|
|
\]
|
|
|
|
converges to $tx + (1 - t)y$ by (TVS1) and (TVS2). Hence $tx + (1 - t)y \in \ol{A}$.
|
|
\end{proof}
|
|
|
|
\begin{proposition}
|
|
\label{proposition:convex-interior-closure}
|
|
Let $E$ be a TVS over $K \in \RC$ and $A \subset E$ be convex. If $A^o \ne \emptyset$, then $\ol{A} = \ol{A^o}$.
|
|
\end{proposition}
|
|
\begin{proof}[Proof, {{\cite[II.1.3]{SchaeferWolff}}}. ]
|
|
Since $A^o \subset A$, $\ol{A^o} \subset \ol{A}$. Let $x \in A^o$, then for any $y \in \ol{A}$,
|
|
\[
|
|
y \in \ol{\bracs{tx + (1 - t)y|t \in (0, 1)}} \subset \ol{A^o}
|
|
\]
|
|
|
|
by \autoref{lemma:convex-interior}.
|
|
\end{proof}
|
|
|
|
|
|
|
|
\begin{definition}[Sublinear Functional]
|
|
\label{definition:sublinear-functional}
|
|
Let $E$ be a vector space over $K \in \RC$, then a \textbf{sublinear functional} is a mapping $\rho: E \to \real$ such that:
|
|
\begin{enumerate}
|
|
\item $\rho(0) = 0$.
|
|
\item For any $x \in E$ and $\lambda \ge 0$, $\rho(\lambda x) = \lambda \rho(x)$.
|
|
\item For any $x, y \in E$, $\rho(x + y) \le \rho(x) + \rho(y)$.
|
|
\end{enumerate}
|
|
\end{definition}
|
|
|
|
|
|
\begin{definition}[Seminorm]
|
|
\label{definition:seminorm}
|
|
Let $E$ be a vector space over $K \in \RC$, then a \textbf{seminorm} on $E$ is a mapping $\rho: E \to [0, \infty)$ such that:
|
|
\begin{enumerate}
|
|
\item[(SN1)] $\rho(0) = 0$.
|
|
\item[(SN2)] For any $x \in E$ and $\lambda \in K$, $\rho(\lambda x) = \abs{\lambda} \rho(x)$.
|
|
\item[(SN3)] For any $x, y \in E$, $\rho(x + y) \le \rho(x) + \rho(y)$.
|
|
\end{enumerate}
|
|
\end{definition}
|
|
|
|
\begin{lemma}
|
|
\label{lemma:continuous-seminorm}
|
|
Let $E$ be a TVS over $K \in \RC$ and $[\cdot]: E \to [0, \infty)$ be a seminorm on $E$, then the following are equivalent:
|
|
\begin{enumerate}
|
|
\item $[\cdot]$ is uniformly continuous.
|
|
\item $[\cdot]$ is continuous.
|
|
\item $[\cdot]$ is continuous at $0$.
|
|
\item $\bracs{x \in E| [x] < 1} \in \cn_E(0)$.
|
|
\item $\bracs{x \in E| [x] \le 1} \in \cn_E(0)$.
|
|
\end{enumerate}
|
|
\end{lemma}
|
|
\begin{proof}
|
|
$(5) \Rightarrow (1)$: Let $x, y \in E$ and $r > 0$. If
|
|
\[
|
|
x - y \in \bracs{x \in E|[x] \le r} = r\bracs{x \in E|[x] \le 1} \in \cn_E(0)
|
|
\]
|
|
|
|
then $[x - y] \le r$.
|
|
\end{proof}
|
|
|
|
|
|
\begin{definition}[Topology Induced by Seminorm]
|
|
\label{definition:seminorm-topology}
|
|
Let $E$ be a vector space over $K \in \RC$ and $\seqi{[\cdot]}$ be seminorms, then:
|
|
\begin{enumerate}
|
|
\item For each $i \in I$, $d_i: E \times E \to [0, \infty)$ defined by $(x, y) \mapsto [x - y]_i$ is a pseudo-metric.
|
|
\item The topology induced by $\seqi{d}$ makes $E$ a topological vector space.
|
|
\item For each $i \in I$, $[\cdot]_i: E \to [0, \infty)$ is continuous.
|
|
\end{enumerate}
|
|
|
|
The topology induced by $\seqi{d}$ is the \textbf{vector space topology induced by} $\seqi{[\cdot]}$. In addition,
|
|
\begin{enumerate}
|
|
\item[(U)] For any family $\bracsn{[\cdot]_j}_{j \in J}$ of continuous seminorms on $E$, the vector space topology induced by $\bracsn{[\cdot]_j}_{j \in J}$ is contained in the vector space topology induced by $\seqi{[\cdot]}$.
|
|
\end{enumerate}
|
|
\end{definition}
|
|
|
|
|
|
\begin{definition}[Gauge/Minkowski Functional]
|
|
\label{definition:gauge}
|
|
Let $E$ be a vector space over $K \in \RC$ and $A \subset E$ be radial, then the mapping
|
|
\[
|
|
[\cdot]_A: E \to [0, \infty) \quad x \mapsto \inf\bracsn{\lambda > 0| \lambda^{-1}x \in A}
|
|
\]
|
|
|
|
is the \textbf{gauge/Minkowski functional} of $A$, and
|
|
\begin{enumerate}
|
|
\item For any $x \in E$ and $\lambda \ge 0$, $[\lambda x]_A = \lambda [x]_A$.
|
|
\item If $A$ is convex, then for any $x, y \in E$, $[x + y]_A \le [x]_A + [y]_A$.
|
|
\item If $A$ is circled, then for any $x \in E$ and $\lambda \in K$, $[\lambda x]_A = \abs{\lambda}[x]_A$.
|
|
\item If $A$ is circled, then $\bracs{\rho < 1} \subseteq A \subseteq \bracs{\rho \le 1} \subseteq \ol A$, with respect to any vector space topology on $E$.
|
|
\end{enumerate}
|
|
|
|
In particular,
|
|
\begin{enumerate}[start=4]
|
|
\item If $A$ is convex, then $[\cdot]_A$ is a sublinear functional.
|
|
\item If $A$ is convex and circled, then $[\cdot]_A$ is a seminorm.
|
|
\end{enumerate}
|
|
\end{definition}
|
|
\begin{proof}
|
|
(2): Let $\lambda, \mu > 0$ such that $\lambda^{-1}x, \mu^{-1}y \in A$. By convexity, $t\lambda^{-1} + (1 - t)\mu^{-1}y \in A$ for all $t \in [0, 1]$. Let $t \in [0, 1]$ such that
|
|
\[
|
|
(\lambda + \mu)^{-1}(x + y) = t\lambda^{-1}x + (1 - t)\mu^{-1}y
|
|
\]
|
|
|
|
then $(\lambda + \mu)^{-1} \in A$, and $\lambda + \mu \ge [x + y]_A$. Thus $[x + y]_A \le [x]_A + [y]_A$.
|
|
|
|
(4): Let $x \in \bracs{\rho \le 1}$, then $\lambda x \in A$ for all $\lambda \in (0, 1)$. Therefore
|
|
\[
|
|
x \in \overline{\bracs{\lambda x|\lambda \in (0, 1)}} \subset A
|
|
\]
|
|
|
|
so $x \in \overline{A}$.
|
|
\end{proof}
|
|
|
|
\begin{definition}[Locally Convex Space]
|
|
\label{definition:locally-convex}
|
|
Let $E$ be a TVS over $K \in \RC$, then the following are equivalent:
|
|
\begin{enumerate}
|
|
\item There exists a fundamental system of neighborhoods at $0$ consisting of convex sets.
|
|
\item There exists a fundamental system of neighbourhoods at $0$ consisting of convex, circled, and radial sets.
|
|
\item There exists a family of seminorms $\seqi{[\cdot]}$ that induces the topology on $E$.
|
|
\end{enumerate}
|
|
|
|
If the above holds, then $E$ is a \textbf{locally convex} space.
|
|
\end{definition}
|
|
\begin{proof}
|
|
$(1) \Rightarrow (2)$: Let $U \in \cn(0)$ be convex. By \autoref{proposition:tvs-good-neighbourhood-base}, there exists $V \in \cn(0)$ circled such that $V + V \subset U$. Let $W = \text{Conv}(V)$ be the convex hull of $V$, then $W \subset U$ is convex and circled.
|
|
|
|
$(2) \Rightarrow (3)$: For each $V \in \cn(0)$ convex, circled, and radial, let $[\cdot]_V: E \to [0, \infty)$ be its gauge, then $[\cdot]_V$ is a seminorm. For each $x, y \in X$ and $r > 0$, $[x - y]_V < r$ if and only if $x - y \in rV$. Thus the uniformity induced by $[\cdot]_V$ corresponds to the uniformity generated by $\bracs{U_{rV}| r > 0}$, where $U_V = \bracs{(x, y) \in E|x - y \in V}$. Since this holds for all $U \in \cn(0)$, the topology of $E$ and the topology induced by $\bracs{[\cdot]_V| V \in \cn(0), \text{ convex, circled, radial}}$ coincide.
|
|
|
|
$(3) \Rightarrow (1)$: For each $i \in I$ and $r > 0$, $\bracs{x \in E| [x]_i < r}$ is convex.
|
|
\end{proof}
|
|
|
|
\begin{definition}[Fréchet Space]
|
|
\label{definition:frechet-space}
|
|
Let $E$ be a locally convex space over $K \in \RC$, then $E$ is a \textbf{Fréchet space} if $E$ is first countable and complete.
|
|
\end{definition}
|
|
|
|
|
|
\begin{definition}[Associated Normed Space]
|
|
\label{definition:lc-associated-normed-space}
|
|
Let $E$ be a separated locally convex space and $A \subset E$ be convex and circled. Let $E_0 = \bigcup_{n \in \natp}nA$, $\rho_0: E_0 \to [0, \infty)$ be the gauge of $A$, and $(E_A, \rho_A)$ be the quotient of $E_0$ by $\bracs{\phi = 0}$, equipped with the quotient norm of $\rho_0$, then
|
|
\begin{enumerate}
|
|
\item $(E_A, \rho_A)$ is a normed space.
|
|
\end{enumerate}
|
|
|
|
If $A$ is radial, then $E_0 = E$ and the map $\pi_A: E \to E_A$ is the \textbf{canonical projection}, and
|
|
\begin{enumerate}[start=1]
|
|
\item If $A \in \cn_E(0)$, then $\pi_A \in L(E; E_A)$.
|
|
\end{enumerate}
|
|
|
|
If $(E_0, \rho_0)$ is separated, then $(E_0, \rho_0) = (E_A, \rho_A)$, and the map $\iota_A: E_A \to E$ is the \textbf{canonical inclusion}. In particular, if $A$ is bounded, then
|
|
\begin{enumerate}[start=2]
|
|
\item $(E_0, \rho_0)$ is separated.
|
|
\item $\iota_A \in L(E_A; E)$.
|
|
\end{enumerate}
|
|
|
|
The space $(E_A, \rho_A)$ is the \textbf{normed space associated with} $A$.
|
|
\end{definition}
|
|
\begin{proof}
|
|
(3): Let $x \in E_0 \setminus \bracs{0}$. Since $E$ is separated, there exists $U \in \cn_E(0)$ such that $x \not\in U$. As $A$ is bounded, there exists $\lambda > 0$ such that $\lambda U \supset A$. In which case, $x \not\in \lambda^{-1}A$, and $E_0$ is separated.
|
|
|
|
(4): Let $U \in \cn_E(0)$, then there exists $\lambda > 0$ such that $\lambda U \supset A$, so $\iota_A^{-1}(U) \supset \lambda^{-1}A \in \cn_{E_A}(0)$, and $\iota_A$ is continuous by \autoref{proposition:tvs-convex-morphism}.
|
|
\end{proof}
|
|
|