Self-organized criticality, magnetic islands, and a self-sustaining loop
A tokamakA doughnut-shaped device that confines plasma using magnetic fields. The name is a Russian acronym. The leading approach to controlled nuclear fusion. confines plasma — ionized gas at 150 million degrees — inside a doughnut-shaped magnetic field. The goal is controlled nuclear fusion. The obstacle is turbulence: the hot plasma generates turbulent fluctuations that carry heat outward, eroding confinement. In the late 1980s, physicists discovered that above a threshold heating power, the plasma spontaneously enters a state of dramatically improved confinement called H-modeHigh-confinement mode — a plasma state with dramatically reduced energy loss, discovered in 1982. The plasma spontaneously transitions from L-mode (low confinement) when heating exceeds a threshold..5 No one designed it. The plasma organized itself.
The key structure in H-mode is the pedestalThe narrow region at the plasma edge where the pressure profile steepens sharply, forming a transport barrier. Named for the step-like shape it creates in the pressure profile.: a narrow region at the plasma edge where the pressure profile steepens dramatically, forming a transport barrierA region where turbulent transport of heat and particles is strongly suppressed, allowing steep pressure gradients to build. The pedestal IS the transport barrier.. The pedestal is not static. It builds up, releases its energy in violent bursts called edge-localized modes (ELMs)Violent periodic bursts that release energy stored in the pedestal. Like pressure-relief valves that fire when the gradient exceeds a stability limit. Damaging to reactor walls., rebuilds, and bursts again. This cycle is governed by three independent critical phenomena converging in a single region of plasma.
The tokamak is the only non-biological system in this series. Its inclusion demonstrates that multi-criticality is not a special property of living matter. It is a property of systems where multiple independent transitions converge — regardless of substrate.
Chapman, Dendy, and Hnat (PRL 2001) demonstrated that the pedestal operates as a self-organized criticalA system that naturally evolves toward a critical state without external tuning — like a sandpile that maintains itself at the angle of repose. Coined by Bak, Tang, and Wiesenfeld (1987). (SOC) sandpile.1 The pressure gradient is the slope of the pile. Turbulent transport events are the avalanches. When the gradient exceeds a local threshold, a transport event fires, reducing the gradient locally — like sand sliding down an oversteep slope. The system self-organizes to maintain the gradient near the critical threshold.
ELMs are the large avalanches. The pedestal builds until the entire edge region exceeds the stability limit, then releases its stored energy in a single catastrophic event. Between ELMs, smaller avalanches continually redistribute energy, maintaining the critical gradient. The avalanche size distribution follows a power lawA mathematical pattern where the probability of an event decreases as a fixed power of its size. No characteristic scale — small events are common, large ones rare but not impossible. — the signature of SOC.
In 2024, Park and colleagues published a result in Nature Physics: direct observation of magnetic islandsClosed loops in the magnetic field that trap plasma, breaking the normal nested-surface geometry. They form at surfaces where field lines close on themselves after a rational number of turns. forming during ELM suppression.2 Magnetic islands are topological structures — regions where the magnetic field lines close on themselves, creating islands of trapped plasma within the otherwise nested flux surfacesNested, doughnut-shaped surfaces traced out by magnetic field lines in a tokamak. In ideal conditions, plasma is confined to these surfaces. Their integrity determines confinement quality..
When external magnetic perturbations (resonant magnetic perturbations, or RMPsExternal magnetic fields applied to deliberately create magnetic islands at specific locations. The primary tool for suppressing ELMs in fusion reactors, including ITER.) are applied to suppress ELMs, they work not by smoothly modifying the pressure profile, but by creating magnetic islands at rational flux surfacesSurfaces in the tokamak where magnetic field lines close on themselves after a rational number of toroidal/poloidal turns (e.g., 3 toroidal turns per 2 poloidal turns). Islands form preferentially here.. The islands break the nested topology of the magnetic field, introducing new pathways for particle and energy transport. This is a topological phase transitionA change in the fundamental connectivity of a field — not a gradual shift but a qualitative restructuring. Creating a magnetic island changes the topology from nested tori to a punctured surface. in the magnetic field structure: from nested, intact flux surfaces to an island-containing topology with fundamentally different transport properties.
The transport barrier does not just exist — it creates the conditions for its own existence. The mechanism is a closed autocatalytic loop:
Steep pressure gradients drive turbulence. Turbulence generates Reynolds stressThe force exerted by turbulent fluctuations on the mean flow. In plasma, turbulent eddies transfer momentum to create large-scale zonal flows. The bridge between turbulence and order.. Reynolds stress produces zonal flowsSheared plasma flows that wrap around the torus in the toroidal direction. Generated by turbulence via Reynolds stress. They suppress the turbulence that created them — a key self-regulating mechanism. — sheared plasma flows that run around the torus.6 Zonal flows suppress the turbulence that created them. Suppressed turbulence allows the pressure gradient to steepen further. The steeper gradient drives more turbulence, completing the cycle.
McMillan and colleagues (2018) identified a self-sustaining quasi-traveling-wave at the plasma edge that exists at the edge of chaosThe boundary between laminar (ordered) and turbulent (chaotic) flow. Structures at this boundary are marginally stable — they persist only by continuously regenerating their own stability conditions.3 — a structure that propagates only because it continuously creates the conditions for its own propagation. Remarkably, our own analysis suggests this cycle requires stochastic noise exceeding roughly 100% signal-to-noise ratioThe ratio of a signal’s strength to the background noise level. An SNR above 100% means the noise is larger than the signal — the system is more random than deterministic.◆ our resultAn original, not-yet-peer-reviewed result from this research program — distinct from established consensus. to sustain itself.4 On this reading the transport barrier is noise-stabilized: it may not exist in a perfectly deterministic system.
Drag the slider to change the stochastic noise level. The transport barrier exists only above a threshold — it is stabilized by noise, not destroyed by it.
The three transitions are not three independent descriptions of the same plasma. They interact in a specific hierarchy:
The autocatalytic loop creates the transport barrier — the steep gradient that defines the pedestal. Without the loop, no pedestal exists. The SOC dynamics govern how the pedestal releases stored energy: small avalanches between ELMs, large avalanches during ELMs. The SOC system sits on top of the autocatalytic structure. The magnetic topology provides the control mechanism: magnetic islands can break the loop by short-circuiting the transport barrier, which is how ELM suppression works.
This creates a three-layer structure: autocatalysis builds the barrier, SOC governs its release dynamics, and topology provides the intervention point. The three layers interact. Magnetic islands modify the gradient profile, which changes the SOC dynamics, which alters the autocatalytic loop’s operating point. The coupling produces behavior — like the noise-stabilized quasi-traveling-wave — that no single layer predicts.
The first five tutorials describe biological or soft-matter systems. The tokamak pedestal demonstrates that multi-criticality is not a property of biological matter. It is a property of systems with multiple independent transitions that couple.
The key parallel is with the brain. Both systems are modular-hierarchical: the brain has cortical modules with heterogeneous dynamics; the tokamak has nested flux surfaces with radially varying profiles. Both produce robust criticality without fine-tuning: the brain through Griffiths phases; the tokamak through the predator-prey loop’s self-regulation. Both contain topological structures that organize transport: cortical spirals in the brain; magnetic islands in the tokamak.
The substrates could not be more different — neurons at 37°C versus hydrogen plasma at 150 million degrees. But the mathematical structure of the multi-critical coupling is the same: architecture produces robustness, topology provides control, and the coupling between them generates emergent dynamics that single-transition physics cannot predict.
Chapman, Dendy & Hnat (PRL 2001) constructed a running sandpile model of the tokamak edge where: (1) the local variable is the pressure gradient at each radial position, (2) a threshold gradient exists at each position (the ideal MHD stability limit), (3) when the local gradient exceeds the threshold, a transport event fires that redistributes pressure to neighboring cells, (4) external heating continuously steepens the gradient at the inner boundary.
The model reproduces the key phenomenology: a pedestal forms naturally (the average gradient profile steepens to near the threshold), ELM-like events occur when a cascade reaches the system boundary (full-width avalanche), and the inter-ELM period is set by the competition between heating rate and avalanche transport. The avalanche size distribution follows a power law with a cutoff at the system size — SOC with finite-size effects.
Crucially, the SOC model does not require the predator-prey loop as input. The loop creates the conditions for SOC (the steep gradient), but the SOC dynamics are self-organizing once the gradient exists. This is the layered structure: the loop creates the pedestal, SOC governs its fluctuations.
Park et al. (Nature Physics, 2024) used electron cyclotron emission imaging to directly observe magnetic islands forming at rational flux surfaces (q = m/n) during RMP-driven ELM suppression. The islands have measurable widths, positions, and rotation frequencies. Their formation coincides with the loss of the pedestal’s steep gradient — the islands break the transport barrier by connecting inner and outer flux surfaces.
This is a topological transition: the magnetic field topology changes from nested, closed flux surfaces (topologically equivalent to nested tori) to a topology containing O-points and X-pointsThe two special points of a magnetic island chain: the O-point is the island’s centre (where field lines close in a small loop), the X-point is where neighbouring field lines cross. Their appearance marks the change in magnetic topology. (islands). The transition is discrete — you either have an island or you don’t — and it produces a qualitative change in transport: from the ambipolar, neoclassical transport of intact surfaces to the streaming transport along the island’s separatrix.
The practical significance: ITER (the international fusion experiment under construction) plans to use RMPs for ELM control. Understanding the topological mechanism — that ELM suppression works by creating magnetic islands, not by smooth gradient modification — is essential for predicting RMP effectiveness.
McMillan et al. (2018) identified a coherent structure at the plasma edge — a quasi-traveling-wave (QTW) — that exists at the edge of chaos: the boundary between laminar and turbulent flow. The QTW propagates poloidally around the tokamak, maintaining the transport barrier as it travels.
The QTW is self-sustaining in the sense described by the predator-prey loop: it generates the zonal flow shear that suppresses the turbulence behind it, while the turbulence ahead of it (driven by the steep gradient) provides the energy that sustains the wave. It propagates only because it continuously creates the conditions for its own propagation.
Our own analysis◆ our resultAn original, not-yet-peer-reviewed result from this research program — distinct from established consensus. suggests that this structure cannot persist in a purely deterministic model: it appears to require stochastic fluctuations exceeding roughly 100% of the mean signal. On this account the noise provides two essential functions: (1) it seeds the turbulence ahead of the wave that the wave needs as fuel, and (2) it prevents the wave from locking into a fixed spatial pattern and decaying. The transport barrier would then be stabilized by noise — a counterintuitive possibility that echoes the role of thermal fluctuations in biological systems at the edge of chaos.
Claim: If the brain–tokamak pairing is real (both are modular-hierarchical systems producing Griffiths-phase criticality), then the avalanche scaling exponents in both systems should belong to the same universality class or show systematic correspondence. The modular-hierarchical architecture should impose the same statistical structure on avalanches regardless of substrate.
Status: Planned. Requires comparison of ELM size distributions from tokamak experiments with neural avalanche distributions from cortical recordings, using the same statistical framework (crackling-noise scaling relations). The comparison is non-trivial because the systems operate at different scales and the “avalanche” definition differs.
Claim: The predator-prey loop (gradient → turbulence → Reynolds stress → zonal flow → suppression → steeper gradient) satisfies the formal criteria for a Reflexively Autocatalytic and Food-generated (RAF) set, where the “food” is external heating and the “reactions” are the physical processes in the loop.
Status: Framework. The loop is qualitatively autocatalytic (each step produces the conditions for the next). Formal RAF analysis requires defining the reaction network precisely: what are the molecular species (physical fields), reactions (transport processes), catalysis relations (which process enables which), and food set (external drives). This analysis has not been performed.
If the loop is formally an RAF, it connects the tokamak to the cardiac calcium network (Tutorial 03) through shared RAF structure — two autocatalytic networks in completely different physical systems.
Connected through Griffiths phase architecture. Both systems have modular-hierarchical structure that produces robust criticality without parameter fine-tuning. The brain’s cortical modules create a Griffiths phase for neural avalanches. The tokamak’s nested transport barriers create a self-sustaining critical region for plasma avalanches. The mathematical mechanism is the same: heterogeneous modules stretch a critical point into a critical phase. Neurons at 37°C and hydrogen plasma at 150 million degrees, coupled by architecture.
Three order parameters on distinct but coupled topologies. The SOC avalanche dynamics operate on the radial pressure profile (1D chain of radial positions). The magnetic topology operates on the poloidal cross-section (2D field of magnetic flux). The predator-prey loop operates on the azimuthal structure (zonal flows are toroidally symmetric).
These are three different topologies coupled through the physics: the radial pressure drives the turbulence (SOC → loop), the turbulence modifies the magnetic topology (loop → topology), and the magnetic islands modify the radial gradient (topology → SOC). The coupling is cyclic.
The question: The noise-stabilized QTW is an emergent structure that exists only because all three transitions are coupled. Can a relational operator on the combined multi-topology field detect the QTW as a cross-component circulation signal — a persistent nonzero coupling between the three order parameters that has a specific spatial structure (the traveling wave)? If so, the QTW provides a concrete example of a multi-critical detection target: a structure invisible to any single-field measurement that appears only in the cross-field coupling.
Chapman, S. C., Dendy, R. O. & Hnat, B. (2001). A sandpile model with tokamak-like properties. Phys. Rev. Lett., 86, 2814.
Park, J.-K. et al. (2024). Direct observation of magnetic islands during ELM suppression. Nature Physics.
McMillan, B. F. et al. (2018). Self-sustaining quasi-traveling-wave at the tokamak edge. Phys. Plasmas.
Noise-stabilization of the edge quasi-traveling-wave (SNR > ~100%) is a finding of this research program; it has not yet been published or independently replicated.
Diamond, P. H. et al. (2005). Zonal flows in plasma — a review. Plasma Phys. Control. Fusion, 47, R35.
Wagner, F. et al. (1982). Regime of improved confinement and high beta in neutral-beam-heated divertor discharges of the ASDEX tokamak. Phys. Rev. Lett., 49, 1408.
Moretti, P. & Muñoz, M. A. (2013). Griffiths phases and the stretching of criticality in brain networks. Nature Communications, 4, 2521.
This series has examined six systems — from glass to plasma, from protein to brain — where multiple independent phase transitions converge. In each case, the coupling between transitions produces observables that no single transition contains: new universality classes (glass bicriticality, active matter η = 1.04), structural robustness (cardiac RAF, brain Griffiths phase), and emergent dynamics (tokamak QTW, protein folding kinetics).
The systems form three pairs, each connected by a shared physical mechanism: Gardner/jamming (Glass ↔ Protein), KT defect unbinding (Active Matter ↔ Cardiac), and Griffiths architecture (Brain ↔ Tokamak). These cross-system connections are not analogies. They are the same mathematics appearing in different substrates.
The question that opens from here: can multi-critical coupling be detected in advance? Single-critical detection — measuring the approach to one transition — provides early warning of one type of failure. Multi-critical detection — measuring the coupling between transitions — would detect the emergent structure that single-critical monitoring cannot see: the class-level vulnerability (cardiac), the geometry-dependent braiding (active matter), the noise-stabilized self-sustaining wave (tokamak), and the bicritical point where two glass transitions merge.
That is the research program these tutorials motivate.