1: \documentclass{ws-procs9x6}
2:
3: %\usepackage{showkeys}
4:
5: \begin{document}
6:
7: \title{A Stochastic Heisenberg Inequality}
8: \footnotetext{
9: {\em 2000 Mathematics Subject Classifications:} 60G44, 60G15.}
10: \footnotetext{{\em Key words and phrases}:
11: martingale, It\^{o}'s formula, commutation relationship, unitary operator,
12: Schwarz inequality, Heisenberg inequality.}
13:
14: \author{C.~Mueller\footnote{\uppercase{S}upported by
15: \uppercase{NSA} and \uppercase{NSF} grants.}}
16: \address{Department of Mathematics,\\
17: University of Rochester,\\
18: Rochester, NY 14627, USA\\
19: E-mail: cmlr@math.rochester.edu}
20:
21: \author{A.~Stan\footnote{\uppercase{S}upported by
22: an \uppercase{NSF} grant.}}
23:
24: \address{Department of Mathematics,\\
25: University of Rochester,\\
26: Rochester, NY 14627, USA\\
27: E-mail: astan@math.rochester.edu}
28:
29: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
30: % You may repeat \author \address as often as necessary %
31: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
32:
33: \maketitle
34:
35:
36:
37: \abstracts{An analogue of the Fourier transform will be introduced for
38: all square integrable continuous martingale processes whose quadratic
39: variation is deterministic. Using this transform we will formulate and
40: prove a stochastic Heisenberg inequality.}
41:
42:
43:
44: \section{Introduction}
45: The Heisenberg inequality in ${\mathbb R}$ (see Folland and
46: Sitaram\cite{fs} and Strichartz\cite{s})
47: says that there exists a positive constant
48: $c$ such that for any $f \in L^{2}({\mathbb R})$ and $a$ and $b$
49: real numbers, we have:
50: \begin{eqnarray}
51: \int_{{\mathbb R}}(x - a)^{2}|f(x)|^{2}dx \cdot
52: \int_{{\mathbb R}}(\gamma - b)^{2}|\hat{f}(\gamma)|^{2}d\gamma
53: & \geq & c\parallel f \parallel_{2}^{4}. \label{cheisenberg}
54: \end{eqnarray}
55: In the above inequality, $\hat{f}$ denotes the Fourier transform
56: of $f$:
57: \[
58: \hat{f}(\gamma)=\int_{\mathbf{R}}e^{-2\pi i\gamma x}f(x)dx
59: \]
60: while $\parallel f \parallel_{2}$ represents the $L^{2}$-norm of $f$.
61: Besides, both integrals from the left-hand side are assumed to be finite.\\
62: \par The mathematical interpretation of inequality (\ref{cheisenberg})
63: is that an $L^{2}$-function
64: and its Fourier transform cannot both be localized.
65: See Folland and Sitaram\cite{fs}.
66: \\
67: \par In this paper the function $f$ will be replaced by a complex valued
68: stochastic process $\{Y_{t}\}_{t \in I}$, where $I$ is an interval of the form
69: $[0, T]$, for some $T > 0$, or $I=[0, \infty)$. The integration with respect
70: to the Lebesgue measure will be replaced by the integration with respect to
71: a real valued square integrable continuous martingale
72: $\{X_{t}\}_{t \in I}$ whose
73: quadratic variation $\{\langle X \rangle_{t}\}_{t \in I}$ is
74: deterministic. The real numbers $a$ and $b$ will be replaced by
75: two real valued deterministic (measurable) functions $g$ and $\tilde{g}$.
76: The Fourier transform will be replaced by a unitary operator $\mathcal{G}$
77: that will be defined in the next section.
78:
79: \section{The ${\mathcal G}$-transform}
80:
81: Let $(\Omega, {\mathcal F}, P)$ be a probability space and
82: $\{{\mathcal F}_{t}\}_{t \in I}$ a filtration over ${\mathcal F}$.
83: Let $\{X_{t}\}_{t \in I}$, where $I = [0, T]$ or $I = [0, \infty)$,
84: be a real valued square integrable continuous martingale defined on
85: $\Omega$, adapted to $\{{\mathcal F}_{t}\}_{t \in I}$,
86: whose quadratic variation process
87: $\{\langle X \rangle_{t}\}_{t \in I}$ is deterministic. That means,
88: for all $t \in I$, $\langle X \rangle_{t}$ is constant a.s.(almost surely).
89: Thus, as a process, $X_t$ is equal in distribution to a time-changed Brownian
90: motion $B_{h(t)}$ with a deterministic time change $h(t)$. We assume that
91: $X_{0} = 0$ a.s.. Let
92: $f : {\mathbb R}^{2} \to {\mathbb R}$
93: be a twice differentiable function with continuous second order partial
94: derivatives. Applying It\^{o}'s formula to the semimartingales
95: $X_{t}^{1} = X_{t}$ and $X_{t}^{2} = \langle X \rangle_{t}$
96: we obtain:
97: \begin{eqnarray}
98: f(X_{t}, \langle X \rangle_{t}) - f(X_{0}, \langle X \rangle_{0})
99: & = &
100: \int_{0}^{t}\frac{\partial f}{\partial x}(X_{s}, \langle X \rangle_{s})
101: dX_{s} \nonumber\\
102: & + & \int_{0}^{t}\frac{\partial f}{\partial y}(X_{s}, \langle X \rangle_{s})
103: d\langle X \rangle_{s} \nonumber\\
104: & + & \frac{1}{2}
105: \int_{0}^{t}\frac{\partial^{2} f}{\partial x^{2}}
106: (X_{s}, \langle X \rangle_{s})d\langle X \rangle_{s}. \label{Ito}
107: \end{eqnarray}
108: From relation (\ref{Ito}), we see that, if $f$ satisfies the
109: differential equation:
110: \begin{eqnarray}
111: \frac{1}{2}\frac{\partial^{2} f}{\partial x^{2}} +
112: \frac{\partial f}{\partial y} & = & 0, \label{Differential}
113: \end{eqnarray}
114: then $f(X_{t}, \langle X \rangle_{t})$ is a local martingale
115: (see Durrett\cite{d}, page 70).\\
116: \par The function $f_{c}(x, y) = e^{cx - \frac{1}{2}c^{2}y}$ satisfies
117: equation (\ref{Differential}), for all $c \in {\mathbb C}$. Thus
118: $f_{c}(X_{t}, \langle X \rangle_{t})$ is a local martingale,
119: for all $c \in {\mathbb C}$. Since
120: $X_{t} \stackrel{\mathcal{D}}{=} B_{h(t)}$ we have $E(e^{cX_{t}}) < \infty$,
121: for all $c \in {\mathbb R}$ and $t \in I$. Thus the process
122: $\mathcal{E}_{c, t} := e^{cX_{t} - \frac{1}{2}c^{2}\langle X \rangle_{t}}$
123: is a martingale ($c$ is fixed).
124:
125: \begin{lemma}\label{lem:product}
126: For all $t \geq 0$ and all $c$ and $d$ complex numbers, we have:
127: \begin{eqnarray}
128: \mathcal{E}_{c, t} \mathcal{E}_{d, t} = e^{cd \langle X \rangle_{t}}
129: \mathcal{E}_{c + d, t}. \label{eq:product}
130: \end{eqnarray}
131: \end{lemma}
132:
133: \begin{proof}
134: \begin{eqnarray*}
135: \mathcal{E}_{c, t} \mathcal{E}_{d, t} & = &
136: e^{cX_{t} - \frac{1}{2}c^{2}\langle X \rangle_{t}}
137: e^{dX_{t} - \frac{1}{2}d^{2}\langle X \rangle_{t}}\\
138: & = & e^{cd \langle X \rangle_{t}}
139: e^{(c + d)X_{t} - \frac{1}{2}(c + d)^{2}\langle X \rangle_{t}}\\
140: & = & e^{cd \langle X \rangle_{t}}\mathcal{E}_{c + d, t}.
141: \end{eqnarray*}
142: \end{proof}
143:
144: For any $t \in I$, let $V_{t}$ be the vector space spanned by all
145: exponential functions
146: $\{\mathcal{E}_{c, t}\}_{c \in {\mathbb C}}$. Lemma \ref{lem:product} proves
147: that $V_{t}$ is closed under multiplication. The constant process $1$ belongs
148: to $V_{t}$ since $1 = \mathcal{E}_{0, t}$. Since $X_{t}$ is a real
149: valued function,
150: we have $\overline{\mathcal{E}_{c, t}} = \mathcal{E}_{\bar{c}, t} \in V_{t}$,
151: for all $c \in {\mathbb C}$.
152: Thus $V_{t}$ is an algebra containing the constant functions and closed under
153: conjugation.
154:
155: \begin{lemma}\label{lem:scalar}
156: For all $s$, $t \geq 0$ and all $c$ and $d$ complex numbers we have:
157: \begin{eqnarray}
158: E\left[\mathcal{E}_{c, s}\mathcal{E}_{d, t}\right] & = &
159: e^{cd \langle X \rangle_{s \wedge t}}, \label{eq:scalar}
160: \end{eqnarray}
161: where $s \wedge t$ denotes the minimum between $s$ and $t$.
162: \end{lemma}
163:
164: \begin{proof}
165: Let us assume that $s \leq t$. We have:
166: \begin{eqnarray*}
167: E\left[\mathcal{E}_{c, s}\mathcal{E}_{d, t}\right] & = &
168: E\left[E\left[\mathcal{E}_{c, s}\mathcal{E}_{d, t} | \mathcal{F}_{s}\right]
169: \right]\\
170: & = & E\left[\mathcal{E}_{c, s}
171: E\left[\mathcal{E}_{d, t} | \mathcal{F}_{s}\right]\right]\\
172: & = & E\left[\mathcal{E}_{c, s}\mathcal{E}_{d, s}\right]\\
173: & = & E\left[e^{cd \langle X \rangle_{s}}
174: \mathcal{E}_{c + d, s}\right]\\
175: & = & e^{cd \langle X \rangle_{s}}
176: E\left[\mathcal{E}_{c + d, s}\right]\\
177: & = & e^{cd \langle X \rangle_{s}}
178: E\left[\left[\mathcal{E}_{c + d, s} | \mathcal{F}_{0}\right]\right]\\
179: & = & e^{cd \langle X \rangle_{s}}
180: E\left[\mathcal{E}_{c + d, 0}\right]\\
181: & = & e^{cd \langle X \rangle_{s}}.
182: \end{eqnarray*}
183: \end{proof}
184:
185: \par We define the function:
186: $\mathcal{G} : \cup_{t \in I}V_{t} \to \cup_{t \in I}V_{t}$,
187: $\mathcal{G}\mathcal{E}_{c, t} := \mathcal{E}_{-ic, t}$. We can see that
188: $\mathcal{G}$ preserves the inner product. Indeed, we have:
189: \begin{eqnarray*}
190: E\left[\mathcal{G}\mathcal{E}_{c, s}
191: \overline{\mathcal{G}\mathcal{E}_{d, t}}\right]
192: & = & \left[\mathcal{E}_{-ic, s}\overline{\mathcal{E}_{-id, t}}\right]\\
193: & = & \left[\mathcal{E}_{-ic, s}\mathcal{E}_{i\bar{d}, t}\right]\\
194: & = & e^{(-ic)(i\bar{d})X_{s \wedge t}}\\
195: & = & e^{c\bar{d}X_{s \wedge t}}\\
196: & = & E\left[\mathcal{E}_{c, s}\mathcal{E}_{\bar{d}, t}\right]\\
197: & = & E\left[\mathcal{E}_{c, s}\overline{\mathcal{E}_{d, t}}\right].
198: \end{eqnarray*}
199: Thus $\mathcal{G}$ can be uniquely extended to a unitary operator, that
200: we denote also by $\mathcal{G}$, from the Hilbert space $\mathcal{H}$
201: into itself, where $\mathcal{H}$ is the closure of the vector space
202: spanned by $\{\mathcal{E}_{c, t} ~|~ c \in {\mathbb C}, t \in I\}$
203: in $L^{2}(\Omega, \mathcal{F}, P)$.
204:
205: \section{The Operator of Multiplication by $X_{t}$}
206:
207: Let $\{X_{t}\}_{t \in I}$ be a square integrable continuous martingale
208: process adapted to the filtration $\mathcal{F}_{t}$ and having a deterministic
209: quadratic variation process $\{\langle X \rangle_{t}\}_{t \in I}$. We know
210: that for each $t \in I$, $X_{t} \stackrel{\mathcal{D}}{=} B_{h(t)}$.
211:
212: We note the following. If $t$ is fixed, then $h(t)$ is fixed, and $X_t$ is
213: normally distributed with mean 0 and variance $h(t)$. Thus,
214: \begin{eqnarray}
215: \label{ltwo}
216: e^{c|X_{t}|} & \in& L^{2}(\Omega, \mathcal{F}, P).
217: \end{eqnarray}
218:
219: Since $|X_{t}| < e^{|X_{t}|}$, we have
220: \begin{eqnarray*}
221: |X_{t}\mathcal{E}_{c, t}| & = & |X_{t}| \cdot
222: |e^{cX_{t} - \frac{c^{2}}{2}\langle X \rangle_{t}}|\\
223: & \leq & e^{|X_{t}|}e^{|c| \cdot |X_{t}|}
224: e^{-\frac{\rm{Re}(c^{2})}{2}\langle X \rangle_{t}}\\
225: & = & e^{-\frac{\rm{Re}(c^{2})}{2}\langle X \rangle_{t}}e^{(|c| + 1)|X_{t}|}.
226: \end{eqnarray*}
227: Since, according to (\ref{ltwo}),
228: $e^{(|c| + 1)|X_{t}|} \in L^{2}(\Omega, \mathcal{F}, P)$, we conclude that
229: $X_{t}\mathcal{E}_{c, t} \in L^{2}(\Omega, \mathcal{F}, P)$, for all
230: $c \in {\mathbb C}$ and $t \in I$.
231:
232: \begin{lemma}
233: For any $t \in I$ and $c \in {\mathbb C}$, we have:
234: \begin{eqnarray}
235: X_{t}\mathcal{E}_{c, t} & = &
236: \lim_{r \to 0}\frac{\mathcal{E}_{r, t} - 1}{r}\mathcal{E}_{c, t}, \quad
237: {\rm in} \ L^{2}-{\rm sense}.
238: \label{Xasaderviative}
239: \end{eqnarray}
240: \end{lemma}
241:
242: \begin{proof}
243: We have:
244: \begin{eqnarray*}
245: \lim_{r \to 0}\frac{\mathcal{E}_{r, t} - 1}{r} & = &
246: \lim_{r \to 0}
247: \frac{e^{rX_{t} - \frac{r^{2}}{2}\langle X \rangle_{t}} - 1}{r}\\
248: & = & \frac{d}{dr}
249: \left(e^{rX_{t} - \frac{r^{2}}{2}\langle X \rangle_{t}}\right)\Big|_{r = 0}\\
250: & = & X_{t}.
251: \end{eqnarray*}
252: Thus $X_{t}\mathcal{E}_{c, t} =
253: \lim_{r \to 0}\frac{\mathcal{E}_{r, t} - 1}{r}\mathcal{E}_{c, t}$ pointwise.
254: By Taylor's formula with Lagrange's remainder,
255: for any fixed $\omega \in \Omega$, there exists
256: a real number $u(\omega)$ between $0$ and
257: $rX_{t}(\omega) - \frac{r^{2}}{2}\langle X \rangle_{t}$ such that
258: \begin{eqnarray*}
259: e^{rX_{t}(\omega) - \frac{r^{2}}{2}\langle X \rangle_{t}} & = &
260: 1 + rX_{t}(\omega) - \frac{r^{2}}{2}\langle X \rangle_{t} +
261: \frac{1}{2}e^{u(\omega)}
262: \left[rX_{t}(\omega) - \frac{r^{2}}{2}\langle X \rangle_{t}\right]^{2}.
263: \end{eqnarray*}
264: This relation can be rewritten as:
265: \begin{eqnarray*}
266: \frac{e^{rX_{t}(\omega) - \frac{r^{2}}{2}\langle X \rangle_{t}} - 1}{r} -
267: X_{t}(\omega) & = &
268: - \frac{r}{2}\langle X \rangle_{t} +
269: \frac{r}{2}e^{u(\omega)}
270: \left[X_{t}(\omega) - \frac{r}{2}\langle X \rangle_{t}\right]^{2}.
271: \end{eqnarray*}
272: From here, using the inequalities
273: $e^{u(\omega)} \leq e^{|r| \cdot |X_{t}(\omega)|}$
274: and $|X_{t}(\omega)| < e^{|X_{t}(\omega)|}$, and (\ref{ltwo}), it follows
275: from Lebesgue's dominated convergence theorem, that
276: $\frac{e^{rX_{t}(\omega) - \frac{r^{2}}{2}\langle X \rangle_{t}} - 1}{r} -
277: X_{t}(\omega) \to 0$ in $L^{2}(\Omega, \mathcal{F}, P)$, as $r \to 0$.
278: \end{proof}
279:
280: \section{The Differential Operator}
281:
282: \begin{definition}
283: For any $t \in I$ and $c \in {\mathbb C}$, we define
284: $D_{t}\mathcal{E}_{c, t} := c\langle X \rangle_{t}\mathcal{E}_{c, t}$.
285: \end{definition}
286: We extend $D_{t}$ by linearity to the vector space $V_{t}$
287: spanned by the exponential functions
288: $\{\mathcal{E}_{c, t}\}_{c \in {\mathbb C}}$. We must check that $D_{t}$ is
289: well-defined.
290:
291: \begin{lemma}
292: $D_{t} : V_{t} \to V_{t}$ is a well-defined operator.
293: \end{lemma}
294:
295: \begin{proof}
296: To check that $D_{t}$ is a well-defined operator from $V_{t}$ into itself
297: we must check that if a function $f$ can be expressed in two different ways
298: as a linear combination of exponential functions, then calculating $D_{t}f$
299: using these two linear combinations we obtain the same result.
300: This reduces to checking that if $\lambda_{1}$, $\lambda_{2}$, $\dots$,
301: $\lambda_{N} \in {\mathbb C}$ and $c_{1}$, $c_{2}$, $\dots$,
302: $c_{N} \in {\mathbb C}$, $c_{i} \neq c_{j}$, for $i \neq j$, such that:
303: $$\sum_{k = 1}^{N}\lambda_{k}\mathcal{E}_{c_{k}, t} = 0,$$
304: then
305: $$\sum_{k = 1}^{N}\lambda_{k}c_{k}\langle X \rangle_{t}
306: \mathcal{E}_{c_{k}, t} = 0.$$
307: If $\sum_{k = 1}^{N}\lambda_{k}\mathcal{E}_{c_{k}, t} = 0$, then
308: for any $n \in \{0$, $1$, $\dots$, $N - 1\}$ we have:
309: $E\left[\left(\sum_{k = 1}^{N}\lambda_{k}\mathcal{E}_{c_{k}, t}\right)
310: \mathcal{E}_{n, t}\right] = 0$. This means, according to formula
311: (\ref{eq:scalar}), that:
312: \begin{eqnarray*}
313: \sum_{k = 1}^{N}\lambda_{k}e^{nc_{k}\langle X \rangle_{t}} & = & 0,
314: \quad \forall n \in \{0, 1, \dots, N - 1\}.
315: \end{eqnarray*}
316: The above relations represent a homogenous linear system of $N$ equations
317: and $N$ unknowns $\lambda_{1}$, $\lambda_{2}$, $\dots$,
318: $\lambda_{N}$, having a Vandermonde determinant. Since
319: $c_{i} \neq c_{j}$, for $i \neq j$, the Vandermonde determinant of the
320: above system is different from zero if $\langle X \rangle_{t} \neq 0$.
321: Thus if $\langle X \rangle_{t} \neq 0$, then
322: $\lambda_{1} = \lambda_{2} = \dots = \lambda_{N} = 0$. Hence, either
323: $\langle X \rangle_{t} = 0$ or
324: $\lambda_{1} = \lambda_{2} = \dots = \lambda_{N} = 0$. Therefore,
325: $\sum_{k = 1}^{N}\lambda_{k}c_{k}\langle X \rangle_{t}
326: \mathcal{E}_{c_{k}, t} = 0$.
327: \end{proof}
328:
329: We call $D_{t}$ {\it the differential operator}. Its initial domain is
330: $V_{t}$ which is a dense subspace of the Hilbert space $\mathcal{H}_{t}$,
331: where $\mathcal{H}_{t}$ is the closure of $V_{t}$ in
332: $L^{2}(\Omega, \mathcal{F}, P)$. Observe that
333: $\mathcal{H}_{t} \subset L^{2}(\Omega, \sigma(X_{t}), P) \subset
334: L^{2}(\Omega, \mathcal{F}_{t}, P)$, where $\sigma(X_{t})$ is the smallest
335: $\sigma$-field on $\Omega$ with respect to which $X_{t}$ is measurable.
336: Since $X_{t} \stackrel{\mathcal{D}}{=} B_{h(t)}$, we have
337: $\mathcal{H}_{t} = L^{2}(\Omega, \sigma(X_{t}), P)$
338: (see Janson\cite{j}, page 19).\\
339: \par We denote the adjoint of $D_{t}$ by $D_{t}^{*}$.
340:
341: \begin{lemma}
342: For any $t \in I$, $V_{t}$ is contained in the domain of $D_{t}^{*}$ and
343: for any $\varphi \in V_{t}$ we have:
344: \begin{eqnarray}
345: X_{t}\varphi & = & D_{t}\varphi + D_{t}^{*}\varphi. \label{x=d+d*}
346: \end{eqnarray}
347: \end{lemma}
348:
349: \begin{proof}
350: Since $V_{t}$ is spanned by the exponential functions we have to check
351: that for all $d \in {\mathbb C}$, $\mathcal{E}_{d, t}$ belongs to the
352: domain of $D_{t}^{*}$ and for any $c \in {\mathbb C}$ we have:
353: \begin{eqnarray}
354: \langle D_{t}\mathcal{E}_{c, t}, \mathcal{E}_{d, t}\rangle & = &
355: \langle \mathcal{E}_{c, t}, (X_{t} - D_{t})\mathcal{E}_{d, t}\rangle,
356: \label{adjoint}
357: \end{eqnarray}
358: where $\langle , \rangle$ denotes the $L^{2}$-inner product.
359: Let us observe that, for any complex number $c$,
360: the function $g_{c} : {\mathbb R}^{2} \to {\mathbb R}$,
361: $g_{c}(x, y) = (x - cy)e^{cx - \frac{c^{2}}{2}y}$ satisfies the differential
362: equation:
363: $$\frac{1}{2}\frac{\partial^{2}g_{c}}{\partial x^{2}} +
364: \frac{\partial g_{c}}{\partial y} = 0.$$
365: It follows from It\^{o}'s formula that the process
366: $\{g_{c}(X_{t}, \langle X \rangle_{t})\}_{t \in I}$ is a martingale.
367: We denote this martingale process by
368: $\{\mathcal{X}_{t}\mathcal{E}_{c, t}\}_{t \in I}$. So,
369: $\mathcal{X}_{t}\mathcal{E}_{c, t} = (X_{t} - c\langle X \rangle_{t})
370: e^{cX_{t} - \frac{c^{2}}{2}\langle X \rangle_{t}}$. From this definition
371: we obtain the formula:
372: \begin{eqnarray}
373: X_{t}\mathcal{E}_{c, t} & = & \mathcal{X}_{t}\mathcal{E}_{c, t} +
374: c\langle X \rangle_{t}\mathcal{E}_{c, t}. \label{madjoint}
375: \end{eqnarray}
376: Thus we have:
377: \begin{eqnarray*}
378: \langle \mathcal{E}_{c, t}, (X_{t} - D_{t})\mathcal{E}_{d, t}\rangle
379: & = & E[X_{t}\mathcal{E}_{c, t}\mathcal{E}_{\bar{d}, t}] -
380: E[\mathcal{E}_{c, t}D_{t}\mathcal{E}_{\bar{d}, t}]\\
381: & = & E[X_{t}\mathcal{E}_{c + \bar{d}, t}e^{c\bar{d}\langle X \rangle_{t}}]
382: - E[\mathcal{E}_{c, t}\bar{d}\langle X \rangle_{t}\mathcal{E}_{\bar{d}, t}]
383: \\
384: & = & e^{c\bar{d}\langle X \rangle_{t}}E[X_{t}\mathcal{E}_{c + \bar{d}, t}]
385: - \bar{d}\langle X \rangle_{t}E[\mathcal{E}_{c, t}\mathcal{E}_{\bar{d}, t}].
386: \end{eqnarray*}
387: Using relations (\ref{madjoint}) and (\ref{eq:scalar}) we obtain:
388: \begin{eqnarray*}
389: & \ & \langle \mathcal{E}_{c, t}, (X_{t} - D_{t})\mathcal{E}_{d, t}\rangle\\
390: & = & e^{c\bar{d}\langle X \rangle_{t}}
391: E[\mathcal{X}_{t}\mathcal{E}_{c + \bar{d}, t} +
392: (c + \bar{d})\langle X \rangle_{t}\mathcal{E}_{c + \bar{d}, t}] -
393: \bar{d}\langle X \rangle_{t}E[\mathcal{E}_{c, t}\mathcal{E}_{\bar{d}, t}]\\
394: & = & e^{c\bar{d}\langle X \rangle_{t}}
395: E[\mathcal{X}_{t}\mathcal{E}_{c + \bar{d}, t}] +
396: (c + \bar{d})\langle X \rangle_{t}e^{c\bar{d}\langle X \rangle_{t}}
397: E[\mathcal{E}_{c + \bar{d}, t}] -
398: \bar{d}\langle X \rangle_{t}e^{c\bar{d}\langle X \rangle_{t}}.
399: \end{eqnarray*}
400: Because $\{\mathcal{X}_{t}\mathcal{E}_{c + \bar{d}, t}\}_{t \in I}$ and
401: $\{\mathcal{E}_{c + \bar{d}, t}\}_{t \in I}$ are martingale processes
402: we have:
403: \begin{eqnarray*}
404: E[\mathcal{X}_{t}\mathcal{E}_{c + \bar{d}, t}] & = &
405: E[\mathcal{X}_{0}\mathcal{E}_{c + \bar{d}, 0}]\\
406: & = & 0
407: \end{eqnarray*}
408: and
409: \begin{eqnarray*}
410: E[\mathcal{E}_{c + \bar{d}, t}] & = & E[\mathcal{E}_{c + \bar{d}, 0}]\\
411: & = & 1.
412: \end{eqnarray*}
413: Thus, we obtain:
414: \begin{eqnarray*}
415: & \ & \langle \mathcal{E}_{c, t}, (X_{t} - D_{t})\mathcal{E}_{d, t}\rangle\\
416: & = & e^{c\bar{d}\langle X \rangle_{t}} \cdot 0 +
417: (c + \bar{d})\langle X \rangle_{t}e^{c\bar{d}\langle X \rangle_{t}} \cdot 1
418: - \bar{d}\langle X \rangle_{t}e^{c\bar{d}\langle X \rangle_{t}}\\
419: & = & c\langle X \rangle_{t}e^{c\bar{d}\langle X \rangle_{t}}\\
420: & = & c\langle X \rangle_{t}E[\mathcal{E}_{c, t}\mathcal{E}_{\bar{d}, t}]\\
421: & = & \langle c\langle X \rangle_{t}\mathcal{E}_{c, t},
422: \mathcal{E}_{d, t}\rangle\\
423: & = & \langle D_{t}\mathcal{E}_{c, t}, \mathcal{E}_{d, t}\rangle.
424: \end{eqnarray*}
425: \end{proof}
426:
427: \begin{corollary}
428: For all $t \in I$, the differential operator $D_{t}$
429: admits a closed extension.
430: \end{corollary}
431:
432: \begin{proof} Let $W_{t}$ be the subspace of all functions $\varphi$
433: in $\mathcal{H}_{t}$ for which there exists a sequence
434: $\{\varphi_{n}\}_{n \geq 1} \subset V_{t}$ such that
435: $\varphi_{n} \to \varphi$ in $L^{2}(\Omega, \mathcal{F}, P)$ and the
436: sequence $\{D_{t}\varphi_{n}\}_{n \geq 1}$ is Cauchy in
437: $L^{2}(\Omega, \mathcal{F}, P)$.
438: We define the operator $\widetilde{D}_{t}$ on $W_{t}$ in the following way:
439: if $\varphi \in W_{t}$ and $\{\varphi_{n}\}_{n \geq 1} \subset V_{t}$ such
440: that $\varphi_{n} \to \varphi$ in $ L^{2}(\Omega, \mathcal{F}, P)$ and
441: $\{D_{t}\varphi_{n}\}_{n \geq 1}$ is Cauchy in
442: $L^{2}(\Omega, \mathcal{F}, P)$, then $\widetilde{D}_{t}\varphi = u$, where
443: $u = \lim_{n \to \infty}D_{t}\varphi_{n}$ in $L^{2}(\Omega, \mathcal{F}, P)$.
444: We notice that $u \in \mathcal{H}_{t}$.
445: \par We need to check that $\widetilde{D}_{t}\varphi$ is well defined.
446: Let $\{\varphi_{n}\}_{n \geq 1} \subset V_{t}$ such
447: that $\varphi_{n} \to \varphi$ in $ L^{2}(\Omega, \mathcal{F}, P)$ and
448: $\{D_{t}\varphi_{n}\}_{n \geq 1}$ is Cauchy in
449: $L^{2}(\Omega, \mathcal{F}, P)$, and
450: $\{\psi_{n}\}_{n \geq 1} \subset V_{t}$ such
451: that $\psi_{n} \to \varphi$ in $ L^{2}(\Omega, \mathcal{F}, P)$ and
452: $\{D_{t}\psi_{n}\}_{n \geq 1}$ is Cauchy in
453: $L^{2}(\Omega, \mathcal{F}, P)$.
454: Let $u := \lim_{n \to \infty}D_{t}\varphi_{n}$ and
455: $v := \lim_{n \to \infty}D_{t}\psi_{n}$. Both limits are in the $L^{2}$-sense.
456: For any $\phi \in V_{t}$ we have:
457: \begin{eqnarray*}
458: \langle u, \phi \rangle & = &
459: \lim_{n \to \infty}\langle D_{t}\varphi_{n}, \phi \rangle\\
460: & = & \lim_{n \to \infty}\langle \varphi_{n}, D_{t}^{*}\phi \rangle\\
461: & = & \langle \varphi, D_{t}^{*}\phi \rangle.
462: \end{eqnarray*}
463: In the same way, we can see that
464: \begin{eqnarray*}
465: \langle v, \phi \rangle & = & \langle \varphi, D_{t}^{*}\phi \rangle.
466: \end{eqnarray*}
467: Thus $\langle u, \phi \rangle = \langle v, \phi \rangle$, for all
468: $\phi \in V_{t}$. Since $V_{t}$ is dense in $\mathcal{H}_{t}$ and
469: $u$, $v \in \mathcal{H}_{t}$ it follows that $u = v$.
470: \end{proof}
471:
472: From now on we will denote the closure of $D_{t}$ by $D_{t}$, too.
473:
474: \section{Commutation Relationships}
475:
476: \begin{lemma}
477: For any $t \in I$, we have:
478: \begin{eqnarray}
479: [D_{t}, X_{t}] & = & \langle X \rangle_{t}I.
480: \end{eqnarray}
481: \end{lemma}
482:
483: \begin{proof}
484: We will check that for any exponential function $\mathcal{E}_{c, t}$ we have:
485: \begin{eqnarray}
486: D_{t}X_{t}\mathcal{E}_{c, t} - X_{t}D_{t}\mathcal{E}_{c, t} & = &
487: \langle X \rangle_{t}\mathcal{E}_{c, t}. \label{comdx}
488: \end{eqnarray}
489: Indeed we have:
490: \begin{eqnarray*}
491: D_{t}X_{t}\mathcal{E}_{c, t} & = & D_{t}\lim_{s \to 0}
492: \frac{\mathcal{E}_{s, t} - 1}{s}\mathcal{E}_{c, t}.
493: \end{eqnarray*}
494: We have seen that, because of (\ref{ltwo}), the limit from
495: the right-hand side of the last equality is both pointwise and in the
496: $L^{2}$-sense. The last equality can be written now as:
497: \begin{eqnarray*}
498: D_{t}X_{t}\mathcal{E}_{c, t} & = & D_{t}\lim_{s \to 0}
499: \frac{\mathcal{E}_{s, t}\mathcal{E}_{c, t} - \mathcal{E}_{c, t}}{s}\\
500: & = & D_{t}\lim_{s \to 0}
501: \frac{\mathcal{E}_{s + c, t}
502: e^{sc\langle X \rangle_{t}} - \mathcal{E}_{c, t}}{s}.
503: \end{eqnarray*}
504: Because we are working with the closure of the operator $D_{t}$ and the
505: limit from the last relation is in the $L^{2}$-sense, we can commute
506: $D_{t}$ with the limit if $D_{t}\frac{\mathcal{E}_{s + c, t}
507: e^{sc\langle X \rangle_{t}} - \mathcal{E}_{c, t}}{s}$ converges in
508: $L^{2}(\Omega, \mathcal{F}, P)$, as $s \to 0$.
509: Commuting $D_{t}$ with the limit we get:
510: \begin{eqnarray*}
511: D_{t}X_{t}\mathcal{E}_{c, t}
512: & = & \lim_{s \to 0}D_{t}\frac{\mathcal{E}_{s + c, t}
513: e^{sc\langle X \rangle_{t}} - \mathcal{E}_{c, t}}{s}\\
514: & = & \lim_{s \to 0}\frac{e^{sc\langle X \rangle_{t}}
515: D_{t}\mathcal{E}_{s + c, t} - D_{t}\mathcal{E}_{c, t}}{s}\\
516: & = & \lim_{s \to 0}\frac{e^{sc\langle X \rangle_{t}}(s + c)
517: \langle X \rangle_{t}\mathcal{E}_{s + c, t} -
518: c\langle X \rangle_{t}\mathcal{E}_{c, t}}{s}\\
519: & = & \langle X \rangle_{t}\lim_{s \to 0}
520: \frac{(s + c)e^{sc\langle X \rangle_{t}}\mathcal{E}_{s + c, t} -
521: c\mathcal{E}_{c, t}}{s}\\
522: & = & \langle X \rangle_{t}\lim_{s \to 0}
523: \frac{(s + c)\mathcal{E}_{c, t}\mathcal{E}_{s, t} -
524: c\mathcal{E}_{c, t}}{s}\\
525: & = & \langle X \rangle_{t}
526: \left[\lim_{s \to 0}\mathcal{E}_{s, t} + c\lim_{s \to 0}
527: \frac{\mathcal{E}_{s, t} - 1}{s}\right]\mathcal{E}_{c, t}\\
528: & = & \langle X \rangle_{t}[\mathcal{E}_{0, t} + cX_{t}]
529: \mathcal{E}_{c, t}\\
530: & = & \langle X \rangle_{t}(1 + cX_{t})\mathcal{E}_{c, t}\\
531: & = & \langle X \rangle_{t}\mathcal{E}_{c, t} +
532: X_{t}(c\langle X \rangle_{t}\mathcal{E}_{c, t})\\
533: & = & \langle X \rangle_{t}\mathcal{E}_{c, t} +
534: X_{t}D_{t}\mathcal{E}_{c, t}.
535: \end{eqnarray*}
536: Because of (\ref{ltwo}) all the above limits are not only
537: pointwise, but also in the $L^{2}$-sense.
538: Thus
539: \begin{eqnarray*}
540: D_{t}X_{t}\mathcal{E}_{c, t} - X_{t}D_{t}\mathcal{E}_{c, t} =
541: \langle X \rangle_{t}\mathcal{E}_{c, t}.
542: \end{eqnarray*}
543: \end{proof}
544:
545: \begin{corollary}
546: For any $t \in I$, we have:
547: \begin{eqnarray}
548: [D_{t}, D_{t}^{*}] & = & \langle X \rangle_{t}I. \label{comdd*}
549: \end{eqnarray}
550: \end{corollary}
551:
552: \begin{proof}
553: According to formula (\ref{x=d+d*}) we have $X_{t} = D_{t} + D_{t}^{*}$.
554: Thus:
555: \begin{eqnarray*}
556: [D_{t}, D_{t}^{*}] & = & [D_{t}, X_{t} - D_{t}]\\
557: & = & [D_{t}, X_{t}] - [D_{t}, D_{t}]\\
558: & = & \langle X \rangle_{t}I - 0\\
559: & = & \langle X \rangle_{t}I.
560: \end{eqnarray*}
561: \end{proof}
562:
563: \begin{lemma}
564: For any $t \in I$, we have:
565: \begin{eqnarray}
566: D_{t}\mathcal{G} & = & -i\mathcal{G}D_{t}. \label{dg}
567: \end{eqnarray}
568: \end{lemma}
569:
570: \begin{proof}
571: For any exponential function $\mathcal{E}_{c, t}$, we have:
572: \begin{eqnarray*}
573: D_{t}\mathcal{G}\mathcal{E}_{c, t} & = & D_{t}\mathcal{E}_{-ic, t}\\
574: & = & -ic\langle X \rangle_{t}\mathcal{E}_{-ic, t}\\
575: & = & -ic\langle X \rangle_{t}\mathcal{G}\mathcal{E}_{c, t}\\
576: & = & -i\mathcal{G}\left(c\langle X \rangle_{t}\mathcal{E}_{c, t}\right)\\
577: & = & -i\mathcal{G}D_{t}\mathcal{E}_{c, t}.
578: \end{eqnarray*}
579: \end{proof}
580:
581: \begin{lemma}
582: For any $t \in I$, we have:
583: \begin{eqnarray}
584: D_{t}^{*}\mathcal{G} & = & i\mathcal{G}D_{t}^{*}. \label{d*g}
585: \end{eqnarray}
586: \end{lemma}
587:
588: \begin{proof}
589: For any two exponential functions $\mathcal{E}_{c, t}$ and
590: $\mathcal{E}_{d, t}$, we have:
591: \begin{eqnarray*}
592: \langle D_{t}^{*}\mathcal{G}\mathcal{E}_{c, t}, \mathcal{E}_{d, t} \rangle
593: & = &
594: \langle \mathcal{G}\mathcal{E}_{c, t}, D_{t}\mathcal{E}_{d, t} \rangle\\
595: & = & \langle \mathcal{E}_{-ic, t},
596: d\langle X \rangle_{t}\mathcal{E}_{d, t} \rangle\\
597: & = & \bar{d}\langle X \rangle_{t}\langle \mathcal{E}_{-ic, t},
598: \mathcal{E}_{d, t} \rangle\\
599: & = & \bar{d}\langle X \rangle_{t}e^{-ic\bar{d}\langle X \rangle_{t}}\\
600: & = & i(-i\bar{d})\langle X \rangle_{t}
601: e^{c\overline{(id)}\langle X \rangle_{t}}\\
602: & = & i(-i\bar{d})\langle X \rangle_{t}
603: \langle \mathcal{E}_{c, t}, \mathcal{E}_{id, t} \rangle\\
604: & = & i\langle \mathcal{E}_{c, t},
605: id\langle X \rangle_{t}\mathcal{E}_{id, t} \rangle\\
606: & = & i\langle \mathcal{E}_{c, t}, D_{t}\mathcal{E}_{id, t} \rangle\\
607: & = & i\langle D_{t}^{*}\mathcal{E}_{c, t}, \mathcal{E}_{id, t} \rangle.
608: \end{eqnarray*}
609: Since $\mathcal{G}$ preserves the inner product, being a unitary operator,
610: we obtain:
611: \begin{eqnarray*}
612: \langle D_{t}^{*}\mathcal{G}\mathcal{E}_{c, t}, \mathcal{E}_{d, t} \rangle
613: & = & i\langle D_{t}^{*}\mathcal{E}_{c, t}, \mathcal{E}_{id, t} \rangle\\
614: & = & i\langle \mathcal{G}D_{t}^{*}\mathcal{E}_{c, t},
615: \mathcal{G}\mathcal{E}_{id, t} \rangle\\
616: & = & i\langle \mathcal{G}D_{t}^{*}\mathcal{E}_{c, t},
617: \mathcal{E}_{-i(id), t} \rangle\\
618: & = & \langle i\mathcal{G}D_{t}^{*}\mathcal{E}_{c, t},
619: \mathcal{E}_{d, t} \rangle.
620: \end{eqnarray*}
621: Thus $D_{t}^{*}\mathcal{G} = i\mathcal{G}D_{t}^{*}$.
622: \end{proof}
623:
624: \section{Heisenberg Inequality}
625:
626: \begin{lemma}
627: Let $t \in I$ and $Y_{t} \in L^{2}(\Omega, \sigma(X_{t}), P)$.
628: Let $c_{t}$, $\tilde{c}_{t} \in {\mathbb R}$.
629: If $Y_{t}$ belongs to the domains of the operators $D_{t}$, $D_{t}^{*}$,
630: $D_{t}D_{t}^{*}$, and $D_{t}^{*}D_{t}$, then
631: \begin{eqnarray}
632: E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
633: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}} & \geq &
634: \langle X \rangle_{t}E[|Y_{t}|^{2}]. \label{h1}
635: \end{eqnarray}
636: \end{lemma}
637:
638: \begin{proof}
639: Since $X_{t} = D_{t} + D_{t}^{*}$, we have:
640: \begin{eqnarray*}
641: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
642: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
643: & = & E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
644: E[|(D_{t} + D_{t}^{*} - \tilde{c}_{t})\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
645: & = & E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
646: E[|(D_{t}\mathcal{G} + D_{t}^{*}\mathcal{G} - \tilde{c}_{t}\mathcal{G})
647: Y_{t}|^{2}]^{\frac{1}{2}}.
648: \end{eqnarray*}
649: Using the commutation relationships (\ref{dg}) and (\ref{d*g}) we obtain:
650: \begin{eqnarray*}
651: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
652: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
653: & = & E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
654: E[|(-i\mathcal{G}D_{t} + i\mathcal{G}D_{t}^{*} - \tilde{c}_{t}\mathcal{G})
655: Y_{t}|^{2}]^{\frac{1}{2}}\\
656: & = & E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
657: E[|i\mathcal{G}(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]^{\frac{1}{2}}.
658: \end{eqnarray*}
659: Since $\mathcal{G}$ is a unitary operator and $i$ has modulus 1, we have
660: $E[|i\mathcal{G}(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}] =
661: E[|(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]$.
662: Thus we obtain:
663: \begin{eqnarray*}
664: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
665: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
666: & = & E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
667: E[|(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]^{\frac{1}{2}}\\
668: & = & \frac{1}{2}\left\{E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
669: E[|(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]^{\frac{1}{2}}\right.\\
670: & + &
671: \left.E[|(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]^{\frac{1}{2}}
672: E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}\right\}.
673: \end{eqnarray*}
674: Applying Schwarz inequality we obtain:
675: \begin{eqnarray*}
676: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
677: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
678: & = & \frac{1}{2}\left\{E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}
679: E[|(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]^{\frac{1}{2}}\right.\\
680: & + &
681: \left.E[|(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}|^{2}]^{\frac{1}{2}}
682: E[|(D_{t} + D_{t}^{*} - c_{t})Y_{t}|^{2}]^{\frac{1}{2}}\right\}.\\
683: & \geq & \frac{1}{2}\left\{\left|\langle(D_{t} + D_{t}^{*} - c_{t})Y_{t},
684: (-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}\rangle\right|\right.\\
685: & + & \left.\left|\langle(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t},
686: (D_{t} + D_{t}^{*} - c_{t})Y_{t}\rangle\right|\right\}.
687: \end{eqnarray*}
688: Applying the triangle inequality we get:
689: \begin{eqnarray*}
690: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
691: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
692: & \geq & \frac{1}{2}\left|\langle(D_{t} + D_{t}^{*} - c_{t})Y_{t},
693: (-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t}\rangle\right.\\
694: & + & \left.\langle(-D_{t} + D_{t}^{*} + i\tilde{c}_{t})Y_{t},
695: (D_{t} + D_{t}^{*} - c_{t})Y_{t}\rangle\right|.
696: \end{eqnarray*}
697: Using the linearity in the first argument and
698: the conjugate linearity in the second argument of the inner product, and the
699: definition of the adjoint operator, because $c_{t}$ and $\tilde{c}_{t}$
700: are real numbers, the last expression can be simplified and we obtain
701: finally:
702: \begin{eqnarray*}
703: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
704: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
705: & \geq &
706: \frac{1}{2}\left|-2\langle D_{t}Y_{t}, D_{t}Y_{t}\rangle
707: + 2\langle D_{t}^{*}Y_{t}, D_{t}^{*}Y_{t}\rangle\right|\\
708: & = &
709: \left|-\langle D_{t}Y_{t}, D_{t}Y_{t}\rangle
710: + \langle D_{t}^{*}Y_{t}, D_{t}^{*}Y_{t}\rangle\right|\\
711: & = &
712: \left|-\langle D_{t}^{*}D_{t}Y_{t}, Y_{t}\rangle
713: + \langle D_{t}D_{t}^{*}Y_{t}, Y_{t}\rangle\right|\\
714: & = &
715: \left|\langle (D_{t}D_{t}^{*} - D_{t}^{*}D_{t})Y_{t}, Y_{t}\rangle\right|\\
716: & = &
717: \left|\langle [D_{t}, D_{t}^{*}]Y_{t}, Y_{t}\rangle\right|.
718: \end{eqnarray*}
719: Using the commutation relationship (\ref{comdd*}) we obtain:
720: \begin{eqnarray*}
721: & \ & E[(X_{t} - c_{t})^{2}|Y_{t}|^{2}]^{\frac{1}{2}}
722: E[(X_{t} - \tilde{c}_{t})^{2}|\mathcal{G}Y_{t}|^{2}]^{\frac{1}{2}}\\
723: & \geq &
724: \left|\langle \langle X \rangle_{t}Y_{t}, Y_{t}\rangle\right|\\
725: & = & \langle X \rangle_{t}\langle Y_{t}, Y_{t}\rangle\\
726: & = & \langle X \rangle_{t}E[|Y_{t}|^{2}].
727: \end{eqnarray*}
728: \end{proof}
729:
730: \begin{theorem}
731: Let $\{Y_{t}\}_{t \in [0, T]}$ be a stochastic process
732: such that for each $t \in [0$, $T]$,
733: $Y_{t} \in L^{2}(\Omega, \sigma(X_{t}), P)$ and
734: $Y_{t}$ belongs to the domain of $D_{t}$, $D_{t}^{*}$,
735: $D_{t}D_{t}^{*}$, and $D_{t}^{*}D_{t}$.
736: Then for any Borel measurable functions
737: $g$, $\tilde{g} : [0, T] \to {\mathbb R}$, we have:
738: \begin{eqnarray}
739: & \ & E\left[\left|\int_{0}^{T}(X_{t} - g(t))Y_{t}dX_{t}
740: \right|^{2}\right]^{\frac{1}{2}}
741: E\left[\left|\int_{0}^{T}(X_{t} - \tilde{g}(t))
742: \mathcal{G}Y_{t}dX_{t}\right|^{2}\right]^{\frac{1}{2}} \nonumber\\
743: & \geq & E\left[\int_{0}^{T}|Y_{t}|^{2}\langle X \rangle_{t}
744: d\langle X \rangle_{t}\right]. \label{h2}
745: \end{eqnarray}
746: We assume that all the measurability and square integrability conditions,
747: necessary for the existence of the above stochastic integrals, hold.
748: \end{theorem}
749:
750: \begin{proof}
751: Using the isomorphism through which the stochastic integral is extended
752: from the simple processes to the square integrable processes we obtain:
753: \begin{eqnarray*}
754: & \ & E\left[\left|\int_{0}^{T}(X_{t} - g(t))Y_{t}dX_{t}
755: \right|^{2}\right]
756: E\left[\left|\int_{0}^{T}(X_{t} - \tilde{g}(t))
757: \mathcal{G}Y_{t}dX_{t}\right|^{2}\right]\\
758: & = &
759: \int_{0}^{T}E\left[|(X_{t} - g(t))Y_{t}|^{2}\right]
760: d\langle X \rangle_{t}
761: \int_{0}^{T}E\left[|(X_{t} - \tilde{g}(t))\mathcal{G}Y_{t}|^{2}\right]
762: d\langle X \rangle_{t}.
763: \end{eqnarray*}
764: Applying Schwarz inequality we obtain:
765: \begin{eqnarray*}
766: & \ & E\left[\left|\int_{0}^{T}(X_{t} - g(t))Y_{t}dX_{t}
767: \right|^{2}\right]
768: E\left[\left|\int_{0}^{T}(X_{t} - \tilde{g}(t))
769: \mathcal{G}Y_{t}dX_{t}\right|^{2}\right]\\
770: & = &
771: \int_{0}^{T}E\left[|(X_{t} - g(t))Y_{t}|^{2}\right]
772: d\langle X \rangle_{t}
773: \int_{0}^{T}E\left[|(X_{t} - \tilde{g}(t))\mathcal{G}Y_{t}|^{2}\right]
774: d\langle X \rangle_{t}\\
775: & \geq & \left\{\int_{0}^{T}
776: E\left[|(X_{t} - g(t))Y_{t}|^{2}\right]^{\frac{1}{2}}
777: E\left[|(X_{t} - \tilde{g}(t))\mathcal{G}Y_{t}|^{2}\right]^{\frac{1}{2}}
778: d\langle X \rangle_{t}\right\}^{2}.
779: \end{eqnarray*}
780: Applying inequality (\ref{h1}), for $c_{t} := g(t)$ and
781: $\tilde{c}_{t} := \tilde{g}(t)$, we obtain:
782: \begin{eqnarray*}
783: & \ & E\left[\left|\int_{0}^{T}(X_{t} - g(t))Y_{t}dX_{t}
784: \right|^{2}\right]^{\frac{1}{2}}
785: E\left[\left|\int_{0}^{T}(X_{t} - \tilde{g}(t))
786: \mathcal{G}Y_{t}dX_{t}\right|^{2}\right]^{\frac{1}{2}}\\
787: & \geq & \int_{0}^{T}
788: E\left[|(X_{t} - g(t))Y_{t}|^{2}\right]^{\frac{1}{2}}
789: E\left[|(X_{t} - \tilde{g}(t))\mathcal{G}Y_{t}|^{2}\right]^{\frac{1}{2}}
790: d\langle X \rangle_{t}\\
791: & \geq & \int_{0}^{T}\langle X \rangle_{t}E[|Y_{t}|^{2}]
792: d\langle X \rangle_{t}\\
793: & = & E\left[\int_{0}^{T}|Y_{t}|^{2}\langle X \rangle_{t}
794: d\langle X \rangle_{t}\right].
795: \end{eqnarray*}
796: \end{proof}
797: If we take, $Y_{t} := \varphi(X_{t})$, where
798: $\varphi : {\mathbb R} \to {\mathbb C}$ is a Borel measurable function,
799: then we obtain the following:
800:
801: \begin{corollary}
802: Let $\varphi : {\mathbb R} \to {\mathbb C}$ be a Borel measurable function
803: such that for each $t \in [0$, $T]$,
804: $\varphi(X_{t})$ belongs to the domain of $D_{t}$, $D_{t}^{*}$,
805: $D_{t}D_{t}^{*}$, and $D_{t}^{*}D_{t}$.
806: Then for any Borel measurable functions
807: $g$, $\tilde{g} : [0, T] \to {\mathbb R}$, we have:
808: \begin{eqnarray}
809: & \ & E\left[\left|\int_{0}^{T}(X_{t} - g(t))\varphi(X_{t})dX_{t}
810: \right|^{2}\right]^{\frac{1}{2}}
811: E\left[\left|\int_{0}^{T}(X_{t} - \tilde{g}(t))
812: \mathcal{G}\varphi(X_{t})dX_{t}\right|^{2}\right]^{\frac{1}{2}} \nonumber\\
813: & \geq & E\left[\int_{0}^{T}|\varphi(X_{t})|^{2}\langle X \rangle_{t}
814: d\langle X \rangle_{t}\right].
815: \end{eqnarray}
816: We assume that all the measurability and square integrability conditions,
817: necessary for the existence of the above stochastic integrals, hold.
818: \end{corollary}
819:
820: Suppose $X_{t} = B_{t}$ is a Brownian motion process. Then for each
821: $t \in [0$, $T]$, $\langle X \rangle_{t} = t$ is deterministic.
822: The inequality (\ref{h2}) becomes:
823:
824: \begin{theorem}
825: Let $\{Y_{t}\}_{t \in [0, T]}$ be a stochastic process
826: such that for each $t \in [0$, $T]$,
827: $Y_{t}$ is measurable with respect to the sigma-field $\sigma(B_{t})$ and
828: $Y_{t}$ belongs to the domain of $D_{t}$, $D_{t}^{*}$,
829: $D_{t}D_{t}^{*}$, and $D_{t}^{*}D_{t}$.
830: Then for any Borel measurable functions
831: $g$, $\tilde{g} : [0, T] \to {\mathbb R}$, we have:
832: \begin{eqnarray}
833: & \ & E\left[\left|\int_{0}^{T}(B_{t} - g(t))Y_{t}dB_{t}
834: \right|^{2}\right]^{\frac{1}{2}}
835: E\left[\left|\int_{0}^{T}(B_{t} - \tilde{g}(t))
836: \mathcal{G}Y_{t}dB_{t}\right|^{2}\right]^{\frac{1}{2}} \nonumber\\
837: & \geq & E\left[\int_{0}^{T}t|Y_{t}|^{2}
838: dt\right].
839: \end{eqnarray}
840: We assume that all the measurability and square integrability conditions,
841: necessary for the existence of the above stochastic integrals, hold.
842: \end{theorem}
843:
844: In all the inequalities from this section we may replace $\int_{0}^{T}$ by
845: $\int_{0}^{\infty}$ and obtain the same results.
846:
847: \bibliographystyle{plain}
848:
849: \begin{thebibliography}{fs}
850: \bibitem{d} Durrett, R.: {\em Brownian Motion and Martingales in
851: Analysis}{\bf .} The Wadsworth Mathematics Series, 1984.
852: \bibitem{dm}
853: Dym, H. and McKean, H. P.: {\em Fourier Series and Integrals}{\bf .}
854: Academic Press, 1972.
855: \bibitem{fs}
856: Folland, G. B. and Sitaram, A.: The Uncertainty Principle:
857: A Mathematical Survey, {\em The Journal of Fourier Analysis and Applications},
858: Vol. 3, Nr. 3, 1997, 207-238.
859: \bibitem{j} Janson, S.: {\em Gaussian Hilbert Spaces}{\bf .} Cambridge
860: Tracts in Mathematics 129, Cambridge University Press, 1997.
861: \bibitem{s}
862: Strichartz, R.: {\em A Guide to Distribution Theory and Fourier
863: Transforms}{\bf .} CRC Press, 1994.
864: \bibitem{v}
865: Von Neumann, J.: {\em Mathematical Foundations of Quantum Mechanics}{\bf .}
866: Princeton University Press, 1955.
867: \end{thebibliography}
868: \end{document}
869: