第 1 版
粒子模拟
3. Macroscopic equilibrium and stability
Avoiding a toroidal plasma current brings great advantages for plasma stability. Indeed, magnetohydrodynamic (MHD) stability plays a far less prominent role in stellarators than in tokamaks, where the toroidal current causes kink modes, sawteeth and resistive and neoclassical tearing modes that all limit the plasma performance. These instabilities are usually absent in stellarators for the simple reason that there is no, or very little, net toroidal plasma current.
There is, of course, a non-zero poloidal plasma current present to satisfy force balance, , and just like in a tokamak, the requirement implies the existence of a Pfirsch–Schluter current parallel to the magnetic¨ field, since the perpendicular current is generally not divergence-free. In addition, at low collisionality a bootstrap current arises for reasons similar to those in a tokamak, although the details are different because the particle orbits are not the same. The total toroidal current arising in this way is, however, usually substantially smaller than the typical (Ohmic) current in a tokamak. Of course, it is quite possible to drive a more substantial current using a transformer, and then tearing modes can be destabilized [6].
In principle neoclassical tearing modes can also exist in a stellarator, since these do not depend on a destabilizing Ohmic current profile for their existence but rather on a non-zero bootstrap current. The latter is reduced when a magnetic island forms and the pressure profile is flattened within it, and in tokamaks this negative current perturbation causes the island to grow further. In stellarators, however, the (global) magnetic shear usually has the opposite sign from that in tokamaks, so that, if an island should form and flatten the pressure profile, the resulting reduction in the bootstrap current makes the island shrink rather than grow. Neoclassical tearing modes are thus nonlinearly stable unless the bootstrap current is negative [7]. (The bootstrap current is taken to be positive if it increases the rotational transform.) In fact, finite plasma pressure often has the tendency to ‘heal’ magnetic islands in stellarators [8–10].
Stellarators do not experience plasma-terminating disruptions when, for instance, a stability limit is approached. The only exception seems to be situations where a transformer is used to induce so much toroidal current that tearing modes are destabilized [11]. After 120 000 plasma discharges, LHD has still not experienced a single current disruption [12].
In stellarators, the plasma density is not limited by the ‘Greenwald’ limit [13] but is instead determined by radiation losses from the plasma core. It is therefore a ‘soft’ limit and depends on the concentration and transport of impurities. Because the Greenwald limit is absent, stellarators often operate at a higher density than do tokamaks. The record is held by LHD, where densities are reached in so-called super-dense core plasmas. The question why stellarators do not have a Greenwald limit cannot be answered without understanding its origin in tokamaks. A recent explanation by Gates and Delgado-Aparicio [14] suggests that the limit is due to the destabilization of magnetic islands (in the outer regions rather than the core) by radiation losses: at high enough density these cannot be overcome by local Ohmic heating, and this leads to further island growth. As the authors themselves note, this mechanism would not operate in stellarators.
The pressure limit is also approached in a different way from that in tokamaks. The limit set by pressure-driven MHD modes is surprisingly ‘soft’; for instance, LHD routinely operates far above the ideal-MHD ballooning limit. It is not entirely clear why this is possible, but it appears that finite-Larmor-radius effects are playing a stabilizing role. Moreover, a significant fraction of the plasma pressure can be produced by suprathermal ions from neutral-beam injection, which has a stabilizing effect on ballooning modes [15]. By shaping the plasma appropriately, it is possible to raise the ideal ballooning limit significantly. In W7-X the volume-averaged normalized pressure limit is about . Because the stability limit is so high and soft, the equilibrium pressure limit is more important than in the tokamak. The Shafranov shift limits the pressure in a classical stellarator to about
where A denotes the aspect ratio. By optimizing the magneticfield geometry so as to reduce the Pfirsch–Schluter current, it¨ is possible to reduce the Shafranov shift significantly, and this has been carried out successfully in the designs of and W7-X. Another equilibrium effect limiting the achievable pressure is the tendency of the magnetic field to become stochastic in the edge region at high beta. Figure 2 shows calculations of this effect in W7-X using the PIES code, which solves the force balance equation without assuming nested flux surfaces [16]. As the volumeaveraged normalized pressure increases from 1% to 5%, the confinement region, i.e. the volume inside the largest closed flux surface, shrinks from 31.7 to 19.3 m<sup>3</sup>.

Figure 2. Poincare sections of two equilibria in W7-X with´ different normalized pressures. The pressure profiles are in both cases of the form , with s the normalized toroidal flux coordinate. The Shafranov shift and a stochastization of the edge region are clearly seen in the equilibrium.
Fast-ion-driven modes, such as toroidal Alfven eigen-´ modes (TAEs), have been observed in most stellarators. Such modes arise in the gaps of the continuous Alfven spectrum ´ that form when a plasma cylinder is bent into a torus. In stellarators, the breaking of axisymmetry gives rise to additional gaps and discrete modes lying therein. There are therefore more types of Alfven eigenmodes in stellarators than in toka- ´ maks, e.g. helicity- and mirror-induced Alfven eigenmodes,´ there are more wave–particle resonances, and thus more scope for instability [17]. For instance, while in a circular tokamak with large aspect ratio the main TAE resonances are and , where is the Alfven speed, it is possi-´ ble to have resonances with in stellarator geometry. On the other hand, the alpha-particle pressure in a reactor is proportional to the slowing-down time,
which is expected to be smaller in stellarators than in tokamaks, thanks to the ability of the former to operate above the Greenwald limit. The fast-particle drive for Alfvenic modes ´ can thus be smaller in the stellarator, if an operating point with high density can be chosen. Note that at constant thermal pressure, or fusion power , the alphaparticle pressure scales as , so increasing the plasma density by a factor of 2.5 leads to ten times lower alphaparticle pressure.
4. Moment-based preconditioning of the multi-dimensional Vlasov-Darwin PIC solver
The final set of equations solved in this study is comprised of the set of field-particle equations, Eqs. 53-56. We invert this system using a JFNK solver with nonlinear elimination, implemented and configured as described in Ref. [20]. In particular, the particle equations are enslaved to the field equations (so-termed particle enslavement) in a way such that only field variables are involved in the nonlinear residual. An advantage of this approach is that only a single copy of the particle variables is required (as in explicit PIC algorithms), and the memory footprint of the nonlinear solver is determined by the low-dimensional field variables. This results in memory requirements for the nonlinear solver comparable to fluid simulations.
For practical simulations, the convergence JFNK must be accelerated for eficiency. For Krylov iterative methods, this task is performed by the preconditioner. In the context of implicit PIC simulations, we seek an inexpensive approximation of the linearized kinetic solution. The basic idea is “physics-based”, i.e., we employ the linear response of and j, as obtained from approximate, linearized moment equations [7] to advance the linearized electromagnetic fields. This idea is at the root of semi-implicit moment methods [26–31], and has already been successfully explored to some degree in fully implicit 1D PIC [22, 24] (with limited models for the current response). Here, we generalize the fluid preconditioner to multi-D, and consider a general current response for moderately magnetized plasmas , namely, the electron plasma frequency is larger than the electron cyclotron frequency). We will demonstrate that the so-derived preconditioner features the correct ambipolar and MHD asymptotic responses, and is therefore suitable for arbitrary mass ratios and system lengths, provided that the plasma does not become strongly magnetized.
3. Macroscopic equilibrium and stability
Avoiding a toroidal plasma current brings great advantages for plasma stability. Indeed, magnetohydrodynamic (MHD) stability plays a far less prominent role in stellarators than in tokamaks, where the toroidal current causes kink modes, sawteeth and resistive and neoclassical tearing modes that all limit the plasma performance. These instabilities are usually absent in stellarators for the simple reason that there is no, or very little, net toroidal plasma current.
There is, of course, a non-zero poloidal plasma current present to satisfy force balance, , and just like in a tokamak, the requirement implies the existence of a Pfirsch–Schluter current parallel to the magnetic¨ field, since the perpendicular current is generally not divergence-free. In addition, at low collisionality a bootstrap current arises for reasons similar to those in a tokamak, although the details are different because the particle orbits are not the same. The total toroidal current arising in this way is, however, usually substantially smaller than the typical (Ohmic) current in a tokamak. Of course, it is quite possible to drive a more substantial current using a transformer, and then tearing modes can be destabilized [6].
In principle neoclassical tearing modes can also exist in a stellarator, since these do not depend on a destabilizing Ohmic current profile for their existence but rather on a non-zero bootstrap current. The latter is reduced when a magnetic island forms and the pressure profile is flattened within it, and in tokamaks this negative current perturbation causes the island to grow further. In stellarators, however, the (global) magnetic shear usually has the opposite sign from that in tokamaks, so that, if an island should form and flatten the pressure profile, the resulting reduction in the bootstrap current makes the island shrink rather than grow. Neoclassical tearing modes are thus nonlinearly stable unless the bootstrap current is negative [7]. (The bootstrap current is taken to be positive if it increases the rotational transform.) In fact, finite plasma pressure often has the tendency to ‘heal’ magnetic islands in stellarators [8–10].
Stellarators do not experience plasma-terminating disruptions when, for instance, a stability limit is approached. The only exception seems to be situations where a transformer is used to induce so much toroidal current that tearing modes are destabilized [11]. After 120 000 plasma discharges, LHD has still not experienced a single current disruption [12].
In stellarators, the plasma density is not limited by the ‘Greenwald’ limit [13] but is instead determined by radiation losses from the plasma core. It is therefore a ‘soft’ limit and depends on the concentration and transport of impurities. Because the Greenwald limit is absent, stellarators often operate at a higher density than do tokamaks. The record is held by LHD, where densities are reached in so-called super-dense core plasmas. The question why stellarators do not have a Greenwald limit cannot be answered without understanding its origin in tokamaks. A recent explanation by Gates and Delgado-Aparicio [14] suggests that the limit is due to the destabilization of magnetic islands (in the outer regions rather than the core) by radiation losses: at high enough density these cannot be overcome by local Ohmic heating, and this leads to further island growth. As the authors themselves note, this mechanism would not operate in stellarators.
The pressure limit is also approached in a different way from that in tokamaks. The limit set by pressure-driven MHD modes is surprisingly ‘soft’; for instance, LHD routinely operates far above the ideal-MHD ballooning limit. It is not entirely clear why this is possible, but it appears that finite-Larmor-radius effects are playing a stabilizing role. Moreover, a significant fraction of the plasma pressure can be produced by suprathermal ions from neutral-beam injection, which has a stabilizing effect on ballooning modes [15]. By shaping the plasma appropriately, it is possible to raise the ideal ballooning limit significantly. In W7-X the volume-averaged normalized pressure limit is about . Because the stability limit is so high and soft, the equilibrium pressure limit is more important than in the tokamak. The Shafranov shift limits the pressure in a classical stellarator to about
where A denotes the aspect ratio. By optimizing the magneticfield geometry so as to reduce the Pfirsch–Schluter current, it¨ is possible to reduce the Shafranov shift significantly, and this has been carried out successfully in the designs of and W7-X. Another equilibrium effect limiting the achievable pressure is the tendency of the magnetic field to become stochastic in the edge region at high beta. Figure 2 shows calculations of this effect in W7-X using the PIES code, which solves the force balance equation without assuming nested flux surfaces [16]. As the volumeaveraged normalized pressure increases from 1% to 5%, the confinement region, i.e. the volume inside the largest closed flux surface, shrinks from 31.7 to 19.3 m<sup>3</sup>.

Figure 2. Poincare sections of two equilibria in W7-X with´ different normalized pressures. The pressure profiles are in both cases of the form , with s the normalized toroidal flux coordinate. The Shafranov shift and a stochastization of the edge region are clearly seen in the equilibrium.
Fast-ion-driven modes, such as toroidal Alfven eigen-´ modes (TAEs), have been observed in most stellarators. Such modes arise in the gaps of the continuous Alfven spectrum ´ that form when a plasma cylinder is bent into a torus. In stellarators, the breaking of axisymmetry gives rise to additional gaps and discrete modes lying therein. There are therefore more types of Alfven eigenmodes in stellarators than in toka- ´ maks, e.g. helicity- and mirror-induced Alfven eigenmodes,´ there are more wave–particle resonances, and thus more scope for instability [17]. For instance, while in a circular tokamak with large aspect ratio the main TAE resonances are and , where is the Alfven speed, it is possi-´ ble to have resonances with in stellarator geometry. On the other hand, the alpha-particle pressure in a reactor is proportional to the slowing-down time,
which is expected to be smaller in stellarators than in tokamaks, thanks to the ability of the former to operate above the Greenwald limit. The fast-particle drive for Alfvenic modes ´ can thus be smaller in the stellarator, if an operating point with high density can be chosen. Note that at constant thermal pressure, or fusion power , the alphaparticle pressure scales as , so increasing the plasma density by a factor of 2.5 leads to ten times lower alphaparticle pressure.