282

Papers

10/10

Roadblocks Active

3

Connections

This conceptual design study anchors a compact HTS tokamak entirely to materials limits rather than grid economics: a 35 displacement-per-atom lifetime cap on the vanadium-alloy vacuum vessel forces fusion power density down to 0.7 MW/m², with a self-consistent 9.29 T, 9.67 MA operating point derived through integrated multi-code modeling. The design targets marine and off-grid markets where smaller, lower-power plants are viable. It matters because it quantifies exactly how materials constraints — not plasma physics — set the ceiling for compact fusion applications, and it gives the field a concrete design envelope against which first-wall and neutronics R&D can be prioritized.

██████████ 0.9 first-wall-materials Preprint

ReadSaveConnections

iGENE re-implements the nonlinear electromagnetic gyrokinetic code GENE in TensorFlow, enabling automatic differentiation through the entire turbulence simulation — so researchers can compute how any output (heat flux, growth rate) changes with any input (temperature gradient, magnetic shear) via backpropagation rather than expensive finite-difference sweeps. Linear benchmarks against the Fortran GENE code across ITG, ETG, and KBM modes confirm correctness, and nonlinear heat flux comparisons use five-window steady-state averaging for uncertainty control. The practical payoff is that transport model calibration against experimental profiles, previously requiring days of manual parameter scans, becomes a gradient descent problem solvable in minutes.

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

ReadSaveConnections

MAESTRO couples the PORTALS surrogate-based optimizer to external physics solvers — MHD equilibrium, pedestal stability, divertor detachment, and heating deposition — to predict steady-state plasma profiles without time-stepping through the full PDE system. A key technical advance is handling discontinuities in turbulent transport fluxes (caused by sudden onset of instabilities in reduced models like TGLF), which historically caused optimization loops to fail. This matters for fusion power plant design because it makes multi-physics profile prediction fast enough to use in iterative design loops, replacing computations that previously took hours with surrogate inference.

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

ReadSaveConnections

This paper trains reinforcement learning controllers for plasma rotation profiles using only historical DIII-D shot data — no physics simulator required — and then deploys them on the real machine's control system. Rotation control matters for fusion because plasma spin stabilizes certain MHD instabilities that degrade confinement, and achieving a target rotation profile requires coordinating multiple neutral beam injectors and microwave heating systems simultaneously. The demonstration that data-driven policies can work on live tokamak plasmas, despite the policy never having seen a physics model during training, is a practical step toward automated machine operation.

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

ReadSaveConnections

Laboratory experiments on a linear magnetized plasma device directly observe how turbulence transitions between organized zonal flows (which suppress cross-field transport) and radially elongated streamers (which enhance it) as neutral gas pressure is varied across three controlled operating points. At low collisionality, coherent zonal flows at 600–700 Hz are nonlinearly driven by drift-wave fluctuations; at intermediate collisionality, streamers emerge via coupling through a mediator mode while zonal flow drive simultaneously weakens. This controlled ground truth is valuable for testing nonlinear turbulence theories that predict tokamak confinement degradation near the L-H transition.

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

ReadSaveConnections

This study builds a forward model of gamma-ray emission expected from SPARC's primary reference discharge (140 MW, Q≈11), using TRANSP plasma profiles and CQL3D RF heating calculations, to guide the design of diagnostic systems before the machine operates. Three gamma-ray-producing nuclear reactions are modeled — T(D,γ)He-5, B-10(He-4,pγ)C-13, and D(He-3,γ)Li-5 — which encode information about DT fusion power, fast-ion velocity distributions, and plasma heating efficiency that cannot be obtained from neutron measurements alone. Getting the diagnostic design right pre-operationally is critical because SPARC will be among the first DT-burning tokamaks to reach Q>1, and incorrect detector placement or shielding would be uncorrectable after construction.

██████████ 0.8 q-engineering Preprint

ReadSaveConnections

This computational design proposes a single experimental device with 288 small dipole-field coils that can access more than 1.66 million optimized magnetic configurations — quasi-axisymmetric, quasi-helically symmetric, quasi-isodynamic, and tokamak-like states — by reprogramming coil currents rather than rebuilding hardware. Monte Carlo tracing of 10,000 fusion-born alpha particles is used to filter configurations by confinement quality. If experimentally realized, such a device could compress decades of magnetic configuration space exploration into a single machine, but the confidence is rated low because no codes or design files are shared, a patent is pending, and all results are computational.

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

ReadSaveConnections

This theoretical paper proves that a plasma controller trained only on macroscopic measurements (e.g., density profiles at sparse sensor locations) can stabilize kinetic plasma instabilities, with the performance gap relative to a fully-observed expert controller bounded by an entropy measure of the initial distribution's complexity. The mechanism is behavior cloning: an expert policy derived from the full kinetic state is distilled into a lightweight controller that runs on real-time diagnostic signals. For disruption avoidance in tokamaks, where full kinetic state measurement is impossible, this provides a rigorous framework for designing controllers that are provably stable despite limited sensor coverage.

██████████ 0.7 plasma-disruption Preprint

ReadSaveConnections

Using GENRAY ray-tracing simulations of ITER's full H-mode scenario, this study finds that viewing electron cyclotron emission at an oblique angle (9.25°) introduces Doppler broadening that can mask the spectral signatures of non-Maxwellian electron populations — which are present whenever RF heating creates fast electrons. The practical implication is that ITER's oblique ECE view will be unreliable for detecting non-thermal distortions, but higher harmonics in either polarization remain clean for standard electron temperature profile reconstruction. This pre-operational simulation directly informs how ITER's diagnostic team should interpret conflicting ECE and Thomson scattering temperature measurements observed on existing tokamaks.

██████████ 0.7 q-engineering Preprint

ReadSaveConnections

Particle tracking simulations show that magnetic turbulence generated by the Weibel instability (driven by electron temperature anisotropy) produces strongly anisotropic electron diffusion whose magnitude depends sensitively on electron temperature, external field strength, and turbulence level. In collisionless conditions, the anomalous electrical conductivity can exceed collisional resistivity by a significant factor. This is relevant to tokamak edge and scrape-off-layer physics, where temperature-anisotropic distributions develop near plasma-facing components and Weibel-type turbulence could enhance cross-field transport beyond classical predictions.

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

ReadSaveConnections