Overdamped Langevin dynamics:

$$ \frac{dx}{dt} = -\frac{1}{\gamma} V'(x) + \sqrt{\frac{2k_B T}{\gamma}} \eta(t) $$

Underdamped Langevin:

$$ m \frac{d^2x}{dt^2} = -\gamma \frac{dx}{dt} - V'(x) + \sqrt{2 \gamma k_B T} \eta(t) $$

where $\eta(t)$ is a random variable.

Why do we use Ito chain rule in the first case and standard chain rule in the second one to compute the time evolution of functions of $x(t)$?

Is it because the last one is a second order equation so the noise term in the solution is of order $dt^{3/2}$ and negligible?


In the first case you have $$\mathrm{d}x_t = -\frac{1}{m\gamma}V'(x_t)\mathrm{d}t + \sqrt{\frac{2k_BT}{m^2\gamma}}dW_t$$ so indeed for $f(t, x)$ you would have by Ito $$\mathrm{d}f_t = \frac{\partial f}{\partial t}\mathrm{d}t + \frac{\partial f}{\partial x} \mathrm{d}x_t + \frac{1}{2}\frac{\partial^2f}{\partial x^2}(\mathrm{d}x_t)^2$$

As for the latter case, we have $$\mathrm{d}\mathbf{x}_t= A(\mathbf{x}_t)\mathrm{d}t + \sigma\mathrm{d}W_t$$ where $\mathbf{x}_t = (x_t, v_t)^\mathrm{T}$, $A(x_t) = \begin{pmatrix}0 & 1 \\ -\frac{V'(x_t)}{m} & -\frac{\gamma}{m}x_t\end{pmatrix}$, $\sigma = (0, \sqrt{2 \gamma k_BT}/m)^\mathrm{T}$.

Now $f(t, x)$ by Ito is: $$\mathrm{d}f_t = \frac{\partial f}{\partial t}\mathrm{d}t + \frac{\partial f}{\partial x} \mathrm{d}x_t + \frac{1}{2}\frac{\partial^2f}{\partial x^2}(\mathrm{d}x_t)^2$$ but $(\mathrm{d}x_t)^2 = 0$ (the noise is in increments to velocity, i.e. $g(t, v)$ would have the noise term in Ito).

| cite | improve this answer | |
  • $\begingroup$ can you expand on why (dx-t)^2=0? $\endgroup$ – iacolippo Oct 2 '16 at 13:29
  • 1
    $\begingroup$ @iacolippo Because $\mathrm{d}x_t = v_t\mathrm{d}t$. $\endgroup$ – alarge Oct 2 '16 at 13:32

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.