mitsuba-renderer/drjit

Higher order derivatives

daseyb opened this issue · 1 comments

daseyb commented

Is there a way to compute higher-order partial derivatives of a multi-dimension function?

That is, for some scalar-valued function $f(x,y)$ I want to compute $\frac{\partial^2 f(x,y)}{\partial x \partial y}$.

I tried some variations of multiple calls to backwards, but I've always gotten a crash complaining about missing gradients.

Thanks!

wjakob commented

It's not supported, Dr.Jit only computes first-order derivatives.