Is the 2D local solver still supported, or do the examples only work with the 2D online solver?

[CosimoMV]

I was testing the “100 Ohm differential stripline benchmark” example, but using the 2D solver that runs locally. I know it does not perform subpixel averaging, but even when I increase the grid resolution significantly, I still get noticeable discrepancies (and also a discontinuity around 10Ghz) compared to the graphs shown in the example.

Is there some basic incompatibility between the 2D local mode solver and the examples in the example library?

[momchil-flex]

Very timely question! We have just released version 2.10.0, and with it an optional package tidy3d-extras, which contains some C++ bindings, and which should allow you to use the local solver to get exactly the same results as the remote one. Could you try something like pip install --upgrade tidy3d[extras] or just pip install tidy3d==2.10.0 && pip install tidy3d-extras==2.10.0 and see what you get?

Practically speaking the reason you're observing this is that the local solver (without -extras) staircases the permittivity, which for PEC structures in particular can be really bad. It's not just that the permittivity is slightly inaccurate, PEC is effectively a boundary condition and that boundary condition is not enforced correctly.

[CosimoMV]

I tried with pip install --upgrade tidy3d[extras] but I still have the same problem with the local mode solver.

[CosimoMV]

I reread the guide and noticed that the API key registration was missing. Now, even when using the local mode solver, I get results that visually appear to be the same as those from the web mode solver.

[momchil-flex]

Great!