This Week in Nuclear Fusion

A busy week across 703 papers spanning plasma physics, engineering design, and machine learning applications to fusion dynamics. Compact tokamak design concepts continued to mature, with hybrid laser-heating architectures entering the conceptual phase targeting Q factors of 5–20. On the experimental front, DIII-D delivered one of the more consequential pedestal results in years: boron powder injection is opening a real path to sustained ELM-free operation. Meanwhile, mathematical tools from Koopman operator theory are quietly becoming relevant to plasma control, offering new ways to linearize the notoriously unruly nonlinear dynamics of burning plasmas. Underlying all three threads is a deeper tension: our reduced models of turbulent transport may be more fundamentally limited than previously assumed.

Top 3 Papers

1. PROJECT HCT-LH — Hybrid Compact Tokamak with Hybrid Lasers A conceptual compact tokamak (R = 2–3 m) integrates four technology pillars, including the Super-X divertor validated on MAST-Upgrade, targeting a Q range of 5–10 in the base case and up to 10–20 under optimistic assumptions. The design represents a credible mid-scale stepping stone between current devices and full pilot plants, leveraging proven divertor physics to manage the heat exhaust problem that has historically plagued compact geometries.

2. Impurity-Driven Turbulence Opens a Pathway to ELM-Free Operation (DIII-D) Controlled boron powder injection on DIII-D progressively suppressed ELM frequency and enabled sustained ELM-free periods of ~300 ms by decoupling the peeling and ballooning stability boundaries at moderate injection levels. The mechanism — selective amplification of low-frequency pedestal turbulence that regulates gradients before they reach the ELM trigger — suggests a self-regulating stability channel that could be actively managed in reactor-relevant regimes.

3. AdaKoop: Efficient Koopman Operator Regression for Nonstationary Dynamics By treating nonlinear plasma-like dynamics as linear transitions in an infinite-dimensional feature space via Koopman operator theory, AdaKoop enables tractable real-time modeling even when the underlying system drifts. An integrated statistical hypothesis test detects abrupt pattern shifts, allowing the model to adapt — a capability directly relevant to disruption precursor detection and real-time tokamak control.

Connection of the Week

Why your turbulence model breaks down near marginal stability — and it's not just truncation error

A theory paper on memory effects in hydrodynamics this week offers an unexpectedly sharp lens on a long-standing problem in fusion transport modeling. The paper demonstrates that when relaxation times become unbounded — as they do near critical gradients — the standard gradient expansion diverges, not because you've truncated it too early, but because slow non-hydrodynamic modes fundamentally prevent local equilibrium from emerging.

The bridge to fusion is direct: reduced transport models like TGLF and QuaLiKiz parameterize turbulent fluxes in the form Γ = f(∇T, ∇n, …), which is structurally a gradient expansion. Their known failures in collisionless regimes and near critical gradients have typically been attributed to insufficient fitting or missing physics terms. This paper's analysis suggests the failure is more fundamental — trapped particle effects and toroidal mode coupling act as slow non-hydrodynamic modes that cannot be resummed away in the gradient-expansion framework. The implication is that next-generation reduced models may need to explicitly retain slow degree-of-freedom dynamics rather than project them out. Confidence: plausible. Active roadblock: turbulence-modeling.

Want More?

This digest covers the headline signal from 703 papers. The full daily Pro digest includes every qualifying paper with connection mapping, Tree-of-Thought reasoning chains showing why each connection holds, confidence tiers, and a live roadblock tracker flagging where the field is genuinely stuck.

Get daily full digests with all connections, ToT reasoning chains, and roadblock tracking. Upgrade to Pro ($9/mo).