PDE filter for Brinkman source term

[gloopydoop]

This is a PDE filter used to smooth the boundaries of a Brinkman source term, to avoid oscillations associated with sharp interfaces + spectral methods.

The theory is based on B. S. Lazarov, O. Sigmund (2010) "Filters in topology optimization based on Helmholtz-type differential equations".

[timfelle]

Suggested change

	implicit none
	implicit none

[timfelle]

Maybe use exponential notation instead?

Suggested change

	0.0000000001_rp)
	1.0e-10_rp)

[timofeymukha]

As a general comment, could we come up with more specific names? filter_t is definitely too generic, since we already have e.g. a filter for LES. PDE_filter is better, but perhaps could be concretized as well. I would also use pde, we generally avoid capitals unless really needed.

[timfelle]

As a general comment, could we come up with more specific names? filter_t is definitely too generic, since we already have e.g. a filter for LES. PDE_filter is better, but perhaps could be concretized as well. I would also use pde, we generally avoid capitals unless really needed.

A counter proposal would be, shouldn't the LES filter derive from the generic filter type then, we will probably add more filters later, meaning the abstract base class should have some name which is easy to understand.

on the PDE_filter name, sure, it might be better to call it a helmholz_filter, since that is the PDE implemented for this particular filter.

[gloopydoop]

Both good suggestions.
I just had a little read of the elementwise_filter_t and I suppose with the exception of fields vs arrays, the elementwise_field_filter_3d fits filter_apply. @timofeymukha, I guess you've done a lot LES stuff, what are your thoughts?

Regarding the names, yeh I sort of expected this to be problematic. We also now have filter.f90 and filters.f90 😅

I'm happy with helmholz_filter, that's a good suggestion.

[timfelle]

Both good suggestions. I just had a little read of the elementwise_filter_t and I suppose with the exception of fields vs arrays, the elementwise_field_filter_3d fits filter_apply. @timofeymukha, I guess you've done a lot LES stuff, what are your thoughts?

Regarding the names, yeh I sort of expected this to be problematic. We also now have filter.f90 and filters.f90 😅

I'm happy with helmholz_filter, that's a good suggestion.

Yes that is also a thing 😅 I guess we need to move the smooth_step functions somewhere else as well.

[timofeymukha]

@Shiyu-Sandy-Du implemented the LES filtering. I think if you guys can figure out a good abstract interface that would fit a reasonably large class of filters, that would be great!