1
$\begingroup$

I would like to understand how the Wiener measure of some simple sets can be evaluated. I will sketch the construction of Wiener measure I have in mind:

We denote the one point compactification of $\mathbb{R}^n$ by $\hat{\mathbb{R}}^n$. Now consider the product space $$\Omega:=\prod\limits_{0\leq t<\infty}\hat{\mathbb{R}}^n.$$ By Tychonoff's theorem this space is a compact and hausdorff space. Now the idea is to construct the wiener measure with the aid of riesz representation theorem. Consider first functions like this $\varphi:\Omega\rightarrow \mathbb{R}, \varphi(w)=F(w(t_1),...,w(t_n))$ with $n\in\mathbb{N}$, $t_1<t_2<...<t_n$ and some continuous F. Now define the following linear functional on all those functions given by:

$$\Lambda(\varphi)=\int....\int p(t_1,x,x_1)p(t_2-t_1,x_1,x_2)...\times p(t_n-t_{n-1},x_{n-1},x_n) F(x_1,...,x_n)dx_1....dx_n,$$

where $p(t,x,y)$ denotes the heat kernel on $\mathbb{R}^n$.

The space of all such $\varphi$ is dense in $C(\Omega)$ and therefore we can extend $\Lambda$ continuously on the whole space $C(\Omega)$. By Riesz representation theorem there exists a regular Borel measure $\mu$, s.t. for all continuous functions $f$ we have:

$$\Lambda(f)=\int f d\mu$$.

This measure is called Wiener measure.

Now I want to understand how to calculate measures like $\mu(G)$, whereas $G:= \lbrace{ w:[0,\infty)\rightarrow \mathbb{R}^n \vert \omega(t)\in U\rbrace}$ with fixed $U\subset \mathbb{R}^n$ open. The problem is that the characteristic function $1_{\lbrace{ w(t)\in G , 0\leq t<\infty\rbrace}}$ is not continuous, therefore I can't calculate it like $\Lambda(1_{{\lbrace{ w(t)\in G , 0\leq t<\infty\rbrace}}})$

I would really appreciate any help!

$\endgroup$
7
  • 2
    $\begingroup$ The usual construction goes via Kolmogorov's extension theorem, followed by Kolmogorov's continuity test. This gives you a Borel measure on the space of continuous functions. The set $\{w\,:\, w(t) \in G\,\forall t\}$ is a perfectly nice Borel set there. In order for it to have non-zero measure, you will need $n \ge 3$ and $G$ unbounded. $\endgroup$ Jul 13, 2014 at 12:35
  • $\begingroup$ There is something wrong with the definition, as $ w (t) $ cannot belong to a set of $ w $'s. $\endgroup$ Jul 13, 2014 at 12:53
  • 1
    $\begingroup$ I don't think you are really asking the right question here. Asking "In general, how do I compute the Wiener measure of sets" is tantamount to asking "In general, how do I prove theorems about Brownian motion." It is a very large theory and there are a great variety of techniques. Some very deep and difficult theorems (as well as open questions) can be stated as asking for the Wiener measure of a particular easily-described Borel set. So if there's a specific set whose measure you want to know, you should ask that by itself. $\endgroup$ Jul 14, 2014 at 1:53
  • $\begingroup$ @Martin Hairer: Thank you very much for your comment. I will consider the construction via Kolmogorov soon. But I'm not a probabilist. $\endgroup$
    – supersnail
    Jul 14, 2014 at 8:23
  • $\begingroup$ @Kjos-Hanssen: Why do you mean the set can't be a Borel set? $\endgroup$
    – supersnail
    Jul 14, 2014 at 8:24

2 Answers 2

2
$\begingroup$

First let $t$ be fixed and look at the set $\{ w \mid w(t) \in U\}$. Take a sequence of continuous functions $f_k$ that are uniformly bounded and converge to the indicator function of a set $U$ in the Borel-Algebra of $\mathbb{R}^n$. Let $\pi_t$ be the projection on the $t$-th component of Omega. Then $$\int_\Omega (\pi_t)^*f_k d \mu = \int_{\mathbb{R}^n} p(t, x, y) f_k(y) d y$$ by construction. Because the $f_k$ are bounded and the measure is finite, they are dominated by an integrable function and by Lebesgue's theorem on bounded convergence, $$\mu(\{w(t) \in U\}) = \lim_{k \rightarrow \infty}\int_\Omega (\pi_t)^*f_k d \mu = \int_{\mathbb{R}^n} p(t, x, y) \lim_{k \rightarrow \infty} f_k(y) d y = \int_{U} p(t, x, y) d y $$ Hence non-continuity is no real problem.

To now tackle your problem, notice that the indicator functions of the cylinder sets $$ G_{N} := \{ w \mid w(k/N) \in U \forall k = 1, \dots, N^2 \}$$ converges a.s. to the indicator function of $G := \{ w \mid w(t) \in U \forall t \geq 0 \}$ as $N \longrightarrow \infty$: Clearly $G \subset G_N$. On the other hand, if $w \notin A$, then $w(q) \notin U$ for some rational number $q$ and therefore $w \notin G_N$ for $N$ big enough.

The dominated convergence theorem gives then $$ \mu(G) = \lim_{N \rightarrow \infty} \int_U \dots \int_U \left(\prod_{k=1}^{N^2} p(1/N, x_{k-1}, x_k) \right) d x_1 \dots d x_{N^2}$$ If this is helpful at all, I cannot say.

$\endgroup$
1
  • $\begingroup$ Thank you very much for your help! I have to think some time about your approach and probably will tell you whether it helps. $\endgroup$
    – supersnail
    Jul 14, 2014 at 8:56
1
$\begingroup$

The probability $h(x,t)$ that the Brownian motion path started from $x$ stays in a given set $U$ for $0<s<t$ satisfies the heat equation: $\Delta_x h-\partial_t h=0$ with initial data $1$ and boundary data zero. The large $t$ asymptotics is governed by the leading eigenvalue of the Laplacian and by projection of the constant function $1$ to the line spanned by the leading eigenfunction.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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