Efficient Differentiable n-d PDE Solvers in JAX.

[Neurips 2024] A benchmark suite for autoregressive neural emulation of PDEs. (≥46 PDEs in 1D, 2D, 3D; Differentiable Physics; Unrolled Training; Rollout Metrics)

Incompressible Navier-Stokes solver

[IROS 2024] [ICML 2024 Workshop Differentiable Almost Everything] MonoForce: Learnable Image-conditioned Physics Engine

Electromagnetics simulation library for moving point charges built on JAX

[NeurIPS 2024] NeuMA: Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics

A simple JAX-powered simulation library for numerical experiments of fuzzy dark matter, stars, gas + more!

This repo contains the differentiable physics simulation module in "PPR: Physically Plausible Reconstruction from Monocular Videos". ICCV 23.

Differentiable Flexible Mechanical Metamaterials

Universal Partial Differential Equations Simulator

Generic framework for building differentiable models.

Simple differentiable approximate ocean models built with JAX.

A library for soft differentiable relaxations of common JAX functions.

Repository for MHPI differentiable hydrological models.

A differentiable astrophysics simulator in JAX

Training methodologies for autoregressive neural operators/emulators in JAX.

Controlling fluids in a reduced-dimensional simulation.

Repository for NeurIPS 2025 paper, "Physics-informed Reduced Order Modeling of Time-dependent PDEs via Differentiable Solvers."

Official code implementation of the paper on adaptive node position in biological transport networks

A library for soft differentiable relaxations of common PyTorch functions.

Stochastic PDE solvers (SPDE) built on top of exponax: Exponential Euler-Maruyama stepper for the stochastic Allen-Cahn equation with additive/multiplicative Q-Wiener noise, tamed nonlinearities, ensemble utilities, Richardson extrapolation, and a Strang-split hybrid SSA scaffold.

A high performance data-driven and differentiable SPH fluid solver

Differentiable cortical folding simulator in JAX for forward mechanics and inverse growth-field recovery.

A JAX framework for differentiable neuromechanical flight optimization. Unifies neural CPG, articulated body dynamics, and unsteady aerodynamics to synthesize Lyapunov-stable insect flight.

rigidRL is a compiled C++ autograd library backed by Eigen. It provides a graph-based differentiation engine with seamless Python bindings, designed specifically for accelerating physics-based optimization tasks like 6-DOF drone control and nonlinear dynamics.

Differentiable Earth System Model (Ocean, Atmos, Land, Ice) in JAX for GPU/TPU High-Performance Simulation.