How Proxima Fusion Works: Stellarator Confinement and Resonant Energy Retention

In ninety seconds: tokamak, stellarator, Proxima

What is a tokamak? A magnetic chamber shaped like a torus, in which extremely hot plasma is held by magnetic fields. A substantial part of the confining configuration is produced by a large electric current flowing through the plasma itself.

Is Proxima Fusion building a tokamak? No. Proxima Fusion develops a stellarator. In a stellarator the confining configuration is produced mainly by external three-dimensional magnetic coils rather than by plasma current. The company works on a quasi-isodynamic (QI) architecture resting on physics developed for decades around the Wendelstein 7-X device.

Why the twisted doughnut? So that charged particles of the hot plasma do not leave too quickly. The field geometry is arranged so that charged particles remain confined over repeated trajectories rather than being lost rapidly to the wall.

What does an electrical resonator have to do with it? Not that they work the same way — the physics is entirely different. But an engineer designing either system faces one abstract question: once the state exists, how fast does it lose energy, and how much energy must be replenished for the state to persist.

Picture two machines. The first holds a megajoule and loses it in a second. The second holds one joule and loses it in an hour. Which one keeps its state better?

Stored energy alone does not answer the question. What matters is the rate at which the energy is lost. So the useful quantity appears on its own:

\[ \text{characteristic time} = \frac{\text{energy held in the state}}{\text{power at which it is lost}} \]

Joules divided by joules per second give seconds. This is the construction that plasma confinement physics and resonant electrical engineering each arrive at independently. The rest of this page is about how far the correspondence can be taken, and where it must stop.

What Proxima Fusion is building today

What it is. A German company spun out of the Max Planck Institute for Plasma Physics (IPP) ecosystem in 2023. It develops quasi-isodynamic stellarators using high-temperature superconducting magnets.

Stellaris. A published, peer-reviewed concept for a commercial power plant: a quasi-isodynamic machine with modular coils, minimised toroidal plasma current, an edge magnetic island chain enabling an island divertor, and high-temperature superconducting magnets. Electromagnetic, structural, thermal and neutronics analysis are brought into one coherent calculation [6].

What is being built now. The company’s public plan: a Stellarator Model Coil as a standalone hardware demonstration of the magnet technology, then the demonstration device Alpha, whose stated goal is net fusion energy gain in steady-state operation.

What does not yet exist. An operating fusion power plant. Alpha is a future demonstration machine, not a commercial facility, and net fusion energy gain in the plasma is not the same thing as net electrical output from the plant.

Fuel. The first generation is based on deuterium–tritium fusion. Tritium is practically absent in nature and has to be bred in a lithium-bearing blanket. The Stellaris concept includes a tritium-breeding blanket; the company itself stresses that this is a concept rather than a completed engineering design, and names tritium technology as one of the unsolved problems of the field.

Why a torus holds plasma at all

First without the jargon. A particle in a magnetic field does not travel along an ideal prescribed line. The field is non-uniform, the particle collides with other particles, and the plasma itself fluctuates. Over time, energy and particles therefore leak across the magnetic surfaces. The task of an optimised stellarator is to shape the three-dimensional field geometry so that one important systematic channel of radial loss of trapped particles becomes as small as possible. Other channels, turbulence above all, are not automatically closed by that choice. Only now do the precise names appear.

Three-dimensional field geometry opens a large configuration space for optimisation, but it introduces a strong temperature dependence into the stellarator’s neoclassical — non-turbulent — energy transport. For decades this dependence was regarded as the principal weakness of the concept.

A key concept in this optimisation is omnigenity. A field is omnigenous when the bounce-averaged radial drift of trapped particles vanishes: the particle follows a complicated orbit but is statistically not displaced outward. The rigorous formulation and its transport consequences are given in Helander’s review [1]; the constructive class of fields with good confinement was introduced by Cary and Shasharina [2].

A quasi-isodynamic configuration is a subclass of omnigenous fields in which contours of constant magnetic field strength close poloidally. In the perfectly quasi-isodynamic limit and under the corresponding conditions, including zero net toroidal current, the bootstrap current in the low-collisionality regime vanishes, and the associated neoclassical problem simplifies substantially and can be reduced to the tokamak one [3].

Two developments moved the approach from theory into design practice. First, analysis of Wendelstein 7-X discharges showed that the record triple-product values in that device are evidence of precisely the reduced neoclassical transport that the design optimisation was meant to deliver [4]. Second, methods appeared for finding QI configurations with simultaneously low neoclassical transport, low fast-particle losses and suppressed turbulence [5].

Tokamak and stellarator: where the difference lies

In a tokamak the confining field is produced by external magnet systems together with a current flowing through the plasma itself. The large toroidal current creates a substantial part of the poloidal magnetic field and so participates in forming the magnetic surfaces. A property of the confined plasma therefore becomes part of the mechanism that confines it.

This creates a class of problems that is greatly reduced where confinement does not rely on a large toroidal plasma current. Current-driven instabilities, operational limits, disruptions with rapid release of stored energy and generation of runaway electrons are collected in the ITER physics basis chapter on MHD stability and disruptions [11]. The second consequence is temporal: an inductively sustained current is finite in duration, and continuous operation requires non-inductive current drive, which itself consumes power and is a design problem of its own [12].

In a stellarator the confining field is formed by external coils. The coil geometry and associated engineering are more complex, but the class of problems tied specifically to sustaining and stabilising a large toroidal plasma current is largely removed.

The precise statement of the difference: in a tokamak, improving confinement and managing stability are addressed in part to the same object.

What an electrical resonator holds

Now the other machine.

In an idealised LC circuit the charge carriers do not pass through the system once. The local ensemble is redistributed many times between two forms of organisation: the electric state of the capacitance and the magnetic state of the inductance. Energy flows rhythmically between \( q^2/2C \) and \( Li^2/2 \), and in an ideal circuit the sum is constant.

A terminological precision is required here, without which errors begin.

This is not pedantry. In a stellarator one may speak literally of spatial confinement of a material ensemble inside the magnetic configuration. In a resonator no analogous spatial circuit exists for the individual carrier: what is preserved is the energy of the organised oscillatory state of the ensemble, not the trajectory of a particle.

This is not a historical analogy. Coupled resonant structures and the resonant transformer remain an active engineering territory rather than an early twentieth-century curiosity. Current work addresses not only the classical two-circuit resonant transformer but multi-resonant architectures, coupling geometry and the formation of several coupled modes: a model of a resonant transformer with an extremely high turns ratio, built on a segmented representation of mutual inductance, was published in 2025 [13], and a compact high-efficiency pulse source on a triple-resonant transformer in 2026 [14]. The name “Tesla transformer” designates a physical class and does not by itself determine a particular electrical behaviour.

How the working resonant state is initially excited belongs to a different engineering stage and is not treated on this page.

The corresponding measure of how long the resonant state persists is the quality factor.

Put simply: the higher the quality factor, the more oscillation cycles the state can persist through before a significant fraction of its energy leaves through losses and load. That is an aid to intuition, not a definition.

The definition: \( 2\pi \) times the ratio of stored energy to the energy lost per cycle. Which losses enter the denominator matters.

The internal quality factor accounts only for internal dissipation, \( Q_i = \omega_0 W / P_{\text{int}} \). The loaded quality factor accounts for every permitted channel through which energy leaves the mode, including outflow into the coupled circuit:

\[ \frac{1}{Q_L} = \frac{1}{Q_i} + \operatorname{Re}\frac{1}{Q_c}, \qquad Q_L = \omega_0\,\frac{W}{P_{\text{int}} + P_{\text{ext}}} \]

where \( Q_c \) corresponds to coupling with the measurement or extraction path [8]. The quality factor is a measured parameter; methods of extracting it from complex transmission data and their comparative accuracy are treated in the standard work of Petersan and Anlage [7].

The bridge: one balance construction, two physics problems

In plasma confinement physics the central quantity is the energy confinement time: the ratio of energy stored in the plasma to the loss power in steady state,

\[ \tau_E = \frac{W}{P_{\text{loss}}} \]

This is the quantity collected in the international stellarator database and described by the ISS04 similarity scaling [9].

Now take the loaded quality factor and divide by \( \omega_0 \). In the standard weak-damping regime this gives the energy decay time of the mode:

\[ \tau_{\text{res},E} = \frac{Q_L}{\omega_0} = \frac{W}{P_{\text{int}} + P_{\text{ext}}} \]

The right-hand sides share a structure, but the quantities do not become physically identical because of it. In both cases the engineer takes the stored energy and divides it by the rate at which that energy irreversibly leaves the chosen state. The result has the dimension of time and answers a shared question: if replenishment stopped, how long could the state persist given the existing loss channels?

Beyond that the two physics diverge. \( \tau_E \) is a transport characteristic of a complex open multi-component system, set by collisional and turbulent transport. \( \tau_{\text{res},E} \) follows from the damping dynamics of one specific mode. What they share is a balance construction, not a theory.

That the construction describes a real object rather than a convenient fraction is visible in superconducting resonator practice: internal quality factors above \( 5\times10^{8} \) are reported together with intrinsic state lifetimes of the order of \( 10^{-2} \) s [10].

Without the formulas one sentence remains, and it needs no preparation: the stellarator fights the escape of energy from hot plasma through transport; the resonator fights the escape of energy from an oscillatory mode through dissipation and load.

The difference appears at normalisation. A resonator is treated as a periodic mode with its own frequency \( \omega_0 \), so its lifetime can be expressed by the dimensionless \( Q_L = \omega_0 \tau_{\text{res},E} \). The energy confinement time of a plasma characterises its entire energy store rather than one distinguished mode; there is no universal \( \omega_0 \) to normalise against. A plasma has many characteristic frequencies, and damping rates are defined for individual plasma oscillations — but that is a different quantity and a different problem.

Five places where the analogy ends

Break 1. Trajectory versus organised state. The stellarator holds particles in space. The resonator preserves the energy of an organised oscillatory state. The phrase “charges run in a circle like plasma round a torus” is physically wrong.

Break 2. An energy source inside the boundary. A fusion device contains a nuclear energy source inside its boundary, and the whole construction of ignition criteria rests on that. A passive resonant circuit has no internal energy source: a store is a store, not a source. Pumping must replenish both the losses and the energy handed to the extraction path; if it supplies less than that sum, the energy stored in the mode decreases. The correspondence of the previous section concerns the discipline of confinement and is not a statement about the origin of energy in either system.

Break 3. Where the useful product goes. In a resonator the useful output is drawn from the very state that is being held: extraction loads the mode, enters \( Q_L \), and competes with confinement. In a fusion device the split runs straight through the products of the reaction:

\[ \mathrm{D} + \mathrm{T} \rightarrow \alpha\,(3.5\ \text{MeV}) + n\,(14.1\ \text{MeV}) \]

The charged alpha particle remains a participant in the plasma energy balance. The neutron is electrically neutral and, once born, is essentially not governed by the magnetic configuration: it leaves the plasma and delivers its energy to the blanket. At the same time the rate at which neutrons are born depends critically on confinement quality, because confinement is what sustains the conditions of the reaction. The magnetic system holds not the product but the conditions of its appearance.

Break 4. The homonym “Q”. In fusion engineering \( Q \) is a power gain, the ratio of fusion power to the heating power invested. In electrical engineering \( Q \) is the quality factor, the ratio of store to loss per cycle. The first is a ratio of two powers characterising fusion gain at the plasma level. The second is a dimensionless characteristic of the lifetime of a resonant mode relative to its period. Conflating the two is a common and consequential error in cross-field comparisons.

Break 5. Where the organising mechanism resides. In a tokamak part of the confining configuration is created by the confined medium itself, and that same property gives rise to a distinct class of instabilities [11]. In a stellarator the field is formed by external coils [3]. In a resonant system the organising mechanism is external by construction: the pump is not part of the state being held. The only thing that follows is that the tokamak-specific problem of relying on a property of the confined plasma to help create its own confining configuration does not arise in the resonant picture and must not be carried there by analogy. Nothing about the stability of a resonant regime follows from the external position of that mechanism; its own limiting mechanisms are established separately.

Only the boundary establishes the result

There is one further coincidence — not physical but accounting, and in practical consequence the most important of all.

In neither field is the internal intensity of the process a measure of the useful result.

Two levels have to be separated at once. The energy confinement time is a plasma-physics quantity, not the external boundary of a power station; after plasma physics, fusion engineering still requires a separate plant-level energy balance. In that balance, between gross thermal power and the electricity delivered to the grid, sits substantial plant auxiliary and recirculating power demand: magnets, plasma heating, pumping, cooling and power conversion. In the tokamak branch the power of non-inductive current drive is added — the price of continuous operation, which cannot be moved outside the balance [12]. The Stellaris concept is built precisely as a coherent end-to-end calculation rather than a collection of separately optimised subsystems [6].

In resonant electrical engineering the same rule holds in its own wording: high internal circulation intensity is not a measure of active power. Inside the circuit, voltage and current are shifted in phase, the product of amplitudes is not active power, and active power at a stated measurement plane is determined by the synchronously acquired time average \( \langle u(t)\,i(t) \rangle \) over the defined observation interval.

From this follows a shared order of work, the same for both programmes. First the problem of the existence of an organised state is solved; only then the problem of useful output. For the fusion machine the chain runs plasma state → confinement → reaction conditions → neutron energy → thermal conversion. For the resonant system it runs oscillatory state → persistence → controlled coupling → electrical output.

The qualifier is obligatory: this concerns a steady or repeating regime averaged over the appropriate time window. A single fast release of a slowly accumulated store does not contradict it and belongs to a different problem. In the fusion programme the constraint shows up through confinement, heating and burn conditions. In the resonant one, through loaded quality factor, losses and extraction. The constraint is shared; the physics that realises it diverges completely. Regime-level accounting rules are set out on the energy model page, and the metrology of active power on the active power metrology page.

Why this analysis is published here

The direct answer: because VENDOR.Energy works with a resonant electrodynamic architecture, and the accounting discipline of the previous section is a working rule of the project rather than a borrowed metaphor. The full device boundary serves as the verification envelope, internal boundaries describe the mechanism, and mixing quantities from different boundaries under one symbol is prohibited.

But first, on scientific ancestry, because here the two programmes are symmetric in a way that is easy to miss.

Proxima Fusion is not reinventing magnetic plasma confinement. Its engineering programme rests on decades of experimental stellarator physics including Wendelstein 7-X; the company’s own contribution is the selection and integration of a particular quasi-isodynamic configuration into a future power plant design.

The same principle applies to VENDOR, with a different literature family. The project does not claim that coupled resonant circuits, resonant transformer dynamics and energy transfer between coupled resonant structures are new physics. These classes exist in the peer-reviewed literature independently of it and continue to develop [13] [14]. The project’s public scientific map places the coupled resonant node of the device in that same territory: coupled resonators, transient and resonant dynamics of resonant transformers, voltage and current transformation in coupled resonant structures, resonant energy transfer and extraction under load. The full literature map is held on the scientific foundations page.

Three levels that must not be glued together: established physics does not prove a particular architecture, and a particular architecture does not automatically prove a system-level result. For the fusion programme the chain reads stellarator physics → the Stellaris and Alpha architecture → a future integrated result. For the electrodynamic one, the physics of coupled resonant structures → the device architecture → independent system validation. The novelty of either engineering programme is to be sought not in the existence of separate physical laws but in the specific composition of known physical operations into a new machine.

And the second question that arises here deserves an equally direct answer.

Three differences of class, visible without a single number.

Fuel cycle. A fusion plant is an object with a material fuel cycle: deuterium and tritium as real reagents, tritium breeding in a lithium blanket, tritium handling, a neutron cycle. A resonant electrical architecture contains no such cycle.

Path topology. A fusion plant first produces nuclear energy, converts a substantial part of it into heat, and only then into electricity: blanket → heat → conversion → electrical output. An electrical system forms its output through an electrical path from the working process to the user interface. Different classes of heat rejection system and different self-consumption balances follow. This difference describes only the topology of conversion downstream of the respective working state and does not answer the question of energy attribution; that question belongs to a separate boundary validation programme.

Deployment topology. The deployment unit of a fusion programme is a power plant. The deployment unit of a distributed electrical architecture is a node. This changes the meaning of the word scaling: in one case the task is to make a single large system good enough that its own infrastructure is paid for by the plant’s output; in the other, to make a single node small, reproducible and modular enough that power is accumulated by the number of units installed.

From this follows why the comparison “a gigawatt against a kilowatt” is uninformative: it sets a future design-stage plant against a prototype of another category, and the two architectures have different optimisation functions. Contemporary deuterium–tritium plant concepts remain large integrated objects: alongside the plasma system they require a vacuum vessel, magnets, cryogenics, heating, a blanket, tritium handling, neutron shielding, heat rejection and the electrical balance of plant. Their minimum useful unit therefore belongs to an entirely different infrastructure class than a distributed electrical node. This is neither a universal law about the economics of the field — the field itself is working on size reduction through high-field magnets — nor a claim of superiority. It is a statement of different objective functions.

What is shown, what is designed, what is still to be verified

No statement about one programme carries over to the other. Every item has its own evidence class.

Applied to the material of this page:

• Measured: reduced neoclassical transport in Wendelstein 7-X, published in a peer-reviewed journal [4].
• Published / modelled: the Stellaris power plant concept [6]: an integrated calculation, not a built machine.
• Design target: the dates of the model coil and the Alpha demonstrator, and the intended net energy gain, per the company’s public statements.
• Published / modelled: the tritium-breeding blanket in Stellaris, which the company itself calls a concept rather than a completed engineering design.

For VENDOR.Energy this page likewise does not replace the status pages. The project has a physical laboratory implementation and internally recorded working results: on the scale above that is Measured as regards internal observation and Not externally validated as regards independent metrological status. Specific regimes, operating hours and numerical results belong to the endurance test page and the validation page and are not reproduced here.

What does not follow from anything on this page:

• That a resonant circuit is physically similar to a plasma column.
• That any analogue of plasma current, bootstrap current or non-inductive current drive exists in a resonant system.
• That plasma confinement and the retention of an oscillatory state are described by one transport theory.
• That any publication cited here examines VENDOR.Max, verifies it, or describes its composition.
• Anything about the origin of energy in any system: attribution of a source is settled by boundary inventory, not by analogy.