betaFeedback

WorksPhysics and Astronomy

Elasticized magnetohydrostatics

Abstract

The modern approach to modeling equilibria in stellarators features two blind spots: lack of foundational agreement between theory and computation, and inability to reliably explore economically-attractive high-ββ concepts. These blind spots originate at a technical level from lack of a useful existence theory for the partial differential equations believed to describe equilibria, and from the encumbrance of crude, inflexible models for the topology of the confining magnetic field. This Article proposes the first stellarator equilibrium model that rigorously guarantees existence of an equilibrium with any chosen magnetic topology. It completely separates determination of the magnetic topology from the equilibrium existence question, and thereby enables importing realistic tearing physics into equilibrium computations. The new model replaces the standard plasma potential energy functional with its ensemble average over rapid kinetic-scale fluctuations. Averaging ``elasticizes" the potential by energetically penalizing separation of field lines. The elasticized potential's super-linear dependence on plasma strain embodies a phenomenological model for profile stiffness, and plays an essential role in the existence theory.

Significance. The design of stellarator fusion reactors requires solving an optimization problem in a very high dimensional space. Analytical theory is meant to serve as a light that guides optimizers through this formidable design space. But it partially fails in this respect due to foundational disagreements between theory and computation and incomplete coverage of high-pressure operation regimes. This work proposes a new framework for modeling steady-state stellarator configurations that repairs the disconnect between theory and computation, and unlocks high-pressure reactor design.

Stellarator fusion is advancing rapidly, driven by the success of Wendelstein 7-X (1) and growing private investment [4, 5] [2, 3]. A central challenge for a stellarator fusion pilot plant is the ability to reliably predict and optimize three-dimensional stellarator equilibria, including the magnetic field BB and plasma pressure pp. However, the current approach to equilibrium modeling features two blind spots: lack of foundational agreement between theory and computation, and inability to reliably explore economically-attractive high-β\beta concepts. Here β\beta denotes the ratio of plasma pressure to magnetic energy density – a number that quantifies how efficiently the magnetic field confines the plasma fuel. These blind spots originate at a technical level from lack of a useful existence theory for the magnetohydrostatics (MHS) equations,

(∇×B)×B=∇p,∇⋅B=0(\nabla\times B) \times B = \nabla p, \qquad\nabla\cdot B = 0

and from the encumbrance of crude, inflexible models for the topology of the confining magnetic field. These blind spots limit the ability of theory to guide design optimization, which is essential for development of a viable fusion pilot plant.

Theory contradicts computation — Notable disagreements between theory and computation touch upon various foundational questions in stellarator confinement. Assuming MHS, Garren-Boozer [6] argued quasisymmetric (QS) stellarator equilibria cannot exist. Using VMEC, Landreman-Paul [8] identified numerical equilibria with precise QS in the entire plasma volume. Assuming MHS, Cary-Shasharina [9, 10] show analytic omnigenity (OM) implies QS. Using VMEC, Goodman et al. [11] find vacuum fields (automatically analytic!) with precise OM and significant deviations from QS. More broadly, the theory of tearing modes predicts smooth current density across rational flux surfaces, while equilibrium computations routinely reveal singular current densities in the same locations [12].

Disagreements between theory and computation of stellarator equilibria plausibly originate from lack of a useful MHS existence theory. In the 1960s Harold Grad recognized severe mathematical difficulties posed by these equations in non-symmetric plasma containers. He conjectured [13, 14, 15] non-existence of smooth solutions with nested flux surfaces. Recent counterexamples [16, 17] show that Grad’s conjecture is false. However, a more practical weakening of Grad’s conjecture still stands. It asserts smooth MHS solutions fail to exist for generic plasma boundary shape and profile functions. A significant body of computational work [12, 18] supports the weakened conjecture. Moreover, the recent counterexamples fall far short of addressing the weakened conjecture because they only include finite-dimensional families of plasma boundary shapes. In spite of the plausible truth of a weakened Grad conjecture, most of stellarator confinement theory continues to assume MHS is actually satisfied in a stellarator. This creates a potential for vacuous theoretical claims and predictions. On the other hand, codes like VMEC do not actually solve the MHS equations. Instead they solve implicitly-defined resolution-dependent regularizations of the MHS model. This sets up ideal conditions for vacuous theoretical predictions manifesting as contradictions between theory and computation.

The mathematical difficulties identified by Grad may not implicate the MHS model per se, but rather the assumption of nested flux surfaces. There is some rigorous mathematical support for this perspective. Bruno-Laurence [19] found nearly-axisymmetric MHS solutions comprising nested toroidal annula in which the magnetic field is Beltrami. Their solutions suffer jump discontinuities across toroidal interfaces, but generally contain magnetic islands and chaos [20]. The equilibrium code SPEC [21] computes this sort of equilibrium in domains far from axisymmetry, where the Bruno-Laurence theory does not apply. More recently, Constantin-Pasqualotto [22] showed that a particular form of magnetic relaxation, similar in spirit to methods implemented in some iterative equilibrium solvers [23, 24, 25], produces non-trivial MHS solutions in any domain. While all of these results are suggestive, they fail to provide a useful MHS existence theory for stellarators. The Bruno-Laurence solutions have only been rigorously established in nearly-axisymmetric domains, and numerical evidence [26] suggests they need not exist in strongly non-axisymmetric domains. While the Constantin-Pasqualotto solutions do exist in stellarator-relevant domains, their suitability for stellarator confinement remains shrouded in mystery. The existence theory does not show these configurations support a sufficient volume of flux surfaces to realistically achieve burning plasma conditions.

Elusive high-β\beta designs — Higher field strengths translate to larger magnetic coils, which translates to much higher construction costs. High-β\beta stellarator designs therefore generally make more efficient use of their applied magnetic field. But most design optimization studies use fast nested-surface equilibrium solvers like VMEC [7] or DESC [27, 28] inside of the optimization loop. These solvers are doomed to fail sufficiently close to the so-called equilibrium β\beta-limit [29, 30], where flux surfaces degrade without triggering an ideal MHD instability. Conditioned on the assumption of piecewise Beltrami fields, the high impact of allowing for non-nested magnetic topology during design optimization with β>0\beta> 0 was demonstrated directly [31] through use of the SPEC in place of VMEC or DESC. If the piecewise-Beltrami assumption was known to accurately reproduce the true tearing physics [32, 33, 34, 35] that governs magnetic topology in stellarators, these results would indicate SPEC offers the key to high-β\beta design. However, no existing stellarator equilibrium model, including SPEC, faithfully reproduces tearing physics due to pervasive use of crude magnetic topology models. High-β\beta designs remain elusive.

The crude nature of current approaches to equilibrium magnetic topology modeling is reflected in the most widespread simulation tools used for stellarator equilibrium modeling. The codes VMEC [7] and DESC [27, 28] assume the plasma-region magnetic field is tangent to concentric toroidal flux surfaces. This assumption is reasonable in low-β\beta configurations, but is known to fail at larger β\beta [29, 30, 36, 37]. Starting in the late 1980s, the potential for non-nested magnetic topologies in stellarators motivated development of non-nested 3D equilibrium solvers. Iterative solvers, such as PIES [38, 39], SIESTA [40], HINT2 [25], and MRX [41] introduce physically-motivated relaxation effects that numerically generate magnetic islands and chaos, usually starting from a nested initial guess. Other solvers like SPEC [21] and SPECTRE [42] target weak solutions of MHS by extending the rigorous Bruno-Laurence construction [19] to strongly-shaped domains [26]. Iterative schemes do not actually simulate the tearing physics responsible for island formation and saturation. Instead they introduce numerical expedients with questionable physics fidelity. In solvers based on Bruno-Laurence theory, the field in between toroidal interfaces is necessarily Beltrami. This leads to a correct topology model only when all ideal constraints except helicity between interfaces are removed. Potential issues with these crude modeling choices are only compounded by their central, inflexible role in the underlying simulation algorithms. For example, users may not freely specify the embedded topology model in any of these codes, even if a better model is available.

A new approach — This Article proposes the first rigorous energy minimization principle for stellarator equilibria with user-specified magnetic topology. Due to its close connections with the theory of hyperelasticity, this equilibrium modeling framework will be referred to as elasticized magnetohydrostatics.

The Article begins by appealing to physics beyond ideal MHD to address a major technical hurdle in minimizing the standard plasma potential energy due to Newcomb [43]: lack of coercivity. § 1 argues that “elasticizing” Newcomb’s magnetic energy by including the mean energy of kinetic-scale magnetic fluctuations improves the functional’s growth properties. This argument extends those used to formulate a related equilibrium model in Ref. [44]. § 2 then argues further that combining the elasticized magnetic energy with a modified thermodynamic internal energy eliminates all apparent obstacles to coercivity. The modified internal energy differs from the usual ideal gas internal energy by a soft vacuum barrier that energetically penalizes excessive rarefaction. Based on the observations from § 1 and § 2, § 3 presents a precise mathematical formulation of elasticized magnetohydrostatics that defines equilibria as minimizers of a modified potential energy functional. As with Newcomb’s variational principle, this model fixes the underlying magnetic topology by specifying a reference magnetic field. Unlike Newcomb’s model, the potential energy functional is the sum of the modified internal energy from § 2 and the elasticized magnetic energy from § 1. Refining the heuristic arguments from § 1 and § 2, § 4 rigorously proves coercivity of the elasticized potential energy functional. Leveraging the coercivity property and analytical techniques developed in the context of elasticity theory, § 5 applies the direct method in the calculus of variations to prove existence of minimizers for the modified potential in the Sobolev space W1,pW^{1,p}, comprising LpL^p back-to-labels maps [45] with gradients that lie in LpL^p. Here the exponent pp is set by the nonlinear response of magnetic fluctuations to large-scale plasma deformations. Notably, Theorem 2 provides a minimum potential energy state for any specified magnetic topology whatsoever, including nested, islanded, and chaotic configurations. We are not aware of any comparable results in the literature.

By allowing a user complete freedom to specify equilibrium magnetic topology, as illustrated in Figure 1, elasticized magnetohydrostatics successfully separates magnetic topology modeling from the equilibrium existence problem. It is therefore the first stellarator equilibrium model that provides a direct interface to sophisticated magnetic topology models based on real tearing physics. It also comes equipped with the first useful existence theory for equilibria in the sense that inputs to their existence theorem directly describe essential features of the solution.

Reference and computed field-line Poincaré maps

Figure 1. Reference (a) and computed (b) field-line Poincaré map using elasticized magnetohydrostatics. Two simulations are compared left-to-right in each figure, differing only by the user-specified magnetic topology. The left simulation features a single 3:13:1 integrable island chain, while the right features overlapping 3:13:1 and 5:25:2 island chains. Euler-Lagrange equation residuals are 1.4×10−71.4 \times10^{-7} in each case.

Elasticized magnetic energy

In Newcomb’s variational formulation of ideal MHD [43], the plasma configuration space comprises diffeomorphisms G:Q→Q0G: Q \to Q_0, where QQ denotes the plasma container and Q0Q_0 denotes a reference plasma container. The diffeomorphism GG is known as a back-to-labels map because its value G(x)G(\boldsymbol{x}), x∈Q\boldsymbol{x} \in Q, denotes the location of the fluid parcel currently located at x\boldsymbol{x} if the entire plasma were moved into its reference configuration. Newcomb’s potential energy W\mathcal{W} is therefore a functional of GG, whose specific form is given by

W(G)=Wmag(G)+Wint(G)\mathcal{W}(G) = \mathcal{W}_{\mathrm{mag}}(G) + \mathcal{W}_{\mathrm{int}}(G)
Wmag(G)=12μ0−1∫Q∣B∣2 d3x,Wint(G)=∫QU(ρ,s) ρ d3x,(1)\mathcal{W}_{\mathrm{mag}}(G) = \frac{1}{2}\mu_0^{-1}\int_Q \lvert\boldsymbol{B}\rvert^2\,d^3x,\qquad\mathcal{W}_{\mathrm{int}}(G) = \int_Q \mathcal{U}(\rho,s)\,\rho\,d^3x, \tag*{(1)}

where the magnetic field, mass density, and entropy are given in terms of the back-to-labels map GG according to

B(x)=A(x)B0(G(x)),ρ(x)=J(x)ρ0(G(x)),s(x)=s0(G(x)),\mathbf{B}(x)=A(x)\mathbf{B}_0(G(x)),\qquad\rho(x)=J(x)\rho_0(G(x)),\qquad s(x)=s_0(G(x)),
G(x)=DG(x),A(x)=adj⁡G(x),J(x)=det⁡G(x).G(x)=DG(x),\qquad A(x)=\operatorname{adj}G(x),\qquad J(x)=\det G(x).

Here DG(x)DG(x) denotes the 3×33 \times3 matrix whose columns are the partial derivatives of GG at xx, adj⁡\operatorname{adj} denotes the matrix adjugate, and B0\mathbf{B}_0, ρ0\rho_0, s0s_0 denote the magnetic field, the mass density, and the entropy in the plasma’s reference configuration. The fact that B0\mathbf{B}_0, ρ0\rho_0, and s0s_0 are treated as fixed fields in Newcomb’s potential energy functional reflects imposition of frozen-in magnetic flux, mass density, and entropy, respectively. Since the formula B(x)=A(x)B0(G(x))\mathbf{B}(x)=A(x)\mathbf{B}_0(G(x)) indicates the magnetic flux 22-form in QQ is the pullback of the magnetic flux 22-form in Q0Q_0, all topological features of the magnetic field B\mathbf{B} are encoded in B0\mathbf{B}_0.

Newcomb’s potential energy functional is mathematically problematic because it fails to be coercive, which is a property that many rigorous applications of the calculus of variations rely upon [46]. The functional W(G)\mathcal{W}(G) would be coercive if W(G)→∞\mathcal{W}(G) \to\infty whenever ∥G∥→∞\lVert G\rVert\to\infty, where ∥⋅∥\lVert\cdot\rVert denotes a suitable norm on the function space containing GG. This function space and its associated norm will be described in detail in § 3. For now it is enough to assume ∥⋅∥\lVert\cdot\rVert measures both values of GG and its first derivative GG.

Lack of coercivity for Newcomb’s functional can be understood heuristically as follows. Consider a 11-parameter family of back-to-labels maps GλG_\lambda, with parameter λ∈R\lambda\in\mathbb{R}. Suppose that on a subregion of QQ with volume V(λ)∝λ−1V(\lambda)\propto\lambda^{-1} the derivative matrix is given by

Gλ=(λ000λ−1/2000λ−1/2),G_\lambda= \begin{pmatrix} \lambda& 0 & 0\\ 0 & \lambda^{-1/2} & 0\\ 0 & 0 & \lambda^{-1/2} \end{pmatrix},

while elsewhere Gλ=O(1)G_\lambda=O(1). Now choose any p>1p>1. The pp-norm of the derivative matrix scales with λ\lambda according to

∫Q∥Gλ∥p d3x∝λp−1.\int_Q \lVert G_\lambda\rVert^p\,d^3x \propto\lambda^{p-1}.

In particular, the pp-norm of the gradient tends to ∞\infty as λ→∞\lambda\to\infty. Since det⁡Gλ=1\det G_\lambda=1 the internal energy Wint(Gλ)\mathcal{W}_{\mathrm{int}}(G_\lambda) is bounded as λ→∞\lambda\to\infty. Coercivity of Newcomb’s functional would therefore require Wmag(Gλ)→∞\mathcal{W}_{\mathrm{mag}}(G_\lambda)\to\infty as λ→∞\lambda\to\infty. To assess whether this actually happens, first note that the adjugate matrix in the region where G\mathbf{G} is large is given by

Aλ=(λ−1000λ1/2000λ1/2).\mathbf{A}_{\lambda} = \begin{pmatrix} \lambda^{-1} & 0 & 0 \\ 0 & \lambda^{1/2} & 0 \\ 0 & 0 & \lambda^{1/2} \end{pmatrix}.

By the quadratic dependence of Wmag(Gλ)\mathcal{W}_{\mathrm{mag}}(\mathbf{G}_{\lambda}) on Aλ\mathbf{A}_{\lambda} the magnetic energy therefore grows no faster than a constant as λ→∞\lambda\to\infty. This shows that Newcomb’s potential cannot control any pp-norm of the derivative matrix G\mathbf{G} with p>1p>1, a particularly dramatic failure of coercivity. The one-parameter family Gλ\mathbf{G}_{\lambda} is an escape sequence. (Note that there is no need to consider p<1p<1 because then (∫Q∣⋅∣p)1/p\left(\int_{\mathcal{Q}}|\cdot|^{p}\right)^{1/p} does not define a norm.)

Lack of coercivity is not a necessary condition for existence of minimizers for WW, as demonstrated by the single-variable function f(x)=−1/(1+x2)f(x)=-1/(1+x^{2}). So it is not fair to infer at this stage that Newcomb’s potential energy functional lacks minimizers in some suitable function space. At the same time, since ideal MHD omits a great deal of important plasma physics [47] it is also not necessary to settle for Newcomb’s potential energy as the only plausible variational description of 3D equilibria. There may be room for improving the coercivity properties of Newcomb’s potential energy by including additional physics effects.

Newcomb’s form of the magnetic energy can be improved by recognizing that real stellarators sustain rapid electromagnetic turbulent fluctuations [48] that contribute to the plasma stress tensor. Here “rapid” refers to a timescale τ\tau much shorter than the equilibrium evolution timescale. Building on Ref. [44], the argument proceeds as follows. Fluctuations of the true magnetic field B\mathbf{B} generally break the frozen-in flux condition. The reference magnetic field

B0(x0)=G(G−1(x0))B(G−1(x0))J(G−1(x0)),\mathbf{B}_{0}(\mathbf{x}_{0}) = \frac{\mathbf{G}(\mathbf{G}^{-1}(\mathbf{x}_{0}))\mathbf{B}(\mathbf{G}^{-1}(\mathbf{x}_{0}))} {\mathcal{J}(\mathbf{G}^{-1}(\mathbf{x}_{0}))},

which would be time-independent in ideal MHD, therefore varies rapidly in time. Suppose averages over the fast time scale can be replaced with ensemble averages defined by some probability measure P(dB0;G)P(d\mathbf{B}_{0};\mathbf{G}). Since changes in the equilibrium generally alter the statistics of gyrokinetic turbulence, this probability measure must depend parametrically on the back-to-labels map G\mathbf{G}. The time average of the magnetic energy then simplifies according to

1τ∫0τWmag(G) dt=E[12μ0−1∫Q∣A(x)B0(G(x))∣2 d3x]=12μ0−1∫Q∣A(x)B‾0(G(x))∣2 d3x+12μ0−1∫Qtr⁡(AT(x)A(x)Σ‾0(G(x))) d3x≡Wmag(G),(2)\begin{aligned} \frac{1}{\tau}\int_{0}^{\tau}\mathcal{W}_{\mathrm{mag}}(\mathbf{G})\,dt = \mathbb{E}\left[ \frac{1}{2}\mu_{0}^{-1}\int_{\mathcal{Q}} \left|\mathbf{A}(\mathbf{x})\mathbf{B}_{0}(\mathbf{G}(\mathbf{x}))\right|^{2}\,d^{3}x \right] \\ &= \frac{1}{2}\mu_{0}^{-1}\int_{\mathcal{Q}} \left|\mathbf{A}(\mathbf{x})\overline{\mathbf{B}}_{0}(\mathbf{G}(\mathbf{x}))\right|^{2}\,d^{3}x + \frac{1}{2}\mu_{0}^{-1}\int_{\mathcal{Q}} \operatorname{tr}\left(\mathbf{A}^{T}(\mathbf{x})\mathbf{A}(\mathbf{x})\overline{\boldsymbol{\Sigma}}_{0}(\mathbf{G}(\mathbf{x}))\right)\,d^{3}x \\ &\equiv\mathcal{W}_{\mathrm{mag}}(\mathbf{G}), \tag*{(2)} \end{aligned}

where the one-point mean and variance of the reference magnetic field are given by

B‾0=∫B0 P(dB0;G),Σ‾0=∫(B0−B‾0)(B0−B‾0)TP(dB0;G).\overline{\mathbf{B}}_{0}=\int\mathbf{B}_{0}\,P(d\mathbf{B}_{0};\mathbf{G}),\qquad \overline{\boldsymbol{\Sigma}}_{0}=\int (\mathbf{B}_{0}-\overline{\mathbf{B}}_{0}) (\mathbf{B}_{0}-\overline{\mathbf{B}}_{0})^{T} P(d\mathbf{B}_{0};\mathbf{G}).

These expressions entail functional integration over the space of reference magnetic fields. Replacing Wmag(G)\mathcal{W}_{\mathrm{mag}}(\mathbf{G}) with Wmag(G)\mathcal{W}_{\mathrm{mag}}(\mathbf{G}) improves on Newcomb’s magnetic energy by incorporating the effects of nonideal electromagnetic fluctuations.

The first term in Eq. (2) represents the magnetic energy stored in the mean magnetic field. The second term gives the excess magnetic energy associated with fluctuations about the mean. Both the energy of the mean field and the excess energy generally exhibit a complicated G\mathbf{G}-dependence due to the presence of expectations over the probability measure P(dB0;G)P(d\mathbf{B}_{0};\mathbf{G}). Deducing the precise form of this G\mathbf{G}-dependence from first principles likely lies beyond the reach of analytical methods. However, there is a plausible functional form motivated by simple phenomenological arguments.

In a stellarator the amplitude of turbulent magnetic fluctuations is small compared to the mean field. It is therefore reasonable to expect that the mean magnetic energy reduces to the usual ideal MHD form in the limit of vanishing fluctuations. Thus, to leading order in the fluctuation strength the mean reference field B‾0\overline{\mathbf{B}}_0 is independent of GG. Denote this GG-independent reference field B‾0=B0\overline{\mathbf{B}}_0 = \mathbf{B}_0. Higher-order corrections to the mean reference field correspond to fluctuation-driven dynamo effects [49, 50]. This work ignores these effects. These considerations completely determine the functional form of the energy of the mean field.

Next consider the excess energy. The instantaneous magnetic energy is proportional to the plasma strain along the reference magnetic field; components of the strain transverse to the reference field make no energetic contribution. See the left panel in Figure 2. In ideal MHD, where the reference field does not change, the plasma therefore behaves as a directed elastic medium. But as the reference field fluctuates it changes direction, activating other components of the slowly-evolving strain tensor. Therefore, from the perspective of an observer on a timescale much longer than the fluctuation timescale, fluctuations produce an effective undirected (although anisotropic) elastic potential energy. See the right panel in Figure 2. This explains the origin and basic physics interpretation of the excess energy term. Its functional dependence on GG is determined by considering the one-point magnetic variance Σ‾0\overline{\Sigma}_0, which is discussed next.

Figure 2. (Left) Instantaneously, magnetic energy can be viewed as elastic potential energy stored in field lines, shown as unconnected lines of springs. (Right) Averaging over nonideal turbulent fluctuations produces an effective, undirected elasticity effect, shown schematically as additional springs connecting neighboring field lines.

Stellarators empirically display a phenomenon known as profile stiffness [51, 52, 53], whereby increasing plasma profile gradients beyond a threshold causes the rate of change of turbulent heat flux with profile gradients to sharply increase. Several previous studies [54, 55, 56] indicate that, for large enough profile gradients, steeper heat flux is accompanied by larger amplitude turbulent fluctuations. Assuming critical balance, asymptotic scaling laws for the amplitude of electrostatic turbulence as a function of profile gradients have been obtained [54, 56]. Although similar scaling laws have yet to be derived for electromagnetic fluctuations, these results suggest that the variance of turbulent magnetic fluctuations Σ‾\overline{\Sigma} also grows with profile gradients when those gradients exceed some threshold. These observations suggest a particular form for the excess energy.

The literature on profile stiffness generally assumes the magnetic field has nested flux surfaces and that the “profiles” are constant on those surfaces. Supposing flux surfaces are level sets of some function ψ\psi, the profiles therefore have the form f(ψ)f(\psi), where ff is a single-variable function. Since ψ=ψ0∘G\psi= \psi_0 \circ G for some reference flux function ψ0:Q0→R\psi_0 : Q_0 \to\mathbb{R} it follows that gradients of f(ψ)f(\psi) are proportional to GG according to

∣∇f(ψ)∣2=∣f′(ψ)∣2∇ψ0T(G(x))G(x)GT(x)∇ψ0(G(x)).\lvert\nabla f(\psi) \rvert^2 = \lvert f'(\psi) \rvert^2 \nabla\psi_0^T(G(x))G(x)G^T(x)\nabla\psi_0(G(x)).

This result, in conjunction with assumed validity of the profile stiffness picture above, implies that the Σ‾\overline{\Sigma} should be an increasing function of plasma strain. For suppose that ∥A∥\lVert A \rVert increases. Since ∥G∥\lVert G \rVert is bounded below by ∥A∥1/2\lVert A \rVert^{1/2} the gradient norm ∥G∥\lVert G \rVert must eventually increase as well. Again invoking the assumed rapid fluctuations in the reference magnetic field, increasing ∥G∥\lVert G \rVert causes the RMS profile gradient, E[∣∇f(ψ)∣2]\sqrt{\mathbb{E}[|\nabla f(\psi)|^2]}, to increase. Profile stiffness then implies that the amplitude of magnetic fluctuations must increase. Thus, the magnetic variance should be an increasing function of ∥A∥\lVert A\rVert. This result applies literally to nested magnetic topologies; in what follows it will be assumed the result remains valid in non-nested topologies. Detailed study of gyrokinetic fluctuation statistics in non-nested topologies remains a largely open area of research.

A simple model that embodies this rough physics picture decomposes the excess energy as the sum of a term linear in ATAA^{\mathsf{T}}A and a homogeneous superlinear term. The linear term corresponds to the magnetic variance Σ0\Sigma_0 in the reference configuration. It is reasonable to assume Σ0\Sigma_0 is uniformly positive definite. The homogeneous superlinear term generally depends on a higher-degree reference tensor field. But under the simplifying assumption that the nonlinear term merely renormalizes the linear term, the following phenomenological model for the mean magnetic energy emerges,

Wmag(G)=∫wmag(G(x),A(x),J(x)) d3x,(3)W_{\mathrm{mag}}(G)=\int w_{\mathrm{mag}}(G(x),A(x),\mathcal{J}(x))\,\mathrm{d}^{3}x, \tag*{(3)}
wmag(G,A,J)=12μ0−1∣B∣2+12μ0−1tr⁡(Σ+Σαmag(ΔB2)αmag−1),(4)w_{\mathrm{mag}}(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\lvert\mathbf{B}\rvert^2+\frac{1}{2}\mu_0^{-1}\operatorname{tr}\left(\Sigma+\frac{\Sigma^{\alpha_{\mathrm{mag}}}}{(\Delta B^2)^{\alpha_{\mathrm{mag}}-1}}\right), \tag*{(4)}
B=AB0(G),Σ=AΣ0(G)AT.\mathbf{B}=A\mathbf{B}_0(G),\qquad\Sigma=A\Sigma_0(G)A^{\mathsf{T}}.

Here αmag>1\alpha_{\mathrm{mag}}>1 is an exponent that sets the homogeneous degree of the nonlinearity, ΔB2\Delta B^2 is a characteristic magnetic variance above which the nonlinearity becomes important, the fields B0\mathbf{B}_0, Σ0\Sigma_0, denote the mean reference magnetic field and the reference magnetic variance, and wmag:Q0×R3×3×R→R∪{+∞}w_{\mathrm{mag}}:Q_0\times\mathbb{R}^{3\times3}\times\mathbb{R}\to\mathbb{R}\cup\{+\infty\} denotes the effective magnetic energy density. The parameters αmag\alpha_{\mathrm{mag}} and ΔB2\Delta B^2 should be fit to the results of high-fidelity microturbulence simulations in various device configurations. The mean reference magnetic field B0\mathbf{B}_0 should be determined by a suitable magnetic topology model. The reference magnetic variance Σ0\Sigma_0 should be determined by requiring the magnetic variance predicted by the equilibrium model agrees with the magnetic variance predicted by a gyrokinetic turbulence model.

When Newcomb’s magnetic energy WmagW_{\mathrm{mag}} in Eq. (1) is replaced by WmagW_{\mathrm{mag}} in Eq. (3) the one-parameter family GλG_\lambda introduced earlier ceases to be an escape sequence. The magnetic energy instead scales as λαmag−1\lambda^{\alpha_{\mathrm{mag}}-1} as λ→∞\lambda\to\infty. Thus, including the excess energy of magnetic fluctuations simultaneously improves the physics fidelity of the equilibrium model and improves the potential energy functional’s growth properties.

Soft vacuum barrier

Including excess magnetic energy does not achieve coercivity by itself. The following argument constructs an escape sequence for Newcomb’s potential with excess magnetic energy included. Let γ∗=1−12αmag\gamma_* = 1-\frac{1}{2\alpha_{\mathrm{mag}}} and note 0<γ∗<10<\gamma_*<1. Consider a 1-parameter family of back-to-labels maps GλG_\lambda with parameter λ∈R\lambda\in\mathbb{R}. Suppose that on a subregion of QQ with volume V(λ)∝λ−1V(\lambda)\propto\lambda^{-1} the derivative matrix is given by

Gλ=(λ000λ−γ∗000λ−1).G_\lambda=\begin{pmatrix} \lambda& 0 & 0\\ 0 & \lambda^{-\gamma_*} & 0\\ 0 & 0 & \lambda^{-1} \end{pmatrix}.

Elsewhere the derivative matrix is O(1)O(1). Now choose any p>1p>1. The pp-norm of the derivative matrix scales with λ\lambda according to

∫Q∥Gλ∥p d3x∝λp−1.\int_Q \lVert G_\lambda\rVert^p\,\mathrm{d}^3x\propto\lambda^{p-1}.

In particular, the pp-norm of the gradient tends to ∞\infty as λ→∞\lambda\to\infty. Coercivity of Newcomb’s functional with excess magnetic energy included would therefore require (Wint+Wmag)(Gλ)→∞(W_{\mathrm{int}}+W_{\mathrm{mag}})(G_\lambda)\to\infty as λ→∞\lambda\to\infty. To assess whether this actually happens, first note that the adjugate matrix and the Jacobian determinant in the region where G\mathbf{G} is large are given by

Aλ=(λ1−γ∗0001000λ1−γ∗),Jλ=λ−γ∗.\mathbf{A}_{\lambda} = \begin{pmatrix} \lambda^{1-\gamma_*} & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & \lambda^{1-\gamma_*} \end{pmatrix}, \qquad \mathcal{J}_{\lambda}=\lambda^{-\gamma_*}.

By the superlinear dependence of Wmag(Gλ)W_{\mathrm{mag}}(G_{\lambda}) on AλTAλ\mathbf{A}_{\lambda}^{\mathsf{T}}\mathbf{A}_{\lambda} the magnetic energy therefore scales according to

Wmag(Gλ)∝λ2αmag(1−γ∗)−1=1,λ→∞.W_{\mathrm{mag}}(G_{\lambda}) \propto\lambda^{2\alpha_{\mathrm{mag}}(1-\gamma_*)-1}=1,\qquad\lambda\to\infty.

Moreover, since Jλ→0\mathcal{J}_{\lambda}\to0 and GλG_{\lambda} is bounded the internal energy tends to 0 as λ→∞\lambda\to\infty. This shows that the one-parameter family GλG_{\lambda} is an escape sequence for Wint+WmagW_{\mathrm{int}}+W_{\mathrm{mag}} for any p>1p>1.

The origin of this escape sequence is failure of the internal energy or the modified magnetic energy to energetically penalize excessive rarefaction in thin regions. Very small densities are a well-known problem in MHD fluid modeling, especially in the high-energy-density plasma simulation community. When low densities become problematic it has become standard practice to introduce a density floor. A recent take on this idea [57] is described by Farmer et. al.. In addition, in the context of stellarator equilibrium modeling there is little evidence to suggest excessive rarefaction actually happens in reactor-relevant configurations. This all suggests that modifying the internal energy functional so that it sharply penalizes excessive rarefaction is practical, mathematically convenient, and physically harmless.

It is in this spirit that the equilibrium model proposed in this work exchanges the standard ideal gas internal energy WintW_{\mathrm{int}} with the modified internal energy Wint(G)=∫Qwint(G,A,J) d3xW_{\mathrm{int}}(G)=\int_Q w_{\mathrm{int}}(G,A,\mathcal{J})\,d^3x, where wint:Q0×R3×R→R∪{+∞}w_{\mathrm{int}}:Q_0\times\mathbb{R}^3\times\mathbb{R}\to\mathbb{R}\cup\{+\infty\} is given by

wint(G,A,J)=ρU∗(ρ,s)={pˉ(s)ρˉγργγ−1+εvac ρvacαvacραvac,ρ>0,+∞,ρ≤0.(5)w_{\mathrm{int}}(G,A,\mathcal{J})=\rho U^*(\rho,s)= \begin{cases} \dfrac{\bar{p}(s)}{\bar{\rho}^{\gamma}}\dfrac{\rho^\gamma}{\gamma-1} +\dfrac{\varepsilon_{\mathrm{vac}}\,\rho_{\mathrm{vac}}^{\alpha_{\mathrm{vac}}}}{\rho^{\alpha_{\mathrm{vac}}}}, & \rho>0,\\ +\infty, & \rho\leq0. \end{cases} \tag*{(5)}
ρ=Jρ0(G),s=s0(G).\rho=\mathcal{J}\rho_0(G),\qquad s=s_0(G).

Here γ>1\gamma>1 denotes the adiabatic index, pˉ(s)∝exp⁡(s/Cv)\bar{p}(s)\propto\exp(s/C_v) denotes a characteristic entropy-dependent pressure, and ρˉ\bar{\rho} denotes a (constant, positive) characteristic mass density. The positive constants εvac\varepsilon_{\mathrm{vac}} and ρvac\rho_{\mathrm{vac}} control the strength and mass density scale for the vacuum barrier; decreasing mass density below ρvac\rho_{\mathrm{vac}} incurs an energy cost proportional to εvac\varepsilon_{\mathrm{vac}}. The barrier abruptness is controlled by the exponent αvac\alpha_{\mathrm{vac}}. Eventually there will be a further constraint on αvac\alpha_{\mathrm{vac}}, but for now it is sufficient to assume αvac>0\alpha_{\mathrm{vac}}>0. Note that wintw_{\mathrm{int}} is manifestly non-negative. Also note that the vacuum barrier term ∝ρ−αvac\propto\rho^{-\alpha_{\mathrm{vac}}} ensures GλG_{\lambda} above is no longer an escape sequence.

It will become clear in § 4 that the pair of modifications to Newcomb’s potential, Wint→W~intW_{\mathrm{int}}\to\widetilde{W}_{\mathrm{int}} and Wmag→W~magW_{\mathrm{mag}}\to\widetilde{W}_{\mathrm{mag}}, together produce a coercive potential energy functional.

Model formulation

Let Q⊂R3Q\subset\mathbb{R}^3 denote a compact region in R3\mathbb{R}^3 with C∞C^\infty boundary. Denote the volume of QQ as L3L^3, which gives LL the interpretation of a characteristic domain length scale. Let W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3) denote the (real) linear space of functions G:Q→R3G:Q\to\mathbb{R}^3 such that the pp-norm of GG and its first derivative are each finite. Eventually the analysis will select a specific pp-value. For now it suffices to assume 3<p<∞3<p<\infty, which ensures in particular that elements of W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3) have continuous representatives. If G(x)=DG(x)G(x)=DG(x) denotes the 3×33\times3 matrix whose columns equal the partial derivatives of GG at x∈Qx\in Q, G∈W1,p(Q,R3)G\in W^{1,p}(Q,\mathbb{R}^3) is equivalent to ∫Q∣G(x)∣p d3x<∞\int_Q |G(x)|^p\,d^3x<\infty and ∫Q∥G(x)∥p d3x<∞\int_Q\|G(x)\|^p\,d^3x<\infty, where ∣⋅∣|\cdot| denotes the standard Euclidean norm on R3\mathbb{R}^3 and ∥⋅∥\|\cdot\| denotes the Frobenius norm on 3×33\times3 matrices. For the allowed range of pp, it is a well-known fact that W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3) is a reflexive Banach space. Of the many possible equivalent norms on W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3) this work uses

∥G∥1,p=(1L3+p∫Q∣G(x)∣p d3x+1L3∫Q∥G(x)∥p d3x)1/p.\|G\|_{1,p}= \left( \frac{1}{L^{3+p}}\int_Q |G(x)|^p\,d^3x+ \frac{1}{L^3}\int_Q \|G(x)\|^p\,d^3x \right)^{1/p}.

which has the benefit of being dimensionless. Let Q0⊂R3Q_0 \subset\mathbb{R}^3 denote another region in R3\mathbb{R}^3 C∞C^\infty-diffeomorphic to QQ. Let Z⊂W1,p(Q,R3)\mathcal{Z} \subset W^{1,p}(Q,\mathbb{R}^3) be the subset of W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3) comprising maps G∈W1,p(Q,R3)G \in W^{1,p}(Q,\mathbb{R}^3) that satisfy the three conditions

BL1: G(Q)⊂Q0G(Q) \subset Q_0

BL2: G(∂Q)⊂∂Q0G(\partial Q) \subset\partial Q_0

BL3:

∫Qdet⁡G(x) d3x≤∫G(Q)d3x0\int_Q \det G(\mathbf{x})\,\mathrm{d}^3x \le\int_{G(Q)} \mathrm{d}^3x_0

Note that BL3 is known as the Ciarlet-Nečas [58] condition in the field of nonlinear elasticity. Also note that the upper bound in BL3 exists because GG is continuous for p>3p > 3, which implies G(Q)G(Q) is measurable. The set Z\mathcal{Z} serves as a viable fluid configuration space for the plasma. Note that GG in the complement of Z\mathcal{Z} are not valid back-to-labels maps because all fluid parcel labels must be contained in the reference spatial domain Q0Q_0.

Elasticized magnetohydrostatics — The potential energy is the elasticized Newcomb potential W:Z→R∪{+∞}W:\mathcal{Z}\to\mathbb{R}\cup\{+\infty\} defined as the sum of the elasticized magnetic energy given in Eq. (3) and the modified internal energy given in Eq. (5). The following provides a detailed specification of the energy functional WW in terms of a potential energy density ww.

Let Sym⁡(3)\operatorname{Sym}(3) denote the space of symmetric 3×33\times3 matrices. Let ρ0:Q0→R\rho_0:Q_0\to\mathbb{R}, s0:Q0→Rs_0:Q_0\to\mathbb{R}, B0:Q0→R3B_0:Q_0\to\mathbb{R}^3, and Σ0:Q0→Sym⁡(3)\Sigma_0:Q_0\to\operatorname{Sym}(3) be fixed fields defined over the reference domain Q0Q_0. The reference fields will always satisfy the following assumption.

Assumption 1. Each of the reference fields ρ0\rho_0, s0s_0, B0B_0, Σ0\Sigma_0 is C∞C^\infty. In addition B0=∇×A0B_0=\nabla\times A_0 for some C∞C^\infty vector potential A0A_0, Σ0\Sigma_0 is uniformly positive definite, and ρ0\rho_0 is positive pointwise.

Remark 1. The field B0B_0 need not be tangent to ∂Q0\partial Q_0, but Assumption 1 does allow for it. Tangency of B0B_0 to ∂Q0\partial Q_0 is a common confinement modeling assumption, but it will play no role in the rigorous existence theory.

The elasticized potential energy density w:Q0×R3×3×R→R∪{+∞}w:Q_0\times\mathbb{R}^{3\times3}\times\mathbb{R}\to\mathbb{R}\cup\{+\infty\} is given by

w(G,A,J)=wint(G,A,J)+wmag(G,A,J),w(G,A,J)=w_{\mathrm{int}}(G,A,J)+w_{\mathrm{mag}}(G,A,J),

where wintw_{\mathrm{int}} and wmagw_{\mathrm{mag}} are defined in Eq. (5) and Eq. (4). Note that both wintw_{\mathrm{int}} and wmagw_{\mathrm{mag}} depend on the reference fields. The corresponding elasticized potential energy is

W(G)=∫Qw(G(x),A(x),J(x)) d3x,(6)W(G)=\int_Q w(G(\mathbf{x}),A(\mathbf{x}),J(\mathbf{x}))\,\mathrm{d}^3x, \tag*{(6)}
J(x)=det⁡G(x),A(x)=adj⁡G(x).J(\mathbf{x})=\det G(\mathbf{x}),\qquad A(\mathbf{x})=\operatorname{adj}G(\mathbf{x}).

Elasticized magnetohydrostatics defines equilibria as follows.

Definition 1. A back-to-labels map G∈ZG\in\mathcal{Z} is an equilibrium if it is a minimizer for WW. Equivalently, GG is an equilibrium if, for every G′∈ZG'\in\mathcal{Z}, W(G)≤W(G′)W(G)\le W(G').

Remark 2. The definition 1 defines equilibria in the model as ground states for the potential energy. The energy functional WW may therefore have local minima not labeled as equilibria by the definition. As stated, the results proved in this work concerning regularity of equilibria do not apply to local minima that are not global minima. Generically one expects a single global minimizer, and therefore a unique equilibrium. However, there may be special model inputs (domain boundary and reference fields) that produce multiple global minimizers.

The potential energy is norm-Coercive

It is clear from the definition of WW that W(G)≥0W(G)\ge0 for all G∈ZG\in\mathcal{Z}, as would be expected for a physical energy. In fact WW satisfies a much stronger positivity property known as coercivity when pp is appropriately related to the vacuum and variance exponents, αvac,αmag\alpha_{\mathrm{vac}},\alpha_{\mathrm{mag}}.

Theorem 1. Suppose 3<αvacαmagαvac+αmag=p3 < \frac{\alpha_{\mathrm{vac}}\alpha_{\mathrm{mag}}}{\alpha_{\mathrm{vac}}+\alpha_{\mathrm{mag}}}=p. Then the functional W:Z→R∪{+∞}W:\mathcal{Z}\to\mathbb{R}\cup\{+\infty\} is W1,pW^{1,p}-norm-coercive. In fact, there are G\mathcal{G}-independent positive constants c0,c1c_0,c_1 such that for each G∈ZG\in\mathcal{Z}

W(G)≥∥G∥1,pp−c0c1.(7)W(G)\geq\frac{\lVert G\rVert_{1,p}^{p}-c_0}{c_1}. \tag*{(7)}

Remark 3. Since αvacαmagαvac+αmag\frac{\alpha_{\mathrm{vac}}\alpha_{\mathrm{mag}}}{\alpha_{\mathrm{vac}}+\alpha_{\mathrm{mag}}} is strictly less than either αvac\alpha_{\mathrm{vac}} or αmag\alpha_{\mathrm{mag}} it follows that the Theorem’s hypothesis requires, in particular, αvac>3\alpha_{\mathrm{vac}}>3 and αmag>3\alpha_{\mathrm{mag}}>3. The vacuum barrier is artificial, so making it more narrow by increasing αvac\alpha_{\mathrm{vac}} is desirable. On the other hand, since the amplitude scaling of electromagnetic fluctuations with supercritical profile gradients remains an open question, it is possible that αmag>3\alpha_{\mathrm{mag}}>3 is unrealistic. This represents a genuine weakness of the rigorous results presented here. However, numerical experiments in § 6 suggest this is a weakness in the analytical methods used to obtain the results, rather than a weakness in the model itself.

Proof. Let G∈ZG\in\mathcal{Z} and let G˙\dot{G} denote its weak derivative. The coercivity inequality Eq. (7) is trivially satisfied if W(G)=+∞W(G)=+\infty. So assume that W(G)<∞W(G)<\infty. This implies that the modified internal energy must be finite, which in turn implies that J(x)>0\mathcal{J}(x)>0 for almost all x∈Qx\in Q.

Observe that if MM is any invertible 3×33\times3 matrix then its Frobenius norm is bounded from above according to

∥M∥2≤13∥adj⁡M∥4(det⁡M)2,(8)\lVert M\rVert^2\leq\frac{1}{3}\frac{\lVert\operatorname{adj}M\rVert^4}{(\det M)^2}, \tag*{(8)}

where adj⁡M\operatorname{adj}M denotes the adjugate of MM. Let 1<a,b<∞1<a,b<\infty be real numbers that satisfy 1/a+1/b=11/a+1/b=1. By Eq. (8) and Hölder’s inequality the pp-norm of GG satisfies

∫Q∥G∥p d3x≤3−p/2∫Q∥A∥2pJp d3x≤3−p/2(∫Q1Jap d3x)1/a(∫Q∥A∥2pb d3x)1/b.(9)\int_Q\lVert G\rVert^p\,\mathrm{d}^3x\leq3^{-p/2}\int_Q\frac{\lVert A\rVert^{2p}}{\mathcal{J}^p}\,\mathrm{d}^3x \leq3^{-p/2}\left(\int_Q\frac{1}{\mathcal{J}^{ap}}\,\mathrm{d}^3x\right)^{1/a}\left(\int_Q\lVert A\rVert^{2pb}\,\mathrm{d}^3x\right)^{1/b}. \tag*{(9)}

Now by the definition Eq. (5) of the modified internal energy density and non-negativity of the modified magnetic energy density,

W(G)≥∫Q(εvacρ0αvacJ0(G)αvac) d3x≥εvacρ0,max⁡αvacρ0,max⁡αvac∫Q1Jαvac d3x,W(G)\geq\int_Q\left(\frac{\varepsilon_{\mathrm{vac}}\rho_0^{\alpha_{\mathrm{vac}}}}{\mathcal{J}_0(G)^{\alpha_{\mathrm{vac}}}}\right)\,\mathrm{d}^3x\geq\frac{\varepsilon_{\mathrm{vac}}\rho_{0,\max}^{\alpha_{\mathrm{vac}}}}{\rho_{0,\max}^{\alpha_{\mathrm{vac}}}}\int_Q\frac{1}{\mathcal{J}^{\alpha_{\mathrm{vac}}}}\,\mathrm{d}^3x,

where ρ0,max⁡\rho_{0,\max} denotes the maximum reference mass density. By assumed continuity and point-wise positivity of ρ0,ρ0αvac>0\rho_0,\rho_0^{\alpha_{\mathrm{vac}}}>0. This bound suggests choosing aa in Hölder’s inequality so that ap=αvacap=\alpha_{\mathrm{vac}}, which is permissible because then a=αvac/p>1a=\alpha_{\mathrm{vac}}/p>1, by hypothesis. (Notice that αvacαmag/(αvac+αmag)<αvac\alpha_{\mathrm{vac}}\alpha_{\mathrm{mag}}/(\alpha_{\mathrm{vac}}+\alpha_{\mathrm{mag}})<\alpha_{\mathrm{vac}}.) Proceeding in this manner, the bound in Eq. (9) on GG’s pp-norm becomes

∫Q∥G∥p d3x≤3−p/2ρ0,max⁡pεvacp/αvacρvacp W(G)p/αvac(∫Q∥A∥2αmag d3x)(αvac−p)/αvac.(10)\int_Q\lVert G\rVert^p\,\mathrm{d}^3x\leq3^{-p/2}\frac{\rho_{0,\max}^{p}}{\varepsilon_{\mathrm{vac}}^{p/\alpha_{\mathrm{vac}}}\rho_{\mathrm{vac}}^{p}}\ W(G)^{p/\alpha_{\mathrm{vac}}}\left(\int_Q\lVert A\rVert^{2\alpha_{\mathrm{mag}}}\,\mathrm{d}^3x\right)^{(\alpha_{\mathrm{vac}}-p)/\alpha_{\mathrm{vac}}}. \tag*{(10)}

Now let σmin⁡\sigma_{\min} be a uniform lower bound on the eigenvalues of Σ0\Sigma_0. The integrand ∥A∥2αmag\lVert A\rVert^{2\alpha_{\mathrm{mag}}} relates to Σαmag\Sigma^{\alpha_{\mathrm{mag}}} according to

∥A∥2αmag=[tr⁡(ATA)]αmag≤1σmin⁡αmag[tr⁡(AΣ0AT)]αmag≤3αmag−1σmin⁡αmagtr⁡([AΣ0AT]αmag)=3αmag−1σmin⁡αmagtr⁡(Σαmag).(11)\lVert A\rVert^{2\alpha_{\mathrm{mag}}}=\left[\operatorname{tr}(A^{T}A)\right]^{\alpha_{\mathrm{mag}}}\leq\frac{1}{\sigma_{\min}^{\alpha_{\mathrm{mag}}}}\left[\operatorname{tr}(A\Sigma_0A^{T})\right]^{\alpha_{\mathrm{mag}}} \leq\frac{3^{\alpha_{\mathrm{mag}}-1}}{\sigma_{\min}^{\alpha_{\mathrm{mag}}}}\operatorname{tr}\left([A\Sigma_0A^{T}]^{\alpha_{\mathrm{mag}}}\right)=\frac{3^{\alpha_{\mathrm{mag}}-1}}{\sigma_{\min}^{\alpha_{\mathrm{mag}}}}\operatorname{tr}(\Sigma^{\alpha_{\mathrm{mag}}}). \tag*{(11)}

Here the first inequality follows from uniform positive definiteness of Σ0\Sigma_0, while the second inequality follows from Jensen’s inequality applied to the convex function f(x)=xαmagf(x)=x^{\alpha_{\mathrm{mag}}}. From Eq. (11) it follows that

W(G)≥12μ0−1∫Qtr⁡(Σαmag)(ΔB2)αmag−1 d3x≥μ0−12⋅3αmag−1σmin⁡αmag(ΔB2)αmag−1∫Q∥A∥2αmag d3x.(12)W(G) \ge\frac{1}{2}\mu_0^{-1}\int_Q \frac{\operatorname{tr}(\Sigma^{\alpha_{\mathrm{mag}}})}{(\Delta B^2)^{\alpha_{\mathrm{mag}}-1}}\,\mathrm{d}^3x \ge\frac{\mu_0^{-1}}{2\cdot3^{\alpha_{\mathrm{mag}}-1}}\frac{\sigma_{\min}^{\alpha_{\mathrm{mag}}}}{(\Delta B^2)^{\alpha_{\mathrm{mag}}-1}}\int_Q \lVert A\rVert^{2\alpha_{\mathrm{mag}}}\,\mathrm{d}^3x. \tag*{(12)}

Now substituting Eq. (12) into Eq. (10) implies

∫Q∥G∥p d3x≤3p/2(ρ0,max⁡ρvacΔB2σmin⁡)p1(εvac)p/αvac1(32μ0−1ΔB2)αvac−pαvacW(G).(13)\int_Q \lVert G\rVert^p\,\mathrm{d}^3x \le3^{p/2}\left(\frac{\rho_{0,\max}}{\rho_{\mathrm{vac}}}\frac{\Delta B^2}{\sigma_{\min}}\right)^p\frac{1}{(\varepsilon_{\mathrm{vac}})^{p/\alpha_{\mathrm{vac}}}}\frac{1}{\left(\frac{3}{2}\mu_0^{-1}\Delta B^2\right)^{\frac{\alpha_{\mathrm{vac}}-p}{\alpha_{\mathrm{vac}}}}}W(G). \tag*{(13)}

To finish the proof, note that boundedness of the reference spatial domain Q0Q_0 and G(Q)⊂Q0G(Q)\subset Q_0 (property BL1) implies that there is some radius R0R_0 such that ∣G(x)∣<R0\lvert G(x)\rvert<R_0 for all x∈Qx\in Q. Therefore ∫Q∣G(x)∣p d3x≤R0pL3\int_Q\lvert G(x)\rvert^p\,\mathrm{d}^3x\le R_0^pL^3. In conjunction with Eq. (13) this implies

∥G∥1,pp=1L3+p∫Q∣G(x)∣p d3x+1L3∫Q∥G(x)∥p d3x≤R0pLp+3p/2L3(ρ0,max⁡ρvacΔB2σmin⁡)p1(εvac)p/αvac1(32μ0−1ΔB2)αvac−pαvacW(G).\lVert G\rVert_{1,p}^p=\frac{1}{L^{3+p}}\int_Q\lvert G(x)\rvert^p\,\mathrm{d}^3x+\frac{1}{L^3}\int_Q\lVert G(x)\rVert^p\,\mathrm{d}^3x \le\frac{R_0^p}{L^p}+\frac{3^{p/2}}{L^3}\left(\frac{\rho_{0,\max}}{\rho_{\mathrm{vac}}}\frac{\Delta B^2}{\sigma_{\min}}\right)^p\frac{1}{(\varepsilon_{\mathrm{vac}})^{p/\alpha_{\mathrm{vac}}}}\frac{1}{\left(\frac{3}{2}\mu_0^{-1}\Delta B^2\right)^{\frac{\alpha_{\mathrm{vac}}-p}{\alpha_{\mathrm{vac}}}}}W(G).

which is the desired result. In particular, the positive constants c0,c1c_0,c_1 are given by

c0=R0pLp(14)c_0=\frac{R_0^p}{L^p} \tag*{(14)}
c1=3p/2L3(ρ0,max⁡ρvacΔB2σmin⁡)p1(εvac)p/αvac1(32μ0−1ΔB2)αvac−pαvac.(15)c_1=\frac{3^{p/2}}{L^3}\left(\frac{\rho_{0,\max}}{\rho_{\mathrm{vac}}}\frac{\Delta B^2}{\sigma_{\min}}\right)^p\frac{1}{(\varepsilon_{\mathrm{vac}})^{p/\alpha_{\mathrm{vac}}}}\frac{1}{\left(\frac{3}{2}\mu_0^{-1}\Delta B^2\right)^{\frac{\alpha_{\mathrm{vac}}-p}{\alpha_{\mathrm{vac}}}}}. \tag*{(15)}

Existence theory

With coercivity of the elasticized potential established, existence of minimizers now follows from standard arguments from mathematical elasticity. First it is necessary to establish the following Lemmas.

Lemma 1. Assume p>3p>3. The optimization domain Z\mathcal{Z} is a weakly sequentially closed subset of W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3).

Proof. Let Gn∈W1,p(Q,R3)G_n\in W^{1,p}(Q,\mathbb{R}^3) be a weakly-convergent sequence such that Gn∈ZG_n\in\mathcal{Z} for all nn. Since we always assume p>3p>3 and QQ is closed the Rellich-Kondrachov theorem implies that GnG_n converges uniformly to a continuous map G:Q→R3G:Q\to\mathbb{R}^3. We must show that the limit map GG satisfies properties BL1-BL3.

BL1: Let x∈Qx\in Q. The sequence Gn(x)G_n(x) is contained in Q0Q_0. By uniform convergence of GnG_n to GG the sequence Gn(x)G_n(x) converges to G(x)G(x). By closedness of Q0Q_0 the limit of Gn(x)G_n(x), which is G(x)G(x), is contained in Q0Q_0. Thus G(Q)⊂Q0G(Q)\subset Q_0.

BL2: Let x∈∂Qx\in\partial Q. The sequence Gn(x)G_n(x) is contained in ∂Q0\partial Q_0. By uniform convergence of GnG_n to GG the sequence Gn(x)G_n(x) converges to G(x)G(x). By closedness of ∂Q0\partial Q_0 the limit of Gn(x)G_n(x), which is G(x)G(x), is contained in ∂Q0\partial Q_0. Thus G(∂Q)⊂∂Q0G(\partial Q)\subset\partial Q_0.

BL3: We recall an argument due to Ciarlet-Nečas [58] from the proof of their Theorem 5, which we will recount in some detail. We do this to clarify that we need not assume JnJ_n is non-negative almost everywhere, which is not explicitly stated by Ciarlet-Nečas. For δ>0\delta> 0 let Bδ(x)B_{\delta}(x) denote the open ball with radius δ\delta at x∈R3x \in\mathbb{R}^{3}. Since GG is continuous and QQ is compact, G(Q)G(Q) is compact, and hence measurable. It follows that for every ϵ>0\epsilon> 0 there is an open set OϵO_{\epsilon} with G(Q)⊂OϵG(Q) \subset O_{\epsilon} and ∫Oϵ−G(Q)d3x<ϵ\int_{O_{\epsilon}-G(Q)} \mathrm{d}^{3}x < \epsilon. In fact there is a δ>0\delta> 0 such that ⋃x∈G(Q)Bδ(x)⊂Oϵ\bigcup_{x \in G(Q)} B_{\delta}(x) \subset O_{\epsilon}. Uniform convergence of GnG_n then implies that Gn(Q)⊂OϵG_n(Q) \subset O_{\epsilon} for all sufficiently large nn. Combining this with ∫QJnd3x≤∫Gn(Q)d3x0\int_Q \mathcal{J}_n \mathrm{d}^{3}x \leq\int_{G_n(Q)} \mathrm{d}^{3}x_0 then gives ∫QJnd3x≤∫Oϵd3x\int_Q \mathcal{J}_n \mathrm{d}^{3}x \leq\int_{O_{\epsilon}} \mathrm{d}^{3}x for all sufficiently large nn.

Let r=p/3r = p/3. The next step is to use the well-known fact that the mapping W1,p(Q,R3)→Lr(Q)W^{1,p}(Q,\mathbb{R}^{3}) \to L^r(Q) given by G↦JG \mapsto\mathcal{J} is sequentially weakly continuous when p>3p > 3. (This is described for example by Ball[59] in his Corollary 6.2.2.) Thus, by weak-convergence of GnG_n in W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^{3}), the sequence Jn\mathcal{J}_n converges weakly in Lr(Q)L^r(Q) to J\mathcal{J}. In particular, since the unit constant function 11 is a member of the continuous dual (Lr(Q))∗(L^r(Q))^*,

∫QJ d3x=lim⁡n→∞∫QJn d3x≤∫Oϵd3x0=∫G(Q)d3x0+∫Oϵ−G(Q)d3x0<∫G(Q)d3x0+ϵ.\int_Q \mathcal{J}\,\mathrm{d}^{3}x = \lim_{n\to\infty}\int_Q \mathcal{J}_n\,\mathrm{d}^{3}x \leq\int_{O_{\epsilon}}\mathrm{d}^{3}x_0 = \int_{G(Q)}\mathrm{d}^{3}x_0+\int_{O_{\epsilon}-G(Q)}\mathrm{d}^{3}x_0 < \int_{G(Q)}\mathrm{d}^{3}x_0+\epsilon.

Since this inequality must be satisfied for all ϵ>0\epsilon> 0 we conclude that property BL3 is satisfied by GG, as claimed.

Lemma 2. Assume p>3p > 3. If GnG_n is a weakly-convergent sequence in W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^{3}) with weak limit GG and Gn∈ZG_n \in\mathcal{Z} for all nn then

W(G)≤lim inf⁡n→∞W(Gn).W(G) \leq\liminf_{n\to\infty} W(G_n).

Proof. The proof is a direct application of Theorem 6.4 from Benešová and Kružíк[60].

We begin by extending the reference domain fields ρ0\rho_0, Σ0\Sigma_0, s0s_0, B0B_0 to all of R3\mathbb{R}^{3} while preserving their constraints. By the Whitney extension theorem there is a C∞C^{\infty} function s~0:R3→R\widetilde{s}_0:\mathbb{R}^{3}\to\mathbb{R} that agrees with s0s_0 on QQ. Since ρ0\rho_0 is positive λ0=ln⁡ρ0\lambda_0=\ln\rho_0 is a C∞C^{\infty} function on Q0Q_0. Again by the Whitney extension theorem there is a C∞C^{\infty} extension λ~0:R3→R\widetilde{\lambda}_0:\mathbb{R}^{3}\to\mathbb{R}. Since ρ~0=exp⁡(λ~0)\widetilde{\rho}_0=\exp(\widetilde{\lambda}_0) is positive and agrees with ρ0\rho_0 on Q0Q_0 it provides a C∞C^{\infty} extension of ρ0\rho_0 that is positive everywhere. Since Σ0\Sigma_0 is uniformly symmetric positive definite we can form its point-wise matrix logarithm Λ0=ln⁡Σ0\Lambda_0=\ln\Sigma_0. The matrix function Λ0:Q0→R3×3\Lambda_0:Q_0\to\mathbb{R}^{3\times3} is C∞C^{\infty} and takes values in symmetric matrices. By the Whitney extension theorem applied to the entries of Λ0\Lambda_0 on the diagonal and above, there is a C∞C^{\infty} extension Λ~0:R3→R3×3\widetilde{\Lambda}_0:\mathbb{R}^{3}\to\mathbb{R}^{3\times3} that also takes values in symmetric matrices. Since Σ~0=exp⁡(Λ~0)\widetilde{\Sigma}_0=\exp(\widetilde{\Lambda}_0) agrees with Σ0\Sigma_0 on Q0Q_0 it provides a C∞C^{\infty} extension of Σ0\Sigma_0 to all of R3\mathbb{R}^{3} that is pointwise symmetric positive definite. Finally, by application of the Whitney extension theorem component-wise to A0A_0 there is a C∞C^{\infty} vector field A~0:R3→R3\widetilde{A}_0:\mathbb{R}^{3}\to\mathbb{R}^{3} that extends A0A_0. Since B~0=∇×A~0\widetilde{B}_0=\nabla\times\widetilde{A}_0 agrees with B0B_0 on Q0Q_0 it provides a divergence-free C∞C^{\infty} extension of B0B_0 to all of R3\mathbb{R}^{3}. Note that the extension Σ~0\widetilde{\Sigma}_0 need not be uniformly positive definite. Also note that the extension B~0\widetilde{B}_0 may have nulls, even if B0B_0 does not.

Next we extend the domain of definition of the potential energy density. As highlighted in § 3, the potential energy density is given by a function w:Q0×R3×3×R→R∪{∞}w:Q_0\times\mathbb{R}^{3\times3}\times\mathbb{R}\to\mathbb{R}\cup\{\infty\}. We decompose this function as a sum of four terms, w=∑i=14wiw=\sum_{i=1}^{4}w_i, where

w1(G,A,J)={p‾(s~0(G))ρ~0γ(Jρ~0(G))γγ−1+ϵvacpvacγ(Jρ~0(G))γvac,J>0,+∞,J≤0w_1(G,A,\mathcal{J}) = \begin{cases} \dfrac{\overline{p}(\widetilde{s}_0(G))}{\widetilde{\rho}_0^\gamma}\dfrac{(\mathcal{J}\widetilde{\rho}_0(G))^\gamma}{\gamma-1}+\dfrac{\epsilon_{\mathrm{vac}}\mathbf{p}_{\mathrm{vac}}^\gamma}{(\mathcal{J}\widetilde{\rho}_0(G))^{\gamma_{\mathrm{vac}}}}, & \mathcal{J}>0,\\ +\infty, & \mathcal{J}\leq0 \end{cases}
w2(G,A,J)=12μ0−1∣AB~0(G)∣2w_2(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\lvert A\widetilde{B}_0(G)\rvert^2
w3(G,A,J)=12μ0−1tr⁡(AΣ~0(G)AT)w_3(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\operatorname{tr}\bigl(A\widetilde{\Sigma}_0(G)A^{\mathsf{T}}\bigr)
w4(G,A,J)=12μ0−1tr⁡([AΣ~0(G)AT]αmag)(ΔB2)αmag−1.w_4(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\frac{\operatorname{tr}\bigl([A\widetilde{\Sigma}_0(G)A^{\mathsf{T}}]^{\alpha_{\mathrm{mag}}}\bigr)}{(\Delta B^2)^{\alpha_{\mathrm{mag}}-1}}.

By replacing the C∞C^\infty reference fields ρ0,s0,Σ0,B0\rho_0,s_0,\Sigma_0,\mathbf{B}_0 with their C∞C^\infty extensions ρ~0,s~0,Σ~0,B~0\widetilde{\rho}_0,\widetilde{s}_0,\widetilde{\Sigma}_0,\widetilde{\mathbf{B}}_0, we obtain extensions w~i:R3×R3×3×R→R3\widetilde{w}_i:\mathbb{R}^3 \times\mathbb{R}^{3\times3} \times\mathbb{R} \to\mathbb{R}^3 given explicitly by

w~1(G,A,J)={p~(s~0(G))ρ~0(G)γ(Jρ~0(G))γ+εvacρvacαvac(Jρ~0(G))αvac,J>0,+∞,J≤0.\widetilde{w}_1(G,A,\mathcal{J}) = \begin{cases} \dfrac{\widetilde{p}(\widetilde{s}_0(G))}{\widetilde{\rho}_0(G)^\gamma}(\mathcal{J}\widetilde{\rho}_0(G))^\gamma+\dfrac{\varepsilon_{\mathrm{vac}}\rho_{\mathrm{vac}}^{\alpha_{\mathrm{vac}}}}{(\mathcal{J}\widetilde{\rho}_0(G))^{\alpha_{\mathrm{vac}}}}, & \mathcal{J}>0,\\ +\infty, & \mathcal{J}\leq0. \end{cases}
w~2(G,A,J)=12μ0−1∣AB~0(G)∣2\widetilde{w}_2(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\lvert A\widetilde{\mathbf{B}}_0(G)\rvert^2
w~3(G,A,J)=12μ0−1tr⁡(AΣ~0(G)AT)\widetilde{w}_3(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\operatorname{tr}\left(A\widetilde{\Sigma}_0(G)A^{\mathsf{T}}\right)
w~4(G,A,J)=12μ0−1tr⁡([AΣ~0(G)AT]αmag)(ΔB2)αmag−1.\widetilde{w}_4(G,A,\mathcal{J})=\frac{1}{2}\mu_0^{-1}\frac{\operatorname{tr}\left([A\widetilde{\Sigma}_0(G)A^{\mathsf{T}}]^{\alpha_{\mathrm{mag}}}\right)}{(\Delta B^2)^{\alpha_{\mathrm{mag}}-1}}.

Notice that each of the extensions is continuous and non-negative when the extended real line R∪{+∞}\mathbb{R}\cup\{+\infty\} is equipped with its usual topology.

Now we establish convexity of w~i(G,A,J)\widetilde{w}_i(G,A,\mathcal{J}) in (A,J)(A,\mathcal{J}) for each fixed G∈R3G\in\mathbb{R}^3. The function w~1\widetilde{w}_1 is convex in (A,J)(A,\mathcal{J}) because it is equal to a convex single-variable function of J\mathcal{J}. Note that this observation makes use of positivity of the extension ρ~0\widetilde{\rho}_0. For fixed G∈RG\in\mathbb{R}, both w~2\widetilde{w}_2 and w~3\widetilde{w}_3 have the form wS(A)=tr⁡(ASAT)w_S(A)=\operatorname{tr}(ASA^{\mathsf{T}}), where SS is symmetric positive semidefinite. Note this uses pointwise positive definiteness of Σ~0\widetilde{\Sigma}_0. There is a unique symmetric positive-semidefinite matrix S1/2S^{1/2} such that S=S1/2S1/2S=S^{1/2}S^{1/2}. Therefore wS(A)=tr⁡(AS1/2[AS1/2]T)=∥AS1/2∥22w_S(A)=\operatorname{tr}(AS^{1/2}[AS^{1/2}]^{\mathsf{T}})=\lVert AS^{1/2}\rVert_2^2. This shows the function wSw_S is given by composing a linear map with a squared norm. Such compositions are known to be convex, which implies w~i\widetilde{w}_i is convex in (A,J)(A,\mathcal{J}) for i=2,3i=2,3. Finally convexity of w~4\widetilde{w}_4 follows from writing

tr⁡([AΣ~0AT]αmag)=tr⁡([(AΣ~01/2)(AΣ~01/2)T]αmag)=∥AΣ~01/2∥2αmag2αmag.\operatorname{tr}\left([A\widetilde{\Sigma}_0A^{\mathsf{T}}]^{\alpha_{\mathrm{mag}}}\right)=\operatorname{tr}\left([(A\widetilde{\Sigma}_0^{1/2})(A\widetilde{\Sigma}_0^{1/2})^{\mathsf{T}}]^{\alpha_{\mathrm{mag}}}\right)=\lVert A\widetilde{\Sigma}_0^{1/2}\rVert_{2\alpha_{\mathrm{mag}}}^{2\alpha_{\mathrm{mag}}}.

where ∥⋅∥2αmag\lVert\cdot\rVert_{2\alpha_{\mathrm{mag}}} denotes the Schatten 2αmag2\alpha_{\mathrm{mag}}-norm. This implies convexity in AA because if ∥⋅∥∗\lVert\cdot\rVert_* is any norm on 3×33\times3 matrices then ∥⋅∥∗r\lVert\cdot\rVert_*^r is convex whenever r>1r>1; in this case r=2αmag>2r=2\alpha_{\mathrm{mag}}>2 because αmag>1\alpha_{\mathrm{mag}}>1.

We now apply Theorem 6.4 from Benešova and Kružík [60] as follows. Define

Φ:Q×R3×(R3×3×R)→R∪{+∞}\Phi:Q\times\mathbb{R}^3\times(\mathbb{R}^{3\times3}\times\mathbb{R})\to\mathbb{R}\cup\{+\infty\}

according to

Φ(x,G,A,J)=∑i=14w~i(G,A,J).\Phi(x,G,A,\mathcal{J})=\sum_{i=1}^{4}\widetilde{w}_i(G,A,\mathcal{J}).

Since Φ\Phi is independent of xx it is clear that Φ(⋅,G,A,J)\Phi(\cdot,G,A,\mathcal{J}) is measurable for each (G,A,J)(G,A,\mathcal{J}). Since each of the w~i\widetilde{w}_i is continuous it is clear that Φ(x,⋅,⋅,⋅)\Phi(x,\cdot,\cdot,\cdot) is continuous for all x∈Qx\in Q. Since each of the w~i\widetilde{w}_i is convex in (A,J)(A,\mathcal{J}) it is also clear that Φ(x,G,⋅,⋅)\Phi(x,G,\cdot,\cdot) is convex for each (x,G)(x,G). In addition, since each of the w~i\widetilde{w}_i is non-negative the function Φ\Phi is bounded below by the zero function, which is a member of L1(Q)L^1(Q). It follows that Φ\Phi satisfies each of the hypothesis specified in Theorem 6.4 from Benešova and Kružík. Because p>3p>3 we know that GnG_n is contained in L1(Q,R3)L^1(Q,\mathbb{R}^3) and converges uniformly to GG. By Ball’s Corollary 6.2.2.[59], AnA_n converges weakly to AA in Lp/2(Q,R3)L^{p/2}(Q,\mathbb{R}^3) and Jn\mathcal{J}_n converges weakly to J\mathcal{J} in Lp/3(Q,R3)L^{p/3}(Q,\mathbb{R}^3). In particular (An,Jn)(A_n,\mathcal{J}_n) converges weakly to (A,J)(A,\mathcal{J}) in L1L^1. Setting zn=Gnz_n=G_n and an=(An,Jn)a_n=(A_n,\mathcal{J}_n) in Theorem 6.4, we therefore infer

∫QΦ(x,G(x),A(x),J(x)) d3x≤lim inf⁡n→∞∫QΦ(x,Gn(x),An(x),Jn(x)) d3x.(16)\int_Q \Phi(x,G(x),A(x),\mathcal{J}(x))\,\mathrm{d}^3x \leq\liminf_{n\to\infty}\int_Q \Phi(x,G_n(x),A_n(x),\mathcal{J}_n(x))\,\mathrm{d}^3x. \tag*{(16)}

To complete the proof we note that Gn∈ZG_n\in\mathcal{Z} by hypothesis and G∈ZG\in\mathcal{Z} by Lemma 1. In light of the fact that w~i:R3×R3×3×R→R∪{+∞}\widetilde{w}_i:\mathbb{R}^3\times\mathbb{R}^{3\times3}\times\mathbb{R}\to\mathbb{R}\cup\{+\infty\} extends wi:Q0×R3×3×R→R∪{+∞}w_i:Q_0\times\mathbb{R}^{3\times3}\times\mathbb{R}\to\mathbb{R}\cup\{+\infty\}, the following equalities therefore hold for almost all x∈Qx\in Q:

Φ(x,G(x),A(x),J(x))=w(G(x),A(x),J(x)),\Phi(x,G(x),A(x),\mathcal{J}(x))=w(G(x),A(x),\mathcal{J}(x)),
Φ(x,Gn(x),An(x),Jn(x))=w(Gn(x),An(x),Jn(x)).\Phi(x,G_n(x),A_n(x),\mathcal{J}_n(x))=w(G_n(x),A_n(x),\mathcal{J}_n(x)).

We conclude, by Eq. (16),

W(G)=∫Qw(G(x),A(x),J(x)) d3x=∫QΦ(x,G(x),A(x),J(x)) d3x≤lim inf⁡n→∞∫QΦ(x,Gn(x),An(x),Jn(x)) d3x=lim inf⁡n→∞∫Qw(Gn(x),An(x),Jn(x)) d3x=lim inf⁡n→∞W(Gn),(17)\begin{aligned} \mathrm{W}(G) &= \int_{\mathcal{Q}} w(G(x),\mathcal{A}(x),\mathcal{J}(x))\,d^3x \\ &= \int_{\mathcal{Q}} \Phi(x,G(x),\mathcal{A}(x),\mathcal{J}(x))\,d^3x \\ &\leq\liminf_{n\to\infty}\int_{\mathcal{Q}} \Phi(x,G_n(x),\mathcal{A}_n(x),\mathcal{J}_n(x))\,d^3x \\ &= \liminf_{n\to\infty}\int_{\mathcal{Q}} w(G_n(x),\mathcal{A}_n(x),\mathcal{J}_n(x))\,d^3x \\ &= \liminf_{n\to\infty}\mathrm{W}(G_n), \tag*{(17)} \end{aligned}

as claimed.

The main existence result for elasticized magnetohydrostatics is then.

Theorem 2. Suppose

3<αvacαmagαvac+αmag=p.3 < \frac{\alpha_{\mathrm{vac}}\alpha_{\mathrm{mag}}}{\alpha_{\mathrm{vac}}+\alpha_{\mathrm{mag}}}=p.

Then there is a G∗∈ZG_* \in\mathcal{Z} such that, for all G∈ZG \in\mathcal{Z}, W(G∗)≤W(G)\mathrm{W}(G_*) \leq\mathrm{W}(G).

Proof. Since W(G)≥0\mathrm{W}(G)\geq0 for all G∈ZG\in\mathcal{Z} the lower bound Wmin⁡=inf⁡{W(G)∣G∈Z}\mathrm{W}_{\min}=\inf\{\mathrm{W}(G)\mid G\in\mathcal{Z}\} is finite. Choose a bounded decreasing sequence of real numbers WN\mathrm{W}_N, N=1,2,…N=1,2,\ldots, with lim⁡N→∞WN=Wmin⁡\lim_{N\to\infty}\mathrm{W}_N=\mathrm{W}_{\min}. Let ZN={G∈Z∣W(G)≤WN}\mathcal{Z}_N=\{G\in\mathcal{Z}\mid\mathrm{W}(G)\leq\mathrm{W}_N\} denote the sublevel set at energy WN\mathrm{W}_N. For each NN the set ZN\mathcal{Z}_N must be non-empty, else WN\mathrm{W}_N would be a lower bound for W\mathrm{W} greater than Wmin⁡\mathrm{W}_{\min}. Therefore there is an infinite sequence GN∈ZG_N\in\mathcal{Z} with GN∈ZNG_N\in\mathcal{Z}_N for each NN. By Theorem 1 and GN∈ZNG_N\in\mathcal{Z}_N, the W1,pW^{1,p}-norm of GNG_N satisfies

∥GN∥1,p≤(c1W(GN)+c0)1/p≤(c1WN+c0)1/p.(18)\lVert G_N\rVert_{1,p}\leq\left(c_1\mathrm{W}(G_N)+c_0\right)^{1/p}\leq\left(c_1\mathrm{W}_N+c_0\right)^{1/p}. \tag*{(18)}

Here c0,c1c_0,c_1 are some positive real numbers given in Eqs. (14) and (15). Since the sequence WN\mathrm{W}_N is bounded the inequality Eq. (18) implies that the sequence GNG_N is bounded in W1,pW^{1,p}.

It is well-known that W1,p(Q,R3)W^{1,p}(\mathcal{Q},\mathbb{R}^3) is reflexive for p>1p>1. Since p>3p>3 by assumption, reflexivity of W1,p(Q,R3)W^{1,p}(\mathcal{Q},\mathbb{R}^3) comes for free. The Banach–Alaoglu theorem and the Eberlein–Šmulian theorem therefore imply that the bounded sequence GNG_N admits a weakly-convergent subsequence GN(n)∈ZG_{N(n)}\in\mathcal{Z}, n=1,2,…n=1,2,\ldots. Let G∗=lim⁡n→∞GN(n)G_*=\lim_{n\to\infty}G_{N(n)}. Since WN\mathrm{W}_N is a decreasing convergent sequence, lim⁡n→∞W(GN(n))=Wmin⁡\lim_{n\to\infty}\mathrm{W}(G_{N(n)})=\mathrm{W}_{\min}. Moreover, by Lemma 2,

W(G∗)≤lim inf⁡n→∞W(GN(n))=Wmin⁡.\mathrm{W}(G_*)\leq\liminf_{n\to\infty}\mathrm{W}(G_{N(n)})=\mathrm{W}_{\min}.

But Wmin⁡\mathrm{W}_{\min} is a lower bound on W\mathrm{W}, which implies Wmin⁡≤W(G∗)\mathrm{W}_{\min}\leq\mathrm{W}(G_*). Thus, W(G∗)=Wmin⁡\mathrm{W}(G_*)=\mathrm{W}_{\min}, and G∗∈ZG_*\in\mathcal{Z} is a global minimizer for W\mathrm{W}, as claimed.

Discussion

The proof presented here for existence of minimizers of the elasticized Newcomb potential W\mathrm{W} largely follows a pattern established in the context of mathematical elasticity [58, 59, 61, 62]. Its most novel mathematical ingredient comprises the method used to establish coercivity, which entails bounding the gradient of the back-to-labels map by its adjugate. The key inequality is given in Eq. (8), which is an elementary result in linear algebra.

The idea of modifying Newcomb’s potential by including the excess magnetic energy associated with turbulent fluctuations originated in earlier work in Ref. [44]. However, this earlier statistical equilibrium model assumed the one-point reference magnetic variance does not depend on the large-scale fluid state GG. It also fell short of establishing a rigorous existence theory. In contrast, the elasticized magnetohydrostatic model presented here hypothesizes that the reference magnetic variance grows with plasma strain due to profile stiffness. This profile stiffness effect, together with the vacuum barrier, are the crucial new ingredients needed for the existence theory presented here.

It is presently unclear whether solutions of elasticized magnetohydrostatics are always smoother than Bruno-Laurence solutions. While the rigorous existence theory presented here only guarantees minimizers in W1,p(Q,R3)W^{1,p}(Q,\mathbb{R}^3), a space that allows for jump discontinuities in the magnetic field, we find that well-resolved numerical solutions, such as those shown in Figure 1, do not exhibit jump discontinuities. Like the existence theory, the difficulty in assessing regularity of minimizers for the elasticized Newcomb potential is largely inherited from the theory of mathematical elasticity, where equilibrium regularity is an open problem in general [61]. This obstacle might be sidestepped by allowing for strain gradient dependence in the mean magnetic energy, which is specifically neglected in the present work. A plausible source of strain gradient dependence is large-scale turbulent fluid motions, as discussed in Holm’s formulation of the Lagrangian-averaged α\alpha-model [63].

The existence theory established here relies on both the vacuum barrier, proportional to ρvac>0\rho_{\mathrm{vac}}>0, and the nonlinear elastic response, proportional to 1/ΔB2>01/\Delta B^2>0. If ρvac→0\rho_{\mathrm{vac}}\to0 or 1/ΔB2→01/\Delta B^2\to0 the elasticized potential loses coercivity and the direct method in the calculus of variations cannot be applied. On the other hand, numerical simulations summarized in Figure 3 indicate that solutions remain well-behaved when either ρvac→0\rho_{\mathrm{vac}}\to0 or 1/ΔB2→01/\Delta B^2\to0. This suggests that neither the vacuum barrier nor the nonlinear elastic response is neccessary to obtain well-behaved equilibria using the elasticized model. This conjecture is also supported both by numerics and formal asymptotics from Ref. [44].

Self-convergence of computed solutions as either the nonlinear elastic response or vacuum barrier is removed

Figure 3. Self-convergence of computed solutions as either the nonlinear elastic response (left) or vacuum barrier (right) is removed. The reference magnetic variance is Σ0=λ2\Sigma_0=\lambda^2 with λ=0.05\lambda=0.05. Plotted curves show W1,pW^{1,p}-distance to a reference solution with ΔB2=104×λ2\Delta B^2=10^4\times\lambda^2 (left) or ρvac=0\rho_{\mathrm{vac}}=0 (right). Convergence studies with various values of ρvac\rho_{\mathrm{vac}} (left) or λ2/ΔB2\lambda^2/\Delta B^2 (right) are indicated with different colors.

If plasma incompressibility is introduced as an additional hypothesis in elasticized magnetohydrostatics, the vacuum barrier is completely unnecessary. It would be interesting to establish existence of incompressible minimizers without the vacuum barrier in future work. It would also be interesting to extend the modeling principles introduced here to other equations of state.

Theorem 2 on existence of minimizers for WW does not establish that minimizers satisfy the Euler-Lagrange equations. The gap between establishing existence of minimizers and establishing validity of the Euler-Lagrange equations is well-known in the calculus of variations, and in the theory of mathematical elasticity in particular [61, 62]. Part of the subtlety is already clear for minimization of the single-variable function f(x)=x2f(x)=x^2, with x≥1x\ge1, whose solution, x=1x=1, is not a critical point for ff. The subtleties are more intricate in infinite dimensions. Given the formal similarity of the elasticized Newcomb potential to potentials for hyperelastic solids, methods for establishing variants of the Euler-Lagrange equations developed within elasticity theory likely apply to the elasticized magnetohydrostatic model. This strategy for establishing some form of the Euler-Lagrange equations would be interesting to explore in future work. The strong-form Euler-Lagrange equations for the functional Eq. (6) are given by ∇⋅T=0\nabla\cdot\mathbf{T}=0, where the stress tensor T\mathbf{T} is given by

\mathbf{T}=-p\mathbf{I}+\mu_0^{-1}\left(\mathbf{B}\mathbf{B}^{\mathsf{T}}-\frac{1}{2}|\mathbf{B}|^2\mathbf{I}\right)+\mu_0^{-1}\left(\mathbf{\Sigma}-\frac{1}{2}\operatorname{tr}(\mathbf{\Sigma})\mathbf{I}\right) +\mu_0^{-1}\frac{1}{(\Delta B^2)^{\alpha_{\mathrm{mag}}-1}}\left(\alpha_{\mathrm{mag}}\mathbf{\Sigma}^{\alpha_{\mathrm{mag}}}-\frac{1}{2}\operatorname{tr}([2\alpha_{\mathrm{mag}}-1]\mathbf{\Sigma}^{\alpha_{\mathrm{mag}})\mathbf{I}\right),}

and the Eulerian fields p\mathbf{p}, B\mathbf{B}, Σ\mathbf{\Sigma} are given in terms of the back-to-labels map G\mathcal{G} by

p(x)=ρ2(x)∂ρU∗(ρ(x),s(x)),ρ(x)=J(x)ρ0(G(x)),s(x)=s0(G(x)),p(\mathbf{x})=\rho^2(\mathbf{x})\partial_\rho U^*(\rho(\mathbf{x}),s(\mathbf{x})),\qquad\rho(\mathbf{x})=\mathcal{J}(\mathbf{x})\rho_0(\mathcal{G}(\mathbf{x})),\qquad s(\mathbf{x})=s_0(\mathcal{G}(\mathbf{x})),
B(x)=A(x)B0(G(x)),Σ(x)=A(x)Σ0(G(x))AT(x).\mathbf{B}(\mathbf{x})=\mathbf{A}(\mathbf{x})\mathbf{B}_0(\mathcal{G}(\mathbf{x})),\qquad\mathbf{\Sigma}(\mathbf{x})=\mathbf{A}(\mathbf{x})\mathbf{\Sigma}_0(\mathcal{G}(\mathbf{x}))\mathbf{A}^{\mathsf{T}}(\mathbf{x}).

Here U∗U^* is defined in Eq. (5).

Data, Materials, and Software Availability. All code used to produce the figures in this manuscript is freely available at https://doi.org/10.5281/zenodo.23649878.

ACKNOWLEDGMENTS. This research was supported by: (I) the U.S. Department of Energy, Office of Science, Office of Fusion Energy Sciences, under contract DE-FG02-04ER54742; (II) a Google.org AI Impact Challenge: AI for Science award; and (III) the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research, as a part of the Mathematical Multifaceted Integrated Capability Centers program, under Award Number DE-SC0023164. The authors used Claude Opus 4.8 and 5.5 during manuscript preparation to assist with verification of analytical calculations. All Article prose was prepared by the authors without the assistance of artificial intelligence. The authors takes full responsibility for all final content, analysis, and interpretations.

References

  1. [4]J. Lion, J.-C. Anglès, L. Bonauer, A. Bañón Navarro, S.A. Cadena Ceron, R. Davies, M. Drevlak, N. Foppiani, J. Geiger, A. Goodman, W. Guo, E. Guiraud, F. Hernández, S. Henneberg, R. Herrero, C. Hintze, H. Höchter, J. Jelonek, F. Jenko, R. Jorge, M. Kaiser, M. Kubie, E. Lascas Neto, H. Laqua, M. Leoni, J.F. Lobsien, V. Maurin, A. Merlo, D. Middleton-Gear, M. Pascu, G.G. Plunk, N. Riva, M. Savchoutkov, F. Sciortino, J. Schilling, J. Shimwell, A. Di Siena, R. Slade, T. Stange, T.N. Todd, L. Wegener, F. Wilms, P. Xanthopoulos, and M. Zheng. Stellaris: A high-field quasi-isodynamic stellarator for a prototypical fusion power plant. Fusion Engineering and Design, 214:114868, May 2025. ISSN 09203796. . URL https://linkinghub.elsevier.com/retrieve/pii/S0920379625000705.
  2. [5]C. P. S. Swanson, S. T. A. Kumar, D. W. Dudt, E. R. Flom, W. B. Kalb, T. G. Kruger, M. F. Martin, J. R. Olatunji, S. Pasmann, L. Z. Tang, J. von der Linden, J. Wasserman, M. Avida, A. S. Basurto, M. Dickerson, N. de Boer, M. J. Donovan, A. H. Doudna Cate, D. Fort, W. Harris, U. Khera, A. Koen, J. A. Labatte, N. Maitra, A. Ottaviano, R. K. Parmar, E. J. Paul, B. Reydel, A. van Riel, P. K. Romano, M. Savastianov, S. Saxena, S. Seethalla, S. Srinivasan, R. H. Wu, D. Nash, J. Priebe, M. Shepchenkov, S. Walsh, B. Berzin, D. A. Gates, and the Thea Energy team. Overview of the Helios Design: A Practical Planar Coil Stellarator Fusion Power Plant, 2025. URL https://arxiv.org/abs/2512.08027. Version Number: 3.
  3. [6]D. A. Garren and A. H. Boozer. Existence of quasishelly symmetric stellarators. Physics of Fluids B: Plasma Physics, 3(10):2822–2834, October 1991. ISSN 0889-8221. . URL https://pubs.aip.org/pfb/article/3/10/2822/841899/Existence-of-quasihelically-symmetric-stellarators.DOI
  4. [7]S. P. Hirshman and J. C. Whitson. Steepest-descent moment method for three-dimensional magnetohydrodynamic equilibria. The Physics of Fluids, 26(12):3553–3568, December 1983. ISSN 0031-9171. . URL https://pubs.aip.org/pfl/article/26/12/3553/809269/Steepest-descent-moment-method-for-three.
  5. [8]Matt Landreman and Elizabeth Paul. Magnetic Fields with Precise Quasisymmetry for Plasma Confinement. Physical Review Letters, 128(3):035001, January 2022. ISSN 0031-9007, 1079-7114. . URL https://link.aps.org/doi/10.1103/PhysRevLett.128.035001.DOI
  6. [9]John R. Cary and Svetlana G. Shasharina. Omnigenity and quasisymmetry in helical plasma confinement systems. Physics of Plasmas, 4(9):3323–3333, September 1997. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/4/9/3323/457323/Omnigenity-and-quasisymmetry-in-helical-plasma.DOI
  7. [10]John R. Cary and Svetlana G. Shasharina. Helical Plasma Confinement Devices with Good Confinement Properties. Physical Review Letters, 78(4):674–677, January 1997. ISSN 0031-9007, 1079-7114. . URL https://link.aps.org/doi/10.1103/PhysRevLett.78.674.
  8. [11]A.G. Goodman, K. Camacho Mata, S.A. Henneberg, R. Jorge, M. Landreman, G.G. Plunk, H.M. Smith, R.J.J. Mackenbach, C.D. Beidler, and P. Helander. Constructing precisely quasi-isodynamic magnetic fields. Journal of Plasma Physics, 89(5):905890504, October 2023. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S002237782300065X/type/journal.article.
  9. [12]Samuel A. Lazerson, Joaquim Loizu, Steven Hirshman, and Stuart R. Hudson. Verification of the ideal magnetohydrodynamic response at rational surfaces in the VMEC code. Physics of Plasmas, 23(1):012507, January 2016. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/23/1/012507/108429/Verification-of-the-ideal-magnetohydrodynamic.DOI
  10. [13]Harold Grad. Toroidal Containment of a Plasma. The Physics of Fluids, 10(1):137–154, January 1967. ISSN 0031-9171. . URL https://doi.org/10.1063/1.1761965.
  11. [14]Peter Costorlan, Theodore D. Drivas, and Daniel Ginsberg. Flexibility and Rigidity in Steady Fluid Motion. Communications in Mathematical Physics, 385(1):521–563, July 2021. ISSN 0010-3616, 1432-0916. . URL https://link.springer.com/10.1007/s00220-021-04048-4.
  12. [15]Robert Cardona, Nathan Duignan, and David Perrella. Asymmetry of MHD Equilibria for Generic Adapted Metrics. Archive for Rational Mechanics and Analysis, 249(1):1, February 2025. ISSN 0003-9527, 1432-0673. . URL https://link.springer.com/10.1007/s00205-024-02075-8.arxiv.org/abs/2312.14368
  13. [16]Matt Landreman. Analytic toroidal 3D MHD equilibria and steady Euler flows with invariant surfaces, September 2026. URL https://arxiv.org/abs/2609.26742. Version Number: 2.
  14. [17]Opal Issan, Jonathan Citrin, Matt Landreman, and Brendan Tracey. Analytic non-axisymmetric 3D MHD equilibria without stellarator symmetry, October 2026. URL https://arxiv.org/abs/2610.07304. Version Number: 1.
  15. [18]M.I. Mikhailov, J. Núhrenberg, and H. Zille. Elimination of current sheets at resonances in three-dimensional toroidal ideal-magnetohydrodynamic equilibria. Nuclear Fusion, 59(6):066002, June 2019. ISSN 0029-5515, 1741-4326. . URL https://iopscience.iop.org/article/10.1088/1741-4326/ab0f50.DOI
  16. [19]Oscar P. Bruno and Peter Laurence. Existence of three-dimensional toroidal MHD equilibria with nonconstant pressure. Communications on Pure and Applied Mathematics, 49(7):717–764, 1996. ISSN 1097-0312. . URL https://onlinelibrary.wiley.com/doi/abs/10.1002/(%29?siC1%291097-0312%28199607%2949%3A7%3C717%3A%3AAID-CPAM3%3E3.0.CO%3B2-C. https://onlinelibrary.wiley.com/doi/pdf/10.1002/%28SICI%291097-0312%28199607%2949%3A7%3C717%3A%3AAID-CPAM3%3E3.0.CO%3B2-C.
  17. [20]John R Cary and Robert G Littlejohn. Noncanonical Hamiltonian mechanics and its application to magnetic field line flow. Annals of Physics, 151(1):1–34, November 1983. ISSN 00034916. . URL https://linkinghub.elsevier.com/retrieve/pii/000349168303135.DOI
  18. [21]S. R. Hudson, R. L. Dewar, G. Dennis, M. J. Hole, M. McGann, G. von Nessi, and S. Lazerson. Computation of multi-region relaxed magnetohydrodynamic equilibria. Physics of Plasmas, 19(11):112502, November 2012. ISSN 1070-664X. . URL https://doi.org/10.1063/1.4765691.
  19. [22]Peter Constantin and Federico Pasqualotto. Magnetic Relaxation of a Voigt–MHD System. Communications in Mathematical Physics, 402(2):1931–1952, September 2023. ISSN 1432-0916. . URL https://doi.org/10.1007/s00220-023-04770-1.
  20. [23]Yi-Min Huang, Justin Kin Jun Hew, Andrew Brown, and Amitava Bhattacharjee. Computation of magnetohydrodynamic equilibria with Voigt regularization. Physics of Plasmas, 32(6):062507, June 2025. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/32/6/062507/3350464/Computation-of-magnetohydrodynamic-equilibria-with.
  21. [24]Andrew Brown, Yi-Min Huang, and Amitava Bhattacharjee. Forced Reconnection in Voigt-Regularized MHD, 2026. URL https://arxiv.org/abs/2603.02432. Version Number: 1.
  22. [25]Yasuhiro Suzuki, Noriyoshi Nakajima, Kiyomasa Watanabe, Yuji Nakamura, and Takaya Hayashi. Development and application of HINT2 to helical system plasmas. Nuclear Fusion, 46(11):L19–L24, November 2006. ISSN 0029-5515, 1741-4326. . URL https://iopscience.iop.org/article/10.1088/0029-5515/46/11/L01.
  23. [26]Z. S. Qu, S. R. Hudson, R. L. Dewar, J. J. Hole, A. M. J. Hole. On the non-existence of stepped-pressure equilibria far from symmetry. Plasma Physics and Controlled Fusion, 63(12):125007, October 2021. ISSN 0741-3335. . URL https://dx.doi.org/10.1088/1361-6587/ac2afc.
  24. [27]D. Panici, R. Conlin, D. W. Dudt, K. Unalmis, and E. Kolemen. The DESC stellarator code suite. Part 1. Quick and accurate equilibria computations. Journal of Plasma Physics, 89(3):955890303, June 2023. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S0022377823000272/type/journal_article.DOI
  25. [28]D. W. Dudt and E. Kolemen. DESC: A stellarator equilibrium solver. Physics of Plasmas, 27(10):102513, October 2020. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/27/10/102513/108268/DESC-A-stellarator-equilibrium-solver.DOI
  26. [29]J. Loizu, S. R. Hudson, C. Núhrenberg, J. Geiger, and P. Helander. Equilibrium beta-limits in classical stellarators. Journal of Plasma Physics, 83(6):715830601, December 2017. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S0022377817000861/type/journal_article.
  27. [30]A. Baillod, J. Loizu, Z.S. H. P. Arbez, and J.P. Graves. Equilibrium beta-limits dependence on bootstrap current in classical stellarators. Journal of Plasma Physics, 89(5):905890508, October 2023. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S0022377823000910/type/journal_article.arxiv.org/abs/2211.12948
  28. [31]C. B. Smiet, J. Loizu, E. Balkovic, and A. Baillod. Efficient single-stage optimization of islands in finite-beta stellarator equilibria. Physics of Plasmas, 32(1):012504, January 2025. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/32/1/012504/3331878/Efficient-single-stage-optimization-of-islands-in.arxiv.org/abs/2407.02097
  29. [32]J. W. Connor, S. C. Cowley, R. J. Hastie, T. C. Hender, A. Hood, and T. J. Martin. Tearing modes in toroidal geometry. The Physics of Fluids, 31(3):577–590, March 1988. ISSN 0031-9171. . URL https://pubs.aip.org/pof/article/31/3/577/903632/Tearing-modes-in-toroidal-geometry.DOI
  30. [33]C. C. Hegna. The physics of neoclassical magnetohydrodynamic tearing modes. Physics of Plasmas, 5(5):1767–1774, May 1998. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/5/5/1767/1069111/The-physics-of-neoclassical-magnetohydrodynamic.DOI
  31. [34]C. C. Hegna. Physics of magnetic islands in stellarators. Physics of Plasmas, 19(5):056101, May 2012. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/19/5/056101/596911/Plasma-flow-healing-of-magnetic-islands-in.
  32. [35]Richard Fitzpatrick. Tearing mode dynamics in tokamak plasmas. IOP ebooks. IOP Publishing, Bristol, UK, version: 202306 edition, 2023. ISBN 978-0-7503-5367-0 978-0-7503-5366-3. .DOI
  33. [36]A. Reimann, M. Zarnstorff, D Monticello, A Weller, J. Geiger, and The W7-As Team. Pressure-induced breaking of equilibrium flux surfaces in the W7AS stellarator. Nuclear Fusion, 47(7):572–578, July 2007. ISSN 0029-5515, 1741-4326. . URL https://iopscience.iop.org/article/10.1088/0029-5515/47/7/008.
  34. [37]Jian Zhang, Ping Zhu, Haifeng Liu, Yasuhiro Suzuki, Chris C. Hegna, and Yuhong Xu. Identification of MHD equilibrium beta limits for CFQS magnetas. June 2026. URL http://arxiv.org/abs/2606.26675. arXiv:2606.26675 [physics.plasm-ph].
  35. [38]A. Reiman and H. Greenside. Calculation of three-dimensional MHD equilibria with islands and stochastic regions. Computer Physics Communications, 43(1):157–167, December 1986. ISSN 00104655. . URL https://linkinghub.elsevier.com/retrieve/pii/0010465586900597.
  36. [39]M. Drevlak, D. Monticello, and A Reiman. 3D MHD equilibria with improved initial conditions. Nuclear Fusion, 45(7):731–740, July 2005. ISSN 0029-5515, 1741-4326. . URL https://iopscience.iop.org/article/10.1088/0029-5515/45/7/022.
  37. [40]C. P. Hirshman, R. Sanchez, and C. R. Cook. SIESTA: A scalable iterative equilibrium solver for toroidal applications. Physics of Plasmas, 18(6):062504, June 2011. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/18/6/062504/387811/SIESTA-A-scalable-iterative-equilibrium-solver-for.
  38. [41]Tobias Bluckhan, Julianne Stratton, and Alan A. Kaptanoglu. MRX: A differentiale 3D MHD equilibrium solver without nested flux surfaces, 2025. URL https://arxiv.org/abs/2510.26986. Version Number: 1.
  39. [42]E. Balkovic, J. Loizu, E. Lanti, C. Smiet, C. Lazzati, R. Ramassamy, A. Goodman, J. Geiger, and J. P. Graves. SPECTRE: A robust solver for 3D equilibria with arbitrary topology, July 2026. URL https://arxiv.org/abs/2607.27135. Version Number: 1.
  40. [43]W. A. Newcomb. Reangangian and Hamiltonian methods in magnetohydrodynamics. Technical Report UCRL-6387, California Univ., Livermore (USA). Lawrence Livermore Lab., June 1961. URL https://www.osti.gov/biblio/5034470.
  41. [44]Maximilian Ruth, Joshua W. Burby, Wrick Sengupta, and Andrew Brown. Statistical equilibrium model for stellarators, 2026. URL https://arxiv.org/abs/2604.08878. Version Number: 1.
  42. [45]C J Cotter, D. D Holm, and P E Hydon. Multisymplectic formulation of fluid dynamics using the inverse map. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 463(2086):2671–2687, October 2007. ISSN 1364-5021, 1471-2946. . URL http://royalsocietypublishing.org/rsfa/article/463/2086/2671/82/.DOI
  43. [46]Bernard Dacorogna. Direct methods in the calculus of variations. Number v. 78 in Applied mathematical sciences. Springer, New York, 2nd ed edition, 2008. ISBN 978-0-387-55249-1.
  44. [47]Jeffrey P. Freidberg. Ideal MHD. Cambridge University Press, 1 edition, June 2014. ISBN 978-1-107-00625-6 978-0-511-79504-6. . URL https://www.cambridge.org/core/product/identifier/9780511795046/type/book.
  45. [48]Alexey Mishchenko, Matthias Borchardt, Roman Hatzky, Ralf Kleiber, Axel Könies, Carolin Nührenberg, Pavlos Xanthopoulos, Gareth Reesberg-Clark, and Gabriel G. Plunk. Global gyrokinetic simulations of electromagnetic turbulence in stellarator plasmas. Journal of Plasma Physics, 89(3):95589034, June 2023. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S0022377823000363/type/journal_article.
  46. [49]E. J. Spence, M. D. Nornberg, R. A. Bayliss, R. D. Kendrick, and C. B. Forest. Fluctuation-driven magnetic fields in the Madison Dynamo Experiment. Physics of Plasmas, 15(5):055910, May 2008. ISSN 1070-664X, 1089-7674. . URL https://pubs.aip.org/pop/article/15/5/055910/1015740/Fluctuation-driven-magnetic-fields-in-the-Madison.DOI
  47. [50]W. Z. Maw, W. X. Ding, H. Q. Liu, A. Ti, D. I. Brewer, H. L. Lian, Y. Q. Chu, L. Zeng, H. L. Zhao, J. Zhang, J. Qian, Q. Zang, X. Z. Gong, L. Q. Hu, L. Q. Xu, C. Zhou, T. Lan, A. D. Liu, J. L. Xie, G. Zhuang, Y. T. Song, B. N. Wang, and J. G. Li. Fluctuation-induced dynamo effect in a magnetic confinement plasma. Physical Review Research, 5(2):L022047, June 2023. ISSN 2643-1564. . URL https://link.aps.org/doi/10.1103/PhysRevResearch.5.L022047.
  48. [51]M.N.A. Beurskens, S.A. Bozhenkov, O. Ford, P. Xanthopoulos, A. Zocco, Y. Turkin, A. Alonso, C. Beidler, I. Calvo, D. Carralero, T. Estrada, G. Fuchert, O. Grulke, M. Hirsch, K. Ida, M. Jakubowski, C. Killer, M. Kryuchowa, S. Kwak, S. Lazercon, A. Langenberg, R. Lunsford, N. Pablant, E. Pasch, A. Pavone, F. Reimold, Th. Romba, A. Von Stechow, H.M. Smith, T. Windisch, M. Yoshimura, D. Zhang, R.C. Wolf, and The W7-X Team. Ion temperature clamping in Wendelstein 7-X electron cyclotron heated plasmas. Nuclear Fusion, 61(11):116072, November 2021. ISSN 0029-5515, 1741-4326. . URL https://iopscience.iop.org/article/10.1088/1741-4326/ac1653.
  49. [52]Sven Stroteich, Pavlos Xanthopoulos, Gabriel Plunk, and Ralf Schneider. Seeking turbulence optimized configurations for the Wendelstein 7-X stellarator: ion temperature gradient and electron temperature gradient turbulence. Journal of Plasma Physics, 88(5):175880501, October 2022. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S0022377822000757/type/journal_article.DOI
  50. [53]X Garbet, P Mantica, F Ryter, G Cordey, F Imbeaux, C Sozzi, A Manini, E Asp, V Parail, R Wolf, and The Jet Efda Contributors. Profile stiffness and global confinement. Plasma Physics and Controlled Fusion, 46(9):1351–1373, September 2004. ISSN 0741-3335, 1361-6587. . URL https://iopscience.iop.org/article/10.1088/0741-3335/46/9/002.
  51. [54]M. Barnes, F. I. Parra, and A. S. Schekochihin. Critically Balanced Ion Temperature Gradient Turbulence in Fusion Plasmas. Physical Review Letters, 107(11):115003, September 2011. ISSN 0031-9007, 1079-7114. . URL https://link.aps.org/doi/10.1103/PhysRevLett.107.115003.DOI
  52. [55]F. Van Wyk, E. G. Highcock, A. A. Schekochihin, C. M. Roach, A. R. Field, and W. Dorland. Transition to subcritical turbulence in a tokamak plasma. Journal of Plasma Physics, 82(6):905820609, December 2016. ISSN 0022-3778, 1469-7807. . URL https://www.cambridge.org/core/product/identifier/S0022377816001148/type/journal_article.
  53. [56]T. Adkins, J. G. Abel, M. Barnes, S. Buller, W. D. Narang, F. P. Ivanov, R. F. Parra, A. A. Schekochihin, and J. Squire. Asymptotic scaling theory of electrostatic turbulent transport in magnetised fusion plasmas, 2026. URL https://arxiv.org/abs/2601.15391. Version Number: 1.
  54. [57]W. A. Farmer, C. L. Ellison, J. H. Hammer, K. R. LeChien, N. B. Meezan, K. S. Raman, V. Svidzinski, and K. Tummel. Numerical Improvements in Magnetohydrodynamic, Pulsed Power Simulations of Near-Target Plasmas. IEEE Transactions on Plasma Science, 52(10):4771–4781, October 2024. ISSN 0093-3813, 1939-9375. . URL https://ieeexplore.ieee.org/document/10413242/.DOI
  55. [58]Philippe G. Ciarlet and Jindřich Nečas. Injectivity and self-contact in nonlinear elasticity. Archive for Rational Mechanics and Analysis, 97(3):171–188, September 1987. ISSN 0003-9107, 1432-0673. . URL http://link.springer.com/10.1007/BF00250807.
  56. [59]John M. Ball. Convexity conditions and existence theorems in nonlinear elasticity. Archive for Rational Mechanics and Analysis, 63(4):337–403, December 1976. ISSN 0003-9107, 1432-0673. . URL http://link.springer.com/10.1007/BF00279992.DOI
  57. [60]Barbora Benešová and Martin Kruží̌k. Weak Lower Semicontinuity of Integral Functionals and Applications. SIAM Review, 59(4):703–766, January 2017. ISSN 0036-1445, 1095-7200. . URL https://epubs.siam.org/doi/10.1137/16M1060947.DOI
  58. [61]John M. Ball. Some Open Problems in Elasticity. In Paul Newton, Philip Holmes, and Alan Weinstein, editors, Geometry, Mechanics, and Dynamics, pages 3–59. Springer-Verlag, New York, 2002. ISBN 978-0-387-95518-6. . URL http://link.springer.com/10.1007/0-387-21791-6_1.
  59. [62]J. M. Ball. Minimizers and the euler-lagrange equations. In Philippe G. Ciarlet and Maurice Roseau, editors, Trends and Applications of Pure Mathematics to Mechanics, volume 195, pages 1–4. Springer Berlin Heidelberg, Berlin, Heidelberg, 1984. ISBN 978-3-540-12916-5 978-3-540-38800-5. . URL http://link.springer.com/10.1007/3-540-12916-2_47. Series Title: Lecture Notes in Physics.DOI
  60. [63]Darryl D. Holm. Lagrangian averages, averaged Lagrangians, and the mean effects of fluctuations in fluid dynamics. Chaos: An Interdisciplinary Journal of Nonlinear Science, 12(2):518–530, June 2002. ISSN 1054-1500, 1089-7682. . URL https://pubs.aip.org/cha/article/12/2/518/134848/Lagrangian-averages-averaged-Lagrangians-and-the-pubmed.ncbi.nlm.nih.gov/12779582

Paper details

vlnc:W_m3f7cbne
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

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

  • Physics and Astronomy

Claude Opus 4.8 (bare), 5.5

View statement
Looking for related papers…
Contents