Probabilistic heat diffusion on spherical shell

78 Views Asked by At

I'd like to sample random paths to approximate the solution to the heat equation on the unit sphere shell. Specifically, I'd like to solve $u_t = \alpha\Delta u$ with some initial condition $u_0$. In spherical coordinates, unless I have made an error, I believe the pde is $$ u_t = \alpha\left(u_{\theta\theta} + \cot(\theta) u_\theta + \csc^2(\theta) u_{\phi\phi}\right). $$

Assuming I have written the pde correctly, then we may write $$ u(\theta,\phi,t) = \mathbb{E}\left[u_0\left(X_t,Y_t\right)\,\middle|\,X_0 =\theta, Y_0 = \phi\right], $$ where $$ \mathrm{d}X_t = \cot(X_t)\mathrm{d}t + \sqrt{\alpha}\mathrm{d}W_t,\qquad\mathrm{d}Y_t = \sqrt{\alpha}\left|\csc(X_t)\right|\mathrm{d}W'_t. $$ I am assuming that $\theta\in[0,\pi]$ measures the angle from the north pole and $\phi\in[-\pi,\pi]$. $W$ and $W'$ are independent white noise processes.

I have two main (related) issues at this point. If I were to implement this numerically, how would I handle sampling paths when $X_t$ and $\theta$ are near zero or $\pi$?

Beyond this, how does one interpret the SDE? Intuitively we should have that for every point $p$ on the sphere, the probability of transitioning to a different area should be relatively the same. So if I were at the north or south pole, the probability transition to nearby areas should be the same as those anywhere else on the sphere. So I should be able to handle the poles by rotation calculations.

At the same time, the SDE for $X_t$ has a drift term that I cannot get an intuition for. Why would heat diffusion be biased to migrate around the sphere in a particular direction? Why wouldn't $Y_t$ show such a drift term too?

Any help you could lend would be greatly appreciated.

Edited to fix a variable typo.