Consider the one-dimensional SDE $$ dX_t = f(X_t)dt + \sigma(X_t)dW_t,\quad X_0 = 1, $$ where $W_t$ is a Brownian motion under the measure $\mathbb{P}$, in its natural filtration. Suppose that $f(x)-\sigma^2(x)/2x=0$ and $\sigma^2(x)\le x^2$. Show that $X_t>0$ almost surely.

My initial attempt was to consider the log-process $\log(X_t)$. Using Ito's lemma one finds $$ d\log(X_t) = \frac{1}{X_t}\left(f(X_t) - \frac{\sigma^2(X_t)}{2X_t}\right)dt + \frac{\sigma(X_t)}{X_t}dW_t = \frac{\sigma(X_t)}{X_t}dW_t. $$ Integrating now leads to $$ \log(X_t)=\int_0^t\frac{\sigma(X_s)}{X_s}dW_s. $$ By the Ito isometry, the variance of the above process is bounded by $t$: $$ \text{Var}\big[\log(X_t)\big] = \mathbb{E}\left[\int_0^t\frac{\sigma^2(X_s)}{X_s^2}ds\right] \le t. $$ I feel like this is heading in the right direction, but I'm struggling to see what the next step is. Any suggestions would be very welcome.


1 Answer 1


Define stopping times by

$$T_k := \inf\{t \geq 0; X_t \leq k^{-1}\}, \qquad k \in \mathbb{N}$$


$$T := \inf\{t \geq 0; X_t \leq 0\}.$$

Following the reasoning in your question we find that

$$\mathbb{E} \left( \left| \log(X_{t \wedge T_k}) \right|^2 \right) \leq t.$$

By the continuity of the sample paths of $(X_t)_{t \geq 0}$, we have $X_{T_k}(\omega) = k^{-1}$ for $\omega \in \{T_k<\infty\}$, and therefore we get that

$$\mathbb{E} \bigg( 1_{\{T_k \leq t\}} \underbrace{|\log(X_{T_k})|^2}_{=|\log(k^{-1})|^2} \bigg) \leq t.$$

Since $\{T \leq t\} \subseteq \{T_k \leq t\}$ this shows, in particular, that

$$\mathbb{E}\left(1_{\{T \leq t\}} |\log(k^{-1})|^2 \right) \leq t,$$


$$\mathbb{P}(T \leq t) \leq \frac{t}{|\log(k^{-1})|^2}.$$

As $|\log(k^{-1})|^2 \to \infty$ as $k \to \infty$ we conclude that $\mathbb{P}(T \leq t)=0$. As $t>0$ is arbritrary, this proves the assertion.

  • 2
    $\begingroup$ How can I be half as good as you at this $\endgroup$
    – user515599
    Jun 7, 2018 at 6:39
  • $\begingroup$ Brilliant, thanks a lot saz! Could I just confirm that here $T:=\inf\{t\ge0 : X_t = 0\}$? $\endgroup$
    – Alex W
    Jun 7, 2018 at 8:28
  • 1
    $\begingroup$ @AlexW Yes, exactly; it seems that I got lost in my thoughts (somehow I thought that you had already introduced $T$ in your question). $\endgroup$
    – saz
    Jun 7, 2018 at 8:29

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.