betaFeedback

WorksPhysics and Astronomy

Exact analytic anisotropic three-dimensional equilibria with nested magnetic surfaces, confined drift orbits and without spatial symmetry

Abstract

Exact toroidal fluid and kinetic equilibria with nested magnetic surfaces and no continuous symmetry are of great interest in laboratory, astrophysical and space plasmas. Their existence and stability have been open in physics and mathematics for decades. We construct such anisotropic-pressure equilibria by applying the Clemente transformation to exact vacuum fields. They connect smoothly to these fields on a fixed regular domain and preserve a regular magnetic axis, perfectly nested surfaces and spatially varying rotational transform. To our knowledge, the sheared configurations are the first exact anisotropic equilibria combining these properties with positive pressures and a positive firehose margin. The fixed-boundary Chew–Goldberger–Low (CGL) quadratic energy splits into a nonnegative vacuum-field term and a local pressure term, nonnegative for p_perp/p_parallel <= (3+\sqrt{21})/2 ~ 3.79. For p_parallel >= p_perp each configuration minimizes the finite-amplitude CGL energy over smooth boundary-fixed ideal material deformations. For the current-free configurations, physical guiding-center distributions produce these pressures within the same bound, with no free energy for magnetic-moment-conserving electrostatic perturbations. Trapped particles inherit the vacuum field's second adiabatic invariant at a rescaled magnetic moment, so local closed action contours carry over. The equilibrium code DESC reproduces them from the exact solution. Kinetic stability for surface-dependent rescalings and against general nonlinear electromagnetic perturbations remains open.

Magnetic confinement fusion aims to hold a plasma hotter than the core of the Sun inside a toroidal magnetic field. The field lines wind around nested toroidal surfaces, and charged particles stream rapidly along them. Two concepts lead the field. The tokamak is axisymmetric and drives a large current through the plasma to twist its field lines. The stellarator twists them with external coils alone, at the cost of a fully three-dimensional (3D) geometry [1]. Symmetry has guided the design of both. In a tokamak, axisymmetry gives every particle a conserved canonical angular momentum, which keeps its slow drifts close to a magnetic surface on which field lines lie. A generic stellarator lacks this protection, and its trapped particles drift out much faster. The way forward was quasisymmetry, which is a type of hidden symmetry: if the field strength, though not the field’s shape, is symmetric in suitable coordinates, guiding centres are confined as well as in a tokamak [2, 3, 4]. A weaker principle, omnigeneity, asks only that the bounce-averaged drift across the surfaces vanish [5, 6], and reduced neoclassical transport has now been demonstrated in an optimized stellarator [7]. Here we ask whether a spatial symmetry is needed at all. We study fields without continuous symmetry whose trapped particles nevertheless drift on local closed contours of the second adiabatic invariant. Such fields trade an exact symmetry of the field for a property of the particle orbits themselves.

The existence of smooth 3D equilibria with nested flux surfaces has remained a long-standing problem in plasma as well as mathematical physics. Grad argued that, with scalar pressure, such equilibria generally require a continuous symmetry [8]. The issue is more pronounced for quasisymmetric fields, which confine guiding centers as well as an axisymmetric field does. Garren and Boozer showed that the expansion of a quasisymmetric equilibrium about its magnetic axis is overdetermined [9, 10]. However, overdetermined equations can have solutions. Exact equilibria that sharpen these questions have recently been constructed. They include 3D equilibria with scalar pressure and nested surfaces [11, 12], some without stellarator symmetry [13] or with an arbitrary number of field periods [14], and vacuum fields with nested tori and no continuous symmetry [15].

Real stellarator plasmas need not have scalar pressure. Neutral-beam injection and ion-cyclotron heating create fast-ion populations with very different parallel and perpendicular pressures, and the guiding-center description of a collisionless plasma gives a gyrotropic pressure tensor [16]. 3D equilibrium codes with anisotropic pressure have been developed and deployed [17, 18, 19]. Anisotropy also bears directly on the equilibrium existence question. Rodríguez and Bhattacharjee [20] imposed quasisymmetry on the Kruskal–Kulsrud variational principle [21] through a Lagrange multiplier and found that the multiplier acts as an anisotropic pressure. A recent variational treatment of guiding-center kinetics [22] showed that the pressures of any kinetic equilibrium built on the invariants of guiding-center motion satisfy parallel force balance identically at the kinetic as well as the fluid level of description, without imposing parallel force balance through an additional Lagrange multiplier.

The existence of an exact finite-pressure equilibrium leaves a separate question: can it be connected continuously to a nonzero vacuum field on a fixed regular toroidal domain? The scalar-pressure families in Ref. [12] demonstrate exact nested-surface equilibria and axisymmetric limits. That work does not establish a continuation to a nonzero vacuum field on a fixed regular toroidal domain.

In this paper, we construct exact nonsymmetric toroidal anisotropic equilibria by applying a classical pressure-field transformation to the new vacuum seeds of Ref. [15]. They connect continuously to 3D nonsymmetric vacuum fields, have perfectly nested flux surfaces through the whole volume with a regular magnetic axis, sheared rotational transform, positive pressures, and no continuous symmetry. A single parameter switches on the anisotropy, the plasma current and the Lorentz force, while the flux surfaces and the transform remain exactly those of the vacuum field. Shear depends on the seed. We show an ellipsoidal configuration with finite magnetic shear (Figure 1 and Appendix § 16) and a second, also sheared, ellipsoidal configuration whose trapped particles have local closed action contours. A shearless paraboloidal benchmark is treated in Methods and the appendices. To our knowledge, the sheared ellipsoidal companions are the first exact anisotropic equilibria that combine nested toroidal magnetic surfaces about a regular axis, given in closed form, with nonzero rotational transform, finite magnetic shear, no continuous symmetry, positive pressures, a positive firehose margin and a continuous connection to a nonzero vacuum field. None of the earlier constructions cited below establish all these properties together, which bring these equilibria a step closer to physical realizability. Three consequences follow, for ideal stability, for kinetic pressure support and for drift confinement.

Figure 1. A nonsymmetric ellipsoidal configuration with finite magnetic shear. a, Outer surface of the N=1N = 1 ellipsoidal vacuum seed, coloured by the exact field strength. b, Cutaway showing nested surfaces and the planar noncircular magnetic axis (black). c, Pressure envelopes of the current-carrying companion f2=1+0.05(1−ρ02)f^2 = 1 + 0.05(1 - \rho_0^2) with P=15P = 15 kPa, sampled over each surface. d, Contours of the seed field strength ∣B∣|B| in Boozer coordinates at ρ0=0.8\rho_0 = 0.8. The companions keep these surfaces and the rotational transform, which falls by 7.9% from axis to edge (Figure A1). This configuration is distinct from the N=3N = 3 action example of Figure 3.

Figure A1. Rotational transform and shear of the configuration of Fig. 1. The exact transform falls by 7.9% from 2/7\sqrt{2}/7 on the axis. Circles are DESC values at (L,M,N)=(14,14,48)(L,M,N)=(14,14,48) with zero net toroidal current. The rescaling leaves the transform unchanged, and derivatives with respect to physical flux include the flux relabeling.

Adding pressure while preserving exact surfaces

The static equilibrium of a plasma with gyrotropic pressure tensor P=p⊥I+(p∥−p⊥)bb\mathbf{P}=p_{\perp}\mathbf{I}+(p_{\parallel}-p_{\perp})\mathbf{b}\mathbf{b} satisfies [16]

∇⋅P=j×B,μ0j=∇×B,∇⋅B=0,(1)\nabla\cdot\mathbf{P}=\mathbf{j}\times\mathbf{B},\qquad\mu_{0}\mathbf{j}=\nabla\times\mathbf{B},\qquad\nabla\cdot\mathbf{B}=0, \tag*{(1)}

equivalently ∇⋅(P+B2I/(2μ0)−BB/μ0)=0\nabla\cdot(\mathbf{P}+B^{2}\mathbf{I}/(2\mu_{0})-\mathbf{B}\mathbf{B}/\mu_{0})=0 with ∇⋅B=0\nabla\cdot\mathbf{B}=0.

Exact solutions of equations (1) follow from any smooth isotropic magnetostatic equilibrium (B0,p0)(\mathbf{B}_{0},p_{0}) with B0≠0\mathbf{B}_{0}\ne0, μ0j0=∇×B0\mu_{0}\mathbf{j}_{0}=\nabla\times\mathbf{B}_{0} and ∇p0=j0×B0\nabla p_{0}=\mathbf{j}_{0}\times\mathbf{B}_{0}. We take a function f>0f>0 with B0⋅∇f=0\mathbf{B}_{0}\cdot\nabla f=0 and set Q=(f2−1)B02/(2μ0)Q=(f^{2}-1)B_{0}^{2}/(2\mu_{0}). For any constant CC, the fields B=fB0\mathbf{B}=f\mathbf{B}_{0}, p∥=p0+C+Qp_{\parallel}=p_{0}+C+Q and p⊥=p0+C−Qp_{\perp}=p_{0}+C-Q solve equations (1) exactly, because the total stress equals the seed’s isotropic stress plus CIC\mathbf{I}. The current is j=fj0+∇f×B0/μ0\mathbf{j}=f\mathbf{j}_{0}+\nabla f\times\mathbf{B}_{0}/\mu_{0}, and both pressures are positive where p0+C>∣Q∣p_{0}+C>|Q|. This is the classical transformation introduced by Clemente for axisymmetric equilibria [23] and given in three

dimensions by Cheviakov, Bogoyavlenskij and Anco [24, 25]. Because ff is constant on field lines, the map keeps the seed’s field lines, magnetic surfaces and rotational transform.

Applied to the exact 3D scalar-pressure equilibria with nested toroidal surfaces of Refs. [11, 12, 13, 14], the map gives exact nonsymmetric anisotropic equilibria with nested surfaces and the seeds’ rotational transform. No continuation of those seeds to a nonzero vacuum field on a fixed domain has been established. Some earlier constructions have sought solutions of the kind given here in different ways, and we state how each differs. Cheviakov applied the map to seeds whose magnetic surfaces are helical cylinders, ellipsoidal shells or open flux tubes [24, 26]. His thesis also applied it to the vacuum field of a nonsymmetric closed current loop outside the plasma, whose field lines were found by numerical tracing to cover nested tori [27]. That is the closest earlier construction, but its surfaces and rotational transform were not established analytically. Sato and Yamada [28] rescaled the harmonic field ∇(φ+ϵx)\nabla(\varphi+\epsilon x) by a function of an explicit family of nested tori with no continuous Euclidean symmetry, which is the same map with a vacuum seed. That seed has no vertical component, so every field line is a closed planar curve and the rotational transform is zero. Its tori are degenerate and only a choice among many invariant families rather than surfaces fixed by the field. Exact anisotropic constructions with exactly zero firehose margin, and related flowing equilibria, are compared in Methods.

We now specialize to vacuum seeds, p0=0p_0=0, with a spatially uniform mean pressure P=CP=C. This choice supplies the explicit nonzero vacuum endpoint and the nonnegative vacuum magnetic term used in our Chew–Goldberger–Low (CGL) [16] proofs. Our kinetic realization also uses the uniform mean pressure and globally constant ff. These stability and kinetic results have not been established here for a general pressure-carrying seed. The constant-ff action identity remains valid for any seed. We use the exact vacuum fields of Ref. [15]. They are curl-free and divergence-free, are given in closed form, have nested toroidal surfaces about a regular magnetic axis and nonzero rotational transform, and have no continuous symmetry.

We take B0\mathbf{B}_0 to be a smooth, nonvanishing vacuum field on a toroidal plasma domain, with a regular magnetic axis and nested surfaces labelled by ψ\psi:

∇×B0=0,∇⋅B0=0,B0⋅∇ψ=0.(2)\nabla\times\mathbf{B}_0 = 0,\qquad\nabla\cdot\mathbf{B}_0 = 0,\qquad\mathbf{B}_0 \cdot\nabla\psi= 0. \tag*{(2)}

For such a seed we choose a positive smooth function f(ψ)f(\psi) and a spatially uniform pressure PP. Then

B=fB0,p∥=P+Q,p⊥=P−Q,Q=(f2−1)B022μ0(3)\mathbf{B} = f\mathbf{B}_0,\qquad p_{\parallel} = P + Q,\qquad p_{\perp} = P - Q,\qquad Q = \frac{(f^2-1)B_0^2}{2\mu_0} \tag*{(3)}

solve the static anisotropic equations (1) exactly. Here P=p⊥+(p∥−p⊥)bbP = p_{\perp} + (p_{\parallel} - p_{\perp})\mathbf{b}\mathbf{b} and b=B/B\mathbf{b} = \mathbf{B}/B. For f=1f=1 the anisotropy vanishes and the magnetic field is the vacuum seed. With P=0P=0 as well this is the vacuum itself, the endpoint η=0\eta=0 of the vacuum-connected family given in Methods. The physical reason is a cancellation of stresses:

P+B22μ0I−BBμ0=PI+B022μ0I−B0B0μ0.(4)P + \frac{B^2}{2\mu_0}\mathbf{I} - \frac{\mathbf{B}\mathbf{B}}{\mu_0} = P\mathbf{I} + \frac{B_0^2}{2\mu_0}\mathbf{I} - \frac{\mathbf{B}_0\mathbf{B}_0}{\mu_0}. \tag*{(4)}

The right-hand side is divergence-free. Pressure anisotropy compensates precisely for the change in magnetic stress. Appendix B gives the derivation.

Because ff is constant along each field line, all field-line trajectories and their rotational transforms are unchanged. This preservation is exact at finite amplitude, with no perturbation expansion. The current is

j=f′(ψ)μ0∇ψ×B0.(5)\mathbf{j} = \frac{f'(\psi)}{\mu_0}\nabla\psi\times\mathbf{B}_0. \tag*{(5)}

This current is already divergence-free, so no driven parallel-current correction is required. The uniform mean pressure removes that source of the current-solvability problem; it does not impose quasisymmetry (Appendix B). Constant ff gives a current-free magnetic field supporting finite anisotropic pressure. A surface-dependent ff gives a perpendicular plasma current. Both cases retain the same physical magnetic surfaces. No continuous spatial symmetry or quasisymmetry is imposed. In Boozer coordinates the field strength of the sheared configuration has isolated maxima and minima enclosed by closed contours (Figure 1d), which quasisymmetry excludes. The Boozer spectrum of the paraboloidal benchmark mixes several helicities (Appendix C). A one-parameter family of these equilibria, equation (11) in Methods, switches on the pressure, the anisotropy and the current from the vacuum while the domain, the surfaces and the transform stay fixed.

A nonsymmetric configuration with finite shear

The N=1N = 1 ellipsoidal configuration of the vacuum family has nested toroidal surfaces around a planar noncircular magnetic axis (Figure 1a,Figure 1b). It has nonzero magnetic shear such that the rotational transform falls by 7.9%, from 2/7\sqrt{2}/7 on the axis to 0.186 at the boundary. Multiplication by a constant or surface-dependent ff keeps the field lines and this transform profile. Derivatives with respect to physical flux change only through the flux relabeling, dι/dψ=(dι0/dψ0)/f\mathrm{d}\iota/\mathrm{d}\psi= (\mathrm{d}\iota_0/\mathrm{d}\psi_0)/f. The companions use P=15P = 15 kPa and f2=1.05f^2 = 1.05 or 1+0.05(1−ρ2)1 + 0.05(1 - \rho^2), the latter shown in Figure 1c, and both have p∥≥p⊥p_\parallel\ge p_\perp. Started from the projected exact solution, DESC [18, 19] runs with zero net toroidal current reproduce both companions with errors near 10−810^{-8}, with the transform and its shear computed as outputs. Appendix A gives the parameters and the DESC errors. We mapped the second adiabatic invariant of this configuration (equation (9) below) over the labels of nearby trapped wells, for several field-line sections and twelve mirror strengths. None of these 36 maps has an interior minimum or maximum, so local drift confinement is shown on a separate, also sheared, N=3N = 3 configuration below.

Stability

The stress identity (4) provides a remarkable organization of the stability results. For a displacement fixed at the boundary, the quadratic energy of every configuration splits into a nonnegative vacuum-seed magnetic term and a local compression and parallel-strain term. In this decomposition the usual pressure-gradient-curvature drive cancels exactly. We use three models. Ideal CGL dynamics gives a sufficient linear criterion for every configuration and a finite-amplitude constrained minimum for p∥≥p⊥p_\parallel\ge p_\perp, indicating nonlinear stability. For globally constant ff, guiding-center kinetic theory gives an electrostatic ground state and a restricted linear electromagnetic criterion. Both kinds of statement are needed. For globally constant ff, the Kruskal–Oberman quadratic form is bounded above by the CGL form for the same displacement, so CGL stability alone does not imply electromagnetic kinetic stability [29, 30]. One pressure inequality, derived first on the kinetic side, governs both.

Kinetic pressure support. An exact pressure tensor is not sufficient to ensure equilibria at the kinetic level since the particles must be able to sustain it. For constant ff, QQ is proportional to B2B^2, so both pressures are functions of BB alone (Appendix D). On a bounded field-strength interval, the guiding-center moment problem can be inverted explicitly. A particularly useful result is that a nonnegative distribution F(E,μ)F(\mathcal{E},\mu), non-increasing in energy at fixed magnetic moment, exists if and only if

p⊥2≤6Pp∥(6)p_\perp^2 \le6Pp_\parallel \tag*{(6)}

throughout that interval, assuming positive pressures. The construction follows from an Abel-type moment representation (Methods and appendices). The necessity is the classical Cauchy–Schwarz moment condition of Taylor and Hastie [29], their Eq. (3.7), specialized to this pressure law. Our result is an exact statement within the standard guiding-center model, where μ\mu is an invariant. It is not an exact finite-gyroadius Vlasov realization.

Linear CGL stability. Anisotropic stability and guiding-center pressure support have a long history [31, 32, 5]. Earlier work studied the linear stability [33] and classical particle transport [34] of anisotropic axisymmetric field-reversed configurations. We take a displacement ξ\boldsymbol{\xi} that vanishes at the boundary and write c=∇⋅ξc=\nabla\cdot\boldsymbol{\xi} and s=bibj∂jξis=b_i b_j\partial_j\xi_i. For every equilibrium (3), including surface-dependent ff, its quadratic potential energy is

δW=12μ0∫∣∇×(ξ×B0)∣2 dV+12∫[2Pc2−2(P+Q)cs+(3P+5Q)s2] dV.(7)\delta W = \frac{1}{2\mu_0}\int\lvert\nabla\times(\boldsymbol{\xi}\times\mathbf{B}_0)\rvert^2\,dV+\frac{1}{2}\int\left[2Pc^2-2(P+Q)cs+(3P+5Q)s^2\right]\,dV. \tag*{(7)}

The first term is the energy of bending and compressing the vacuum seed. The remaining term is a local compression and parallel-strain energy, the quadratic form

2Pc2−2(P+Q)cs+(3P+5Q)s2=(cs)(2P−(P+Q)−(P+Q)3P+5Q)(cs).2Pc^2-2(P+Q)cs+(3P+5Q)s^2=(c\quad s)\begin{pmatrix}2P&-(P+Q)\\-(P+Q)&3P+5Q\end{pmatrix}\begin{pmatrix}c\\s\end{pmatrix}.

Its matrix has leading entry 2P>02P>0 and determinant 6Pp∥−p⊥26Pp_{\parallel}-p_{\perp}^2. Condition (6) therefore makes every term nonnegative. Appendix E gives the derivation.

The shared inequality is not a coincidence. With P=(p∥+p⊥)/2P=(p_{\parallel}+p_{\perp})/2, the local form in equation (7) can be written 2P(c−s)2+2p⊥(c−s)s+3p∥s22P(c-s)^2+2p_{\perp}(c-s)s+3p_{\parallel}s^2. For the energy-monotone distributions of a globally constant ff, its three coefficients are the moments ∑s∫(−∂EFs)(μB)2 d3v=2P\sum_s\int(-\partial_E F_s)(\mu B)^2\,d^3v=2P, ∑s∫(−∂EFs)(μB)(msv∥2) d3v=p⊥\sum_s\int(-\partial_EF_s)(\mu B)(m_s v_{\parallel}^2)\,d^3v=p_{\perp} and ∑s∫(−∂EFs)(msv∥2)2 d3v=3p∥\sum_s\int(-\partial_EF_s)(m_s v_{\parallel}^2)^2\,d^3v=3p_{\parallel} (Appendix F). Whenever such a distribution exists, the local CGL energy is a nonnegative sum of squares weighted by −∂EFs-\partial_EF_s. Its determinant 6Pp∥−p⊥26Pp_{\parallel}-p_{\perp}^2 is the Cauchy–Schwarz moment margin. This is why condition (6) governs both kinetic realization and nonnegative CGL energy (Figure 2).

Figure 2. The same pressure inequality controls two distinct questions. For r=p⊥/p∥r = p_{\perp}/p_{\parallel}, the normalized CGL compression matrix has diagonal entries 1+r1+r and 4−r4-r and off-diagonal entries −1-1. The solid blue curve is its minimum eigenvalue. The dashed orange curve is its determinant 3(1+r)−r23(1+r)-r^2, the kinetic moment margin, divided by 4. Both vanish at r=(3+21)/2≈3.79r=(3+\sqrt{21})/2\approx3.79 (dotted line). In the blue region the local CGL energy is nonnegative. For the globally constant-ff pressure law, an energy-monotone kinetic realization exists if and only if the bound holds throughout the field-strength interval. A value in the grey region leaves the CGL test inconclusive and excludes such a kinetic realization. The CGL statement is sufficient for the entire vacuum-companion family satisfying this bound, and failure of the bound alone does not imply instability. The curves are analytic criteria.

This is a sufficient fixed-boundary ideal CGL result, not a necessary criterion. It includes localized ballooning and internal kink stability. With P=(p∥+p⊥)/2P=(p_{\parallel}+p_{\perp})/2 and r=p⊥/p∥r=p_{\perp}/p_{\parallel}, condition (6) reads r2−3r−3≤0r^2-3r-3\leq0, that is 0<r≤(3+21)/2≈3.790<r\leq(3+\sqrt{21})/2\approx3.79 (Figure 2). It holds automatically for r≤1r\leq1, which includes every case with Q≥0Q \ge0, so it restricts only Q<0Q < 0. Methods discusses the equality case and the firehose margin. Although the fluid bound applies to radial ff, the distribution theorem established here applies to constant ff.

Finite-amplitude CGL minimum. For 0≤Q<P0 \le Q < P, the fluid family has a further finite-amplitude property. Every case minimizes CGL potential energy over smooth ideal material deformations connected to identity and fixed pointwise at the boundary. The conserved constraints are material mass, frozen magnetic flux and the two material invariants p⊥/(nB)p_\perp/(nB) and p∥B2/n3p_\parallel B^2/n^3. The exact finite-deformation identity is

ECGL[X]−ECGL[id]=12μ0∫∣B0,X−B0∣2 dV+∫G d3a≥0,(8)E_{\mathrm{CGL}}[\mathbf{X}] - E_{\mathrm{CGL}}[\mathrm{id}] = \frac{1}{2\mu_0}\int\lvert\mathbf{B}_{0,\mathbf{X}}-\mathbf{B}_0\rvert^2\,dV + \int G\,d^3a \ge0, \tag*{(8)}

where B0,X\mathbf{B}_{0,\mathbf{X}} is the frozen image of the vacuum seed and G≥0G \ge0 is the local compression/strain remainder derived in Appendix G, which also compares this minimum with Taylor’s relaxation theory [35], which has found extensive applications in laboratory, astrophysical and solar plasmas. Adding kinetic energy gives a conserved nonnegative relative energy, which is a nonlinear Lyapunov functional for stability. It bounds a perturbed ideal motion of these equilibria, measured in this energy, for as long as that motion remains smooth.

Sampled finite boundary-fixed deformations raised the material CGL energy of the sheared ellipsoidal configuration and of the paraboloidal benchmark, apart from signed values at roundoff level (Appendices A and C).

Kinetic stability. For globally constant ff satisfying equation (6), we choose compatible quasineutral species distributions Fs(Es,μ)F_s(\mathcal{E}_s,\mu) non-increasing in energy. In a fixed magnetic field, these minimize particle energy among phase-space-volume-preserving rearrangements at each μ\mu. With a closed, energy-conserving boundary, nonnegative electrostatic field energy makes the combined state a constrained ground state in the Hamiltonian guiding-center–Poisson model. This applies Gardner’s and Helander’s framework [36, 37], consistent with the kinetic variational viewpoint of Ref. [22]; neither spatial symmetry nor conservation of bounce action is required. For smooth strictly decreasing distributions, a positive quadratic invariant of the linearized fixed-field guiding-center–Poisson model excludes growing electrostatic modes in its norm, and a convex energy–Casimir functional extends this to nonlinear energetic stability (Methods and Appendix H).

For globally constant ff, classical Taylor–Hastie theory [29] supplies a sufficient linear electromagnetic stability argument in the low-frequency small-Larmor-radius guiding-center model [31]. For these equilibria its coefficients reduce exactly, and its potential energy is the vacuum-seed magnetic energy plus a nonnegative kinetic term (Appendix F). Its application requires energy-monotone compatible species, finite moments, a closed supported conservative boundary and well-defined orbit averages. The explicit nonsmooth distributions require a finite quadratic-form or justified limiting interpretation. Kinetic stability for radial ff and general nonlinear electromagnetic or finite-Larmor-radius stability are not established here.

Confined drift contours without a symmetry requirement

Magnetic flux surfaces constrain parallel motion, but particles experiencing magnetic drifts need not remain on them. At fixed energy and magnetic moment, leading bounce-averaged drift trajectories follow level contours of the second adiabatic invariant defined as [5, 6]

J(E,μ)=2∫ℓ−ℓ+2mp(E−μB) dℓ.(9)J(\mathcal{E},\mu) = 2\int_{\ell_-}^{\ell_+}\sqrt{2m_p(\mathcal{E}-\mu B)}\,d\ell. \tag*{(9)}

A nondegenerate definite extremum of this smooth action has nearby closed regular contours, provided the reduced drift coefficient is nonsingular and the same mirror well persists. This local confinement mechanism does not require the action to be constant on every magnetic surface.

We now consider the N=3N = 3 ellipsoidal configuration, which differs from the N=1N = 1 case shown in Figure 1. This configuration is also sheared. Tracing the field lines of the exact seed gives a rotational transform that falls from 2/7\sqrt{2}/7 on the axis to between 0.1838 and 0.1847 on the surfaces that carry the contours of Figure 3a, about 9% lower (Appendix I). Each of 32 mirror strengths from Bm=28.5B_m = 28.5 to 31.6 gives a numerical action minimum with positive Hessian eigenvalues (Figure 3b), and Figure 3a shows the contours for three of them. Magnetic flux surface labels vary along these contours, demonstrating the distinction between a closed drift contour and omnigeneity. For a globally constant ff, this confinement structure transfers particularly simply:

Figure 3. Local iso-action confinement across magnetic surfaces. a, Sixteen computed contour roots for each of three seed mirror strengths in the N=3N = 3 ellipsoidal configuration. The dashed lines are quadratic Hessian approximations, not integrated particle orbits. Grey curves are magnetic surface labels on the section. Coordinates and field strengths use the seed’s native normalization. b, Smallest Hessian eigenvalue at the numerical minima for 32 seed mirror strengths. The three coloured points are the mirror strengths of panel a. For constant f=1.01f = \sqrt{1.01}, equation (10) transfers the action contours to the anisotropic companion. This configuration, which is also sheared, differs from the N=1N = 1 configuration of Figure 1.

Jf(E,μ;ψ,α)=J0(E,fμ;ψ,α).(10)J_f(\mathcal{E}, \mu; \psi, \alpha) = J_0(\mathcal{E}, f\mu; \psi, \alpha). \tag*{(10)}

Thus closed contours of the seed map to companion contours at a rescaled magnetic moment. Methods states the scope of this transfer.

Any exact vacuum field with nested magnetic surfaces therefore supplies a family of anisotropic equilibria with the same field lines and rotational transform, and finite shear is retained. Both ellipsoidal configurations shown here are sheared, and the N=3N = 3 configuration also has local closed action contours. Kinetic stability for surface-dependent rescalings and against general nonlinear electromagnetic perturbations remains open, as does whether such states can be observed in experiments or in nature.

Methods

Exact construction and domain

The seed is restricted to an embedded toroidal domain in which its coordinate map and surface foliation are regular and B0>0B_0 > 0. Multiplication by a positive smooth flux function preserves divergence freedom and regularity. The divergence of equation (4) establishes force balance without solving an additional differential equation. Positive pressures require P>∣Q∣P > |Q| pointwise. The dimensionless anisotropy parameter is [38, 23] σ=μ0(p∥−p⊥)/B2=1−f−2\sigma= \mu_0(p_{\parallel} - p_{\perp})/B^2 = 1 - f^{-2}. Constant PP is essential. The positive pressures remain nonzero at the boundary. These equilibria require boundary support. No plasma–vacuum matching is constructed here.

Following Mercier and Cotsaftis [38], the dimensionless anisotropy parameter σ=μ0(p∥−p⊥)/B2\sigma= \mu_0(p_{\parallel} - p_{\perp})/B^2 is constant along field lines in this construction, although the pressures themselves generally are not. DESC’s anisotropic solver corroborates the constructed equilibria. Rodríguez and Bhattacharjee related quasisymmetry to a more general anisotropic equilibrium formulation [20]. Our construction does not allow equilibria with field-line-varying σ\sigma. It therefore cannot resolve the overdetermination of quasisymmetry in the way proposed by Rodríguez and Bhattacharjee.

Zero-firehose-margin class and flowing equilibria

Exact anisotropic force balance also holds for every divergence-free field when p∥−p⊥=B2/μ0p_{\parallel} - p_{\perp} = B^2/\mu_0 and p∥+p⊥p_{\parallel} + p_{\perp} is uniform [39]. Sato et al. used this class for local quasisymmetric fields and for fields without rotational transform in asymmetric toroidal domains [40, 41]. Kaltsas et al. [42] used it to build toroidally asymmetric equilibria and examined their surfaces by field-line tracing. Some of their parameter choices develop outer islands or stochastic regions. The firehose margin 1−μ0(p∥−p⊥)/B21 - \mu_0(p_{\parallel} - p_{\perp})/B^2 of that class is exactly zero, so it lies outside our family with finite positive ff. Related axisymmetric rotating and cylindrical flowing equilibria have also been constructed [43, 44]. Extending the present nonsymmetric toroidal construction to flow would require a separate derivation. We focus on cases without flow, but future extensions of our method to rotating equilibria are possible [43, 44].

Continuous connection to the vacuum field

To exhibit the vacuum connection explicitly, we choose a bounded smooth h(ψ)h(\psi) and set

fη2=1+ηh(ψ),Pη=ηP∗,0≤η≤η∗,(11)f_\eta^2 = 1 + \eta h(\psi), \qquad P_\eta= \eta P_\ast, \qquad0 \le\eta\le\eta_\ast, \tag*{(11)}

Here η≥0\eta\ge0 is the loading parameter, and η=0\eta= 0 is the vacuum. The function h(ψ)h(\psi) fixes the radial shape of the anisotropy, and η∗\eta_\ast is any value for which 1+ηh>01 + \eta h > 0 throughout, and P∗P_\ast is the mean pressure per unit η\eta, so Pη=ηP∗P_\eta= \eta P_\ast. The condition P∗>sup⁡∣hB02∣/(2μ0)P_\ast> \sup|hB_0^2|/(2\mu_0) keeps both pressures positive. The pressures and current are then

p∥,η=η(P∗+hB022μ0),p⊥,η=η(P∗−hB022μ0),(12)p_{\parallel,\eta} = \eta\left(P_\ast+ \frac{hB_0^2}{2\mu_0}\right), \qquad p_{\perp,\eta} = \eta\left(P_\ast- \frac{hB_0^2}{2\mu_0}\right), \tag*{(12)}
jη=ηh′(ψ)2μ01+ηh(ψ)∇ψ×B0.(13)\mathbf{j}_\eta= \frac{\eta h'(\psi)}{2\mu_0\sqrt{1+\eta h(\psi)}}\nabla\psi\times\mathbf{B}_0. \tag*{(13)}

Both pressures are positive for η>0\eta> 0. The additional stability restrictions are stated below. On the same fixed toroidal domain,

Bη⟶B0≠0,p∥,η,p⊥,η⟶0,jη⟶0(η⟶0).(14)\mathbf{B}_\eta\longrightarrow\mathbf{B}_0 \ne0, \qquad p_{\parallel,\eta},p_{\perp,\eta} \longrightarrow0, \qquad\mathbf{j}_\eta\longrightarrow0 \qquad(\eta\longrightarrow0). \tag*{(14)}

For smooth seed data this convergence is smooth. No surface or axis degenerates in the limit. If hh is not a constant, every configuration with η>0\eta> 0 carries plasma current. The rotational transform and physical field-line trajectories are unchanged for every η\eta, not merely to leading order.

This regular vacuum limit distinguishes the family from finite-pressure constructions without such an endpoint. The pressure load is supported by anisotropic stress with spatially uniform arithmetic mean Pη=(p∥,η+p⊥,η)/2P_\eta= (p_{\parallel,\eta} + p_{\perp,\eta})/2; an arbitrary scalar-pressure gradient cannot be added to the cancellation. At an isotropic case Q=0Q = 0, equation (3) returns f=1f = 1 and uniform pressure. Each value of η\eta gives an equilibrium with its boundary support. The family is not an ideal dynamical trajectory from one value to another.

Guiding-center moment inversion

For E=mpv∥2/2+μB\mathcal{E} = m_p v_\parallel^2/2 + \mu B, we take F=w(μ)G^(E/μ)F = w(\mu)\widehat{G}(\mathcal{E}/\mu), where w≥0w \ge0 has finite positive μ1/2\mu^{1/2} and μ3/2\mu^{3/2} moments, as an exponential does, and G^(y)=∫y∞ν(b) db\widehat{G}(y) = \int_y^\infty\nu(b)\,db, with ν\nu a nonnegative measure. With Cw=4π2mp−3/2∫0∞w(μ)μ3/2dμC_w = 4\pi\sqrt{2}m_p^{-3/2}\int_0^\infty w(\mu)\mu^{3/2}d\mu, its moments are

p∥=23CwB2∫ν(b)(b−B)+3/2 db(15)p_\parallel= \frac{2}{3}C_w B^2 \int\nu(b)(b-B)_+^{3/2}\,db \tag*{(15)}
p⊥=CwB2∫ν(b)(b−B)+1/2 db(16)p_\perp= C_w B^2 \int\nu(b)(b-B)_+^{1/2}\,db \tag*{(16)}

Cauchy–Schwarz applied to these moments and their derivatives gives equation (6). Conversely, a positive atom above Bmax⁡B_{\max} plus a nonnegative density on [Bmin⁡,Bmax⁡][B_{\min}, B_{\max}] reproduces the target pressure law whenever the bound holds. The explicit inversion is in Appendix § 19. More general energy-monotone distributions can be integrated over μ\mu to obtain the same positive-measure representation. In terms of Rμ=(∫wμ1/2dμ)/(∫wμ3/2dμ)R_\mu= (\int w\mu^{1/2}d\mu)/(\int w\mu^{3/2}d\mu), the physical moments obey n=Rμp⊥/Bn = R_\mu p_\perp/B, T⊥=B/RμT_\perp= B/R_\mu and T∥=Bp∥/(Rμp⊥)T_\parallel= Bp_\parallel/(R_\mu p_\perp) in energy units. They generally vary with BB. Appendix § 21 gives a compatible ion–electron normalization with pointwise neutrality. It also shows that for constant ff the guiding-center and magnetization currents cancel, as the curl-free field requires.

Pressure bound and firehose margin

At equality the local form is semidefinite. The bound is sufficient for nonnegative CGL energy because the nonnegative seed magnetic term was not used. It is not a necessary mirror threshold. The ideal firehose margin is 1−μ0(p∥−p⊥)/B2=f−2>01 - \mu_0(p_\parallel- p_\perp)/B^2 = f^{-2} > 0 for every positive ff. For f<1f < 1 the perpendicular pressure is larger, so condition (6) must also be checked.

Electrostatic invariant

For smooth strictly decreasing distributions the linearized equations of the fixed-field, μ\mu-conserving guiding-center–Poisson model with a conservative boundary conserve the quadratic invariant derived in Appendix § 23 (for energy–Casimir background, see [22] and references therein)

Wes=12∑s∫∣δfs∣2−∂EFs dΓs+ϵ02∫∣∇δϕ∣2 dV.(17)\mathcal{W}_{\mathrm{es}} = \frac{1}{2}\sum_s \int\frac{|\delta f_s|^2}{-\partial_{\mathcal{E}}F_s}\,d\Gamma_s + \frac{\epsilon_0}{2}\int|\nabla\delta\phi|^2\,dV. \tag*{(17)}

The kinetic and electrostatic parts change at equal and opposite rates. At the energy–Casimir level this corresponds to Cs′′(Fs,μ)=1/(−∂EFs)C_s''(F_s,\mu) = 1/(-\partial_{\mathcal{E}}F_s) [45]. This positive invariant excludes growing modes in its norm. A convex energy–Casimir functional extends it to nonlinear energetic stability. Our explicit distributions can have plateaus and steps: their ground-state statement uses rearrangement and allows neutral rearrangements (Appendix § 23). The theorem does not cover electromagnetic perturbations or processes that break conservation of μ\mu.

Ideal constraints and second variation

For a material deformation F\mathbf{F}, we write J=det⁡F\mathcal{J} = \det\mathbf{F} and ℓ=∣Fb∣\ell= \lvert\mathbf{F}\mathbf{b}\rvert. Frozen flux gives B′=FB/J\mathbf{B}' = \mathbf{F}\mathbf{B}/\mathcal{J}; mass conservation gives n′=n/Jn' = n/\mathcal{J}. Conservation of p⊥/(nB)p_{\perp}/(nB) and p∥B2/n3p_{\parallel}B^{2}/n^{3} gives the CGL material energy

U(F)=B2ℓ22μ0J+p⊥ℓJ+p∥2ℓ2.(18)U(\mathbf{F}) = \frac{B^{2}\ell^{2}}{2\mu_{0}\mathcal{J}} + \frac{p_{\perp}\ell}{\mathcal{J}} + \frac{p_{\parallel}}{2\ell^{2}}. \tag*{(18)}

Its first variation gives force balance, the anisotropic counterpart of an ideal constrained equilibrium formulation [21]. Standard variational background is in Refs. [30, 46]. Expanding F=I+ϵ∇ξ\mathbf{F} = \mathbf{I} + \epsilon\nabla\boldsymbol{\xi} to second order, subtracting the vacuum-seed magnetic energy and using ξ=0\boldsymbol{\xi} = 0 on the boundary yields equation (7). The integrated residual P[tr⁡(∇ξ)2−(∇⋅ξ)2]/2P[\operatorname{tr}(\nabla\boldsymbol{\xi})^{2} - (\nabla\cdot\boldsymbol{\xi})^{2}]/2 vanishes by integration by parts. Each constructed state is therefore an equilibrium of the CGL energy under the ideal constraints. This does not show that it is the only such equilibrium, or that an ideal motion can carry the plasma from one configuration of the family to another.

Action calculations

The action example uses the ellipsoidal seed with k=0.05k = 0.05, N=3N = 3, κ=1.5\kappa= 1.5, phase K(1−k2)K(1-k^{2}) and ua=va≈1.662556u_{a} = v_{a} \approx1.662556. Its axis transform is 2/7\sqrt{2}/7. Numerical quadrature evaluates j=∫Bm−B dℓj = \int\sqrt{B_{m}-B}\,\mathrm{d}\ell, with J=22mpμ jJ = 2\sqrt{2m_{p}\mu}\,j. At each of 32 values Bm=28.5,28.6,…,31.6B_{m} = 28.5, 28.6, \ldots, 31.6, action minimization is followed by Hessian estimation and sixteen radial contour-root solves. Closed contours represent leading bounce-averaged drift confinement only while mirror points remain simple and the drift reduction is regular. The contours in Figure 3 are not direct integrations of guiding-center or finite-gyroradius orbits. For a surface-dependent ff, equation (10) still holds on each field line, but the rescaled moment fμf\mu differs between the surfaces that a drift contour crosses. A contour of JfJ_{f} then need not be a contour of J0J_{0} at a single magnetic moment, so the seed’s closed contours do not carry over directly. Appendix § 24 gives the contour data and the rotational transform.

Configurations treated in the appendices

Appendix § 16 gives the parameters, the DESC errors, the finite-deformation tests and the action maps of the sheared configuration of Figure 1. An exact paraboloidal vacuum configuration with exactly constant rotational transform, with constant and surface-dependent ff, serves as a shearless benchmark. Appendix § 18 gives its geometry, Boozer harmonics, DESC results and finite-deformation tests of the energy minimum.

Use of AI tools

Under the authors’ direction, Anthropic Claude Code and OpenAI Codex assisted with analytical derivations, symbolic and numerical checks, equilibrium-code comparisons, scientific plotting code, literature searches and manuscript drafting and editing. The authors set the research questions, selected the examples and reviewed intermediate results and drafts. Checks included analytical identities, symbolic calculations, convergence tests and comparisons between separate agent sessions. The authors retain responsibility for the scientific claims and the final manuscript.

Data availability

The data that support the figures and the validation results, including the DESC comparison data, the finite-deformation results and single-mode minimization checks, the action-curvature scan and the field-line calculation of the rotational transform of the N=3N = 3 configuration, are provided in a source archive (source_archive_v12.5.zip) that is available on request and will be deposited in a public repository with a persistent identifier before publication. Full numerical equilibrium files are retained in a separate DESC reproducibility package and are available from the corresponding author on request.

Code availability

The scripts that evaluate the exact fields, run the finite-deformation tests, trace the field lines, drive the DESC calculations and generate the figures are contained in the same source archive, which is available on request and will be deposited with the data. DESC is publicly available [18, 19].

Acknowledgements

This research was supported by the Simons Foundation/SFARI (Grant No. 560651, AB) and the US Department of Energy (Award Nos. DE-SC0024548 and DE-AC02-09CH11466).

Correspondence

Correspondence and requests for materials should be addressed to [email protected].

The sheared ellipsoidal configuration of Fig. 1

Parameters. The configuration has (N,M,G)=(1,2,2)(N,M,G)=(1,2,2), k=kg=3−22k=k_g=3-2\sqrt{2}, κ=1/2\kappa=1/2, axis transform 2/7\sqrt{2}/7 and boundary surface-label excess 0.1001112822. Native lengths are read as metres, and the field is scaled by 0.012033783 T per native unit, so the maximum seed field is about 0.616 T. The companions use P=15P=15 kPa and f2=1.05f^2=1.05 or 1+0.05(1−ρ02)1+0.05(1-\rho_0^2). The transform falls from 0.2020305 on the axis to 0.1861083 at the boundary, and dι/dsd\iota/ds with s=ψ/ψbs=\psi/\psi_b runs from about −0.0162-0.0162 to −0.0157-0.0157. Both pressures are positive. The local β=2μ0P/B2\beta=2\mu_0P/B^2 ranges from 0.094 to 5.09 for constant ff and from 0.099 to 5.35 for radial ff.

DESC. Each run starts from the exact solution projected on DESC’s basis (DESC 0.17.1, two field periods). The net toroidal current is set to zero, so the transform is an output. At (L,M,N)=(14,14,48)(L,M,N)=(14,14,48) the maximum relative field error against the exact solution is 1.1×10−81.1\times10^{-8} for both companions, ∣Δι∣≤2.9×10−11|\Delta\iota|\leq2.9\times10^{-11}, and the relative error in dι/dψd\iota/d\psi is at most 2.6×10−82.6\times10^{-8}. At (12,12,40)(12,12,40) the field error is 9.6×10−89.6\times10^{-8}, the same as for a vacuum run.

Energy and action maps. Finite boundary-fixed deformations raise the material CGL energy of this configuration, as for the benchmark of Appendix C. A mixed deformation raises it by about 937 J, or 0.042 PVPV. For this configuration, none of 36 maps of the second adiabatic invariant over the labels of nearby trapped wells has an interior minimum or maximum. The maps use well-centre and displaced sections and twelve mirror levels from 0.10 to 0.40 T.

Exact equilibrium and current

We write the total stress as T=PI+B2I/(2μ0)−BB/μ0\mathbf{T}=P\mathbf{I}+B^2\mathbf{I}/(2\mu_0)-\mathbf{B}\mathbf{B}/\mu_0. For B=f(ψ)B0\mathbf{B}=f(\psi)\mathbf{B}_0, p∥=P+Qp_\parallel=P+Q, p⊥=P−Qp_\perp=P-Q and Q=(f2−1)B02/(2μ0)Q=(f^2-1)B_0^2/(2\mu_0), the identity (p∥−p⊥)bb=(f2−1)B0B0/μ0(p_\parallel-p_\perp)\mathbf{b}\mathbf{b}=(f^2-1)\mathbf{B}_0\mathbf{B}_0/\mu_0 gives

T=PI+B022μ0I−B0B0μ0.(19)\mathbf{T}=P\mathbf{I}+\frac{B_0^2}{2\mu_0}\mathbf{I}-\frac{\mathbf{B}_0\mathbf{B}_0}{\mu_0}. \tag*{(19)}

For uniform PP and a curl-free, divergence-free seed, ∇⋅T=0\nabla\cdot\mathbf{T} = 0. Also ∇⋅(fB0)=0\nabla\cdot(f\mathbf{B}_0) = 0, because ff is constant on field lines. Smoothness at the axis requires ff to be a smooth function of a regular surface label. In the notation of Fu and Bhattacharjee (Ref. [22]), the family has p(ψ)=Pp(\psi) = P and K(ψ)=(1−f−2)/(2μ0)K(\psi) = (1 - f^{-2})/(2\mu_0), so p∥,⊥=P±KB2p_{\parallel,\perp} = P \pm K B^2.

Ampère’s law gives

j=f′(ψ)μ0∇ψ×B0,j⋅B=j⋅∇ψ=0.(20)\mathbf{j} = \frac{f'(\psi)}{\mu_0}\nabla\psi\times\mathbf{B}_0,\qquad\mathbf{j}\cdot\mathbf{B} = \mathbf{j}\cdot\nabla\psi= 0. \tag*{(20)}

Since ∇×B0=0\nabla\times\mathbf{B}_0 = 0, ∇⋅(∇ψ×B0)=0\nabla\cdot(\nabla\psi\times\mathbf{B}_0) = 0, and differentiating f′(ψ)f'(\psi) adds a multiple of ∇ψ⋅(∇ψ×B0)=0\nabla\psi\cdot(\nabla\psi\times\mathbf{B}_0) = 0. Hence ∇⋅j⊥=0\nabla\cdot\mathbf{j}_{\perp} = 0 and B⋅∇(j∥/B)=0\mathbf{B}\cdot\nabla(j_{\parallel}/B) = 0. The family has j∥=0j_{\parallel} = 0 without solving a magnetic differential equation, and no quasisymmetry condition enters.

The shearless paraboloidal benchmark

Figure C1. Shearless paraboloidal benchmark. a, Outer surface of the seed, coloured by the exact field strength. b, Cutaway with two nested interior surfaces and the nonplanar magnetic axis (black). c, Pressure envelopes of the current-carrying companion f2=1+0.05(1−ρ02)f^2 = 1 + 0.05(1 - \rho_0^2) with P=15P = 15 kPa, sampled over each surface. d, Contours of the seed field strength ∣B∣|B| in Boozer coordinates at ρ=0.8\rho= 0.8, over one field period.

Geometry. The position map is

x=(sinh⁡ucos⁡z−sinh⁡vsin⁡z, sinh⁡usin⁡z−sinh⁡vcos⁡z, sinh⁡usinh⁡v+12sin⁡2z),(21)\mathbf{x}=\left(\sinh u\cos z-\sinh v\sin z,\ \sinh u\sin z-\sinh v\cos z,\ \sinh u\sinh v+\frac{1}{2}\sin2z\right), \tag*{(21)}

with k=0.5k=0.5, ua=1u_a=1, va=0v_a=0, κ=2K(1−k2)/π\kappa=2K(1-k^2)/\pi, zero profile phase and magnetic scale 0.2 in SI units. The surface label runs from 1 on the axis to 1.02. The transform is the constant 0.1103880277, the toroidal flux is 0.0582526266 Wb, and the axis, with Za=12sin⁡2zZ_a=\frac{1}{2}\sin2z, is not planar.

Boozer harmonics. At ρ=0.8\rho=0.8 the Boozer spectrum B=∑m,nBmnccos⁡(mθB−nζB)+…B=\sum_{m,n}B_{mn}^{c}\cos(m\theta_B-n\zeta_B)+\ldots, with nn the full-torus mode number, has B00=0.4288B_{00}=0.4288 T. The leading amplitudes ∣Bmnc∣/B00\lvert B_{mn}^{c}\rvert/B_{00} are 0.117 for (m,n)=(1,0)(m,n)=(1,0), 0.111 for (0,4)(0,4), 0.054 for (1,−2)(1,-2), 0.036 for (1,2)(1,2), 0.028 for (1,−4)(1,-4) and 0.018 for (1,4)(1,4). Several helicities contribute, so the field is not quasisymmetric on this surface.

DESC. DESC 0.17.3 with two field periods solves the vacuum and the configurations of equation (11) with h=1h=1 (constant ff) or h=1−ρ02h=1-\rho_0^2 (radial ff), P∗=3×105P_* = 3\times10^5 Pa and η=0.05\eta=0.05, so P=15P=15 kPa. The transform is prescribed, and the three-dimensional p⊥p_\perp is fitted in Fourier–Zernike modes. The displayed runs start from the exact geometry fitted on DESC’s basis with λ=0\lambda=0. From (L,M,N)=(6,4,4)(L,M,N)=(6,4,4) to (11,9,11)(11,9,11) the maximum relative field error against the exact solution falls from at most 1.2×10−21.2\times10^{-2} to 2.4×10−62.4\times10^{-6} for the vacuum, 2.8×10−62.8\times10^{-6} for constant ff and 2.4×10−62.4\times10^{-6} for radial ff (Figure C2). At (11,9,11)(11,9,11) the normalized force residuals are at most 2.3×10−62.3\times10^{-6}. Vacuum points at N≥7N\geq7 reach the 100-iteration limit. The (12,10,12)(12,10,12) runs are excluded because the six-radius initial fit is rank deficient and gives an invalid starting geometry. For radial ff at (8,6,8)(8,6,8) the computed current agrees with equation (5) to 0.46% RMS.

Figure C2. DESC equilibria against the exact paraboloidal solution. Maximum relative vector-field error at common physical points on an independent validation grid, at eight resolutions from (L,M,N)=(6,4,4)(L,M,N)=(6,4,4) to (11,9,11)(11,9,11). The transform is prescribed.

Finite deformations. We evaluate the material CGL energy, equation (29) below, directly after boundary-fixed deformations X=x+∑iaiξi\mathbf{X}=\mathbf{x}+\sum_i a_i\boldsymbol{\xi}_i of both companions, with the physical volume V=1.2033 m3V=1.2033\ \mathrm{m}^3. Flux, mass and both CGL invariants are advected, and the pressure law is not imposed again. With xa(z)=x(1,0,z)\mathbf{x}_a(z)=\mathbf{x}(1,0,z), d=x−xa\mathbf{d}=\mathbf{x}-\mathbf{x}_a, dR=d⋅eRd_R=\mathbf{d}\cdot\mathbf{e}_R, dZ=d⋅eZd_Z=\mathbf{d}\cdot\mathbf{e}_Z, the envelope w=[1−(ψ−1)/0.02]2w=[1-(\psi-1)/0.02]^2 and L=0.1L=0.1 m, the six modes are

ξ1=wLcos⁡z eR,ξ2=wLsin⁡z eZ,ξ3=w(dReR+dZeZ),ξ4=wLsin⁡2z eϕ,ξ5=wL(cos⁡3z eR+sin⁡3z eZ),ξ6=w(dReR−dZeZ).(22)\begin{aligned} \boldsymbol{\xi}_1&=wL\cos z\,\mathbf{e}_R, & \boldsymbol{\xi}_2&=wL\sin z\,\mathbf{e}_Z, & \boldsymbol{\xi}_3&=w(d_R\mathbf{e}_R+d_Z\mathbf{e}_Z),\\ \boldsymbol{\xi}_4&=wL\sin2z\,\mathbf{e}_\phi, & \boldsymbol{\xi}_5&=wL(\cos3z\,\mathbf{e}_R+\sin3z\,\mathbf{e}_Z), & \boldsymbol{\xi}_6&=w(d_R\mathbf{e}_R-d_Z\mathbf{e}_Z). \tag*{(22)} \end{aligned}

For a=(0.15,−0.12,0.10,0.10,−0.08,0.08)\mathbf{a}=(0.15,-0.12,0.10,0.10,-0.08,0.08), the energy rises by 0.05802 PV for both companions. This agrees with the identity (G3) to 2×10−102\times10^{-10} on the finest grid. Forty random mixed deformations give ΔE/(PV)\Delta E/(PV) between 0.0049 and 0.113. Both signs of every mode raise the energy for coefficients from 10−410^{-4} to 0.3. For the energy minimizations we start from each mode alone at amplitude 0.3. These starts displace the plasma by up to 3.0 cm, keep the minimum Jacobian above 0.63 and raise the energy by 0.0089 PVPV (mode 4) to 0.209 PVPV (mode 3). BFGS minimization returns all twelve runs, six on each companion, to the reference’s numerical neighbourhood within 9 to 16 iterations, with coefficient norms below 1.4×10−91.4 \times10^{-9} and ∣ΔE∣/(PV)<5×10−17|\Delta E|/(PV) < 5 \times10^{-17} on the finest grid (Figure C3). The Hessian in this subspace has eigenvalues from 0.196 to 5.19.

Figure C3. Finite deformations raise the CGL energy. a, Sections at ϕ=0\phi= 0 of the six single-mode deformations ξ1\xi_1 to ξ6\xi_6 at amplitude 0.3. Modes 2 and 4 vary as sin⁡z\sin z and sin⁡2z\sin2z and are nearly invisible in this section. b, Energy during BFGS minimization from these six starts on both companions. All twelve runs return to the unperturbed reference. Energies are normalized by PV=18.05PV = 18.05 kJ and clipped at 10−1510^{-15} for display. The initial deformations displace the plasma by up to 3.0 cm.

Energy-monotone kinetic realization

We take constant ff and K=(1−f−2)/(2μ0)K = (1-f^{-2})/(2\mu_0), so that p∥=P+KB2p_{\parallel} = P + KB^2 and p⊥=P−KB2p_{\perp} = P - KB^2 on 0<Bmin⁡≤B≤Bmax⁡0 < B_{\min} \le B \le B_{\max}, in the guiding-center model with no electric potential. For F(E,μ)≥0F(\mathcal{E},\mu) \ge0 non-increasing in E\mathcal{E}, a superposition of energy steps gives

p∥=23CwB∫ν(b)(b−B)+3/2 db(23)p_{\parallel} = \frac{2}{3}C_wB \int\nu(b)(b-B)_{+}^{3/2}\,db \tag*{(23)}
p⊥=CwB2∫ν(b)(b−B)+1/2 db(24)p_{\perp} = C_wB^2 \int\nu(b)(b-B)_{+}^{1/2}\,db \tag*{(24)}

with a nonnegative measure ν\nu. One realization is F=w(μ)∫E/μ∞ν(b) dbF = w(\mu)\int_{\mathcal{E}/\mu}^{\infty}\nu(b)\,db with w≥0w \ge0 and Cw=4π2 mp−3/2∫0∞wμ3/2 dμC_w = 4\pi\sqrt{2}\,m_p^{-3/2}\int_0^\infty w\mu^{3/2}\,d\mu. A general μ\mu-dependent mixture gives the same representation after integration over μ\mu.

Necessity. With ϕ=p∥/B\phi= p_{\parallel}/B, the representation gives ϕ′=−p⊥/B2\phi' = -p_{\perp}/B^2 and ϕ′′=(2p⊥−B∂Bp⊥)/B3\phi'' = (2p_{\perp} - B\partial_Bp_{\perp})/B^3. Cauchy–Schwarz gives (ϕ′)2≤3ϕϕ′′(\phi')^2 \le3\phi\phi'', that is p⊥2≤3p∥(2p⊥−B∂Bp⊥)p_{\perp}^2 \le3p_{\parallel}(2p_{\perp} - B\partial_Bp_{\perp}). For the target law the bracket is 2P2P, which gives p⊥2≤6Pp∥p_{\perp}^2 \le6Pp_{\parallel}.

Sufficiency. Hats denote values at Bmax⁡B_{\max}. We set

Ψ(B)=3(P+KB2)2CwB,d=3Bmax⁡p^∥2p^⊥,a=Ψ(Bmax⁡)d−3/2,(25)\Psi(B)=\frac{3(P+KB^2)}{2C_wB},\qquad d=\frac{3B_{\max}\hat{p}_{\parallel}}{2\hat{p}_{\perp}},\qquad a=\Psi(B_{\max})d^{-3/2}, \tag*{(25)}
x=Bmax⁡−B,r(x)=Ψ(Bmax⁡−x)−a(d+x)3/2 andx=B_{\max}-B,\quad r(x)=\Psi(B_{\max}-x)-a(d+x)^{3/2}\text{ and}
ρ(x)=r′′(0)x−1/2+∫0xr′′′(t)(x−t)−1/2 dtπ Γ(5/2).(26)\rho(x)=\frac{r^{\prime\prime}(0)x^{-1/2}+\int_0^x r^{\prime\prime\prime}(t)(x-t)^{-1/2}\,dt}{\sqrt{\pi}\,\Gamma(5/2)}. \tag*{(26)}

The measure is the density ρ(Bmax⁡−b)\rho(B_{\max}-b) on [Bmin⁡,Bmax⁡][B_{\min},B_{\max}] plus an atom aδ(b−Bmax⁡−d)a\delta(b-B_{\max}-d). The choices of dd and aa give r(0)=r′(0)=0r(0)=r'(0)=0, and

r′′′(t)=9PCw(Bmax⁡−t)4+3a8(d+t)3/2>0,r′′(0)=3PCwBmax⁡3(1−p^⊥ 26Pp^∥).(27)r^{\prime\prime\prime}(t)=\frac{9P}{C_w(B_{\max}-t)^4}+\frac{3a}{8(d+t)^{3/2}}>0,\qquad r^{\prime\prime}(0)=\frac{3P}{C_wB_{\max}^3}\left(1-\frac{\hat{p}_{\perp}^{\,2}}{6P\hat{p}_{\parallel}}\right). \tag*{(27)}

The bound therefore makes ρ≥0\rho\geq0. Fractional integration gives ∫ν(b)(b−B)+3/2 db=Ψ(B)\int\nu(b)(b-B)_{+}^{3/2}\,db=\Psi(B), which is equation (23), and differentiation gives equation (24). For K<0K<0, p⊥/p∥p_{\perp}/p_{\parallel} increases with BB, so the bound at Bmax⁡B_{\max} implies it on the whole interval. For K≥0K\geq0 the bound holds automatically. A quadrature check at f=1.01f=\sqrt{1.01} reproduces the analytic moments to 4×10−154\times10^{-15}.

CGL quadratic energy

Particle number, frozen flux and the two CGL invariants p⊥/(nB)p_{\perp}/(nB) and p∥B2/n3p_{\parallel}B^2/n^3 are conserved. For a deformation gradient F\mathbf{F} with J=det⁡F\mathcal{J}=\det\mathbf{F} and ℓ=∣Fb∣\ell=|\mathbf{F}\mathbf{b}|,

n′=nJ,B′=BℓJ,p⊥′=p⊥ℓJ2,p∥′=p∥Jℓ2.(28)n'=\frac{n}{\mathcal{J}},\qquad B'=\frac{B\ell}{\mathcal{J}},\qquad p'_{\perp}=\frac{p_{\perp}\ell}{\mathcal{J}^2},\qquad p'_{\parallel}=\frac{p_{\parallel}}{\mathcal{J}\ell^2}. \tag*{(28)}

The internal energy density is p⊥′+p∥′/2p'_{\perp}+p'_{\parallel}/2, and the material potential energy per initial volume is

U=B2ℓ22μ0J+p⊥ℓJ+p∥2ℓ2.(29)U=\frac{B^2\ell^2}{2\mu_0\mathcal{J}}+\frac{p_{\perp}\ell}{\mathcal{J}}+\frac{p_{\parallel}}{2\ell^2}. \tag*{(29)}

Its first variation is the virtual work of the total stress, so the stress identity of Appendix § 17 gives stationarity.

For F=I+ϵA\mathbf{F}=\mathbf{I}+\epsilon\mathbf{A} with Aij=∂jξiA_{ij}=\partial_j\xi_i, we write c=tr⁡Ac=\operatorname{tr}A, s=biAijbjs=b_iA_{ij}b_j and t=∣Ab∣2t=|A\mathbf{b}|^2. Then

J=1+ϵc+ϵ22(c2−tr⁡A2)+O(ϵ3),ℓ2=1+2ϵs+ϵ2t.(30)\mathcal{J}=1+\epsilon c+\frac{\epsilon^2}{2}(c^2-\operatorname{tr}A^2)+O(\epsilon^3),\qquad\ell^2=1+2\epsilon s+\epsilon^2t. \tag*{(30)}

We subtract the magnetic energy of the seed frozen through the same deformation, B02ℓ2/(2μ0J)B_0^2\ell^2/(2\mu_0\mathcal{J}). With p∥,⊥=P±Qp_{\parallel,\perp}=P\pm Q and Q=(B2−B02)/(2μ0)Q=(B^2-B_0^2)/(2\mu_0), the second-order coefficient is

12[2Pc2−2(P+Q)cs+(3P+5Q)s2]+P2(tr⁡A2−c2).(31)\frac{1}{2}\left[2Pc^2-2(P+Q)cs+(3P+5Q)s^2\right]+\frac{P}{2}(\operatorname{tr}A^2-c^2). \tag*{(31)}

For constant PP and ξ=0\boldsymbol{\xi}=0 on the boundary, the last term integrates to zero. The seed term gives ∫∣∇×(ξ×B0)∣2dV/(2μ0)\int|\nabla\times(\boldsymbol{\xi}\times\mathbf{B}_0)|^2dV/(2\mu_0), which is equation (7). The local matrix

M=(2P−(P+Q)−(P+Q)3P+5Q),det⁡M=5P2+8PQ−Q2=6Pp∥−p⊥2,(32)\mathbf{M}=\begin{pmatrix}2P&-(P+Q)\\-(P+Q)&3P+5Q\end{pmatrix},\qquad\det\mathbf{M}=5P^2+8PQ-Q^2=6Pp_{\parallel}-p_{\perp}^2, \tag*{(32)}

is nonnegative under the bound, so δW≥0\delta W\geq0. Positive mass density then excludes exponentially growing ideal CGL normal modes in this boundary class.

Kinetic closure and Taylor–Hastie stability for constant ff

We use the low-frequency small-Larmor-radius guiding-center energy principle of Kruskal and Oberman (Ref. [31]) and Taylor and Hastie (Ref. [29]), with globally constant ff, no equilibrium electric potential, ξ=0\boldsymbol{\xi}=0 on the boundary and closed particle dynamics that conserve energy, μ\mu and phase-space volume.

Coefficients. Parallel force balance gives

B∂Bp∥=p∥−p⊥,B∂Bp⊥−2p⊥=−2P,1+μ0(p⊥−p∥)B2=1+μ0B∂Bp⊥=f−2.(33)B\partial_Bp_{\parallel}=p_{\parallel}-p_{\perp},\quad B\partial_Bp_{\perp}-2p_{\perp}=-2P,\quad1+\frac{\mu_0(p_{\perp}-p_{\parallel})}{B^2}=1+\frac{\mu_0}{B}\partial_Bp_{\perp}=f^{-2}. \tag*{(33)}

These are exact, not low-beta approximations.

Moment identity. With c=∇⋅ξc=\nabla\cdot\boldsymbol{\xi}, h=bibj∂jξih=b_i b_j\partial_j\xi_i and As=msv∥2h+μB(c−h)A_s=m_s v_{\parallel}^2h+\mu B(c-h), integration by parts in v∥v_{\parallel} and summation over species give

∑s∫(−Fs,E)(μB)2 d3v=2P,∑s∫(−Fs,E)(μB)(msv∥2) d3v=p⊥,∑s∫(−Fs,E)(msv∥2)2 d3v=3p∥.(34)\sum_s\int(-F_{s,\mathcal{E}})(\mu B)^2\,d^3v=2P,\quad\sum_s\int(-F_{s,\mathcal{E}})(\mu B)(m_sv_{\parallel}^2)\,d^3v=p_{\perp},\quad\sum_s\int(-F_{s,\mathcal{E}})(m_sv_{\parallel}^2)^2\,d^3v=3p_{\parallel}. \tag*{(34)}

The Gram determinant is 6Pp∥−p⊥26Pp_{\parallel}-p_{\perp}^2, and

2P(c−h)2+2p⊥(c−h)h+3p∥h2=∑s∫(−Fs,E)As2 d3v.(35)2P(c-h)^2+2p_{\perp}(c-h)h+3p_{\parallel}h^2=\sum_s\int(-F_{s,\mathcal{E}})A_s^2\,d^3v. \tag*{(35)}

The Kruskal–Oberman energy replaces As2A_s^2 by the square of its orbit average,

δWKO=12μ0∫∣∇×(ξ×B0)∣2 dV+12∑s∫(−Fs,E)⟨As⟩2 dΓs≥0,(36)\delta W_{\mathrm{KO}}=\frac{1}{2\mu_0}\int\lvert\nabla\times(\boldsymbol{\xi}\times\mathbf{B}_0)\rvert^2\,dV+\frac{1}{2}\sum_s\int(-F_{s,\mathcal{E}})\langle A_s\rangle^2\,d\Gamma_s\ge0, \tag*{(36)}

and δWCGL−δWKO=12∑s∫(−Fs,E)(As−⟨As⟩)2dΓs≥0\delta W_{\mathrm{CGL}}-\delta W_{\mathrm{KO}}=\frac{1}{2}\sum_s\int(-F_{s,\mathcal{E}})(A_s-\langle A_s\rangle)^2d\Gamma_s\ge0. For the step distributions of Appendix D, −Fs,E-F_{s,\mathcal{E}} is a nonnegative measure and enters linearly, so these integrals hold in the Stieltjes sense when they are finite. As stated in the main text, using them as a dynamical statement requires a finite quadratic-form or justified limiting interpretation.

Current and neutrality. With κB=(b⋅∇)b\boldsymbol{\kappa}_B=(\mathbf{b}\cdot\nabla)\mathbf{b}, an even distribution in v∥v_{\parallel} gives

Jgc=p⊥B2b×∇B+p∥Bb×κB,M=−p⊥Bb,Jgc+∇×M=1Bb×[∇p⊥+(p∥−p⊥)κB]=0,(37)\mathbf{J}_{\mathrm{gc}}=\frac{p_{\perp}}{B^2}\mathbf{b}\times\nabla B+\frac{p_{\parallel}}{B}\mathbf{b}\times\boldsymbol{\kappa}_B,\quad\mathbf{M}=-\frac{p_{\perp}}{B}\mathbf{b},\quad\mathbf{J}_{\mathrm{gc}}+\nabla\times\mathbf{M}=\frac{1}{B}\mathbf{b}\times[\nabla p_{\perp}+(p_{\parallel}-p_{\perp})\boldsymbol{\kappa}_B]=0, \tag*{(37)}

using the vacuum curvature κB=∇⊥ln⁡B\boldsymbol{\kappa}_B=\nabla_{\perp}\ln B and p⊥=P−KB2p_{\perp}=P-KB^2. This agrees with ∇×B=0\nabla\times\mathbf{B}=0. For singly charged ions and electrons with the same step measure in E/μ\mathcal{E}/\mu and the same moment ratio RμR_{\mu}, the amplitudes can be chosen so that Pi=Pe=P/2P_i=P_e=P/2 and Ki=Ke=K/2K_i=K_e=K/2. Then ns=Rμp⊥s/Bn_s=R_{\mu}p_{\perp s}/B gives ni=nen_i=n_e at every point.

Finite-deformation CGL minimum

We take smooth orientation-preserving material maps X(a)\mathbf{X}(\mathbf{a}) of the domain, connected to the identity through maps fixed at the boundary, with F=∂X/∂a\mathbf{F}=\partial\mathbf{X}/\partial\mathbf{a}, J>0\mathcal{J}>0 and ℓ>0\ell>0. Mass, flux and the two CGL invariants are advected. For uniform P>0P>0 and 0≤Q<P0\le Q<P, which allows radial ff, the reference state minimizes the potential energy (E2).

Proof. We freeze the seed through the same map. The remaining density is

R(J,ℓ)=Qℓ2+(P−Q)ℓJ+P+Q2ℓ2.(38)R(\mathcal{J},\ell)=\frac{Q\ell^2+(P-Q)\ell}{\mathcal{J}}+\frac{P+Q}{2\ell^2}. \tag*{(38)}

with R(1,1)=(3P+Q)/2R(1,1)=(3P+Q)/2. The domain is unchanged, so ∫Jd3a=∫d3a\int_{\mathcal{J}}d^3a=\int d^3a, and we may add P(J−1)P(\mathcal{J}-1). With G=R+PJ−(5P+Q)/2G=R+P\mathcal{J}-(5P+Q)/2, q=Q/P∈[0,1)q=Q/P\in[0,1) and Aq=qℓ2+(1−q)ℓA_q=q\ell^2+(1-q)\ell, the arithmetic–geometric mean inequality and Aq≥qℓ+(1−q)ℓ\sqrt{A_q}\ge q\ell+(1-q)\sqrt{\ell} give

GP≥2Aq+1+q2ℓ2−5+q2≥q(ℓ−1)2(2ℓ+1)ℓ2+(1−q)(x−1)2(4x3+3x2+2x+1)2x4≥0,(39)\frac{G}{P}\ge2\sqrt{A_q}+\frac{1+q}{2\ell^2}-\frac{5+q}{2}\ge q\frac{(\ell-1)^2(2\ell+1)}{\ell^2}+(1-q)\frac{(x-1)^2(4x^3+3x^2+2x+1)}{2x^4}\ge0, \tag*{(39)}

with x=ℓx=\sqrt{\ell}. Equality requires ℓ=J=1\ell=\mathcal{J}=1. For the seed, along an ideal path with v=0\mathbf{v}=0 on the boundary, ∇×B0=0\nabla\times\mathbf{B}_0=0 gives d∫B0⋅B0,tdV/dt=0d\int\mathbf{B}_0\cdot\mathbf{B}_{0,t}dV/dt=0, with no single-valued scalar potential needed. Hence

ECGL[X]−ECGL[id]=12μ0∫∣B0,X−B0∣2dV+∫G d3a≥0.(40)E_{\mathrm{CGL}}[\mathbf{X}]-E_{\mathrm{CGL}}[\mathrm{id}]=\frac{1}{2\mu_0}\int\lvert\mathbf{B}_{0,\mathbf{X}}-\mathbf{B}_0\rvert^2dV+\int G\,d^3a\ge0. \tag*{(40)}

Lyapunov functional and Taylor relaxation. Adding the kinetic energy 12∫ρ0∣X˙∣2d3a\frac{1}{2}\int\rho_0\lvert\dot{\mathbf{X}}\rvert^2d^3a gives a nonnegative functional that smooth ideal CGL motion conserves. It bounds the kinetic energy and ∫∣B0,X−B0∣2dV\int\lvert\mathbf{B}_{0,\mathbf{X}}-\mathbf{B}_0\rvert^2dV while the motion stays smooth. Taylor relaxation (Ref. [35]) fixes magnetic helicity and gives a force-free field. Here the material flux topology, the mass and the two CGL invariants are fixed instead, and the minimum carries perpendicular current and anisotropic pressure.

Electrostatic stability

We fix the magnetic field and use the Hamiltonian guiding-center–Poisson model with μ\mu conserved, an invariant phase-space measure dΓsd\Gamma_s and a closed, conservative boundary that does no work. The equilibrium is f0s=Fs(Es,μ)f_{0s}=F_s(\mathcal{E}_s,\mu) with ϕ0=0\phi_0=0 and charge balance, and the energy is

Hes=∑s∫Esfs dΓs+ϵ02∫∣∇ϕ∣2dV,−ϵ0∇2ϕ=∑sqs∫fs dΓv,s.(41)\mathcal{H}_{\mathrm{es}}=\sum_s\int\mathcal{E}_sf_s\,d\Gamma_s+\frac{\epsilon_0}{2}\int\lvert\nabla\phi\rvert^2dV,\qquad-\epsilon_0\nabla^2\phi=\sum_s q_s\int f_s\,d\Gamma_{v,s}. \tag*{(41)}

Smooth decreasing distributions. For Fs,E<0F_{s,\mathcal{E}}<0, the unperturbed advection LsL_s is antisymmetric with LsFs,E=0L_sF_{s,\mathcal{E}}=0, and ∂tδfs+Lsδfs=qsFs,ELsδϕ\partial_t\delta f_s+L_s\delta f_s=q_sF_{s,\mathcal{E}}L_s\delta\phi. Multiplying by δfs/(−Fs,E)\delta f_s/(-F_{s,\mathcal{E}}) and integrating gives a rate ∫δj⋅δE dV\int\delta\mathbf{j}\cdot\delta\mathbf{E}\,dV, and the field energy changes at the opposite rate. Hence equation (17) is conserved and positive, and no mode with finite norm grows. The convex Casimirs CsC_s with Cs,f(Fs,μ)=−EC_{s,f}(F_s,\mu)=-\mathcal{E} and Cs,ff=−1/Fs,EC_{s,ff}=-1/F_{s,\mathcal{E}} give the conserved relative energy–Casimir functional

Les=∑s∫[Cs(fs)−Cs(f0s)−Cs,f(f0s)(fs−f0s)]dΓs+ϵ02∫∣∇ϕ∣2dV≥0,(42)\mathcal{L}_{\mathrm{es}}=\sum_s\int\left[C_s(f_s)-C_s(f_{0s})-C_{s,f}(f_{0s})(f_s-f_{0s})\right]d\Gamma_s+\frac{\epsilon_0}{2}\int\lvert\nabla\phi\rvert^2dV\ge0, \tag*{(42)}

whose quadratic part is that invariant.

Steps and plateaus. The distributions of Appendix D need not be strictly decreasing. At each μ\mu the phase-space volume of every superlevel set {fs>a}\{f_s>a\} is conserved, and filling the lowest energies minimizes ∫EsdΓs\int\mathcal{E}_s d\Gamma_s at fixed volume. The superlevel sets of a non-increasing FsF_s already have this form, so integration over aa gives ∑s∫Es(fs−f0s)dΓs≥0\sum_s\int\mathcal{E}_s(f_s-f_{0s})d\Gamma_s\ge0 for every such rearrangement. With the field energy added,

ϵ02∫∣∇ϕ(t)∣2dV≤Hes[f(0),ϕ(0)]−Hes[f0,0].(43)\frac{\epsilon_0}{2}\int\lvert\nabla\phi(t)\rvert^2dV\le\mathcal{H}_{\mathrm{es}}[f(0),\phi(0)]-\mathcal{H}_{\mathrm{es}}[f_0,0]. \tag*{(43)}

Action contours and rotational transform of the N=3N=3 configuration

At fixed (E,μ)(\mathcal{E},\mu) the bounce-averaged drift over the labels (q1,q2)(q_1,q_2) of nearby trapped wells is q˙1=C∂q2J\dot q_1=C\partial_{q_2}J, q˙2=−C∂q1J\dot q_2=-C\partial_{q_1}J, with a smooth nonzero CC, so J˙=0\dot J=0. A definite nondegenerate Hessian at a stationary point gives nearby closed contours, as long as the mirror points stay simple and the well persists.

At Bm=29.9B_m = 29.9, 30 and 30.1 the reduced action jj has minima 0.1488355, 0.1545035 and 0.1601559. The smallest Hessian eigenvalues are 0.760, 0.762 and 0.764, and the largest are 52.0, 52.4 and 52.9. The sixteen contour roots lie on surface labels [20.3555, 20.3709], [20.3654, 20.3809] and [20.3753, 20.3908], with refinement residuals below 1.2×10−91.2 \times10^{-9}. Over all 32 mirror strengths Bm=28.5,28.6,…,31.6B_m = 28.5, 28.6, \ldots, 31.6 of Figure 3b, the smallest eigenvalue rises from 0.737 to 0.798 and the largest from 48.0 to 63.2. Every sampled boundary gap is positive, the refinement residuals stay below 3.6×10−93.6 \times10^{-9}, and the contour labels move outward from [20.213, 20.228] to [20.515, 20.530]. The magnetic axis has label 19.

Field lines of the exact seed, traced over 160 periods and averaged with a weighted Birkhoff average, give rotational transforms 0.1968195, 0.1915942, 0.1846528 and 0.1837743 on the labels 19.6, 20.0, 20.355 and 20.39. The axis value is 2/7=0.2020305\sqrt{2}/7 = 0.2020305. Half-length averages agree to 4×10−64 \times10^{-6} on the labels 20.355 and 20.39. The transform therefore falls by 8.6 to 9.0% from the axis to the surfaces of the contours in Figure 3a, and this configuration is sheared.

References

  1. [1]Helander, P. Theory of plasma confinement in non-axisymmetric magnetic fields. Rep. Prog. Phys. 77, 087001 (2014).DOI
  2. [2]Boozer, A. H. Transport and isomorphic equilibria. Phys. Fluids 26, 1288–1291 (1983).
  3. [3]Nührenberg, J. & Zille, R. Quasi-helically symmetric toroidal stellarators. Phys. Lett. A 129, 113–117 (1988).DOI
  4. [4]Landreman, M. & Paul, E. Magnetic fields with precise quasisymmetry for plasma confinement. Phys. Rev. Lett. 128,035001 (2022).DOI
  5. [5]Hall, L. S. & McNamara, B. Three-dimensional equilibrium of the anisotropic, finite-pressure guiding-center plasma: Theory of the magnetic plasma. Phys. Fluids 18, 552–565 (1975).
  6. [6]Cary, J. R. & Shasharina, S. G. Omnigenity and quasihelicity in helical plasma confinement systems. Phys. Plasmas 4, 3323–3333 (1997).DOI
  7. [7]Beidler, C. D. et al. Demonstration of reduced neoclassical energy transport in Wendelstein 7-X. Nature 596, 221–226 (2021).DOI
  8. [8]Grad, H. Toroidal containment of a plasma. Phys. Fluids 10, 137–154 (1967).DOI
  9. [9]Garren, D. A. & Boozer, A. H. Existence of quasihelically symmetric stellarators. Phys. Fluids B 3, 2822–2834 (1991).DOI
  10. [10]Landreman, M. & Sengupta, W. Constructing stellarators with quasisymmetry to high order. J. Plasma Phys. 85, 815850601 (2019).DOI
  11. [11]Gómez-Serrano, J. and Liehr, L. and Taylor, M. A. Counterexamples to Grad’s conjecture. arXiv:2609.24739 (2026).
  12. [12]Landreman, M. Analytic toroidal 3D MHD equilibria and steady Euler flows with invariant surfaces. arXiv:2609.26742v2 (2026).arxiv.org/abs/2609.26742
  13. [13]Issan, O., Citrin, J., Landreman, M. & Tracey, B. Analytic non-axisymmetric 3D MHD equilibria without stellarator symmetry. Preprint at https://arxiv.org/abs/2610.07304 (2026).
  14. [14]Issan, O., José, L., Landreman, M., Citrin, J. & Tracey, B. Non-axisymmetric MHD equilibria with an arbitrary number of field periods. Preprint at https://arxiv.org/abs/2611.07672 (2026).
  15. [15]Sengupta, W. et al. Exact Nonsymmetric Vacuum Magnetic Fields with Nested Toroidal Surfaces. Preprint at https://arxiv.org/abs/2610.28466 (2026).
  16. [16]Chew, G. F., Goldberger, M. L. & Low, F. E. The Boltzmann equation and the one-fluid hydromagnetic equations in the absence of particle collisions. Proc. R. Soc. Lond. A 236, 112–118 (1956).DOI
  17. [17]Cooper, W. A. et al. Anisotropic pressure bi-Maxwellian distribution function model for three-dimensional equilibria. Nucl. Fusion 46, 683 (2006).DOI
  18. [18]Dudt, D. W. & Kolemen, E. DESC: A stellarator equilibrium solver. Phys. Plasmas 27, 102513 (2020).DOI
  19. [19]Panici, D., Conlin, R., Dudt, D. W., Unalmis, K. & Kolemen, E. The DESC stellarator code suite. Part 1. Quick and accurate equilibria computations. J. Plasma Phys. 89, 955890303 (2023).DOI
  20. [20]Rodríguez, E. & Bhattacharjee, A. Connection between quasisymmetric magnetic fields and anisotropic pressure equilibria in fusion plasmas. Phys. Rev. E 104, 015213 (2021).DOI
  21. [21]Kruskal, M. D. & Kulsrud, R. M. Equilibrium of a magnetically confined plasma in a toroid. Phys. Fluids 1, 265–274 (1958).
  22. [22]Fu, L. & Bhattacharjee, A. A Variational Framework for Guiding-Center Kinetics, Anisotropic Equilibria, and Quasisymmetry in Stellarators. Preprint at https://arxiv.org/abs/2607.06903v2 (2026).
  23. [23]Clemente, R. A. Anisotropic axisymmetric equilibria via an analytic method. Nucl. Fusion 33, 963–965 (1993).DOI
  24. [24]Cheviakov, A. F. & Bogoyavlenskij, O. I. Exact anisotropic MHD equilibria. J. Phys. A: Math. Gen. 37, 7593–7607 (2004).DOI
  25. [25]Cheviakov, A. F. & Anco, S. C. Analytical properties and exact solutions of static plasma equilibrium systems in three dimensions. Phys. Lett. A 372, 1363–1373 (2008).DOI
  26. [26]Cheviakov, A. F. Plasma equilibrium equations in coordinates connected with magnetic surfaces. Exact equilibrium solutions. Preprint at https://arxiv.org/abs/math-ph/0410045 (2004).
  27. [27]Cheviakov, A. F. Symmetries and exact solutions of plasma equilibrium equations. PhD thesis, Queen’s University, Kingston, Ontario (2004).
  28. [28]Sato, N. & Yamada, M. Nested invariant tori foliating a vector field and its curl: toward MHD equilibria and steady Euler flows in toroidal domains without continuous Euclidean isometries. J. Math. Phys. 64, 081505 (2023).DOI
  29. [29]Taylor, J. B. & Hastie, R. J. Maximum plasma pressure for stability in magnetic fields with finite minima. Phys. Fluids 8, 323–332 (1965).DOI
  30. [30]Newcomb, W. A. Lagrangian and Hamiltonian methods in magnetohydrodynamics. Nucl. Fusion Suppl. Part 2, 451–463 (1962).
  31. [31]Kruskal, M. D. & Oberman, C. R. On the stability of plasma in static equilibrium. Phys. Fluids 1, 275–280 (1958).
  32. [32]Taylor, J. B. Some stable plasma equilibria in combined mirror-cusp fields. Phys. Fluids 6, 1529–1536 (1963).DOI
  33. [33]Clemente, R. A., Sakanaka, P. H. & Rincoski, C. R. M. On the stability of anisotropic field-reversed configurations. Plasma Phys. Control. Fusion 37, 1381–1388 (1995).DOI
  34. [34]Clemente, R. A. & Cesar, R. G. F. Anisotropic effects on classical particle confinement time in field-reversed configurations. Plasma Phys. Control. Fusion 37, 137–143 (1995).DOI
  35. [35]Taylor, J. B. Relaxation and magnetic reconnection in plasmas. Rev. Mod. Phys. 58, 741–763 (1986).DOI
  36. [36]Gardner, C. S. Bound on the energy available from a plasma. Phys. Fluids 6, 839–840 (1963).DOI
  37. [37]Helander, P. Available energy and ground states of collisionless plasmas. J. Plasma Phys. 83, 715830401 (2017).
  38. [38]Mercier, C. and Cotsafitis, M. Equilibrium and stability of an axially symmetric plasma with anisotropic pressure. Nucl. Fusion 1, 121 (1961).
  39. [39]Lortz, D. & Spies, G. O. Anisotropic magnetohydrostatic equilibria with arbitrary magnetic fields. Phys. Plasmas 10, 553 (2003).DOI
  40. [40]Sato, N., Qu, Z., Pfefferlé, D. & Dewar, R. L. Quasisymmetric magnetic fields in asymmetric toroidal domains. Phys. Plasmas 28, 112507 (2021).DOI
  41. [41]Sato, N. Existence of weakly quasisymmetric magnetic fields without rotational transform in asymmetric toroidal domains. Sci. Rep. 12, 11322 (2022).DOI
  42. [42]Kaltsas, D. A., Kuirokouidis, A. I. & Throumoulopoulos, G. N. A special exact class of three-dimensional equilibria of a magnetized plasma exhibiting closed and nested toroidal magnetic surfaces. Preprint at https://arxiv.org/abs/2603.20982 (2026).
  43. [43]Clemente, R. A. A method for obtaining rotating anisotropic toroidal equilibria. Plasma Phys. Control. Fusion 36, 707–710 (1994).DOI
  44. [44]Clemente, R. A. & Sterzo, D. Anisotropic ideal magnetohydrodynamic cylindrical equilibria with incompressible adiabatic flow. Plasma Phys. Control. Fusion 51, 085011 (2009).DOI
  45. [45]Holm, D. D., Marsden, J. E., Ratiu, T. & Weinstein, A. Nonlinear stability of fluid and plasma equilibria. Phys. Rep. 123, 1–116 (1985).DOI
  46. [46]Webb, G. M., Anco, S. C., Meleshko, S. V. & Zank, G. P. Action principles and conservation laws for Chew–Goldberger–Low anisotropic plasmas. J. Plasma Phys. 88, 835880402 (2022).DOI

Paper details

vlnc:W_hcjfk5dz
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

Claude Code, Codex

View statement
Looking for related papers…
Contents