JAX - Differentiable Physics & PDEs
JAX is uniquely suited for physics because it can differentiate through numerical solvers. This guide covers how to implement traditional PDE solvers that are "optimization-friendly" and how to build neural-hybrid physical models.
When to Use
- Solving Navier-Stokes, Wave, or Heat equations on GPU.
- Implementing Physics-Informed Neural Networks (PINNs).
- Performing Inverse Design (finding material properties from observations).
- Creating differentiable simulations for robotics or climate modeling.
- Sensitivity analysis of physical systems.
Core Principles
1. Differentiation through the Solver
In JAX, if you write an Euler or Runge-Kutta integrator using jax.numpy, you can automatically calculate ∂Result/∂InitialCondition or ∂Result/∂Viscosity.
2. Staggered Grids & Vmap
Physical fields (velocity, pressure) are often stored on grids. JAX's vmap allows you to parallelize solvers across different boundary conditions or parameter sets instantly.
3. The Adjoint Method
For very large systems, JAX's reverse-mode autodiff effectively implements the "Adjoint State Method" used in traditional CFD/Geophysics for gradient calculation.
Implementation Patterns
1. PINNs (Physics-Informed Neural Networks)
import jax.numpy as jnp
from jax import grad, vmap
# A simple MLP representing the solution u(x, t)
def model(params, x, t):
# standard neural net logic...
return result
# Residual of the PDE: u_t + u*u_x - nu*u_xx = 0 (Burgers Equation)
def pde_loss(params, x, t, nu):
u = lambda x, t: model(params, x, t)
# Automatic derivatives of the MODEL
u_t = grad(u, argnums=1)(x, t)
u_x = grad(u, argnums=0)(x, t)
u_xx = grad(grad(u, argnums=0), argnums=0)(x, t)
return jnp.mean((u_t + u * u_x - nu * u_xx)**2)2. Differentiable Finite Difference Solver
@jit
def update_step(u, dt, dx, nu):
"""One step of a diffusion solver."""
# Vectorized Laplacian using shifts (Zero-copy views)
u_left = jnp.roll(u, -1)
u_right = jnp.roll(u, 1)
laplacian = (u_left + u_right - 2*u) / (dx**2)
return u + dt * nu * laplacian
# We can now differentiate this solver!
def loss(initial_u, target_u):
final_u = integrate_pde(initial_u) # Loop of update_step
return jnp.sum((final_u - target_u)**2)
grad_initial_condition = grad(loss)(initial_u, target_u)Critical Rules
✅ DO
- Use jax.lax.scan for time loops - Standard Python for loops create massive XLA graphs.
scancompiles the loop into a single efficient kernel. - Normalize your Grids - Like ML, PINNs converge faster if x, t are scaled to [0,1] or [-1,1].
- Combine Data and Physics - Use PINNs where you have some sensor data + the physical law to "fill the gaps".
- Use Double Precision for Physics - Use
jax.config.update("jax_enable_x64", True)for sensitive numerical solvers.
❌ DON'T
- Don't use PINNs for everything - Traditional solvers (FDM/FEM) are much faster for "forward" problems. PINNs excel at "inverse" problems.
- Don't ignore Boundary Conditions (BCs) - In PINNs, BCs must be added to the loss function: Loss = PDE_loss + BC_loss.
- Don't forget the 'Ghost Cells' - When implementing FDM, handle boundaries carefully to avoid artifacts.
Practical Workflows: Inverse Problem
Finding Viscosity from a Video of Fluid
def objective(nu_guess):
# 1. Run simulation with nu_guess
final_state = run_simulation(initial_state, nu_guess)
# 2. Compare with experimental data
return jnp.mean((final_state - experimental_frame)**2)
# Gradient descent to find the real physical property
optimal_nu = optimize(grad(objective))JAX PDE transforms physics from a static simulation into a dynamic, optimizable landscape. It allows researchers to ask "What physical parameters produced this result?" and find the answer through the power of gradients.