285

Papers

9/9

Roadblocks Active

4

Connections

A neural network trained on Doppler backscattering spectrograms — which measure density fluctuations at the plasma edge — can predict the first ELM crash after an L-to-H mode transition with 100 ms advance warning. ELMs are explosive energy bursts that can erode reactor wall materials; 100 ms is long enough for mitigation hardware like resonant magnetic perturbation coils or pellet injectors to respond. The approach uses a survival-analysis architecture (DeepHit) that outputs a probability distribution over time windows rather than a single number, which is better suited to real-time risk-based control decisions.

██████████ 0.9 elm-control Preprint

ReadSaveConnections

Using a three-barrel shattered pellet injection (SPI) system on ASDEX Upgrade, this study maps the reproducible sequence of events that unfold during a mitigated disruption: first light, main fragment arrival, plasma movement, MARFE formation, thermal quench, current quench, and vertical displacement. Crucially, it shows that the shape of the plasma current trace during the current quench — convex for poor mitigation, concave for good mitigation — provides a real-time metric of mitigation quality. Increasing assimilated neon fraction shifts behavior along this spectrum, giving operators a tunable handle on disruption severity.

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

ReadSaveConnections

Injecting a small amount of tungsten (roughly 3 parts in 10,000) into DIII-D tokamak plasmas cooled electrons relative to ions, which stabilized a class of plasma turbulence called trapped-electron modes. This stabilization reduced how quickly heat and momentum escaped the plasma, causing the ion temperature to peak and toroidal rotation to roughly double — both desirable outcomes for confinement. The result suggests that a controlled level of high-Z impurity radiation, rather than being purely harmful, can shift the turbulence regime in a favorable direction, with direct implications for high-radiation scenarios planned for WEST and future reactors.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

The open-source HISP framework couples plasma edge simulation outputs directly to a 1D hydrogen transport solver (FESTIM) to predict where and how much tritium accumulates inside ITER's first wall and divertor components. After 10 days of deuterium-tritium pulses, modeled tritium inventory reaches roughly 35 grams — nearly 80% trapped in co-deposited boron layers in the divertor. Baking the components at elevated temperature removed up to 88% of tritium from tungsten surfaces, pointing to baking as the most effective routine recovery method and informing how ITER will manage its tritium safety limits.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

A three-dimensional convolutional neural network trained on quantum-mechanical energy landscapes around interstitial sites can predict how easily a hydrogen atom moves through tungsten — a key quantity for understanding tritium trapping in reactor walls — with an accuracy of 0.124 eV and in just 2.7 milliseconds, versus hours for conventional calculations. The 23,000-fold speedup makes it feasible to embed this model inside larger-scale simulations that track hydrogen behavior across an entire plasma-facing component over reactor timescales. This matters because tritium retention in tungsten walls is both a fuel-cycle efficiency issue and a regulatory safety limit.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

Zonal flows — large-scale rotating structures that suppress turbulent transport in tokamaks — are continuously disrupted by bursts of energy flowing back from zonal to turbulent scales, and this back-transfer sets a fundamental limit on how steady and coherent those stabilizing flows can be. Gyrokinetic simulations using the GENE code show that plasmas with negative triangularity (a shape where the plasma cross-section flares outward) experience significantly less back-transfer, explaining why negative triangularity configurations consistently show lower turbulent transport. This mechanistic link between plasma shape and zonal flow persistence gives designers a concrete physical lever to optimize confinement.

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

ReadSaveConnections

gyaradax is a new GPU-accelerated gyrokinetic turbulence code written in JAX — a modern Python framework — that reproduces results from the established GKW code in about 3,000 lines instead of 30,000 lines of Fortran, with substantial speed improvements from native GPU parallelism. The key practical advance is that JAX's automatic differentiation capability allows gradients of simulation outputs to flow back through the physics solver, enabling machine-learning models to be trained end-to-end against gyrokinetic data. Because it is publicly available on GitHub, it substantially lowers the barrier for coupling turbulence physics to data-driven optimization and control methods.

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

ReadSaveConnections

Ion temperature in some stellarators stops rising even as heating power increases — a phenomenon called temperature clamping that limits confinement quality. This paper shows mathematically that the turbulent modes responsible (ion-temperature-gradient modes) undergo Anderson localization, a wave-physics effect borrowed from condensed-matter physics, because the magnetic curvature in aperiodic stellarator geometry acts like a disordered lattice. The mapping to the well-studied Aubry-André-Harper equation provides a predictive threshold for when localization occurs, offering a new design criterion for stellarator optimization using real W7-X magnetic geometry data.

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

ReadSaveConnections

This paper derives an economic gain factor for fusion power plants — analogous to the well-known plasma gain factor Q — built from ten normalized design parameters that are independent of the specific fusion technology or absolute plant scale. By normalizing all costs and revenues to the energy-capture surface area rather than raw power output, the framework reveals which engineering choices dominate economic viability across different reactor concepts. This kind of structured economic metric is useful for comparing private fusion ventures and for identifying which physics or engineering improvements yield the largest reduction in cost of electricity.

██████████ 0.8 q-engineering Preprint

ReadSaveConnections

Proton-boron (p-11B) fusion produces no neutrons and is therefore appealing for a reactor that avoids neutron-induced material damage and radioactive waste, but it requires much higher plasma temperatures and densities than deuterium-tritium. This paper develops a multi-fluid equilibrium model for spherical tokamak p-11B plasmas that separately tracks thermal ions, suprathermal ions at ~0.5 MeV, and relativistic electrons in the MeV range — finding that these energetic populations meaningfully enhance the fusion reaction rate by accessing favorable peaks in the p-11B cross-section. The work provides the first force-balance framework for this plasma regime, which is being targeted by the EXL-50 device, though confidence in the results is limited by the model's reliance on unvalidated assumptions about suprathermal particle distributions.

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

ReadSaveConnections