DeepScience
All digests
ResearchersENNuclear Fusiondaily

[Nuclear Fusion] Daily digest — 289 papers, 0 strong connections (2026-04-21)

DeepScience — Nuclear Fusion
DeepScience
Nuclear Fusion · Daily Digest
April 21, 2026
289
Papers
9/9
Roadblocks Active
4
Connections
⚡ Signal of the Day
• The clearest engineering advances today are in HTS magnet reliability and runaway electron disruption control: both fields received papers with specific, actionable mechanistic findings.
• REBCO coil contact resistivity can be stabilized to prescribed values across 30,000 mechanical stress cycles using a 2–3 µm PbSn solder coating or conductive filler — directly applicable to SPARC-era no-insulation coil designs that must survive repeated magnet ramps. Separately, the long-debated mechanism behind benign runaway electron termination via gas injection was clarified: it works by raising bulk resistivity, not by reducing free electron density, which refocuses disruption mitigation engineering on achieving the right resistivity window rather than maximizing particle recombination.
• Watch for experimental validation of the two large-speedup surrogate models (PINN for NTV torque, pLaSDI for NLTE atomic kinetics): both report 10,000–100,000× computational gains and are structured to plug into real-time control loops — the next milestone is demonstrating them on live tokamak discharges rather than post-processed data.
📄 Top 10 Papers
Control of turn-to-turn contact resistivity in resistively insulated REBCO coils
High-temperature superconducting (REBCO) magnets for fusion reactors use tape-wound coils where the electrical resistance between tape layers governs how safely the magnet survives faults. This paper shows that mechanical pressure cycling — unavoidable during magnet operation — can drop contact resistance by up to 1000× unless the tape surfaces are treated: a thin tin-lead solder coating (2–3 µm) or conductive filler stabilizes resistance across 30,000 cycles at liquid-helium temperature. For compact high-field fusion devices like SPARC, this provides concrete process parameters to prevent the runaway resistive heating that has limited no-insulation coil designs.
██████████ 0.9 hts-magnets Preprint
Mechanism Behind the Recombination Requirement for Benign Termination of Relativistic Electron Beams
When a tokamak plasma disrupts, runaway electrons can form a relativistic beam capable of melting reactor components; standard mitigation injects neutral gas, but why a specific quantity is needed was unclear. This paper uses kinetic modeling and nonlinear MHD simulations (JOREK code) to show the gas works by raising bulk plasma resistivity — not by removing free electrons — and that this resistivity increase preferentially excites edge tearing modes that deconfine the beam before it hits the wall. The result directly prescribes what to optimize in disruption mitigation systems: achieving the right resistivity window, not maximizing recombination.
██████████ 0.9 plasma-disruption Preprint
A Data-Free, Physics-Informed Surrogate Solver for Drift Kinetic Equation: Enabling Fast Neoclassical Toroidal Viscosity Torque Modeling in Tokamaks
Plasma rotation in a tokamak suppresses turbulence and instabilities, but calculating the neoclassical toroidal viscosity (NTV) torque that damps that rotation requires solving a complex kinetic equation — too slow for real-time use. This paper shows that a physics-informed neural network trained on physical constraints alone (no labeled data) can solve this equation with accuracy matching a validated finite-difference solver, running more than 10,000× faster on conditions from China's EAST tokamak. Embedding this surrogate in a control loop would make rotation-profile feedback physically rigorous and computationally feasible for the first time.
██████████ 0.9 q-engineering Preprint
On nonlinear saturation of toroidal Alfvén eigenmode due to thermal plasma nonlinearities
Toroidal Alfvén Eigenmodes (TAEs) are plasma oscillations that scatter the alpha particles which sustain fusion burn; knowing what limits their amplitude is critical for predicting alpha particle losses. Using gyrokinetic particle-in-cell simulations, this paper identifies a threshold (linear drive γ_L/ω > 0.47%) above which the thermal plasma itself — not just the fast-particle drive — governs saturation, and shows the saturated amplitude is 'stiff' with weak dependence on drive strength. This stiffness means TAE-induced alpha losses in ITER-scale plasmas may be more predictable and bounded than previously assumed.
██████████ 0.8 long-confinement Preprint
Nonlinear Energy Transfer Analysis in Developing Plasma Turbulence
Turbulent transport in a fusion plasma involves complex nonlinear energy exchanges between oscillation modes that standard spectral diagnostics cannot capture. This paper applies two bispectral analysis methods (Ritz and Kim) to a laboratory plasma device and demonstrates energy flowing from Rayleigh-Taylor modes into low-frequency drift-wave modes via three-wave coupling — a process invisible to linear analysis. Crucially, it systematically maps the statistical conditions under which each method is valid, providing a practical guide for interpreting turbulence measurements in fusion experiments where data quality varies across radial positions.
██████████ 0.8 turbulence-modeling Preprint
Physics-Informed Latent Space Dynamics Identification for Time-Dependent NLTE Atomic Kinetics
Modeling how plasma radiates energy requires tracking atomic charge states across thousands of rate equations at every time step — a calculation so expensive it bottlenecks integrated reactor simulations. This paper presents pLaSDI, a physics-informed reduced-order model that compresses these calculations into a low-dimensional latent space, achieving charge-state errors below 2% while running 50,000–100,000× faster than conventional NLTE solvers. For fusion, this speedup is sufficient to include accurate radiation physics in real-time plasma control and rapid design-optimization loops for the divertor and plasma edge.
██████████ 0.8 divertor-thermal Preprint
Laboratory evidence of electron pressure anisotropy driving plasmoid mediated magnetic reconnection
Magnetic reconnection drives impulsive heat deposition on divertor surfaces during edge-localized modes (ELMs) in fusion plasmas; understanding what triggers and sustains reconnection in the edge current sheet is therefore practically important. This paper combines laser-driven plasma experiments at LULI and 3D hybrid particle-in-cell simulations to show that electron pressure anisotropy — different pressures parallel and perpendicular to the magnetic field — is the primary driver of reconnection and plasmoid formation, even in the absence of classical resistivity. Incorporating this mechanism into edge stability codes could improve ELM heat-pulse predictions and guide the design of resonant magnetic perturbation coils that suppress plasmoid formation.
██████████ 0.7 elm-control Preprint
Autoregressive prediction of 2D MHD dynamics inferred from deep learning modeling
Running full magnetohydrodynamic simulations of plasma instabilities is computationally prohibitive for real-time use in fusion reactors. This paper trains two deep learning architectures — a Koopman-based Transformer and a ConvLSTM-UNet — to autoregressively predict the evolution of a 2D MHD Kelvin-Helmholtz instability across varying magnetic field strengths, reproducing both the growth phase and nonlinear saturation while preserving global conservation laws. Accurate, fast MHD surrogates are a prerequisite for real-time disruption prediction systems and rapid exploration of reactor operating scenarios.
██████████ 0.7 plasma-disruption Preprint
A tensor invariant approach to energy flux in magnetohydrodynamic turbulence
In magnetized plasmas, energy cascades through turbulence via multiple competing mechanisms — inertial flow, magnetic Maxwell stresses, advection, and dynamo action — but directly measuring these fluxes from experiment is difficult. This paper derives exact relationships between tensor invariants of the velocity and magnetic field gradients (quantities computable from simulation or diagnostic data) and the energy flux in MHD turbulence, providing rigorous upper bounds and proxy relationships for each mechanism. The framework could be applied to diagnose which turbulent transport channels dominate in tokamak edge and core measurements without needing full-field decompositions.
██████████ 0.6 turbulence-modeling Preprint
FlowRefiner: Flow Matching-Based Iterative Refinement for 3D Turbulent Flow Simulation
Predicting turbulent flow step-by-step into the future using neural networks tends to accumulate errors that destabilize the simulation; standard fixes using stochastic diffusion models struggle in the small-noise regime. This paper replaces stochastic refinement with a deterministic flow-matching correction and introduces a decoupled noise schedule, achieving state-of-the-art rollout accuracy on two 3D turbulence benchmarks (isotropic turbulence and Taylor-Green vortex). For fusion, the stability improvements are relevant to developing turbulent transport surrogates capable of the long time-horizon predictions needed for integrated reactor simulations.
██████████ 0.6 turbulence-modeling Preprint
🔬 Roadblock Activity
Roadblock Papers Status Signal
Plasma Turbulence Modeling 35 Active High paper volume today but dominated by theoretical frameworks and astrophysical MHD; the most fusion-relevant contribution is experimental bispectral evidence of nonlinear mode coupling in a laboratory plasma, with several new analytical and ML surrogate tools providing longer-term utility.
Plasma-Wall Interaction 10 Active Weak day for direct plasma-wall advances; most contributions are indirect, arriving via MHD turbulence and dense plasma kinetics work rather than material erosion or sputtering experiments.
Disruption Prediction and Mitigation 6 Open A strong mechanistic result on runaway electron beam termination clarifies what gas injection actually optimizes, directly informing ITER and future-device disruption mitigation system design.
Engineering Gain (Q > 1) 4 Open A physics-informed neural network surrogate for neoclassical toroidal viscosity torque demonstrates >10,000× speedup, enabling rotation-profile physics to enter real-time plasma control for the first time.
High-Temperature Superconducting Magnets 4 Open The strongest engineering result of the day: specific coatings and fillers stabilize REBCO turn-to-turn contact resistance across 30,000 mechanical cycles at 4.2 K, with coil-level experimental validation supporting direct application in fusion magnet design.
ELM Control and Suppression 3 Open Indirect progress via the electron pressure anisotropy reconnection paper, which improves the physical picture of how ELM current sheets fragment and deposit heat, but no direct ELM suppression experiment today.
Divertor Thermal Management 2 Low Thin day; the pLaSDI NLTE atomic kinetics surrogate provides a computational pathway to include radiation physics in divertor thermal modeling at speeds compatible with design optimization.
Long-Pulse Plasma Confinement 2 Low The TAE saturation stiffness result is the notable contribution, suggesting alpha particle losses from Alfvénic modes may be more predictable in burning plasmas than previously assumed.
First Wall Materials 2 Low No direct first-wall materials result today; the two papers touching this roadblock are general plasma and transport theory with only marginal relevance.
View Full Analysis
DeepScience — Cross-domain scientific intelligence
Sources: arXiv · OpenAlex · Unpaywall
deepsci.io