Martin Schubert

Hi @xpdong22, @HiHatDoDot, you might be interested in [fmmax](https://github.com/facebookresearch/fmmax), which implements FMM/RCWA and supports anisotropic magnetic materials and uniaxial PML. We have one example involving PML, which consists of a...

I developed a workaround for my use case, which involves using the `jax.experimental.host_callback` module. Just sharing it in case it's useful. ``` def _eig_host(matrix: jnp.ndarray) -> Tuple[jnp.ndarray, jnp.ndarray]: """Wraps jnp.linalg.eig...

A brief update to this: we have a slightly modified version of this which avoids the device specification in the call to `jax.jit`, which is the new recommended practice: ```python...

> As `pure_callback` does not seem to support fp64 at the moment, you need additional tricks (in case you are using fp32). We don't seem to have issues supporting fp32...

Yes, I agree some form of leaderboard would be very useful. This could be based on the [imageruler](https://github.com/NanoComp/imageruler/tree/main/imageruler), although one would need to think carefully e.g. about how to handle...

The leaderboard with manufacturing metrics is now live, here: docs: https://invrs-io.github.io/leaderboard/readme.html repo: https://github.com/invrs-io/leaderboard

@oskooi @stevengj We discussed that problems where large LDOS enhancement is sought would be of interest, similar to the cavity design problem in the [photonics-opt-testbed](https://github.com/NanoComp/photonics-opt-testbed/tree/main/cavity_design). A 3D version could be...

I did some tests comparing `eig` performance for scipy, numpy, jax, and torch and found that they can differ quite a bit, with torch generally being the fastest. In lieu...

I was planning on it, but haven’t found the way to automatically document the entire package yet. I am using jupyterbooks for my notebooks.

@smartalecH Should we close this? I guess it's no longer needed?