Custom epsilon array or map (in h5py format) in FDTDx?

[aa16oaslak]

Hey,

I have a basic question- Can we define random geometries in a array and give that as a input to the simulation? (similar to the epsilon_input_file function in MEEP FDTD code)

[ymahlau]

So sorry for the late reply! I saw your post in the discussion section, but then life happened and i forgot to reply.

To answer your question: Yes, this is possible. There are two ways you can go about this:

• The place_objects function returns (among other things) an ArrayContainer object, which contains the ArrayContainer.permittivity and ArrayContainer.permeability fields as 3d jax arrays of shape (size_x, size_y, size_z). You can freely hardcode anything you want in there before you run the simulation.

• In fdtdx, there is a special object called Device, which can map parameters to simulation materials. You can take a look at the ceviche corner example to see how this is done: https://github.com/ymahlau/fdtdx/blob/main/examples/optimize_ceviche_corner.py. This is very useful if you want to automatically optimize your design by calculating the gradient through the differentiable simulation. If you don't define any mapping function, then you can simply put the material indices (ordered from lowest to highest permittivity) in the params structure to get your desired structure.

[hao14chasing]

Hello @aa16oaslak, have you succeed in implementing the custom epsilon array? If so, how do you visualize that custom shape? On my end, it always shows the strcuture as of the partial_real_shape, even though I used ceviche_corner example. I wonder if you were able to resolve this. Thanks!

[ymahlau]

For full coverage, I am writing here again, you can use the plot_material function to visualize the actual permitivitty distribution. This is a new function that is not yet included in the latest pip version, so you will need to work with the version from the main branch.