betaFeedback

WorksPhysics and Astronomy

Collisionality scaling of the tokamak density limit

Abstract

The threshold at which density-limiting plasma instabilities occur in tokamak fusion reactors is not well understood. Here, we present the first large-scale empirical evidence of a common collisionality threshold present across multiple devices and confinement regimes. Machine-learning-guided analysis of a diverse experimental database identifies a scaling in edge effective collisionality (ν∗ν_*) that bears resemblance to the widely utilized Greenwald limit but yields a substantially lower false positive rate. This threshold decreases with dimensionless pressure (ββ), consistent with theoretical expectations for enhanced transport from electromagnetic effects on resistive ballooning mode turbulence. These results suggest that future burning plasma devices such as ITER and ARC could potentially achieve higher density, and thus fusion power, than previously expected.

Plasma density is a key parameter in nuclear fusion, as fusion power scales with the square of the density (Pfus∝n2P_{\mathrm{fus}} \propto n^2). In tokamak reactors, the maximum achievable density is limited by an anomalous collapse of the plasma edge temperature, which results in either a major loss of energy confinement or a magnetohydrodynamic (MHD) instability. The physical mechanism responsible for this “density limit” remains contested. In practice, the operational boundary is often estimated using the empirical Greenwald limit [1],

nG=Ipπa2,(1)n_G = \frac{I_p}{\pi a^2}, \tag*{(1)}

where nGn_G is the predicted maximum line-averaged density (in units of 1020 m−310^{20}\,\mathrm{m}^{-3}), IpI_p is the plasma current (units of MA), and aa is the minor radius (units of m). The Greenwald limit was originally identified as an approximate, empirical upper bound on the operational range of tokamak density, rather than as a precise instability predictor. For this reason, the scaling captures general trends across devices, but deviations of 50% or more are observed in some circumstances [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14]. This leads to uncertainty regarding the expected performance of future experiments, such as ITER [15] and ARC [16], and requirements for their mission-critical instability avoidance schemes [17, 18].

Here, we take a complementary approach: we explicitly optimize a scaling to discriminate density limit precursor states from stable operation.

Experimental [1, 19], numerical [20, 21], and theoretical studies [22, 23, 24] have suggested that transport processes play a central role in setting the density limit. However, past experimental validations are limited in scope, often focusing on individual devices or confinement regimes, and lack quantitative assessment of the false positive rate. Additionally, correlations with radiative phenomena such as MARFEs [25] and X-point radiators [26] make it difficult to isolate the underlying mechanism.

In this Letter, we provide the first large-scale, multi-device experimental evidence of how edge collisionality

plays a central role in setting the tokamak density limit across confinement regimes. We analyze a newly assembled database spanning several tokamaks: Alcator C-Mod, ASDEX Upgrade (AUG), DIII-D, JET, and TCV (illustrated in Figure 1). These experiments span a wide range in density (nˉe=0.1\bar{n}_{e} = 0.1–2.7⋅1020 m−32.7 \cdot10^{20}\ \mathrm{m}^{-3}), magnetic field (BT=1.4B_{T} = 1.4–7.6 T7.6\ \mathrm{T}), device size (R0=0.68R_{0} = 0.68–2.9 m2.9\ \mathrm{m}), plasma heating sources (ohmic, NBI, ICRF), triangularity (positive and negative), and wall material (carbon, molybdenum, tungsten, beryllium). There are 82 density limits in H-mode, 243 in L-mode, and 4,359 stable discharges gathered via the DEFUSE [27] and DisruptionPy [28] frameworks. As with previous analyses [14, 29], the precursor phase for L-mode discharges is defined as the points in time between the formation of the MARFE and the point at which the MARFE is destabilized. For H-mode density limits, the precursor phase is determined by a fixed time window before the H/L back-transition (100 ms for the larger devices—DIII-D, AUG, JET—and 30 ms for smaller devices with faster inherent time scales—TCV, C-Mod). We have excluded discharges where there are disruptions or macroscopic instabilities caused by other mechanisms or that cannot be clearly associated with a “pure” density limit. We also ignore discharge phases where diagnostic coverage is limited, including the dynamic ramp-up and ramp-down stages at the very beginning and end of the experiments.

Figure 1. Data points associated with the precursor phase of the density limit for five tokamaks plotted in terms of edge collisionality (ν∗\nu_*) and Greenwald fraction. Stable data points are represented by gray hexbins. Nearly all density limit time points lie in the collisional regime, ν∗>1\nu_* > 1.

We apply an interpretable machine learning workflow to identify power laws that accurately separate the stable and density limit plasma states. The algorithm searches over a standard set of dimensionless variables chosen to parameterize edge turbulence: edge effective collisionality ν∗=νeiq∗R0/(vthϵ3/2)\nu_{*} = \nu_{ei}q_{*}R_{0}/(v_{th}\epsilon^{3/2}), edge dimensionless pressure β=pedge/(B2/2μ0)\beta= p_{\mathrm{edge}}/(B^{2}/2\mu_{0}), edge safety factor q∗=2πμ0BTa2IpR01+κ22q_{*} = \frac{2\pi}{\mu_{0}}\frac{B_{T}a^{2}}{I_{p}R_{0}}\frac{1+\kappa^{2}}{2}, and normalized edge poloidal gyroradius ρ∗,pol=mDvth/(eBpola)\rho_{*,\mathrm{pol}} = m_{D}v_{th}/(eB_{\mathrm{pol}}a). We crudely assume a purely deuterium plasma as edge ZeffZ_{\mathrm{eff}} measurements are not available across the database. We also include two other dimensionless parameters related to radiation: normalized edge ionization energy T∗=Tedge/EionT_{*} = T_{\mathrm{edge}}/E_{\mathrm{ion}} and global radiated fraction frad=Prad/Pinf_{\mathrm{rad}} = P_{\mathrm{rad}}/P_{\mathrm{in}}. Edge quantities are evaluated using Thomson scattering measurements averaged over the radial interval 0.85≤ρpol<0.950.85 \le\rho_{\mathrm{pol}} < 0.95, as in previous analyses [14, 24].

The machine learning workflow used in this study is summarized in Figure 2. For a given set of variables, a Linear Support Vector Machine (LSVM) fits a linear classification boundary to the data. Notably, LSVMs have previously been used to identify a physically interpretable boundary for the computational fidelity of a class of simulations [30]; here, we apply this algorithm to find an accurate power law decision boundary. Model selection is performed using forward feature selection, and five-fold cross-validation is applied to provide error bars on predictive output. The primary metric for assessing model performance is the area under the receiver operating characteristic curve (AUC), a standard metric that quantifies classification performance across sensitivity levels. Predictions are evaluated on a per-discharge basis, as is standard in instability prediction studies, rather than on a per-time-point basis.

Figure 2. Candidate power-law solutions are explored via Linear Support Vector Machine (LSVM) classification and forward feature selection. Results are reported over five-fold train/test splits.

As a baseline, we evaluate the predictive capability of the Greenwald fraction. We find that the line-averaged Greenwald fraction achieves an AUC of 0.918±0.0120.918 \pm0.012, which is a moderate degree of prediction accuracy. We further find that the edge density variant of the Greenwald fraction is a less accurate descriptor of the data, with an AUC of 0.853±0.0150.853 \pm0.015. We do not find an improvement in prediction accuracy from re-optimizing the Greenwald fraction variables (nˉ\bar{n}, IpI_{p}, aa) by training an LSVM on the dataset. Fitting a heating power dependence likewise yields no significant improvement (LSVM-optimized scaling: nˉlimit∼nGPin0.19\bar{n}^{\mathrm{limit}} \sim nG P_{\mathrm{in}}^{0.19}, AUC =0.922±0.011= 0.922 \pm0.011).

The results of the LSVM-guided search are summarized in Figure 3. The first parameter selected by the model is the edge effective collisionality, ν∗\nu_{*}. Remarkably, this single parameter achieves predictive accuracy comparable to the line-averaged Greenwald fraction. The preference for collisionality is insensitive to the choice of input variables; including edge density as an independent variable does not improve prediction accuracy. Figure 1 illustrates that the density limit occurs in the collisional regime, as nearly all precursor data points satisfy ν∗>1\nu_{*} > 1. Following collisionality, the second parameter selected is β\beta, which significantly improves predictive accuracy to an AUC of 0.956±0.0060.956 \pm0.006. From the state space of these parameters shown in Figure 4, we make two observations: first, that the density limit occurs at the high-collisionality edge of the operating space, and second, that the collisionality threshold at which the density limit occurs decreases with β\beta. We also note that the L-mode and H-mode density limit events form a single continuous band in this state space. The third parameter selected is the edge safety factor, leading to an AUC of 0.962±0.0070.962 \pm0.007. As shown in Figure 3, additional parameters beyond this point provide no improvements in predictive accuracy, including the radiation-related variables (T∗T_{*} and fradf_{\mathrm{rad}}).

Predictive accuracy of candidate density limit scalings

Figure 3. Predictive accuracy of candidate density limit scalings, including the line-averaged Greenwald fraction (green, dashed), edge Greenwald fraction (green, dotted), scalings from the LSVM (blue), and the ν∗βq∗\nu_\ast\beta q_\ast scaling (orange). The largest improvement occurs when adding collisionality and normalized pressure, and performance saturates after three parameters. The simple physics-motivated metric ν∗βq∗\nu_\ast\beta q_\ast achieves comparable accuracy to the ML-optimized scalings.

The edge collisionality-pressure operating space of this dataset

Figure 4. The edge collisionality-pressure operating space of this dataset. Stable data time points are plotted in gray hexbins, with density limit precursor time points plotted on top (averaged per shot for visualization, marker shape same as in Figure 1). The density limit points cluster at the high-collisionality, high-β\beta edge of the operating space.

The central role of collisionality in setting the density limit is consistent with expectations from fluid models of the plasma edge and past experimental observations. Prior numerical analysis identified the onset of catastrophic transport in high-collisionality and high-β\beta regimes, hypothesized to be the origin of the density limit [20]. Later experimental studies found the density limit in both H- and L-mode correlated with enhanced transport at high edge collisionality [19, 31, 32, 33, 34]. Low edge temperature [35, 36, 37] and low electron adiabaticity [38, 39] are also reported when approaching the density limit, consistent with a collisionality-based mechanism.

Recent theoretical work proposed that strongly enhanced energy transport should occur when the characteristic wavenumber of resistive ballooning modes kRBM=αc/2CωB∼(ν∗q∗)−1/2k_{\mathrm{RBM}} = \sqrt{\alpha_c/2C}\sqrt{\omega_B} \sim(\nu_\ast q_\ast)^{-1/2} at the separatrix matches the electromagnetic wavenumber kEM=βe/μ∼β1/2k_{\mathrm{EM}} = \sqrt{\beta_e/\mu} \sim\beta^{1/2} [22], where C≡0.51μνei(q∗R)2/(csλ⊥)C \equiv0.51\mu\nu_{ei}(q_\ast R)^2/(c_s\lambda_\perp) is a normalized collision frequency and ωB≡2λ⊥/R\omega_B \equiv2\lambda_\perp/R is a normalized perpendicular gradient length. This mechanism has been proposed to explain the density limit in L-mode. This critical threshold can be expressed as

ν∗βq∗∣sep>1.4αcμ1/2ϵ3/2λ⊥R0≈0.3λ⊥R0,(2)\nu_\ast\beta q_\ast\big|_{\mathrm{sep}} > \frac{1.4\alpha_c\mu^{1/2}}{\epsilon^{3/2}}\sqrt{\frac{\lambda_\perp}{R_0}} \approx0.3\sqrt{\frac{\lambda_\perp}{R_0}}, \tag*{(2)}

where the right-hand side is approximated under typical conditions (ϵ≈1/3\epsilon\approx1/3, αc≈2.4\alpha_c \approx2.4 [22]). The left-hand side can also be written in terms of the turbulence parameter αt≡(1+τi)CωB∼ν∗q∗\alpha_t \equiv(1+\tau_i)C\omega_B \sim\nu_\ast q_\ast [22] of the drift-Alfvén (DALF) model. Beyond this critical threshold, electromagnetic effects on resistive ballooning modes can increase transport both directly, through magnetic flutter, and indirectly, as induction weakens the diamagnetic stabilization provided by the parallel electron response [40].

Motivated by this prediction, we test the simplified scaling ν∗βq∗\nu_\ast\beta q_\ast on this dataset. This yields an AUC of 0.965 ± 0.007, which is as accurate as the best LSVM-optimized scalings and a significant improvement upon the Greenwald fraction. When calibrated to a true positive rate of 95% (the ITER operational requirement for thermal quench prediction [41]), the false positive rate of ν∗βq∗\nu_\ast\beta q_\ast is 1.6× lower than the Greenwald fraction (16.6%, compared with 27.0%); at a true positive rate of 90%, this improvement increases to 2.6× (8.6%, compared with 22.2%). Applied to the L-mode and H-mode density limit data separately, ν∗βq∗\nu_\ast\beta q_\ast is likewise a better predictor than the Greenwald fraction and on par with the best LSVM-optimized scaling for each regime.

Figure 5 visually compares how ν∗βq∗\nu_\ast\beta q_\ast and the Greenwald fraction separate the density limit and stable data in this study. The Greenwald fraction provides a moderate degree of separation between the stable and density limit regimes; this is to be expected as this parameter was derived as a rule-of-thumb rather than a precise instability predictor. The physically motivated ν∗βq∗\nu_\ast\beta q_\ast scaling produces a substantially clearer distinction between the stable and unstable regimes. We also plot the expected range of enhanced transport under typical conditions approaching the density limit: ne,edge≈2nsepn_{e,\mathrm{edge}} \approx2 n_{\mathrm{sep}} [42], αc≈2.4\alpha_c \approx2.4 [22], and a conservative range of λ⊥/R0≈3⋅10−4\lambda_\perp/R_0 \approx3\cdot10^{-4}–3⋅10−23\cdot10^{-2}. There is agreement between the expected range where enhanced transport should occur, ν∗βq∗∣edge≳0.02\nu_\ast\beta q_\ast|_{\mathrm{edge}} \gtrsim0.02–0.20.2, and the observed range of density limit precursor states.

Distributions of stable, L-mode density limit precursor, and H-mode density limit precursor data points plotted in Greenwald fraction and $\nu_\ast\beta q_\ast$

Figure 5. Distribution of stable, L-mode density limit precursor, and H-mode density limit precursor data points plotted in both Greenwald fraction and ν∗βq∗\nu_\ast\beta q_\ast. The greater predictive power of ν∗βq∗\nu_\ast\beta q_\ast is reflected in the smaller overlap between stable and density limit points. Also shown is the expected regime (red) where enhanced transport should begin under typical conditions approaching the density limit: nedge≈2⋅nsepn_{\mathrm{edge}} \approx2 \cdot n_{\mathrm{sep}} [42], αc≈2.4\alpha_{\mathrm{c}} \approx2.4 [22], and λ⊥/R0≈3⋅10−4\lambda_\perp/R_0 \approx3 \cdot10^{-4}–3⋅10−23 \cdot10^{-2}.

The ν∗βq∗\nu_\ast\beta q_\ast threshold is compatible with decades of experimental observations of a density limit that generally scales like the Greenwald limit. Rewriting in terms of edge density, a modified Greenwald scaling emerges,

ν∗βq∗∼(nnG1T1/21+κ2a1/2ϵ1/4)2,(3)\nu_\ast\beta q_\ast\sim\left(\frac{n}{n_G}\frac{1}{T^{1/2}}\frac{1+\kappa^2}{a^{1/2}\epsilon^{1/4}}\right)^2, \tag*{(3)}

where nn is the edge density, κ\kappa is the plasma elongation, TT is the edge temperature, and ϵ\epsilon is the inverse aspect ratio. In this expression, the edge Greenwald fraction n/nGn/n_{\mathrm{G}} is a key parameter, with additional dependencies on temperature, machine size, and shaping. The edge temperature dependence (nlimit∼T1/2n_{\mathrm{limit}} \sim T^{1/2}) is the most consequential deviation from the standard Greenwald limit, as this leads to a density limit that grows with edge temperature and hence, in general, with heating power. In the parallel conduction-limited regime (Tsep∼PSOL2/7T_{\mathrm{sep}} \sim P_{\mathrm{SOL}}^{2/7}),

the threshold on edge density scales like PSOL1/7P_{\mathrm{SOL}}^{1/7}. More generally, the response of the edge temperature to total heating power can vary between plasma scenarios, which may explain why a range of heating power sensitivities has been reported [2, 3, 5, 9, 13, 29, 35, 43, 44, 45, 46] and why no single heating power scaling improves the prediction accuracy of the Greenwald fraction across the database. The additional shaping quantities vary modestly across operating scenarios, leading to little effect on the density limit threshold.

We note two deviations from the original theoretical predictions. First, we evaluate the quantity ν∗βq∗\nu_\ast\beta q_\ast utilizing density and temperature measurements in the region 0.85≤ρpol<0.950.85 \le\rho_{\mathrm{pol}} < 0.95, as opposed to at the separatrix as prescribed by theory. However, it has been shown that the edge and separatrix conditions approach each other near the density limit, even in H-mode [19, 42, 44]. Additionally, this definition significantly eases data collection, enabling the large statistics of this study and opening the door to real-time plasma control schemes [29]. The second departure from the original theory is that this scaling unexpectedly predicts the H-mode density limit as well. While theoretical treatments of the density limit generally prescribe mechanisms specific to L-mode or H-mode conditions [22, 24, 47, 48, 49, 50], past experimental observations of a Greenwald-limit-like threshold in both regimes and the shared threshold in ν∗βq∗\nu_\ast\beta q_\ast observed in this study are consistent with a common mechanism causing enhanced transport.

Projecting to future tokamaks, these results suggest robust stability to the density limit in burning plasmas. When organized in terms of the Greenwald fraction, the estimated operating point of ITER (nˉ/nG=0.85\bar{n}/n_{\mathrm{G}} = 0.85) is well within the range of observed H-mode and L-mode density limit cases in present experiments. In contrast, the high edge temperature of ITER’s intended operating regime (nedge≈8⋅1019 m−3n_{\mathrm{edge}} \approx8 \cdot10^{19}\ \mathrm{m}^{-3} and Tedge≈6 keVT_{\mathrm{edge}} \approx6\ \mathrm{keV} [51]) results in relatively low values of ν∗βq∗\nu_\ast\beta q_\ast, far from where density limit events occur in the database. This is true for several major tokamak reactor concepts examined, including ARC [16], MANTA [52], and EU-DEMO [53]; all have relatively high Greenwald fractions (nˉ/nG>0.85\bar{n}/n_{\mathrm{G}} > 0.85) but low edge ν∗βq∗\nu_\ast\beta q_\ast values (ν∗βq∗≪10−2\nu_\ast\beta q_\ast\ll10^{-2}), suggesting a substantial margin of stability to the density limit. While extrapolation to reactor regimes must be treated with caution, this suggests that higher densities, and therefore fusion power, may be possible.

Further analysis can enhance our understanding of this important operational limit. Although the ν∗βq∗\nu_\ast\beta q_\ast metric is an improved predictor of the density limit across the multi-machine dataset studied here, some misclassification between stable and density limit states remains, particularly in H-mode. This could be a result of other mechanisms [42, 54] playing a secondary role or due to sensitivities in the theoretical model that are not captured in the present analysis, such as isotope mass, gradient scale length, and ion dilution. Moreover, the edge conditions used here represent averages over a region, not precise measurements at the separatrix. While this definition facilitates a consistent multi-device comparison, as the location of the separatrix can be challenging to determine, future studies can probe this sensitivity.

In summary, we find that the onset of the tokamak density limit across a multi-device (Alcator C-Mod, AUG, DIII-D, JET, and TCV) experimental database is well described by a combination of three dimensionless parameters measured in the plasma edge: ν∗\nu_*, β\beta, and q∗q_*. A machine-learning-guided search of the dataset selects these three parameters to maximize the separation between stable and density limit precursor states. A simple scaling in these parameters, ν∗βq∗\nu_*\beta q_*, previously predicted to correspond to a threshold for enhanced transport [22], achieves higher predictive power than the Greenwald fraction in terms of AUC (0.965±0.007\pm0.007 vs. 0.918±0.012\pm0.012) and false positive rate (1.6×1.6 \times lower at a true positive rate of 95% and 2.6×2.6 \times lower at 90%). The ν∗βq∗\nu_*\beta q_* scaling is consistent with past observations of a Greenwald-like density limit, with a heating power sensitivity entering through the edge temperature. This work illustrates that precise measurements of the separatrix are not needed for improved density limit prediction, substantially easing requirements for real-time instability avoidance. The agreement between the empirical trends and theoretical predictions provides new experimental support for models in which the density limit arises from enhanced turbulent transport. These results suggest that the density limit may be more fundamentally understood as a threshold in edge collisionality and β\beta, rather than simply a critical density, with important implications for the achievable performance of future burning plasma devices.

Acknowledgments

The authors would like to thank D. Humphreys, A. Hubbard, J. Hughes, and N. Logan for providing insightful feedback on this study. We also thank J. Dunsmore and A. Miller for guidance in gathering C-Mod Thomson scattering data. This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Fusion Energy Sciences, using the DIII-D National Fusion Facility, a DOE Office of Science user facility, under Awards DE-FC02-04ER54698 and DE-SC0014264.

This work has been carried out within the framework of the EUROfusion Consortium, via the Euratom Research and Training Programme (Grant Agreement No 101052200 — EUROfusion) and funded by the Swiss State Secretariat for Education, Research, and Innovation (SERI). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union, the European Commission, or SERI. Neither the European Union nor the European Commission nor SERI can be held responsible for them.

Disclaimer: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.

References

  1. [1]M. Greenwald, J. L. Terry, S. M. Wolfe, S. Ejima, M. G. Bell, S. M. Kaye, and G. H. Neilson, A new look at density limits in tokamaks, Nuclear Fusion 28, 2199 (1988).DOI
  2. [2]M. G. Bell, G. L. Schmidt, P. C. Efthimion, B. Grek, R. A. Hulse, D. K. Owens, H. K. Park, A. T. Ramsey, J. F. Schivell, and G. Taylor, Attainment of high plasma densities in TFTR with injection of multiple deuterium pellets, Nuclear Fusion 32, 1585 (1992).DOI
  3. [3]A. Stabler, K. McCormick, V. Mertens, E. R. Muller, J. Neuhauser, H. Niedermeyer, K.-H. Steuer, H. Zohm, F. Dollinger, A. Eberhagen, et al., Density limit investigations on ASDEX, Nuclear Fusion 32, 1557 (1992).DOI
  4. [4]G. Pucella, O. Tudisco, M. L. Apicella, G. Apruzzese, G. Artaserse, F. Belli, W. Bin, L. Boncagni, A. Botrugno, P. Buratti, et al., Density limit experiments on FTU, Nuclear Fusion 53, 083002 (2013).DOI
  5. [5]P. C. De Vries, J. Rapp, F. C. Schuller, M. Z. Tokar, et al., Influence of recycling on the density limit in TEXTOR 94, Physical Review Letters 80, 3519 (1998).
  6. [6]T. H. Osborne, A. W. Leonard, M. A. Mahdavi, M. Chu, M. E. Fenstermacher, R. La Haye, G. McKee, T. W. Petrie, E. Doyle, G. Staebler, et al., Gas puff fueled H-mode discharges with good energy confinement above the Greenwald density limit on DIII-D, Physics of Plasmas 8, 2017 (2001).
  7. [7]M. Greenwald, Density limits in toroidal plasmas, Plasma Physics and Controlled Fusion 44, R27 (2002).DOI
  8. [8]R. Maingi and M. A. Mahdavi, Review of DIII-D H-mode density limit studies, Fusion Science and Technology 48, 1117 (2005).DOI
  9. [9]A. Huber, S. Brezinsek, M. Groth, P. C. De Vries, V. Riccardo, G. Van Rooij, G. Sergienko, G. Arnoux, A. Boboc, P. Bilkova, et al., Impact of the ITER-like wall on divertor detachment and on the density limit in the JET tokamak, Journal of Nuclear Materials 438, S139 (2013).
  10. [10]P. T. Lang, A. Burckhart, M. Bernert, L. Casali, R. Fis- cher, O. Kardaun, G. Kocsis, M. Maraschek, A. Mlynek, B. Plöckl, et al., ELM pacing and high-density operation using pellet injection in the ASDEX Upgrade all-metal-wall tokamak, Nuclear Fusion 54, 083009 (2014).
  11. [11]G. Pucella, O. D’Arcangelo, O. Tudisco, F. Belli, W. Bin, A. Botrugno, P. Buratti, G. Calabrò, B. Esposito, E. Giovannozzi, et al., Analytical relation between peripheral and central density limit on FTU, Plasma Physics and Controlled Fusion 59, 085011 (2017).DOI
  12. [12]S. Ding, A. M. Garofalo, H. Q. Wang, D. B. Weisberg, Z. Y. Li, X. J. Xian, D. Eldon, B. S. Victor, A. Marinoni, Q. M. Hu, et al., A high-density and high-confinement tokamak plasma regime for fusion energy, Nature 629, 555 (2024).DOI
  13. [13]O. Sauter, R. Hong, A. Marinoni, F. Scotti, P. H. Diamond, C. Paz-Soldan, D. Shiraki, K. E. Thome, M. A. Van Zeeland, H. Q. Wang, et al., Operation above the Greenwald density limit in high performance DIII-D negative triangularity discharges, Plasma Physics and Controlled Fusion 67, 075009 (2025).arxiv.org/abs/2503.10854
  14. [14]A. D. Maris, C. Rea, A. Pau, W. Hu, B. Xiao, R. Granetz, E. Marmar, EUROfusion Tokamak Exploitation Team, Alcator C-Mod Team, ASDEX Upgrade Team, et al., Correlation of the L-mode density limit with edge collisionality, Nuclear Fusion 65, 016051 (2025).arxiv.org/abs/2406.18442
  15. [15]K. Ikeda, Progress in the ITER physics basis, Nuclear Fusion 47, E01 (2007).DOI
  16. [16]J. C. Hillesheim, A. J. Creely, T. H. Eich, N. T. Howard, N. Leuthold, R. Sweeney, A. Leviness, A. O. Nelson, L. Nichols, R. A. Tinguely, et al., Overview of the physics basis for the ARC fusion power plant, Journal of Plasma Physics 92, E69 (2026).DOI
  17. [17]M. Maraschek, A. Gude, V. Igochine, H. Zohm, E. Alessi, M. Bernert, C. Cianfarani, S. Coda, B. Duval, B. Esposito, et al., Path-oriented early reaction to approaching disruptions in ASDEX Upgrade and TCV in view of the future needs for ITER and DEMO, Plasma Physics and Controlled Fusion 60, 014047 (2018).DOI
  18. [18]B. Sieglin, M. Maraschek, A. Gude, F. Klossek, F. Felici, M. Bernert, O. Kudlacek, A. Pau, W. Treutterer, ASDEX Upgrade Team, et al., Disruption avoidance and investigation of the H-Mode density limit in ASDEX Upgrade, Plasma Physics and Controlled Fusion 66, 025004 (2024).
  19. [19]B. LaBombard, R. L. Boivin, M. Greenwald, J. Hughes, B. Lipschultz, D. Mossessian, C. S. Pitcher, J. L. Terry, S. J. Zweben, and Alcator Group, Particle transport in the scrape-off layer and its relationship to discharge density limit in Alcator C-Mod, Physics of Plasmas 8, 2107 (2001).DOI
  20. [20]B. N. Rogers, J. F. Drake, and A. Zeiler, Phase space of tokamak edge turbulence, the L-H transition, and the formation of the edge pedestal, Physical Review Letters 81, 4396 (1998).
  21. [21]X. Q. Xu, W. M. Nevins, T. D. Rognlien, R. H. Bulmer, M. Q. Greenwald, A. Mahdavi, L. D. Pearlstein, and P. Snyder, Transitions of turbulence in plasma density limits, Physics of Plasmas 10, 1773 (2003).DOI
  22. [22]T. Eich, P. Manz, ASDEX Upgrade Team, et al., The separatrix operational space of ASDEX Upgrade due to interchange-drift-Alfvén turbulence, Nuclear Fusion 61, 086017 (2021).DOI
  23. [23]R. Singh and P. H. Diamond, Zonal shear layer collapse and the power scaling of the density limit: old LH wine in new bottles, Plasma Physics and Controlled Fusion 64, 084004 (2022).DOI
  24. [24]M. Giacomin, A. Pau, P. Ricci, O. Sauter, T. Eich, JET Contributors, ASDEX Upgrade Team, et al., First-principles density limit scaling in tokamaks based on edge turbulent transport and implications for ITER, Physical Review Letters 128, 185003 (2022).DOI
  25. [25]B. Lipschultz, B. LaBombard, E. S. Marmar, M. M. Pickrell, J. L. Terry, R. Watterson, and S. M. Wolfe, MARFE: an edge plasma phenomenon, Nuclear Fusion 24, 977 (1984).DOI
  26. [26]M. Bernert, F. Janky, B. Sieglin, A. Kallenbach, B. Lipschultz, F. Reimold, M. Wischmeier, M. Cavedon, P. David, M. G. Dunne, et al., X-point radiation, its control and an ELM suppressed radiating regime at the ASDEX Upgrade tokamak, Nuclear Fusion 61, 024001 (2021).DOI
  27. [27]A. Pau et al., A modern framework to support disruption studies: the EUROfusion disruption database, in Preprint: 29th IAEA Fusion Energy Conference (FEC 2023), London IAEA-CN-316-2057 (2023).
  28. [28]G. L. Trevisan, Y. Wei, A. M. Decker, J. Lorincz, S. L. Jackson, C. Rea, R. S. Granetz, et al., DisruptionPy: an open-source physics-based scientific framework for disruption analysis, Journal of Open Source Software 11, 9364 (2026).DOI
  29. [29]A. D. Maris, C. Rea, A. Pau, J. Barr, K. Erickson, L. Schmitz, Z. Yan, R. Granetz, E. Marmar, and EUROfusion Tokamak Exploitation Team, Real-time avoidance of the L-mode and H-mode density limit via machine-learned stability metrics, Nuclear Fusion 66, 056002 (2026).DOI
  30. [30]M. S. Murillo, M. Marciante, and L. G. Stanton, Machine learning discovery of computational model efficacy boundaries, Physical Review Letters 125, 085503 (2020).
  31. [31]W. Suttrop, M. Kaufmann, H. J. De Blank, B. Brüsehaber, K. Lackner, V. Mertens, H. Murmann, J. Neuhauser, F. Ryter, H. Salzmann, et al., Identification of plasma-edge-related operational regime boundaries and the effect of edge instability on confinement in ASDEX Upgrade, Plasma Physics and Controlled Fusion 39, 2051 (1997).DOI
  32. [32]H. J. Sun, R. J. Goldston, A. Huber, X. Q. Xu, J. Flanagan, D. C. McDonald, E. de la Luna, M. Maslov, J. R. Harrison, F. Militello, et al., The role of edge plasma parameters in H-mode density limit on the JET-ILW, Nuclear Fusion 61, 066009 (2021).DOI
  33. [33]P. Wang, G. H. Hu, L. Chen, R. Chen, L. Wang, G. S. Xu, J. Cheng, X. S. Jiang, X. D. Ye, L. L. Shan, et al., Evolution of edge turbulence and shear flow approaching the density limit in L-mode plasmas on EAST, Nuclear Fusion 66, 026014 (2026).DOI
  34. [34]M. A. Miller, J. W. Hughes, T. Eich, G. R. Tynan, P. Manz, A. E. Hubbard, B. LaBombard, and J. Dunsmore, Experimental evidence for increased particle fluxes due to a change in transport at the separatrix near density limits on Alcator C-Mod, Nuclear Fusion 66, 124003 (2026).arxiv.org/abs/2603.24512
  35. [35]T. W. Petrie, A. G. Kellman, and M. A. Mahdavi, Plasma density limits during Ohmic L mode and ELming H mode operation in DIII-D, Nuclear Fusion 33, 929 (1993).DOI
  36. [36]J. Rapp, P. C. De Vries, F. C. Schüller, W. Biel, R. Jaspers, H. R. Koslowski, A. Krämer-Flecken, A. Kreter, M. Lehnen, A. Pospieszczyk, et al., Density limits in TEXTOR-94 auxiliary heated discharges, Nu-clear Fusion 39, 765 (1999).
  37. [37]H. Yamada, H. Takenaga, T. Suzuki, T. Fujita, T. Takizuka, Y. Kamada, N. Asakura, T. Tuda, M. Takechi, G. Matsunaga, et al., Density limit in discharges with high internal inductance on JT-60U, Nuclear Fusion 47, 1418 (2007).DOI
  38. [38]R. Hong, G. R. Tynan, P. H. Diamond, L. Nie, D. Guo, T. Long, R. Ke, Y. Wu, B. Yuan, M. Xu, et al., Edge shear flows and particle transport near the density limit of the HL-2A tokamak, Nuclear Fusion 58, 016041 (2018).DOI
  39. [39]R. Hong, P. H. Diamond, O. Sauter, J. Chen, F. Khabarov, Z. Li, D. Liu, A. Marinoni, G. R. McKee, T. L. Rhodes, et al., Disentangling core and edge mechanisms of the density limit in DIII-D negative triangularity plasmas, Nuclear Fusion 66, 016027 (2026).DOI
  40. [40]P. Manz, T. Eich, O. Grover, How turbulence sets boundaries for tokamak operation, Reviews of Modern Plasma Physics 9, 5 (2025).DOI
  41. [41]P. C. De Vries, G. Pautasso, D. Humphreys, M. Lehnen, S. Maruyama, J. A. Snipes, A. Vergara, and L. Zabeo, Requirements for triggering the ITER disruption mitigation system, Fusion Science and Technology 69, 471 (2016).
  42. [42]P. Manz, T. Eich, O. Grover, ASDEX Upgrade Team, et al., The power dependence of the maximum achievable H-mode and (disruptive) L-mode separatrix density in ASDEX Upgrade, Nuclear Fusion 63, 076026 (2023).DOI
  43. [43]G. Duensing, First results of neutral beam heating on JET, Plasma Physics and Controlled Fusion 28, 1429 (1986).
  44. [44]M. Bernert, T. Eich, A. Kallenbach, D. Carralero, A. Huber, P. T. Lang, S. Potzel, F. Reimold, J. Schweinzer, E. Viezzer, et al., The H-mode density limit in the full tungsten ASDEX Upgrade tokamak, Plasma Physics and Controlled Fusion 57, 014038 (2015).DOI
  45. [45]A. Huber, M. Bernert, S. Brezinsek, A. V. Chankin, G. Sergienko, V. Huber, S. Wiesen, P. Abreu, M. N. A. Beurskens, A. Boboc, et al., Comparative H-mode density limit studies in JET and AUG, Nuclear Materials and Energy 12, 100 (2017).DOI
  46. [46]L. Schmitz et al., Accessing and maintaining robust H-mode in ITER pre-fusion power operation (PFPO) plasmas, in Preprint submitted: 2023 IAEA Fusion Energy Conf. (London) (2023).
  47. [47]P. Zanca, F. Sattin, D. F. Escande, and JET Contributors, A power-balance model of the density limit in fusion plasmas: application to the L-mode tokamak, Nuclear Fusion 59, 126011 (2019).DOI
  48. [48]R. Singh and P. H. Diamond, Bounds on edge shear layer persistence while approaching the density limit, Nuclear Fusion 61, 076009 (2021).DOI
  49. [49]A. O. Brown and R. J. Goldston, Generalization of the Heuristic Drift SOL model for finite collisionality and effect on flow shearing rate vs. interchange growth rate, Nuclear Materials and Energy 27, 101002 (2021).DOI
  50. [50]U. Stroth, M. Bernert, D. Brida, M. Cavedon, R. Dux, E. Huett, T. Lunt, O. Pan, M. Wischmeier, ASDEX Upgrade Team, et al., Model for access and stability of the X-point radiator and the threshold for MARFEs in tokamak plasmas, Nuclear Fusion 62, 076008 (2022).DOI
  51. [51]F. Koechl, R. Ambrosino, P. Belo, M. Cavinato, G. Corrigan, L. Garzotti, D. Harting, A. Kukushkin, A. Loarte, M. Mattei, et al., Evaluation of fuelling requirements for core density and divertor heat load control in non-stationary phases of the ITER DT 15 MA baseline scenario, Nuclear Fusion 60, 066015 (2020).DOI
  52. [52]G. Rutherford, H. S. Wilson, A. Saltzman, D. Arnold, J. L. Ball, S. Benjamin, R. Bielajew, N. de Boucaud, M. Calvo-Carrera, R. Chandra, et al., MANTA: a negative-triangularity NASEM-compliant fusion pilot plant, Plasma Physics and Controlled Fusion 66, 105006 (2024).arxiv.org/abs/2405.20243
  53. [53]G. Federici, C. Bachmann, L. Barucca, C. Baylard, W. Biel, L. V. Boccaccini, C. Bustreo, S. Ciattaglia, F. Cismondi, V. Corato, et al., Overview of the DEMO staged design approach in Europe, Nuclear Fusion 59, 066013 (2019).DOI
  54. [54]P. Manz, Density and radiation operational limits to H-mode sustainment, Nuclear Fusion 65, 092001 (2025).DOI

Paper details

vlnc:W_wq2eyyn3
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents