search

soanagno / rba-pinns

Implementation of fast PINN optimization with RBA weights
MIT License
40 stars 3 forks source link

Two questions #4

Closed xztu2121 closed 2 months ago

xztu2121 commented 3 months ago
  1. In the file "allen-cahn_rba-sa.ipynb," I found that your boundary is directly read from the data file, but the upper spatial boundary in the data file is not equal to 1.
  2. In the file "allen-cahn_rba-sa.ipynb," why is loss_ub set to 0? There doesn't seem to be any related handling of the boundary in the code . Is there an issue with this setting?
soanagno commented 3 months ago

Hi, thanks for reaching out.

  1. The upper and lower boundary velocities read from the data file (uu2, uu3) are not used anywhere for solving AC, they were there for tests so we removed them. Also there is no Dirichlet condition equal to 1 for AC.

  2. We've made a correction to the symmetry bc, you can now activate it in the updated code. Note that this bc is not needed for good convergence.

xztu2121 commented 3 months ago

Thank you very much for your answer. Can I understand that the inclusion of boundary loss in the loss function may not significantly affect the overall convergence,? In fact, the boundaries provide supervisory information, which appears to have a certain impact.

soanagno commented 2 months ago

Yes, for this problem the information from the initial condition is enough.