search

lululxvi / deepxde

A library for scientific machine learning and physics-informed learning
https://deepxde.readthedocs.io
GNU Lesser General Public License v2.1
2.68k stars 748 forks source link

Euler Equations - Sod Problem - Shock Capturing Issue #126

Closed alexpapados closed 3 years ago

alexpapados commented 4 years ago

Hello!

I hope all is well.

I need some advice on how to obtain better results for the Sod problem. I have looked through the FAQ on how to train the network better when training fails. I've introduced loss weights, increased the number of points in the domain and increased the number of iterations. Unfortunately, I am not able to capture the proper solutions to this problem. Each solution to rho,u, and p experiences too much dissipation or solutions have random jumps. I have attached my code below. Any advice moving forward would be greatly appreciated.

Euler_Eq.py.zip

lululxvi commented 4 years ago

Euler Equation is not easy to learn because of the shock. You may try put more points near the shock, see https://doi.org/10.1016/j.cma.2019.112789

kmache commented 1 year ago

Dear Lu Lu Please I am also working on PDEs with shock, how can I put more points near the shock? can you provide a piece of code?

lululxvi commented 1 year ago

See FAQ Q: How does DeepXDE choose the training points? How can I use some specific training points?