Implementing a CUDA kernel for solving parabolic partial differential equations


Some idea on how to implement a CUDA kernel for solving 2D or even better 3D time dependent heat conduction (solving parabolic partial differential equations)?


I have a similar problem, but with higher order derivatives. I compute the derivatives in k -space and employ a semi -implicit integraation in time. I use cufft combined with my kernels. For real space there are some efficient finte difference algorithm, easy to implement.