DeepScience — Nuclear Fusion

DeepScience · Nuclear Fusion · Daily Digest

Better Walls, a Backwards Simulator, and a 1-GW Blueprint

Three papers this week show fusion progress is quiet, methodical, and real — not a headline waiting to happen.

            July 16, 2026
          

Hey, I picked today's three stories from a batch of 283 papers — and I'll be honest, none of them announce a miracle. What they do is something arguably more useful: one team shows that coating a reactor wall differently doubles its operating range, another builds a simulator you can run in reverse, and a third publishes an actual blueprint for a one-gigawatt fusion power plant. Let me walk you through each.

Today's stories

              01 / 03
            

Coating a Fusion Reactor Wall with Boron Doubled Its Operating Range

Change what you coat the walls with, and suddenly your fusion reactor works in twice as many conditions.

Picture a non-stick frying pan. The coating doesn't generate heat — it changes how the surface behaves, which changes everything about what you can cook. The team running China's EAST tokamak did something analogous: they switched from coating the reactor's inner wall with lithium to coating it with boron, using a process called boronization. The question was whether this would change how the plasma behaves. It did — dramatically. The measurement they care about is called the Greenwald fraction — essentially how dense the plasma is, as a percentage of the theoretical maximum before things fall apart. Under lithium walls, EAST could run a promising operating mode called I-mode only within a narrow density band: roughly 35% to 54% of maximum. Under boron walls, that band expanded to 26–77%. Nearly double the range. Even more striking: the configuration that plasma physicists most want for a real reactor — called the favorable configuration — was achievable in 51% of boronized discharges, compared to just 8% under lithium. Why does this matter? I-mode is a regime fusion researchers genuinely like. It keeps the plasma well-insulated (good energy confinement) without triggering violent edge instabilities called ELMs — which are the thing that could damage reactor walls in a future power plant. Wider access to I-mode is a real operational win. The catch: this is one machine, one magnetic field strength, 37 boronized discharges compared against 48 lithium ones. The team proposes that higher recycling of fuel at the wall explains the expanded density range, but that mechanism is not yet confirmed. And what works on EAST's geometry may need to be relearned on different machines.

Glossary

Greenwald fraction — A number from 0 to 1 expressing how close a plasma's density is to the theoretical limit where it becomes too dense to control — higher is riskier.

I-mode — An operating regime for fusion reactors that combines good energy confinement with naturally suppressed edge instabilities, making it attractive for power plant designs.

ELMs (Edge-Localised Modes) — Periodic bursts of energy from the plasma edge — like small internal explosions — that can erode reactor walls if not controlled.

boronization — A wall-conditioning technique where a thin boron film is deposited on the inside of a fusion reactor to reduce impurities and change how the surface interacts with the plasma.

Source: Boronization-enabled I-mode on EAST tokamak with an expanded density window and favorable-configuration access

              02 / 03
            

A Fusion Simulator You Can Run Backwards to Design Better Reactors

What if instead of guessing the right dials for a fusion reactor, you could just tell the simulator what you want and let it work backwards?

When you use GPS navigation, you type in your destination and the app figures out the route. You don't specify every turn — you specify the goal. Most engineering simulation works the other way: you set all the parameters, run the simulation, see what happens, adjust, and repeat. That trial-and-error loop is expensive when you're modeling a fusion reactor — each run can take hours or days. TokaGrad, built by a team using a programming framework called JAX, is designed to break that loop. It's a full simulator of a tokamak plasma — covering how the plasma heats up, how it transitions from a poorly-confined state (L-mode) to a well-confined state (H-mode), and how the edge builds up — all connected in a single chain of calculations. The key feature is that this chain is differentiable, meaning the simulator can compute, mathematically, exactly how changing any input knob affects any output you care about. That lets you run it like a GPS: tell the simulator you want maximum confinement at the end of a discharge, and gradient-based optimizers — the same kind used in AI training — can automatically tune all the actuator waveforms to get you there. The team benchmarked the forward simulation against established tools and demonstrated actuator optimization on scenarios relevant to ITER. The honest limit here: this is a computational tool paper. The benchmarks are against other simulators, not against real experimental data. And 'differentiable' doesn't mean 'fast enough for real-time control' — at least not yet. But as a design and planning tool, the ability to propagate gradients through an entire discharge scenario is a genuine capability step forward.

Glossary

differentiable simulator — A simulation where the computer can automatically calculate how a small change in any input (like a heating power setting) affects any output — enabling mathematical optimization rather than trial and error.

L-mode to H-mode transition — The switch from a poorly-confined plasma state to a well-confined one — a critical moment in a fusion reactor's operation that determines how much energy is retained.

actuator waveforms — The time-varying control signals sent to a reactor's heating systems, magnetic coils, and fuel injectors — effectively the recipe the reactor follows over the course of a discharge.

JAX — A Google-developed programming library that makes it easy to compute derivatives automatically through complex numerical calculations, widely used in machine learning and increasingly in scientific simulation.

Source: TokaGrad: End-to-end differentiable tokamak simulator for L-to-H full scenario optimization

              03 / 03
            

Here Are the Blueprints for a One-Gigawatt Fusion Power Plant

Before you build a house, an architect draws every wall and beam on paper — and someone just did that for a fusion power plant big enough to power a city.

An architect's drawing isn't a house. But without a credible drawing, no one can check whether the house is even physically possible before a brick is laid. That's more or less what a team has published for GIGA — a proposed stellarator fusion power plant designed to generate one gigawatt of electricity. A stellarator is a type of fusion device with a twisted, pretzel-like magnetic cage instead of the doughnut shape of a tokamak. The advantage: it doesn't need a current running through the plasma to maintain its magnetic field, which makes it inherently more stable. The disadvantage: designing the shape is enormously complicated. The GIGA team started from the geometry of the Wendelstein 7-X stellarator — currently operating in Germany and the most advanced stellarator in the world — and scaled and optimized it up to a plasma volume of 1,500 cubic metres, targeting 3 gigawatts of fusion power (of which 1 GW would reach the grid). The optimization hit its key targets: more than 85% of the energy from fusion-born alpha particles is confined (not lost to the walls), the neoclassical ripple — a measure of how much particles leak out due to the magnetic geometry — is below 0.01, and the bootstrap current stays under 50 kA. To be clear: this is a computer-optimized plasma shape. It's an equilibrium design, not an engineering design. The magnets, the blanket, the exhaust system, the structural materials — none of that is in this paper. It's the architect's first drawing: rigorous, necessary, and still very far from a building.

Glossary

stellarator — A fusion device that confines plasma using a carefully twisted magnetic field generated entirely by external coils, with no need for a current running through the plasma itself.

alpha particles — Helium nuclei produced by the fusion reaction — they carry most of the energy and are meant to stay inside the plasma long enough to heat it; losing them to the wall wastes that energy.

neoclassical ripple — A measure of how much the magnetic field's imperfections cause particles to drift out of the confined region — lower is better.

bootstrap current — A self-generated electric current that arises naturally in a confined plasma; in a stellarator it's generally undesirable because it can distort the carefully designed magnetic geometry.

Source: The fixed boundary plasma equilibrium basis for a one Gigawatt electric stellarator power plant

The bigger picture

Look at what these three papers are collectively doing. One is tuning the physical surface where plasma meets machine. One is building the computational tools to design plasma behaviour more systematically. One is drafting the shape that a full-scale power plant plasma would need to take. These aren't three random topics — they're three layers of the same problem: you need a plasma that behaves well (I-mode, good confinement), you need tools to design and optimize how to get there (differentiable simulators, gradient-based control), and you need credible reactor-scale designs to aim at (the GIGA blueprint). None of these papers alone gets fusion to the grid. But they are working on adjacent parts of the same machine. If I had to take a position: the weakest link right now is still the middle layer — the engineering translation from 'we know what we want' to 'we know how to build it and operate it reliably.' That's where tools like TokaGrad matter most.

What to watch next

Keep an eye on ITER's schedule updates — the machine is now in assembly and any new commissioning milestone will test whether simulators like TokaGrad actually predict real machine behaviour. On the stellarator side, Wendelstein 7-X is expected to publish further high-performance plasma results in late 2026; those results will directly inform whether the GIGA design assumptions are physically grounded. The open question I'd most want answered: does boronization generalize to other machines, or is it something specific to EAST's wall geometry and plasma conditions?