Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $B$ be a Brownian motion. I'm trying to show that

$$\left(\int_0^te^{B_s}ds\right)^\frac{1}{\sqrt{t}}$$

convergences in distribution as $t \to \infty$. As a hint, we are told to consider the limit as $p\to\infty$ of

$$\left(\int_0^1|f(x)|^pdx\right)^\frac{1}{p}$$

I can show that this is the essential supremum of $|f|$ by a sandwiching argument, but I'm having trouble relating the hint to the question.

As an alternative approach, I showed the laws were tight using Prokhorov's Theorem, but don't know how to make the step from tightness to convergence.

Thank you.

share|improve this question

2 Answers 2

up vote 4 down vote accepted

this is essentially a "Laplace Method" argument that allows you to conclude that : $\left(\int_0^te^{B_s}ds\right)^\frac{1}{\sqrt{t}}$ converges in law to $exp(S_1)=e^{\sup_{s\leq 1}\textit{B}_s}$.

you have the following equalities (sometime only in law) $(\int_0^te^{B_s}ds)^\frac{1}{\sqrt{t}}= (t\int_{0}^{1}e^{\sqrt{t}\textit{B}_s}ds)^\frac{1}{\sqrt{t}}=t^\frac{1}{\sqrt{t}}.||e^{\textit{B}_.}||_{L^\sqrt{t}[0,1]} $ The Laplace Method allows you to conclude that $Lim_{t\to \infty}||e^{\textit{B}_.}||_{L^\sqrt{t}[0,1]}=||e^{\textit{B}_.}||_{L^\infty[0,1]}=e^{\sup_{s\leq 1}\textit{B}_s}$

The only thing you need to apply Laplace Method is that the function is $L^{\infty}$ which is true a.s. for every path of $e^{\textit{B}_.}$ over the interval $[0,1]$

Regards

Edit proof of Laplace Method : Let's take a positve $f\in L_\mu^{\infty}(X)$, then first we have for any $p>1$ :
$\|f\|_{L^p}\leq \|f\|_{L^{\infty}}\cdot (\mu(X))^{1/p}$ (here $\mu(X)=1$ but this works for finite measures aswell)

Which gives half of the result.

Second, let's fix $\epsilon>0$, and note $A_\epsilon=\{x \in X s.t.|f(x)|\geq \|f\|_{L^{\infty}}-\epsilon\}$, and such that $\mu(A_{\epsilon})>0$ (there exists such an $\epsilon$ and for every values inferior to this value, the corresponding set has non zero measure, this can be seen by absurd)

Then we have:

$\int_{X}{|f(s)|}^p d\mu(s)=\int_{A_{\epsilon}}{|f(s)|}^p d\mu(s)+\int_{A_{\epsilon}^c}{|f(s)|}^p d\mu(s) \geq (\|f\|_{L^{\infty}}-\epsilon)^p \cdot \mu(A_{\epsilon}) $

Taking the p-th root we get :
$(\|f\|_{L^{\infty}} - \epsilon)\cdot (\mu(A_{\epsilon}))^{1/p} \leq \|f\|_{L^p}$

Letting p going to $\infty$ ($\epsilon$ still fixed) we get :

$ \lim_{p \rightarrow +\infty}(\|f\|_{L^{\infty}} - \epsilon)\cdot (\mu(A_{\epsilon}))^{1/p} = \|f\|_{L^{\infty}} - \epsilon \leq \lim_{p \rightarrow +\infty} \|f\|_{L^p}$.(*)

As this is true for any $\epsilon>0$ we get the desired result.

(*)$1\ge\mu(A_{\epsilon})>0$

share|improve this answer
    
Thank you. This is a good idea, although I can't see how to make it rigorous. –  Ben Derrett Sep 17 '11 at 20:09
    
@Ben Derrett: Hi where is it exactly you think this proof is not rigourous enough ? Regards –  TheBridge Sep 18 '11 at 9:20
    
To prove this, we need an analogue of Laplace's method that can handle a non-$C^2$ function. I can't see how to find this without some kind of sandwiching argument. The function being in $L^\infty$ is not sufficient to apply Laplace's Method - what happens if the function has no unique global maximum? –  Ben Derrett Sep 18 '11 at 15:40

I guess that the indication is to mimick the sandwiching argument, starting from something like the fact that the random variable $X_t$ you are interested in is such that $$ A_t^{1/\sqrt{t}}\mathrm e^{(M_t-1)/\sqrt{t}}\le X_t\le t^{1/\sqrt{t}}\mathrm e^{M_t/\sqrt{t}}, $$ where $M_t=\max\{B_s\mid 0\le s\le t\}$ and $A_t=\displaystyle\int\limits_0^t[B_s\ge M_t-1]\mathrm d s$.

In the RHS, $t^{1/\sqrt{t}}\to1$ and $M_t/\sqrt{t}$ equals $M_1$ in distribution.

In the LHS, $(M_t-1)/\sqrt{t}$ converges to $M_1$ in distribution. If $\log A_t=o(\sqrt{t})$ almost surely (a fact which should follow from Williams's decomposition theorems and/or a Ray-Knight theorem for the local time of $B$), one sees that $X_t$ converges in distribution to $\mathrm e^{M_1}$.

share|improve this answer
    
Thank you - this must be the meaning of the hint. :) Perhaps there's a more elementary way of showing $\log A_t = o(\sqrt{t})$? –  Ben Derrett Sep 17 '11 at 19:34

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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