A preprint proposes a different way to design stellarators: require the drift surface selected by an orbit to close inside the plasma, even when it is misaligned with flux surfaces. The proposed iso-action criterion shifts attention away from keeping local branch actions constant and does not require quasisymmetry, omnigenity or piecewise omnigenity.
The document is arXiv preprint version 1, dated 20 August 2026; no journal or DOI is reported. It tests the idea in a solvable Treibich–Verdier B3 field model and in selected numerical orbit calculations, including five alpha-particle-optimized configurations.
A new target for the calculation
Here, “action” refers to the reduced bounce action used to describe an orbit. The framework asks whether the orbit-selected characteristic returns to its starting branch and section while its radial reach stays inside the confined region. That allows the action contour to miss the usual flux surfaces.
In the solvable B3 example, the actions on its daughter branches varied with field-line label, but their difference, J*, remained independent of both energy and field-line label. The field therefore was not piecewise omnigenous.
The B3 test
At model energy u = 4.0, the occupied-well action varied by 40% of its launch value. Even so, drift cancellation occurred over the full orbit rather than within each branch.
A direct fast–slow integration brought the bounce center back to its starting section after three B3 periods, with a displacement below 2 × 10−6. The paper describes this as numerical evidence from embedding the construction on a torus, not as a guiding-center orbit in a realized three-dimensional MHD equilibrium.
Checks on optimized configurations
The authors evaluated five alpha-particle-optimized configurations. Across them, peak-to-peak action modulation ranged from 0.9% to 56%, and none was omnigenous or piecewise omnigenous.
In these configurations, action contours closed inside the plasma and independently traced guiding-center orbits lay on those contours. In the strongest-modulation case, the residual was 0.38% of orbit width, while action conservation deviated by about 1%.
The paper calls its ΓW measure a time-limited estimate of how far an orbit reaches radially. It follows a bounce-averaged action-contour characteristic using the magnetic field alone and deterministic branch following; it does not include finite-orbit-width effects beyond guiding-center physics.
On the five alpha-optimized configurations, ΓW reproduced the traced radial reach within 1%.
Where the proxy met harder cases
In 250 coil perturbations of a precise quasihelical, or QH, configuration, ΓW had a reported Spearman correlation—a rank-based comparison of two orderings—of 0.656 with SIMPLE losses. The QS error reached 0.736, and median loss tracking across perturbation-amplitude bins reached 0.985, although the pooled trend was partly driven by perturbation amplitude.
A separate comparison used three matched marker cohorts, each containing 26 markers: promptly lost, confined and near-omnigenous. Their mean reaches were 0.701, 0.394 and 0.076, respectively; the promptly lost markers remained in a single well.
In the reactor-scale set, the proxy ranked the lossy quasihelical case highest but under-ranked lossy quasiaxisymmetric cases whose losses required finite-orbit-width information. The quasihelical case was outside the adiabatic regime over the quoted time window.
The boundary of the claim
The agreement applies to the tested adiabatic regime. ΓW is calculated from the field and action contour rather than full guiding-center tracing, and the perturbation trend was partly amplitude-driven.
The bounce-averaged framework does not cover the trapped-passing layer, bounce-precession resonances, rational-surface drift islands or finite-orbit-width loss. Separatrix crossing changes the adiabatic invariant, and the assumed adiabaticity can fail.
The paper also distinguishes its collisionless-closure target from exact omnigenity’s stronger low-collisionality 1/ν transport property. At small finite collisionality, tolerated action-contour misalignment comes with a small nonzero neoclassical flux, so thermal and energetic-particle measures are not interchangeable.
Taken together, the preprint presents a design principle tested in selected models and configurations, with further work needed in the omitted regimes.
Funding
The authors report support from the Department of Energy under Award No. DE-SC0024548.
Paper data and sources
Original title: Symmetry-agnostic stellarators for collisionless confinement
Authors: W. Sengupta, A. Bhattacharjee, S. Buller
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text