Physics-Informed Neural Networks in Python
Wavefunction |ψ(x,t)| evolution over time
This animation was generated only using PhysAI's built-in Schrodinger Residual Function and trained on the pre-built PINN on just 500 Epochs!
PhysAI is a Python package for solving ODEs and PDEs using Physics-Informed Neural Networks (PINNs). It integrates physics directly into neural network training, allowing the solution of classic physics problems without relying on traditional numerical solvers.
PhysAI supports a wide range of physics problems including:
Key features:
git clone https://github.com/yourusername/physai.git
cd physai
pip install -r requirements.txt
Python >= 3.10 recommended.
physai/
├── __init__.py
├── models.py
├── trainer.py
├── visualization.py
├── pde_residual.py
├── losses.py
├── utils.py
examples/
├── example_schrodinger.py
├── example_newton_cooling.py
├── example_markov.py
├── example_photoelectric.py
├── example_planck.py
README.md
requirements.txt
.gitignore
import torch
from physai.models import PINN
from physai.pde_residual import pde_residual
from physai.losses import pinn_loss
from physai.visualization import plot_1d_solution
# Define a 1D logistic growth ODE
def logistic(x, y):
r, K = 1.0, 1.0
return torch.autograd.grad(y, x, grad_outputs=torch.ones_like(y), create_graph=True)[0] - r*y*(1 - y/K)
# Create a PINN model
model = PINN(layers=[1, 20, 20, 1], activation='tanh')
# Training points
x_train = torch.linspace(0, 5, 100).reshape(-1,1)
# Train the model
from physai.trainer import Trainer
trainer = Trainer(model, collocation_points=x_train, pde_type='logistic')
history = trainer.train(epochs=500, lr=1e-3)
# Plot solution
plot_1d_solution(model, x_train, title='Logistic Growth')