293

Papers

10/10

Roadblocks Active

4

Connections

Researchers fired shattered pellets of frozen neon into ASDEX Upgrade tokamak plasmas and mapped out the precise timing and sequence of every disruption phase, from first pellet light through the final vertical displacement event. A key practical finding is that the shape of the plasma current decay curve — convex for poorly mitigated disruptions, concave for well-mitigated ones — serves as a real-time quality indicator for how much radiation-driven suppression was achieved. This gives future fusion reactors a measurable target for evaluating whether their disruption mitigation systems are working, and shows that fragment size and velocity are tunable knobs that shift the outcome.

██████████ 1.0 plasma-disruption Preprint

ReadSaveConnections

Injecting small amounts of tungsten into DIII-D tokamak plasmas cooled the electrons enough to suppress a class of turbulence (trapped electron modes), which unexpectedly improved plasma rotation by a factor of two and sharply reduced ion heat losses. The mechanism is a chain reaction: lower electron temperature reduces the electron-to-ion temperature ratio, which stabilizes the turbulence that was previously driving momentum and heat out of the plasma. This matters for future reactors because tungsten will inevitably enter the plasma from the walls, and this result suggests the contamination may partially self-regulate transport rather than purely degrading performance.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

A neural network trained on radar-like measurements of plasma edge turbulence can predict the first large ELM energy burst 100 milliseconds before it occurs — enough time for active mitigation systems to respond. The model uses a survival-analysis architecture originally developed for medical event prediction, adapted to output a probability distribution over when the ELM will arrive rather than a single yes/no answer. This is significant because the first post-transition ELM is often the largest and most damaging to the divertor, and current diagnostics do not reliably anticipate it.

██████████ 0.9 elm-control Preprint

ReadSaveConnections

HISP is an open-source computational tool that chains together plasma edge simulations with hydrogen transport calculations to estimate how much tritium accumulates in ITER's reactor walls and divertor over time. After ten days of deuterium-tritium pulses, the model predicts roughly 35 grams of tritium retained — nearly 80% locked in co-deposited boron layers in the divertor, which baking removes only 30% of, compared to 88% removal from tungsten surfaces. Knowing where tritium hides and how well different cleaning methods work is essential for fuel accounting, regulatory safety limits, and tritium breeding balance.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

Using high-fidelity gyrokinetic simulations, this study identifies that the fundamental limit on how long zonal flows can sustain their shearing effect is set by intermittent bursts of energy flowing back from these flows into the surrounding turbulence. Zonal flows act as a natural turbulence suppressor in fusion plasmas, and understanding what caps their coherence lifetime directly informs how much transport reduction they can realistically provide. Notably, negative triangularity plasma shapes — already of practical interest for their lower turbulence — show substantially less back-transfer, helping to explain their favorable confinement properties from a new angle.

██████████ 0.9 turbulence-modeling Preprint

ReadSaveConnections

gyaradax is a new open-source GPU-accelerated gyrokinetic simulation code written in JAX that reproduces results from the established GKW Fortran code while running substantially faster. Beyond speed, the JAX framework provides automatic differentiation, meaning the code can compute exact gradients of turbulent transport coefficients with respect to plasma parameters — enabling gradient-based optimization and sensitivity studies that were previously impractical. Validated against standard community benchmarks (Rosenbluth-Hinton zonal flow and Cyclone Base Case), it is immediately usable and lowers the barrier for groups without access to large HPC clusters.

██████████ 0.9 turbulence-modeling Preprint

ReadSaveConnections

A three-dimensional convolutional neural network trained on quantum-chemistry calculations can predict how easily hydrogen atoms migrate through tungsten — the planned armor material for fusion reactor walls — with a mean error of 0.124 eV and in just 2.7 milliseconds per prediction. Conventional calculations of the same quantity take hours, so this 23,000-fold speedup enables large-scale statistical surveys of hydrogen trapping behavior across different tungsten microstructures. Better models of hydrogen migration directly feed into predictions of tritium retention and permeation, which are safety-critical quantities for any burning plasma device.

██████████ 0.8 plasma-wall Preprint

ReadSaveConnections

This theoretical paper proposes that the puzzling observation of ion temperatures plateauing in stellarators — so-called ion temperature clamping — is caused by the same physics that makes electrons stop conducting in disordered materials (Anderson localization). The three-dimensional, non-repeating magnetic geometry of stellarators like W7-X creates a quasiperiodic potential in the equations governing plasma turbulence, and beyond a critical temperature gradient the unstable modes become spatially trapped rather than spreading, limiting how much drive can be fed into turbulence. This is a conceptual advance because it reframes a device-specific empirical puzzle using a well-developed framework from condensed matter physics, potentially enabling stellarator geometries to be optimized to exploit or tune this effect.

██████████ 0.8 turbulence-modeling Preprint

ReadSaveConnections

This paper develops a detailed equilibrium model for a spherical tokamak burning proton-boron-11 fuel, a reaction that produces no neutrons but requires far hotter plasmas than conventional deuterium-tritium fusion. The model treats the plasma as multiple interacting fluids — thermal ions, fast ions near 0.5 MeV, and relativistic electrons — and finds that the double-peaked shape of the p-11B fusion cross section can be exploited by maintaining a population of suprathermal particles to boost reaction rates. The work is theoretical and tied to ENN's EXL-50 program, with no experimental validation of the burning plasma regime yet, but it provides a framework for assessing whether compact spherical tokamaks could viably pursue aneutronic fusion.

██████████ 0.8 long-confinement Preprint

ReadSaveConnections

Plasma GraphRAG builds a knowledge graph from fusion physics literature and uses it to guide a language model in selecting input parameters for gyrokinetic turbulence simulations, outperforming standard retrieval-augmented generation by over 10% on quality metrics and reducing hallucination rates by up to 25%. Setting up gyrokinetic simulations correctly requires expert knowledge of how dozens of plasma parameters interact, and mistakes lead to unphysical results, so automated assistance could significantly reduce the expertise barrier for using these codes. The approach is still early-stage — evaluation relies on heuristic LLM-based scoring rather than hard numerical benchmarks — but demonstrates a path toward AI-assisted fusion simulation workflows.

██████████ 0.8 turbulence-modeling Preprint

ReadSaveConnections