Let $B=(B_t)_{t}$ be a standard Brownian motion and let $X=(B_{t/2})_t$. I want to disprove that $\mu_X\ll\mu_B$ where $\mu_\cdot$. denotes the respective distribution. The standard way might be to note first that $\langle X\rangle_t=t/2$ whilst $\langle B\rangle_t=t$ and thus for $A=\{f\in C[0,1]\;|\;\langle f\rangle_1=1/2\}$ $$ \mu_X(A)=1,\qquad\mu_B(A)=0. $$ Is there another way of proving this?

  • 1
    $\begingroup$ They are singular, not absolutely continuous. Your method works, another one might be to set for instance $$A=\bigg\{ f: \limsup_{t\to 0} \frac{f(t)}{\sqrt{2t\log\log t}}=1 \bigg\}$$ or $$A:=\bigg\{f: \lim_{n\to \infty} \frac{1}{n} \sum_{k=1}^n 2^{k+1}(f(2^{-k})-f(2^{-k-1}))^2=1\bigg\}$$A standard result of Cameron-Martin theory states that dilations of Gaussian measures on infinite-dimensional banach spaces are always mutually singular. $\endgroup$ – Shalop Feb 6 '18 at 21:28
  • $\begingroup$ *should be $\log \log(1/t)$ $\endgroup$ – Shalop Feb 6 '18 at 23:00
  • $\begingroup$ Let us now consider $X_t=B_t+\int_0^t \sin(B_s)\;ds$. Why are the distributions in this case equivalent? $\endgroup$ – julian Feb 7 '18 at 16:18
  • $\begingroup$ Hint: Fix some $T>0$, and define a martingale $M_t:= -\int_0^t \sin(B_s) dB_s$, and define a measure $Q$ on $\mathcal F_T$ by $Q(A) = E_P[e^{M_T-\frac{1}{2}\langle M \rangle_T}\cdot 1_A]$. Since $B$ is a $P$-BM, it follows from Girsanov that $B-\langle M,B \rangle$ is a $Q$-BM. $\endgroup$ – Shalop Feb 7 '18 at 20:43

Let $$D=\Bigl\{f\in C[0,1]: \lim_{n\to\infty}\sum_{k=1}^{2^n}[f(\tfrac{k}{2^n})-f(\tfrac{k-1}{2^n})]^2=1\Bigr\}.\qquad A=C[0,1]\setminus D. $$ Then \begin{gather} \mu_X(D)=0,\qquad \mu_B(D)=1.\\ \mu_X(A)=1, \qquad \mu_B(A)=0. \end{gather}


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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