simulation.subsection to include PML

[weiliangjin2021]

ElectromagneticFieldData.dot to reindex coords for aligning almost identical coordinates

[weiliangjin2021]

One test still errored. So I think this implementation is not robust. I'll try the custom grid idea.

[weiliangjin2021]

Please check out the new implementation. CustomGridBoundaries is risky as we don't validate if the coordinate covers exactly the entire simulation domain. So for now I plan not to expose it to external users: it will not be added to doc API, and not added to tidy3d namespace. It can only be used in the imported classmethod.

[tylerflex]

This will make things so much easier! Thanks @weiliangjin2021

Dont forget changelog

[weiliangjin2021]

Please check out the new implementation. CustomGridBoundaries is risky as we don't validate if the coordinate covers exactly the entire simulation domain. So for now I plan not to expose it to external users: it will not be added to doc API, and not added to tidy3d namespace. It can only be used in the imported classmethod.

Maybe I can validate it in simulation's root validator: sim.simulation_bounds align with the grid coords?

[weiliangjin2021]

@dbochkov-flexcompute now that I added the validator to validate the simulation domain boundaries match the grid boundaries, mode_solver.reduced_simulation_copy will fail the validation because it removes PML layers. Should I add it back?

[dbochkov-flexcompute]

@dbochkov-flexcompute now that I added the validator to validate the simulation domain boundaries match the grid boundaries, mode_solver.reduced_simulation_copy will fail the validation because it removes PML layers. Should I add it back?

hmm, not sure if that would resolve the issue. I think we should make CustomGridBoundaries just to match only original sim bounds, without pml, so that pml layers are constructed as usual. That would also avoid an issue when subsection is taken in region of non-uniform grid and that wouldn't have uniform spacing near boundaries like pml. Btw, are coordinates outside of target region discarded? Can't find the place where it is done.

[weiliangjin2021]

hmm, not sure if that would resolve the issue. I think we should make CustomGridBoundaries just to match only original sim bounds, without pml, so that pml layers are constructed as usual.

Well, that can be achieved with CustomGrid. In my test, the grid in original sim bounds with CustomGrid matches exactly. However, once we start to construct pml layers, there is machine precision issue (the grid differ by around 1e-15). That is reason for this CustomGridBoundaries to include the complete grids without any further constructions.

Btw, are coordinates outside of target region discarded? Can't find the place where it is done.

No postprocessing is currently performed. The coordinate must span exactly the simulation domain size.

[weiliangjin2021]

hmm, not sure if that would resolve the issue. I think we should make CustomGridBoundaries just to match only original sim bounds, without pml, so that pml layers are constructed as usual.

Well, that can be achieved with CustomGrid. In my test, the grid in original sim bounds with CustomGrid matches exactly. However, once we start to construct pml layers, there is machine precision issue (the grid differ by around 1e-15). That is reason for this CustomGridBoundaries to include the complete grids without any further constructions.

You are right. There is something wrong in my test setup. I'll refactor it to optionally construct pml layers.

[weiliangjin2021]

@dbochkov-flexcompute now that I added the validator to validate the simulation domain boundaries match the grid boundaries, mode_solver.reduced_simulation_copy will fail the validation because it removes PML layers. Should I add it back?

hmm, not sure if that would resolve the issue. I think we should make CustomGridBoundaries just to match only original sim bounds, without pml, so that pml layers are constructed as usual. That would also avoid an issue when subsection is taken in region of non-uniform grid and that wouldn't have uniform spacing near boundaries like pml. Btw, are coordinates outside of target region discarded? Can't find the place where it is done.

Now CustomGridBoundaries works almost identical to CustomGrid, except that it takes grid boundaries rather than grid size.

[dbochkov-flexcompute]

Looks great! Thanks for fixing this and making it more robust. Just a few minor comments below

[tylerflex]

couple minor things.