Syver Døving Agdestein

drag · one velocity component from a turbulence simulation — left the full field u, right the filtered field ū that a large-eddy simulation gets to see. My research lives in the difference.

Turbulent flows cannot be simulated eddy-for-eddy at realistic scales, so we filter: resolve the large motions, and model the effect of everything the filter removed. I'm a postdoctoral researcher in the Scientific Computing group at CWI in Amsterdam working on that model — the closure — with a current focus on probabilistic approaches to simulating turbulence. I did my PhD at Eindhoven University of Technology on data-driven closure models for large-eddy simulation, treating the numerical discretization as part of the physics rather than an afterthought.

The papers, talks, and interactive essays here all pull on that one thread, and the software behind them is open: differentiable, GPU-accelerated solvers in Julia. Away from the keyboard I'm drawn to history, languages, and the piano — I've lived in Asker, Gembloux, Toulouse, Saint Petersburg, Paris, and now Amsterdam.

research publications talks software blog cv

Play with the research

Each recent paper ships with an interactive companion — the same ideas, with the knobs exposed:

What symmetry buys a learned turbulence model — try to break equivariance yourself
Your grid is a filter — why the most-modeled stress tensor in CFD is the wrong target
Your time integrator is a filter — the largest coarse-simulation error nobody models