286

Papers

10/10

Roadblocks Active

6

Connections

A neural network trained on edge turbulence measurements from DIII-D can predict the first ELM crash 100 ms before it happens — enough lead time for automated mitigation systems to act. The model uses 50 ms of Doppler backscattering spectrogram data and the DeepHit survival-analysis architecture to output ELM crash probability in future time windows. ELMs are repetitive plasma instabilities that concentrate intense heat loads on the divertor wall; reliable early warning is a prerequisite for protecting ITER and future reactors.

██████████ 1.0 elm-control Preprint

ReadSaveConnections

This ASDEX Upgrade experiment systematically maps how varying neon fraction and shattering geometry in shattered pellet injection (SPI) affects the full disruption sequence, from first pellet light through thermal and current quench to vertical displacement. Higher neon assimilation shifts the plasma current decay from a convex shape — associated with dangerous electromagnetic wall loads — to a concave, radiation-dominated shape indicative of good mitigation. Characterizing this parameter space directly informs the design of ITER's disruption mitigation system, which will rely on SPI to protect the first wall.

██████████ 0.9 plasma-disruption Preprint

ReadSaveConnections

A three-dimensional convolutional neural network predicts hydrogen migration barriers in tungsten — the energy cost for a hydrogen atom to hop between trapping sites — with a mean absolute error of 0.124 eV and in just 2.7 milliseconds per barrier, a roughly 23,000-fold speedup over conventional nudged-elastic-band calculations. The model takes a volumetric map of the local atomic potential as input and was trained on standard interatomic potential calculations. This speed-up makes large-scale simulation of tritium retention and transport through radiation-damaged tungsten walls computationally tractable for the first time.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

Controlled tungsten injection into a DIII-D plasma lowered the electron-to-ion temperature ratio, which stabilized trapped-electron-mode turbulence and produced a factor-of-two increase in toroidal plasma rotation through enhanced E×B shear. The suppressed turbulence reduced ion heat diffusivity and improved energy confinement in the outer plasma region. This is significant because ITER will operate with tungsten plasma-facing components that unavoidably release tungsten impurities — this experiment suggests that, under the right conditions, that radiation can be beneficial rather than purely harmful to plasma performance.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

The open-source HISP framework couples ITER plasma edge codes to a 1D hydrogen transport solver and predicts approximately 35 g of tritium accumulating in ITER's first wall and divertor after 10 days of DT operation, with nearly 80% stored in co-deposited boron layers in the divertor. Baking the first wall removes 88% of retained tritium from tungsten components but only about 30% from boron co-deposits, pointing to boron layers as the dominant retention challenge. Controlling tritium inventory is a safety and fuel-cycle requirement for any reactor, and this tool provides the first integrated, open-source framework for quantifying it across full operational scenarios.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

This theoretical paper maps the ion-temperature-gradient (ITG) turbulence equation in a non-symmetric stellarator onto the Aubry-André-Harper model from condensed matter physics, which is known to exhibit a sharp localization transition. The result analytically explains why ion temperature in stellarators saturates at a fixed fraction of electron temperature regardless of additional heating — a longstanding experimental puzzle called ion temperature clamping. If the localization threshold can be engineered through magnetic geometry choices, stellarator designs could be tuned to intrinsically suppress turbulent ion heat loss.

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

ReadSaveConnections

gyaradax is a full reimplementation of the established gyrokinetic turbulence code GKW in JAX/CUDA, enabling native GPU acceleration and, crucially, automatic differentiation through the simulation. Validation against the standard Cyclone Base Case and Rosenbluth-Hinton benchmarks shows statistical agreement with GKW results. The public GitHub release and automatic differentiation capability open a direct path to gradient-based optimization of plasma profiles and magnetic geometry to minimize turbulent heat losses — something not previously available in production gyrokinetic codes.

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

ReadSaveConnections

This analytical paper derives an economic gain factor (Q_econ) for fusion power plants, analogous to the plasma physics Q-factor, expressed through ten normalized design parameters that are independent of absolute plant size or specific fusion technology. The framework separates plant operation phases from component-replacement phases and normalizes all costs to the energy-capturing surface area, making it applicable to tokamaks, stellarators, and inertial confinement concepts alike. While purely theoretical with no empirical validation, it provides a technology-neutral common yardstick for comparing competing fusion approaches on economic grounds.

██████████ 0.8 q-engineering Preprint

ReadSaveConnections

Standard magnetohydrodynamic equilibrium theory predicts unphysical singular currents on resonant magnetic surfaces in three-dimensional (non-symmetric) configurations like stellarators, a mathematical inconsistency that has challenged stellarator design for decades. This paper replaces that framework with a statistical equilibrium model formed by averaging over fast, small-amplitude magnetic fluctuations, producing smooth solutions verified both analytically in a slab geometry and numerically in the nonlinear regime. Because all stellarator performance predictions — from confinement time to instability thresholds — rest on equilibrium calculations, resolving this inconsistency has direct consequences for how reliably stellarator designs can be optimized.

██████████ 0.7 long-confinement Preprint

ReadSaveConnections

Plasma GraphRAG builds a physics knowledge graph from fusion literature — with nodes for parameters, device metadata, and sources, and edges encoding physical couplings — and uses it to help users select correct input parameters for gyrokinetic turbulence simulations. Compared to standard retrieval-augmented generation, it reduces hallucination rates by up to 25% and improves overall answer quality by over 10% on controlled benchmarks. The practical value is lowering the barrier for setting up complex turbulence simulations correctly, which currently requires deep specialist knowledge and is a bottleneck for broadening the user base of these codes.

██████████ 0.6 turbulence-modeling Preprint

ReadSaveConnections