An in-depth investigation of the braking of tearing mode rotation in tokamak plasmas via eddy currents induced in external ferromagnetic conducting structures is performed. In general, there is a “forbidden band” of tearing mode rotation frequencies that separates a branch of high-frequency solutions from a branch of low-frequency solutions. When a high-frequency solution crosses the upper boundary of the forbidden band, there is a bifurcation to a low-frequency solution, and vice versa. The bifurcation thresholds predicted by simple torque-balance theory (which takes into account the electromagnetic braking torque acting on the plasma, as well as the plasma viscous restoring torque, but neglects plasma inertia) are found to be essentially the same as those predicted by more complicated time-dependent mode braking theory (which takes inertia into account). Significant ferromagnetism causes otherwise electromagnetically thin conducting structures to become electromagnetically thick and also markedly decreases the critical tearing mode amplitude above which the mode “locks” to the conducting structures (i.e., the high-frequency to low-frequency bifurcation is triggered). On the other hand, if the ferromagnetism becomes too large, then the forbidden band of mode rotation frequencies is suppressed, and the mode frequency consequently varies smoothly and reversibly with the mode amplitude.

## I. INTRODUCTION

A tokamak is a device whose purpose is to confine a high-temperature plasma on a set of nested toroidal magnetic flux-surfaces.^{1} Charged particles are free to circulate rapidly around the flux-surfaces, but can only diffuse slowly across them, as a consequence of their relatively small gyroradii.

Tokamak plasmas are subject to a number of macroscopic instabilities that limit their effectiveness. Such instabilities can be divided into two broad classes. So-called *ideal instabilities* are non-reconnecting modes that disrupt the plasma in a matter of microseconds.^{2} However, such instabilities can usually be avoided by limiting the plasma pressure and/or by tailoring the toroidal current profile.^{3} *Tearing modes*, on the other hand, are relatively slowly growing instabilities that are more difficult to prevent.^{3,4} These instabilities tend to saturate at relatively low levels,^{5–8} in the process reconnecting magnetic flux-surfaces to form helical structures known as *magnetic islands*. Magnetic islands are radially localized structures centered on so-called rational flux-surfaces that satisfy $k\xb7B=0$, where $k$ is the wave vector of the instability, and $B$ the equilibrium magnetic field. Magnetic islands degrade plasma confinement because they enable heat and particles to flow very rapidly along field-lines from their inner to their outer radii, implying an almost complete loss of confinement in the region lying between these radii.^{9}

Toroidal plasma rotation plays an important, and in some cases a critical, role in tokamaks. One reason for this is that, with sufficiently fast rotation, magnetic islands co-rotate with the plasma at their associated rational surfaces.^{10} However, if the plasma rotation becomes too small, then such islands can lock (i.e., become stationary in the laboratory frame of reference).^{10–15} Mode locking has a variety of negative consequences. In particular, locking often results in a disruption (i.e., a total loss of plasma containment).^{16,17}

Mode locking usually takes place in two distinct stages.^{10} First, the island rotation frequency is reduced to a small fraction of its unperturbed value via electromagnetic torques associated with eddy currents induced in the conducting structures that inevitably surround the plasma. Typical examples of such structures include the vacuum vessel and passive plates used to stabilize kink modes. Henceforth, these structures are simply referred to as the “wall.” Second, the island locks to a static resonant error-field. (Such fields are always present in tokamak plasmas due to magnetic field-coil misalignments, poorly compensated coil feeds, etc.) Generally speaking, unless the error-field in question is unusually large, mode locking is not possible without the prior braking action of wall eddy currents.

A theory of tearing mode rotation braking due to eddy currents excited in the wall was developed by Nave and Wesson.^{13} According to this theory, the electromagnetic braking torque acting on the island is balanced by plasma inertia (because a significant fraction of the plasma is forced to co-rotate with the island as a consequence of the inability of the plasma to easily cross the island's magnetic separatrix, as well as the action of plasma perpendicular viscosity). In Nave and Wesson's theory, the island rotation frequency decelerates smoothly and reversibly as the island width increases. An alternative theory was subsequently developed by Fitzpatrick in which the electromagnetic braking torque is balanced by plasma viscosity, and plasma inertia plays a negligible role.^{10} Nave and Wesson's theory is appropriate to islands that grow comparatively rapidly, i.e., on a timescale that is small compared to the plasma momentum confinement time. Fitzpatrick's theory, on the other hand, is appropriate to islands that grow comparatively slowly: i.e., on a timescale that is long compared to the momentum confinement time. According to Fitzpatrick's theory, if the intrinsic plasma rotation at the rational surface is sufficiently large, then there is a “forbidden band” of island rotation frequencies. This band (which has been directly observed experimentally^{18}) separates a branch of high-frequency solutions from a branch of low-frequency solutions. If the island width becomes sufficiently large that a high-frequency solution crosses the upper boundary of the forbidden band, then there is a bifurcation to a low-frequency solution. This bifurcation is associated with a sudden collapse in the island rotation frequency to a comparatively low level. Such a collapse is almost certain to lead to mode locking.

Fitzpatrick's theory of tearing mode rotation braking is based on the hypothesis that solutions of the steady-state torque balance equation (the two torques in question being the electromagnetic braking torque and the viscous restoring torque) for which the island rotation frequency decreases with increasing island width are dynamically stable, whereas solutions for which the island frequency increases with increasing island width are dynamically unstable. The first aim of this paper is to explicitly test this hypothesis by comparing the predictions of simple torque-balance theory with those of the more complicated time-dependent mode braking equations (which take plasma inertia into account).

For the sake of simplicity, Fitzpatrick's theory of wall-induced tearing mode rotation braking assumes that the wall is electromagnetically thin (i.e., that the radial thickness of the wall is much less than the electromagnetic skin-depth in the wall material). This assumption is reasonable for present-day tokamaks, which tend to have relatively thin walls, but is not appropriate to next-step devices, such as ITER,^{19} which will necessarily have thick walls (for engineering reasons). Thus, the second aim of this paper is to generalize Fitzpatrick's theory to allow for electromagnetically thick walls.

Tokamak vacuum vessels are generally fabricated from a non-ferromagnetic metal such as copper, stainless steel, or Inconel. Up to now, it has been standard practice to avoid (whenever possible) incorporating ferromagnetic materials into tokamaks; first, because the magnetic fields generated by such materials greatly complicate plasma control; and, second, because ferromagnetic materials are known to destabilize ideal external-kink modes [and other magnetohydrodynamical (MHD) instabilities].^{20–22} However, economic and environmental considerations in future tokamak reactors demand that the amount of radioactive waste be kept to an absolute minimum. Hence, it is vitally important that the first wall and blanket, which will play the role of the wall in such devices, be fabricated from a low-activation material. Unfortunately, all of the most promising candidate materials, such as F82H steel,^{23} are ferromagnetic. Thus, the final aim of this paper is to assess the impact of wall ferromagnetism on wall-induced tearing mode rotation braking.

This paper is organized as follows. In Sec. II, we determine the response of a ferromagnetic wall, which can be either electromagnetically thin or thick, to the magnetic field generated by a rotating tearing mode. This permits us to write the plasma toroidal angular equation of motion, and, hence, to determine the effect of the wall on the mode rotation frequency. In Sec. III, we examine the limit in which the wall is electromagnetically thin. In Sec. IV, we examine the opposite limit in which the wall is electromagnetically thick. Finally, we summarize our findings in Sec. V.

## II. PRELIMINARY ANALYSIS

### A. Coordinates

Let us adopt the standard right-handed cylindrical coordinates: *r*, *θ*, *z*. The system is assumed to be periodic in the *z*-direction with period $2\pi \u2009R0$, where *R*_{0} is the simulated major radius.

### B. Plasma equilibrium

The equilibrium magnetic field is written as $B=(0,\u2009B\theta (r),\u2009Bz)$, whereas the equilibrium current density takes the form $J=(0,\u20090,\u2009Jz(r))$, with *J _{z}* = 0 for

*r*>

*a*. Here,

*a*is the plasma minor radius. Let

It follows that

where

is the safety-factor profile.

In the following, we shall adopt standard, large aspect-ratio, tokamak orderings, according to which $Bz\u226bB\theta ,\u2009R0\u226ba$, and $q\u223cO(1)$.^{10}

### C. Perturbed magnetic field

The helical magnetic field associated with a tearing mode is written

where

is the inverse aspect-ratio of the plasma. Let

where

is a simulated toroidal angle. Here, *ψ* is the helical magnetic flux (normalized with respect to $\u03f5a\u2009Bz\u2009a$) associated with the tearing mode, *m* > 0 is the poloidal mode number, and *n* > 0 the toroidal mode number. It follows that

### D. Marginally stable ideal-MHD physics

Except in the immediate vicinity of the rational surface, the tearing perturbation within the plasma is governed by the equations of marginally-stable, ideal-MHD, which easily yield the well-known *cylindrical tearing mode equation*:^{10}

Here, $\u2032\u2261d/dr$, and $qs=m/n$. We can treat any vacuum regions outside the plasma as zero-inertia plasmas characterized by *J* = 0. In this case, the previous equation simplifies to give

### E. Wall physics

Suppose that the plasma is surrounded by rigid wall of electrical conductivity *σ _{w}*, and relative magnetic permeability

*μ*, which extends from

_{w}*r*=

*r*to $r=rw+\delta w$, where $rw\u2265a$. Inside the wall, standard electrodynamics reveals that

_{w}### F. Eigenfunctions and stability parameters

Outside the wall, without loss of generality, we can write

Here, $\psi \u0302s(r)$ and $\psi \u0302w(r)$ are real piecewise continuous functions (which are analytic in the segments $0<r<rs\u2212,\u2009rs+<r<rw\u2212,\u2009rw+<r<\u221e$), whereas $\Psi s$ and $\Psi w$ are complex parameters. Furthermore,

and

where $q(rs)=qs$. Here, *r _{s}* is the minor radius of the rational surface, $\Psi s$ is the (normalized) reconnected magnetic flux at the rational surface, and $\Psi w$ is the (normalized) helical magnetic flux that penetrates the wall.

Let

Here, *E _{ss}* is the conventional ideal-wall tearing stability index (normalized to the plasma minor radius),

^{4}

*E*is an analogous index that governs the stability of the ideal-plasma resistive wall mode (which is assumed to be stable),

_{ww}^{24}and

*E*and

_{sw}*E*parameterize the electromagnetic coupling between the rational surface and the wall. Now, it follows from Eq. (9) that

_{ws}for $rs+\u2264r\u2264rw\u2212$. Hence, we deduce that

or

### G. Asymptotic matching

In the vacuum region lying outside the wall, for which $r>rw+\delta w$, the appropriate solution of Eq. (10) is

Hence,

Matching the normal and tangential components of the perturbed magnetic field across $r=rw+\delta w$ yields

which implies that the appropriate boundary condition (in the wall material) at the wall's outer radius is

Matching the normal and tangential components of the perturbed magnetic field across *r* = *r _{w}* yields

Hence, the appropriate boundary condition (in the wall material) at the wall's inner radius is

Moreover,

where the so-called *wall response function*, $\Delta w$, is defined

The wall response function (or, to be more exact, $\Delta \Psi w\u2261\Delta w\u2009\Psi w$) parameterizes the helical eddy currents excited in the wall. Equation (32) can be rearranged to give

Finally, asymptotic matching across the rational surface yields

Here, $\Delta \Psi s$ parameterizes the helical currents flowing in the immediate vicinity of the rational surface.^{25}

### H. Electromagnetic braking torque

The net toroidal electromagnetic braking torque acting in the immediate vicinity of the rational surface, as a consequence of eddy currents excited in the wall, is written^{10}

where

where

Here, *R _{w}* and

*I*are the real and imaginary parts of the wall response function, respectively.

_{w}### I. Magnetic island width

The reconnected magnetic flux at the rational surface is associated with an island chain whose full radial width (normalized to the minor radius) is^{10}

where

is the magnetic shear at the rational surface.

### J. Calculation of wall response function

Inside the wall, let us search for a separable solution of the form

where

Finally, Eq. (33) gives

According to Eq. (43),

where $z=p\u2009r,\u2009p=(\gamma \u2009\mu 0\u2009\mu w\u2009\sigma w)1/2$, *A* and *B* are arbitrary constants, and *I _{m}* and

*K*are modified Bessel functions. It follows from Eq. (45) that

_{m}where $z2=p\u2009(rw+\delta w)$, and

Hence, Eq. (44) gives

and

where $z1=p\u2009rw$, and

Note that $z2\u2009P(z2,z2)=k\u2009Q(z2,z2)$, in accordance with Eq. (45). It follows from Eqs. (47) and (52) that

Let

so that $z1=(\gamma \u2009\tau d)1/2/\u03f5w$ and $z2=z1\u2009(1+\u03f5w)$. Here, *τ _{d}* is the

*wall diffusion time*(i.e., the time required for magnetic flux to diffuse from the wall's inner to its outer radius), whereas

*ϵ*parameterizes the wall thickness. Now, in the limit $|z|\u226b1$,

_{w}Hence, in the limit $|z1|\u226b1$, or

we obtain

where $\gamma \u0302=\gamma \u2009\tau d,\u2009s=(rw+\delta w\u2212r)/\delta w=(z2\u2212z)/\gamma \u03021/2$, and it is assumed that

It follows from Eq. (55) that

Note that there is a small inconsistency in the analysis presented in this section. Equation (42) asserts that *ψ*_{0} is not a function of *t*, whereas it is clear from Eq. (51) that $\psi 0\u2261\psi 0(z)$ does depend on *t*, because *z* depends on *γ*, and *γ* is time-dependent in a non-steady-state situation. It is easily demonstrated that our expression for the wall response function is not invalidated by this inconsistency as long as

which reduces to

### K. Plasma equation of toroidal angular motion

Let

be the plasma toroidal angular velocity profile. Here, $\Omega \varphi (0)(r)$ is the steady-state profile in the absence of the braking torque ultimately induced by the helical eddy currents excited in the wall, whereas $\Delta \Omega \varphi (r,t)$ is the modification to the rotation profile generated by these currents. The plasma equation of toroidal angular motion is written as^{10}

where *ρ* and *μ* are the plasma mass density and perpendicular viscosity, respectively. For the sake of simplicity, there quantities are assumed to be uniform across the plasma. Finally, the boundary conditions satisfied by the perturbed toroidal angular velocity profile are^{10}

### L. No-slip constraint

The conventional *no-slip constraint* demands that the island chain co-rotate with the plasma at the rational surface.^{10} (This constraint holds as long as the island width is much greater than the linear layer width at the rational surface.) In other words,

where

Here,

is the steady-state island rotation frequency in the absence of the braking torque. Incidentally, we are assuming that the plasma poloidal angular velocity profile is fixed, as a consequence of strong poloidal flow damping.^{10}

### M. Normalized plasma equation of toroidal angular motion

Let $r\u0302=r/a,\u2009r\u0302s=rs/a,\u2009\Omega (r\u0302,t)=n\u2009\Delta \Omega \varphi /\omega 0$, $\tau H=(R0/n\u2009ss)\u2009(\mu 0\u2009\rho /Bz\u20092)1/2$, and $\tau M=\rho \u2009a\u20092/\mu $. Here, *τ _{H}* and

*τ*are the conventional

_{M}*hydromagnetic*and

*momentum confinement*timescales, respectively.

^{10}Equations (36), (38), (40), (68), and (71) yield the normalized no-slip constraint

as well as the normalized plasma equation of toroidal angular motion

where

Here,

and

Finally, it follows from Eq. (69) that

Note that $Rw=Rw(\gamma )$ and $Iw=Iw(\gamma )$, where

### N. Solution of normalized plasma equation of toroidal angular motion

Let

where $j0,k$ denotes the *k*th zero of the *J*_{0} Bessel function. It is easily demonstrated that

and

Let us write

### O. Rutherford island width evolution equation

The saturated width of the island chain is determined by the Rutherford island width evolution equation, which takes the form^{5,10}

where $\tau R=0.8227\u2009\mu 0\u2009rs\u20092/\eta $, and *η* is the plasma resistivity. It follows from Eqs. (34), (35), and (39) that

Here, we have explicitly taken account of the fact that *E _{ss}* (the conventional tearing $\Delta \u2032$ parameter) is generally a (fairly complicated) function of the island width,

^{6–8}whereas (in a steady state) the real and imaginary parts of the wall response function,

*R*and

_{w}*I*, respectively, are functions of the island frequency,

_{w}*ω*. In the following, for the sake of simplicity, we shall neglect any dependence of the saturated island width on the island frequency. This is equivalent to neglecting the second term on the right-hand side of the previous equation with respect to the first. Thus, the saturated island width is determined by the solution of

Note that it would be a fairly straightforward (but tedious) task to incorporate the frequency dependence of the island width into the following analysis.

## III. THIN-WALL REGIME

### A. Governing equations

Consider the *thin-wall regime*, which is characterized by $|\gamma \u2009\tau d|\u226a1$. In this regime, the wall is “electromagnetically thin”: i.e., its radial thickness is much less that the skin depth in the wall material (calculated with the growth-rate *γ*). It follows from Eq. (64) that the wall response function reduces to

in the thin-wall regime, where

is the *wall time-constant* (i.e., the effective L/R time of the wall). Here, for the sake of simplicity, it is assumed that $k\u2243m\u2009\u2009\mu w\u226b1$. The approximations used in deriving Eq. (89) are valid provided

Let us suppose that the first term on the right-hand side of Eq. (79) is dominant. It follows that

which implies that [see Eq. (39)]

Thus, the approximations employed in writing Eq. (92) are valid provided

### B. Thin-wall rotation-braking equations

Let

Here, for the sake of simplicity, *E _{ww}* has been given the conventional value

which corresponds to the absence of plasma current in the region $r>rs+$. In Eqs. (97) and (98), $t\u0302$ is the normalized time, $\omega \u0302$ is the normalized island rotation frequency, $r\u0302s$ is the normalized minor radius of the rational surface, *κ* (primarily) parameterizes the intrinsic plasma rotation, and *X* parameterizes the island width.

### C. Steady-state solutions

In a steady-state (i.e., $d/dt\u0302=0$), Eq. (97) yields

Here, use has been made of the identity

The right-hand side of Eq. (101) represents the electromagnetic braking torque, due to eddy currents induced in the wall, which acts to reduce the island rotation frequency, whereas the left-hand side represents the viscous restoring torque that acts to maintain the rotation.

Equation (101) yields

In Refs. 10, 27, and 28, it is argued that solutions of Eq. (101) for which $dX/d\omega \u0302<0$ (i.e., solutions for which the island rotation frequency decreases as the island width increases) are dynamically stable, whereas solutions for which the $dX/d\omega \u0302>0$ are dynamically unstable. This argument leads to the conclusion that, when $\kappa <\kappa c=1/27$ (i.e., when the intrinsic plasma rotation is sufficiently high), the general solution of the torque-balance equation exhibits a “forbidden band” of island rotation frequencies. [Incidentally, the value of *κ _{c}* is obtained from the simultaneous solution of $dX/d\omega \u0302=d\u20092X/d\omega \u0302\u20092=0$, with $dX/d\omega \u0302$ and $d\u20092X/d\omega \u0302\u20092=0$ specified by Eqs. (103) and (104), respectively. At the critical point, $\omega \u0302=\omega \u0302c=1/3$ and $X=Xc=32/27$.] This band separates a branch of dynamically stable low-frequency solutions from a branch of dynamically stable high-frequency solutions. Thus, when a low-frequency solution crosses the lower boundary of the forbidden band, it becomes dynamically unstable, and there is (presumably) a bifurcation to a high-frequency solution (characterized by the same values of

*X*and

*κ*). Likewise, when a high-frequency solution crosses the upper boundary of the forbidden band, it becomes dynamically unstable, and there is (presumably) a bifurcation to a low-frequency solution.

The critical values of *X* and $\omega \u0302$ where the aforementioned bifurcations take place, which are numerically determined from the solution of $dX/d\omega \u0302=0$, with $dX/d\omega \u0302$ specified by Eq. (103), are shown in Figs. 1 and 2. It can be seen that, to a good approximation, there is a bifurcation from a high-frequency solution [i.e., $\omega \u0302\u223cO(1)$] to a low-frequency solution [i.e., $\omega \u0302\u223cO(\kappa )$] when *X* exceeds the critical value

Just before the bifurcation, $\omega \u0302$ takes the value

On the other hand, there is a bifurcation from a low-frequency to a high-frequency solution when *X* falls below the critical value

Just before the bifurcation, $\omega \u0302$ takes the value

Of course, Eqs. (105)–(108) are only valid when $\kappa <\kappa c$. For $\kappa \u2265\kappa c$, there are no bifurcations, which implies that $\omega \u0302$ varies smoothly and reversibly with *X*. The fact that $X\u2212<X+$ suggests that the high-frequency/low-frequency bifurcation cycle exhibits considerable hysteresis.^{10} Note, incidentally, that Eqs. (105)–(108) are analytic approximations to the exact solutions (which cannot be expressed in closed forms).

### D. Time-dependent solutions

Figure 3 shows a typical solution of the thin-wall rotation-braking equations, (97)–(98), in which the island width parameter, *X*, is very slowly (compared to the momentum confinement timescale) ramped up from zero to some maximum value that exceeds that at which the bifurcation from high-frequency to low-frequency solution branches is predicted to occur (according to torque-balance theory), and then very slowly ramped back down to zero. It can be seen that, at first, the normalized plasma rotation frequency, $\omega \u0302$, decreases smoothly as *X* increases. Eventually, however, when the rotation frequency has been reduced to about half of its original value ($\omega \u0302=1$), the frequency drops precipitously to a very low value (compared to its original value). This is the high-frequency to low-frequency bifurcation. Conversely, the rotation frequency initially rises smoothly, as *X* decreases, although it still takes a comparatively low value. Eventually, however, the frequency rises precipitously, attaining a value that is comparable with its original one. This is the low-frequency to high-frequency bifurcation. Incidentally, the hysteresis in the high-frequency/ low-frequency bifurcation cycle is illustrated by the strong asymmetry (about $t\u0302=100$) evident in the $\omega \u0302$–$t\u0302$ curve plotted in Fig. 3.

Let $\Gamma =|d\omega \u0302/dt\u0302|$ be the normalized acceleration/deceleration in the island rotation frequency. In the following, the start and end times of the high-to-low-frequency bifurcation are conveniently defined as the times at which $\Gamma $ first exceeds the critical value 0.05, and subsequently first falls below this value, respectively. (Note that, except during the bifurcations, $\Gamma $ is typically much smaller than 0.05.) The start and end times of the low-to-high-frequency bifurcation are defined in an analogous manner.

Figures 4 and 5 show the bifurcation thresholds (i.e., the values of *X* and $\omega \u0302$ at the start times of the high-to-low-frequency and low-to-high-frequency bifurcations) determined from the thin-wall rotation-braking equations, as a function of the parameter *κ*. In all cases, *X*(*t*) is ramped up and down again, in the manner described in the caption to Fig. 3. Also, shown are the analytic approximations (105)–(108) that were previously derived from torque-balance theory. It is clear that the analytic approximations are fairly accurate, which confirms, first, that torque-balance theory is capable of correctly predicting the occurrence of the high-to-low-frequency and low-to-high-frequency bifurcations, and, second, that solutions of the thin-wall torque-balance equation characterized by $dX/d\omega \u0302>0$ are indeed dynamically unstable.

Of course, torque-balance theory is incapable of predicting the duration of a given bifurcation. Figure 6 shows the durations (i.e., the difference between the end and start times, as defined previously) of the high-to-low-frequency and low-to-high-frequency bifurcations determined from the thin-wall rotation-braking equations, as a function of the parameter *κ*. As before, *X*(*t*) is ramped up and down again, in the manner described in the caption to Fig. 3. It is clear that, in order to reach completion, both types of bifurcations require a time interval whose duration is of order a momentum confinement time. Moreover, at small *κ*, the low-to-high-frequency bifurcation is slightly faster than the high-to-low-frequency bifurcation, whereas the opposite is true at high *κ*.

It is apparent, from the preceding discussion, that [see Eq. (92)]

Hence, the inequalities (95) yield

where *τ _{W}* is the timescale on which the island width varies. Furthermore, the inequality (66) reduces to (assuming that $k\u2009\u03f5w\u226a1$—see Sec. III E)

Finally, according to Eq. (91), the thin-wall regime only holds when (using $k\u2243m\u2009\mu w$)

Thus, the previous three inequalities are the criteria for the validity of the theory presented in this section.

### E. Effect of wall ferromagnetism

It follows from Eqs. (50), (57), (77), (96), (105), and (107) that, in the thin-wall regime, the critical island width above which the high-to-low-frequency bifurcation is triggered can be written as

where

Likewise, the critical island width below which the low-to-high-frequency bifurcation is triggered takes the form

Incidentally, assuming, for the sake of simplicity, that there is negligible plasma current in the region $r>rs+$, it is easily demonstrated that

where $\zeta =(rs/rw)\u20092\u2009m$. This result has been incorporated into Eq. (114).

There are bifurcations between high-frequency and low-frequency solution branches when *λ*, which parameterizes the degree of wall ferromagnetism, lies in the range $0<\lambda <\lambda c$. For $\lambda \u2265\lambda c$, there are no bifurcations, and the island frequency consequently varies smoothly and reversibly as the island width varies. If $\lambda c<0$ then there are no bifurcations, irrespective of the value of *λ*. According to Eq. (91), the thin-wall ordering is only valid provided

Thus, in order to have bifurcations in the thin-wall regime, we require

However, it is clear that if $0<\lambda <\lambda c$, and the previous inequality is satisfied, then $0<\lambda \u226a1$. In this case, it follows from Eqs. (113) and (117) that the bifurcation thresholds only exhibit a very weak dependence on *λ*. In other words, in the thin-wall regime, wall ferromagnetism has little effect on the bifurcation thresholds. In fact, the main effect of increasing ferromagnetism (i.e., increasing *λ*) is to cause the inequality (119) to be violated, and, hence, to push the system into the thick-wall regime described in Sec. IV.

Incidentally, in the previous expressions, *ω*_{0} is the island rotation frequency in the absence of wall braking, *τ _{H}* and

*τ*are the hydromagnetic and momentum confinement timescales, respectively (see Sec. II M),

_{M}*τ*is the wall time-constant [see Eq. (90)],

_{w}*r*is the wall minor radius,

_{w}*δ*is the wall radial thickness,

_{w}*μ*is the wall relative permeability,

_{w}*m*is the poloidal mode number of the tearing mode,

*q*is the safety-factor at the associated rational surface,

_{s}*r*is the minor radius of the rational surface, and

_{s}*a*is the plasma minor radius.

## IV. THICK-WALL REGIME

### A. Governing equations

Consider the *thick-wall regime*, which is characterized by $|\gamma \u2009\tau d|\u226b1$. In this regime, the wall is “electromagnetically thick”: i.e., its radial thickness is much greater that the skin depth in the wall material (calculated with the growth-rate *γ*). It follows from Eq. (64) that the wall response function reduces to

in the thick-wall regime. Here, it is again assumed that $k\u2243m\u2009\mu w\u226b1$. The approximations used in deriving Eq. (121) are valid provided

Let us suppose that the first term on the right-hand side of Eq. (79) is dominant. It follows that

which implies that

Thus, the approximations employed in writing Eq. (123) are valid provided

### B. Thick-wall rotation-braking equations

Let

Here, *E _{ww}* has again been given the conventional value specified in Eq. (99). As before, $t\u0302$ is the normalized time, $\omega \u0302$ is the normalized island rotation frequency, $r\u0302s$ is the normalized minor radius of the rational surface, $\kappa \u2032$ parameterizes the intrinsic plasma rotation, and $X\u2032$ parameterizes the island width.

### C. Steady-state solutions

In a steady-state (i.e., $d/dt\u0302=0$), Eq. (128) yields

which can be combined with Eq. (129) to give the *thick-wall torque-balance equation*

Here, use has been made of the identity (102). As before, the right-hand side of Eq. (131) represents the electromagnetic braking torque, due to eddy currents induced in the wall, which acts to reduce the island rotation frequency, whereas the left-hand side represents the viscous restoring torque that acts to maintain the rotation.

Equation (101) yields

By analogy with the analysis of Sec. III C, we expect solutions of Eq. (131) for which $dX\u2032/d\omega \u0302<0$ to be dynamically stable, and solutions for which $dX\u2032/d\omega \u0302>0$ to be dynamically unstable. This argument again leads to the conclusion that, when $\kappa \u2032<\kappa c\u2032=1/21.56$ (i.e., when the intrinsic plasma rotation is sufficiently high), the general solution of the torque-balance equation exhibits a “forbidden band” of island rotation frequencies. [Incidentally, the value of $\kappa c\u2032$ is obtained from the simultaneous solution of $dX\u2032/d\omega \u0302=d\u20092X\u2032/d\omega \u0302\u20092=0$, with $dX\u2032/d\omega \u0302$ and $d\u20092X\u2032/d\omega \u0302\u20092=0$ given by Eqs. (132) and (133), respectively. At the critical point, $\omega \u0302=\omega \u0302c=1/9.198$ and $X\u2032=Xc\u2032=1.795$.] This band separates a branch of dynamically stable low-frequency solutions from a branch of dynamically stable high-frequency solutions. As before, when a low-frequency solution crosses the lower boundary of the forbidden band, it becomes dynamically unstable, and there is (presumably) a bifurcation to a high-frequency solution (characterized by the same values of $X\u2032$ and $\kappa \u2032$), and vice versa.

The critical values of $X\u2032$ and $\omega \u0302$ where the aforementioned bifurcations take place, which are numerically determined from the solution of $dX\u2032/d\omega \u0302=0$, with $dX\u2032/d\omega \u0302$ specified by Eq. (132), are shown in Figs. 7 and 8. It can be seen that, to a good approximation, there is a bifurcation from a high-frequency solution [i.e., $\omega \u0302\u223cO(1)$] to a low-frequency solution [i.e., $\omega \u0302\u223cO(\kappa \u2032)$] when $X\u2032$ exceeds the critical value

Just before the bifurcation, $\omega \u0302$ takes the value

On the other hand, there is a bifurcation from a low-frequency to a high-frequency solution when $X\u2032$ falls below the critical value

Just before the bifurcation, $\omega \u0302$ takes the value

Of course, Eqs. (134)–(137) are only valid when $\kappa \u2032<\kappa c\u2032$. For $\kappa \u2032\u2265\kappa c\u2032$, there are no bifurcations, which implies that $\omega \u0302$ varies smoothly and reversibly with $X\u2032$. As before, the fact that $X\u2212\u2032<X+\u2032$ suggests that the high-frequency/low-frequency bifurcation cycle exhibits considerable hysteresis. Note, incidentally, that Eqs. (134)–(137) are analytic approximations to the exact solutions (which cannot be expressed in closed forms).

### D. Time-dependent solutions

Figure 9 shows a typical solution of the thick-wall rotation-braking equations (128) and (129), in which the island width parameter, $X\u2032$, is very slowly (compared to the momentum confinement timescale) ramped up from zero to some maximum value that exceeds that at which the bifurcation from high-frequency to low-frequency solution branches is predicted to occur (according to torque-balance theory) and then very slowly ramped back down to zero. It can be seen that, at first, the normalized plasma rotation frequency, $\omega \u0302$, decreases smoothly as $X\u2032$ increases. Eventually, however, when the rotation frequency has been reduced to about one third of its original value ($\omega \u0302=1$), the frequency drops precipitously to a very low value (compared to its original value.). This is the high-frequency to low-frequency bifurcation. Conversely, the rotation frequency initially rises smoothly, as *X* decreases, although it still takes a comparatively low value. Eventually, however, the frequency rises precipitously, attaining a value that is comparable with its original one. This is the low-frequency to high-frequency bifurcation. As before, the hysteresis in the high-frequency/low-frequency bifurcation cycle is illustrated by the strong asymmetry (about $t\u0302=100$) evident in the $\omega \u0302$–$t\u0302$ curve plotted in Fig. 9.

In the following, the start and end times of the high-to-low-frequency bifurcation are conveniently defined as the times at which $\Gamma \u2261|d\omega \u0302/dt\u0302|$ first exceeds the critical value 0.075, and subsequently first falls below this value, respectively. (Note that, except during the bifurcations, $\Gamma $ is typically much smaller than 0.075.) The start and end times of the low-to-high-frequency bifurcation are defined in an analogous manner.

Figures 10 and 11 show the bifurcation thresholds (i.e., the values of $X\u2032$ and $\omega \u0302$ at the start times of the high-to-low-frequency and low-to-high-frequency bifurcations) determined from the thick-wall rotation-braking equations, as a function of the parameter $\kappa \u2032$. In all cases, $X\u2032(t)$ is ramped up and down again, in the manner described in the caption to Fig. 9. Also, shown are the analytic approximations (134)–(137) that were previously derived from torque-balance theory. It is again clear that the analytic approximations are fairly accurate, which confirms, first, that torque-balance theory is capable of correctly predicting the occurrence of the high-to-low-frequency and low-to-high-frequency bifurcations, and, second, that solutions of the thick-wall torque-balance equation characterized by $dX\u2032/d\omega \u0302>0$ are indeed dynamically unstable.

As before, torque-balance theory is incapable of predicting the duration of a given bifurcation. Figure 12 shows the durations (i.e., the difference between the end and start times, as defined previously) of the high-to-low-frequency and low-to-high-frequency bifurcations determined from the thick-wall rotation-braking equations, as a function of the parameter $\kappa \u2032$. Here, $X\u2032(t)$ is ramped up and down again, in the manner described in the caption to Fig. 9. It is clear that, in order to reach completion, both types of bifurcations require a time interval whose duration is of order a momentum confinement time. Moreover, at small $\kappa \u2032$, the low-to-high-frequency bifurcation is slightly faster than the high-to-low-frequency bifurcation, whereas the opposite is true at high $\kappa \u2032$.

It is again apparent from the preceding discussion that [see Eq. (123)]

Hence, the inequalities (126) yield

where *τ _{W}* is the timescale on which the island width varies. Furthermore, the inequality (66) reduces to

Finally, according to Eq. (122), the thick-wall regime only holds when (using $k\u2243m\u2009\mu w$)

Thus, the previous three inequalities are the criteria for the validity of the theory presented in this section.

### E. Effect of wall ferromagnetism

It follows from Eqs. (50), (57), (77), (118), (127), (134), and (136) that, in the thick-wall regime, the critical island width above which the high-to-low-frequency bifurcation is triggered can be written

where

and $\kappa c\u2032=1/21.56$. Likewise, the critical island width below which the low-to-high-frequency bifurcation is triggered takes the form

Here, *W*_{2}, *λ*, *ω*_{0}, *τ _{w}*,

*m*, and

*ζ*are defined in Sec. III E.

There are bifurcations between high-frequency and low-frequency solution branches when *λ*, which parameterizes the degree of wall ferromagnetism, lies in the range $0<\lambda <\lambda c$. For $\lambda \u2265\lambda c$, there are no bifurcations, and the island frequency consequently varies smoothly and reversibly as the island width varies. According to Eq. (122), the thick-wall ordering is only valid provided

Thus, in order to have bifurcations in the thick-wall regime, we require

This criterion is easily satisfied provided that $\omega 0\u2009\tau w$ is sufficiently large (i.e., provided that the intrinsic plasma rotation is sufficiently large).

Figure 13 shows $W+/W3$ and $W\u2212/W3$ plotted as functions of $\lambda /\lambda c$. It can be seen that, for $\lambda \u226a\lambda c$, increasing wall ferromagnetism (i.e., increasing *λ*) causes a marked decrease in $W+$, but leaves $W\u2212$ virtually unaffected. In other words, ferromagnetism significantly reduces the critical island width above which a rotating tearing mode “locks” to the wall, but makes little difference to the critical width below which the mode “unlocks.” On the other hand, $W+$ asymptotes to $W\u2212$ (which remains essentially independent of *λ*) from above as $\lambda \u2192\lambda c$. Of course, there are no bifurcations for $\lambda \u2265\lambda c$. Hence, increasing wall ferromagnetism eventually suppresses the bifurcations, in which case the island rotation frequency varies smoothly and reversibly with the island width.

## V. SUMMARY

This paper contains an in-depth investigation of the braking of tearing mode rotation in tokamak plasmas via eddy currents induced in external conducting structures. The analysis is subject to a number of limitations. First, it only applies to large-aspect ratio, low-*β*, circular cross-section, tokamak plasmas. Second, it assumes that the tearing mode amplitude varies on a timescale that is much longer than the plasma momentum confinement timescale. Third, it assumes that the plasma density and perpendicular viscosity are uniform across the plasma. Fourth, it assumes that the plasma poloidal rotation profile is fixed due to the action of strong poloidal flow damping. Fifth, it neglects the transient response of the eddy currents to the magnetic field generated by the rotating tearing mode. Finally, it assumes that the conducting structure, or “wall,” is of uniform thickness, resistivity, and magnetic permeability, is physically thin (i.e., its radial thickness is much less than its minor radius), and concentric with the plasma.

As is well-known,^{10} for sufficiently large intrinsic plasma rotation, there is a “forbidden band” of tearing mode rotation frequencies that separates a branch of high-frequency solutions from a branch of low-frequency solutions. When a high-frequency solution crosses the upper boundary of the forbidden band, there is a bifurcation to a low-frequency solution, and vice versa. There is considerable hysteresis in this process, because the critical tearing mode amplitude above which the high-frequency to low-frequency bifurcation is triggered is much larger than the critical mode amplitude below which the opposite bifurcation is triggered.

The aims of this paper were threefold. First, to verify that the bifurcation thresholds predicted by simple torque-balance theory (which takes into account the electromagnetic braking torque acting on the plasma, and well as the plasma viscous restoring torque, but neglects plasma inertia) are the same as those predicted by more complicated time-dependent mode braking theory (which takes inertia into account). This was, indeed, found to be the case (see Secs. III D and IV D). Second, to generalize existing theory to allow for electromagnetically thick conducting structures. This was achieved in Sec. IV. Finally, to generalize the theory to allow the conducting structures to be ferromagnetic. The results of this generalization are described in Secs. III E and IV E. It is found that significant ferromagnetism causes otherwise electromagnetically thin conducting structures to become electromagnetically thick, and also markedly decreases the critical tearing mode amplitude above which the mode “locks” to the conducting structures (i.e., the high-frequency to low-frequency bifurcation is triggered). On the other hand, if the ferromagnetism becomes too large, then the forbidden band of mode rotation frequencies is suppressed, and the mode frequency consequently varies smoothly and reversibly with the mode amplitude.

## ACKNOWLEDGMENTS

This research was funded by the U.S. Department of Energy under Contract No. DE-FG02-04ER-54742.