Software

Most of my research runs on software I build and maintain in the open. If a paper of mine makes a claim, there is a repository where you can rerun it.

IncompressibleNavierStokes.jl

IncompressibleNavierStokes.jl is the main package: a differentiable, GPU-accelerated incompressible Navier–Stokes solver in Julia, designed for large-eddy simulation and data-driven closure modeling. Being differentiable end-to-end means neural closure models can be trained through the solver ("a-posteriori" training), not just on precomputed snapshots.

It is described in the paper A differentiable software suite for accelerated simulation of turbulent flows and in the blog post Writing a differentiable fluid solver in Julia.

Research codes

Each paper ships with the code that produced it:

• SymmetryCode.jl — equivariant closure models and the experiments of the symmetry paper
• ExactClosure.jl — exact unresolved-stress expressions for finite-volume LES (paper)
• DLES15CompanionCode — time integration as filtering (paper)
• DivergenceConsistency — divergence-consistent neural closure models (paper)
• DiscreteFiltering.jl — discrete filtering in 1D (paper)

Tutorials

NeuralClosureTutorials contains pedagogical notebooks on learning neural closure models for fluid flows, used in lectures and schools.

Contributions

I have contributed to GEMSEO (multidisciplinary design optimization), SpinDoctor (diffusion-MRI simulation), and Artery.FE (1D blood-flow simulation in FEniCS, JOSS paper).