Issue Description:
Description: I am working on implementing a Physics-Informed Neural Network (PINN) in PyTorch to solve the 1D acoustic wave equation. My goal is to train the PINN to approximate the solution of the wave equation using a combination of boundary conditions and physics loss (PDE residual). However, despite my efforts, the PINN prediction tends to align with the physics and boundary data points but diverges from the exact solution.
PINN.zip
Issue: While training, the PINN prediction closely fits the boundary and physics points, but it does not generalize well to match the exact solution of the 1D wave equation across the entire domain. The issue persists despite increasing the number of training iterations and adjusting hyperparameters such as the weight of the physics loss term.
Issue Description:
Description: I am working on implementing a Physics-Informed Neural Network (PINN) in PyTorch to solve the 1D acoustic wave equation. My goal is to train the PINN to approximate the solution of the wave equation using a combination of boundary conditions and physics loss (PDE residual). However, despite my efforts, the PINN prediction tends to align with the physics and boundary data points but diverges from the exact solution.
PINN.zip
Issue: While training, the PINN prediction closely fits the boundary and physics points, but it does not generalize well to match the exact solution of the 1D wave equation across the entire domain. The issue persists despite increasing the number of training iterations and adjusting hyperparameters such as the weight of the physics loss term.