282

Papers

11/11

Roadblocks Active

5

Connections

A zero-shot disruption prediction framework uses MHD simulations (NIMROD) to generate synthetic diagnostic signals that mimic the hardware layout of a new target tokamak, then transfers a model trained on EAST data to J-TEXT without any prior J-TEXT experimental disruption data, improving early warning rates from 50% to 57%. This matters because new fusion devices cannot wait years to accumulate enough disruption events to train predictive models — reliable disruption protection is needed from the first plasma. The domain adaptation pipeline combining Fourier-based frequency alignment and deep covariance matching addresses the sim-to-real gap that has limited previous transfer approaches.

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

ReadSaveConnections

Self-consistent simulations of the ARC compact fusion plant using coupled transport, turbulence, and atomic physics codes show that fusion power in the 750–1000 MW range is achievable in H-mode while maintaining divertor temperatures below 2 eV through argon impurity seeding. This matters because it is the first demonstration that high fusion power and manageable divertor heat loads can coexist in a compact device without sacrificing one for the other. Fusion power is most sensitive to separatrix density, giving reactor operators a practical real-time control knob, while argon seeding provides the most robust path to simultaneous H-mode access and divertor detachment.

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

ReadSaveConnections

Physics-informed neural networks trained on the adjoint of the relativistic Fokker-Planck equation predict runaway electron current, mean energy, and full energy distribution across broad plasma conditions with orders-of-magnitude speedup over traditional solvers. This matters because runaway electron beams generated during disruptions carry enough energy to melt plasma-facing components, and fast prediction is a prerequisite for real-time mitigation. The adjoint formulation is the key innovation: once trained, the network can produce predictions for any initial electron distribution without re-solving the full forward problem, making it suitable for online control applications.

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

ReadSaveConnections

A computational design study shows that a single array of 32 planar HTS dipole coils arranged axisymmetrically around the HBT-EP experiment footprint can produce a wide range of plasma configurations — from quasi-axisymmetric stellarators to H-mode-shaped tokamaks with elongation 1.7 — using realistic HTS engineering limits. This matters because it proposes a physically compact way to experimentally test the confinement advantages of both stellarators and tokamaks in one machine without rebuilding the coil structure. The use of planar rather than non-planar coils significantly reduces manufacturing complexity and cost for HTS-based experiment construction.

██████████ 0.8 hts-magnets Preprint

ReadSaveConnections

The ECHO algorithm combines a neural network surrogate of a microwave ray-tracing code with a genetic optimizer to control where electron cyclotron heating is deposited in the plasma in real time on DIII-D, achieving the target radial deposition profile even when individual gyrotrons fail mid-shot. This matters because precise heating at specific radial locations is the primary tool for stabilizing neoclassical tearing modes — dangerous current-driven instabilities that degrade confinement and can trigger disruptions. Demonstrated robustness to hardware failures and large plasma parameter swings makes this approach viable for use in future devices where such failures cannot be anticipated.

██████████ 0.8 mhd-instability Preprint

ReadSaveConnections

A Ka-band Doppler backscattering diagnostic designed for the Pegasus-III spherical tokamak can measure ion-scale density fluctuations (wavenumbers 1–8 cm⁻¹) from the outer core out to just beyond the plasma boundary, a radial range not easily covered by other diagnostics. This matters because turbulence at these scales is a primary driver of anomalous heat and particle transport that degrades confinement — quantifying it in spherical tokamak geometry fills an important measurement gap relevant to NSTX-U and ST40. The design also enables inference of magnetic field pitch angle from mode-conversion windows, providing equilibrium information from a single diagnostic system.

██████████ 0.7 turbulence-modeling Preprint

ReadSaveConnections

A Kolmogorov-Arnold Network surrogate reduces the time to compute neoclassical toroidal viscosity torque — a quantity governing how external magnetic perturbations affect plasma rotation — from 36 seconds to under 4 seconds while outperforming standard neural network baselines in prediction accuracy on EAST tokamak parameters. This matters because neoclassical toroidal viscosity is central to ELM suppression strategies that use resonant magnetic perturbations, and faster computation enables these effects to be included in transport codes and control workflows that currently cannot afford the runtime. Extrapolation tests beyond the EAST training regime provide initial evidence of generalization to other devices.

██████████ 0.7 turbulence-modeling Preprint

ReadSaveConnections

A physics-informed neural operator surrogate of the LIUQE equilibrium reconstruction code achieves inference below 100 microseconds — enabling 10 kHz plasma shape control on TCV — trained on approximately 10,000 discharges and validated in closed-loop on the actual TCV plasma control system. This matters because real-time knowledge of plasma shape determines strike-point location on the divertor and is prerequisite for controlling plasma-wall interactions, but existing equilibrium codes run too slowly for high-bandwidth feedback. Physics-informed regularization via automatic differentiation ensures the predicted flux maps respect Maxwell's equations, making the model more reliable across the wide range of shapes TCV can produce.

██████████ 0.7 plasma-wall Preprint

ReadSaveConnections

An analytical theory paper derives how the toroidal magnetic flux enclosed by magnetic surfaces relates to loop voltage and the slippage between poloidal and toroidal flux, with separatrix effects modeled using a Cartesian periodicity simplification that yields explicit bounds on flux-surface behavior. This matters because the separatrix defines the boundary between confined and unconfined plasma, and understanding how flux evolves there during current quenches is central to disruption physics and runaway electron generation. The framework provides cleaner theoretical grounding for interpreting loop voltage and flux measurements from existing tokamaks like JET and TFTR.

██████████ 0.6 plasma-disruption Preprint

ReadSaveConnections

A new empirical formula fitted to experimental data for heavy nuclei (mass numbers 146–238) predicts (n,3n) reaction cross sections at 14–15 MeV neutron energies — exactly the range produced by deuterium-tritium fusion. This matters for blanket design because these reactions contribute secondary neutrons that improve tritium breeding ratios and determine activation rates in structural materials like tungsten and steel. Agreement with the TALYS 1.95 nuclear code provides a cross-check for the neutronics calculations used in blanket optimization, filling gaps where direct 14 MeV measurements are scarce.

██████████ 0.6 first-wall-materials Peer-reviewed

Read