I have to solve a bunch of fairly high dimensional SDE's, but I only have a diagonal diffusion matrix/control term.
As it turns out this is a fairly big bottleneck in my case, because I have already optimized the ode term.
Is there a way to use a sparse matrix to speed up the integration?
I tried to sparsify() diffeqsolve with the jax.sparse module, but I could not get it to work.
Passing just the diagonal as vector results in a the dot product being used for every dimension and one gets the same noise realization for each dimension.
Is there a way to cleverly pass only the diagonal array?
Hi,
I have to solve a bunch of fairly high dimensional SDE's, but I only have a diagonal diffusion matrix/control term. As it turns out this is a fairly big bottleneck in my case, because I have already optimized the ode term.
Is there a way to use a sparse matrix to speed up the integration? I tried to sparsify() diffeqsolve with the jax.sparse module, but I could not get it to work.
Passing just the diagonal as vector results in a the dot product being used for every dimension and one gets the same noise realization for each dimension. Is there a way to cleverly pass only the diagonal array?
Thanks for the help!