Fiest draft of nuclear operators.

This commit is contained in:
Bokuan Li
2026-07-13 15:26:05 -04:00
parent 3113e1da04
commit 84316a2059
5 changed files with 163 additions and 3 deletions

158
src/fa/lc/nuclear.tex Normal file
View File

@@ -0,0 +1,158 @@
\section{Nuclear Operators}
\label{section:nuclear-operator}
\begin{definition}[Nuclear Operator Between Banach Spaces]
\label{definition:nuclear-operator-normed}
Let $E, F$ be Banach spaces, $E^*$ be the dual of $E$, equipped with the uniform topology, and $T \in L(E; F)$, then $T$ is \textbf{nuclear} if there exists $\seq{\phi_n} \subset E^*$ and $\seq{y_n} \subset F$ such that:
\begin{enumerate}
\item For each $x \in E$, $Tx = \sum_{n = 1}^\infty y_n \dpn{x, \phi_n}{E}$.
\item $\sum_{n \in \natp}\norm{y_n}_F\norm{\phi_n}_{E^*} < \infty$.
\end{enumerate}
The set $N(E; F)$ is the \textbf{space of nuclear operators} from $E$ to $F$. For each $T \in N(E; F)$, let
\[
\norm{T}_{N(E; F)} = \inf\bracs{\sum_{n \in \natp}\norm{y_n}_F\norm{\phi_n}_{E^*} \bigg | Tx = \sum_{n = 1}^\infty y_n \dpn{x, \phi_n}{E} \forall x \in E}
\]
then $\norm{\cdot}_{N(E; F)}$ is a norm on $N(E; F)$, and $N(E; F)$ is a Banach space.
\end{definition}
\begin{lemma}
\label{lemma:nuclear-operator-normed-tensor}
Let $E, F$ be Banach spaces, $E^*$ be the dual of $E$, equipped with the uniform topology, then the mapping
\[
E^* \otimes F \to N(E; F) \quad \sum_{j = 1}^n \phi_j \otimes y_j \mapsto \sum_{j = 1}^n y_j\dpn{\cdot, \phi_j}{E}
\]
extends continuously into a surjective linear map $E^* \tilde \otimes_\pi F \to N(E; F)$.
\end{lemma}
\begin{definition}[Nuclear Operator]
\label{definition:nuclear-operator}
Let $E, F$ be separated locally convex spaces over $K \in \RC$ and $T \in L(E; F)$, then the following are equivalent:
\begin{enumerate}
\item There exists convex and circled sets $U \in \cn_E(0)$ and $B \in B(F)$ such that:
\begin{enumerate}[label=(\alph*)]
\item The auxiliary space $F_B$ is a Banach space.
\item $T(U) \subset B$.
\item The induced map $\tilde E_U \to F_B$ is nuclear.
\end{enumerate}
\item There exists an equicontinuous sequence $\seq{\phi_n} \subset E^*$, a convex, circled, and bounded subset $B \subset F$, $\seq{y_n} \subset B$, and $\seq{\lambda_n} \subset K$ such that
\begin{enumerate}[label=(\alph*)]
\item The auxiliary space $F_B$ is a Banach space.
\item For each $x \in E$, $Tx = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n}{E}$.
\item $\sum_{n \in \natp}|\lambda_n| < \infty$.
\end{enumerate}
\end{enumerate}
If the above holds, then $T$ is \textbf{nuclear}.
\end{definition}
\begin{proof}[Proof, {{\cite[Theorem III.7.1]{SchaeferWolff}}}. ]
(1) $\Rightarrow$ (2): Let $\pi: E \to E_U$ be the canonical projection map associated with $E_U$ and $\iota: F_B \to F$ be the canonical inclusion map associated with $F_B$. By assumption (1b), there exists an induced map $\hat T: E_U \to F_B$ such that the following diagram commutes:
\[
\xymatrix{
E \ar@{->}[r]^{T} \ar@{->}[d]_{\pi} & F \\
E_U \ar@{->}[r]_{\hat T} & F_B \ar@{->}[u]_{\iota}
}
\]
By the \hyperref[linear extension theorem]{theorem:linear-extension-theorem-normed}, $E_U^* = (\tilde E_U)^*$. Assume without loss of generality that $E_U$ is a Banach space, then (1c) implies that $\hat T \in L(E_U; F_B)$ is a nuclear operator. By \autoref{lemma:nuclear-operator-normed-tensor} and \autoref{theorem:metrisable-tensor-product}, there exists $\seq{\phi_n} \subset E_U^*$, $\seq{y_n} \subset F_B$, and $\seq{\lambda_n} \subset K$ such that:
\begin{enumerate}[label=(\roman*)]
\item $\sum_{n \in \natp}|\lambda_n| < \infty$.
\item $\limv{n}\phi_n = 0$ and $\limv{n}y_n = 0$.
\item For each $x \in E_U$, $\hat Tx = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n}{E_U}$.
\end{enumerate}
By (ii), $\sup_{n \in \natp}\norm{\phi_n}_{E_U^*} < \infty$ and $\sup_{n \in \natp}\norm{y_n}_{F_B} < \infty$, so $\seq{\phi_n}$ is equicontinuous, and there exists $R > 0$ such that $\seq{y_n} \subset RB$. After rescaling, assume without loss of generality that $\seq{y_n} \subset B$. By unraveling the factorisation, (iii) shows that for each $x \in E$,
\[
Tx = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n \circ \pi}{E}
\]
Therefore the decomposition using $\seq{\phi_n \circ \pi} \subset E^*$, $B \subset F$, $\seq{y_n} \subset B$, and $\seq{\lambda_n} \subset K$ given above satisfies (2).
(2) $\Rightarrow$ (1): Since $\seq{\phi_n}$ is equicontinuous, $U = \bigcap_{n \in \natp}\phi_n^{-1}(B_K(0, 1))$ is a convex and circled neighbourhood of $0$ in $E$.
(1b): Using assumption (2c) and rescaling, assume without loss of generality that $\sum_{n = 1}^\infty |\lambda_n| < 1$. Let $\rho: F_B \to [0, \infty)$ be the gauge of $B$, then for any $x \in U$,
\begin{align*}
\rho(Tx) &= \rho\braks{\sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n}{E}} \le \sum_{n \in \natp} |\lambda_n| \cdot \underbrace{|\dpn{x, \phi_n}{E}|}_{\le 1} \cdot \underbrace{\rho(y_n)}_{\le 1} \\
&\le \sum_{n \in \natp}|\lambda_n| < 1
\end{align*}
so $\rho(Tx) < 1$ and $Tx \in B$. Therefore $T(U) \subset B$.
(1c): Let $\pi: E \to E_U$ be the canonical projection map associated with $E_U$ and $n \in \natp$. By construction, $U \subset \phi_n^{-1}(B_K(0, 1))$, so there exists $\hat \phi_n \in E_U^*$ such that the following diagram commutes:
\[
\xymatrix{
E \ar@{->}[d]_{\pi} \ar@{->}[rd]^{\phi_n} & \\
E_U \ar@{->}[r]_{\hat \phi_n} & K
}
\]
Thus for each $x \in E$,
\[
Tx = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n}{E} = \sum_{n = 1}^\infty y_n \dpn{\pi(x), \lambda_n \hat \phi_n}{E}
\]
and the induced map $\hat T: \tilde E_U \to F_B$ takes the form $\hat Tx = \sum_{n = 1}^\infty y_n \dpn{x, \lambda_n \hat \phi_n}{\tilde E_U}$. Finally, for each $n \in \natp$, $U \subset \phi_n^{-1}(B_K(0, 1))$, and $\normn{\hat \phi_n}_{E_U^*} \le 1$. Similarly, since $y_n \in B$, $\norm{y_n}_{F_B} \le 1$ as well. Therefore
\[
\norm{\hat T}_{N(\tilde E_U; F_B)} \le \sum_{n \in \natp}|\lambda_n| \cdot \norm{y_n}_{F_B} \cdot \normn{\hat \phi_n}_{E_U^*} \le \sum_{n \in \natp}|\lambda_n| < \infty
\]
and $\hat T: \tilde E_U \to F_B$ is nuclear.
\end{proof}
\begin{proposition}
\label{proposition:nuclear-gymnastics}
Let $E, F, G, H$ be separated locally convex spaces and $S \in N(F; G)$, then:
\begin{enumerate}
\item $T$ is compact.
\item For any $T \in L(E; F)$, $S \circ T \in N(E; G)$.
\item For any $R \in L(G; H)$, $R \circ S \in N(F; H)$.
\item There exists a unique $\tilde S \in L(\tilde F; G)$ such that $\tilde S|_{F} = S$. Moreover, $\tilde S \in N(\tilde F; G)$.
\end{enumerate}
\end{proposition}
\begin{proof}[Proof, {{\cite[Corollary III.7.1.1-III.7.1.3]{SchaeferWolff}}}. ]
Let $\seq{\phi_n} \subset F^*$ be an equicontinuous sequence, $B \in B(G)$ be convex and circled, $\seq{y_n} \subset B$, and $\seq{\lambda_n} \subset K$ such that
\begin{enumerate}[label=(\alph*)]
\item The auxiliary space $G_B$ is a Banach space.
\item For each $x \in F$, $Tx = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n}{F}$.
\item $\sum_{n \in \natp}|\lambda_n| < \infty$.
\end{enumerate}
(1): Let $U = \bigcap_{n \in \natp}\phi_n^{-1}(B_K(0, 1))$, then since $\seq{\phi_n}$ is equicontinuous, $U$ is a convex and circled neighbourhood of $0$ in $F$. Given that $G_B$ is complete, $S$ is the following composition of continuous maps:
\[
\begin{CD}
U @>{\prod_{n \in \natp} \phi_n}>> \overline{B_K(0,1)}^{\natp} @>{x \mapsto \sum_{n=1}^\infty \lambda_n x_n y_n}>> G_B @>>> G
\end{CD}
\]
By \hyperref[Tychonoff's Theorem]{theorem:tychonoff}, $\overline{B_K(0,1)}^{\natp}$ is compact. Since $T(U)$ is contained in its image in the above diagram, $T(U)$ is relatively compact.
(2): Since $T \in L(E; F)$, for each $\seq{\phi_n \circ T} \subset E^*$ is equicontinuous. Thus for any $x \in E$,
\[
(S \circ T)x = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \phi_n \circ T}{E}
\]
and $S \circ T \in N(E; G)$.
(3): Using (1), assume without loss of generality that $B$ is also compact. In which case, $R(C)$ is a convex, circled, and compact set in $H$ containing $0$. Thus $H_{R(C)}$ is a Banach space. For each $x \in F$,
\[
(R \circ S)x = \sum_{n = 1}^\infty \lambda_n R(y_n) \dpn{x, \phi_n}{F}
\]
and $R \circ S \in N(F; H)$.
(4): By the \hyperref[linear extension theorem]{theorem:linear-extension-theorem-tvs}, $\seq{\phi_n} \subset F^*$ extend into an equicontinuous family $\bracsn{\tilde \phi_n}_1^\infty \subset \tilde F^*$. In which case, since $G_B$ is complete, the extension $\tilde S \in L(\tilde F; G)$ takes the form
\[
\tilde S x = \sum_{n = 1}^\infty \lambda_n y_n \dpn{x, \tilde \phi_n}{F}
\]
Therefore $\tilde S \in N(\tilde F; G)$.
\end{proof}

View File

@@ -135,7 +135,7 @@
(5): By (6) of \autoref{definition:projective-tensor-product}. (5): By (6) of \autoref{definition:projective-tensor-product}.
\end{proof} \end{proof}
\begin{theorem}[{{\cite[III.6.4]{SchaeferWolff}}}] \begin{theorem}
\label{theorem:metrisable-tensor-product} \label{theorem:metrisable-tensor-product}
Let $E, F$ be metrisable locally convex spaces over $K \in \RC$, then for any $z \in E \td{\otimes}_\pi F$, there exists $\seq{\lambda_n} \subset K$ and $\seq{(x_j, y_j)} \subset E \times F$ such that: Let $E, F$ be metrisable locally convex spaces over $K \in \RC$, then for any $z \in E \td{\otimes}_\pi F$, there exists $\seq{\lambda_n} \subset K$ and $\seq{(x_j, y_j)} \subset E \times F$ such that:
\begin{enumerate} \begin{enumerate}
@@ -146,7 +146,7 @@
\end{theorem} \end{theorem}
\begin{proof} \begin{proof}[Proof, {{\cite[III.6.4]{SchaeferWolff}}}.]
Let $\seq{p_n}$ and $\seq{q_n}$ be increasing sequences of continuous seminorms that induce the topology on $E$ and $F$, respectively. For each $n \in \natp$, let $r_n = p_n \otimes q_n$, and $\td r_n$ be the continuous extension of $r_n$ to $E \td{\otimes}_\pi F$. Let $\seq{p_n}$ and $\seq{q_n}$ be increasing sequences of continuous seminorms that induce the topology on $E$ and $F$, respectively. For each $n \in \natp$, let $r_n = p_n \otimes q_n$, and $\td r_n$ be the continuous extension of $r_n$ to $E \td{\otimes}_\pi F$.
Let $u \in E \td{\otimes}_\pi F$, then there exists $\seq{u_n} \subset E \otimes_\pi F$ such that $\td r_n(u - u_n) < 2^{-n}/n^2$ for all $n \in \natp$. For each $N \in \natp$, let $v_N = u_{N+1} - u_N$, then Let $u \in E \td{\otimes}_\pi F$, then there exists $\seq{u_n} \subset E \otimes_\pi F$ such that $\td r_n(u - u_n) < 2^{-n}/n^2$ for all $n \in \natp$. For each $N \in \natp$, let $v_N = u_{N+1} - u_N$, then

View File

@@ -26,6 +26,7 @@
$E \otimes_\pi F$ & Projective tensor product of $E$ and $F$. & \autoref{definition:projective-tensor-product} \\ $E \otimes_\pi F$ & Projective tensor product of $E$ and $F$. & \autoref{definition:projective-tensor-product} \\
$E \,\widetilde{\otimes}_\pi F$ & Projective completion of $E$ and $F$. & \autoref{definition:projective-tensor-product} \\ $E \,\widetilde{\otimes}_\pi F$ & Projective completion of $E$ and $F$. & \autoref{definition:projective-tensor-product} \\
$p \otimes q$ & Cross seminorm of $p$ and $q$. & \autoref{definition:cross-seminorm} \\ $p \otimes q$ & Cross seminorm of $p$ and $q$. & \autoref{definition:cross-seminorm} \\
$N(E; F)$ & Nuclear mappings from $E$ to $F$. & \autoref{definition:nuclear-operator-normed} \\
% ---- Order Structures ---- % ---- Order Structures ----
$x \vee y$, $x \wedge y$ & $\sup$ and $\inf$ in vector lattice. & \autoref{definition:vector-lattice} \\ $x \vee y$, $x \wedge y$ & $\sup$ and $\inf$ in vector lattice. & \autoref{definition:vector-lattice} \\
$|x|$ & Absolute value $x \vee (-x)$ in a vector lattice. & \autoref{definition:order-absolute-value} \\ $|x|$ & Absolute value $x \vee (-x)$ in a vector lattice. & \autoref{definition:order-absolute-value} \\

View File

@@ -119,7 +119,7 @@
\phi: A \to \complex \quad x \mapsto \dpn{\pi(x)\xi, \xi}{H} \phi: A \to \complex \quad x \mapsto \dpn{\pi(x)\xi, \xi}{H}
\] \]
is a state on $A$. Let $(H_\phi, \pi_\phi, \xi_\phi)$ be the GNS triple associated with $\phi$, then there exists a unitary equivalence $U: H \to H_\phi$ such that $U\xi = \xi_\phi$. is a state on $A$. Moreover, if $(H_\phi, \pi_\phi, \xi_\phi)$ is the GNS triple associated with $\phi$, then there exists a unitary equivalence $U: H \to H_\phi$ such that $U\xi = \xi_\phi$.
\item For each $\mathcal{S} \subset S(A)$, the mapping \item For each $\mathcal{S} \subset S(A)$, the mapping
\[ \[
\pi_{\mathcal{S}}: A \to B([l^2(\mathcal{S}); H_\phi]) \quad \pi_{\mathcal{S}}(x)(\eta)_\phi = \pi_{\phi}(x)(\eta_\phi) \pi_{\mathcal{S}}: A \to B([l^2(\mathcal{S}); H_\phi]) \quad \pi_{\mathcal{S}}(x)(\eta)_\phi = \pi_{\phi}(x)(\eta_\phi)

View File

@@ -18,6 +18,7 @@
$S(A)$ & State space of a $C^*$-algebra $A$. & \autoref{definition:cstar-state} \\ $S(A)$ & State space of a $C^*$-algebra $A$. & \autoref{definition:cstar-state} \\
$P(A)$ & Pure state space of a $C^*$-algebra $A$. & \autoref{definition:pure-state} \\ $P(A)$ & Pure state space of a $C^*$-algebra $A$. & \autoref{definition:pure-state} \\
$\dpn{x, y}{\phi}$ & Defined as $\dpn{y^*x, \phi}{A}$, the pseudo inner product associated to a positive linear functional. & \autoref{definition:cstar-state-pseudo-inner-product} \\ $\dpn{x, y}{\phi}$ & Defined as $\dpn{y^*x, \phi}{A}$, the pseudo inner product associated to a positive linear functional. & \autoref{definition:cstar-state-pseudo-inner-product} \\
$(H_\phi, \pi_\phi, \xi_\phi)$ & GNS triple associated with $\phi \in S(A)$. & \autoref{definition:gns-triple} \\
$M_n(\complex)$ & Algebra of $n \times n$ matrices over $\complex$. & \autoref{definition:matrix-algebra} \\ $M_n(\complex)$ & Algebra of $n \times n$ matrices over $\complex$. & \autoref{definition:matrix-algebra} \\
$B(H)$ & Algebra of bounded operators on a Hilbert space. & \autoref{definition:hilbert-endomorphism} \\ $B(H)$ & Algebra of bounded operators on a Hilbert space. & \autoref{definition:hilbert-endomorphism} \\