274

Papers

10/10

Roadblocks Active

2

Connections

Using the ORB5 global gyrokinetic code, this study shows that consistently including the Shafranov shift — the outward displacement of flux surfaces caused by plasma pressure — in the MHD equilibrium reduces Toroidal Alfvén Eigenmode (TAE) growth rates by up to 90%, with the effect being stronger at longer wavelengths and lower toroidal mode numbers. The result matters because TAEs can kick fast ions (alpha particles from fusion reactions) out of the plasma before they transfer their energy to the fuel, reducing heating efficiency; prior simulations that ignored this shift were therefore predicting a more unstable situation than reality. This finding suggests burning plasma devices like ITER may have more margin against fast-ion losses than conservative models indicated, though nonlinear confirmation with zonal flows included is still needed.

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

ReadSaveConnections

Classical molecular dynamics simulations of the liquid lead-lithium alloy used as a tritium breeding blanket coolant reveal that helium — produced by neutron reactions in the blanket — has extremely low solubility and rapidly forms bubbles, generating local pressure inhomogeneities at the liquid-metal interface that vary systematically with temperature up to 1021 K. Understanding bubble nucleation and interfacial tension in PbLi is critical because uncontrolled helium bubble accumulation could impair tritium extraction, alter flow dynamics, and damage blanket structures. The simulated interfacial properties across the relevant temperature range provide quantitative inputs for blanket design codes that currently rely on uncertain empirical estimates.

██████████ 0.9 tritium-breeding Preprint

ReadSaveConnections

Experiments on the IMPED linear plasma device show that increasing the magnetic field from 600 to 1000 G suppresses large-scale zonal flows while amplifying mean flows, simultaneously shifting spectral power from low-frequency (0.1–1 kHz) to high-frequency (1–300 kHz) turbulent fluctuations. This matters because zonal flows are self-generated structures that ordinarily suppress cross-field transport, so their suppression by stronger magnetic fields — the opposite of the naively expected stabilizing effect — indicates a non-trivial competitive mechanism that turbulence models must capture. The result provides a controlled experimental benchmark for validating gyrofluid and gyrokinetic codes that predict confinement degradation at high field.

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

ReadSaveConnections

Machine learning models trained on synthetic Ion Cyclotron Emission (ICE) spectra generated by particle-in-cell simulations of the magnetoacoustic cyclotron instability can accurately recover the background magnetic field strength, plasma density, and fast-ion velocity pitch and concentration in near real-time (~100 ms). ICE is a passive emission produced by energetic ions (alpha particles, NBI ions) as they gyrate, meaning this method could give real-time information about fast-ion populations without additional hardware — currently a major gap in burning plasma diagnostics. A pipeline-level connection identified in today's analysis links this directly to first-wall protection: real-time fast-ion loss inference could trigger feedback control to reduce wall bombardment before damage accumulates.

██████████ 0.8 first-wall-materials Preprint

ReadSaveConnections

This study benchmarks three tomographic inversion algorithms — Minimum Fisher Information, Phillips-Tikhonov regularization, and Maximum-Likelihood Expectation-Maximization — for reconstructing radiated power distributions from bolometer measurements on the ADITYA-U tokamak, using synthetic phantoms that represent central, hollow, asymmetric, and divertor-localized emission profiles. Accurate, real-time radiated power maps are essential for tracking power exhaust to the divertor and detecting impurity accumulation, both of which must be controlled to protect plasma-facing components. The head-to-head comparison under realistic noise conditions gives practical guidance on algorithm selection for current and future machines where divertor heat loads are a primary engineering constraint.

██████████ 0.7 divertor-thermal Preprint

ReadSaveConnections

Neural network emulators trained on over one million synthetic Grad-Shafranov equilibria for the MAST-U tokamak can derive 'virtual circuits' — decoupled control modes linking coil currents to specific plasma shape parameters — in real-time via automatic differentiation, replacing fixed precomputed schedules that become inaccurate as plasma conditions evolve. Plasma shape control determines boundary strike-point positions on divertor tiles; losing accurate shape control can cause localized overheating or uncontrolled disruptions. Making virtual circuits state-aware and continuously updated is a practical step toward the adaptive control systems required for steady-state operation in future reactors.

██████████ 0.7 elm-control Preprint

ReadSaveConnections

A reinforcement learning controller trained with an asymmetric actor-critic architecture — where the critic sees the full equilibrium state during training but the actor operates only on magnetic sensor readings — achieves 2.01 cm mean shape error while tolerating random dropout of 30% of magnetic sensors per episode, eliminating the need for separate backup controllers when sensors fail. Sensor failures are a realistic operational risk in high-radiation fusion environments where diagnostics degrade over time; a single robust policy that gracefully handles arbitrary sensor loss reduces control system complexity and improves uptime. The approach avoids the computationally expensive explicit equilibrium reconstruction step, making real-time deployment on faster timescales feasible.

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

ReadSaveConnections

This study derives the widely used Boris integrator and the Rodrigues rotation scheme from a common exponential integrator framework for solving the Newton-Lorentz equation, showing that the Rodrigues scheme evaluates the matrix exponential exactly without further approximation while both schemes perform comparably when the magnetic field must be recomputed at each timestep. Boris-type algorithms are the workhorses of test-particle simulations used to study energetic ion transport and anomalous diffusion in turbulent fusion plasmas; placing them in a unified mathematical framework clarifies their error properties and opens paths to higher-order variants. The practical takeaway is that algorithm choice should account for the field-recomputation cost: in turbulent-field simulations where this cost dominates, the theoretical accuracy advantage of Rodrigues does not translate to a net speedup.

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

ReadSaveConnections

FIRM3D is an open-source Python/C++/CUDA framework for Monte Carlo guiding-center orbit integration of energetic particles in full 3D magnetic field geometries, with symplectic and Runge-Kutta integrator options and GPU acceleration benchmarked on NERSC Perlmutter hardware showing substantial throughput gains over CPU runs. Energetic particle losses to the first wall in stellarators and tokamaks with 3D field perturbations (ELM suppression coils, ripple, error fields) are difficult to quantify without 3D orbit codes; FIRM3D provides a publicly available, validated tool for this calculation, verified against the SIMPLE code. The ability to couple to MHD wave solvers (AE3D, FAR3D) enables self-consistent studies of how Alfvén waves re-distribute fast ions — relevant to both heating efficiency and wall loading.

██████████ 0.6 q-engineering Preprint

ReadSaveConnections

This theoretical feasibility study proposes that multi-ampere spin-polarized deuterium negative ion beams suitable for neutral beam injection can be produced by passing hydrogen isotope beams through cesium vapor optically pumped to 80% polarization using off-resonant Raman techniques with kilowatt-scale narrowband lasers in a 1 m long, 10 cm diameter cell. Spin-polarized fuel injection could increase fusion reactivity by up to ~50% relative to unpolarized D-T plasmas, potentially reducing the plasma current or density required to reach ignition and directly improving the Q-factor margin. The study is purely analytical with no experiments performed, so the claimed cross-section estimates and depolarization rates require experimental validation before engineering implications can be drawn.

██████████ 0.5 q-engineering Preprint

ReadSaveConnections