Experimenting with boundary conditions

[kerim371]

Hi,

I feel like default boundary condition requires to greatly expand computational grid so it absorbs waves.
I would like to play around this.
Especially I wanted to compare Absorbing Boundary Conditions (ABC) and Perfectly Matched Layers (PML).
Could you please guide me where it is implemented in JUDI and how can I start start modifying BC?
Also for now JUDI uses ABC or PML?

[mloubout]

The current boundary condition is a diffusion mechanism that translate to solving the equation

$$ m \frac{d^2 u}{dt^2} - \Delta u + \eta \frac{du}{dt} = 0 $$

where $\eta$. is zero in the physical domain and increasing in the boundary layer to progressively damp and diffuse the waves in it.

The damping profile is implemented here

and the pde in kernels.py

If you want to look at something fancier like PMLs you might wanna look at this devito tutorials

https://github.com/devitocodes/devito/tree/master/examples/seismic/abc_methods

[kerim371]

@mloubout according to the Devito examples, PML expects ndims*2 (or probably ndims if they are pairwisely equal) damping fields.
Thus I guess we have to overwrite Model.damp from Function type to Tuple(Function).
Then I probably have to reimplement damp_op method so it return Tuple(Operator) (one operator for damp field) and call each of them within initialize_damp() method.
Finally modify kernel for each of type of equation.

May this work?

[kerim371]

Draft PR #210