284

Papers

10/10

Roadblocks Active

4

Connections

iGENE reimplements the established GENE gyrokinetic code in TensorFlow so that gradients of any plasma output (e.g., heat flux) can be automatically computed with respect to any input (e.g., temperature gradient). This matters because fusion plasma turbulence is a major cause of energy loss in tokamaks, and being able to efficiently optimize plasma profiles against turbulence predictions — rather than relying on expensive trial-and-error simulation — could substantially reduce confinement losses. The approach works even through noisy, chaotic nonlinear turbulence simulations, which was previously considered a fundamental barrier.

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

ReadSaveConnections

This study designs a compact high-temperature superconducting (HTS) tokamak anchored to what the structural material can actually survive: a V-4Cr-4Ti vacuum vessel limited to 35 displacements-per-atom (DPA) over a 20-year lifetime drives a fusion power density of just 0.7 MW/m². The key finding is that the vacuum vessel is the lifetime-limiting component — the HTS magnets last roughly 16 times longer — which reframes reactor economics and maintenance strategy. For off-grid and marine applications where grid-scale power is not the goal, this materials-first design philosophy may be more practical than pushing toward maximum power output.

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

ReadSaveConnections

Solid plasma-facing materials erode under fusion plasma bombardment, motivating interest in self-replenishing liquid metal surfaces. This simulation study shows that adding small amounts of oxygen or hydrogen causes low-atomic-number elements (lithium or aluminum) to preferentially migrate to the surface of liquid tin alloys, making the surface behave like a lighter-element material that sputters less aggressively. Sn-Al and Sn-Li alloys are identified as viable candidates, with the surface enrichment mechanism quantified using a newly developed molecular dynamics force field validated against quantum chemistry data.

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

ReadSaveConnections

In tokamaks, self-organized 'zonal flows' can suppress turbulence and reduce energy losses — but when collisions between ions and neutral atoms increase (e.g., near the plasma edge), these beneficial flows can break down and be replaced by 'streamers' that transport energy and particles radially outward. This experiment on a linear plasma device directly maps that transition by varying neutral pressure across three conditions, identifying the nonlinear coupling mechanism through which streamers emerge. Understanding when and why this transition occurs is directly relevant to controlling edge turbulence in reactor-grade plasmas.

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

ReadSaveConnections

SPARC is expected to produce around 140 MW of fusion power, and this paper builds a forward model predicting what gamma-ray signals specific detector designs would see during actual deuterium-tritium experiments. The model accounts for neutron background noise and identifies which nuclear reactions (particularly T+D→He-5+γ) can be used to infer fusion power, fast-ion behavior, and heating efficiency in near-real-time. Validated detector designs that can measure performance under intense neutron flux are a prerequisite for the feedback control systems SPARC will need to operate safely.

██████████ 0.8 q-engineering Preprint

ReadSaveConnections

Plasma rotation suppresses turbulence and impurity influx in tokamaks, but controlling the full radial rotation profile — not just a single value — requires coordinating four actuators simultaneously with no reliable physics simulator available. This work trains a reinforcement learning policy using only historical shot data from DIII-D (avoiding inaccurate simulators), then deploys it on the real machine, achieving promising profile control. The result demonstrates that offline RL is a practical path to complex, multi-variable plasma control without requiring accurate first-principles models.

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

ReadSaveConnections

Stellarators can confine plasma without continuous current drive but have historically required extremely precise, fixed coil shapes. This paper proposes adding 288 programmable dipole coils to a tokamak-like coil set, enabling the same physical device to generate over 1.66 million distinct optimized magnetic configurations spanning three major symmetry classes. The concept could dramatically accelerate stellarator physics exploration by allowing a single device to scan configuration space that would otherwise require building many separate machines.

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

ReadSaveConnections

Conducting shells around tokamak plasmas slow down disruptions by opposing rapid changes in magnetic flux — but cutting the shell into toroidal segments (to reduce eddy current losses during normal operation) weakens this protection. This peer-reviewed paper derives an analytical formula for how segmentation degrades the shell's response time, showing the effect grows quadratically with the number of gaps. This gives engineers a concrete tool to optimize the trade-off between disruption mitigation performance and operational efficiency in the 'Armadillo' shell geometry.

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

ReadSaveConnections

When modeling the shape of a fusion plasma in 3D (as in stellarators), the choice of where to place the computational boundary matters enormously — placing it at the plasma edge rather than in the surrounding vacuum region introduces errors. This paper derives the correct placement strategy (in the low-current, low-pressure vacuum region exterior to the scrape-off layer) and provides a compatible equilibrium solver algorithm. The work also identifies that existing stellarator coil optimization routines must be updated to remain consistent with this correct boundary treatment.

██████████ 0.7 plasma-wall Preprint

ReadSaveConnections

Controlling plasma instabilities in fusion devices requires acting on limited sensor measurements, yet the underlying physics evolves in full 6D phase space. This theoretical paper shows that a controller trained by imitating a 'perfect-information' expert — using only sparse density measurements at sensor locations — can still achieve provable stability guarantees, with performance degradation bounded by a measurable property of the initial plasma distribution. While still theoretical and validated only on a 1D model system, the result provides a rigorous foundation for building certifiably stable plasma controllers from limited diagnostics.

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

ReadSaveConnections