Firedrake
Firedrake is an automated system for the solution of partial differential equations using the finite element method (FEM). HydroGym's Firedrake backend provides the five 2-D canonical flow-control environments (Cylinder, RotaryCylinder, Pinball, Cavity, Step) and is the recommended starting point for users new to HydroGym, because it can run on a single workstation without any GPU or large cluster.
MAIA
m-AIA (Multi-physics Aachen code) is a high-performance CFD framework developed at the Institute of Aerodynamics (AIA) at RWTH Aachen University. It couples finite-volume, discontinuous Galerkin, Lattice Boltzmann (LBM), and level-set methods on Cartesian meshes and targets massively parallel simulations on systems ranging from single workstations to leadership-class GPU clusters. HydroGym's MAIA backend supports both the LBM and structured finite-volume solvers and provides over 60 flow-control environments in 2-D and 3-D.
NEK5000
NEK5000 is a spectral-element CFD code developed at Argonne National Laboratory. It solves the incompressible Navier-Stokes equations on unstructured quadrilateral (2-D) and hexahedral (3-D) meshes with high-order polynomial bases. HydroGym's NEK5000 backend currently provides a turbulent channel flow environment at $Re_{\tau} = 180$.
JAX
JAX is a high-performance array computing library from Google that combines NumPy-compatible array operations with automatic differentiation, JIT compilation via XLA, and native support for CPUs, GPUs, and TPUs. HydroGym's JAX backend provides fully differentiable spectral solvers for 2-D Kolmogorov turbulence and 3-D turbulent channel flow — examples that are feasible to run on a laptop without any GPU, but run very fast and allow for direct experimentation on a capable GPU.
JAX-Fluids
JAX-Fluids is a fully differentiable compressible CFD solver built on JAX, developed at the Technical University of Munich. It supports high-order spatial reconstruction (WENO, TENO), multi-stage time integration, two-phase flow via level-set and diffuse-interface methods, and immersed boundary methods. Because the entire solver is written in JAX, it supports automatic differentiation through the simulation and scales to 512+ NVIDIA A100 GPUs and 2 048+ TPU-v3 cores. HydroGym's JAX-Fluids backend provides compressible jet engine control environments in 2-D and 3-D.