Chapter 7

Path properties of Brownian motion

[8/0%]
LaTeX
Problem 7.1

Here you are asked to find a more precise description of the modulus of continuity of Brownian paths. Prove that

lim⁡δ→0sup⁡s,t∈[0,1],0<∣t−s∣<δ∣Wt−Ws∣δlog⁡(1/δ)<∞, a.s.  \lim _{\delta \rightarrow 0} \sup _{s, t \in [0,1], 0<|t-s|<\delta } \frac{\left|W_{t}-W_{s}\right|}{\sqrt{\delta \log (1 / \delta )}}<\infty , \quad \text{ a.s. }

Hint: Imitate the proof of Theorem 7.1.

?
Problem 7.2

The following is part of what is known as Chung's law of the iterated logarithm. We will see in Section 40.3 that there exists c1c_{1} such that

P(sup⁡s≤t∣Ws∣≤λ)≤c1e−π2t/8λ2 \mathbb {P}\left(\sup _{s \leq t}\left|W_{s}\right| \leq \lambda \right) \leq c_{1} e^{-\pi ^{2} t / 8 \lambda ^{2}}

for t/λ2t / \lambda^{2} sufficiently large. Prove that

lim inf⁡t→∞sup⁡s≤t∣Ws∣t/log⁡log⁡t<∞, a.s.  \liminf _{t \rightarrow \infty } \frac{\sup _{s \leq t}\left|W_{s}\right|}{\sqrt{t / \log \log t}}<\infty , \quad \text{ a.s. }
?
Problem 7.3

Let WtW_{t} be a one-dimensional Brownian motion. We will see in Section 40.3 that there exists c2c_{2} such that

P(sup⁡s≤t∣Ws∣≤λ)≥c2e−π2t/βλ2 \mathbb {P}\left(\sup _{s \leq t}\left|W_{s}\right| \leq \lambda \right) \geq c_{2} e^{-\pi ^{2} t / \beta \lambda ^{2}}

if t/λ2t / \lambda^{2} is sufficiently large. Prove that

lim inf⁡t→∞sup⁡s≤t∣Ws∣t/log⁡log⁡t>0, a.s.  \liminf _{t \rightarrow \infty } \frac{\sup _{s \leq t}\left|W_{s}\right|}{\sqrt{t / \log \log t}}>0, \quad \text{ a.s. }

This is the other half of Chung's law of the iterated logarithm. In fact,

lim inf⁡t→∞sup⁡s≤t∣Ws∣t/log⁡log⁡t=c, a.s. (7.6) \liminf _{t \rightarrow \infty } \frac{\sup _{s \leq t}\left|W_{s}\right|}{\sqrt{t / \log \log t}}=c, \quad \text{ a.s. } \tag {7.6}

Identify cc and prove (7.6).

?
Problem 7.4

A function ff is Hölder continuous of order α\alpha at a point tt if there exists cc such that ∣f(u)−f(t)∣≤|f(u)-f(t)| \leq c∣u−t∣αc|u-t|^{\alpha } for all uu. Suppose α>1/2\alpha >1 / 2 and WtW_{t} is a Brownian motion. Show that the event

A={∃t∈[0,1]:W is Ho¨lder continuous of order α at t) A=\{ \exists t \in [0,1]: W \text{ is Hölder continuous of order } \alpha \text{ at } t)

has probability 0 .

Hint: Imitate the proof of nowhere differentiability, but use more than three time intervals.

?
Problem 7.5

Let WW be a one-dimensional Brownian motion and let Mt=sup⁡s≤tWsM_{t}=\sup_{s \leq t} W_{s} (with no absolute value signs). Prove that if ε>0\varepsilon >0, then

lim inf⁡t→∞Mtt/(log⁡t)1+ε>0, a.s.  \liminf _{t \rightarrow \infty } \frac{M_{t}}{\sqrt{t} /(\log t)^{1+\varepsilon }}>0, \quad \text{ a.s. }
?
Problem 7.6

This is a complement to Exercise 4.10. Prove that if p>2p>2 and WW is a Brownian motion, then the pp-variation of WW, defined in Exercise 4.10 , is finite, a.s.

Hint: Use the fact that the paths of Brownian motion are Hölder continuous of order α\alpha if α<1/2\alpha <1 / 2.

?
Problem 7.7

Let WW be a Brownian motion and let ZZ be the zero set: Z={t∈[0,1]:Wt=0}Z=\left\{ t \in [0,1]: W_{t}=0\right\}.

?
(1)

Show there exists a constant cc not depending on xx or δ\delta such that

P(∃s≤δ:Ws=−x)≤P(sup⁡s≤δ∣Ws∣≥∣x∣)≤ce−x2/2δ \mathbb {P}\left(\exists s \leq \delta : W_{s}=-x\right) \leq \mathbb {P}\left(\sup _{s \leq \delta }\left|W_{s}\right| \geq |x|\right) \leq c e^{-x^{2} / 2 \delta }
(2)

Use the Markov property of Brownian motion to show that there exists a constant cc not depending on ss or tt such that

P(Z∩[s,t]≠∅)≤c(1∧t−ss) \mathbb {P}\left(Z \cap [s, t] \neq \emptyset \right) \leq c\left(1 \wedge \sqrt{\frac{t-s}{s}}\right)
Problem 7.8

Given a Borel measurable subset AA of [0,1][0,1], define

Hγ(A)=lim⁡δ→0sup⁡[inf⁡{∑i=1∞[bi−ai]γ:A⊂∪i=1∞[ai,bi],sup⁡i∣bi−ai∣≤δ}] H_{\gamma }(A)=\lim _{\delta \rightarrow 0} \sup \left[\inf \left\{ \sum _{i=1}^{\infty }\left[b_{i}-a_{i}\right]^{\gamma }: A \subset \cup _{i=1}^{\infty }\left[a_{i}, b_{i}\right], \sup _{i}\left|b_{i}-a_{i}\right| \leq \delta \right\} \right]

In other words, cover AA by the union of intervals [ai,bi]\left[a_{i}, b_{i}\right] and define the analog of Lebesgue measure. The differences are that we look at ∣bi−ai∣γ\left|b_{i}-a_{i}\right|^{\gamma } but do not require that γ\gamma be one, and we require that none of the intervals be longer than δ\delta. The quantity Hγ(A)H_{\gamma }(A) is called the Hausdorff measure of AA with respect to the function xγx^{\gamma }. The Hausdorff dimension of a set AA is defined to be

inf⁡{γ:Hγ(A)>0}=sup⁡{γ:Hγ(A)=∞} \inf \left\{ \gamma : H_{\gamma }(A)>0\right\} =\sup \left\{ \gamma : H_{\gamma }(A)=\infty \right\}

(For subsets of Rd\mathbb {R}^{d}, we replace the intervals [ai,bi]\left[a_{i}, b_{i}\right] by balls of radius ri⋅r_{i \cdot }.) As a warm-up to this exercise, prove that the Hausdorff dimension of the standard Cantor set in [0,1][0,1] is log⁡2/log⁡3\log 2 / \log 3.

The purpose of this exercise is to show that if WW is a Brownian motion and Z={t∈[0,1]Z=\{ t \in [0,1] : Wt=0}\left.W_{t}=0\right\} is the zero set, then the Hausdorff dimension of ZZ is no more than 1/21 / 2.

?
(1)

For each nn, let Cn\mathcal{C}_{n} be the collection of intervals [i/2n,(i+1)/2n]\left[i / 2^{n},(i+1) / 2^{n}\right] contained in [0,1][0,1] that intersect ZZ. (Cn\left(\mathcal{C}_{n}\right. is random.) If #Cn\# \mathcal{C}_{n} is the cardinality of Cn\mathcal{C}_{n}, use Exercise 7.7 to show

E[#Cn]≤∑i=02n−1P(Z∩[i/2n,(i+1)/2n]≠∅)≤c2n/2 \mathbb {E}\left[\# \mathcal{C}_{n}\right] \leq \sum _{i=0}^{2^{n}-1} \mathbb {P}\left(Z \cap \left[i / 2^{n},(i+1) / 2^{n}\right] \neq \emptyset \right) \leq c 2^{n / 2}
(2)

Write

∑[i/2n,(i+1)/2n]∈Cn∣2−n∣γ=2−nγ#Cn \sum _{\left[i / 2^{n},(i+1) / 2^{n}\right] \in \mathcal{C}_{n}}\left|2^{-n}\right|^{\gamma }=2^{-n \gamma } \# \mathcal{C}_{n}

Use the Chebyshev inequality and (1) to conclude that the Hausdorff dimension of ZZ is less than or equal to 1/21 / 2, a.s. (We will show that it is at least 1/21 / 2 in Exercise 14.10.)