EPFL theory predicts the power a tokamak needs to enter H-mode
Category: Simulations, Tokamak


TCV is one of the machines the EPFL L-H power threshold scaling was tested against, shown here with a diverted plasma
(Image courtesy of EPFL Swiss Plasma Center)
Researchers at EPFL’s Swiss Plasma Center have derived a physics-based scaling for the L-H transition power threshold, with exponents that come from the theory and not from regression against machine data. The model also predicts the density minimum and the roughly twofold power penalty for the unfavourable field direction at high density. Applied to ITER and SPARC in a Physical Review Letters paper published this month, it returns minimum H-mode powers with explicit error bars.
Drift waves generate the flow behind the L-H transition
In an L-H transition, the plasma switches from low to high confinement mode, and the power threshold is the heating power needed to trigger it. The group ran power-ramp simulations with GBS, a global flux-driven code that solves the drift-reduced Braginskii equations, in a diverted lower-single-null geometry. Starting from quasi-steady L-mode turbulence, they raised core heating by a factor of about 1.4 at constant density source, split equally between ions and electrons. In the favourable configuration, where the ion ∇B drift points toward the X-point, a fast transport bifurcation followed, and the energy confinement time rose by about a factor of two before settling roughly 1.5 times above its starting value.

The favourable run made the L-H transition on the matched ramp, while the unfavourable run needed roughly twice the L-mode input power
(All images courtesy of B. De Lucca et al., EPFL Swiss Plasma Center)
The unfavourable configuration at matched power showed neither better confinement nor a significant change in the radial electric field. It transitioned only when the team repeated the ramp at roughly twice the L-mode input power, and the electric field well then deepened much as it had in the favourable case.
Electromagnetic effects proved essential in these simulations, because the same ramps at zero beta produced no transition. The authors identify two mechanisms, linear stabilisation of drift waves through an enhanced electron adiabatic response and nonlinear generation of E×B shear by finite-beta drift waves. Removing the flux-surface-averaged E×B flow from the model immediately restored high transport, which confirms its role in forming the barrier.
Finite collisionality sets the field-direction asymmetry
To explain why one field direction needs less power than the other, the group built a quasilinear model of the turbulent momentum flux. It yields a critical value of βν, a parameter that combines plasma beta with collisionality, near 0.07 in the favourable direction and varying by about 15 percent across the experimentally relevant range of the other parameters. The unfavourable direction needs roughly twice that value, which gives an asymmetry ratio of about two.
In the simulations, the favourable case crossed into the suppressed-transport regime at a βν of about 0.09 at matched power, while the opposite configuration stayed put and transitioned at about 0.13 only once the team added more power, a pattern the authors say agrees well with the quasilinear estimates. Reversing the plasma current, by contrast, leaves the threshold unchanged, as in experiments, and the GBS results confirm it.
Four conditions must hold together for the asymmetry to appear, namely up-down geometric asymmetry, finite E×B shear, finite magnetic shear and finite collisionality. The authors say earlier time-independent Reynolds-stress analyses identify the first three conditions but treat the perpendicular wavenumber as constant in time, so they miss collisionality. One analysis in that group is the gyrokinetic ASDEX Upgrade study published in Physical Review Letters in July, which also traces the deeper electric field well in the favourable direction to turbulence-driven poloidal flows, so both studies tie the deeper well to turbulence-driven mean flow.
The derived L-H power threshold tracks the data
Combining the transition criteria with a mixing-length estimate of turbulent heat flux and parallel electron conduction losses gives closed-form expressions for the threshold power. In the high-density branch it scales as aR^1.1 B^0.6 n^1.05, where a and R are the minor and major radii, B is the field strength and n the density. Every exponent comes from the derivation, while experiments fix only the order-one constants.

The 2008 empirical scaling reaches an R² of 0.83 on the ITPA high-density database., against 0.87 for the theory-derived scaling
That form lines up with the revised empirical scaling for ITER-like-wall devices, which runs roughly as aRB^0.58 n^1.08. The older 2008 ITPA scaling rests mainly on carbon-wall machines, and the revision grew out of like-for-like thresholds falling by about 30 percent on ASDEX Upgrade and JET after the first wall changed material. Against the ITPA database of high-density favourable-configuration transitions, the fitted theory gives an R² of 0.87, while the 2008 empirical scaling gives 0.83, and combining the asymmetry ratio with the scaling yields a power penalty factor of about two for the unfavourable direction in the high-density branch.
Below the density minimum, where inertial drift waves dominate, the threshold rises as density falls, following n^-2.25 in the conduction-limited regime and a shallower n^-0.75 once sheath losses take over at higher power. The sheath-limited form describes ASDEX Upgrade data at three plasma currents, and the high- and low-density expressions both agree well with TCV. Because the threshold climbs with density on one branch and falls on the other, the two meet at a minimum, and the model gives that minimum power as rising with field strength as B^1.75 and falling with ion mass as M^-1.25, the latter closely matching the empirical isotope exponent.
Present-device fits set the ITER and SPARC error bars
Extrapolated from present devices to ITER’s full-field scenario of 15 MA and 5.3 T, with an effective charge of 1.5 and a mass number of 2.5, the minimum-power scaling gives a density minimum of (5.8 ± 1.0) × 10¹⁹ m⁻³ and a minimum power of 43 ± 11 MW, with uncertainties set by the scalings’ root-mean-square errors. The group concludes that ITER should reach H-mode within its baseline 73 MW of auxiliary heating.
Where ITER’s 43 ± 11 MW sits inside that 73 MW, the same method puts SPARC’s full-field scenario of 8.7 MA and 12.2 T at (2.6 ± 0.4) × 10²⁰ m⁻³ and 26 ± 7 MW. The authors describe that as leaving little margin over the planned 25 MW heating capacity, a concern they trace to Hughes and colleagues in 2025 and rederive from theory. An earlier SPARC projection from MIT-led authors in 2020 found a clear window for H-mode access at full field with 25 MW of design auxiliary power, and noted that additional alpha heating is likely needed to sustain it.

Predicted minimum power and density track Delabie and colleagues’ reference values (R² 0.84 and 0.88), with ITER and SPARC at the high-power end
In TCV the measured minimum power exceeds the prediction, which the authors plausibly attribute to unsubtracted radiative losses. The isotope dependence of the high-density threshold, M^-0.55, is weaker than the observed 1/M trend, and the low-density branch rests on quasilinear theory because GBS did not simulate it.
A 2025 conference abstract from MIT, Oak Ridge and Commonwealth Fusion Systems authors says SPARC will sit at the extremes of field and density covered by existing devices, so the extrapolation uncertainties will need threshold measurements on SPARC itself. The EPFL projection is likewise calibrated on present devices, and its 26 ± 7 MW carries error bars taken from the scalings’ fit errors.
This article is based on:
B. De Lucca, P. Ricci, B. Labit, D. Mancini, L. N. Stenger and Z. Tecchiolli, “Theory for the Power Threshold of the L-H Transition in Tokamaks,” Physical Review Letters 137, 205101 (2026), published 13 November 2026. DOI 10.1103/rw25-wy15. Figures checked against the authors’ arXiv manuscript, arXiv:2605.00624v3.