Stellarator: Twisted Coils That Braid a Plasma Into Place

Why a Magnetic Doughnut Cannot Hold Anything

Bend a solenoid into a closed ring and you seem to have solved confinement: a torus of field with no ends to leak from. It fails in about a millisecond. Ampère's law for a ring of N coils carrying I gives B_φ = μ₀NI / 2πR — crowded on the inboard side, thin on the outboard side, falling as 1/R across the plasma. A field that varies in space pushes guiding centres sideways.

Two drifts do the damage, and in a torus both point vertically. The grad-B drift, of size W_⊥/(qB) × (∇B/B), comes from a gyro-orbit tighter on one side than the other; the curvature drift, 2W_∥/(qBR), is the centrifugal push on a particle streaming along a bent line. Their sum for a thermal particle is roughly v_d ≈ W/(qBR). In Wendelstein 7-X numbers — a 3 keV deuteron, B = 2.5 T, major radius R = 5.5 m — that is 3000 V ÷ (2.5 × 5.5 T·m) ≈ 220 m/s.

The lethal detail is the q in the denominator: ions drift up, electrons drift down. Charge separates vertically, and nothing undoes it, because a field line in a purely toroidal field is a circle at constant radius and constant height — it never visits both the top and the bottom, so electrons cannot stream along B to short the separation out. A vertical electric field grows instead, and every particle then feels the E×B drift, v = E×B/B², which depends on neither charge nor mass: ions and electrons move outward together, quasineutrality intact, no restoring force anywhere. The vertical drift alone crosses W7-X's 0.53 m minor radius in about 2.5 ms, and the E×B march is faster. A pure toroidal field is not a poor confinement scheme — it is not one at all.

Rotational Transform: One Number That Rescues the Torus

The cure is to make every field line helical, so that following one carries you from the top of the plasma to the bottom and back. The measure is the rotational transform ι: poloidal turns per toroidal lap, formally ι = dΨ_pol/dΨ_tor. (Some texts write ι/2π for the same thing; tokamak people quote its reciprocal, the safety factor q = 1/ι.) W7-X runs at ι ≈ 0.8–1.0 with almost no magnetic shear; a tokamak edge sits nearer ι ≈ 0.3, q ≈ 3.

Transform saves the plasma twice. A particle on a twisted line spends half of each circuit above the midplane, drifting outward, and half below, drifting inward: the excursion averages to zero and the orbit closes. Faster still, the line now connects the charged top to the charged bottom, so electrons stream along it and short the separation out before a field can build — those neutralising parallel flows are the Pfirsch–Schlüter currents. Check the timescale: a 3 keV deuteron moves at √(2W/m) = √(9.6×10⁻¹⁶ J ÷ 3.34×10⁻²⁷ kg) ≈ 5.4×10⁵ m/s, and one lap of W7-X is 2πR ≈ 35 m, so it circulates every ~65 μs; electrons at 10 keV take under a microsecond. Against a 220 m/s leak, that is not a race.

What the helical lines weave is nested flux surfaces. Force balance J×B = ∇p gives B·∇p = 0 — pressure cannot change along a field line — so each line is trapped inside a constant-pressure surface, and for irrational ι it never closes and ergodically paints that whole surface. The onion of surfaces is the confinement. Where ι hits a low-order rational n/m the line closes after m laps, and a tiny resonant error field tears the surface open into a chain of magnetic islands. W7-X turns the vice into its exhaust: the ι = 1 edge surface makes a 5/5 island chain that ten island-divertor units deliberately intersect.

Two Ways to Twist — and the Confusion With the Tokamak

Nature cares only that ι exists, and there are two engineering answers. The tokamak drives a toroidal current through the plasma, making the plasma the secondary winding of a transformer. That current generates the poloidal field, heats the plasma ohmically, and leaves the machine axisymmetric, so the coils are simple hoops — ITER's design value is 15 MA. The bill arrives later: a transformer pushes its flux one way, so the discharge is a pulse and steady operation needs inefficient external current drive; and a multi-mega-amp current stored in a hot, unstable conductor can dump itself in milliseconds — a disruption, with forces, runaway electrons and heat loads that scale badly toward reactor size.

The stellarator puts all of ι into the external coils and asks the plasma for nothing. No net toroidal current means steady state by construction, no current-driven kink, no sawteeth, no disruption — and no Greenwald density limit, which is a current-based scaling; stellarators run above it, bounded instead by radiation (the Sudo limit). The confining field also exists before the plasma does, which makes start-up placid.

External coils make transform in only two ways, usually combined. Torsion of the magnetic axis: if the axis of the torus is non-planar, field lines rotate as they follow it — Spitzer's figure-8. Rotating elongation: give the cross-section a bean or ellipse shape and rotate its long axis around the torus, dragging the lines poloidally with it. W7-X's cross-section morphs from bean to triangle and back, five times around.

The price is symmetry itself. An axisymmetric torus has an exactly conserved canonical angular momentum that alone keeps collisionless orbits bounded; a stellarator throws that theorem away, so its coils must be individually shaped, non-planar and built to millimetre tolerance. Do not confuse the relatives: a heliotron or torsatron such as Japan's Large Helical Device twists with continuous helical windings rather than modular coils, while the reversed-field pinch and spheromak take most of their field from plasma current and are not stellarators at all.

Spitzer's Figure-8, the Wilderness Years, and Quasi-Symmetry

Lyman Spitzer Jr., the Princeton astrophysicist the space telescope is named for, invented the concept in spring 1951 after press reports of Ronald Richter's bogus Argentine fusion claim — reputedly working the figure-8 out on a ski lift to Aspen. His classified July 1951 report was titled A Proposed Stellarator, from stella, star. Project Matterhorn became the Princeton Plasma Physics Laboratory in 1961; the tabletop glass figure-8 Model A ran in 1952–53, followed by Model B and the large Model C (1961–1969).

They confined badly — losses near the Bohm rate, D_B = kT/16eB, orders of magnitude above collisional theory. In 1968 the Soviet T-3 tokamak reported kilovolt electrons, confirmed in 1969 by a Culham team who shipped a Thomson-scattering laser to Moscow. Princeton cut Model C apart and rebuilt it as the Symmetric Tokamak in 1969–70, and the stellarator entered two decades of eclipse.

Those Bohm-level losses were anomalous — turbulence nobody could then calculate — but the structural defect that would have doomed a hot stellarator anyway was neoclassical. In a helical device |B| ripples not only with 1/R but with the windings, so particles can be mirror-trapped in a local well; a locally trapped particle never circulates poloidally, so its vertical drift is never averaged away and it walks straight out. In the low-collisionality 1/ν regime the diffusivity goes as D ∝ ε_eff^3/2 v_d²/ν, and since v_d ∝ T while ν ∝ n/T^3/2, that is D ∝ ε_eff^3/2 T^7/2/n — losses worsening exactly as you heat toward reactor conditions.

The escape was mathematical. Allen Boozer showed in the early 1980s that in the right magnetic coordinates, guiding-centre drift orbits depend on the field only through the magnitude |B|, not the shape of the coils producing it. A field can therefore be three-dimensional while |B| is symmetric, and particles cannot tell. In 1988 Jürgen Nührenberg and Reinhard Zille at IPP Garching found such configurations numerically — quasi-helically symmetric stellarators with |B| = B(ψ, mθ − nφ) — turning design into an optimisation problem. The Helically Symmetric Experiment (HSX) at Wisconsin–Madison (first plasma 1999, R = 1.2 m, B ≈ 1 T) then measured the predicted collapse in neoclassical loss. W7-X takes the sibling route, quasi-isodynamic or omnigenous: |B| is not symmetric, but its contours are shaped so the bounce-averaged radial drift of trapped particles vanishes and they precess poloidally instead. Effective ripple ε_eff falls to ~1% or below, against ~5–10% unoptimised.

Wendelstein 7-X: Photographing a Field Good to One Part in 100,000

W7-X sits at the Max Planck Institute for Plasma Physics in Greifswald. Approved in 1994 and assembled between 2005 and 2014 for of order a billion euros and roughly a million person-hours, it made its first helium plasma on 10 December 2015 and its first hydrogen plasma on 3 February 2016. Its magnet system is 50 non-planar superconducting coils in five shapes, ten of each, plus 20 planar coils for flexibility — NbTi cooled to about 3.9 K, carrying up to ~17.6 kA. Five field periods; R = 5.5 m, average minor radius 0.53 m, plasma volume ~30 m³, B = 2.5 T on axis. Ten 1 MW gyrotrons heat at 140 GHz: the electron cyclotron frequency is 28 GHz per tesla, so 2.5 T puts f_ce at 70 GHz and 140 GHz is its second harmonic. Ten actively water-cooled carbon divertor units rated near 10 MW/m² catch the edge islands.

How do you verify that a field this baroque is the one you designed? You photograph it. Before first plasma, W7-X ran at reduced field with an electron gun firing a low-energy beam along a single field line while a fluorescent rod was swept through the vessel; wherever the rod crossed the surface the beam paints, it glowed, and a camera recorded the intersections. Thomas Sunn Pedersen and colleagues published the result in Nature Communications in 2016: the topology matched the design, and residual error fields were below 1 part in 10⁵ — a tolerance five normal-conducting trim coils exist to preserve.

The rest is standard high-temperature diagnostics — Thomson scattering for electron temperature and density profiles, dispersion interferometry for line density, an X-ray imaging crystal spectrometer and charge-exchange recombination spectroscopy for ion temperature and rotation, bolometry and infrared thermography for power balance. In 2021 Craig Beidler and co-authors reported in Nature the measurement the programme was built to make: neoclassical energy transport in W7-X really is reduced against a deliberately unoptimised configuration. Typical performance is T_i ≈ 3 keV with T_e reaching ~10 keV, densities of order 10²⁰ m⁻³ and energy confinement times ~0.2 s. In February 2023 the machine sustained an eight-minute discharge with ~1.3 GJ of energy turnover; the 2024 campaign reached ~1.8 GJ, against a design target of 30 minutes at 10 MW. Japan's Large Helical Device (Toki, first plasma 31 March 1998, R = 3.9 m, continuous superconducting helical coils) is the other flagship, holding a 54-minute discharge from 2006 and ion temperatures near 10 keV in deuterium in 2018.

Where It Still Fails, and What Nobody Knows Yet

Solving neoclassical transport promotes turbulence to the bottleneck. With ripple losses optimised away, W7-X confinement is set by drift-wave turbulence — chiefly ion-temperature-gradient modes, since the optimisation happened to suppress trapped-electron modes. The first stellarators were not designed against turbulence at all; doing so is now the leading edge of the field.

The plasma is not current-free inside. A common overstatement is that a stellarator carries no current. It carries no net driven toroidal current, but Pfirsch–Schlüter currents flow along field lines and the pressure-gradient-driven bootstrap current still appears — a few kiloamps in W7-X, deliberately minimised, because a larger one would shift ι and drag the island divertor's strike lines off the targets.

Pressure limits and exhaust. Stellarators cannot disrupt, but they still meet interchange and ballooning limits: W7-X is designed for a volume-averaged β near 5%, W7-AS reached ⟨β⟩ ≈ 3.4%, and LHD has run near 5%. Radiative collapse and impurity accumulation are the realistic long-pulse failure modes, though W7-X has seen turbulence-driven impurity flushing rather than the feared neoclassical inward pinch.

Fast alphas are the great unknown. No stellarator has burned deuterium–tritium fuel; W7-X has no tritium capability and is a physics machine, not a power experiment. In a reactor the 3.5 MeV alphas must stay confined long enough to heat the fuel, and their drift orbits are far wider than thermal ions', so symmetry-breaking that is harmless at 3 keV can spray megaelectronvolt helium onto the first wall. That is the chief physics risk in any stellarator power-plant design.

Engineering tolerance is the historical killer. Princeton's National Compact Stellarator Experiment was cancelled in 2008 over cost and coil-tolerance overruns, and W7-X's own coil-manufacturing crisis in the early 2000s nearly ended that project; reactor-scale machines must also be maintainable inside an activated, three-dimensionally obstructed geometry. The tools have transformed, though: Matt Landreman and Elizabeth Paul showed in 2022 that configurations with precise quasisymmetry can be found directly by optimisation, and high-temperature REBCO superconductor permits higher fields with simpler coils — the combination several private ventures are now betting on.

Three ways to arrange a toroidal magnetic field — and what each one costs

Property	Pure toroidal field	Tokamak	Stellarator (W7-X)
Source of the twist (ι)	None — ι = 0	Toroidal plasma current (ITER: 15 MA)	External 3D coils (50 non-planar)
Net toroidal plasma current	None	1–15 MA, inductively driven	~kA bootstrap only, deliberately minimised
Pulse length	Plasma lost in ~1 ms	Seconds to minutes (ITER goal ~400 s)	Steady state by construction (8 min achieved, 30 min goal)
Disruptions	Not applicable	Yes — current-driven, MJ-scale	None — there is no current to disrupt
Symmetry	Axisymmetric	Axisymmetric	3D, five field periods
Dominant loss channel	Gross E×B drift to the wall	Turbulence (ITG / trapped-electron modes)	Helical-ripple neoclassical unless optimised, then turbulence