Differentiable Simulation
for Fusion Plasmas

Independent research foundation building differentiable plasma turbulence solvers and gradient-based inference methods, validated against published benchmarks.

Registered non-profit (stichting) — Amsterdam, the Netherlands

Research

We develop fully differentiable pseudo-spectral solvers for plasma turbulence — enabling gradient-based parameter inference, state estimation, and data assimilation directly through the turbulent dynamics. Our Hasegawa-Wakatani solver is validated against Biskamp (1994) and Camargo (1995) benchmarks at resolutions up to 1024², and cross-validated against independent implementations. Built in JAX; all methods and code will be published openly.

Results

Direct numerical simulation of Hasegawa-Wakatani drift-wave turbulence. C = 1.0, N = 512, pseudo-spectral IMEX ETD-RK4 solver. Fully differentiable — gradients flow through every timestep.

< 1%
Parameter recovery
7/7 C values
0.024%
Field reconstruction
φ relative error
10⁵×
Error reduction
14 DA cycles
0.01
Biskamp k⁻³ slope
accuracy
Quantity C = 1.0 (ours) C = 1.0 (ref) Error C = 5.0 (ours) C = 5.0 (ref) Error
Total energy E 3.82 ± 0.21 4.4 13% 6.76 ± 0.38 8.3 19%
Particle flux Γₙ 0.61 ± 0.05 0.73 17% 0.129 ± 0.019 0.13 1%
Partition Eᴺ/E 0.551 ± 0.009 0.57 3% 0.753 ± 0.018 0.77 2%
Eₖ slope −2.30 ± 0.20 −2.9 −3.47 ± 0.06 −3.5 1%
Eₖᴺ slope −2.67 ± 0.18 −2.7 1% −4.56 ± 0.08 −4.2 9%

Camargo, Biskamp & Scott, Phys. Plasmas 2 (1995) 48–62, Tables I & II. N = 512, ν = 10⁻⁸, ∇⁶ hyperviscosity. Cross-validated against hw2d (Greif, github.com/the-rccg/hw2d) — both solvers agree to within 1–2%, confirming the ~15% offset from Camargo 1995 is a systematic numerical-protocol difference, not a coding error.

Full turbulent field recovered from spectral observations via gradient-based optimization — 2×3 grid showing true, reconstructed, and error fields for potential and density.
Full turbulent field recovered from spectral observations via gradient-based optimization. Potential φ relative error 0.024%, density n relative error 0.009%.

Activity

Lecture session at the 14th ITER International School in Aix-en-Provence, with a slide on burning-plasma core physics projected on screen.
Lecture session — 14th ITER International School, Aix-en-Provence, July 2025.

14th ITER International School — 2025

Tulp Fusion served on the organizing committee for the 14th ITER International School on Integrated Modelling, held in Aix-en-Provence — bringing together over 200 students and lecturers from 39 countries. ITER Newsline →

People

Scientific Leadership

Prof. Sadruddin Benkadda
Scientific Director

CNRS Research Director at Aix-Marseille Universite. Founder of the ITER International School. Co-author of ITER Physics. Published extensively on MHD turbulence, anomalous transport, and nonlinear dynamics in magnetized plasmas.

Dr. Xavier Litaudon
Scientific Advisor

CEA Research Director. EUROfusion project leader for Preparation of ITER Operation. Led the JET Task Force and the final 2023 deuterium-tritium campaigns. Founded CICLOP, the IEA-IAEA coordination network on long-pulse fusion operation. Pioneer of the CRONOS integrated tokamak modelling suite. 250+ publications.

Advisory Board

Christopher Springham
Strategic Advisor

Over 30 years in energy sector communications. Former VP at GE Renewable Energy. Director of Media Relations at Rolls-Royce. White House Media Staff, Clinton administration.

Foundation

Richard Kraaijenhagen
Director

Founded Owlin. Engineering lead for optical diagnostics at Nostics. Develops differentiable PDE solvers and data assimilation methods for plasma turbulence. Validated the foundation's Hasegawa-Wakatani solver against Biskamp and Camargo benchmarks.

Linsey Groot
Board Member

Philosopher of science with a career spanning enterprise risk management, high-impact philanthropy in biosecurity, and nuclear risk. Leads governance and policy for the Foundation. Currently advising in the medical industry on the deployment of emerging technologies.

Contact

Collaboration and Research Inquiries

Location

Amsterdam, the Netherlands