1: %%!TEX root = OTERN.tex
2:
3:
4: \chapter{Some functional analysis}
5: \label{sec:functional-analysis}
6:
7: Since this presentation addresses disparate audiences, we found it helpful to organize tools from functional analysis and the theory of unbounded operators in appendix sections. The reader may find the references \cite{Rud91, KadisonRingroseI, DuSc88} to be helpful.
8:
9: %A recurrent theme is that the energy Hilbert space \HE has a number of separate but unitarily equivalent incarnations. These different facets are necessary for computing the resistance metric in different ways, and several other applications.
10:
11: The magic of Hilbert space resides in the following innocent-looking axioms:
12: \begin{enumerate}
13: \item A complex vector space \sH.
14: \item A complex-valued function on $\sH \times \sH$, denoted by $\la \cdot,\cdot\ra$ and satisfying
15: \begin{enumerate}[(a)]
16: \item For every $v \in \sH$, $\la v, \cdot\ra : \sH \to \bC$ is linear.
17: \item For every $v_1,v_2 \in \sH$, $\la v_1,v_2\ra = \cj{\la v_2,v_1\ra}$.
18: \item For every $v \in \sH$, $\la v,v\ra \geq 0$, with equality if and only if $v=0$.
19: \end{enumerate}
20: \item Under the norm defined by $\|v\|_\sH := \la v,v\ra^{1/2}$, \sH is complete.
21: \end{enumerate}
22:
23: \begin{exm}[Square-summable sequences]
24: For a \bC-valued function $v$ on the integers, let $\|v\|_2 := \left(\sum_{n \in \bZ} |v(n)|^2\right)^{1/2}$. Then
25: \[\ell^2(\bZ) := \{v:\bZ \to \bC \suth \|v\|_2 < \iy\}\]
26: is a Hilbert space.
27: \end{exm}
28:
29: \begin{exm}[Classical $L^2$-spaces]
30: For a measurable \bC-valued function $v$ on a measure space $(X,\gm)$, let $\|v\|_2 := \left(\int_X |v(x)|^2 \,d\gm(x) \right)^{1/2}$. Then
31: \[L^2(\gm) := \{v:X \to \bC \suth \|v\|_2 < \iy\}\]
32: is a Hilbert space.
33: \end{exm}
34:
35: \section{von Neumann's embedding theorem}
36: \label{sec:von-Neumann's-embedding-theorem}
37:
38: \begin{theorem}[von Neumann]\label{thm:vonNeumann's-embedding-thm}
39: Suppose $(X,d)$ is a metric space. There exists a Hilbert space $\sH$ and an embedding $w:(X,d) \to \sH$ sending $x \mapsto w_x$ and satisfying
40: \linenopax
41: \begin{equation}\label{eqn:vonNeumann's-embedding-thm}
42: d(x,y) = \|w_x - w_y\|_\sH
43: \end{equation}
44: if and only if $d^2$ is negative semidefinite.
45: \end{theorem}
46:
47: \begin{defn}\label{def:negative-semidefinite}
48: A function $d:X \times X \to \bR$ is \emph{negative semidefinite} iff for any $f:X \to \bR$ satisfying $\sum_{x \in X} f(x) = 0$, one has
49: \linenopax
50: \begin{equation}\label{eqn:negative-semidefinite}
51: \sum_{x,y \in F} f(x) d^2(x,y) f(y) \leq 0,
52: \end{equation}
53: where $F$ is any finite subset of $X$.
54: \end{defn}
55:
56: von Neumann's theorem is constructive, and provides a method for obtaining the embedding, which we briefly describe, continuing in the notation of Theorem~\ref{thm:vonNeumann's-embedding-thm}.
57:
58: \begin{FlatList}
59: \item Schwarz inequality. If $d$ is a negative semidefinite function on $X \times X$, then define a positive semidefinite bilinear form on functions $f,g:X \to \bC$ by
60: \linenopax
61: \begin{align} \label{eqn:vNeu-inner-prod}
62: Q(f,g) = \la f,g\ra_Q := - \sum_{x,y} f(x) d^2(x,y) g(y).
63: \end{align}
64: One obtains a quadratic form $Q(f) := Q(f,f)$, and checks that the generalized Schwarz inequality holds $Q(f,g)^2 \leq Q(f) Q(g)$ by elementary methods.
65: \item The kernel of $Q$. Denote the collection of finitely supported functions on $X$ by $\Fin(X)$ and define
66: \linenopax
67: \begin{align}\label{eqn:def:Fin0(X)}
68: \Fin_0(X) := \{f \in \Fin(X) \suth {\textstyle \sum}_x f(x) = 0\}.
69: \end{align}
70: The idea is to complete $\Fin_0(X)$ with respect to $Q$, but first one needs to identify functions that $Q$ cannot distinguish. Define
71: \linenopax
72: \begin{align}\label{eqn:def:ker(Q)}
73: \ker Q = \{f \in \Fin_0(X) \suth Q(f)=0\}.
74: \end{align}
75: It is easy to see that $\ker Q$ will be a subspace of $\Fin_0(X)$.
76: \item Pass to quotient. Define $\tilde Q$ to be the induced quadratic form on the quotient space $\Fin_0(X)/\ker Q$. One may then verify that $\tilde Q$ is \emph{strictly positive definite} on the quotient space. As a consequence, $\|\gf\|_{\sH_{vN}} := -\tilde Q(\gf)$ will be a bona fide norm.
77: \item Complete. When the quotient space is completed with respect to $\tilde Q$, one obtains a Hilbert space
78: \linenopax
79: \begin{align}\label{eqn:vNeu's-quotient-completion}
80: \sH_{vN} &:= \left(\frac{\Fin_0(X)}{\ker Q}\right)^\sim,
81: \q\text{with}\q
82: \la \gf, \gy \ra_{\sH_{vN}} = -\tilde Q(\gf,\gy).
83: \end{align}
84: \item Embed $(X,d)$ into $\sH_{vN}$. Fix some point $o \in X$ to act as the origin; it will be mapped to the origin of $\sH_{vN}$ under the embedding. Then define
85: \linenopax
86: \begin{align*}
87: w:(X,d) \to \sH_{vN} \q\text{by}\q
88: x \mapsto w_x := \tfrac1{\sqrt2}(\gd_x - \gd_o).
89: \end{align*}
90: Now $w$ gives an embedding of $(X,d)$ into the Hilbert space $H_{vN}$, and
91: \linenopax
92: \begin{align}\label{eqn:embedding-check-app}
93: \|w_x - w_y\|_{vN}^2
94: &= \la w_x - w_y, w_x - w_y \ra_{vN} \\
95: &= \la w_x, w_x\ra_{vN} - \la w_x, w_y \ra_{vN} + \tfrac12 \la w_y, w_y \ra_{vN} \notag \\
96: &= d^2(x,o) + \left(d^2(x,y) - d^2(x,o) - d^2(y,o)\right) + d^2(y,o) \notag \\
97: &= d^2(x,y),\notag
98: \end{align}
99: which verifes \eqref{eqn:vonNeumann's-embedding-thm}. The third equality follows by three computations of the form
100: \linenopax
101: \begin{align}\label{eqn:crossterm-sums-of-madness}
102: \la w_x, w_y\ra_{vN}
103: &= -\sum_{a,b} w_x(a) d^2(a,b) w_y(b) \notag \\
104: &= -\sum_{a,b} d^2(a,b) \tfrac1{\sqrt2}(\gd_x(a)-\gd_o(a))(\gd_y(b)-\gd_o(b)) \notag \\
105: &= \dots = d^2(x,y) - d^2(x,o) - d^2(y,o),
106: \end{align}
107: noting that $d(a,a)=0$, etc.
108: \end{FlatList}
109:
110: von Neumann's theorem also has a form of uniqueness which may be thought of as a universal property.
111:
112: \begin{theorem}\label{thm:uniqueness-of-vNeu-embedding}
113: If there is another Hilbert space \sK and an embedding $k:\sH \to \sK$, with $\|k_x-k_y\|_\sK = d(x,y)$ and $\{k_x\}_{x \in X}$ dense in \sK, then there exists a unique unitary isomorphism $U:\sH \to \sK$.
114: \begin{proof}
115: We show that $U:w \mapsto k$ by $U(\sum \gx_x w_x) = \sum \gx_x k_x$ is the required isometric isomorphism. Let $\sum \gx_x = 0$. It is conceivable that $U$ fails to be well-defined because of linear dependency; we show this is not the case:
116: \linenopax
117: \begin{align}
118: \left\|\sum_{x \in X} \gx_x w_x\right\|^2
119: &= \sum_{x,y \in X} \gx_x \tilde Q(w_x, w_y) \gx_y \notag \\
120: &= \sum_{x,y \in X} \gx_x \left(d^2(x,y) - d^2(x,o) - d^2(y,o) \right) \gx_y \qq\text{by \eqref{eqn:crossterm-sums-of-madness}} \notag \\
121: &= \sum_{x,y \in X} \gx_x \gx_y d^2(x,y)
122: - \sum_{x \in X} \gx_x d^2(x,o) \cancel{\sum_{y \in X} \gx_y}
123: - \sum_{y \in X} \gx_y d^2(y,o) \cancel{\sum_{x \in X} \gx_x} \notag \\
124: &= -\sum_{x,y \in X} \gx_x \gx_y d(x,y), \label{eqn:uniqueness-of-vNeu-embedding}
125: \end{align}
126: since $\sum_x \gx_x = 0$ by choice of \gx. However, the same computation may be applied to $k$ with the same result; note that \eqref{eqn:uniqueness-of-vNeu-embedding} does not depend on $w$. Thus, $\|w\|_\sH = \|k\|_\sK$ and $U$ is an isometry. Since it is an isometry from a dense set in \sH to a dense set in \sK, we have an isomorphism and are finished.
127: \end{proof}
128: \end{theorem}
129:
130: \version{}{\marginpar{Is this paragraph correct/ useful?}}
131: The importance of using $\Fin_0(X)$ in the above construction is that the finitely supported functions $\Fin(X)$ are in duality with the bounded functions $B(X)$ via
132: \linenopax
133: \begin{align}\label{eqn:vNeu-duality-bracket}
134: \la f,\gb\ra := \sum_{x \in X} f(x) \gb(x) < \iy
135: \qq f \in \Fin(X), \gb \in B(X).
136: \end{align}
137: The constant function $\gb_1 := \one$ is a canonical bounded function. With respect to the pairing in \eqref{eqn:vNeu-duality-bracket}, its orthogonal complement is
138: \linenopax
139: \begin{equation*}
140: \gb_1^\perp = \{\gf \suth \la \gf,\gb_1\ra = 0\} = \Fin_0(X).
141: \end{equation*}
142:
143:
144:
145: \section{Remarks and references}
146: \label{sec:Remarks-and-References-App-A}
147:
148: The material here is collected to help make our presentation self-contained. The reader may find the references \cite{Rud91, KadisonRingroseI, DuSc88, ReedSimonII, Arveson:spectral-theory, vN55} to be helpful. The further references \cite{BrRo79}, and \cite{Kat95} may also be useful.
149:
150: