288

Papers

10/10

Roadblocks Active

4

Connections

Engineers developed a low-thermal-mass lithium evaporator that can deposit a fresh lithium film on plasma-facing surfaces in minutes rather than hours, and showed that loading lithium in vacuum (rather than in air) reduces impurity outgassing by more than 10-fold. Fresh lithium surfaces act as a getter pump, rapidly absorbing oxygen and hydrogen before they can reach the plasma core and cool it. For future reactors, rapid lithium refresh offers a practical route to continuous wall conditioning — reducing impurity influx and potentially controlling tritium co-deposition inventory — without lengthy machine down-time.

██████████ 0.9 plasma-wall Preprint

ReadSaveConnections

Three-dimensional resistive MHD simulations using the NIMROD code show that once a CFETR-scale plasma exceeds the no-wall beta limit, an n=1 resistive wall mode grows, stochastizes magnetic flux surfaces, and triggers a full thermal and current quench. The growth rate is highly sensitive to how the conducting wall responds but barely changes with plasma resistivity at reactor Lundquist numbers, meaning wall engineering choices — not plasma resistivity — dominate disruption risk in this regime. This gives reactor designers a quantitative target: the wall time constant and geometry must be chosen to keep RWM growth rates manageable, or active feedback coils must compensate within the simulated timescale.

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

ReadSaveConnections

Turbulent shear flows in an inertial confinement fusion hot spot can distort the ion velocity distribution away from a Maxwellian, creating a suprathermal tail that boosts fusion reaction rates beyond standard predictions. However, the study finds that using the more physically realistic Fokker-Planck collision operator roughly halves the predicted reactivity enhancement compared to the simpler BGK operator commonly used in prior estimates. This is a quantitative correction to optimistic projections: ICF designers relying on BGK-based estimates of turbulence-driven yield gain should downgrade those numbers, and experimental diagnostics should be designed to distinguish the two collision-physics regimes.

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

ReadSaveConnections

By treating all plasma diagnostics as nodes in a network and tracking how correlated they are during a shot, the authors show that a single graph metric — algebraic connectivity, the Fiedler eigenvalue of the diagnostic correlation matrix — predicts every disruption in the MAST dataset with zero misses and an average warning time equal to 49% of the remaining flat-top duration. A related metric, eigenvector centrality, also correlates significantly with the H98 confinement quality factor (r=0.405), meaning the same framework tracks both danger and performance in real time. The approach is model-free and could in principle be deployed on any tokamak diagnostic set without retraining.

██████████ 0.9 plasma-disruption Peer-reviewed

Read

A three-stage deep learning pipeline replaces the most expensive step in atomistic simulations — computing energy barriers for each possible hydrogen hop — allowing kinetic Monte Carlo simulations of hydrogen transport in polycrystalline tungsten to reach macroscopic timescales that were previously inaccessible. The method correctly reproduces preferential hydrogen accumulation at grain boundaries, which is the mechanism responsible for hydrogen embrittlement in tungsten first-wall tiles. For fusion materials qualification, this provides a computational tool to screen how radiation-induced grain boundary damage changes tritium retention and mechanical properties without requiring years of irradiation experiments.

██████████ 0.8 plasma-wall Preprint

ReadSaveConnections

This companion Zenodo record (version deposit of the same work as zenodo.20499620) reports the same correlation-graph state machine achieving perfect disruption recall on MAST with a 49% flat-top lead time and a statistically significant link between graph centrality and H98 confinement enhancement. Code and data are stated to be available via GitHub and the public FAIR-MAST dataset, which in principle allows the community to reproduce and extend the analysis to other machines. The low-overhead, physics-agnostic nature of the method makes it a plausible real-time monitoring layer on top of existing diagnostic infrastructure.

██████████ 0.8 plasma-disruption Peer-reviewed

Read

The EXTRA method solves a longstanding computational bottleneck in designing no-insulation high-temperature superconducting magnets: simulating the full 3D electromagnetic and thermal behavior of a 10,000-turn coil without the prohibitive cost of resolving every turn individually. By homogenising most of the winding but explicitly resolving defect-adjacent and boundary turns, the method achieves a 13× speedup over the reference model while accurately reproducing AC losses and the thermal runaway that destroys such magnets. HTS magnets at this scale are the leading candidate for the compact, high-field coils needed in tokamak and stellarator reactors, so reliable quench simulation tools are a prerequisite for their qualification.

██████████ 0.8 hts-magnets Preprint

ReadSaveConnections

Global electromagnetic gyrokinetic simulations identify a new instability — the weak-shear Alfvenic ion temperature gradient (WSAITG) mode — that appears specifically in reversed magnetic shear profiles used in advanced tokamak scenarios to suppress core turbulence. The mode is electromagnetic and has a substantially higher real frequency than conventional ion temperature gradient turbulence, suggesting it would drive different cross-field transport channels and potentially undermine the confinement improvement that reversed shear is intended to provide. Reversed shear profiles are a key strategy for sustaining high-performance, steady-state plasmas in future reactors, so characterizing new instabilities they harbor is essential for scenario design.

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

ReadSaveConnections

This paper introduces a class of neural network surrogates that automatically satisfy physical conservation laws — such as the divergence-free condition on magnetic fields — by encoding these constraints into the geometry of the network's message-passing structure rather than penalising violations in the loss function. Applied to 3D turbulent MHD and fusion equilibrium prediction benchmarks, the approach outperforms standard neural operators on structure-sensitive metrics including divergence error and long-rollout stability. For fusion, this matters because surrogate models used in real-time plasma control or design optimization must conserve fluxes to remain physically meaningful, a requirement that loss-penalty approaches routinely fail to guarantee.

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

ReadSaveConnections

Gyrotrons — the high-power microwave sources used for electron cyclotron heating in tokamaks — suffer from unexplained performance degradation traced to electron clouds in their magnetron injection guns (MIGs); this paper identifies the diocotron instability as the root cause, using a purpose-built trapped-electron experiment (T-REX) and 3D particle-in-cell simulations. The instability drives periodic collapse and reformation of the electron cloud with measurable rotating structures, explaining discrepancies between simulations and experimental gun behaviour that had previously limited gyrotron design fidelity. Better gyrotron modelling translates directly to more reliable high-power heating systems, which are a critical component of reaching and sustaining ignition conditions.

██████████ 0.6 q-engineering Preprint

ReadSaveConnections