lululxvi / deepxde

A library for scientific machine learning and physics-informed learning
https://deepxde.readthedocs.io
GNU Lesser General Public License v2.1
2.6k stars 734 forks source link

DeepOnet for the Linear Wave Equation #679

Open engsbk opened 2 years ago

engsbk commented 2 years ago

Hello @lululxvi and all other contributors,

thank you for the great work and continuous support! I have a couple of questions about DeepOnet:

  1. Is there an example for using DeepXDE's DeepOnet implementation for a PDE. This will greatly help in implementing my problem of the wave equation.

  2. Is it necessary to use classical methods for obtaining a solution for training process of DeepOnet?

  3. Can we consider DeepOnet a way to use PINN for extrapolating prediction? If we have the solution of the wave equation for t[0,5] and we feed that to DeepOnet, can DeepOnet infer an operator for predicting the solution for t[5,10]?

The equation I'm referring to is the linear wave equation with a time-dependent source function:

f(x,t):= u_tt-cu_xx-s(x,t) where, s(x,t) = sin(2pif t)

Looking forward for your reply!

lululxvi commented 2 years ago

Please see

Yes, you need a dataset for training DeepONet.

This can be done by the newly added feature of physics-informed DeepONet (PINN + DeepONet), but I haven't added an example online yet.