source shape as geometric object

[oskooi]

Currently, the shape of a Source can only be a rectangular prism defined by its size property. Although the amplitude function property (amp_func) does enable arbitrary shapes, this is cumbersome to set up for simple geometries. It would be useful if the source shape could also be defined by a GeometricObject (e.g., Sphere, Cylinder, Prism, etc.).

[soamaven]

This is really interesting. I am curious, what situations would this be useful in, specifically a homogeneous volumetric source? LEDs, gain systems? I've only seen volumetric sources utilized in situations when a known field profile is to be injected, which as you said, is provided by amp_func.

[oskooi]

One application would be for designing grating in-couplers for silicon photonics where a cleaved fiber is used as the input source and placed in the near-field of the grating. See e.g., Optics Express, Vol. 18, pp. 18278-83, 2010. Another application would be to model a VCSEL source, which also tend to have cylindrical cross sections, for use in e.g., next-generation displays such as AR/MR smart glasses.

Once this feature is ready, we will then need to modify the eigenmode source such that even though it is computed from a given volume/cross-section it can be placed anywhere in the cell.

[stevengj]

@oskooi, I don't think we actually need a geometric-object source for that application:

• As you say, the eigenmode source would have to be modified anyway to map it to an oblique slice, and once you did that you probably wouldn't need the geometric-object source.

• Even easier, you can take a Cartesian x/y/z slice of an oblique waveguide and use an ordinary eigenmode source, with one modification — tell MPB that the third lattice vector is in the oblique direction, not orthogonal to the plane. This should make it compute an oblique cross-section of the mode traveling down your oblique waveguide, and create the corresponding current source via the principle of equivalence.

[oskooi]

The following schematic demonstrates the functionality that we are trying to enable. An eigenmode source is defined for a Cartesian x/y/z slice of an oblique waveguide which generates a unidirectional propagating mode. Telling MPB that the mode's wavevector is not orthogonal to the slice is not enough; we would also like to be able to specify its eig_parity somehow just as we would if the waveguide is oriented along the x axis and we could specify either EVEN_Y or ODD_Y.

[stevengj]

It looks like #675 addresses the oblique-waveguide source motivation of this issue.

However, we might as well leave this open. There are occasionally circumstances where you want a finite-volume source instead of a point source, and it would be nice to support a spherical source volume etcetera with subpixel averaging. Implementing subpixel averaging of the edges is something that requires a low-level C++ implementation, where it is straightforward as I commented in #635, and cannot easily be done in user-level Python code.

But it doesn't seem like a high-priority feature.