283

Papers

10/10

Roadblocks Active

2

Connections

Using a chain of coupled plasma transport codes, this study simulates the compact ARC tokamak and finds that seeding argon gas into the exhaust region can simultaneously achieve 750–1000 MW of fusion power and keep the divertor surface below 2 eV — cold enough to dramatically reduce sputtering damage to the tungsten walls. Neon seeding underperforms because it migrates into the plasma core and dilutes the fuel. This directly validates a candidate operational scenario for ARC and similar compact tokamaks, where squeezing high fusion power and wall protection into a small machine is the central engineering challenge.

██████████ 1.0 divertor-thermal Preprint

ReadSaveConnections

A deep learning disruption predictor trained entirely on EAST tokamak data was adapted to predict disruptions on the physically different J-TEXT tokamak without using any J-TEXT experimental shots, by augmenting training with synthetic signals generated from MHD simulations of the target machine. Early warning accuracy improved by 7 percentage points over the unadapted baseline. This matters for ITER and future devices, which cannot safely accumulate the thousands of disruptive shots needed to train conventional data-driven predictors — this approach offers a path to pre-deploying safety systems before a machine is turned on.

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

ReadSaveConnections

Physics-informed neural networks trained on the adjoint of the relativistic Fokker-Planck equation can predict runaway electron current, average energy, and energy distribution orders of magnitude faster than conventional kinetic solvers, across a wide range of plasma conditions. The adjoint formulation is key: one trained network handles any initial electron distribution without retraining. Fast, accurate runaway electron modeling is essential for real-time disruption mitigation systems, where a runaway beam can deposit megajoules into the first wall within milliseconds if not suppressed.

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

ReadSaveConnections

A computational design study shows that mounting a toroidal array of high-temperature superconducting dipole coils on an existing axisymmetric tokamak vessel can produce flexible hybrid operation ranging from a standard tokamak to a quasi-axisymmetric stellarator with rotational transform up to ι=1 under realistic plasma pressure. The coil forces remain within the tensile limits of current HTS tape. This matters because it offers a low-cost pathway to study stellarator confinement physics — which intrinsically avoids disruptions — inside an existing facility without building a dedicated non-axisymmetric machine.

██████████ 0.8 hts-magnets Preprint

ReadSaveConnections

A new empirical formula, fit to experimental data and benchmarked against the TALYS nuclear code, predicts (n,3n) reaction cross sections for heavy nuclei (mass 146–238) at precisely the 14.1 MeV neutron energies produced by D-T fusion. These reactions govern both neutron multiplication in beryllium- and lead-based breeding blankets and displacement damage in structural materials. Reducing cross-section uncertainty at 14 MeV directly tightens predictions of tritium breeding ratio margins and first-wall damage rates — two numbers that determine whether a reactor blanket will produce enough fuel to sustain itself.

██████████ 0.8 tritium-breeding Peer-reviewed

Read

The ECHO algorithm, deployed on the DIII-D plasma control system, uses a neural network surrogate of a microwave ray-tracing code to optimize gyrotron mirror angles and power in real time, hitting target radial heating profiles even when individual gyrotrons fail mid-shot. Validated across multiple experimental discharges with varying plasma conditions, it replaces a computationally prohibitive real-time ray-trace calculation. Precise control of where microwave heating is deposited is a primary tool for stabilizing neoclassical tearing modes and other instabilities that degrade confinement or trigger disruptions.

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

ReadSaveConnections

Beam-tracing simulations for the Pegasus-III spherical tokamak show that a single Doppler backscattering diagnostic can simultaneously measure plasma rotation velocity, ion-scale density fluctuations (wavenumbers 1–8 cm⁻¹), and magnetic field pitch angle — a combination not typically available from one instrument. Spherical tokamaks operate at low aspect ratio where turbulence and flow measurements are hard to make, and Pegasus-III is a testbed for high-beta plasmas relevant to compact fusion concepts. Multi-purpose diagnostics reduce cost and port access demands on small devices.

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

ReadSaveConnections

This theoretical paper demonstrates that toroidal magnetic flux enclosed within plasma surfaces has been systematically underused in tokamak analysis, and that omitting it leads to errors when applying Faraday's Law near an X-point separatrix — the magnetic boundary that defines the plasma edge in diverted operation. The analysis clarifies how poloidal flux can slip through the plasma boundary via loop voltage, which has direct implications for equilibrium reconstruction accuracy. Equilibrium reconstruction underpins every plasma shape and position control system, so analytical errors here propagate into control performance.

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

ReadSaveConnections

A physics-informed neural operator trained on approximately 10,000 TCV discharges reconstructs the 2D magnetic equilibrium — the map of where field lines sit inside the plasma — in under 100 microseconds, enabling 10 kHz real-time shape control. The modular branch-trunk architecture enforces physical consistency of the predicted flux map via automatic differentiation. Fast equilibrium reconstruction is the foundation of all plasma position and shape control; conventional solvers are too slow for high-bandwidth control loops needed to suppress edge instabilities.

██████████ 0.6 long-confinement Preprint

ReadSaveConnections

Automatic differentiation through a full inertial confinement fusion implosion physics model enables gradient-based optimization of 500-parameter laser pulse shapes for 25 kJ OMEGA-scale direct-drive targets. Without being told what pulse shape to use, the optimizer recovers near-isentropic compression pulses — the theoretically ideal form — purely from physics gradients. Compressing the laser pulse design cycle from weeks of hand-tuning to automated gradient descent has direct implications for achieving ignition margins and for designing the target shots needed to validate gain models at NIF scale.

██████████ 0.6 q-engineering Preprint

ReadSaveConnections