MathIsimple

Stochastic Processes – Problem 48: Prove that if is a three-dimensional standard Brownian motion, then almost surely

Question

Prove that if (Bt)t0(B_{t})_{t\geq0} is a three-dimensional standard Brownian motion, then almost surely limtBt=\lim_{t\to\infty}|B_{t}|=\infty.

Step-by-step solution

Step 1. Three-dimensional standard Brownian motion Bt=(Bt(1),Bt(2),Bt(3))B_t=(B_t^{(1)},B_t^{(2)},B_t^{(3)}) has independent one-dimensional standard Brownian motion components, continuous paths, and modulus Bt=(Bt(1))2+(Bt(2))2+(Bt(3))2|B_t|=\sqrt{(B_t^{(1)})^2+(B_t^{(2)})^2+(B_t^{(3)})^2}. The key harmonic function is f(x)=1xf(x)=\frac{1}{|x|} on R3{0}\mathbb{R}^3\setminus\{0\}, satisfying Δf(x)=0\Delta f(x)=0. By Ito's formula, f(Bt)f(B_t) is a local martingale when Bt0B_t\neq 0.

Step 2. For 0<r<R0<r<R, define the exit time Tr,R=inf{t0Btr or BtR}T_{r,R}=\inf\{t\geq0 \mid |B_t|\leq r \text{ or } |B_t|\geq R\}. The process f(BtTr,R)f(B_{t\wedge T_{r,R}}) is a bounded martingale since 1Rf1r\frac{1}{R}\le f\le\frac{1}{r} on the annulus.

Step 3. By the optional stopping theorem, E[f(BTr,R)]=f(B0)E[f(B_{T_{r,R}})]=f(B_0). After adjusting for the singularity at the origin using the first exit time TrT_r from the ball of radius rr, we obtain the boundary probability equation. Setting P1P_1 as the probability of returning to the inner sphere and P2P_2 as reaching the outer sphere: P11r+P21R=1rP_1\cdot\frac{1}{r}+P_2\cdot\frac{1}{R}=\frac{1}{r}, which gives P1=0P_1=0 and P2=1P_2=1.

Step 4. Letting RR\to\infty, the probability of returning to the ball of radius rr after leaving it is zero. This means three-dimensional Brownian motion is transient.

Step 5. For any M>0M>0, take r=Mr=M. By transience, there exists a stopping time Tr<T_r<\infty a.s. such that for all t>Trt>T_r, Bt>M|B_t|>M a.s. Since MM is arbitrary, limtBt=\lim_{t\to\infty}|B_t|=\infty a.s.

Final answer

Three-dimensional standard Brownian motion (Bt)t0(B_t)_{t\geq0} almost surely satisfies limtBt=a.s.\boxed{\lim_{t\to\infty}|B_t|=\infty \quad \text{a.s.}} QED.

Marking scheme

The following is the rubric for this problem. Please score according to the guidelines below.


1. Checkpoints (max 7 pts)

Apply exactly one path for scoring (if the student mixes methods, take the highest-scoring single path).

Chain A: Martingale and Harmonic Function Method (Official Solution)

  • Key function and martingale property [2 pts]
  • State that f(x)=1/xf(x) = 1/|x| is harmonic on R3{0}\mathbb{R}^3\setminus\{0\} (i.e., Δf=0\Delta f = 0), or
  • Use Ito's formula to show f(Bt)f(B_t) is a local martingale.
  • *Note: If only writing the function without mentioning harmonicity or the martingale property, 0 pts.*
  • Construct stopping time and boundedness argument [2 pts]
  • Introduce the annular region {x:r<x<R}\{x : r < |x| < R\} and the first exit time T=inf{t:Btr or BtR}T = \inf\{t : |B_t| \le r \text{ or } |B_t| \ge R\}. (1 pt)
  • State that the process is bounded before this stopping time, hence the optional stopping theorem applies. (1 pt)
  • Establish probability equation [2 pts]
  • Apply the optional stopping theorem to set up the equation relating boundary probabilities, e.g., P(hit r)1r+P(hit R)1R=1xstartP(\text{hit } r) \cdot \frac{1}{r} + P(\text{hit } R) \cdot \frac{1}{R} = \frac{1}{|x_{start}|}.
  • Take limit and final conclusion [1 pt]
  • Let RR \to \infty to deduce the escape probability is 1 (or the return probability is <1<1 combined with the Markov property), concluding limtBt=\lim_{t\to\infty} |B_t| = \infty.

Chain B: Green's Function / Analytic Method (Alternative)

  • Integral / Green's function computation [3 pts]
  • Correctly write and compute the time integral of the 3D Brownian motion transition density (heat kernel), or directly cite the Green's function G(x,y)C/xyG(x,y) \sim C/|x-y|.
  • Key point: show 1t3/2dt<\int_1^\infty t^{-3/2} dt < \infty (convergence).
  • Transience derivation [2 pts]
  • Argue that the finite integral implies: for any bounded set KK, the total occupation time in KK is finite (01K(Bt)dt<\int_0^\infty 1_K(B_t) dt < \infty a.s.).
  • Final conclusion [2 pts]
  • From "finite occupation time in any bounded set" logically deduce limtBt=\lim_{t\to\infty} |B_t| = \infty.

2. Zero-credit items

  • Merely copying definitions of Brownian motion or standard properties (normal increments, independence) without substantive derivation.
  • Merely citing the conclusion "Brownian motion is transient for d3d \ge 3" without proof.
  • Incorrectly using the 2D harmonic function f(x)=logxf(x) = \log|x| or the 1D function f(x)=xf(x)=x.
  • Claiming limtBt\lim_{t\to\infty} B_t (vector limit) exists or equals infinity (conceptual error; it is the modulus that diverges).

3. Deductions

Note: Deductions cannot reduce the total below 0. Apply the most severe single deduction.

  • Ignoring singularity/domain issues (-1 pt): Directly setting initial point B0=0B_0=0 when applying 1/x1/|x| without any treatment (e.g., starting from ϵ\epsilon or using a stopping time to avoid the origin), causing a division-by-zero logical gap.
  • Martingale rigor deficiency (-1 pt): Claiming 1/Bt1/|B_t| is a martingale on the entire interval [0,)[0, \infty) (it is only a local martingale; it must be truncated/stopped to form a bounded martingale).
  • Logical confusion (-2 pts): Confusing "transience" (limBt=\lim |B_t|=\infty) with "recurrence" (infinitely many returns to a region), e.g., computing return probability as 1 yet concluding the modulus diverges.
Ask AI ✨