Let $W_t$ denote a Brownian Motion, use ito's formula to compute the differential of the following stochastic process: $$X_t=e^{t/2}\cos(W_t)$$

Ito's formula is supposed to be used when we know a diffusion differential equation like $dx=\mu dt+\sigma dW_t$, then we can apply it to solve a $dG(x,t)$ by: $$dG=\left(\frac{\partial G}{\partial t}+\frac{\partial G}{\partial x}\mu+\frac{1}{2}\frac{\partial^2G}{\partial x^2}\sigma^2\right)dt+\frac{1}{2}\frac{\partial G}{\partial x}\sigma dW_t.$$ But for this question, why we can solve the $dX_t$ directly?

Also, the answer to this question suggests that we should use the following Ito's formula to solve $dX_t$: $$dX(t,W_t)=\frac{\partial X(t,W_t)}{\partial t}dt+\frac{\partial X(t,W_t)}{\partial W}dW_t+\frac{1}{2}\frac{\partial ^2 X(t,W_t)}{\partial W^2}dt$$

What's the logic behind this Ito formula?

  • 1
    $\begingroup$ You should try to use precise language and notation. Personally, I don't really see what your point is here. $\endgroup$
    – Tobsn
    Mar 26, 2021 at 12:56
  • 2
    $\begingroup$ The form $dX(t,W_t)=\dots$ that you wrote is just a special case of the more general form you wrote further up with $\mu \equiv 0$ and $\sigma \equiv 1$. $\endgroup$
    – Ian
    Mar 26, 2021 at 13:00

2 Answers 2


I think you have a slight misunderstanding of Itô's lemma. What it says (in a simplified version) is that for any drift-diffusion process $(Y_t)_t$ which is solution of $dY_t = \mu dt +\sigma dW_t $, and any twice-differentiable function $f(t,x)$, it holds that the process $(f(t,Y_t))_t$ is a solution of the SDE : $$df(t,Y_t) = \frac{\partial f}{\partial t}(t,Y_t)dt+\frac{\partial f}{\partial x}(t,Y_t)dW_t+\frac{1}{2}\frac{\partial ^2 f}{\partial x^2}(t,Y_t)\cdot\sigma^2dt$$

In your case, the process $Y_t := W_t$ is indeed a drift-diffusion process and $f:(t,x)\mapsto e^{t/2}\cos(x)$ is twice-differentiable w.r.t. its arguments so Itô's lemma applies.


The usual form of It's formula you note is applied like chain rule, when you need to find $dG(x,t)$ but you only know the dynamics of $dx$. The "regular-Calculus" analog of this is when you know $x(t)$ and $x'(t)$ but need to find $\frac{d}{dx} g(x(t))$.

In the specific case you bring, you definitely could differentiate directly, as your second formula suggests.

Not sure what else you are asking, please post comments or update the question.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .