2D convolution with user_function about meep_adjoint_simple

[smartalecH]

I'm able to circumvent the issue by filtering the parameters, rather than the actual permittivity values (as we discussed). However, if I simulate a higher density of design parameters than meep pixels, there are several gradients that go to zero. This seems intuitively correct, but the finite-difference gradients are not zero. They are typically 6 orders of magnitide smaller than similar simulations when I don't oversample my design space. Is this neglibible? Or is there something else we should be aware of?

[stevengj]

Seems fine to me, but it seems like in practice you will always use a filter radius that is larger than a Meep pixel, in which case no gradients should vanish.

[smartalecH]

Good point, thanks!