Massive Gas Injection in Tokamak Disruptions

• Massive Gas Injection (MGI) is a disruption-mitigation technique that rapidly injects neutral gas into plasmas to induce controlled radiative cooling and thermal quench.
• MGI alters the pre-thermal (pre-TQ) phase by modifying edge pressure, resistivity, and triggering MHD instabilities such as tearing modes.
• The method’s effectiveness relies on injection species, quantity, and localization, which together influence impurity penetration, current-profile relaxation, and runaway electron dynamics.

Searching arXiv for the cited MGI papers to ground the article in current records. Search query: "Massive Gas Injection tokamak disruption mitigation" Massive Gas Injection (MGI) is a disruption-mitigation technique in tokamaks in which a large quantity of neutral gas is rapidly injected into the plasma to deliberately radiate energy and terminate discharges in a controlled way. In the literature considered here, MGI is treated not only as a radiative shutdown actuator, but also as a source of poloidally localized pressure asymmetries at the edge, a trigger for low-$n$ MHD activity and impurity penetration, a control knob for thermal-quench and current-quench dynamics, and a strong modifier of runaway-electron generation. Closely related work also uses the broader term massive material injection (MMI), which includes gas injection and pellet-based variants such as SPI (Aydemir et al., 2019, Zafar et al., 2021, Zeng et al., 2020, Vallhagen et al., 2020, Linder et al., 2020).

1. Operational role and studied implementations

MGI is introduced in these studies as a technique intended to spread the thermal energy loss more uniformly and radiatively over the vessel, control the time scales of the thermal quench (TQ) and current quench (CQ), and increase plasma density and collisionality so as to reduce or prevent runaway-electron generation. The injected species differ by device and objective: helium MGI is studied on EAST as a low-$Z$ mitigation case; argon MGI is used in ASDEX Upgrade and in J-TEXT-like simulations to produce strongly radiative disruptions; ASDEX Upgrade datasets also include Ne, Kr, and Ar+$\mathrm{D}_2$ cases; ITER-oriented and SPARC-oriented studies examine deuterium–noble-gas mixtures and neon-only scenarios (Zafar et al., 2021, Zeng et al., 2020, Heinrich et al., 2024, Vallhagen et al., 2020, Datta et al., 5 Dec 2025).

Study or device	Gas or scenario	Reported emphasis
EAST	He MGI	Injection-level dependence and critical impurity level
J-TEXT-like / NIMROD	Ar MGI	$2/1$ tearing, current sheet, cold bubble, TQ onset
ASDEX Upgrade	Ar MGI	Self-consistent TQ-CQ-RE evolution
ASDEX Upgrade	Predominantly Ar MGI	CQ Alfvénic activity and GAE identification
ITER DT phase	D + Ne or D + Ar MMI	Coupled temperature, electric field, and avalanche dynamics
SPARC	Ne-only and $\mathrm{D}_2+$Ne MGI	Coupled 3D MHD and runaway-electron evolution

A common distinction across the papers is between the pre-TQ phase, the TQ, and the CQ. In the MGI context, the pre-TQ phase is the interval in which injected material ionizes, cools the edge, modifies resistivity and current profiles, and amplifies MHD activity. The TQ is the interval of rapid thermal collapse, and the CQ is the subsequent decay of plasma current. In several studies, the eventual disruption path depends sensitively on the amount of injected material, the injected species, the location of the perturbation, and the pre-existing $q$-profile (Zafar et al., 2021, Zeng et al., 2022).

2. Edge asymmetry, force balance, and geometry-dependent flows

One MHD-based formulation treats MGI as a particular way of creating strong, localized poloidal pressure asymmetries at the tokamak edge. In that framework, the pressure is written as

$p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$

with $\delta p$ localized in $\theta$, and the central conceptual statement is that an MHD equilibrium in which the plasma pressure is not a flux function can be maintained only by contributions from mass flows. The analysis is explicitly about the effects of a given asymmetry rather than its microscopic cause (Aydemir et al., 2019).

The equilibrium flow is decomposed into a parallel flow and rigid toroidal rotation,

$$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$

and the radial electric field follows from ideal MHD as

$Z$0

In this picture, a localized pressure bump or pressure hole on a flux surface creates a tangential pressure gradient along the poloidal direction, and toroidal geometry converts that tangential force into a net toroidal torque. For narrow perturbations, the sign of the flux-surface-averaged torque depends only on the poloidal location and on the sign of $Z$1, and is independent of toroidal field or plasma current (Aydemir et al., 2019).

This geometrical dependence is used to explain several edge phenomena. Positive $Z$2 in the lower half-plane drives a negative edge $Z$3 well just inside the separatrix and is described as favorable for L-H access, while positive $Z$4 in the upper half-plane produces a positive edge $Z$5 just inside the separatrix and is described as unfavorable for L-H. By symmetry, negative $Z$6 relevant to MGI produces flows opposite to those of positive $Z$7 at the same location. The same argument is extended to fueling-port placement: if ITER fueling ports above the midplane generate a positive pressure asymmetry at the edge, they are described as misplaced and liable to increase input power requirements (Aydemir et al., 2019).

For MGI and SPI specifically, the paper argues that rapid radiative cooling produces localized negative pressure perturbations, $Z$8, despite the increase in particle density. This gives a geometry-driven explanation for the observed upper–lower asymmetry of poloidal flow directions during disruption mitigation. A reported consequence is that the poloidal flow direction after upper or lower low-field-side injection does not flip with toroidal-field reversal, because the torque is geometric rather than $Z$9-driven (Aydemir et al., 2019).

3. Impurity penetration, tearing modes, and thermal-quench initiation

A recurring theme in MGI research is that injected impurities are initially localized at the edge and then diffuse and are convected toward the core. In extended resistive-MHD simulations with impurity radiation physics, the edge is cooled first, resistivity increases where the plasma is cooled, and low-$\mathrm{D}_2$0 tearing activity grows as the current profile contracts and rational surfaces become more unstable. In helium MGI simulations for EAST, a $\mathrm{D}_2$1 surface is present in the plasma, and the evolving profiles allow $\mathrm{D}_2$2, $\mathrm{D}_2$3, and $\mathrm{D}_2$4 tearing activity; the identified critical injection level is

$\mathrm{D}_2$5

Above this level, a single MHD activity is able to induce a complete core temperature collapse; below it, a series of multiple minor disruptions occur before the complete TQ (Zafar et al., 2021).

In the high-injection EAST case, the pre-TQ lasts about $\mathrm{D}_2$6, radiated power peaks at $\mathrm{D}_2$7, the core temperature starts to collapse at $\mathrm{D}_2$8, and the TQ duration in the illustrative high-injection case is $\mathrm{D}_2$9. More than $2/1$0 of thermal energy is lost already in the pre-TQ phase via edge cooling, and by the end of TQ the total radiated energy is $2/1$1, about $2/1$2 of lost thermal energy. In the low-injection case $2/1$3, the overall collapse of core temperature and stored energy takes $2/1$4, and the radiated energy is $2/1$5 (Zafar et al., 2021).

A more localized mechanism is described in simulations of argon MGI. There, the impurity cold front reaches the $2/1$6 and then $2/1$7 rational surfaces, producing $2/1$8 and $2/1$9 islands successively. At the impurity–plasma interface, a local thin current sheet forms due to an enhanced local pressure gradient and moves inward following the gas cold front. The dominant $\mathrm{D}_2+$0 mode is interpreted as an internal tearing mode rather than an unstable external $\mathrm{D}_2+$1 kink. Following the growth of the $\mathrm{D}_2+$2 tearing mode, impurity penetration into the core region inside the $\mathrm{D}_2+$3 surface gives rise to the formation of the cold bubble temperature structure and initiates the final TQ (Zeng et al., 2020).

The cold bubble is a localized, coherent low-temperature structure whose dominant $\mathrm{D}_2+$4 component in temperature becomes $\mathrm{D}_2+$5 at the onset of TQ, while the magnetic perturbation remains dominated by $\mathrm{D}_2+$6. The causal chain established in that work is

$\mathrm{D}_2+$7

A subdominant $\mathrm{D}_2+$8 mode does not by itself cause cold-bubble formation, although the manner of the preceding impurity penetration depends on whether the $\mathrm{D}_2+$9 mode is kink-tearing or quasi-interchange (Zeng et al., 2020).

A self-consistent 1.5D transport study of argon MGI in ASDEX Upgrade reaches a related conclusion by a different route. There, neutral Ar is applied as a source just outside the LCFS, neoclassical transport alone is too slow to reproduce the observed TQ timing, and additional transient impurity transport is prescribed inside the $q$0 surface to represent rapid MHD-driven mixing. With

$q$1

the simulation reproduces the measured density rise, current decay rate, and final runaway current, and the thermal collapse is induced through strong impurity radiation once impurity ions penetrate rapidly into the core (Linder et al., 2020).

4. Current-profile relaxation, current spikes, and current-quench eigenmodes

MGI-driven disruptions often exhibit a plasma current spike at the end of the TQ and the beginning of the CQ. In systematic NIMROD + KPRAD simulations, the formation of this spike is found to strongly correlate with the onset of $q$2 kink-tearing reconnection. The relevant flux-conservation argument is written as

$q$3

where $q$4 is the poloidal flux and $q$5 is the internal inductance. During the pre-TQ phase, radiation cooling contracts the current profile, so $q$6 increases while $q$7 decreases slightly. At TQ, when the $q$8 mode reconnects strongly, the current profile broadens rapidly, $q$9 drops abruptly, and partial conservation of $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$0 drives a transient increase in $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$1, producing the spike (Zeng et al., 2022).

This mechanism depends strongly on the central safety factor. When $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$2, a resonant $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$3 surface exists, the core mode is $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$4 kink-tearing, and the plasma undergoes a single major disruption with a single, strong current spike. When $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$5, the core mode becomes quasi-interchange-like, the disruption becomes a sequence of successive minor disruptions, the onset of complete TQ is delayed, and the current trace shows two or more smaller spikes. In a species scan at fixed $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$6, He and Ar injections show a clear $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$7 tearing island and a pronounced current spike, whereas Ne yields a quasi-interchange-like core mode and no apparent spike (Zeng et al., 2022).

During the CQ itself, ASDEX Upgrade has developed multiple MGI scenarios to study runaway-electron dynamics, and Alfvénic activity is observed in the $p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$8–$p(\psi,\theta)=p_0(\psi)+\delta p(\psi,\theta),$9 range. A mode tracing algorithm based on Fourier spectrograms was applied to $\delta p$0 discharges, the modes were identified as global Alfvén eigenmodes by linear gyrokinetic MHD simulations, and changes in the Alfvén continuum during the quench were proposed as explanation for the strong frequency sweep. The modes appear only during the CQ, after the initial current spike, and vanish before or at the onset of the runaway plateau. A systematic statistical analysis shows no significant connection of the mode characteristics to the dynamics of the subsequent runaway-electron beams (Heinrich et al., 2024).

The AUG study is important because it excludes a common assumption that any prominent CQ activity necessarily mitigates runaways. In that database, the appearance and amplitude of the CQ modes do not seem to affect the potential subsequent runaway beam, and similar Alfvénic activity is also observed in natural disruptions with no runaway beam forming. This suggests that the CQ modes are generic features of the current quench rather than a robust RE-control mechanism in the explored regime (Heinrich et al., 2024).

5. Runaway electrons under MGI and MMI

MGI and the broader class MMI can mitigate and drive runaway-electron formation simultaneously. In the ITER DT phase, a 1D self-consistent model of deuterium–noble-gas mixtures shows substantial runaway currents unless the current-quench time is intolerably long. The mechanism identified there is that cooling associated with the injected material leads to high induced electric fields and, in combination with significant recombination of hydrogen isotopes, leads to large avalanche generation. The avalanche source is enhanced in partially ionized plasmas through the factor $\delta p$1, and accurate modeling of temperature evolution based on energy balance is described as crucial for quantitative predictions (Vallhagen et al., 2020).

The same study scans injected D + Ne and D + Ar concentrations and finds large runaway currents over wide parameter ranges. For pure Ne injection with partial screening, $\delta p$2 is obtained for $\delta p$3. Mixed cases show a non-monotonic dependence on deuterium content: moderate deuterium can suppress runaway growth by keeping the plasma at higher temperature, whereas very large deuterium content can cool the outer plasma to $\delta p$4, strongly recombine hydrogen, increase $\delta p$5, and regenerate strong off-axis avalanche (Vallhagen et al., 2020).

A self-consistent ASTRA–STRAHL study of argon MGI in ASDEX Upgrade likewise finds that MGI can facilitate rather than suppress a substantial runaway beam. In that discharge, the final RE current is $\delta p$6, about $\delta p$7 below the measured $\delta p$8, and approximately $\delta p$9 of the post-CQ RE current is avalanche-generated. Using classical Connor–Hastie and Rosenbluth–Putvinski models instead of partial-screening models gives $\theta$0, or $\theta$1 above experiment, and produces RE current too early relative to the hard-X-ray signal. In this case, partial screening is essential to reproduce both the magnitude and timing of the RE beam (Linder et al., 2020).

Coupled 3D extended-MHD and RE-fluid simulations for SPARC sharpen the dependence on injected material. Ne-only injection produces large RE plateaus: $\theta$2 are obtained with Ne-only injection $\theta$3. Combined $\theta$4Ne injection $\theta$5 produces a lower RE current $\theta$6, and with $\theta$7Ne injection a post-thermal-quench “cold” VDE terminates the RE beam, preventing a steady plateau. The study also reports that 3D MHD can initially increase RE generation through local $\theta$8 enhancement, then cause stochastic losses, and later allow RE confinement and plateau formation through re-healing of flux surfaces (Datta et al., 5 Dec 2025).

Taken together, these studies show that MGI is not equivalent to automatic RE suppression. Neon-only scenarios can produce large RE plateaus in SPARC; D + noble-gas mixtures in ITER can still yield several mega-amperes of runaway current unless the CQ is very long; and experimentally benchmarked argon MGI in ASDEX Upgrade produces an avalanche-dominated RE beam rather than suppressing it (Datta et al., 5 Dec 2025, Vallhagen et al., 2020, Linder et al., 2020).

6. Design implications, misconceptions, and modeling limits

The reviewed literature identifies several control parameters as decisive. Injection level is one: on EAST, $\theta$9 He atoms separates a single large MHD event and one-step core temperature collapse from a sequence of multiple minor disruptions. Injection composition is another: helium is a relatively weak radiator, and even high He injection yields only $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$0 radiated fraction in the illustrated EAST case, whereas higher-$$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$1 gases radiate more strongly but can also accelerate TQ and alter core-mode structure. Injection location is a third: edge pressure holes above or below the midplane change the sign of MGI-driven poloidal flows and the early evolution of the radiating front, while upper versus lower injection in the low-field side changes whether radiation moves over the top or remains weak or downward near the X-point (Zafar et al., 2021, Aydemir et al., 2019).

A common misconception is that MGI is only a cooling pulse. Several papers instead treat it as a fully coupled perturbation of edge force balance, resistivity, current profile, impurity transport, and MHD stability. Another misconception is that any MGI-induced CQ fluctuation necessarily assists RE mitigation; the ASDEX Upgrade statistical study of CQ Alfvénic activity finds no significant connection between mode characteristics and subsequent RE-beam dynamics. A related misconception is that more injected material is always better. The EAST helium scan, the ITER DT mixture scans, and the SPARC Ne versus $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$2Ne comparison all show that larger or more strongly radiating injections can move the discharge into qualitatively different regimes rather than monotonically improving mitigation (Heinrich et al., 2024, Zafar et al., 2021, Vallhagen et al., 2020, Datta et al., 5 Dec 2025).

The limitations of current modeling are also explicit. Several studies use single-fluid resistive MHD, isotropic or simplified viscosity, prescribed or pre-distributed impurity sources, forced or axisymmetric TQ models, or ad hoc transport coefficients intended to represent unresolved MHD mixing. Some do not include hot-tail seed formation, runaway-electron generation and loss, two-fluid effects, detailed wall response, or full 3D valve-by-valve gas-plume dynamics. In the edge-asymmetry model, the perturbation $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$3 is prescribed and stationary, and the feedback of flows on the pressure asymmetry is not computed. In the SPARC study, the TQ is imposed axisymmetrically and 3D MHD is followed during the CQ; in the ITER DT study, hot-tail generation is omitted and the analysis is 1D; in the ASDEX Upgrade transport study, rapid mixing inside $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$4 is represented through prescribed transport coefficients rather than calculated directly (Aydemir et al., 2019, Datta et al., 5 Dec 2025, Vallhagen et al., 2020, Linder et al., 2020).

These constraints do not remove the central result shared by the cited works: MGI is a strongly nonlinear actuator whose effects depend on species, quantity, geometry, and equilibrium state. In present formulations it can generate edge shear flows and $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$5, trigger $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$6, $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$7, and $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$8 activity, produce cold-bubble-mediated TQ onset, reshape current-spike dynamics through $$\mathbf{u}=\frac{\Phi(\psi)}{\rho_m}\mathbf{B}+\Omega(\psi)R^2\nabla\zeta,$$9 reconnection, and either suppress or amplify runaway-electron formation depending on how density, temperature, resistivity, and impurity charge-state evolution co-evolve (Aydemir et al., 2019, Zeng et al., 2020, Zeng et al., 2022, Datta et al., 5 Dec 2025).