Enable snapping_points in grid generation

[weiliangjin2021]

TODO:

• Add tests and update CHANGELOG
• Stress tests and check edge cases

Feature Description

Enforce placing grid points at certain locations. This is helpful in improving simulation accuracy by enforcing grid points at certain geometry singularities. Around the snapping point, the maximal step size is unmodified.

Implementation:

• Start from intervals returned by parse_structures. If any interval contains a snapping point, split the interval. The maximal grid step size will be the same at the two splitted intervals.
• We don't split if the insertion results in very small interval. Currently I set the threshold as np.amin(max_dl_list) * 0.5, but it's an open question and worth more investigation.

Example

Here is an example comparing the grids after the insertion of snapping points:

medium = td.Medium(permittivity = 16)

box = td.Structure(
    geometry=td.Box(size=[0.1,]*3, center = (-0.12, 0, 0)), 
    medium=medium,
)

point1 = [0.123, 0.145, 0]
point2 = [-0.148, 0.1, 0]

point_outside = [1, 0.13, 0]

grid_spec=td.GridSpec.auto(
    min_steps_per_wvl=10,
    wavelength=1,
)

grid_spec_snapping = grid_spec.updated_copy(snapping_points=[point1, point2, point_outside])

sim = td.Simulation(
    size=(0.6,) * 3,
    grid_spec=grid_spec,
    run_time=1e-13,
    structures=[box],
    boundary_spec = td.BoundarySpec.all_sides(boundary=td.PECBoundary()),
)
sim_snapping = sim.updated_copy(grid_spec = grid_spec_snapping)

fig, ax = plt.subplots(1, 2, figsize=(8, 5), tight_layout=True)
sim.plot(z=0, ax=ax[0])
sim.plot_grid(z=0, ax=ax[0], colors='red', linewidth=1)

sim_snapping.plot(z=0, ax=ax[1])
sim_snapping.plot_grid(z=0, ax=ax[1], colors='red', linewidth=1)

[tylerflex]

Thanks @weiliangjin2021 see my comments.

[QimingFlex]

Thanks, @weiliangjin2021 ! Finally, the snapping points should replace the diagonally placed meshoverride boxes to handle edges. Regarding the open question about the threshold, from my understanding, it is currently determined by the largest allowable step for each structure interval. Adding two more parameters—one for the smallest allowable step and another for the enforced step at the point—might be useful. Perhaps these two parameters could be combined into one.

[weiliangjin2021]

it is currently determined by the largest allowable step for each structure interval.

Not really. Along a direction, the simulation domain is divided into multiple intervals. Each interval has its own maximal allowed step size. The threshold value is half of the minimal of those steps sizes.

Adding two more parameters—one for the smallest allowable step and another for the enforced step at the point

Could you explain a bit more?

[QimingFlex]

I see. I misunderstood it as the smallest allowed step. So currently, the minimum step is given by 'min_steps_per_wvl' and the largest time step is given by these intervals. Is this correct?

[weiliangjin2021]

I see. I misunderstood it as the smallest allowed step. So currently, the minimum step is given by 'min_steps_per_wvl' and the largest time step is given by these intervals. Is this correct?

Is there a typo in your sentence (how does time step enter here)? 'min_steps_per_wvl' determines the maximal allowed step size in each interval.

[QimingFlex]

Sorry it is a typo, ‘space’ step.

…

On Wed, May 29, 2024 at 5:03 PM weiliangjin2021 ***@***.***> wrote: I see. I misunderstood it as the smallest allowed step. So currently, the minimum step is given by 'min_steps_per_wvl' and the largest time step is given by these intervals. Is this correct? Is there a typo in your sentence (how does time step enter here)? 'min_steps_per_wvl' determines the maximal allowed step size in each interval. — Reply to this email directly, view it on GitHub <#1719 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BGTANRPWAR6MNTL7OCCPIN3ZEY7ANAVCNFSM6AAAAABIIGEIN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMZYGI2TSNJWGU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[weiliangjin2021]

Sorry it is a typo, ‘space’ step.

Well, then I still don't get your question. I guess it should be clear that the threshold is np.amin(max_dl_list) * 0.5?

[QimingFlex]

Sorry it is a typo, ‘space’ step.

Well, then I still don't get your question. I guess it should be clear that the threshold is np.amin(max_dl_list) * 0.5?

I tested with several examples and understand the threshold. The parameter looks good to me!

[momchil-flex]

Looks good, main comment from usability perspective is that in some cases you might want to set the point along e.g. one of the axes only, but I think it's fine, we don't need to complicate things so much for now.