277

Papers

10/10

Roadblocks Active

1

Connections

Injecting a controlled amount of tungsten into a DIII-D tokamak plasma cooled the electrons faster than the ions, shifting the temperature ratio in a way that shut down a major turbulence mode (trapped-electron-mode). This stabilization halved the ion heat flux escaping the plasma and doubled the plasma's toroidal rotation speed — both desirable outcomes. The result matters because future reactors like ITER will unavoidably have tungsten walls, and this suggests high-radiation operation may be self-reinforcing rather than purely damaging to confinement.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

A neural network trained on microwave reflectometry measurements of the plasma edge can predict the first violent energy burst (ELM) roughly 100 milliseconds before it occurs — enough time to trigger a mitigation response. The model processes a 50 ms window of spectral data and outputs a probability distribution over when the ELM will arrive, framing the problem as a survival-analysis task. Early ELM prediction is a prerequisite for protecting divertor surfaces in ITER, where unmitigated ELMs could erode the target plates within seconds of operation.

██████████ 0.9 elm-control Preprint

ReadSaveConnections

FIREFLY is a fast computational tool that estimates how much heat and how many particles will hit a fusion reactor's exhaust chamber (the divertor) for a given magnetic geometry, running orders of magnitude faster than full-physics codes by using simplified transport equations on a field-line mesh. It integrates the established EIRENE neutral-particle tracker to handle the complex gas physics near the wall. Rapid design iteration is currently a bottleneck for divertor engineering, particularly for complex configurations like stellarators and advanced tokamaks, and tools like this allow many candidate geometries to be screened before committing to expensive simulations.

██████████ 0.9 divertor-thermal Preprint

ReadSaveConnections

A new multiscale algorithm for gyrokinetic plasma simulations achieves a 30,000-fold speed-up by exploiting the vast separation in timescales between particle orbits and collisional equilibration in a high-field magnetic mirror device (WHAM, with 17 T peak field). The method alternates between tracking full particle dynamics and an orbit-averaged approximation, making previously intractable long-timescale calculations feasible. This matters because magnetic mirror concepts are seeing renewed commercial interest as compact fusion devices, and validated simulation tools are essential for understanding confinement and designing near-term experiments.

██████████ 0.8 hts-magnets Preprint

ReadSaveConnections

gyaradax is an open-source reimplementation of the established gyrokinetic turbulence code GKW in Python/JAX, enabling native GPU acceleration and automatic differentiation without sacrificing physics fidelity — benchmarks against GKW show statistical parity across multiple plasma configurations. GPU acceleration dramatically reduces the wall-clock time for turbulence simulations that underpin transport predictions in reactor design. The automatic differentiation capability opens the door to gradient-based optimization of plasma parameters, which is currently impractical with Fortran-based legacy codes.

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

ReadSaveConnections

This theoretical study explains a long-puzzling observation in stellarators — that ion temperature tends to clamp at a threshold value — by mapping the turbulence eigenvalue problem onto a well-known condensed-matter model (the Aubry-André-Harper equation) where disorder causes wave localization. Because stellarators have aperiodic magnetic geometry, the curvature experienced by turbulent eddies is quasi-random, naturally suppressing large-scale turbulent transport through the same physics that localizes electrons in disordered crystals. The insight suggests that stellarator aperiodicity is not just a geometric quirk but a confinement asset, and provides an analytic tool for optimizing turbulence suppression in new designs.

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

ReadSaveConnections

A 3D convolutional neural network trained on quantum-mechanical calculations can predict how easily hydrogen atoms migrate through a tungsten lattice (a key step in tritium retention and wall damage), achieving mean absolute error of 0.124 eV while running 23,000 times faster than the reference calculation method. The model takes a 3D map of the local potential-energy landscape around a hydrogen atom as input and directly outputs the energy barrier for jumping to the next site. This speed-up makes it practical to simulate hydrogen transport across realistic wall surfaces at scale, which is essential for predicting tritium inventory and helium-bubble formation in reactor first-wall materials.

██████████ 0.8 plasma-wall Preprint

ReadSaveConnections

This analytical study extends the theory of how a plasma's resonant layer responds to externally applied magnetic field perturbations — the same perturbations used to suppress ELMs — by incorporating ion parallel flow, which was previously neglected. The new 'nested boundary layer' framework predicts that ions actively shield the plasma against field penetration in parameter regimes relevant to future reactors, agreeing with prior numerical results from the GPEC/SLAYER code. Getting this shielding physics right matters because it determines whether a given coil configuration will actually penetrate into the plasma and suppress ELMs, or be screened out and do nothing.

██████████ 0.8 plasma-disruption Preprint

ReadSaveConnections

Using analytical mechanics and particle-in-cell simulations, this paper shows that runaway electrons — relativistic particles accelerated by the electric field during a plasma disruption, capable of punching through reactor walls — can be stopped by injecting a circularly polarized radio-frequency wave that acts as a 'firewall' at the Doppler-shifted cyclotron resonance. Once a particle reaches the resonance, the wave traps it and reverses the parallel acceleration, halting runaway growth. Runaway electrons are one of the most dangerous disruption consequences for a reactor, and an active RF-based mitigation scheme would complement existing strategies like massive gas injection.

██████████ 0.8 plasma-disruption Preprint

ReadSaveConnections

This computational study demonstrates that a stellarator — a twisty magnetic confinement device — can be built using only simple circular coils by tilting them at varying angles, replacing the complex three-dimensional coil shapes that make conventional stellarators expensive to manufacture. A parameter scan over 45 configurations identified designs with nested flux surfaces, low neoclassical transport, and good confinement of both thermal and fusion-born alpha particles (3.5 MeV). Reducing coil complexity is a major driver of stellarator cost and manufacturing risk, so configurations that retain good physics with simpler hardware are directly relevant to commercial viability.

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

ReadSaveConnections