rippl 0.5.0

High-performance physics compiler and PINN engine.

rippl

The physics compiler for neural networks.

Welcome to rippl. If you've ever felt that training neural networks to solve Partial Differential Equations (PDEs) requires too much boilerplate, math-heavy configuration, and bloated code, you are exactly where you need to be.

Traditional simulation engines take days to compute on supercomputers. rippl is a modern, hyper-optimized PINN (Physics-Informed Neural Network) compiler built on PyTorch. It trains an AI to understand the fundamental laws of physics, reducing inference time to milliseconds.

The best part? You don't need a PhD in numerical methods to use it. If you can snap Lego blocks together, you can build a physics solver.

---

📖 The Official Guide

👉 Click here to read the full Documentation (docs.md)

This repository contains a definitive, beginner-friendly learning guide. We highly recommend starting there. It will take you from a basic 1D heat wave all the way to operator learning and multi-objective thermodynamics.

---

⚡ Quickstart: The 7-Line Promise

We believe in showing, not telling. Here is the absolute minimum code required to solve the Heat Equation ($\frac{\partial u}{\partial t} = 0.1 \frac{\partial^2 u}{\partial x^2}$) using rippl:

import torch
import rippl

# 1. Physics
eq = rippl.Equation([(1.0, rippl.TimeDerivative(order=1)), (-0.1, rippl.Laplacian())])

# 2. Domain
domain = rippl.Domain(bounds=[(0, 1)])

# 3. Constraints
bcs = [rippl.Dirichlet(0, value=0.0), rippl.Dirichlet(1, value=0.0)]
ic  = rippl.Initial(lambda x: torch.sin(torch.pi * x[:, 0:1]))
constraints = rippl.Constraints(bcs + [ic], fields=["u"])

# 4. Engine
solver = rippl.Solver(equation=eq, domain=domain, constraints=constraints)

# 5. Execute
model = solver.solve(epochs=3000)
You just built a PyTorch neural network, enforced boundaries algebraically, sampled coordinate points, and trained an AI solver.

🧰 What's in the Box? While the core API is simple, the underlying engine is industrial-grade. rippl gives you access to:

Operator Learning: Out-of-the-box support for DeepONets and 1D Fourier Neural Operators (FNOs) so your network learns the rules of physics, not just a single solution.

Mesh-Free Geometry: Build complex 2D domains using Constructive Solid Geometry (CSG) math blocks—no painful meshing required.

Intelligent Training: Native support for Causal Training (respecting time), Adaptive Collocation Sampling, and Pareto Front Balancing for coupled multi-physics systems.

The Migration Engine: Instantly transpile your old, bloated legacy scripts (like DeepXDE) into optimized rippl syntax using rippl.migrate().

🚀 Next Steps Install the package: pip install rippl

Dive into the docs: Open docs.md to see how these simple blocks scale up to 2D aerodynamics, shockwave capturing, and complex thermal stress simulations.