Using DeepONet for Boundary Condition Mapping and Custom Function Spaces

[skhaleghir]

Hello, I'd like to first thank you Professor Lu for creating and sharing this remarkable tool with the community.

I am currently working on training a neural operator to map the flux at a boundary condition (as a function of time) to the concentration profile in a time-dependent 1D diffusion problem (details attached). While implementing this, I encountered the following error:
"flux_function() missing 1 required positional argument: 'v'"

I would greatly appreciate your guidance on resolving this issue.

Additionally, I have a second question: Is there documentation or guidance on defining custom function spaces? I aim to train the model under specific flux conditions that cannot be represented using the existing options, such as "PowerSeries" or GRF.

Thank you for your time and assistance!

Saeed

[lululxvi]

RobinBC doesn't support input function not. I guess we just need to change this line

deepxde/deepxde/icbc/boundary_conditions.py

Line 105 in 3544fdf

X[beg:end], outputs[beg:end]

to

X[beg:end], outputs[beg:end], aux_var[beg:end]

You need to implement a new class to generate functions. Check the code of PowerSeries or GRF, and it should be easy to impelement.

[skhaleghir]

Thank you, Lu, for the suggestions. I will try them and let you know if I have any further questions.