Question about periodic boundary loss in reaction-diffusion

[leastima]

Hello, I'd like to ask why the grad of x in periodic boundary is added to boundary loss when nu > 0 in reaction-diffusion system.

I'm using the code to test basic PINN on the reaction-diffusion equation. I've noticed that adding the loss term associated with "u_pred_lb_x - u_pred_ub_x" calculated on line 174 of net_pbc.py to loss_b seems to make the model worse. For example, when using parameters nu=4 and rho=5, the average l2 error of 5 seeds with line 174 is approximately 0.8, while that without line 174 is only 0.1. I feel a little confused about why this term is added as a supplement to the periodic boundary loss when nu > 0.

Thank you for your time.