Home Search Collections Journals About Contact us My IOPscience

Chapter 3: MHD stability, operational limits and disruptions

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2007 Nucl. Fusion 47 S128

(http://iopscience.iop.org/0029-5515/47/6/S03)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 128.206.9.138
The article was downloaded on 06/07/2013 at 19:49

Please note that terms and conditions apply.

IOP PUBLISHING and INTERNATIONAL ATOMIC ENERGY AGENCY NUCLEAR FUSION
Nucl. Fusion 47 (2007) S128–S202 doi:10.1088/0029-5515/47/6/S03

Chapter 3: MHD stability, operational

limits and disruptions
T.C. Hender1,a , J.C Wesley2 , J. Bialek3 , A. Bondeson4 , A.H. Boozer3 , R.J. Buttery1 ,
A. Garofalo3 , T.P Goodman5 , R.S. Granetz6 , Y. Gribov7 , O. Gruber8 , M. Gryaznevich1 ,
G. Giruzzi9 , S. Günter8 , N. Hayashi10 , P. Helander11 , C.C. Hegna12 , D.F. Howell1 ,
D.A. Humphreys2 , G.T.A. Huysmans9 , A.W. Hyatt2 , A. Isayama10 , S.C. Jardin13 ,
Y. Kawano10 , A. Kellman2 , C. Kessel13 , H.R. Koslowski14 , R.J. La Haye2 , E. Lazzaro15 ,
Y.Q. Liu4 , V. Lukash16 , J. Manickam13 , S. Medvedev17 , V. Mertens8 , S.V. Mirnov18 ,
Y. Nakamura10 , G. Navratil3 , M. Okabayashi13 , T. Ozeki10 , R. Paccagnella19 ,
G. Pautasso8 , F. Porcelli20 , V.D. Pustovitov16 , V. Riccardo1 , M. Sato10 , O. Sauter5 ,
M.J. Schaffer2 , M. Shimada7 , P. Sonato19 , E.J. Strait2 , M. Sugihara7 , M. Takechi10 ,
A.D. Turnbull2 , E. Westerhof21 , D.G. Whyte12 , R. Yoshino10 , H. Zohm8 and the ITPA
MHD, Disruption and Magnetic Control Topical Group
1
EURATOM/UKAEA Fusion Association, Culham Science Centre, Abingdon, Oxon, OX14 3DB, UK
2
General Atomics, PO Box 85608, San Diego, California 92186-5608, USA
3
Columbia University, New York, New York 10027, USA
4
EURATOM/VR Fusion Association, Chalmers University of Technology, Göteborg, Sweden
5
Ecole Polytechnique Fédérale de Lausanne (EPFL), Centre de Recherches en Physique des Plasmas, Association
Euratom-Confédération Suisse, CH-1015 Lausanne, Switzerland
6
Massachusetts Institute of Technology, Plasma Science and Fusion Center, Cambridge, MA 02139, USA
7
ITER Organization, Cadarache Centre, 13108 St Paul lez Durance, France
8
Max-Planck Institut für Plasmaphysik, EURATOM Association, D-85748 Garching, Germany
9
Association Euratom-CEA, CEA Cadarache, F-13108 St Paul-lez-Durance, France
10
Japan Atomic Energy Agency, Naka, Ibaraki-ken, 311-0193, Japan
11
Max-Planck Institut für Plasmaphysik, EURATOM Association, D-17491 Greifswald, Germany
12
Department of Engineering Physics, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA
13
Princeton Plasma Physics Laboratory, PO Box 451, Princeton, NJ 08543, USA
14
Forschungszentrum Jülich GmbH, Association EURATOM-FZ Jülich, Institut für Plasmaphysik, Trilateral Euregio Cluster,
D-52425 Jülich, Germany
15
IFP CNR, Assoc. Euratom-ENEA-CNR, Via R. Cozzi 53, Milan, Italy
16
Nuclear Fusion Institute, Russian Research Centre ’Kurchatov Institute’, Moscow, Russian Federation
17
Keldysh Institute of Applied Mathematics, Moscow, Russian Federation
18
State Research Centre TRINITI, Troitsk, Moscow Region, 142190 Russian Federation
19
Consorzio RFX, Assoc. Euratom-ENEA, I-35127 Padova, Italy
20
Burning Plasma Research Group, Politecnico di Torino, Torino 10129, Italy
21
FOM-Institute for Plasma Physics Rijnhuizen, Association EURATOM-FOM, Trilateral Euregio Cluster, PO Box 1207,
3430 BE Nieuwegein, Netherlands
E-mail: [email protected]

Received 20 June 2006, accepted for publication 8 January 2007

Published 1 June 2007
Online at stacks.iop.org/NF/47/S128
Abstract
Progress in the area of MHD stability and disruptions, since the publication of the 1999 ITER Physics Basis
document (1999 Nucl. Fusion 39 2137–2664), is reviewed. Recent theoretical and experimental research has
made important advances in both understanding and control of MHD stability in tokamak plasmas. Sawteeth are
anticipated in the ITER baseline ELMy H-mode scenario, but the tools exist to avoid or control them through localized
current drive or fast ion generation. Active control of other MHD instabilities will most likely be also required in
ITER. Extrapolation from existing experiments indicates that stabilization of neoclassical tearing modes by highly
localized feedback-controlled current drive should be possible in ITER. Resistive wall modes are a key issue for

a Author to whom any correspondence should be addressed.

0029-5515/07/060128+75$30.00 © 2007 IAEA, Vienna Printed in the UK S128

 Chapter 3: MHD stability, operational limits and disruptions

advanced scenarios, but again, existing experiments indicate that these modes can be stabilized by a combination of
plasma rotation and direct feedback control with non-axisymmetric coils. Reduction of error fields is a requirement
for avoiding non-rotating magnetic island formation and for maintaining plasma rotation to help stabilize resistive
wall modes. Recent experiments have shown the feasibility of reducing error fields to an acceptable level by means
of non-axisymmetric coils, possibly controlled by feedback. The MHD stability limits associated with advanced
scenarios are becoming well understood theoretically, and can be extended by tailoring of the pressure and current
density profiles as well as by other techniques mentioned here. There have been significant advances also in the
control of disruptions, most notably by injection of massive quantities of gas, leading to reduced halo current fractions
and a larger fraction of the total thermal and magnetic energy dissipated by radiation. These advances in disruption
control are supported by the development of means to predict impending disruption, most notably using neural
networks. In addition to these advances in means to control or ameliorate the consequences of MHD instabilities,
there has been significant progress in improving physics understanding and modelling. This progress has been in
areas including the mechanisms governing NTM growth and seeding, in understanding the damping controlling
RWM stability and in modelling RWM feedback schemes. For disruptions there has been continued progress on the
instability mechanisms that underlie various classes of disruption, on the detailed modelling of halo currents and
forces and in refining predictions of quench rates and disruption power loads. Overall the studies reviewed in this
chapter demonstrate that MHD instabilities can be controlled, avoided or ameliorated to the extent that they should
not compromise ITER operation, though they will necessarily impose a range of constraints.

PACS numbers: 52.55.Tn, 52.55.Fa

(Some figures in this article are in colour only in the electronic version)

Contents
1. Introduction
2. MHD stability
2.1. Sawtooth oscillations
2.1.1. Physics of sawtooth oscillations
2.1.2. Sawtooth control
2.1.3. Central MHD activity expected in ITER
2.2. Neoclassical tearing modes
2.2.1. Physics of neoclassical tearing modes
2.2.2. Active control of NTMs
2.2.3. Mitigation of NTMs
2.2.4. NTMs expected in ITER and their control
2.3. Resistive wall modes
2.3.1. Physics of resistive wall modes
2.3.2. Control of RWMs
2.3.3. RWMs expected in ITER and their control
2.4. Error fields
2.4.1. Effect of error fields on plasmas
2.4.2. Error fields expected in ITER and their correction
2.5. MHD stability in advanced scenarios
2.5.1. MHD stability in plasmas with weak magnetic shear
2.5.2. MHD stability in plasmas with strong negative magnetic shear
2.5.3. Expectations for ITER
2.6. Summary
3. Disruptions
3.1. Disruption characteristics, causes and frequency
3.2. Thermal quench energy loss and deposition
3.3. Current quench dynamics
3.3.1. Global toroidal current decay
3.3.2. Vertical instability, halo currents and mechanical forces
3.4. Runaway electrons generated by disruptions
3.4.1. Observations in present tokamaks
3.4.2. Interaction of runaway electrons with plasma facing components
3.4.3. Confinement, termination and mitigation of runaway electrons
3.4.4. Summary and implications for ITER

S129
T.C. Hender et al

3.5. Integrated modelling and simulation

3.5.1. ITER disruption simulations
3.5.2. DINA disruption modelling basis
3.5.3. Predictions for major disruptions and VDEs in ITER
3.5.4. Calculation of non-axisymmetric halo current effects in ITER by M3D code
3.5.5. Summary and future needs
3.6. Disruption avoidance, prediction and mitigation
3.6.1. Disruption prediction
Discussion and future work
3.6.2. Disruption avoidance
Passive avoidance
Disruption avoidance strategies
3.6.3. Disruption mitigation
Methods of disruption mitigation
Pellet injection
Massive gas injection
Liquid jets
Other concepts
Application to ITER
3.7. Summary and R&D needs

1. Introduction these disruption predictions to one or more means for ‘fast-

shutdown’ intervention which (ideally) will make it possible
Magnetohydrodynamic (MHD) stability and the microscopic to either avoid the occurrence of disruption or to mitigate or
and macroscopic effects of various classes of MHD soften the consequences of those disruptions which cannot be
instability underlie essentially all aspects of achievable avoided.
plasma performance in tokamaks and determine the principal A comprehensive discussion of the various aspects of
operational limits for tokamaks—maximum plasma current MHD stability that contribute to the limits on plasma current
and plasma pressure (beta) and pressure gradient—and plays and beta and pressure gradient anticipated in ITER, together
an important role in the closely related operational limit on with a discussion of the specialized nomenclature that applies
maximum achievable or usable plasma density. The beta to consideration of MHD stability in tokamak systems, can
and density limits are two of the three key physics basis be found in chapter 3 of the ITER Physics Basis (IPB)
considerations—energy confinement is the third—that govern [1]. The discussion therein proceeds from a fundamental
the basic design and plasma performance of reactor tokamaks. understanding of ideal and resistive MHD stability theory to the
The combination of these three considerations determines the more specific manifestations and consequences (operational
achievable fusion power and neutron wall loading and fusion limits) of various major categories of ideal and/or resistive
power gain (Q = Pfus /Paux ; with Pfus being the fusion power MHD instabilities: e.g. sawtooth oscillations, neoclassical
tearing modes, error-field-induced locked tearing modes, edge-
and Paux the auxiliary power heating of the plasma).
localized modes (ELMs) and fast-growing internal modes
In contrast, disruptions, which arise from the immediate
localized near regions of high internal pressure gradient
or eventual consequences of MHD instability, set a second
in advanced-performance plasmas with internal transport
type of constraint on the operational feasibility of reactor
barriers (ITBs).
tokamaks—they can determine the operational lifetime of The MHD stability content of this chapter 3 of this issue
certain components, especially those associated with plasma ‘Progress in the ITER Physics Basis’ follows the same format
power and particle exhaust. Since occurrence of at least and approach as the IPB chapter 3. Section 2.1 discusses the
a small number of disruptions is inevitable in any tokamak present understanding of the physics, control and application
that seeks to explore burning plasma science in the reactor to ITER of the m = 1, n = 1 internal reconnection instability
regime, the at-risk components and systems—especially the localized near the q = 1 surface that commonly manifests
plasma-facing-component substrates and surfaces—must be itself as sawtooth oscillations. The principal concern about
designed to accommodate the immediate effects of a finite sawtooth oscillations in ITER lies in their potential to provide
number of disruptions without premature failure or loss of ‘seed islands’ (due to the occurrence of a radial magnetic
function. Furthermore, since the consequences of disruption field resonant at given q = m/n surfaces) that in turn can
will likely shorten the usable operation life of plasma-facing- lead to onset of neoclassical tearing mode (NTM) instability.
components, it is highly desirable to (1) minimize both Section 2.2 addresses the physics, avoidance, control and
the number and severity of disruptions that do occur and effect on ITER of this resistive MHD instability. In ITER,
(2) ameliorate or ‘soften’ the consequences of disruptions uncontrolled growth of certain classes of NTM instability
that cannot otherwise be avoided. To this end, reactor is predicted to lead to deterioration of energy confinement
tokamaks and ITER require (1) means to reliably predict the and possibly, at high-enough plasma pressure, to the onset
pending occurrence of disruption and (2) provision to couple of a disruption. The ITER ‘reference case’ ELMy H-mode

S130
 Chapter 3: MHD stability, operational limits and disruptions

plasma operation scenario, with a positive magnetic shear Basis. Here, section 3.1 provides an introduction to disruption
profile, is particularly susceptible to various NTMs. The phenomenology and characteristics, and an overview of
present understanding of the physics for the onset of NTMs how these disruptions impact on various ITER systems and
in ITER suggests that NTMs in positive-shear scenarios will components. Sections 3.2 and 3.3 address details of the
be inevitable if large enough seed instabilities (e.g. sawteeth) thermal quench and current quench phases of a disruption,
occur. Thus as Section 2.2.4 details, the present ITER design respectively. Section 3.3 also includes discussion of the effect
includes provision of electron cyclotron current drive (ECCD) of vertical instability (the so-called vertical displacement event
for NTM control which is highly desirable for ensuring that or VDE) and the resulting generation of circulating poloidal
ITER is able to reach its sustained burning plasma operation currents (halo currents) in the ITER blanket and vacuum
goal using the ELMy H-mode scenario. vessel structures. Section 3.4 addresses the critical area of
Section 2.3 addresses the broad topic of resistive wall runaway electron generation in an ITER-class tokamak and
modes (RWMs), which are expected to be the principal beta- the means/prospects for runaway avoidance/mitigation.
limiting instability in ‘advanced-performance’ steady-state- Section 3.5 addresses the status of modelling of
capable plasmas, based on ‘reversed shear’ or ‘negative shear’ disruptions and VDEs, and of halo currents and runaway
operation. Resistive wall modes, which arise in any tokamak conversion in ITER. Section 3.6 explores the closely related
plasma when the plasma pressure exceeds the ideal n = 1 subjects of disruption prediction and the prospects for taking
external kink ‘no-wall’ beta limit, are expected to set the action to avoid a pending disruption or to implement a fast-
beta limit (and disruption-onset beta) in ITER steady-state shutdown action to avoid or at least partially mitigate the most
scenarios. As section 2.3 explains, RWMs can be controlled serious consequences of disruption.
either by plasma rotation or by providing feedback control of Section 3.7 concludes the disruption part of the
the mode’s non-axisymmetric field, or by a combination of chapter with a summary and synthesis of the expected
the two. Section 2.3.3 details the critical-issue for advanced ITER disruption characteristics and prospects for achieving
tokamak (AT) scenarios of how rotational and feedback the disruption prediction, avoidance and disruption effect
stabilization will affect ITER’s capability for AT steady-state mitigation needed for successful operation of ITER. ‘Critical
operation. path’ R&D needs in each of the disruption characterization
The effects of small departures of the ITER magnetic and prediction/avoidance/mitigation topical areas are also
field configuration from exact axisymmetry is addressed in identified here.
section 2.4. A strategy for ITER that identifies construction
and alignment accuracies for the toroidal and poloidal field 2. MHD stability
coil systems, and the provision of a multi-mode ‘error
field correction’ coil system is expected to allow successful 2.1. Sawtooth oscillations
mitigation of error-field-induced instabilities and disruptions.
2.1.1. Physics of sawtooth oscillations. Sawtooth
Such a practice is now routine in many present generation
oscillations are periodic, MHD initiated mixing events that
tokamaks.
occur in a tokamak plasma in the near axis region where
The effect of localized internal MHD instabilities excited
the safety factor q is less than or equal to unity. Small
by excessive pressure gradients associated with ITBs in sawteeth, with mixing radii of about 20–40% of the plasma
ITER plasmas is addressed in section 2.5. The principal minor radius and temperature drops of the order of a fraction
conclusion here is that to avoid such instabilities becoming of a keV, can be survived easily by the plasma discharge.
the limiting factor during ITER operation with ITB plasmas Indeed, small sawteeth may even be beneficial in preventing
(as is envisioned for the reference case ITER steady-state the accumulation of impurities and helium ash in the plasma
scenario), undue localization of the barrier must be avoided. centre. On the other hand, large sawteeth with mixing radii
This requirement may, in turn, mean that the ITB pressure of 50% or more of the plasma minor radius and temperature
gradient of such a plasma must be controlled. This subject of drops of one or more keV represent a serious threat to ITER
ITB pressure gradient control is addressed in chapter 6 of this operation, since these large sawtooth events may couple to
issue [2]. NTMs [4] and to edge-localized modes [5], resulting in a
Section 2.6 provides a summary and synthesis of the serious loss of plasma energy and confinement degradation.
prospects for achieving the MHD stability and the control An additional worry, in a fusion burning plasma, is that if the
of instabilities needed for successful operation in the various sawtooth period is shorter than the slowing down time of the
scenarios envisioned for ITER. ‘Critical path’ R&D needs in fusion alpha particles then fusion alphas may be scattered, and
each of the MHD topical areas are also identified here. perhaps lost, before they have time to transfer their energy to
The related subjects of edge-localized modes (ELMs) the thermal plasma. Therefore, the best sawtooth regime for
and density limit associated with the plasma boundary power ITER would be one where the sawtooth period is intermediate
balance are treated in chapter 4 of this issue [3], while that of between the alpha particle slowing down time and the plasma
energetic-particle-modes (toroidal Alfvén eigenmodes (TAE) heating (or energy confinement) time. Thus, in recent years
and the like) is treated in chapter 6 of this issue [2]. a large effort has been dedicated to developing techniques for
The remainder of the chapter (section 3) covers sawtooth control, reviewed in section 2.1.2, and in predicting
disruption physics and characteristics, disruption modelling the sawtooth behaviour in projected ITER discharges, as
and simulation and disruption avoidance, prediction and discussed in section 2.1.3. In this section, we review briefly the
mitigation. Again, the organization and nomenclature largely present theoretical understanding of the sawtooth phenomenon
follows the corresponding section 3.4 of the 1999 ITER Physics and indicate areas where further theoretical work is needed.

S131
T.C. Hender et al

The underlying mechanism responsible for the sawtooth grow very rapidly as soon as it can be detected, in apparent
crash is MHD instability of an internal kink mode, with toroidal contradiction with the slowly evolving equilibrium plasma
mode number n = 1 and dominant poloidal mode number parameters near the threshold and the fact that, near threshold,
m = 1. However, it is clear from experiments that ideal the linear growth rate is very small (= 0 exactly at threshold).
MHD theory alone is not accurate enough in predicting the One possible resolution of this problem is that diamagnetic
threshold for the onset of the sawtooth crash. Non-ideal effects, effects, which provide an effective stabilization mechanism,
such as kinetic effects associated with energetic particles are quenched as soon as the mode grows to a relatively
(including the fusion alphas), finite electrical resistivity and small amplitude, such that the corresponding magnetic island
viscosity, diamagnetic effects, finite ion Larmor radius, becomes comparable with the ion Larmor radius [8]. Another
electron inertia, electron compressibility, play an important possibility is that a secondary instability (such as a ballooning
role in determining the actual stability threshold for the mode or a resistive interchange mode) is triggered when the
sawtooth crash. These effects have been incorporated in a magnetic island reaches a small width [9]. In both cases, the
phenomenological sawtooth trigger criterion, first proposed linear stability threshold proposed in [6] would still be fairly
in [6] (see also section 2.1.3, where the model is described adequate in predicting the effective onset of the sawtooth crash,
in some detail), and the reader is referred to this reference as these secondary effects (diamagnetic quenching, secondary
and to other references cited therein for a discussion of the instabilities) still require the internal kink mode to go linearly
non-ideal physics that determines the actual linear stability unstable before the crash can occur.
threshold of m = n = 1 modes in realistic tokamak In conclusion, we may say that our level of theoretical
discharges. This threshold depends on the actual collisionality understanding of the sawtooth phenomenon is certainly
regime (i.e. collisional, semi-collisional or collisionless), on incomplete, but is probably sufficient to define a credible
the distribution of energetic ions in phase space and on the strategy for the prediction and the control of sawteeth in
radial profiles of the plasma current density and pressure, ITER. This statement is corroborated by our proven ability
including their local gradients near the q = 1 surface. Thus of controlling the sawtooth period and amplitude in existing
theoretical studies suggest that an efficient way to control experiments, as described in the next section.
sawteeth would be by affecting the value of magnetic shear
near the q = 1 surface, for instance by means of localized 2.1.2. Sawtooth control. Sawtooth control refers to the
heating and/or current drive. ability of an actuator system—here, some form of additional
The sawtooth trigger criterion has been incorporated in heating and/or current drive—to alter the period of the sawtooth
transport codes and has proved to be relatively successful in instability. Since the ITER Physics Basis [1] there has
predicting quantitatively the sawtooth period and amplitude been considerable experimental and theoretical progress on
in existing tokamak experiments, as discussed in section 2.1.2. controlling sawteeth.
However, it is important to remark that this trigger criterion has There are two main approaches to sawtooth control. In
a number of limitations. First of all, being predicated on linear the first, one attempts to eliminate or avoid the sawtooth crash
theory, it may not be adequate to describe situations where for the duration of the discharge by lengthening the sawtooth
long-lived precursor oscillations of relatively large amplitude period (stabilization); while in the second, the goal is to
exist during part of the sawtooth ramp, as in these situations increase the rate of the sawteeth (destabilization) to reduce
a non-linear trigger condition would be more appropriate. the perturbation to the plasma at each individual crash.
Secondly, the sawtooth trigger condition says nothing about In recent years, sawtooth control has been studied and used
the non-linear consequences of the instability, and in particular to advantage (e.g. see section 2.2) on many experiments. The
on the relaxed q profile after the sawtooth crash. In existing experiments fall into two broad categories with either central
codes, the Kadomtsev prescription [7] for the relaxed q profile, or off-axis power deposition. All heating and/or current drive
based on full reconnection of all magnetic surfaces inside the methods—NBI, ICRF and ECRF—acting alone can alter the
q = 1 volume, is often used. However, many experiments sawtooth period (as can combinations of the different heating
indicate that partial reconnection may also occur. A partial methods).
reconnection model was proposed in [6], which contains a Sawtooth modelling (e.g. [6, 10–14]) is often carried out
free parameter, Wcrit /r1 , where Wcrit is a critical island width using (a) transport simulations (e.g. PRETOR or ASTRA
for the onset of widespread magnetic turbulence and r1 is the codes) which evolve the electron, ion temperatures and current
q =1 minor radius (full Kadomtsev reconnection corresponds profiles between crashes, (b) sawtooth trigger criteria (e.g.
to Wcrit /r1
2). As this free parameter (Wcrit /r1 ), which is [14] and references therein) and (c) profile relaxation rules
not determined within this theoretical model, is reduced, so the at the crash—based for example on the Kadomtsev magnetic
predicted sawtooth period becomes shorter.Thus a theoretical reconnection model. Partial reconnection is sometimes
determination of Wcrit /r1 in realistic plasma regimes would modelled but, at least for NBI stabilization [15], the general
certainly improve our degree of confidence in predicting dependence of the period on the power remains the same
the sawtooth behaviour in ITER. Consideration of three- as for full reconnection: the absolute sawtooth period is
dimensional space effects on sawtooth magnetic reconnection shorter with the former since the crash triggering condition
may be the way to resolve this long-standing theoretical issue. is reached more quickly after the partial reconnection. The
There is another serious concern associated with the so- simulated sawtooth period is found to be in good agreement
called trigger problem, namely, the non-linear development with many experiments in which the actuator parameters are
of the internal kink mode once the linear stability threshold varied once the free parameter(s) of the model is(are) fixed—
is crossed. Experimentally, the mode is often observed to for example, by matching the measured sawtooth period during

S132
 Chapter 3: MHD stability, operational limits and disruptions

the purely Ohmic heating phase. Simulations can also separate [14] to be strongly destabilizing as observed on JT-60U [23].
the effects due to heating alone or current drive alone to (Detailed modelling would need to be carried out on ASDEX
elucidate complimentary or competing effects [14] which are Upgrade and JT-60U to ensure that the predictions, which were
especially important in the complicated ion cyclotron current made for TCV, are relevant in the experimental conditions of
drive (ICCD) case [16, 13]. these machines.) Considerably stronger ‘partial’ stabilization
In experiments, for localized off-axis heating, the outside q = 1 and destabilization inside the q = 1 surface,
deposition location is generally swept across the q = 1 with current drive (compared with ECH aiming), has also been
region. In the simulations, it is difficult to predict the reported on DIII-D [24]. In all experiments and simulations,
optimum actuator settings for stabilization with sufficient strong changes in sawtooth period are found for very small
accuracy. This is due, first, to the combined uncertainties changes in deposition location (e.g. of the order of the EC
in the control parameters (e.g. launcher aiming, ray-tracing, deposition width consisting of a few centimetres) near the
etc) and equilibrium quantities (plasma position, q profile, q = 1 surface.
etc) and second, to the strong sensitivity of the stabilization Finally, simulations show that the stabilization process
on the deposition location. However, while it may not be can be non-linear in power density; displaying the threshold
possible to predict exactly how the actuators must be set for behaviour observed in early stabilization experiments, e.g.
stabilization prior to the experiments, it is important to note that [25]. Hanada et al showed experimentally that the threshold
simulations appear to be sufficiently advanced to accurately PEC /P
above which full stabilization occurs depends on the
predict whether a setting exists for which the desired control electron density and heating location (here PEC = the applied
is possible: the experimental deposition sweeps are employed ECRH power and P
= the Ohmic power). The threshold
to determine the optimum settings. was found to be lowest for deposition near q = 1 from the
The sweeps are performed using the magnetic field, high field side (HFS) in low density discharges. It is actually
plasma position, or launcher mirrors—all sweeps are typically the local shear that is changed so the relevant modification to
of small amplitude. The possibility of using mirrors is one of consider is that of the local current density with heating power.
the main differences between ECRH and other heating/current To compare sawtooth control between machines having
drive methods (although antenna phasing might provide some vastly different baseline sawtooth periods without reference
flexibility during ICRF). Control of the launcher mirrors to simulations, the sawtooth period during the EC heated
provides an external actuator in a feedback-control loop. For ‘controlled’ phase can be normalized to that in the non-EC
this reason we begin the discussion with EC sawtooth control heated ‘non-controlled’ phase: τsN ≡ τsEC /τs [17]. With
then continue with NBI and ICRF. fixed actuators and very careful placement of the deposition
location the sawtooth instability can be avoided during the
ECRH/ECCD. During ECH, the change in temperature and entire gyrotron pulse length in WT-3 (0.03 s) [25], TEXTOR
thus conductivity leads to a change in local current density (0.2 s) [26], JT-60U (1.5 s) [27] and ASDEX Upgrade (2.0 s)
and magnetic shear (s = (r/q)dq/dr). With ECCD, both [28], corresponding to an increase in τsN to
75,
11,
6
heating and current drive occur simultaneously. For deposition and
20, respectively. Only WT-3 specifies that the power
sweeps near the q = 1 surface both must be taken into account threshold has been reached in these experiments. In TCV
to properly simulate changes in shear and critical shear (for with ECH deposition just outside the HFS q = 1 surface
instability) and thus in the sawtooth period, e.g. [12]. The EC at power levels above the power threshold, the normalized
driven current is unipolar, and the current and power deposition sawtooth period reached values of τsN ∼ 50 (∼110 ms) [29]
profiles are generally nearly aligned to each other. corresponding to ∼50 electron energy confinement times and
ECH just outside the q = 1 surface stabilizes sawteeth, roughly one current redistribution time [14]. Here, however,
illustrating the first method of sawtooth control (stabilization). a sawtooth crash occurred before the end of the 2 s gyrotron
The location of optimal destabilization with ECH is predicted pulse, as the current profile evolved and the q = 1 location
by simulations and confirmed by experiments [14, 17], to be moved relative to the deposition location. Similar evolution
close to but inside the q = 1 surface. This demonstrates the effects may be at play in JT-60U in which a long period
second method of control. sawtooth is followed by sawtooth periods which scale as
3/2
Experiments show that even small amounts (<1% of the expected with Te [30]. These results indicate that, as in
plasma current distributed over 10% of the minor radius in NTM stabilization experiments, feedback-controlled launcher
TCV) of co-ECCD just outside the q = 1 surface enhances the angles and real-time calculations of the q profile are needed to
stabilization effect while counter-ECCD diminishes the effect ensure proper deposition for the entire pulse, if this method of
[12, 18–21]. Simulations also predict that counter-ECCD close sawtooth crash avoidance is to be used for ITER.
to but inside the q = 1 surface should stabilize sawteeth. The success of the sawtooth period modelling on TCV,
This has been confirmed on ASDEX Upgrade [20, 22] and JET, FTU, etc gives confidence that differences between
perhaps on JT-60U. On ASDEX Upgrade, moving counter- machines can be adequately taken into account in simulations
ECCD further inside q = 1 leads to complete elimination to be able to give reliable predictions of the sawtooth control
of the sawtooth crashes; however, from the changed MHD possibilities of the ITER EC actuators, provided that the
signature, it can be concluded that the stabilizing mechanism sensitivity to the model free parameters is not too high
is different than that due to off-axis co-ECCD. On JT-60U, in the ITER plasmas and that the heating and especially
3/2
the sawtooth period matches well the expected Te scaling driven current are correctly calculated as inputs to the model.
[23]: the sawtooth period is slightly longer than in the non- For example, WT-3 reports that the power threshold for
EC heated plasma. Co-ECCD just inside q = 1 is predicted optimum stabilization with low-field-side (LFS) deposition

S133
T.C. Hender et al

was 1.7 times larger than on the HFS, and the ITER EC design [4,39] to lengthen (shorten) the sawtooth period by heating near
does not allow for HFS q = 1 heating (where most present- the q = 1 surface with an antenna phasing of +90◦ (−90◦ );
day stabilization experiments are made). The difference in taking advantage of the ICRF-induced pinch of the resonating
HFS and LFS stabilization was presumably due to electron trapped ions to increase (decrease) the fast particle pressure
trapping effects which can be taken into account by present- inside the q = 1 surface and thereby alter the stabilization of
day ECCD modelling codes [31]. Note that it is crucial that the sawtooth instability [40].
if multiple EC beams are used, they must overlap precisely to Recently, the sawtooth period has been successfully
benefit fully from the higher total power density. controlled in JET using the ITER-relevant 2nd harmonic
Using the second control scenario (destabilization), τsN minority ion frequency scheme. As in many EC experiments,
can be reduced to between 0.4 and 0.5 when heating (TEXTOR, a sweep of the ICRH deposition near the q = 1 surface
ASDEX Upgrade, TCV, JT-60U) or co-ECCD (ASDEX was made varying the toroidal field. Both ±90˚ phasing and
Upgrade, TCV) is done inside but near the q = 1 radius. both HFS and LFS locations were studied during a minority
JT-60U has demonstrated a reduction to 1/6 the NB heated density scan. An NBI ramp was performed at fixed toroidal
sawtooth period [23] with co-ECCD just inside q = 1. This field and plasma current leading to additional effects for HFS
method is likely to be the most robust (as well as the most deposition as the NBI deuterons are resonant with the ICRF
efficient) destabilization [14]) EC sawtooth control scenario, waves near the plasma centre [13]. With ICCD alone, the
as it is less sensitive to the deposition location. driven current (protons) is of the diamagnetic type caused by
the finite orbit widths of trapped resonating ions and results
NBI. NBI can also significantly affect sawtooth behaviour. in a bipolar current perturbation as well as collisional electron
Trapped and/or co-circulating-passing NBI-generated fast heating. Two maxima in the sawtooth period are observed as
particles are expected to provide the stabilization, as these the resonance location passes through the LFS q = 1 inversion
populations typically provide an energy sink, or additional radius, with a further maximum as the resonance approaches
potential well. On JET, dedicated experiments were carried out the plasma centre [41]. Simulations [16] indicate that, first,
to investigate the sawtooth period increase during central NBI electron heating well outside q = 1 stabilizes the sawteeth,
heating [32]. The discharges were in the weakly collisional next, a more equal distribution of power near the q = 1 surface
ion-kinetic regime and the time of the crash at high NBI
reduces the stabilization and, finally, the combined effects
power was determined by the condition −cρ ρ̂ < −δ Ŵ (see
of co- (counter-) ICCD outside (inside) q = 1 are together
section 2.1.3 for details on this notation). The same scenario is
stabilizing enough to again increase the sawtooth period: that
considered most likely in ITER [6]. Detailed sawtooth period
is, modifications in local shear dominate in the crash trigger
simulations [16] provide a good match to the experimental
as in the case of ECH/CD. As the resonance moves further
sawtooth periods. The fast particle contribution to δW is
towards the magnetic axis, the increase in fast ion pressure,
inversely proportional to s1 (the shear at q = 1), so methods to
which modifies δW , appears to be the stabilization mechanism,
alter the evolution of s1 , in order to alter the sawtooth period, as
as with NBI. Good sawtooth control is found on the LFS, nearly
described for ECH, can also be used in the case of fast particle
independent of antenna phasing (±90◦ ; though still due to
stabilized sawteeth (see ICCD section); [33].
ICCD from trapped resonating ions) by tuning the resonance
Both the current profile and/or the pressure profile can be
location. This results in a modification of the local shear at
modified by NBI as noted by JT-60U and ASDEX Upgrade.
q =1 which can be used to delay the triggering of the 3/2 NTM
Negative ion NBI in JT-60U produces a fast co-passing
mode during power ramp-up experiments.
population that extends the sawtooth period [34–36] and has
been shown theoretically to be stabilizing [37]; whereas, In JT-60U detailed measurements of Alfvén eigenmode
counter-circulating ions are predicted to be destabilizing. frequencies during second harmonic minority heating indicate
ASDEX Upgrade found that the sawtooth period varied as a that q(0) < 1 during the entire sawtooth period. An alternative
function of the different individual NB injectors being used full reconnection model of Kolesnichenko et al [42] has been
(8 are available). In particular, each of the two tangential beams used to account for this observation. This model has not
deposit fast ions at about midradius and, except near the plasma yet been implemented in a code to track the evolution of the
edge, these ions are co-circulating. These beams are found to sawtooth period and to study the control of sawteeth.
be stabilizing [28]. Combining several beams, the sawtooth Most sawtooth period modelling uses a 1D transport
period is not only increased but fishbones are eliminated and model; details of the sawtooth crash process are lost.
are largely replaced by an m = n = 1 mode. The mode has two Additional 2D effects, observed in some experiments, such
frequencies and two radial locations suggesting the existence as stable island rotation during tangential NBI or near central
of two q = 1 surfaces in the plasma [38] and consequently ECH/ECCD are also not possible with the 1D modelling.
a modification of the current profile. Recent theoretical work Recent simulations [43 and references therein] recover many
involving asymmetric circulating ions suggests that on-axis detailed 2D effects seen in experiments, which can be due to
counter-injection, or off-axis co-injection, of neutral beams can incomplete reconnection, by introducing an ad hoc m = n = 1
destabilise sawteeth, and thus offset the stabilising properties island displacement function; however, at present no model
of trapped alpha particles in ITER [37]. exists to predict the evolution of the island displacement.
Nevertheless, the success of simple 1D models in predicting
ICRH/ICCD. In the ion cyclotron range of frequencies sawtooth control in present-day experiments, gives confidence
(ICRF), resonant heating (ICRH) has been used to control that sufficient physics is contained in the model for predictions
the sawtooth period on JET and JT-60U. ICRH has been used of the control capabilities of actuators in the ITER device.

S134
 Chapter 3: MHD stability, operational limits and disruptions

2.1.3. Central MHD activity expected in ITER. As discussed unity, the value of the toroidal flux at the inversion surface, 1 ,
in the preceding section, there are now improved tools for is calculated from the integral equation:
modelling sawteeth, which have been better benchmarked with
1 
existing experiments. These have been used to evaluate the 1
− 1 d = 0. (4)
effects of sawteeth in the present ITER design. 0 q( )
A comprehensive model of the sawtooth trigger and For the duration of the sawtooth crash time τCRASH , the
relaxation oscillation has been developed by Porcelli et al [6]. thermal conductivity and the hyper-resistivity in the central
We adopt the notation used in that paper and examine the region < 1 are defined to be: χ = r12 /τCRASH and
consequences of this model for ITER. An effective potential λ = λ0 B0 r1 /τCRASH . A value of λ0 = 0.1 effectively
2 4
energy functional, associated with the macroscopic drive, can causes a Kadomtsev reconnection to occur [46] in the time
be defined by t = τCRASH , which is taken to be 10 ms. This value
δW = δWcore + δWfast . (1) was chosen as a compromise to minimize numerical time-
Here δWcore = δWMHD +δWKO , δWMHD is the ideal MHD term resolution requirements. Even though the actual experimental
containing destabilizing terms due both to plasma pressure and value may be considerably less than this, we do not expect this
parallel current density [44], δWKO is the ‘Kruskal–Oberman’ choice to affect the global result since it is much less than the
term [45] which represents the (normally stabilizing) effect energy confinement time, τCRASH  τE . By lowering λ0 to
of collisionless thermal trapped ions, and δWfast represents the 0.001, an incomplete reconnection can be modelled where the
kinetic effects related to the high energy fusion-produced alpha temperature profile flattens but the current and flux do not fully
particles. reconnect.
It has previously been reported [15] that the Porcelli
Reference [6] gives three conditions, any of which means
sawtooth model described here has been implemented in the
triggering the sawtooth crash if it is met. These conditions,
PRETOR code and compared in detail with JET experiments
which we call the Porcelli criteria, are
in several different regimes with different levels of NBI power.
− δ Ŵcore > ch ωDh τA , It was reported that in every case analysed, the simulated
sawtooth periods are within 20% of the experimentally
− δ Ŵ > 0.5ω∗i τA , observed periods, even as the period varies by more than a
factor of 5 during a given shot [15]. This same Porcelli
− cρ ρ̂ < −δ Ŵ < 0.5ω∗i τA and ω∗i < c∗ γρ . (2)
sawtooth model has been implemented in the Tokamak
Here ch and cρ and c∗ are numerical factors of order unity, Simulation Code (TSC) [47] and its consequences on transport
ρ̂ = ρi /r̄1 with r̄1 being the average radius of the q = 1 and ignition in ITER have been investigated. In order to
surface, ρi = vthi /
ci is the thermal ion Larmor radius, ω∗i is predict the profiles and performance, two of the leading H-
the ion diamagnetic frequency, γρ is the characteristic growth mode models are implemented in TSC: (A) the multi-mode
rate of the internal kink mode in the ion-kinetic regime, ωDh model [48], and (B) the Gyro-Landau fluid model GLF23 [49].
is the precessional drift frequency of the high energy particles, These core transport models must be supplemented by
and the potential energy is normalized here according to boundary and edge models.
The H-mode models (A) and (B) are only applied in the
4δW central region 0 < < 0.75, where is the normalized
δ Ŵ ≡ − (3) toroidal magnetic flux that is zero at the magnetic axis and
s1 ξ 2 ε12 RB 2
unity at the plasma/vacuum separatrix. In the edge region
with s being the magnetic shear, ξ the radial displacement of 0.75 < < 1.0, a transport model χI = χe = C/ne is
the magnetic axis, ε the inverse aspect ratio and the subscript 1 used, where ne is the local electron density and C is a constant
denotes values at the q = 1 surface. chosen so as to make the pressure gradient in this region below
The first of these criteria (2) is that the high-energy trapped the infinite-n ballooning mode stability criteria. This leads
particles complete many precessional drift orbits within a to electron and ion temperatures at the top of the pedestal,
characteristic magnetic perturbation time, which is of order = 0.75, of 3–5 KeV. The density profile is not advanced in
−1
|δ Ŵcore |τA , with τA being the Alfvén time. The second is time in these simulations, but is rather a prescribed function of
that the diamagnetic rotation is not sufficient to stabilize the normalized poloidal flux, ψ, and time, t. The electron density
mode. The third criterion is for the case when the energy during the current flattop is taken to be
drive is sufficiently weak that the mode is stable according to
ne (ψ, t) = n0 (t) × [(1 − ψ β )α + redge ], (5)
the first two criteria. In this case, the m = 1 mode structure
changes its nature from a global internal kink to a drift-tearing with parameters α = 0.25, β = 8.00, n0 = 0.75 × 1020 m−3
mode localized near the q = 1 surface. This is normally and redge = 0.4 chosen to correspond to a reference ITER
stable because of kinetic layer effects, but the layer effects will ELMy H-mode case.
be insufficient when the normalized potential energy exceeds The Gyro-Landau fluid model transport GLF23 is used.
the normalized ion Larmor radius and rotation effects are A feedback system on the ICRH power designed to keep the
sufficiently weak (note that the choice of normalizations has total stored energy W constant at 320 MJ is utilized. A uniform
made this comparison appropriate). distribution of 2% beryllium and 0.12% argon impurities is
When the sawtooth is predicted to be triggered by the included, which together with the He build-up (assuming
Porcelli criteria, the transport coefficients are modified in two τP = 18.5 s), leads to a value of ZEFF ∼ 1.65 during the flattop.
ways. Assuming that the central safety factor q0 is less than The constant in the edge region is chosen as C = 2.5 × 1019 .

S135
T.C. Hender et al

Pulse No: 114494

(a) Pulse No: 114504
1.0 7000
q-axis

0.8
PINJ (KW)
0
0.6 (b)
300 3

200 Reduction in β Due to

βN due to 3/2 NTM 3/2 & 2/1 NTMS
100 0
(c)
20
0 n=2 MIRNOV (G)
20000 10
Te (0)

15000
0
10000 (d)
20
5000 n=1 MIRNOV (G)
0
100 200 300 200 300 10
Time (s) Time (s) 0
1500 2000 2500 3000 3500 4000 4500 5000
Figure 1. ITER simulations using the GLF23 transport model and Time (ms)
both the complete (left) and incomplete (right) reconnection models. Locks
The top frames show the axial q value, in the middle frames the rf on (114494 only)
α-power is the solid line and stored energy the broken line and the (114504 only) 2/1 NTM onset
bottom frames show the axial temperature.
Figure 2. DIII–D discharges with (114504, dotted lines) and
without (114494, solid lines) ECCD suppression of an m/n = 3/2
The results of two simulations are shown in figure 1, one neoclassical tearing mode. (a) Neutral beam power, (b) βN , (c)
using complete reconnection and the other with incomplete n = 2 Mirnov |B̃θ |, (d) n = 1 Mirnov |B̃θ |. The degradation in
reconnection. Note that the sawtooth period is about 50 s energy confinement due to the NTM from 3/2 and 2/1 NTMs can be
seen in the effect on βN .
during the flattop for the complete reconnection, and 2–3 times
that frequent for the incomplete reconnection. Both the alpha
power and stored energy are essentially independent of the island, large enough to increase the local radial transport and
sawtooth period, since the period is always longer than the flatten the pressure profile. The most significant NTMs are
energy confinement time, the core electron temperature is so those with m/n = 3/2 or 2/1 (with m the poloidal mode
high, and the magnetic diffusion time is long compared with number and n the toroidal mode number). The effect of
the energy confinement time. The q = 1 radius is about 42% these NTMs on energy confinement is nicely illustrated by
of the minor radius. comparison of two discharges one of which suffers from a 3/2
It is concluded that in ITER the sawtooth will lead to and then a 2/1 NTM, and an otherwise identical discharge in
periodic oscillations on a time that is considerably longer which electron cyclotron current drive (ECCD) stabilizes the
than the energy confinement time, τSAW  τE , and that the 3/2 NTM and a 2/1 NTM does not occur (see section 2.2.2 for
temperature at the q = 1 surface is sufficiently high that a discussion of NTM stabilization), as shown in figure 2.
the sawteeth oscillations have negligible effect on both the Since the previous report [1], the studies have concentrated
stored energy and the rate of neutron production (as can be either on the capabilities to predict the onset of NTMs
seen from figure 1). The incomplete reconnection sawteeth in ITER and on the possibilities to stabilize the modes if
have a period about half that of the complete reconnection, they are triggered. Scalings at the mode onset and decay
but still long compared with τE . A consequence of these long including both collisionality and Larmor radius have been
period sawteeth is the possibility of destabilizing NTMs (see extensively investigated, e.g. in ASDEX Upgrade [50, 51],
section 2.2.3) and thus sawtooth control (see section 2.1.2) is DIII-D [52], JET [53,54], JT-60U [55] and T-10 [56], leading to
important to consider. Also, heat pulses from sawteeth can be a consensus on the stronger, approximately linear, dependence
linked to ELM triggering. on Larmor radius compared with collisionality. Cross-machine
comparisons of onset conditions have led to a scaling for the
2.2. Neoclassical tearing modes onset beta of the m = 3/n = 2 NTM depending on νi∗
and mainly ρi∗ [57]. The difficulty of such a scaling is that
2.2.1. Physics of neoclassical tearing modes. The presence it needs to combine the seed island formation physics and
of persistent magnetic islands in the plasma core is an important the NTM physics. A particular assumption made in [57] is
issue for burning plasmas, as they can significantly limit the that the seed island width can be described as a function of
performance of both the standard ELMy H-mode and advanced βp and 1/S = τA /τR , the inverse of the magnetic Reynolds
scenarios. The neoclassical tearing modes (NTMs) are driven number. Recent JET experiments have shown however that
by the local reduction of the bootstrap current due to the large seed islands and hence NTMs, can be triggered at the
pressure flattening across the island. This drive is inherently a sawtooth crash after long sawtooth free periods [4], even at
non-linear process as it relies on the existence of a fully formed low β. Therefore more recent studies have concentrated on

S136
 Chapter 3: MHD stability, operational limits and disruptions

15 the marginal beta limit above which NTMs are metastable, is

βp onset
very low but generally the sawteeth (or other seeds) do not
10 form a large enough island (W < Wcrit ) until βp increases well
above βp,marg . Since βp,marg scales approximately linearly with
5
ρ ∗ [54] ITER is predicted to have βp > βp,marg as soon as it
dw/dt [cm/s]

0 is in the H-mode. Therefore the existence of NTMs in ITER

βp marg does not depend on β as such, but rather on the triggering of a
–5 seed island Wseed > Wcrit . Thus, the predictions of seed island
widths and of the value of Wmarg are each of great importance
–10
for burning plasmas.
–15 The prediction of Wmarg indicates that its size normalized
0 w w 8 w 16
crit marg sat by the minor radius will be much smaller in ITER than in
w [cm]
present experiments. Its value depends on all the terms
Figure 3. Sketch of the time evolution of the island growth rate in equation (6) and their dependence at small W . The
as given by equation (6) at the onset of the NTM when the understanding and relevance of each of these terms have been
critical seed island size (Wcrit ) is exceeded and an NTM forms further developed since [1]:
at βp,onset . A slow decrease in beta from βp,onset to βp,marg (when
max(dW/dt) = 0) is assumed, as in power ramp-down experiments, – The first term is the classical  term, which has a weak
such that dW/dt ≈ 0 (reproduced from [54] ‘Marginal β-limit for dependence on W . It has been shown that classical
neoclassical tearing modes in JET H-mode discharges’). tearing modes can provide the seed islands for NTMs
[63, 64], and this may be one of the possible explanations
the seed island formation or on the NTM physics, as they are of the ‘triggerless NTMs’ observed in other machines
not necessarily related. However, additional effects on NTM like ASDEX Upgrade [65], JT-60U [55], T-10 [56] and
onset arise from resonant error fields which can seed NTMs TFTR [66]. This usually happens when the current
and slow plasma rotation [58]. Preliminary analysis suggests profile is modified [63], and could become the main seed
decreased plasma rotation could be reducing the small island island trigger mechanism in hybrid scenarios, or when
polarization current threshold and thus making the 2/1 NTM β approaches the ideal limit, as  can become large and
unstable at lower β, but further experiments are required to positive [64] and thus could become important in advanced
elucidate this effect. scenarios. In addition, using fast power shut-off, leading
To understand the physics mechanisms at play, it is best to a rapid vanishing of the terms proportional to βp in
to describe in some detail the modified Rutherford equation, equation (6), it was possible to show the linear  (W ) on
which can be written symbolically as follows for the island W in TCV [17].
growth rate: – The second term is the bootstrap drive, which is reduced
τR dW at small island width due to two main effects. First,
= rs  (W ) + rs βp (BS − GGJ − pol ) + rs CD . the ratio of perpendicular to parallel heating becomes
rs dt
non-negligible and the pressure profile is not flattened
(6)
completely, reducing the perturbed bootstrap current [61,
Here W is the width of a magnetic island occurring at a radius 67]. Anomalous perpendicular viscosity can also affect
rs and τR is the local resistive diffusion time;  is the stability the bootstrap drive. Its effect is frequency dependent and
index of the equilibrium current profile, BS is the bootstrap can be stabilizing or destabilizing depending on the sign
drive term, and  CGJ and pol are the stabilizing curvature [59] of ω/ω∗pi [68], where ω is the mode frequency in the
and polarization terms [60], respectively. The effect of current electron frame and ω∗pi is the ion diamagnetic frequency.
drive represented by CD will be discussed in the next section. Another effect which reduces the perturbed ion bootstrap
The island width dependence is BS ∝ W/(W 2 + Wd2 ) and current even more is finite ion Larmor radius effects [69].
pol ∝ Wpol2
/W 3 , where Wd describes a stabilizing effect at When the island width is less than ∼5ρb (ρb = ion banana
small island width due to perpendicular thermal conduction width) ions are still affected by the pressure gradient inside
[61] and Wpol is a constant related to the stabilizing polarization and outside the island, leading to a finite bootstrap current
effect. A fuller description of these terms, used to compare within the island.
with experimental data, can be found in [54,62] and references – The third term describes the stabilization due to the effect
therein (see also [1]). The typical evolution of the island growth of curvature and is usually smaller than the bootstrap term
rate in a full discharge, assuming a slow ramp-down of the in present tokamak scenarios and therefore has often been
power, and thus a slow decrease in the terms proportional to neglected in the past. It has been confirmed in MAST to
βp in equation (5), is shown in figure 3. At a given time in the be significant for tight aspect ratio scenarios [70]. On the
discharge, an island is triggered at a beta value βp,onset , in most other hand, it has been shown to yield a finite stabilizing
cases much larger than βp,marg , and subsequently grows to a term for small island width [71] and therefore can be
relatively large saturated island width. When βp  βp,marg , the significant at small island widths in present tokamaks and
mode is stabilized and the growth rate becomes rapidly very for ITER aspect ratio as well.
negative. The hysteresis, ratio βp,onset /βp,marg , is significant – The fourth term is due to the polarization current,
in standard scenarios with modest size sawteeth, it has been resulting from the fluctuating electric field driven by the
measured much above unity in ASDEX Upgrade [51], DIII- different electron and ion responses to the rotating island.
D [52], JET [4] and JT-60U [55]. This occurs because βp,marg , Therefore it involves diamagnetic effects, effective mode

S137
T.C. Hender et al

frequency and ion and electron kinetic effects as the island cross-machine scalings are possible, in particular as these
width of interest is of the order of the ion orbit width. measurements of Wmarg do not depend on the seed island
Excitation of drift waves has recently been included as formation mechanism. Such studies have been started [58]
well [72] and the effects of the ion sound [73]. It has on ASDEX Upgrade [51], DIII-D, JET [54] and JT-60U, but
also been noted that in toroidal geometry, the variations much more work is required in order to extend the database
of plasma pressure within the perturbed magnetic surfaces to a larger range in ρ∗ and collisionalities. Another domain
(due to ion sound waves and/or finite parallel heat for which further physics understanding should be developed
conductivity) lead to a new type of frequency dependent is the effects of fast particles on NTMs and their interaction
polarization terms that are important for determining the with large islands [87]. In summary, the NTMs are expected
threshold value of the seed island [74]. The complexity to be metastable in ITER, i.e. β > βmarg , however we are
of determining this contribution has been highlighted in unable to predict with certainty if large enough seed islands
the recent review [62]. The polarization current term can will be triggered and destabilize these modes. Therefore it
change sign, depending on the effective mode frequency, remains prudent to have a means to stabilize them as discussed
which is very difficult to measure. It is observed to be in sections 2.2.2–2.2.4.
in the ion drift direction in most cases in DIII-D [75]
and JT-60U [76]. A common feature however of all the 2.2.2. Active control of NTMs.
different models, is that the polarization current rapidly
decreases with increasing island size and it scales with Status at time of IPB. In the ITER Physics Basis, the
ρ∗. Therefore the polarization current effect could be possibility of NTM stabilization was only briefly described in
less important in burning plasmas if Wmarg is of the the ECCD chapter. No experimental results were available
order of a few poloidal Larmor radius. However present at that time. This subsection describes direct stabilization
understanding of experimental results require a better of the NTM. Other methods that work on stability of other
quantitative theoretical knowledge of this term. For modes that influence the NTM (sawteeth to suppress seed
example it has been shown recently that finite ion orbit island and (m + 1, n + 1) to trigger FIR regime) are discussed
widths also strongly affect this term [77]. in section 2.2.3. Here, theory and experimental verification as
The prediction of the seed island formation is much more well as elements of a feedback-control scheme are considered.
complicated, also because it can have different origins for The extrapolation to ITER is given in section 2.2.4.
different plasma scenarios like non-linear mode coupling [78],
link to the current profile of 1/1 activity [53], fishbones [65], Theory. NTMs may be stabilized by local CD at the resonant
turbulence [79], error fields [58, 80] in addition to the causes surface q = m/n due to two effects: The equilibrium (i.e. (0,0))
mentioned above. Further changes in the rotation profile component of the driven current alters  , while the (m, n)
between the location of the source of the seed island and component and its multiples directly counteract the reduction
the NTM resonant surface can also affect the onset β— of bootstrap current within the island due to the flattening of the
experiments in JET suggest that decreased rotation reduces pressure gradient there. This has been studied with a variety
the NTM onset β-limit [81]. For ITER, which will have of theoretical approaches including non-linear resistive MHD
sawteeth stabilized by α-particles, the main perturbation will codes or analysis based on the Rutherford equation with a CD-
most probably be at the sawtooth crash, as is the case in high term (see e.g. [88]). The basic physics results are in agreement,
field/current scenarios in JET [4]. The seed island can be so that here, for the sake of simplicity, the Rutherford equation
formed by the direct non-linear coupling to the 1/1 mode [82]. is discussed (equation (6) of section 2.2.1). The CD-term in
It should be mentioned that 2/1 modes can be directly triggered this equation can be written as
at the sawtooth crash at low beta in JET [83]. Therefore better  
 ICD(m,n) 1 ICD(0,0) 1
understanding is clearly needed in this domain, together with CD = −Lq am,n ηCD + a0,0 ,
Ip (rs ) W 2 2
Ip (rs ) δec
sawtooth control experiments (see section 2.1 and 2.2.3).
(7)
The effects of increased shaping, elongation and
triangularity, are in general weak for NTMs but they can where ICD(m,n) is the total helical current of same helicity as
affect different terms. Therefore they can manifest themselves the island, ICD(0,0) is the (0,0) component of the externally
differently in local and global parameters [84]. Since the driven current changing the equilibrium stability index  in
global confinement properties are modified, both the current equation (6) of section 2.2.1, and Lq = q/(dq/dr). The term
and pressure profiles are affected and therefore also the  (W ) with ηCD takes into account the conversion of the externally
and BS terms. Shaping also directly modifies the curvature driven current into a helical current as the current generated
term, GGJ , but it is expected to be significant only for tight locally rapidly distributes on the island flux surfaces. For a
aspect ratio tokamaks in conjunction with small magnetic δ-function in the O-point (i.e. the extreme limit of modulated
shear [85, 86]. Both elongation and triangularity can also ECCD), ηCD = 1. Due to the redistribution of current on
be important in the seed island trigger mechanisms, as they the island flux surfaces, a source that is constant and does not
influence the sawtooth stability (see section 2.1) and the depend on the helical angle may still create a helical current
coupling between the sawteeth and the m/n flux surface. if it is deposited in a radial region smaller than the island
The power ramp-down experiments, as illustrated in width. Conversely, if the deposition width δec is much bigger
figure 3, are important since they can provide accurate than W , a uniform source only creates a constant current that
measurements with NTM island widths of the order of Wmarg , does not have a helical component. However, if the power is
where all the stabilizing terms are important. Therefore modulated, a helical current is created even if δec  W . This

S138
 Chapter 3: MHD stability, operational limits and disruptions

10
NBI Power (MW) ECRH Power (MW*10) #14989 5
n=2 amplitude( a.u)
0

Toroidal Field -2.15

-2.25
3
Normalised Beta 1
1.5
Neutron Rate (1e15/s) 0.5

H-Factor ITER98(p,y) 1
0.8
1.6 2 2.4 2.8 3.2 3.6 4 4.4
time (s)

Figure 4. Complete stabilization of a (3,2) NTM in ASDEX Upgrade. βN can be increased well above the onset level without reappearance
of the mode. At even higher βN , the mode comes back due to a deposition mismatch induced by the Shafranov shift.

effect can usually not be seen in present-day experiments since mode. Usually, if ECCD is switched off, the next sawtooth
the width Wmarg,CD at which dW/dt becomes negative during or fishbone usually triggers the mode again. Figure 4 shows
stabilization is comparable to δec . In ITER, however, this may an example from ASDEX Upgrade (an example from DIII-D
be different and it may be required to modulate the ECCD is shown in figure 2).
power in phase with the island O-point for complete mode In the experiment shown in figure 4, the mode ultimately
stabilization. The requirement for good localization leads comes back due to the Shafranov shift at higher β which
2 moves the resonant surface away from the ECCD deposition.
to the figure of merit (ICD /Ip (rs ))(rs Lq /δec ) for equilibrium
current profile modification as well as for helical current This clearly calls for a feedback control of the deposition (see
generation if δec > Wmarg,CD and the power is injected below). In some experiments, β did not recover to the previous
continuously. For helical current generation in the case value and NBI power had to be increased to recover the β. This
δec > Wmarg,CD , but with modulation, (ICD /Ip (rs ))(Lq /δec ) is due to a confinement reduction when using ECCD that is a
is the relevant figure of merit. Since the current localization is generic feature of strong electron heating in scenarios with
only good if the current is deposited in the island, the source Ti > Te (see chapter 2 of this issue [95]).
also has to be localized precisely in radial direction. Theory Several experiments have proven the need for exact
suggests that if the mismatch is of the order of the island width, localization of the ECCD by sweeping ECCD over the resonant
stabilization becomes ineffective [89]. surface using a slow Bt -ramp or antenna mirror steering. It
One goal of the theoretical description is the prediction of is found that ECCD is only effective in reducing the mode
the power level required for complete stabilization of a given amplitude when it is deposited close to the resonant surface;
mode. Here, the main uncertainties lie in the terms describing if deposition is outside the island, no stabilization occurs [88].
the mode dynamics rather than in the CD part, which seems to Experiments were also performed on ASDEX Upgrade with
agree quite well with the experiment. Thus, we have to improve variation of the toroidal injection angle. For perpendicular
our predictive capability for the NTM stability itself to improve injection, only a weak stabilizing effect is found at a power
the predictions of the required power for stabilization as given level that leads to full stabilization for optimum toroidal
in section 2.2.4. angle [96]. This proves the important role of the direct CD as
opposed to the current generated by local heating. As pointed
Experimental results. Due to the requirements for localiza- 2
out above, theory predicts that ICD /δec or ICD /δec is the figure
tion, ECCD is clearly the best method for NTM control, but of merit for stabilization. This means that although the driven
LHCD has also been used to modify the global current pro- current increases monotonically with toroidal injection angle
file, thus altering NTM stability mainly through a change in φ, the largest toroidal injection angle is not necessarily the best
 [90]. In the experiments, a local current is generated by the for stabilization, because δec also increases with it. Usually,
waves, but also the local heating within the island may lead to a IECCD increases more weakly than linearly with φ whereas
local change in resistivity and therefore generate an additional δec increases more strongly than linear, so that an optimum
helical current. angle exists. This was shown experimentally on ASDEX
Successful ECCD experiments have been carried out Upgrade where above a certain toroidal injection angle, the
in ASDEX Upgrade [91], JT-60U [92] and DIII-D [93, 94] mode could no longer be completely stabilized at the power
demonstrating complete stabilization of both the (3,2) and the level sufficient at smaller angle in spite of the higher driven
(2,1) NTM. The required power was of the order of 10–20% of current at larger φ [97].
the total heating power, with the (2,1) stabilization requiring Experimentally, it is difficult to sort out the stabilizing
more power than the (3,2) stabilization. The NTM remained contributions due to the (0,0) component of the EC current
stabilized even in the presence of sawteeth and fishbones, and due to the helical component of the EC current since both
which provide seed islands, and β could be increased by cannot directly be measured. The fact that the localization
20–30% above the onset beta without reappearance of the within the island is required to obtain good stabilization

S139
T.C. Hender et al

50 Another element of a future NTM control scheme is the

PCS optimization
starting from ∆R ≈ –2 cm
early detection of the mode: the power needed to reduce the
40
No PCS mode is predicted to be small at mode onset. Figure 6 shows an
n=2 Mirnov Amplitude (T/s)

optimization
Dwell experiment in JT-60U where ECCD was applied before mode
onset [102,103]. In this situation, the mode only grows to small
30
amplitude after triggering, whereas the same power applied to
a saturated mode could not reduce the mode amplitude to such
20 Search small values.
NTM
Dwell In summary, local ECCD has successfully been used to
restrikes
10 on a Search suppress NTMs and recover β-values at modest ECCD power
sawtooth level. With ECCD, stable operation without NTMs at β above
crash
PCS reset at 4500 ms the onset β-value without ECCD, has been demonstrated.
0
1.680 1.685 1.690 1.695 1.700 Although the qualitative physics picture of the process is
Rsurf (m) confirmed by the experiments, better theory and more detailed
Major Radius comparison between theory and experiment is needed to better
predict the power requirements for ITER in a quantitative
Figure 5. Search and suppress algorithm in DIII-D (Reprinted with
manner.
permission from [93]. Copyright 2002, American Institute of
Physics): The major radius is adjusted in steps by the plasma control
system (PCS) until complete stabilization occurs. After reset of the 2.2.3. Mitigation of NTMs.
position, the mode reappears at a sawtooth crash. A discharge with
no PCS optimization, but with ECCD applied at the optimum major Seed island control. The neoclassical tearing modes (NTMs)
radius found by the PCS, shows complete suppression
have a relatively low marginal β and are therefore metastable,
typically as soon as the discharge is in the H-mode, with
efficiency points towards the importance of the helical current βN,marg
0.5–1, for standard q95 ∼ 3 ELMy H-mode cases
for the ECCD schemes. A hint towards the importance of (section 2.2.1). This means that the key to avoiding the onset of
 is the observation that in scans of the deposition across NTMs is not necessarily the exact β value [104], as one usually
the resonant surface, a stabilizing effect is found for counter- approaches the β-limit set by MHD, but it is the control of the
ECCD deposited at a radius significantly smaller than the size of seed islands and thus of intrinsic plasma perturbations.
resonant surface [98]: while the helical current produced by If these are smaller than the marginal island size, no NTMs are
counter-ECCD would be destabilizing, the (0,0) component of triggered (see figure 3). The main source of perturbations is
the ECCD current in this case gives a stabilizing contribution
the sawtooth activity, either the precursor/postcursor activity or
to  . Modelling of ECCD experiments suggests that the two
the 1/1 mode responsible for the crash itself and its associated
effects may be of same order in ASDEX Upgrade. In LHCD
magnetic reconnection. It has been shown in ASDEX Upgrade
experiments on COMPASS-D, the deposition width was so
[105] and DIII-D [106] that by avoiding sawteeth, with q
1,
large compared with the island size that  changes were the
the β onset could be raised by typically 20%, at which point
only reasonable explanation for the observed stabilization [90].
NTMs are triggered by fishbones. On the other hand if the q
The importance of both contributions is of special interest for
profile is modified, as in q > 1 hybrid scenarios, because of
the extrapolation to ITER, where usually, the  term is not yet
the presence of impurities in the centre or for example due to
taken into account (see, e.g. the discussion in section 2.2.4).
ECCD current tailoring as shown in TCV, a seed island can be
Elements of a control scheme. The experimental results provided from a classical tearing mode with  > 0 [63].
discussed so far show that local ECCD is a good candidate for However, a recent key result with respect to predictions
an NTM control scheme in ITER. However, the high sensitivity for ITER shows that NTMs can be triggered at low beta, just
to the correct localization calls for a feedback-control scheme above βmarg , when very long sawtooth periods are sustained
of the deposition. This is required when changes in the using fast particles stabilization in JET [2, 37, 104]. As α-
equilibrium, e.g. due to the Shafranov shift variation with β, particles will strongly stabilize sawteeth (section 2.1.1), this
move the resonant surface with respect to the deposition. is an important issue for ITER. However it has been shown
Several experiments have explored the possibilities for that these stabilized sawteeth do not substantially deteriorate
such a control. In DIII-D, a so-called ‘search and suppress’ the overall plasma stability and control, but rather trigger 3/2
algorithm was used [93] (see figure 5): the deposition radius or even 2/1 NTMs at the sawtooth crash. An example from
was varied by a small step through either adjusting Bt or R0 . JET is shown in figure 7, where a relatively large island is
Based on the effect of this step on the mode, it was decided how triggered at the sawtooth crash directly with the width of the
the next step should be taken. In JT-60U, the mode position saturated island [4]. As these modes will grow in typically 30 s
was determined online from ECE measurements from the local on ITER, they mainly cause concern because they can lead to
minimum of mode amplitude (which occurs at the rational the performance degradation.
surface) and the poloidal steering angle of the ECCD launcher The triggering of a 3/2 or 2/1 seed island by a sawtooth
was adjusted to track the mode [23, 99]. More sophisticated crash at q = 1 can result from global current profile
ideas on mode detection have been proposed and in some cases modification [108] or from direct toroidal and non-linear
tested, e.g. on FTU using ECE detection [100, 101]. More coupling as obtained in 3D-MHD simulations [82]. The direct
experimental research is needed to clarify the applicability of coupling of the 1/1 mode, causing the crash with 3/2 and
feedback-controlled schemes to ITER. 2/1 surfaces has been confirmed in MAST experiments where

S140
 Chapter 3: MHD stability, operational limits and disruptions

Figure 6. Application of early ECCD (lefthand figure) in JT-60U leads to a much smaller mode amplitude than application of the same
ECCD power to a saturated mode (righthand figure) [102]. In pulse E41693 the EC injection angle is fixed so that the EC current is driven
around the q = 3/2 surface. In pulse E41650 the injection angle is controlled in real time by searching the magnetic island position.

JET Pulse No: 53285 higher β onset values have been obtained using ICCD to
0.6 destabilize sawteeth, albeit in scenarios without strong fast
n=1
particles. The possibility of destabilizing sawteeth which are
0.4 stabilized by fast particles, in scenarios similar to the ones
n=2 expected in burning plasmas, has been demonstrated recently
0.2 on JET [33]. This also confirms the possibility of decreasing
δWfast by increasing the shear at q = 1 (section 2.1.1).
0 Therefore in addition to direct control there is also the
6000 SXR possibility of using ECCD on ITER to control the sawteeth and
thus the seed islands triggering NTMs, as α-particle stabilized
4000 sawteeth are likely to be the main perturbation source. ECCD
can be used to either stabilize or destabilize the sawteeth,
2000 by either slowing down or accelerating the build-up of the
shear at q = 1 after a crash [14]. The best scenario would
15 16 17 18 19
be to shorten the sawtooth periods in ITER, as sawteeth are
Time (s)
helpful in avoiding impurity or He ash accumulation in the
6 centre. However at present it is not known how to predict
the level of destabilization required. Experiments are being
W (cm)

4 performed to test this issue in ASDEX Upgrade [20], DIII-D

and JET. Another possibility is to fully stabilize the sawteeth
2
JG02.42-1c

and thereby avoid NTMs onset, provided that fishbones or other

0
perturbations are not sufficiently large to trigger NTMs, as is
18.5 19.0 19.5 the case at high field and current in JET. However the relative
Time (s) size of the perturbations needed to trigger NTMs in ITER is
Figure 7. m = 3/n = 2 NTM triggered by a sawtooth crash being expected to be much smaller than that in JET. Nevertheless,
delayed (achieving a sawtooth period of 1.5 s) by the presence of it is possible that the peaked electron heating provided by the
ICRF fast particles at Ip = 3.3 MA, B = 3.3 T using minority H self-heating mechanism will be sufficient to avoid impurity and
heating in the centre with f = 51 MHz. The island is directly He ash accumulation in sawtooth free scenarios in ITER.
triggered at a saturated size of 6 cm. Note that there is no significant
n = 1 activity before the crash, and the relative SXR crash of about
15% is not much larger than the previous sawtooth crashes. Profile control and plasma shaping. According to the
(Reprinted figure with permission from [4]. Copyright 2002 by the relevant contributions for the drive of NTMs as described
American Physical Society.) in section 2.2.1, their growth and thus the resulting drop in
confinement can be influenced by tailoring the radial profiles
tight aspect ratio increases the coupling and NTMs have been of current, pressure or density. The influence of the current
triggered by sawteeth at low β [70]. This has also been used to and pressure profiles has been shown for example on JT-60U,
explain the lower onset β observed in low q95 discharges [54], where either the NTM’s rational surface was shifted inside
because of the larger q = 1 radius. Therefore one needs to (resulting in both smaller local current gradients and pressure
destabilize sawteeth in order to avoid the large seed islands and gradients [55]), or the pressure peaking was changed by using
this has been demonstrated on JET [4, 39], where significantly more central (negative NBI) or more off-axis peaked (positive

S141
T.C. Hender et al

0.4
E036715(w/ NNB)
E036706(w/ o NNB)
4 3 0.3

∆W
W
0.2
3
−∇P [105Pa/m]

2
0.1
2

q
1 0
1 1.5 2 2.5 3 3.5 4
1 β N,onset

#10560 #9800

√ B [a.u.] dB/dt [a.u.]

40
20
0 0 0 0
0.3 0.4 0.5 0.6 0.7 0.8
ρ

~
-40 -20
0.02
0.03 (a) (b)
Figure 8. Two discharges from JT-60U with essentially identical

~
safety factor (q) profiles measured by MSE. In the discharge with 0.0 0.0
NNB injection a more centrally peaked pressure profile results and 3.14 3.24 3.34 1.95 2.05 2.15
t [s] t [s]
so the pressure gradient (∇P ) is lower at q = 2 than in the discharge
without NNB injection. This allows discharge E036715 to achieve a Figure 9. Comparison of reduction in energy confinement
higher βN without a 2/1 NTM being destabilized (which does occur (W/W ) due to (3,2) NTMs on ASDEX Upgrade (open symbols)
in discharge E036706) [55]. and JET (full symbols). Very good agreement is seen, both in the
relative confinement degradation as well as in the βN value above
which FIR-NTMs cause less energy losses. The lower figure shows
NBI) heating [99]. Figure 8 shows an example from JT-60U
the NTM behaviour for two ASDEX Upgrade discharges at about
[55] where negative ion neutral beam (NNB) injection is used βN = 2.3. The time-averaged amplitude for the FIR-NTM is
to achieve a more peaked pressure profile with lower pressure significantly smaller (b) than the saturated amplitude of the
gradient at q = 2. This allows a higher βN to be realized in the smoothly growing mode (a).
an otherwise equivalent discharge with positive NBI heating
(w/o NNB in figure 8). Plasma shaping can also have a direct In figure 9 the fractional confinement degradation of
effect via the tearing stability parameter  , as well as by its the plasma stored energy (WP /WP ) due to the onset of
influence on the local pressure gradient: increased triangularity NTMs is given as function of the βN value at mode onset for
leads to higher pedestal pressure and thus to a smaller pressure ASDEX Upgrade and JET discharges, covering a wide range
gradient at the rational surface for a given βN . In addition, the of plasma parameters (ASDEX Upgrade: Btor = 1.5–2.5 T,
larger plasma current at given value of q95 causes a smaller Ip = 0.8–1.0 MA, q95 = 3.0–4.3; JET: Btor = 1.0–2.7 T,
bootstrap current fraction and thus a smaller NTM drive [109]. Ip = 1.0–2.5 MA, q95 = 2.5–4.7) and for only dominantly
As the density gradient contributes more strongly to the NBI heated discharges. It is seen that for low values of βN,onset
bootstrap current than the temperature gradient, tailoring of the confinement degradation increases linearly with the onset
the density profile is a very effective method to either avoid pressure as expected from NTM theory. At βN ≈ 2.3 the
the NTM onset or to reduce its size. On ASDEX Upgrade degradation in energy confinement is suddenly reduced (from
density profile control has been demonstrated and is now used about 30% to less than 10%). As the average H-factor at
as a general tool to avoid NTM onset or to reduce its saturated the mode onset for the discharges considered is often larger
size [38]. For weakly collisional plasmas (ν ∗ /ωD < 1, ωD : than 1, for βN values between 2.3 and 2.45 the observed energy
curvature drift frequency) density profiles tend to become confinement is very close to that predicted by the ITER scaling.
peaked [110]. Central electron heating has been shown to As seen in figure 9(a), for low βN values the NTM grows
be an effective tool to control this peaking as strong electron smoothly until it saturates at a finite amplitude. In discharges
heating leads to an increased particle transport. This effect has with βN , onset>2.3 however, the growth of the NTMs is
been attributed to trapped electron modes being the dominant often interrupted by sudden drops in NTM amplitude (see
instability under such conditions [111, 112]. figure 9(b)). As the NTM growth time is larger than the time
between two subsequent amplitude drops, the NTMs cannot
The frequently interrupted regime of NTMs (FIR-NTMs). As reach their saturated amplitude. Thus, the time-averaged NTM
NTMs are driven by the loss of bootstrap current inside amplitude is significantly reduced, reduced, resulting in a lower
the magnetic islands, their width is expected to grow nearly confinement reduction [114]. These sudden drops in NTM
proportional to the plasma pressure for given density and amplitude have been explained by non-linear coupling between
temperature profiles. On ASDEX Upgrade, however, a regime the (3,2) NTM, (1,1) and (4,3) mode activity which are locked
of (3,2) NTMs, the frequently interrupted regime (FIR), has in phase [113]. The (4,3) mode typically only occurs at higher
been discovered [113, 114]. In this regime, the confinement βN and grows rapidly, suggesting it has ideal MHD origin.
degradation due to the (3,2) NTMs is strongly reduced in It has been proven that by changing the linear stability
discharges with sufficiently high normalized plasma pressure properties of the ideal (4,3) mode, the critical βN value for
at mode onset. the transition to FIR-NTMs can be modified. Lower magnetic

S142
 Chapter 3: MHD stability, operational limits and disruptions

shear at the q = 4/3 radius allows a transition into the FIR 114504
3
regime at smaller βN values. This has been demonstrated on ECHPWR (a)
ASDEX Upgrade using localized co-ECCD just outside the 2 (MW)
q = 4/3 radius [115]. On JET a more global flattening of the 1
central magnetic shear by early LHCD also resulted in strong 0
1.740
FIR-NTMs [115]. RSURF (m) “Search and Suppress” (b)
Although the ITER operation point is planned to be Adjusts Alignment of
1.730
somewhat below the observed βN ∼ = 2.3 transition to the FIR
ECCD on q=3/2
NTM regime, it might be possible to consider operation at a 1.720
4
little higher normalized plasma pressure. Thus, (3,2) NTMs Island Decay (c)
m/n=3/2 Marginal Island
might not be a very serious problem for ITER for βN > 2.3; 2 Full Island Width
Reached
though current and density profiles may influence the exact βN W (cm)
Island Suppressed
for transition to the FIR NTM regime. A weak magnetic shear 0
that is characteristic of the recently developed advanced H- 2900 3000 3100 3200 3300 3400
Time (ms)
modes [106,116] would be additionally beneficial as it reduces
the influence of the NTMs on confinement. The weak shear Figure 10. Detail of a DIII-D discharge, 114504, in which ECCD is
is destabilizing for the (4,3) mode and thus might decrease used to suppress a 3/2 NTM. Showing (a) rf power, (b) plasma
the critical βN value for the FIR-NTMs. The reduced current major radius adjusted to align q = 3/2 on ECCD, (c) m/n = 3/2
gradient at the NTM rational surface has a stabilizing influence full island width calculated from n = 2 Mirnov signal.
on the NTM.
Whereas in the discharges considered here, the non-linear ITER scenario 2 and which uses ECCD control to stabilize the
coupling between a (3,2) NTM, (4,3) and (1,1) mode activity 3/2 NTM, has been used to benchmark ECCD control in ITER
results in a reduction of the NTM amplitude, it has also Scenario 2.
been found recently that non-linear coupling of (3,2) NTMs Modelling is done by fitting to the modified Rutherford
to other MHD modes can have a similar effect. In [117] equation, equation (6) of section 2.2.1. For benchmarking
a reduction in the (3,2) NTM amplitude due to coupling to to ITER, a form of the modified Rutherford equation is used
(4,3) and (7,5) modes with a corresponding improvement of where all current densities jbs , j|| and jec are flux surface
confinement has been shown. Furthermore, large ELMs are averaged, and all scale lengths W , δec , rs and Lq are taken
also able to reduce the NTM amplitude [115]. The interaction at the outboard midplane rational surface (here j|| is the total
between an NTM and external error fields of different helicity parallel equilibrium current density, and jec is the peak rf
is able to affect small neoclassical islands [118]. This has been current density). The modified Rutherford equation for NTM
verified experimentally on DIII-D [93]. The non-resonant field island growth or decay in this format is
control method for NTMs has the advantage that no precise 
τR dW jbs Lq Wm2 arg
positioning is needed and that it can also act on multiple = rs  + a2 1−
possible NTMs simultaneously. The disadvantage found in rs dt j|| W 3W 2
the DIII-D experiments [93] was that the plasma rotation was 
jec
greatly reduced (see section 2.4.1). −K1 (W/δec , R/W ) (8)
jbs
Conclusions. Besides active NTM stabilization by localized with, for example in DIII-D, jbs from ONETWO [119],
ECCD, as foreseen for ITER, in present-day experiments the j|| from EFIT [120], jec from TORAY-GA [121] and a2
onset of NTMs has been demonstrated to be influenced by expected to be a constant of order unity is fitted to the
seed island control, control of pressure and current gradients saturated no rf island width. For large
√ islands the effective
at the island’s rational surface, as well as by plasma shaping. marginal island width is Wmarg ≈ 3(Wpol 2
+ Wd2 )1/2 which
Seed island control allows for increasing the βN value at NTM incorporates both the stabilizing effects of perpendicular
onset well above the marginal βN value, whereas profile control thermal conductivity and polarization terms. K1 ∝ ηδec /W
and plasma shaping affect the NTM drive. For βN > 2.3, a is the ECCD effectiveness numerically evaluated as a function
favourable NTM regime (FIR-NTMs) has been found, leading of current drive width W/δec and misalignment R/W for
to only relatively benign confinement reduction due to (3,2) unmodulated ECCD [122]. Parameters for the successful
and (4,3) NTMs. If operated at sufficiently high normalized DIII–D ECCD stabilization of the m/n = 3/2 NTM are given
plasma pressures (as, e.g. anticipated for the hybrid scenarios), in table 1. The inferred value before ECCD is a2 = 2.9, using
there might be no need to stabilize (3,2) NTMs in ITER. rs  = −m = −3 and Lq = 21 cm. Given a2 and Wmarg
Thus, the effort for active stabilization might be predominantly then equation (8) determines the island evolution with ECCD
concentrated on the more dangerous (2,1) NTM. with no free parameters. Note that the possible additional
stabilizing effect of the ECCD reducing  has not been
2.2.4. NTMs expected in ITER and their control. directly experimentally verified to date and is not included here,
though studies indicate this could be an important effect [123],
Benchmarking NTMs for modelling of control in ITER. It as was demonstrated, in particular, in [90].
should be possible to extrapolate from results of present
devices, the saturated NTM island size and requirements on ITER Scenario 2, m/n = 3/2 NTM. ITER plans to have
the ECCD power for NTM stabilization in ITER. The DIII- 20 MW of injected power at 170 GHz, from a launching
D discharge 114504 (figure 10), whose properties resemble point at R = 6.485 m, Z = 4.110 m. The relatively high

S143
T.C. Hender et al

Table 1. Parameters for ECCD stabilization. transport with Bohm-like perpendicular transport to give
DIII–D, ITER, ITER, Wd [57]. Alternatively Wmarg in ITER can be deduced from the
m/n = 3/2 m/n = 3/2 m/n = 2/1 observation that it varies as twice the ion banana width when
data from DIII-D, ASDEX Upgrade and JT-60U are compared
Wmarg (cm) 2.5 3.2 4.3
δec (FWHM) (cm) 2.7 10.0 8.2 [123]; this scaling gives a smaller value for Wmarg ∼ 1.8 cm in
Wmarg /δec 0.93 0.32 0.52 ITER compared with value from the theoretical scaling, which
jbs /j 0.22 0.16 0.19 is given in table 1 along with other parameters. Modulation of
jec /jbc 1.9 2.2 1.7 the ECCD as the O-point of the islands rotate by the launcher
could increase the effectiveness about a factor of 2.5 [129].
ITER, m/n=3/2, =1.84
This would reduce the required ECCD power but the island
N
3 would have to be maintained near the peak of the green curve
in figure 11 since modulation requires a non-zero island width
2 for synchronization. However, this would represent a 75%
saturated island reduction in island width which should substantially mitigate
Island Growth Rate ( R/r)*dw/dt

without rf
the effects of the NTM.
1
A narrower current drive in ITER would both increase jec
and decrease δec making the match to the ‘marginal’ threshold
0 better and reducing the required power. A factor of two
narrower ECCD deposition width would roughly increase the
–1 ECCD effectiveness a factor of two [122] and double the jec
with rf, unmod. peak, which would result in about four times less ECCD power
–2
required for complete stabilization, i.e. reducing the power
required from 36 to 9 MW.
–3
ITER Scenario 2, m/n = 2/1 NTM. ECCD for stabilizing
0 2 4 6 8 10 12
the m/n = 2/1 mode in ITER actually has two advantages over
m/n=3/2 Full Island Width w (cm)
stabilizing the m/n = 3/2 NTM. First, the q = 2 flux surface
Figure 11. Island growth rate versus island width for the is further out allowing the ECCD to point more perpendicular,
m/n = 3/2 NTM in ITER with no rf (red) and with just enough thus making the current drive narrower. Second, the higher
unmodulated rf power (39 MW) to achieve complete stabilization collisionality at q = 2, puts q = 2 part way into the high
(green). Using 16 MW of modulated ECCD power instead would
require operating at an island size where the green curve just collisionality polarization threshold regime increasing Wpol
touches dW/dt = 0. Parameters are jbs /j|| = 0.16,  r = −3, and thus Wmarg . Radial widths and current density ratios are
wmarg /δec = 0.32, R/wmarg = 0, jec /jbs = 2.2. given in table 1 and are taken from DIII–D experiments (a new
value of a2 appropriate to the m/n=2/1 is deduced by fitting
(Z/a > 2) launching position makes the ray trajectories not DIII-D data) and from the ITER Scenario 2 equilibrium.
very perpendicular to the q = 3/2 or 2/1 surfaces. Thus, The calculation of the modified Rutherford equation for
narrow peaked current drive is difficult to achieve. Earlier m/n = 2/1 in ITER is shown in figure 12 for no rf and for jec
work evaluated launching both at a somewhat lower elevation adjusted to give complete stability. Modulated ECCD power
as well as from the midplane for a range of gyrotron frequencies would reduce the required peak power but the island would
[124]. In calculations, the tangential injection of EC beam have to be maintained at the peak of the green curve in figure 12.
Again, improvements in the launcher to make narrower current
was found most effective for the local current drive [125].
drive can reduce the required ECCD power, i.e. a factor of 4
A simple criterion of peak ECCD current density comparable
in power for a factor of 2 in current drive width.
to the bootstrap current found that 20 MW of unmodulated
injected power from an upper location from 170 to 210 GHz Summary. The proposed 20 MW, 170 GHz ECCD system
would be adequate at q = 2. Another study making different including the ‘broad’ current drive is adequate to substantially
assumptions found that either 28 MW or 30 MW of EC power control the dangerous m/n = 2/1 neoclassical tearing mode
would be sufficient to suppress both m/n = 3/2 and 2/1 NTMs and could be improved by ‘narrower’ current drive. Precise
simultaneously [126]. This exceeds the planned 20 MW. alignment on q = 2 and modulation of the peak ECCD
Neither of these studies was benchmarked to actual successful power to drive current on the O-point and not the X-point is
experiments. However, experiments in JT-60U, extrapolated required to leave enough power (after mitigating the m/n =
to ITER, concluded that 30 MW is required in ITER to stabilize 2/1 instability) for m/n = 3/2 NTM stabilization and
both modes simultaneously without further optimization of the sawtooth control. Further common benchmarking of ECCD
injection angles [127, 128]. stabilization to experiments on ASDEX Upgrade, DIII-D, JET
The calculation of the modified Rutherford equation for and JT-60U is in progress [123] and initial results support the
the m/n = 3/2 mode for ITER from benchmarking to DIII–D conclusions given here.
experiments is shown in figure 11 for no ECCD and for jec
(assuming R/δec ≡ 0) adjusted to give complete stability.
2.3. Resistive wall modes
For this case a2 is taken from the DIII-D benchmarking
and Wmarg is deduced from the low v/εωe ∗ scaling of the 2.3.1. Physics of resistive wall modes. Steady-state high
polarization term [52] and by assuming convective parallel beta fusion plasmas with a high fraction of self-generated

S144
 Chapter 3: MHD stability, operational limits and disruptions

ITER, m/n=2/1, N=1.84 improved. A key element responsible for the improved
6 agreement was the recognition of the important role of error
fields [136, 137]. Garofalo et al [138 and 139], showed
4 that the error fields slow down the plasma rotation, and also
that reduction of the error fields by correction coils makes it
Island Growth Rate ( R/r)*dw/dt

saturated island
2 without rf
possible to maintain the rotation and stability of discharges
above the no-wall beta limit. The no-wall β-limit is important
because advanced regimes in ITER are expected to operate
0 somewhat above this limit, in the domain where RWMs can
be unstable. Rotational stabilization of the RWM occurs
–2 up to the with-wall limit if the rotation is above a critical
value which depends on β and q95 . At the with-wall limit
with rf, unmod.
–4 an ideal instability co-rotating with the plasma occurs. The
effects of β and rotation on RWM stability have all been
observed experimentally in DIII-D and JET, where discharges
–6
0 6 12 18 24
were operated for a long duration over the no-wall limit,
m/n=2/1 Full Island Width w (cm) and produced measurements of the critical rotation speed
[138, 140–142]. It has been pointed out that in steady state
Figure 12. Island growth rate versus island width for the the error fields should be strongly amplified by plasmas near
m/n = 2/1 NTM in ITER with no rf (red) and with just enough
the no-wall stability limit [137] and [143] has treated the effect
unmodulated rf power (25 MW) to achieve complete stabilization
(green). Using 10 MW of modulated rf power instead would require as being time dependent.
operating at an island size where the green curve just touches A computational study [144], using several different
dW/dt = 0. Parameters are jbs /j|| = 0.19,  r = −2, rotation profiles, indicated that for typical advanced tokamak
wmarg /δec = 0.52, R/wmarg = 0, jec /jbs = 1.7. profiles, the plasma rotation in the q ≈ 2 region tends to be
an important factor for stability. These calculations, where
bootstrap current are likely to be unstable to long-wavelength the damping was conservatively evaluated by using a small
external kink modes [1]. Such kink modes can in principle parallel viscosity, overestimated the critical rotation only by
be stabilized by a nearby conducting wall, allowing beta to about 50% compared with experimental rotation profiles.
increase above the no-wall stability limit. However, if the wall Further studies of the critical rotation velocity have been
has a finite conductivity, the growth rate remains positive and made using MARS [145, 146], modelling the ion Landau
the kink mode is converted to a slowly growing resistive wall damping either as a parallel viscosity with an adjustable
mode (RWM). By the time of the ITER Physics Basis [1], coefficient or by a semi-kinetic model without free parameters.
the possibility of stabilizing RWMs by plasma rotation in In the semi-kinetic model, the drag is introduced as a local force
DIII-D discharges [130, 131], had been demonstrated. The on the displacement perpendicular to the field lines, derived
experimental results had been fairly well reproduced by ideal from the imaginary part of the potential energy, δWk , arising
MHD based models [132,133], which indicated that the critical from kinetic calculations. δWk was evaluated by an analytical
rotation velocity, above which the RWM is stabilized, scales large-aspect-ratio expansion [134] for circular plasma cross
as a small fraction (a few percent) of the Alfvén speed. The section, ignoring drift frequencies. An important conclusion
computational predictions for the critical rotation tended to from this model is that when the rotation is subsonic, the ion
Landau damping comes predominantly from regions where the
be somewhat higher than experimental observations, and the
parallel phase velocity is near sonic, i.e. close to the resonant
discrepancies were attributed to uncertainties in the drag, or
surfaces where |m/q − n|vth ,i/R ∼
rot [145]. Here, toroidal
dissipation, model [1]. The stabilizing drag on the RWM
trapping effects [134, 136] are insignificant. The parallel
comes both from the Alfvén continuum damping and ion
motion involves the toroidal side-bands; e.g. an m = 2, n = 1
Landau damping. It had been pointed out that in a torus,
magnetic perturbation couples to parallel flows with m = 1, 3,
trapping effects strongly reduce the ion Landau damping when
n = 1, and for small rotation speeds, these side-bands produce
the parallel phase velocity of the RWM in the plasma frame
damping around the rational surfaces q = 1 and 3. Already
is subsonic [134, 135]. MHD computations with the MARS
when the rotation speed is a few percent of vA , the kinetic
code, where the ion Landau damping was modelled as a
damping is spread out across the entire cross section.
parallel viscosity with an adjustable coefficient, showed that Modelling of the kinetic damping is work in progress.
a rather large viscosity coefficient (about half the cylindrical The fluid model with a small parallel viscosity usually
value) gives critical rotation velocities close to experimental overestimates the critical rotation, by the order of 50% [140].
results [133]. The discrepancy can be removed by a larger parallel viscosity.
Prior to the ITER Physics Basis [1], active feedback of However, there are experimental discharges that are manifestly
the RWM had only been suggested theoretically. The main stable [139, 145], where a large parallel viscosity fails to
option at that time was a ‘smart-shell’ arrangement, with a large produce complete RWM stabilization. The kinetic model gives
number of active coils covering most of the plasma surface. lower critical rotation that can be close to the experimental
Subsequent developments in this area are discussed below. values, or underestimate them by as much as 40% [140, 142].
Kurita et al [147] showed that ferromagnetic walls give lower
Stabilization by rotation. Since 1999, the agreement between beta limits, but this effect will be small in ITER because of
experiment and theory on rotational stabilization has been saturation.

S145
T.C. Hender et al

Another important aspect of RWM physics, where P (s) [157] is defined, giving the normalized flux or the
experiment and theory can be compared, is the resonant field magnetic field at the sensor position, resulting from a current
amplification (RFA), which is defined as the relative change in in the active coils:
perturbed radial magnetic field from its vacuum value. Here,
single-mode theory ignoring rotation [137] predicted that the P (s) = bs (s)/b0 If (s). (9)
steady-state RFA should peak (in fact, go to infinity [148])
Here, bs (s) is the field measured by the sensors, If (s) is the
at the marginal stability for a kink like perturbation, which
current in the feedback coils and s is the Laplace transform
was interpreted as ‘the no-wall limit against the external kink’
variable. The normalization constant b0 can be chosen as the
in [149], while in experiments with rotating plasmas the RFA
field at the sensors produced by a 1 A current in the active coils
increases significantly with pressure [141,150–152] above the
in the absence of the walls and the plasma. In the simplest
no-wall limit. An alternative theory [143] shows that RFA is
feedback system, the magnetic field detected by the sensors is
largest (though always finite) just at the stability boundary.
fed back with a negative multiplier −K(s), where K(s) is the
Calculations with MARS using a fluid model with small
gain. The characteristic equation for the closed-loop is
parallel viscosity also makes the RFA peak, inconsistently with
experiment, near the no-wall limit, while models with stronger 1 + K(s)P (s) = 0. (10)
damping (fluid model and cylindrical parallel viscosity or the
semi-kinetic model) reproduce the experimental RFA for static In actual control systems using voltage control, it is useful to
error fields rather well [145, 151]. introduce another transfer function that measures the ‘loaded’
There are other aspects of the RFA experiments where self-inductance of the feedback coil [157]. But for the control
the modelling is as yet less successful, e.g. the models in system currently envisaged for ITER, the active coils are far
MARS tend to give too large real frequencies of stable from the resistive wall, and voltage control gives only minor
RWMs. In [153], a single-mode model was used, where modifications over current control [158].
the plasma dynamics is represented by a single complex The simplest theories consider cylindrical geometry and
stability parameter, which determines both the RFA and the assume that active and sensor coils can interact via a single
RWM growth-rate. According to the single-mode model, toroidal and poloidal mode (m, n) [159, 160]. Although
the experimental results in [153] imply that for stable single-mode models ignore the geometry of the coil system,
RWM, the angular real frequency is typically τw−1 times a they give valuable insights into the main properties of the
small number of the order 0.1–0.2, while MARS generally different types of sensors. Three main types of sensors have
predicts larger frequencies. The semi-kinetic model currently been analysed [145, 161]: radial, poloidal and radial sensors
implemented in MARS contains a number of approximations with the direct field from the active coils subtracted [162]
that can be removed; e.g. by including drift frequencies, (including the ac screening effect of the wall). One drawback
kinetic modifications of plasma inertia, shaping effects on the of radial sensors located on the wall is that the signal from
damping, etc, and work in this direction is in progress. the unstable mode vanishes as the RWM approaches the ideal-
Concerning scaling for extrapolation to ITER, both the wall stability limit [145, 163]. Poloidal detectors inside the
fluid and the semi-kinetic model ignoring drift frequencies first wall avoid this difficulty [145, 157, 161, 164, 165]. In the
imply that the critical rotation velocity scales as the Alfvén single-mode approximation, transfer functions for radial and
speed (or equivalently, at fixed beta, the ion sound speed), poloidal sensors for a mode with poloidal mode number have
and this agrees with DIII-D experiments [140]. (Among the the form [158]
approximations in the semi-kinetic model, the neglect of the
{Pmr (s), Pmp (s)} = |m|(rw /rf )|m|−1 {1, (2m +|m|)/m}/(s−m )
drift frequencies may be the most important to remove, as it
(11)
will give deviations from the Alfvénic scaling.) In DIII-D, the
with a single pole at s = m = γm τw which is the
critical rotation speed at q = 2 is typically of order 1% of
normalized growth rate of the RWM in the absence of feedback.
the Alfvén velocity (or less in recent experiments [154, 155]),
Equation (11) shows that when the system approaches the
and this is predicted to be marginally achieved by transport
ideal-wall stability limit, m → ∞, the transfer function for
calculations for ITER with the ASTRA code [156]. Therefore,
radial sensors vanishes, while that for poloidal sensors remains
the strategy for ITER is to develop active feedback stabilization
finite.
and also study its interaction with plasma rotation.
Multimode cylindrical theories [145, 158, 166] take into
account the shape of sensor and active coils by decomposing
Active feedback. RWM feedback systems use magnetic
the coil currents into poloidal Fourier components. The total
sensor (and sometimes other) signals to gain information on
transfer functions are sums over single-m transfer functions:
the amplitude and phase of the unstable RWM, and apply
this signal to control currents in a set of non-axisymmetric P (s) = m cm Pm (s)sm , (12)
feedback coils. Models have been developed using ideal MHD.
These models combine the force balance equation in the plasma where the cm and sm are geometrical coefficients characterizing
(−∇p1 + j1 × B0 + J0 × b1 = 0) for static equilibria, where the feedback and the sensor coils, respectively.
inertia can be ignored) with j1 = 0 in the vacuum region and The multimode theories show that radial sensors fail when
jump conditions for b1 × n on the walls, which are modelled m is large [145, 158, 163], because of interference from the
as thin resistive shells. stable RWM, although ac compensation somewhat improves
To determine the closed-loop growth rates, following the performance of radial sensors [145, 161, 162]. Coupling to
practice in control theory, a non-dimensional transfer function the stable RWM is not a major problem for poloidal sensors.

S146
 Chapter 3: MHD stability, operational limits and disruptions

2 In summary, a simple coil system with poloidal sensors

JET EQ and a single feedback coil in the poloidal direction allow robust
r =1.2a, r =1.5a control with respect to changes in plasma pressure (up to a
w f
5.46
internal bθ sensor certain limit), plasma current and rotation, and can readily
1.5
deal with the double wall structure for ITER [146].
4.59

2.3.2. Control of RWMs. As summarized in section 2.3.1,

cr

1 theory predicts that the RWM can be stabilized by rotation

K

3.87
of the plasma of order of a per cent of the Alfvén velocity
3.07 or by feedback control with non-axisymmetric coils or by a
0.5 combination of rotation and feedback. At the time of writing of
the ITER Physics Basis [1], early experiments in DIII-D [130]
q95=2.88
and PBX-M [172] had demonstrated that stable plasmas could
be achieved with beta above the no-wall beta limit, by means of
0 rapid plasma rotation driven by neutral beam injection. Such
0.1 0.15 0.2 0.25 0.3 0.35
λ discharges were sustained at beta above the no-wall limit for
up to 50 times the resistive magnetic field penetration time
Figure 13. Critical gains as functions of the poloidal width of the of the wall, demonstrating the principle of wall stabilization.
coil for equilibria with pressure half-way between the no-wall and
the ideal-wall limits and different plasma currents. The coil width is Since then several other tokamaks, including NSTX [173],
given by λ which is the fraction of the poloidal circumference JT-60U [174] and JET [175], have also achieved beta values
subtended by the active coils (from [146] ‘Robust control of above the calculated no-wall stability limit. The critical
resistive wall modes in tokamaks’). rotation velocities for stability are consistent with the expected
values of O(1)% of the Alfvén velocity [176]. However, in
Even the double wall structure of ITER can be readily dealt these experiments, extension to longer pulses proved difficult.
with by a system using poloidal sensors [158, 167, 168]. Typically the plasma rotation became slower as beta exceeded
The most realistic modelling of active feedback so the no-wall limit, eventually allowing the growth of a resistive
far has been made with toroidal stability codes: MARS-F wall mode.
[145, 157, 168–170], the combination of DCON for the plasma Significant progress has been made in the last five
and VACUUM for the vacuum-resistive wall region [164,165], years towards the goal of sustained wall stabilization of
the KINX code [143,171] and DCON combined with VALEN high-beta plasmas, using both rapid plasma rotation and
for three-dimensional modelling of the resistive wall [164]. direct feedback control. Recent experiments have developed
Currently, benchmarks between these codes are in progress. techniques to sustain rotational stabilization for long pulses.
Also for toroidal systems, the transfer functions are sums Feedback control using non-axisymmetric coils has proven
over infinitely many isolated poles [165, 168], corresponding valuable in combination with plasma rotation, and initial
to the RWM growth rates in the absence of feedback. The experiments in feedback control with little or no rotation appear
transfer functions can be well approximated in the entire promising.
unstable half-plane by lumping several poles, and a three- (or An important recent discovery related to RWM control has
for poloidal sensors even two-) pole approximation is generally been the ‘error field amplification’ effect [136, 137, 177–179],
very accurate [168]. also known as ‘resonant field amplification.’ This phenomenon
Calculations with several codes [157, 164, 165] have is a consequence of the fact that plasma rotation provides only
shown superior performance of poloidal sensors. Work with weak stabilization and rotation of the RWM, so that a small
MARS-F has shown that the control can be made robust force can produce a relatively large (but stable) displacement.
with respect to variations in plasma pressure, current and Simple lumped-parameter models for resistive wall mode
rotation [145, 158, 168], if the active coils have a single coil in stability [138,150,160,162,180] lead to an equation of the form
the poloidal direction located at the outboard midplane (e.g.
as the C-coil in DIII-D). This is because such coils couple τw dBm /dt − γ0 τw Bm − H Bext = 0, (13)
well to the high-beta n = 1 kinks which balloon on the
outboard side of the torus. Figure 13 shows how the minimum where Bm is the amplitude of the total perturbed magnetic
proportional gain needed for stabilizing equilibria, about half- field evaluated at the resistive wall, Bext is the amplitude of
way between the no-wall and the ideal-wall limits, depends an externally applied field that is resonant with the plasma
on the poloidal width of the coil for different plasma currents mode, γ0 is the growth rate of the mode in the absence of
(q95 -values). A single coil array covering about 20% of the an externally applied field, τw is the inductive time constant
poloidal circumference [157] works well over a large range of for the resonant field to penetrate the resistive wall (in the
plasma currents. Multiple active coils in the poloidal direction cylindrical approximation τw = µw σ dr/2m, where σ , d and
are useful when radial sensors are used [146, 164, 165], but do r are the conductivity, the thickness and the radius of the
not bring substantial improvements in poloidal sensors [170]. wall and m is the poloidal mode number) and H = (1 +
The combined effect of feedback and rotation has also Cγ0 τw )/(1 − C). Here C is a factor that expresses the relative
been studied [146,168]. These studies show a synergistic effect inductive coupling of the plasma (p), wall (w) and coils (c):
if the toroidal phase of the feedback is shifted so that it pushes C = 1 − Mpw Mwc /Lw Mpc , where M and L are the mutual and
the mode in the same direction as the plasma rotation. self-inductances. In the simple cases of a slab or a cylindrical

S147
T.C. Hender et al

114813 114531
4 I-Coil Current (kA)

2 Feedback Control
I–Coil Turned Off
3
βN
2
1 2.4 i ≈ No-Wall Limit
200
150
Vφ at q ≈ 2 (km/s)
100
50
200
150 δBr (Gauss)
100
50
0
500 1000 1500 2000 2500 3000 3500 4000 4500
Time (ms)

Figure 15. Feedback-controlled error field correction in DIII-D

(discharge 114531, solid line) and a similar discharge without error
correction (114813, dotted line). Shown are (a) I-coil current, which
is feedback-controlled after t = 1300 ms in discharge 114531,
(b) normalized beta and the estimated no-wall limit, (c) toroidal
Figure 14. Discharge #59223, an error field amplification pulse in rotation speed at r/a = 0.6 and (d) amplitude of the RWM.
JET, showing time evolution of (a) current in the internal saddle (Reprinted with permission from [190]. Copyright 2004, American
coils, (b) normalized beta compared with 3.4li , and (c) plasma Institute of Physics.)
response measured at the wall [175].

no-wall limit using external control coils [183], and RWM

geometry with external coils, C → 0 and H reduces to unity. stabilization by plasma rotation has become a routine element
In steady state, a weakly damped mode (γ0 < 0) has a finite- of DIII-D advanced tokamak experiments [187]. The recently
amplitude response Bm = (−γ0 τw )−1 H Bext to a resonant installed internal control coils [188,189] have provided similar
external magnetic perturbation such as a magnetic field error. performance [190], with wall stabilization sustained for more
This enhanced plasma response, which becomes larger as the than 2.5 s (see figure 15). (In this discharge, a very broad
mode approaches marginal stability (γ0 → 0), can cause strong current density profile and low triangularity lead to a relatively
damping of the plasma rotation. The slowing of the rotation low beta limit; ideal MHD stability calculations show that the
can then lead to loss of stability, as observed in DIII-D [181] no-wall stability limit is well approximated by βN = 2.4li with
and more recently in NSTX [173] and JET [142, 175]. (In li = Bp2 /Bp 2 ). Similarly, reduction of the intrinsic field
contrast, a strong slowing of rotation is not observed at high errors in NSTX by an adjustment of poloidal field coil positions
beta in JT-60U [182], perhaps because the resonant component has led to improved values of both toroidal beta and normalized
of the error field is smaller.) beta [191, 192]. The low-activation properties of ferritic
Resonant field amplification has been demonstrated steel are attractive for fusion reactors, but its ferromagnetism
explicitly by the application of pulsed magnetic perturbations introduces the possibility of additional field errors (see also
using non-axisymmetric coils [141, 175, 183–185]. Figure 14 section 2.4.1). Recent encouraging results from JFT-2M
shows an example from JET in which a series of square indicate that a ferritic steel wall can provide stabilization at
pulses was applied using internal saddle coils [175]. Here high beta [193], although modelling predicts some reduction
the no-wall ideal MHD stability limit is approximated by in the ideal-wall limit [194]. In both DIII-D and NSTX, a fast
βN = 3.4li (here the internal inductance of the plasma beta collapse or disruption with an ideal time scale magnetic
is li = 2V Bp2 /(µ20 Ip2 R) with V = plasma volume), as precursor has been observed in some discharges, consistent
confirmed by stability calculations and measurements [141]. with stability calculations indicating that in these cases beta is
The plasma response, measured at a location far from the near the ideal MHD wall-stabilized beta limit [139, 192].
driving coils, is almost undetectable at low beta, but increases Despite the success of rotational stabilization, it is also
strongly as beta approaches and exceeds the no-wall stability necessary to develop direct feedback control of the RWM with
limit, where the RWM is only weakly damped. non-axisymmetric coils. Burning fusion plasmas, lacking the
The new understanding of resonant response to error large neutral beam torque of present experiments, may not
fields has been used to improve performance in the wall- reach the critical rotation frequency for RWM stabilization.
stabilized regime, and the feasibility of stabilization by plasma Modelling with the MARS code [132,133,195] indicates that in
rotation now seems well established. Feedback-driven control the regime of subcritical plasma rotation, both feedback control
of error fields has been effective in reducing magnetic field and rotation can contribute to stability [169, 196]. As the
asymmetries in DIII-D, allowing continued rotation above rotation becomes small, stabilization must rely on feedback
the critical value for RWM stabilization. The resonant alone.
response of the stable RWM enhances the detection of small Recent experiments have yielded encouraging results
error fields, and the feedback system then corrects the error for the feasibility of feedback stabilization of the RWM.
field by minimizing the enhanced plasma response [186]. Experiments in HBT-EP have shown that RWM instabilities
This approach has allowed sustained operation above the related to current-driven kink modes can be stabilized with

S148
 Chapter 3: MHD stability, operational limits and disruptions

direct feedback control [197, 198]. Feedback control of ideal wall

1.0
the RWM at high beta has been demonstrated in DIII-D limit I-coil Gp/Gd
MARS Prediction
[150, 162], using a set of six external control coils originally q=2 114819 40/20
114820 40/40
designed for error field correction. The performance of the

no-wall

— βNno-wall
114340 1500 ms 114821 40/0
feedback system is in good qualitative agreement with lumped- Rotation C-COIL 114822 40/80

— βN
10
parameter models [138, 150, 162, 180]. Here equation (13) Frequency 114817 No FB
(kHz)
becomes 0

ideal-wall
Cβ 0.0 0.4
ρ
0.8
τw dBm /dt − γ0 τw Bm − Gm (ω)H Bm = 0, (14)

βN

N
β
Time
where the feedback-controlled external field is Bext =

Cβ =
Magnetic
Gm (ω)Bm . (In terms of the discussion in the preceding section, Braking
this equation is equivalent to the characteristic equation (10) NO
FEEDBACK
combined with the single-mode transfer function (11). Here no wall
limit 0.0
the time derivative is written explicitly.) The stability of the 0 50 1 00 150
system depends strongly on the mutual inductive couplings Rotation (km/s)
of the sensors, mode, wall and external coils (the factor H
Figure 16. Trajectories of several DIII-D feedback-stabilized
defined above) and on the response of the feedback system,
discharges in beta and rotation (at r/a ∼ 0.6), compared with a
which is expressed here as the effective ‘gain’ function Gm (ω). representative stability limit predicted by MARS. Here beta is
(The function Gm (ω) includes the transfer function of the plotted as Cβ , the fraction of the no-wall to the ideal-wall interval.
sensors, the gain of the control system and the response of A discharge without feedback is also shown. Open circles indicate
the coil–amplifier system.) For example, at high frequency, the onset of the RWM. Inset shows the toroidal rotation velocity
induced currents in the wall reduce the coupling of the mode profile for the ‘magnetic braking’ case, shortly before the onset of
the RWM. (Reprinted with permission from [190]. Copyright 2004,
to radial field sensors but not to poloidal field sensors inside American Institute of Physics.)
the wall, requiring the use of a larger gain with radial field
sensors as γ0 increases. The stronger coupling of poloidal comparison is only qualitative. Nevertheless, the experimental
field sensors may be particularly advantageous in multimode results are reasonably consistent with the MARS prediction:
systems, where coupling to stable modes must be minimized. a discharge without feedback becomes unstable near the
Indeed, experiments show that feedback control is improved by calculated stability boundary, while discharges with feedback
the use of internal poloidal field sensors [161] over radial field are able to cross the calculated boundary and continue to
sensors, as predicted by numerical modelling [157, 158, 199] significantly higher beta and lower rotation.
and analytic theory [161, 163, 180]. The goal of future work is a full realization of the
The internal control coils [188,189] have further improved feedback control of resistive wall modes and confident
DIII-D feedback performance [190], providing stability extrapolation to burning plasmas. This will require validation
against RWMs with higher growth rates. Internal coils are of feedback models with and without plasma rotation [133,
more effective because even at high frequency, their field is 169, 196, 199], using experimental plasmas with rotation
not completely cancelled by induced wall currents, allowing rates above and below the critical value. Active MHD
a faster response to rapidly growing RWMs. This can be spectroscopy [183,184,201,202] is a promising new technique
expressed quantitatively using the coupling factor defined for experimental investigation of the physics of rotational
above: C = 1 − Mpw Mwc /Lw Mpc . In order for feedback stabilization. Modelling indicates that a three-fold increase
control to provide the same stabilization as an ideal wall, in the bandwidth of the DIII-D feedback system should allow
the condition C
0 must be satisfied [159, 200]. That feedback stabilization up to the ideal-wall beta limit in the
is, the coupling of the coils to the plasma via induced wall absence of plasma rotation [190].
currents must not exceed the direct coupling of the coils to
the plasma. In a cylindrical model with helical currents, 2.3.3. RWMs expected in ITER and their control. The RWM
the case of external coils is just marginal: C = 0, while instability is expected in ITER steady-state scenarios with high
internal coils give C = 1 − (ac /aw )2m > 0, where ac and values of βN and low values of internal inductance. A typical
aw are the minor radii of the coils and the wall and m is the representative of ITER steady-state scenarios is the Design
poloidal mode number. The improved performance in DIII- Scenario 4 (see [203] where it is termed Scenario 1). This
D is qualitatively consistent with modelling using the VALEN is a 9 MA highly shaped plasma with weak negative shear,
code [164], which has predicted that in the actual experimental producing about 300 MW of fusion power with Q = 5 for
geometry, without the benefit of plasma rotation, internal 3000 s (Rp = 6.35 m, ap = 1.85 m, κsep = 1.97, δsep = 0.58,
control coils should allow feedback stabilization essentially to βN = 2.57, li = 0.63). The plasma cross-section is shown in
the ideal wall-stabilized limit. As shown in figure 16, internal figure 17. A set of Scenario 4-like plasmas was considered in
coils have provided stability at higher beta and lower plasma the studies of RWMs expected in ITER steady-state scenarios.
rotation than was possible with external coils. In the figure, These plasmas have the same current, shape, about the same
the trajectories of several experimental discharges in beta and profile of q, but different values of βN . To provide this βN -scan,
rotation are compared with a representative stability boundary the plasma toroidal current was specified as
predicted by the MARS [132, 195] code in the absence of  
feedback [133, 169]. The MARS calculation uses parameters R Rp
jtor = j0 αG(ψp ) + [H (ψp ) − αG(ψp )] , (15)
typical of DIII-D but not specific to these discharges, so the Rp R

S149
T.C. Hender et al

6 Table 2. Values of q and βN as functions of the parameter α.

α q0 qmin q95 βN
4 1.00 3.24 2.43 5.12 2.57
1.10 3.17 2.39 5.15 2.87
1.20 3.11 2.35 5.19 3.17
2 1.30 3.04 2.31 5.23 3.48
Z, m

0 Table 3. The ‘no-wall’ and the ‘ideal-wall’ βN limits for n = 1 kink

modes.
Code βN (no wall) βN (ideal wall)
-2
MARS-F [146] 2.45 3.65
KINX [207] 2.33 3.6
-4 VALEN [208] 2.52 3.5
PEST-2 [209] 2.45 3.65

-6
2 4 6 8 10 12 30
R, m RWM growth rate x 0.188s
Figure 17. Plasma, vacuum vessel, magnetic sensor (circle) and
coils(squares) used in the study of RWM control. 25
Growth rate x 0.188s(MARS-F)
20 Growth rate x 0.188s(KINX)
Growth rate x 0.188s(VALEN)
3Cbeta/(1-Cbeta)
15

10

0
0 0.2 0.4 0.6 0.8 1
Cβ
Figure 19. RWM growth rate normalized on τw = 0.188 s versus
Figure 18. Profiles of plasma pressure, Peq , and q for the plasma of Cβ for ITER Scenario 4 plasmas calculated with codes MARS-F,
Design Scenario 4 [146]. KINX, VALEN and the analytical scaling. A 2D model of the
vacuum vessel, no blanket.

where j0 is adjusted to get the plasma current of 9 MA, ψp The degree of RWM instability is characterized by a
is the normalized poloidal magnetic flux and the functions dimensionless parameter:
H (ψp ) and G(ψp ) [146] are obtained in the simulation of
Scenario 4, corresponding to the case α = 1, with the transport βN − βNno wall
code ASTRA [203]. Different values of βN are obtained by a Cβ ≡ , (16)
βNwall − βNno wall
variation in the parameter α. The profiles of plasma pressure,
Peq , and q are shown in figure 18. Table 2 shows the change in where βNno-wall and βNno-wall are the ‘no-wall’ and the ‘ideal-wall’
q and βN with variation in α. For these the ‘no-wall’ βN limit limits for the n = 1 kink mode. In Scenario 4, which has
for n = 1 kink modes is about 2.4. βN = 2.57, Cβ  1 and only weakly unstable RWMs are
The main conducting structure affecting RWMs is the expected.
vacuum vessel consisting of the inner and the outer walls shown Figure 19 shows the dependence of the RWM growth
in figure 17. The vacuum vessel time constant, τw , for the mode rate  on Cβ calculated with the codes MARS-F (circles),
m = 1, n = 1 is 0.25 s for an axisymmetric 2D model and KINX (triangles) and VALEN (squares) for the 2D model of
0.188 s for a 3D model, which takes into account the opening the vacuum vessel inner wall. The solid smooth line shows
for the ports [204]. For Scenario 4 plasmas, the ‘ideal-wall’ the scaling τ w = 3Cβ /(1 − Cβ ) [205]. The port openings
βN limit for n = 1 kink modes is about 3.6. The stability limits reduce the ‘ideal-wall’ βN limit by about 7% and increase the
calculated for the 2D model of the vacuum vessel inner wall RWM growth rate by about a factor of 2, if the values of βN
using different codes are given in table 3. between 3 and 3.4 are considered [204]. The RWM growth

S150
 Chapter 3: MHD stability, operational limits and disruptions

VALEN: VV(3D) & BM(3D, 5ms)

10000.00
MARS-F: VV(2D) & BM(2D, 5ms)
KINX: VV(2D) & BM(2D, 5ms)

1000.00

RWM growth rate, s-1

100.00

10.00

1.00
2 2.5 3 3.5 4 4.5 5
BetaN
Figure 20. RWM growth rate normalized on τw = 0.188 s versus
Cβ for ITER Scenario 4 plasmas calculated with code KINX for Figure 21. RWM growth rate versus βN for ITER Scenario 4
different models of the vacuum vessel: G2-two walls, G1-inner wall, plasmas calculated with codes MARS-F, KINX, VALEN. The
G-scaling. blanket modules are taken into account. No plasma rotation.

rate as a function of Cβ calculated with the KINX code with in figure 17, a pair of coils produce the radial magnetic field of
(the line with open circles) and without (the line with squares) 0.1 T/MAturn. Each coil has 28 superconducting turns. The
account of the outer wall is shown in figure 20 [171]. The line resistance of feedback circuit is determined by the resistance of
without markers shows the scaling mentioned above. The outer busbars, 4 m
. This makes the time constant of the feedback
wall only slightly reduces the RWM growth rate; however, it circuit to be about 10 s, which is much higher than the vacuum
deteriorates RWM active stabilization using external feedback vessel time constant τw . Therefore the resistive voltage drop
coils (see figure 24) by screening the magnetic field produced in the busbars can be neglected in the studies of RWM control.
by the coils [167]. The magnetic sensors used in ITER in the feedback loop
The blanket modules are located between the vacuum of the RWM suppression are located on the plasma side of the
inner wall and the plasma. Each module is insulated from vacuum vessel inner wall. The position of the sensors, used
the other modules and consists of a thin first-wall panel and in the studies of RWM control, is shown in figure 17. They
a thick shield block. The first-wall panel has many cuts, measure the poloidal magnetic field of the modes n = 1. The
which strongly reduce the time constant of penetration of the expected amplitude of the noise in the sensors (signals other
magnetic field normal to the first wall. The time constant of than RWM and ELMs) is less than 0.5 mT.
a shield block is significantly higher, about 5 ms. The effect The study of RWM feedback control in ITER was done
of blanket modules on the RWM growth rate was studied with with the codes MARS-F and VALEN. The code MARS-F
the codes VALEN, MARS-F and KINX using very optimistic considers all RWM modes, but uses a 2D model of the feedback
models, from the point of view of RWM stabilization. In coils (effectively like a large number of coils) and a 2D model
the VALEN code, the blanket modules were represented by of the vacuum vessel. The VALEN code considers only a single
insulated conducting plates placed on the first-wall panel. mode (the most unstable), but uses more realistic 3D models of
Each plate has the time constant 5 ms. The codes MARS-F the feedback coils, the vacuum vessel and the blanket modules.
and KINX used axisymmetric models of the blanket modules. Work to improve this modelling is in progress.
The blanket modules were represented by an axisymmetric The studies of RWM control with MARS-F were done
conducting surface placed on the first-wall panels having the in three steps. Firstly, the transfer function, mapping the
time constant 5 ms for the mode m = 1, n = 1. The coil voltages to the magnetic field on sensors, was obtained.
dependence of the RWM growth rate on the value of βN , Then this transfer function was used for the design of
calculated with these codes with the simplified models of controllers (second step). And thirdly, the controllers were
blanket modules taken into account, is shown in figure 21. validated in the simulations of stabilization of RWM. The
The side correction coils (see section 2.4.2) may be stabilization starts when the magnetic field on the sensor is
used for the RWM control in ITER. There are three pairs above B0 = 1.5 mT. The studies have shown that, even with
of toroidally opposite coils connected to produce a magnetic the double wall conducting structure, a simple coil system,
field mainly with n = 1. Each pair has an independent power with a single coil set in the poloidal direction, ensures robust
supply. The maximum coil current is 280 MAturns, the voltage RWM control with respect to changes in plasma pressure
insulation limit is 360 V/turns and the inductance of a pair of the (up to a certain limit) and rotation. The results obtained for
coils is 50 µH/turns2 . About 120 MAturns are required for the three controllers having different performances are presented
correction of the error field expected in ITER (see section 2.5.2) in figure 22 [206]. The figure shows the maximum voltage
and more than 160 MAturns are available for the feedback required for the stabilization of RWM as a function of Cβ .
stabilization of RWM. At the position of the sensor shown The voltage is proportional to B0 . The studies done with

S151
T.C. Hender et al

induced) based on studies on JET, COMPASS-D [212] and

DIII-D [213]:
α
Bpen /Bt ∝ nαn BtαB q95q R αR . (17)

Data from Compass-D, DIII-D and JET were used to establish

all the scaling coefficients except αR which could then be
determined from the Ohmic scale-invariance constraint, 8αn +
5αB − 4αR = 0. There was agreement between the three
machines that αn ∼ 1, but αB ∼ −2.9 on Compass-D and
αB ∼ −1 on JET and DIII-D. In addition in the IPB it was
reported that the harmonic content of the error field (not just
the magnitude of the m = 2, n = 1 component) affected
the threshold, following from the previous studies [214, 215]
and that as the ideal stability β-limit was approached, even in
the presence of strong co-injected NBI heating (and therefore
strong induced plasma rotation), the error field threshold
decreased.
Figure 22. Maximum voltage required for stabilization of RWM by Since the IPB further error field experiments have been
side correction coils versus Cβ for three controllers with different
performances. The stabilization starts when B0 exceeds 1.5 mT.
conducted on JET and DIII-D, and experiments have started
Code MARS-F and no plasma rotation [206]. on Alcator C-Mod [216] and TEXTOR [217] which amongst
other things have confirmed the linear density dependence
of the error field threshold. On Alcator C-Mod, which
MARS-F have shown that RWMs with Cβ close to 1 can be
has linear dimensions of a factor of 4.3 smaller than JET,
stabilized, if the coil voltage is not limited. For stabilization
the typical locked error field threshold values, relative to
of the highly unstable RWM, the feedback voltage should
toroidal field, are very similar to JET and DIII-D. Its threshold
be proportional to a combination of the first and the second
measurements also agree with the scaling based on scale-
derivatives of the magnetic perturbation on the probes. Plasma
invariance arguments [1] to within a factor of 2, thus providing
rotation below the critical velocity does not significantly affect
additional confirmation, based on direct size scaling. These
the performance of RWM feedback stabilization by the side
further experiments on JET, DIII-D, Alcator C-Mod and
correction coils.
TEXTOR have been aimed at refining the understanding of
The studies done with the code VALEN [204] show the physics of how error fields apply a braking torque to the
lower efficiency of the side correction coils with proportional plasma, additional studies of the effect of harmonic content
and derivative feedback stabilization than was demonstrated and the effects of error fields near the m = 2, n = 1 NTM
in MARS-F simulations: RWMs with Cβ < 0.68 can be β-limit. In several cases these studies are still incomplete and
stabilized. More detailed studies of the RWM stabilization so the results presented below represent a progress report.
are ongoing.
Physics of error field braking. JET experiments [218] show
2.4. Error fields a good match to theory, with locked mode formation being
precipitated after the EM torque has slowed the plasma
2.4.1. Effect of error fields on plasmas. to approximately 1/2 its original frequency. Comparisons
against various theory models indicate the ideal viscous model
Status as of the 1999 ITER Physics Basis. The effect of [219] best represents the JET data, with the error field threshold
non-axisymmetric error fields in inducing locked modes in √
scaling with the unperturbed rotation as ω0 . In Alcator
tokamaks was discussed in chapter 3, section 2.5 of the C-Mod clearly visible island structures occur in the electron
1999 IPB [1]. The error fields occur due to poloidal and temperature profiles at q = 2 when the locked mode is formed
toroidal field coil imperfections and misalignments, from the [216], confirming previous observations on JET [210]. The
current feeds to these coils and from ferritic material in the new JET results [218] have also shed light on the important
vicinity of the plasma. In the present experiments, to study physics of plasma rotation braking from applied error fields.
their properties, error fields may also be deliberately applied A viscous drag model with the torque applied solely at the
using non-axisymmetric coil systems. Error field locked island (q = 2) surface would predict a uniform reduction of the
modes are operationally important because they tend to persist rotation within q = 2, but the observations on JET indicate an
once established and then limit the performance or cause a approximately self-similar reduction within q = 2 (figure 23).
disruption; conversely their correction can reduce disruptivity A new model associated with a toroidal viscous drag
and allow lower density to be accessed [210, 211]. Error field originating from non-axisymmetric fields, in particular due to
locked modes are understood to arise from the braking torque the non-resonant m = 0, n = 1 mode, seems to qualitatively
applied to the plasma from a static error field, which can bring match these observations [220].
the rotating plasma to rest (most notably at q = 2) and allow Experiments on DIII-D [93] have also shown an error field
islands to form (again most notably m = 2, n = 1). braking effect due to a largely non-resonant applied error field
In the 1999 IPB an empirical scaling was reported for (dominantly n = 3, m = 1), and such an effect has also been
the error field threshold Bpen (above which locked modes are discussed theoretically [222]. It is found that the n = 3 field

S152
 Chapter 3: MHD stability, operational limits and disruptions

necessary to determine the intrinsic error field content (due

R = 3.0 to 3.7m
to coil misalignments, etc) to add to the applied error field.
1 Measurements in 2001 using an in-vessel rig determined the
Vφ change

main sources of field error in DIII-D [228] and so the fields

from these can be combined with those from the error field
coils to give the total non-axisymmetric field. However, it
should be noted that there are still difficulties in reconciling
0 the applied fields required to empirically correct error fields
with those determined using the measuring rig [227, 229].
The 1999 ITER Physics Basis expression for the effective
error field from m = 1–3, n = 1 was [230]
2
Bpen = 0.28Br1,1
2
+ (0.25Br11 + Br21 − 0.05Br31 )2 + 0.51Br31
2
,
(18)
where Brm,n is the unweighted (m, n) Fourier component

JG03.83-4c
evaluated in straight field line coordinates at q = 2. This
form was based on the theoretically expected form, whereas a
2 seconds previous simpler empirical expression (equation (19)) has been
used extensively in the design of the ITER error correction coils
Figure 23. (Upper) Change in toroidal velocity for JET pulse 52067 (see section 2.4.2):
(normalized to its value just before the error field is applied) for
radii within q = 2. (Lower) Measured radial magnetic field. The 2
Bpen = B2,1
2 2
+ 0.8Br3,1 2
+ 0.2Br1,1 . (19)
error field locked mode is induced when this signal departs from a
linear ramp (indicated by the broken line) expected from a direct Recent fits based on DIII-D I-coil data are given in [227].
response to the linearly ramped error field. As the error field is
turned off, a slowly rotating island occurs, which slowly decays and
Comparison of all the fits for 3-mode coupling against the
rotates with increasing frequency. The plasma toroidal velocity then DIII-D and the COMPASS-D data shows at present that there is
returns approximately to its initial value [221]. no satisfactory fit covering both machines, possibly reflecting
the different toroidal field scaling found in the two machines.
causes an 83% reduction of the momentum confinement time Overall the simplest empirical fit (equation (19)) best describes
and that this is consistent with the expected braking from the the data in both machines. This ‘3-mode’ criterion is applied
n = 3 field ripple. However, in this case the n = 3, m = 1 (left- to the design of the ITER correction coils (see section 2.4.2)
and right-handed) perturbation has Br1,3 /Bt = 2.4 × 10−3 , where the requirement is that B3-mode /Bt < 5 × 10−5 for
compared with an otherwise equivalent case where an n = 1, avoiding locked modes. The choice of the constant (5 × 10−5 )
m = 2 perturbation of only Br2,1 /Bt = 4 × 10−4 causes is conservative relative to the predicted error field sensitivity
a 50% rotation reduction and subsequent locking [223] at of ITER [210].
similar βN ; this indicates that while these n = 3 non-resonant
fields can indeed slow the plasma rotation, resonant fields Error fields due to ferritic material. In order to reduce ripple
have a greater drag at the same relative applied perturbation. losses, regions of ferritic steel have been installed in stages
Recent experiments have also shown the non-resonant effect in JFT-2M. First, ferritic plates (FPs) were installed outside
on NSTX [224]. the vacuum vessel. Ferromagnetism of the FPs could be a
These results indicate that field ripple and neoclassical problem because of possible low-n error fields which might
viscosity effects play a role in braking the plasma whereas be induced. In order to examine the effects of error fields due
previous models for locked mode thresholds, e.g. [225], have to the FPs, experiments were performed with an unbalanced
only considered the EM torque applied in the vicinity of the placement of FPs, where 40% of the FPs contiguous in the
relevant resonant surface. toroidal direction were removed, to increase the error field of
the n = 1 mode artificially [231]. The error field with the
Effect of error field harmonic content. The poloidal mode unbalanced FPs was evaluated, and, for example, the magnetic
coupling effect means that n = 1 harmonics, other than m = 2, field from FPs magnetized by the vertical field was estimated
affect the EM torque at q = 2. The EM torque applied at other to be 1.5 mT. The evaluated error field was Br2,1 /Bt = (0.6–
resonant surfaces can also lead to a torque at q = 2 through 1) × 10−4 for Bt of 0.8–1.3 T. This value corresponds to the
perpendicular viscosity. Further, the neoclassical viscosity 20–40% of the critical value for the locked mode and so was
effects described above mean that non-resonant harmonics can not an operational issue for JFT-2M.
lead to a torque. In the second stage, the inside wall of the vacuum vessel
At the time of the IPB in 1999, experiments on was fully covered with the ferritic steel (termed a ‘ferritic inside
COMPASS-D [212] and DIII-D [215] had established wall’ (FIW)) as a simulation of the blanket wall of a reactor,
empirical expressions for the torque (and thus the locking in JFT-2M. To study possible error field effects the entire FIW
threshold) at q = 2 in terms of the m = 1–3, n = 1 harmonics. is assumed to be rigidly shifted by 20 mm, much more than
Subsequently the flexibility of varying the harmonic content the estimated actual shift of (∼3 mm). The induced error field
using the six recently installed in-vessel coils on DIII-D, due to the shift of 20 mm, using the 3-mode coupling formula
the I-coils, has been exploited to continue studies of error (equation (19)) is less than Bpen /Bt ∼ 4 × 10−4 . This is
field harmonic effects [226, 227]. In these studies it is lower than the allowable limit for an Ohmically heated plasma

S153
T.C. Hender et al

Bpen /Bt = 5 × 10−4 , and experimentally the region of locked

mode occurrence was not affected by the FIW. These tests thus
increase confidence that there are no unexpected error field
effects from ferritic materials, such as will be used in the test
blanket modules in ITER (see section 2.4.2).

Other issues. Initially as β is increased with NBI heating

the error threshold rises; this effect is studied in [218] where
it is shown at low-β that the error field threshold rise with
√
the beam induced rotation is best fitted as ω0 . In contrast,
equivalent plasmas where solely ICRF heating was used in JET
to increase β showed only a small increase in the error field
threshold [210]; though it should be noted that recent results
from TEXTOR do show an increase in the error field threshold
with ICRF heating power [217]. At higher β as the ideal
stability limit is approached the error field threshold decreases
with rising β, as reported in the IPB. It is now understood Figure 24. ITER CS and PF system with error field correction coils.
that as the ideal β-limit is approached in low collisionality
ELMy H-mode plasmas, m = 2, n = 1 NTMs are likely
99.9%. The contribution to error fields from the CS and PF
to be destabilized just below the ideal limit [232]. Studies
coils is approximately the same as it is from the TF coils.
are in progress on how the 2/1 NTM threshold is affected by
Error fields due to the CS, PF and TF coil joints, feeders
error fields or conversely how the error field threshold changes
and terminal regions were calculated using simplified models
as the 2/1 NTM limit is approached. Initial studies on DIII-
of the current-carrying parts of these elements. Each pair of
D [58] indicate that error fields can substantially lower the 2/1
TF coils gives an error field comparable to the error field from
NTM β-limit (by ∼ 40%), with error fields causing rotating
a PF coil. However, the symmetry of the TF coil systems
2/1 NTMs to be formed (since they rotate this eliminates the
provides a mutual compensation of the individual error fields
possibility that they are error field driven modes). Initial
analysis of similar studies on JET [58] indicates a reduction in generated by the coil pairs, so the total error field is negligibly
the error field threshold near the 2/1 NTM β-limit, but there low. The total ‘3-mode’ error field from the elements of CS
is no evidence of rotating modes being formed in this case. and PF coils is 3.0 × 10−5 of the toroidal magnetic field. The
Previous JET experiments [210] have shown that error-field- main contributors are the terminals, joints and feeders of six
induced modes can act as the seeds for 2/1 NTMs. CS modules.
Studies on Alcator C-Mod [216] have extended the range Six test blanket modules (TBM) are planned to be used in
of toroidal field over which the error-field-induced locked ITER. Five of them will have ferromagnetic structure material
modes have been studied to ∼8 T (i.e. encompassing the ITER (martensitic steel). The ‘3-mode’ error field from the five
range). This is important because the toroidal field scaling of TBMs is 3.1 × 10−5 . However, when this error field is
the error field locked mode threshold is one of the remaining superimposed on the error field from the coil joints, feeders
uncertainties. An identity experiment (same ρ ∗ , ν ∗ ) with JET and terminal regions, the total ‘3-mode’ error field is only
also shows good agreement on locked mode thresholds and 4.0 × 10−5 .
further studies are ongoing in this area [216]. There are two neutral beam (NB) injectors and one
diagnostic NB injector in the present design of ITER. To avoid
2.4.2. Error fields expected in ITER and their correction. deflection of the ion beam, the magnetic field inside the NB
injector should be very low during its operation. Each injector
Expected error fields. Analysis of ITER error fields and their has a magnetic field reduction system (MFRS) designed to
correction is reported in [233, 234]. The plasma corresponds reduce the tokamak poloidal field inside the injector to an
to the initial phase of current flattop in the 15 MA inductive acceptable level. The MFRS comprises active coils and passive
scenario (Scenario 2). ferromagnetic shields. The ‘3-mode’ error field due to the
Misalignments of current centrelines of the toroidal field NB MFRSs (two NB injectors and the diagnostic injector) in
(TF) coils, central solenoid (CS) and poloidal field (PF) Scenario 2 is about 1.1 × 10−5 .
coils, arising during the coils manufacture, installation and
assembly, are the main contributors to the error field. A Error field correction The correction coils (CC) system has
tolerance analysis was used to estimate the expected deviation been designed for ITER to reduce the error fields to an
of the coil centrelines from the nominal position and shape. acceptable level. The coils are shown in figure 24. The CC
The deviations are in the range 2–3 mm. The Monte Carlo system consists of three sets of superconducting coils (top, side
simulation, used for error field analysis, included 228 degrees and bottom) distributed poloidally to provide control of several
of freedom in deviation of the current centrelines. The m mode numbers. Each coil is connected to the toroidally
total number of simulations was 250 000, which guarantees opposite coil in series, to produce an n = 1 asymmetric mode
reasonable accuracy of the results. The expected ‘3-mode’ and to avoid coupling with axi-symmetrical magnetic fields.
error field (equation (19)) due to misalignment of the PF, CS Three power supplies are used for each CC set to provide
and TF coils is less than 11.9 × 10−5 with a probability of toroidal rotation of the correcting field.

S154
 Chapter 3: MHD stability, operational limits and disruptions

Table 4. Currents (kAt) required for correction of ITER error fields and the nominal currents.
Source of error field Top CC Side CC Bottom CC Comment
NBI MFRSs, 5TBMs and Coil joints, feeders, terminals 20.6 12.5 20.0 Correction to 0
TF, CS and PF Coils misalignment 103.6 104.2 160.1 Correction from 11.9 × 10−5 to 5 × 10−5
All sources 124.2 116.7 180.1 Simple summation
Nominal currents 140 280 180

The currents required for correction of the error field but discharges with high li have drawbacks for steady-state
expected at SOF are summarized in table 4. The table also operation. However, theoretical optimization studies [235]
shows the nominal currents in the CCs. The side CC have suggest the existence of a regime of moderate li , with q0 near
additionally about 160 kAt available for control of resistive unity and low central magnetic shear, more suitable for steady
wall modes. state while retaining some of the advantages of a peaked current
density profile. With a sufficiently broad pressure profile
2.5. MHD stability in advanced scenarios (pressure peaking factor p(0)/ < p >∼ 3) the beta limit
in such cases can be raised to βN = 3.5–4.0 without wall
2.5.1. MHD stability in plasmas with weak magnetic shear. stabilization, through strong plasma shaping with triangularity
At the time of the ITER Physics Basis [1], two types of up to 0.7. An additional feature of such a configuration is the
advanced scenario had been identified, associated with a relatively high-bootstrap current, with fbs up to 70% and good
centrally peaked current density profile (high li ) or a very broad alignment to the total plasma current density.
current density profile (reverse shear). The high-li scenario had Candidates have been developed recently for high-
good confinement and a high-beta limit (βN ∼ 4li , with li ∼ fluence or ‘hybrid’ operation in ITER, building on earlier
1.0–1.5), but because of high values of edge safety factor and experiments in stationary H-mode discharges with improved
poor alignment of the bootstrap current with the desired current core confinement [236–238]. These operating regimes
density profile, it was considered a less than optimum candidate have some features in common with the moderate-li regime
for steady-state operation. Reverse shear plasmas generally described above. (Hybrid mode operation is discussed in more
have better bootstrap current alignment for steady state, but detail in chapter 6 of this issue [2].) Typically, a stationary
somewhat lower li (∼0.7–0.8) and correspondingly lower β- q profile with low central shear is maintained with q(0) and
limits without wall stabilization. Furthermore, discharges with qmin just above unity, but with very small sawtooth oscillations
strongly reversed central magnetic shear were often observed or no sawteeth at all [99, 116, 239, 240]. The effective beta
to form a strong internal transport barrier, leading to strong limit is set by the 2/1 tearing mode. However, in the absence of
local pressure gradients which can further reduce the beta limit sawteeth this mode is encountered only when beta approaches
to βN ∼ 2. The limiting MHD mechanism in this case was the no-wall ideal kink mode limit at βN
3, a significantly
an ideal instability, the infernal mode, with its mode structure larger beta than the typical threshold for 2/1 tearing modes in
having a maximum near the region of minimum safety factor the presence of sawteeth. Avoidance of peaking of the current
(qmin ). (In addition to these β-limiting ideal instabilities, in density profile leading to sawteeth is related to the presence of
some cases, resistive neoclassical tearing modes were observed a benign 3/2 tearing mode [99,240,241] or in some cases a 1/1
in regions of positive shear. TAE instabilities have also been fishbone instability [116, 242]. This has allowed, for example
observed.) By avoiding a strong internal transport barrier, a quasi-stationary equilibrium with qmin > 1 to be maintained
a high-β scenario with weak central shear, broad pressure at βN ∼ 3 for over 6 s in DIII-D (see [241] and chapter 6 of
profile and wall stabilization of the n = 1 kink mode was this issue [2]).
identified experimentally. In this scenario, βN ∼ 4.5 was The strong internal transport barriers associated with
roughly 1.3 times the free boundary ideal MHD limit. This strongly negative magnetic shear can also occur in discharges
case was considered a promising steady-state candidate, but at with weak central shear. In such a case, the main stability
that time it was uncertain whether wall stabilization could be limitation is set by the n = 1 global kink, driven by the
sustained in long pulses (see section 2.3.2). combination of a peaked pressure profile and a weak shear.
Recent experiments and modelling have extended the Typical beta limits of βN ∼ 2, set by a fast-growing n = 1
performance and the understanding of stability limits in the mode, have been observed in many experiments [243–246].
‘high-li ’ and ‘reverse shear’ regimes, and plans for ITER now Analysis of experiments on JT-60U [244, 247] confirm that
include advanced scenarios related to these two regimes. In when qmin has a rational value, a variety of MHD events
this section progress in the macroscopic stability of advanced are triggered, including partial beta collapse, disruptions and
regimes with weak central magnetic shear is discussed. The energetic particle driven instabilities. Partial collapse is
case with strongly negative shear is addressed in section 2.5.2. associated with localized spikes in the pressure gradient, and
(Here central magnetic shear refers to the shear between the disruptions occur when there is a strong coupling to the edge,
magnetic axis and qmin ; weak shear scenarios typically have caused either by a low shear or a broader pressure profile [76].
q0 (safety factor on-axis) in the range 1 < q0 < 2, with qmin Experiments have shown clearly that the beta limit due to
less than or equal to q0 .) the n = 1 ideal kink mode can be significantly improved with
Discharges with centrally peaked current density profile a broader pressure profile [243, 245] in good agreement with
are of interest because of their potential for high confinement ideal MHD stability modelling predictions [245, 248]. The
and high-beta limits without the need for wall stabilization, transition from an L-mode to an H-mode edge is an effective

S155
T.C. Hender et al

means of broadening the pressure profile, but edge-localized

4.0 Broadened P
modes (ELMs) that occur during H-mode often degrade the Modelled
internal transport barrier [249]. Some experiments have n=1 no wall
reported a role for energetic particle driven modes, which may stability limit Transient
either trigger formation of an internal transport barrier [246] 3.5
or trigger a transition to the H-mode [245]. With a broad H-
Maximum
mode edge pressure profile and wall stabilization, the n = 1 experimental βN
kink stability limit increases to βN ∼ 4, and in some cases Long duration

βN
3.0
significantly greater.
A promising related scenario is the quiescent double
barrier (QDB) mode, achieved with counter neutral beam
injection [250–252]. These discharges are in the H-mode but 2.5 Measured
without the deleterious ELMs, allowing an internal transport no-wall βN limit
barrier to be maintained. Regulation of the heating power
allows the volume within the internal transport barrier to be
expanded slowly, without creating local pressure gradients 2.0
1.5 2.0 2.5
strong enough to destabilize the n = 1 kink mode [250]. In
q min
these discharges, normalized beta up to βN ∼ 2.9 is achieved.
Full characterization of the global stability limits of the QDB Figure 25. The measured and modelled dependence of the βN limit
mode remains an open issue. on the minimum safety factor in DIII-D. Squares are the
Stability modelling and experiments since the ITER experimentally measured no-wall limit and triangles are the n = 1
Physics Basis [1] have influenced the redesign of ITER with no-wall limit calculated for equilibria with profiles similar, but not
identical, to the profiles in the experimental discharges. Diamonds
higher triangularity. In general, experimental results are in are the maximum βN at which discharges have been operated for the
good agreement with predictions that the n = 1 ideal kink duration of the machine pulse without significant instability, circles
stability limit depends strongly on the discharge shaping, the are the maximum βN that have been obtained in steady-state
current density profile, the pressure profile and the presence scenario discharges. The cross represents the maximum βN obtained
of a conducting wall, often in a synergistic way [248]. without adding extra gas to broaden the pressure profile. (Reprinted
with permission from [255]. Copyright 2005, American Institute of
Numerous modelling efforts have shown that the βN limit Physics.)
improves with triangularity, but only when given sufficient
elongation [248, 253] and a sufficiently broad pressure profile
to strongly peaked pressure profiles and drive ideal kink
[248]. Experimental results [254,255] are consistent with these
instabilities which result in a disruption in JT-60U [174, 257],
predictions. As discussed above, the βN limit varies inversely
JET [245, 258], DIII-D [259] and ASDEX Upgrade [260].
with the peaking of the pressure profile in both modelling and
Unique to the reversed shear regime is a dominantly n =
experiment [191, 243, 245]. Very high values of βN have been
1 mode, which has multiple harmonics. This mode is a
achieved in recent spherical tokamak experiments [191, 256],
seemingly common limit to achieving the highest performance
consistent with modelling predictions on the βN variation with
plasmas. Ideal MHD stability analyses revealed that the
aspect ratio [253].
upper limit of the achievable βN is in agreement with the
The weak shear or reverse shear regime with low li
calculated ideal stability limit of the n = 1 kink-ballooning
and qmin > 1 which is proposed for steady-state advanced
mode [245, 257–259, 274]. The region of major collapses, or
scenarios in ITER is likely to require wall stabilization. disruptions, in reversed shear discharges in JT-60U is shown
In such discharges the experimentally observed no-wall βN on the βN − q∗ plane in figure 26. Here, q ∗ is the safety
limit decreases with increasing qmin [255], consistent with factor near the edge defined as q ∗ = (5a 2 BT /2RIp )[1 +
modelling predictions (see figure 25). Experiments are also κ 2 (1 + 2δ 2 − 1.2δ 3 )], here κ and δ are the elongation and the
consistent with the prediction that the beta limit can be triangularity of the plasma poloidal cross-section, respectively.
significantly increased by wall stabilization, particularly in The phenomenology of these disruptions is identical to those
the case of a broad current density profile (high qmin ) and a found in low shear advanced regimes.
broad pressure profile. Although the cases shown in figure 25 In DIII-D, an L-mode negative central shear (NCS)
with beta well above the no-wall beta limit are transient, other discharge exhibited a strong ITB and continually increasing
experiments have demonstrated the feasibility of long pulses pressure peaking. Early calculations suggested the MHD burst
with wall stabilization (See section 2.3.2). prior to the disruption correlated with a resistive interchange
instability; this instability was believed to reduce core rotation
2.5.2. MHD stability in plasmas with strong negative thereby triggering a more global double tearing mode that
magnetic shear. Reversed shear discharges are characterized was thought responsible for the actual disruption. Re-analysis
by negative magnetic shear in the core region, with the internal of the discharge just before disruption, however, found that
transport barrier near the minimum q region. Therefore, a a small (∼15%) steepening of the pressure gradient inside
large pressure gradient exists in a weak or negative magnetic ρ ∼ 0.3 can lower the predicted ideal beta limit to that of
shear region, and a large current density can exist near the the observed disruption limit, implying that the disruption can
plasma edge. These characteristics often cause harmful MHD be consistent in this case with an ideal instability [259]. A
instabilities over a wide range of βN . It has been observed new quasilinear ideal MHD stability model [261], based on the
in reversed shear discharges that the resulting ITBs can lead evolution of a discharge through an ideal stability boundary,

S156
 Chapter 3: MHD stability, operational limits and disruptions

2.0
1 7
1.0
B (b)
p 6

Z (m)
A
0
5
-1.0 0.5
4
-2.0 (a)
B: d/a<
2.5 3.5 4.5 q
3 R (m) 1.3 3
n=1 2
0
0 0.5 1
ρ
βN (%mT/MA)

2 1
A: d/a~ (c) m=3 m=4
2.0

ρξρ
1 m=5
Reversed shear m=6
discharges
no clear precursor m=7
with precursor
0 0
1 2 3 4 0 0.5 1
q min ρ (=√ψ)

Figure 26. (a) Region of achieved βN just before major collapse in the reversed shear discharges with d/a ∼ 2.0 in JT-60U. The solid line
indicates the computed stability boundary for the n = 1 kink-ballooning mode with the wall at infinity. An open square shows data with
d/a ∼ 1.3 and the hatched region indicate the limit in the discharges with d/a < 1.3. Radial profiles of (b) the pressure p and the safety
factor q and (c) eigenfunctions of ideal n = 1 modes are for a discharge indicated as a square in (a). Reprinted with permission from [279].

(a) 15
1000
λw = 1.5 p0 /<p> =2.4
Mirnov δ B
(Gauss)

SXR
δB

p0 /<p> =2.4
10
100
βN

p0 /<p> =4.8
30
1681 1682 1683
5
Time (ms) p0 /<p> = 4.8

Figure 27. Time evolution of Mirnov δB (open circles) and SXR

(filled circles) signals during the disruption in DIII-D showing the
exp[(t/t)3/2 ] dependence, from [259].
0
1 2 5 10
then predicts quantitatively the non-exponential growth of the Rwall / Rplasma
SXR and the Mirnov signals preceding the disruption, as shown
Figure 28. Dependence of the βN limit on the conformal wall
in figure 27 [259]. The final disruption therefore appears to position λw = Rwall /Rplasma for the four extreme cases with circular
be quite consistent with the onset of an ideal instability, in cross-section peaked (dotted line) and broad (solid line) pressure
terms of growth rate and mode structure. Further in ASDEX profile and similarly for the D-shaped with κ = 1.8 and δ = 0.7.
Upgrade, the ideal kink eigenfunction from the CASTOR code The scale for λw is logarithmic. The nominal wall position at
was found to be in good agreement with the experimental one λw = 1.5 used for most of the calculations is indicated by the
vertical line [248].
derived from ECE just before the disruption [260].
Optimization of profiles of pressure, q, shaping and wall
position is a key factor for improving the beta limit. Systematic and highly shaped) cross sections and two sets of pressure
calculations by varying the pressure peaking, the cross-section profile. Experimentally, an H-mode transition is used in DIII-
shape and the wall position showed clearly that there is a D [262] and JT-60U [257] to broaden the pressure profile and
significant gain in all the figures of merit related to β, from this results in a significantly increased βN limit, consistent with
the broadening of the pressure profile and strong shaping the systematic study and specific numerical calculation [265].
[248,262,263]. Stabilization by pressure broadening is similar The effect of the edge current on global stability was also
to that in normal shear plasmas [264]; however, the gain in investigated for reversed shear discharges. The edge current
the βN limit from a broad pressure profile was synergistic density was shown to play an important role in coupling the
with additional large gains from strong shaping and wall internal ideal low n kink modes to the plasma edge, thus
stabilization. The effect from all three is multiplicative instead decreasing the ideal stability limit significantly [244]. The
of additive. This is summarized in figure 28 showing the βN operational βN limit in DIII-D has been found to increase with
limit as a function of the wall position for two extreme (circular the shape factor from increased elongation and triangularity

S157
T.C. Hender et al

for high-performance discharges with weak or intermediate along the Z-direction. The equilibrium is stabilized by the
shear [239, 254, 266]. This is consistent qualitatively with the elongation coils when the current in the ATMI region is limited
systematic stability calculations [248]. Higher elongation is to be small. The idea was numerically confirmed in [274].
beneficial to all modes, while the combination of triangularity The linear analysis of MHD instability due to the negative
and outer squareness has a varying effect [267]. toroidal current in the current hole shows that m/n = 1/0
Localized minor collapses were observed near the resistive interchange modes with the growth rate ∼ η1/3
transport barrier in JT-60U and TFTR [247] and q > 1 are unconditionally unstable, whereas n = 0, m > 1
sawtooth-like events [258] in JET. The barrier localized mode tearing instabilities occur if  > 0 [275, 276]. Nonlinear
(BLM) [244] was first observed in the high-βp mode discharges simulations using the reduced MHD equations, including off-
in JT-60U, and the BLM was also observed in reversed shear axis current drive, show that the n = 0 activity leads to cyclic
discharges which was identified as an ideal MHD n = 1 reconnection events which clamp the central current density
instability [247]. The BLM does not degrade the improved at approximately zero. Simulations in full toroidal geometry
confinement significantly but rather relaxes the strong ITB. using the M3D code [277] also show a similar picture of
Theoretical models are used to identify the dependence of the cyclic n = 0 reconnection events, though these should be
instability on the salient features of the plasma profiles; these interpreted as being due to an axisymmetric equilibria not
are found to be the local pressure gradient, local shear in the existing in this case [274]. As yet there is no direct observation
q profile and the proximity of rational surfaces. The mode of the MHD events causing the current clamping to zero. It is
width is shown to depend on the value of q where the pressure also important to clarify the beta limit for the strongly hollow
gradient is largest—it increases as q decreases. Sawtooth-like current profiles due to the global MHD instability. The stability
events with q > 1 occur most commonly in the lower hybrid analysis of ideal MHD modes shows a possibility of significant
prelude phase, but can also persist into the main NBI/ICRF improvement in the beta by adjusting the pressure profile [278].
heating phase in JET [245, 258]. The sawteeth are associated
with transient collapses of the ITB and are visible in the 2.5.3. Expectations for ITER. Ideal external kink modes
electron temperature during the LH prelude phase. In general, coupled to ‘infernal’ modes due to the presence of low shear
no precursor activity is observed before these sawteeth, but regions inside the plasma may limit the value of βN in ITER
n = 1 post-cursors to the sawteeth are quite common and advanced scenarios. The steady-state (SS) operation with
generally show no phase inversion of the electron temperature Q > 5 requires a high βN > 2.5, low internal inductance
oscillations with radius, suggesting an ideal instability [258]. li ∼ 0.5–0.7, high fraction of the bootstrap current (∼50%),
Other instabilities, including ‘snakes’ at the outer q = 3 off-axis current density maximum and reversed magnetic shear
surface, are also observed to limit the performance of reversed (RS) [280]. The required βN > 2.5 is close to or above the
magnetic shear ITB regimes. upper semi-empirical estimate of the no-wall stability limit
MHD modes such as double tearing modes (DTMs) or a βNno-wall / li = 4 [1] for this range of parameters (and slightly
mode coupling between two resistive modes on the inner and exceeds the calculated values), see table 3, section 2.3.3.
outer mode rational surfaces may also cause a major collapse Although the profile optimization for RS tokamak equilibria
[174,257,268]. On ASDEX Upgrade, a DTM occurring during would allow for much higher values of βNno-wall in ITER [248,
the current ramp-up in strongly reversed shear scenarios led to 281, 282], realistic scenarios will restrict the accessible profile
a transient loss of the ITB. This could be avoided by additional family. Therefore, resistive wall mode (RWM) stabilization is
ECR heating, which, according to modelling, prevents DTM required (see section 2.3).
formation due to increased conductivity [260]. The SS scenarios in ITER were investigated in [203]
Nonlinear extended MHD simulations were performed with plasma transport simulated by the ASTRA transport
using the NFTC code [269] to resolve the cause and the role of code [283]. The ideal MHD stability analysis was performed
the early MHD bursts. These found a linearly unstable resistive for external kink modes using the KINX code assuming a
interchange mode at the inner resonant surface of the negative separatrix at the plasma boundary [284]. The RS scenarios
shear region and a double tearing mode, but the latter was with different current profiles were provided by the variation
stabilized by rotation. Without additional drive, the resistive of the lower hybrid (LH) current drive and the auxiliary heating
interchange modes were found to saturate at low amplitudes, and current drive (CD) by the neutral beam (NB) injection. The
but they can be sustained in a metastable state. internal transport barrier (ITB) was modelled by a fast drop
Motional Stark effect measurements of the poloidal of transport coefficients to the level of the ion neoclassical
magnetic field in JT-60U [270] and JET [271] show that with diffusivity in the reversed shear zone. Since the neoclassical
large local bootstrap current or off-axis LHCD, the central diffusivity in the RS zone increases with the safety factor,
current density reduces to near zero. Theoretical work has the increase in q0 and qmin (qmin = 2.12–2.43) reduces the
progressed on understanding the current hole equilibria. The pressure peaking factor (PPF) p0 / < p > (the ratio of central
standard equilibrium theory was re-examined and extended to to average pressure) from 3.1 to 2.7 (figure 29). The reduction
include equilibria with a central current hole with no toroidal of PPF together with the deviation of qmin from 2 improves the
or poloidal current [272]. It was shown that equilibria with a stability, although it reduces the power multiplication factor Q.
central current hole can exist. One possible explanation for the For the considered SS scenarios the possible operational limits
equilibrium of a current hole is the ‘axisymmetric tri-magnetic- were determined by rescaling the pressure profile keeping the
islands (ATMI) equilibrium’, as proposed in [273] which has current density and safety factor profiles unchanged. For such
three islands along the R direction (a central-negative current equilibrium series the values of the limiting wall position
island and two side-positive-current islands) and two x-points conformal to the plasma boundary for n = 1 mode were

S158
 Chapter 3: MHD stability, operational limits and disruptions

1.5 6 4.5

5
4
1.0 4
3.5
P/P01

βN
0.5 2 3

1
2.5
0.0 0
0 0.2 0.4 0.6 0.8 1
2
X

Figure 29. Normalized pressure and safety factor profiles for SS

operational scenarios: solid line (N1): qmin = 2.43, βNSS = 2.56, 1.5
0.3 0.4 0.5 0.6
dashed line (N2): qmin = 2.25, βNSS = 2.7, dotted line (N3):
qmin = 2.12, βNSS = 2.82; p01 corresponds to central pressure in the
1/q min
scenario N1 [203].
Figure 31. Limiting βN versus the minimal safety factor value for
the equilibrium series related to a full-bore scenario with a = 2 m.
3 2 1 Open symbols correspond to βNno-wall , solid symbols to βNITER-wall ;
p0 /p = 2.55, li = 0.58.
2
1.0 10
1.9 3
jtot 8
1.8 0.8

jnb
1.7 0.6 6
j, MA/m 2

1
aw/a

q
1.6

q
a w , I T E R/a 0.4 4
1.5 jbs
2 0.2 2
1.4
jec
1.3 0.0 0
2 2.5 3 3.5 4 4.5 0 0.2 0.4 0.6 0.8 1
βN X

Figure 32. Current density and safety factor profiles for the SS
Figure 30. Stabilizing wall position aw /a versus normalized beta equilibrium with βNSS = 2.67, q0 /qmin /q95 = 1.6/1.45/5.25,
βN or q = const scan of SS operational points N1,2,3. The no-wall p0 /p = 3.5, li = 0.8.
limits are shown by vertical dashed lines [286].

lower qmin , the wall stabilization becomes stronger for higher

calculated (figure 30). The stability calculations showed IN . This can be explained by the fact that the mode structure
that the stabilizing influence of the ITER vacuum vessel on changes with increasingIN : the mode becomes more extended
the n=1 mode is well described by the conformal wall with to the low-field-side and couples more strongly to the wall. For
awITER /a = 1.375 for plasma minor radius a = 1.85 m (plasma the lowest value of qmin = 1.6 the increase in βNITER-wall = 4.4
centre R = 6.35 m). (βNno-wall = 2.0) corresponds to Cβ = 0.67 provided that the
A full-bore scenario with Q = 8 was also considered steady-state value βNSS is preserved.
(a = 2 m, R = 6.2 m, awITER /a = 1.35) with higher βNSS = 3.6 Recent modelling of the SS scenarios with higher internal
at the SS operational point. Due the smaller the plasma– inductance li ≈0.8 and lower qmin ≈ 1.5 resulted in safety
wall distance and the lower value of PPF p0 / < p >= 2.55 factor profiles with weakly reversed shear in the plasma
(qmin = 2.16) the value of βNITER-wall = 3.8 is also center [285] (figure 32). As expected, it gives an increase
higher in this case (βNno-wall = 2.7). It gives quite a high in the βN limits against an external n=1 mode (βNno-wall ≈ 3),
value of Cβ = (βN − βNno-wall )/(βNITER-wall − βNno-wall ), which but this configuration is unstable against the n = 2 mode once
determines a degree of RWM instability (see section 2.3.3): the resonant surface q = 1.5 is in plasma. The n = 2 RWM
Cβ = 0.82 for βNSS = 3.6. The stability study for different stabilization is an issue and NTM mode stabilization can be
values of normalized current IN was performed in [207] using also required for such profiles.
the profiles from the transport calculations. The results in the In summary, the SS operation in ITER requires RWM
(βN , 1/qmin ) parameter plane (figure 31, IN ≈ 2.5/qmin ) show stabilization for the scenarios with qmin > 2. The
that despite an overall decrease in the βNno-wall values along with SS operational point with relatively low pressure peaking

S159
T.C. Hender et al

p0 / < p >≈ 2.5 should be considered to provide a wider Neoclassical tearing modes represent a significant
range of permitted pressure excursions βNITER-wall /βNSS ≈ 1.5. challenge for present tokamaks and future burning plasmas,
In the scenarios with higher li , the values of βNno-wall /βNSS ≈ 1.2 but the tools now appear to be at hand to control or suppress
can be expected even with higher PPF and without wall these instabilities. Recent experimental studies and simple
stabilization provided that qmin > 1.5. In both cases, scaling models based on modified Rutherford theory indicate
sufficiently accurate profile control would be desirable to avoid that the critical normalized beta above which NTMs can
shrinking of the operational space. potentially be destabilized scales approximately linearly with
the normalized ion gyroradius ρi∗ = ρi /a. This scaling,
2.6. Summary although unfavourable for large burning plasma experiments,
does not directly imply lower NTM beta limits, since the
Recent theoretical and experimental research has made scaling and nature of the NTM seeding mechanisms is the
important advances in both understanding and control of key factor and this issue is not yet fully resolved. In any
MHD stability in tokamak plasmas. Central m/n = 1/1 event, recent experiments have been quite successful in using
sawteeth are anticipated in ITER, but the tools exist to avoid localized electron cyclotron current drive to stabilize the
or control this instability through localized current drive or m/n = 3/2 and 2/1 NTMs, allowing beta to be raised above the
fast ion generation. Active control of other MHD instabilities threshold value observed in the absence of current drive. Other
will most likely be required in ITER. Extrapolation from
promising techniques for avoidance or mitigation of NTMs
existing experiments indicates that stabilization of neoclassical
include the control of sawteeth (and hence the generation of
tearing modes by highly localized, feedback-controlled current
seed islands), the control of the plasma shape and profiles,
drive should be possible in ITER. Resistive wall modes are
elimination of low-order rational surfaces (e.g. 3/2 and 2/1)
a key issue for advanced scenarios, but again extrapolation
from the plasma and operation in the ‘frequently interrupted
from existing experiments indicates that these modes can be
regime’ where nonlinear coupling by other MHD modes limits
stabilized by a combination of plasma rotation and direct
the amplitude of the longer wavelength, more deleterious
feedback control with non-axisymmetric coils. Reduction of
NTMs. Given an appropriate launcher and sufficient power
error fields is a requirement for avoiding non-rotating magnetic
for electron cyclotron current drive, it should be possible in
island formation and for maintaining plasma rotation to help
ITER to keep NTMs at a small amplitude or to stabilize them
stabilize resistive wall modes. Recent experiments have shown
completely.
the feasibility of reducing error fields to an acceptable level
Advanced scenarios in present tokamaks and in ITER
by means of non-axisymmetric coils, possibly controlled by
feedback. The MHD stability limits associated with advanced often rely on wall stabilization of the n = 1 kink mode at
scenarios are becoming well understood theoretically and can high beta. In the presence of a wall with finite conductivity, an
be extended by the tailoring of the pressure and current density external kink that would be stabilized by a perfectly conducting
profiles as well as by the other techniques mentioned here. wall instead becomes a slow-growing resistive wall mode.
The m = 1 internal kink instability is an important In present experiments the resistive wall mode can often be
issue for present tokamak experiments and future burning stabilized by plasma flow (at resonant surfaces such as q = 2
plasmas. A large sawtooth crash can trigger MHD instabilities or q = 3) on the order of a per cent of the Alfvén speed
such as the neoclassical tearing mode, leading to confinement or less [154, 155]. This result is in reasonable agreement
degradation, locked modes and/or disruptions. Therefore, with theoretical predictions, although precise predictions are
it is important to develop methods for controlling sawteeth sensitive to the plasma profiles and the choice of model for
in present and future experiments. A wide range of dissipation in the plasma. Optimized correction of magnetic
techniques have been demonstrated for eliminating sawteeth or field errors can help in maintaining the plasma rotation.
reducing their impact, including stabilization by fast particles, However, a burning plasma will have little external torque
modification of the current density profile with localized applied to it, and the rotation rate is likely to be marginal at best.
current drive or stimulation of small amplitude sawteeth with Recent experiments have shown success in active feedback
local heating or current drive. Neutral beam heating, ion control of the resistive wall mode, using non-axisymmetric
cyclotron heating, electron cyclotron current drive and lower coils both external and internal to the resistive wall, and
hybrid current drive have all been used for this purpose. also show that a sub threshold rotation can be combined
Of particular interest to burning plasmas is the effect of a with feedback stabilization for increased stability. Modelling
significant population of fast alpha particles on the stability predicts, and early experimental results confirm, that internal
properties of the m = 1 mode and the redistribution of the coils are capable of stabilization at higher beta. Modelling
alpha population following a sawtooth crash. Although a indicates that the modest improvement in beta over the no-wall
detailed physics understanding of the m = 1 mode remains an limit needed for advanced scenarios in ITER can be realized
open subject, a predictive capability is emerging in the form of with feedback stabilization using the external error correction
models for the triggering of m = 1 instabilities that are based coils. However, a significant reduction in the current and the
on MHD stability with kinetic effects of thermal and fast ion voltage of the control coils, as well as a greater improvement
populations. These models are coupled with transport codes in beta, could be achieved with internal coils.
to predict the effects of sawteeth on the total stored energy and Recent theories and experiments have emphasized the
fusion power. The results, benchmarked to experimental data, need to reduce magnetic field errors. Relatively small non-
suggest that in a burning plasma such as ITER sawteeth have axisymmetric magnetic fields having components that are
little effect on the total stored energy or time-averaged fusion resonant with low-order rational surfaces in the plasma can
power. slow plasma rotation, cause locked modes or generate seed

S160
 Chapter 3: MHD stability, operational limits and disruptions

islands for neoclassical tearing mode growth. Slowly rotating inherent limits on the thermal energy handling capabilities
plasmas are particularly susceptible to the penetration of of materials available for plasma-facing component (PFC)
field errors. Present experiments have been quite successful surfaces. On the structural engineering side, a burning-plasma-
at reducing the effects of error fields by the use of non- capable tokamak based on niobium–tin superconducting
axisymmetric correction coils, and such coils have become magnet technology allows a plasma-axis toroidal magnetic
a feature of most major tokamaks. Analytic estimates field of about 6 T, an increase in field relative to the 3–4 T
combined with scaling from present experiments give a low but fields routinely employed in the present generation of large,
achievable level of an acceptable error field for ITER. The error high-performance tokamaks. Further the associated increase
field limits become somewhat smaller for advanced scenarios in plasma current to about 15 MA brings with it a moderate
in order to avoid enhanced rotational drag from the ‘error field 2
(∼3) increase in the electromagnetic (EM) Bpol /2µo pressure
amplification effect,’ a resonant response by marginally stable loadings on the torus vacuum vessel (the comparison here
resistive wall modes in plasmas above the no-wall stability is with the corresponding magnetic pressure loadings in the
limit. However, reduction of error fields to a tolerable level present generation of large tokamaks, see table 5). While the
should still be achievable with the correction coil systems structural loading implications of these higher EM forces are
planned for ITER. non-trivial, acceptable engineering design solutions have been
The special features of advanced scenarios present specific identified for past [287] and present [288] ITER designs.
challenges to MHD stability, some of which have been Disruption-produced thermal energy loading on the PFC
mentioned above. Non-inductive operation without large surfaces of an ITER-class tokamak poses a less easily solved
recirculating power requires a large bootstrap current, which challenge [287]. Here the predicted time-normalized surface
in turn requires operation at lower plasma currents and higher energy loading (U /(A × t 0.5 ), where U is the deposited
normalized beta. The bootstrap current tends to produce energy, A is the area of deposit and t is the deposition
a broad current density profile with elevated central safety time) on the divertor targets will be high enough that some
factor, leading to small or negative magnetic shear in the core localized melting and/or vaporization of the affected surfaces
of the plasma. The broad current density profile requires (typically either W or C) will be likely. While the time-
wall stabilization of the external kink mode at high beta, as normalized energy loadings for the present 15 MA ITER design
discussed above. A central region of negative magnetic shear concept are appreciably lower than those predicted on the
is predicted to be stable to neoclassical tearing modes, but same basis for the 21 MA EDA design (see table 5), and
may be unstable to resistive interchange modes; experimental while, in any case, the depth of the melt- or vaporization-
results indicate that the interchange modes may place limits affected layer produced by a single disruption will be small
on the central peaking of pressure, the rotation and the safety (∼100 µm), the erosive effects of a series of unmitigated
factor. Double tearing modes are also a potential problem disruptions in the present ITER would still likely impact the
in discharges with negative central shear but are probably useable lifetime of the divertor PFC surfaces. Furthermore, the
stabilized by rotational shear in most present experiments. predicted sublimation of carbon following a disruption raises
A more serious issue is the ideal kink or the ‘infernal’ serious plasma operation concerns about de-conditioning of
mode which may be destabilized in a central region of weak the divertor and the torus PFC surfaces and tritium retention
shear and strong pressure gradient; internal transport barriers in co-deposited carbon layers. Accordingly, there will be
can contribute to strong local pressure gradients. Several a strong operational incentive to both reduce the frequency
experiments have observed disruptions resulting from such of occurrence of disruption and (at very least) reduce or
instabilities, and experiment and theory show that the beta limit ameliorate the thermal loading consequences of disruptions
varies inversely with the peaking of the pressure profile. The that cannot otherwise be avoided.
experiments also show that such disruptions can be avoided The production during disruptions of relativistic (run-
through the control of the pressure profile, by means of off- away) electrons poses a second type of threat to the integrity of
axis heating or by profile broadening with a controlled H-mode the ITER PFC surfaces. ITER, like any high-current reactor-
transition. Accordingly, it is expected that these instabilities regime tokamak, will be inherently susceptible to efficient
can be avoided in ITER through wall stabilization and control conversion, by Coulomb-collision avalanche multiplication,
of the pressure profile peaking. of plasma current to relativistic (runaway) electron current
In summary, recent research has deepened our understand- [289,290]. Such conversion, of up to 70% of the initial plasma
ing of MHD stability in tokamaks and has demonstrated ap- current, is predicted to occur following either a naturally occur-
proaches to avoidance or direct control of instabilities. Al- ring disruption or an artificially induced fast plasma shutdown.
though MHD stability remains an important issue for future The subsequent uncontrolled interaction of this magnitude of
experiments such as ITER, we can say with some confidence runaway current with PFC surfaces has the potential to produce
that MHD stability limits will not be a fundamental obstacle. local damage to PFC surfaces and their underlying substrate
structures. Accordingly, ITER (and future reactor tokamaks)
3. Disruptions will almost certainly need to have disruption mitigation and
fast-shutdown means that can simultaneously ameliorate dis-
Disruptions and their consequences pose significant design ruption thermal loading and runaway conversion.
and plasma operation challenges for reactor-regime tokamaks Table 5 compares various disruption-related parameters
in general and for ITER in particular. The magnitude and for JET (chosen here as a representative ‘large-tokamak’
scope of these challenges arise from a combination of physics, example), ITER [288] and the 21 MA ITER-EDA [287]
structural and thermal engineering considerations and from designs. The data in Table 5 are presented here to give a

S161
T.C. Hender et al

Table 5. Disruption and disruption consequences for JET, ITER and ITER-EDA.
Parameter JET ITER ITER-EDA Basis or comment
R (m) 2.9 6.2 8.14 Major radius
a (m) 0.95 2.0 2.8 Minor radius
κ 95 1.6 1.7 1.6 Vertical elongation
V (m3 ) 86 831 2000 Plasma volume
S (m2 ) 145 683 1200 Plasma surface area
BT (T) 3.45 5.35 5.68 Toroidal field
Ip (MA) 4.0 15 21 Plasma current
q95 3.0 3.0 3.0 Edge safety factor
Wmag (MJ) ∼11 395 1100 Poloidal field energy inside separatrix
Wth (MJ) ∼12 353 1070 βN = 2, with ‘ITER-like’ p(r) profiles
Magnetic and current quench related attributes
Bp  (T) 0.60 1.07 1.13 Average poloidal field
Bp 2 /2µo (MPa) 0.143 0.454 0.507 Torus vacuum vessel magnetic pressure
tCQ (ms) 9.4 35.6 65.7 Minimum current quench duration
BT ∗ dBp /dt (T2 s−1 ) 220 161 98 Relative force due to induced eddy currents
Wmag /(AFW ∗ tCQ
0.5
) (MJ m−2 s−0.5 ) 0.78 3.1 3.6 cf Be melt onset at ∼15 MJ m−2 s−0.5
Ihalo /Ip
0.45 (data)
0.4 (est)
0.4 (est) Halo current fraction
TPF
1.7 (data)
2 (est)
2 (est) Toroidal peaking factor
Thermal quench and divertor energy loading attributes
Adiv (m2 ) ∼1.6a ∼3.5 ∼4.6a Effective divertor target area, for H-mode
−2
UTQ = Wth /7Adiv (MJ m ) 1.07 14.1 33 For 7-x SOL expansion during-disruption TQ
tTQ (ms) 0.32 0.70 1.0 As per figure 54 of [1]
(0.5)
UTQ /tTQ (MJ m−2 ) 60 530 1040 C or W vapour/melt onset at 40–60 MJ m−2 s−0.5
Runaway electron conversion and mitigation attributes
Eint (V m−1 ) 38.3 38 28.8 In-plasma E-field
ne,RB (m−3 ) 4.2 × 1022 4.2 × 1022 3.2 × 1022 ne to suppress avalanche growth
Gavalanche 2.2 × 104 1.9 × 1016 6 × 1022 Coulomb avalanche gain = exp[2.5 × I (MA)]
IRA, seed (A) 90 4 × 10−10 1.8 × 10−16 Seed current for IRA = 0.5Ip
tfs (ms) 0.030 1.2 3.5 Minimum Wth shutdown time to avoid Be FW melt
a
Divertor area estimates for JET and ITER-EDA assume an R 1 scaling of Adiv .

comparative assessment of the relative ‘challenge’ of the three understanding the bounds on the maximum current quench
major ITER disruption consequence issues—EM loading, rate and the corresponding dBp /dt expected in the present
thermal loading and runaway conversion. The physics basis ITER design. New findings in this regard are described in
considerations that enter into table 5 have been extensively section 3.3.1.
detailed in chapter 3, section 4 (disruption and disruption The present status of the physics basis for VDEs and for
effects) of the ITER Physics Basis (IPB) [1]. Further estimating halo current magnitude and toroidal asymmetries
discussion of the present status of this physics basis and of new (typically described in terms of the toroidal peaking factor or
interpretations relative to what is detailed in the IPB follow in TPF) and the resulting vacuum vessel forces due to VDEs are
sections 3.1–3.6. described in section 3.3.2. The general conclusion reached
Table 5 demonstrates several of the points noted above: here and in [288] is that there has been little change in the
EM pressures on the ITER torus vacuum vessel associated physics basis status and the recommendations for ITER design
with the plasma current quench are about three times higher basis guidelines on the maximum halo current magnitude
than the corresponding pressures in JET (if JET had an ITER- (Ihalo,max /Ip0 ), TPF and the product (Ihalo,max /Ip0 ) × T P F .
like low-resistance vacuum vessel), and the local induced eddy The third current quench phase related subject, that
current forces on ITER first wall and RF antenna structures of runaway electron generation, amplification and loss (to
will be somewhat lower than the corresponding forces on PFC surfaces) is addressed in section 3.4. As the section
JET structures (of otherwise similar construction). Finally, details, in a high-current tokamak such as ITER, the effect
the thermal loading on the ITER first wall due to radiative of Coulomb avalanche multiplication is predicted to provide
dissipation of the plasma magnetic energy during the current a much stronger coupling between the effects of toroidal
quench phase is well below the W /(AFW × tCQ 0.5
) threshold plasma current decay and equilibrium dynamics (VDE) and
−2 −0.5 runaway production and ultimate loss to PFC surfaces.
(∼15 MJ m s ) that applies to the onset of surface melting
of beryllium. Accordingly, the conclusion originally given The surface-damage potential of interaction of a multi-MA
in the ITER Physics Basis [1] that the EM loading and runaway current with localized portions of the at-risk PFC
structural engineering and first-wall PFC challenges, posed by surfaces leads to serious concerns about the high levels of
the fastest-expected ITER current quench can be successfully runaway conversion following naturally occurring disruption
accommodated will continue to apply for the present 15 MA and intentional ‘fast-shutdown’ actions, intended to ameliorate
ITER design. Note, however, that the eddy current-induced disruptions.
loadings in ITER are now somewhat higher than those that are The three current quench related sections (3.3.1, 3.3.2 and
applied for the ITER-EDA. Hence, there is renewed interest in 3.4) are preceded by two preliminary sections. Section 3.1

S162
 Chapter 3: MHD stability, operational limits and disruptions

addresses both the overt causes for disruption and the resulting maximum total normalized plasma pressure (β) by three basic
internal MHD reconnection and flux surface destruction ‘operational limit’ considerations:
that leads to thermal quench, current quench and VDE
– Current ‘limit’: set by a requirement for a plasma edge
‘consequences’ of disruption. The subject of disruption
safety factor, q95
∼2.
frequency and ‘causes’, operational and otherwise, is also
– Density ‘limit’: set by a requirement that the plasma den-
addressed in this section.
sity should not appreciably exceed the empirical ‘Green-
Section 3.2 addresses the physics basis and consequences
wald’ density limit nGW (1020 m−3 ) = I (MA)/π a 2 (m2 ).
of the ‘thermal quench’ phase of a disruption. Here thermal
– Pressure ‘limit’: set by a requirement that the normalized
quench denotes the rapid internal redistribution and subsequent
volume-average toroidal beta, βN = β% /(I (MA)/a(m)
loss to PFC surfaces of the plasma thermal energy, Wth , that
B(T)), should not exceed the ‘Troyon’ ideal MHD beta-
occurs immediately before the onset of disruption. The well-
limit of approximately 3.5% MA m−1 T−1 .
known feature of this phase is the rapidity of the internal
redistribution and/or thermal energy loss, in times that are The plasma current and pressure limits reflect the
predicted to be as short as ∼1 ms for an ITER-class tokamak. immediate effects of onset of ideal MHD instability, whereas
The thermal-quench portion of table 5 summarizes some of the ‘Greenwald limit’ manifestation of the plasma density
the key attributes for present tokamaks and ITER designs. The limit, which appears to be the result, in part, of an underlying
table demonstrates that while time-normalized thermal quench deterioration of plasma energy and particle confinement, may
loadings for ITER are now significantly reduced relative to have a more indirect MHD origin that in turn leads to an MHD-
those projected on the same basis for ITER-EDA, it is still initiated disruption. The subject of plasma density limits has
likely that some divertor PFC surface melting or vaporization been discussed in detail in chapter 3 of IPB [1] and also,
might occur. However, the depth of the melt- or vaporization- with regard to the role of intermediate MHD instability, in
affected zone will be reduced relative to the depth projected the material that follows below.
for ITER-EDA, and hence the cumulative erosive effects of a Attempting plasma operation that approaches or exceeds
given frequency of unmitigated disruptions may be less PFC- the limiting values of any of these three operational boundary
lifetime limiting than was projected—on a worst-case basis— ‘limits’ typically initiates an increase in MHD activity that
for the ITER-EDA. Furthermore, new physics basis details on eventually results in a major disruption, wherein the tokamak
the time dynamics and spatial partitioning (among divertor and magnetic configuration becomes globally unstable to helical
other PFC surfaces) and the spatial spreading of plasma thermal perturbations of the form ζ (r,t) = ζ (r) exp i(mθ −nφ), where
energy deposition on the divertor surfaces during disruption θ and φ are the poloidal and toroidal angles and m and n are
suggest that melt- or vaporization-free divertor PFC operation the corresponding poloidal and toroidal integer mode numbers.
may be obtainable in at least some of the proposed ITER Once the final non-linear growth phase of this helical instability
plasma operation regimes. These important new findings about begins, global destruction of the nominal integrity of the nested
thermal quench characteristics and discussion of the resulting flux surfaces develops on a time scale that can be as short as
implications for the ITER divertor PFC response are given in 100 µs in the present generation of medium-size tokamaks.
section 3.2. Extrapolation (see chapter 3, section 4 of [1]) to ITER predicts
Section 3.5 discusses recent advances—in most cases, a corresponding time scale for thermal energy redistribution
using some degree of an integrated model—in the numerical within the plasma volume that is ∼1 ms.
simulation of ITER disruptions and VDEs, and their resulting As a result of this rapid growth of global MHD instability,
EM loading and runaway electron generation consequences. major disruptions in present tokamaks can expel most of the
Representative examples that contribute to reaching the plasma thermal energy on a time scale that is also typically
conclusions noted above about the structural integrity of ITER ∼100 µs. The ensuing increase in plasma resistivity that this
vacuum vessel and PFC substrate systems are presented. ‘thermal quench’ cooling produces then precipitates a rapid
Progress in incorporating more-fundamental MHD instability decay of the plasma current (typically described as a ‘current
considerations into these types of simulations is also quench’) and, in a vertically elongated tokamak, simultaneous
addressed. development of vertical instability, typically described as
Finally, section 3.6 takes up the closely-connected topics vertical displacement event (VDE) and sometimes also as a
of disruption prediction, avoidance, and mitigation means vertical disruption or vertically unstable disruption. Current
and their projection to ITER. The importance of suppressing quench and VDE development occur on a time scale that
runaway avalanching underlies such projection and sets rather is typically ∼5 ms in present medium-size tokamaks. The
stringent limits on whatever technique is to be implemented. corresponding current quench time scale extrapolated to ITER
Section 3.7 provides a summary of the present is about 35 ms.
understanding of the physics basis for disruption and disruption The rapid current quench that follows disruption also
avoidance and mitigation in ITER and gives recommendations gives rise to a high in-plasma electric field that can generate
as to needs for future physics and technology R&D. superthermal (runaway) electrons, with energies in the 10–
100 MeV range. In some cases, conversion of up to about 50%
3.1. Disruption characteristics, causes and frequency of the before-disruption plasma current to runaway current
is observed in present ‘large’ tokamaks (see, e.g. [291] and
It is well known that stable sustainable (disruption-free) section 3.4). Once created, the subsequent interaction of such
operation in a tokamak system is limited with regard to a ‘runaway discharge’ with the background thermal plasma
maximum plasma current, maximum electron density and is weak and the runaways typically remain well confined for

S163
T.C. Hender et al

as long as the plasma equilibrium is maintained. The eventual

decay of the plasma equilibrium leads to impact of the runaway #2841 Loop Voltage (measured)
electrons on the PFC surfaces. Such impact can cause localized 2
surface and substrate (below-surface) heating, vaporization 0
and thermal shock damage and, in metals, recrystallization
and crack development in the heat-affected zone. -2
The three operational limits (‘boundaries’) cited above -4
as defining the disruption-free operation domain for tokamak
-6
plasmas are indicative, but not definitive. This means that
while approaching or crossing one or more of these boundaries 0.8 Plasma current (MA)
will usually lead to onset of disruption, there do exist examples
0.6
where one or more of these ‘conventional’ operational
boundaries can apparently be exceeded, and conversely, other 0.4
examples where disruption occurs under conditions remote
from all three boundaries e.g. disruptions initiated by tearing 0.2
modes in high-q Ohmic ‘start-up’ plasmas (see chapter 8, 0.0
section 3.2 of [1]) or disruption initiated by neoclassical tearing
mode (NTM) onset in moderate-βN positive-shear H-mode 0.14 NPA Flux [3.61keV H] (arb)
0.12
plasmas (see section 2.2), or by onset of pressure-gradient- 0.10
driven modes in high-q ‘advanced-performance’ plasmas with 0.08
internal transport barriers (see section 2.5 and e.g. [245]). 0.06
The complexity of the underlying MHD stability and the 0.04
possibility for ‘secondary’ energy confinement and/or MHD 0.02
0.00
instabilities to in turn initiate the onset of primary ideal-MHD 0.10 0.15 0.20
instability and disruption, make disruption prediction a much time (s)
more challenging matter than simply evaluating proximity
of the plasma attributes to the three ‘limit’ parameters (see Figure 33. Edge loop voltage, plasma current and NPA flux at
section 3.6). 3.61 keV in a MAST discharge with internal reconnection events.
(Reprinted figure with permission from [296]. Copyright 2002 by
the American Physical Society.)
Mechanisms for major disruption. A major disruption is
similar to a large-scale interchange-type MHD instability
that simultaneously envelops both the plasma centre and its readily observed at the plasma edge. The presence of
periphery. This type of interchange can be explained in such a large toroidal voltage at the plasma centre has been
two ways: magnetic reconnection [292] and non-linear ideal- recently confirmed with non-magnetic evidence: figure 33
helical instability (cold bubble) in regions of low shear [293]. shows evidence for the toroidal acceleration of protons in the
Magnetic reconnection of plasma areas with different helicity centre of a MAST plasma during internal reconnection events
can occur during the development of non-linear kink and (partial disruptions) [296].
tearing modes (m = 1/n = 1, m = 2/n = 1). These modes A detailed account of the dynamics of major disruptions
result in convective and conductive plasma transport, which for plasmas close to the ideal MHD beta-limit can be found
leads to a rapid loss of thermal and magnetic energy in the in [297]. These dynamics include the development of NTMs
central area of the plasma column. This internal disruption and ballooning modes, followed by a deep ‘internal disruption’
is similar to a sawtooth oscillation (see section 2.1), but it and finally by a conventional major disruption (the typical
modifies the current distribution and magnetic shear not only two-step major disruption model cited in chapter 3, section 4
close to the radius of inversion (q(r) = 1), but over most (or of [1]). The outer shell of hot dense plasma between the
all) of the plasma radius. plasma edge and the zone of development of the sawtooth
In the ‘cold bubble’ model, helical flux tubes can internal disruption may play a role in screening interaction
be captured in the periphery and transported into the hot with external m = 2–4, n = 1 modes [298]. A numerical
centre. This interchange ultimately leads to an increase modelling effort devoted to investigate the non-linear evolution
in global resistivity and to the plasma current quench. of ballooning modes in high-temperature plasmas elucidates
The experimentally observed fast impurity transport during some of the MHD mixing mechanisms believed to be present.
sawteeth [294] is in agreement with this interchange-type The numerical modelling shows that convection cells develop
mechanism. between the hot central plasma and the edge in ridges whose
The thermal quench that this mixing precipitates is two-dimensional poloidal projection resembles fingers [299].
typically followed by a redistribution of toroidal current, As the plasma resistivity decreases (is reduced as a parameter
leading to a flattened current profile (and hence to a decrease in in the simulation), the number of fingers increases, while the
the dimensionless plasma internal inductance, li ). In order to width of each individual finger becomes narrower (figure 34).
conserve flux [295], the plasma current increases, producing Because of these effects, the rate at which energy is transported
a characteristic current overshoot or ‘spike’ (∼10% Ip0 for to the wall (inferred from the simulation) is unchanged as the
conventional aspect ratio tokamaks) and a large transient resistivity is reduced. The magnetic field lines are decoupled
negative toroidal loop voltage. This transient voltage is from the plasma even when the resistivity approaches zero,

S164
 Chapter 3: MHD stability, operational limits and disruptions

Disruption causes and frequency. As the discussion

presented above, and preceding discussions developed in
the IPB make clear, onset of rapidly growing global MHD
instability is always the penultimate cause of a major
disruption. However, from a plasma operation point of view,
there is also a proximate and usually clearly identifiable
precursor ‘cause’ for each disruption that occurs. Such causes
can be categorized either in terms of (1) the precursor MHD
instability or plasma energy confinement or energy balance
disturbance event that triggers the final global instability onset
or (2) the operational or tokamak system event that ‘causes’
the disruption. Categorization in terms of item (1) leads to
Figure 34. Temporal evolution of the pressure in the poloidal plane
at a fixed toroidal angle. The normalized resistivity used in the terminologies that include ‘density limit’ disruptions, ‘cold-
simulations is 3 × 10−4 (left) and 3 × 10−5 (right). (Reprinted with edge’ disruptions, ‘beta-limit’ disruptions (either from ideal
permission from [299]. Copyright 2001, American Institute of MHD or, more frequently in present experiments, from NTM
Physics.) or tearing mode growth and/or mode locking), and ‘internal
pressure-gradient’ (ITB) triggered disruptions. Since the
resulting in the plasma fingers being readily convected across time scale for confinement deterioration can approach the
the magnetic field. Disruptions at high-β have also been (degraded) energy confinement time scale, the pre-MHD-
simulated using the NIMROD code [300]. onset phase of these types of events are often described as
‘thermal collapse’. Categorization in terms of item (2) leads
This prediction of fast mixing even for a low-resistivity
to a long list of hardware and/or operation-associated causes
plasma is in contrast to the predictions of disruption models
that can include hardware ‘failures’, e.g. premature turn-off
based on magnetic stochasticity and/or magnetic reconnection.
of the Ohmic drive power supply or neutral beam heating,
In resistive MHD, the rate of reconnection of magnetic field
excessive gas fuelling input rates, release of impurities from
lines scales with the square root of the local resistivity (η) and PFCs owing to excessive power loading, debris falling from in-
there is a large reduction in reconnection rate for η magnitudes torus surfaces into the plasma, poor wall conditions (perhaps
that are characteristic of the initial (before-thermal quench) as the result of a previous disruption), plasma control system
phase of disruption. Therefore, reconnection models tend failure or inadvertent mis-programming (i.e. human error),
to predict reconnection times that increase with increased intentional ‘killer pellet’ or massive gas injection, and so forth.
plasma temperature and/or reduced η. In contrast, non- These lists can be quite detailed and hardware specific and are
linear simulations demonstrate that confinement is destroyed not necessary universal, in terms of categories and terminology,
by ballooning modes at a rate that is independent of the η across the present spectrum of operating tokamaks.
and on time scales that are commensurate with the sudden Most presently operating tokamaks now keep shot logs
loss of thermal energy observed during β-limit disruptions. that describe both disruption ‘type’ or physics cause(s) and
Although these results strictly apply only to ballooning modes also the underlying operations-related event that the operator
in a tokamak, these results may also be relevant to other MHD may identify as being the ‘cause’ of the disruption. These
instabilities involving convection cells [301]. Their inclusion logbook entries constitute a ‘database’ that can be used
in an ‘integrated’ numerical simulation model (see section 3.5) to compile ‘statistics’ that quantify disruption-type, causes
may ultimately provide a first-principle basis for modelling the (physics, operation intent and/or hardware and human factors),
onset and thermal quench phase of a major disruption. development sequences (e.g. ne-limit → thermal collapse →
A sudden internal degradation of the energy confinement— disruption) and frequencies of global and/or by-type or by-
a frequently observed feature of ‘density limit’ disruptions (e.g. category occurrence. Interpretation of such log-book data to
a disruption caused by proximity to the Greenwald limit and/or search for physics-based correlations for disruption likelihood
(‘disruptivity’) can, however, prove problematical. Figure 35
by an excessive plasma edge fuelling rate)—has now been ob-
shows one such long-term assessment for DIII-D: there is
served to initiate close to the q = 2 surface in small [302],
little evidence for any strong correlation of long-term average
medium [303] and large-sized [304] tokamaks. In each in-
disruptivity with any of the traditional ‘operation limit’ plasma
stance, a flattening of the electron temperature profile spreads
parameters [305]. The three-year average disruptivity (with
inwards from the outboard (large-R) O-point of the 2/1 mode.
obvious hardware failure causes screened out) is about 13%.
The plasma periphery cooling that this spreading causes ini- Statistical assessments of the same ∼3600 discharge data set
tiates destruction of the global energy confinement. The ob- shows that the per-unit-time disruptivity of long-pulse high-βN
servation of the same onset phenomena in three distinct-size discharges, that successfully reach a stable stationary condition
tokamak plasmas indicates that it may be an intrinsic charac- after about 4 s (from heating initiation), tends, within the
teristic of this type of disruption (perhaps with an underlying statistical accuracy possible in the 300-discharge dataset, to
MHD cause). An increase in density fluctuation levels and zero as the discharge duration increases towards 7 s (set by
an abrupt increase in the frequency of fluctuations were also hardware limits). This per-unit-time disruptivity decrease
observed to coincide with the onset of the electron tempera- rules out, on a statistical basis, that disruptions occurring
ture erosion. There are future prospects for incorporating such during this 4–7 s interval are precipitated by a random process.
considerations in models of the precursor and onset phases of Interpretation of these types of statistics on disruption
the major disruption. causes and occurrence frequency continues to be applied to

S165
T.C. Hender et al

3.0 and saturated 3/2 NTMs) for a flattop duration of more than 9 s,
n e /2 or about 9 current profile relaxation times (the relaxation-time-
2.5
q 95/2 equivalent duration in ITER would be ∼3600 s) [241]; see also
chapter 6 of this issue [2] for a fuller discussion of hybrid H-
βN modes. There is no indication that the duration of this type of
2.0
enhanced-performance discharge is limited by anything other
than the pulse-duration capabilities of the tokamak systems
1.5 needed to produce and sustain it.
Disruptivity
1.0 3.2. Thermal quench energy loss and deposition
li
Data averaged over 500-shot window
New divertor
hardware
The discussion and data presentation that follows focuses on
0.5
1998 1999
installed:
Carbon
2000
present understanding and estimates of the energy deposition
impurities
significantly
reduced.
magnitudes and time scales expected on the ITER divertor
0.0
95 000 97 000 99 000 1 01 000 1 03 000 1 05 000
and first-wall (FW) PFC surfaces owing to the effects of
Shot Number unmitigated disruptions. Figure 36 illustrates the major plasma
magnetic and thermal energy sources and sinks (deposition
Figure 35. Three-year ‘disruptivity’ history for DIII-D. Here locations) relevant to consideration of plasma energy flow and
disruptivity (D) denotes the per-shot likelihood of having a
disruption during the current flattop phase of the discharge. D = 1.0 deposit during disruption. Estimated magnitudes of the various
corresponds to 10% per-pulse likelihood. There is no evidence of a energies, deposition-affected areas and time scales are given
positive correlation of D with the long-term average values of βN , li in Table 6 (data from [307]). The purpose of introducing the
or q95 . The most evident effect is a long-term decrease in D that various considerations and symbolic definitions in figure 36 is
followed installation of a new set of divertor tiles. These tiles to provide a quantitative basis for evaluating present data and
provided reduced tile edge heating and hence reduced C impurity
generation during high-power plasma operation. While there may making predictions for ITER and beyond.
be evidence of a positive long-term correlation of D with ne (in units Figure 36, and the data in table 6, also indicate a number of
of 1019 m−3 ) the disruptivity data alone does not allow unambiguous important considerations that apply to estimates of the resulting
identification the underlying mechanism(s) responsible [305]. ITER divertor PFC and FW energy deposition magnitudes and
time scales. These considerations and the present experimental
projections of what may occur in ITER. These projections basis for figure 36 and the values in table 6, are discussed below.
are subject to caveats that what is seen in present tokamaks
vis-à-vis global frequency of disruption (typically about 10% Plasma energy magnitudes and thermal energies at disruption.
overall, in terms of the disruptions per shot) and the balance The plasma thermal energy, Wth and the in-plasma poloidal
between ‘physics’ and ‘external’ causes and the breakdowns magnetic energy Wmag for a representative Q = 10 full-
performance ITER ‘ELMy’ H-mode (Scenario 2) plasma are
within these categories are clearly influenced by both the
estimated to be about 350 MJ and 395 MJ, respectively [307].
nature of experimental programmes being conducted, the
The magnetic energy cited in table 6 does not include the
present reliability of hardware systems and the presently
additional magnetic energy, external to the separatrix but
limited ability of human operators to fully predict the
internal to the torus vacuum vessel that contributes to in-
outcome of experimental campaigns that intentionally push
vessel energy deposition during the current decay phase of
plasma operation close to the one or more of the many
a disruption (see section 3.3). This energy is estimated to be
operational limits inherent in achieving high-performance
∼200 MJ. While the time scales for the thermal quench and
plasma operation. In simple terms, this means that campaigns
current quench phases of an ITER disruption are predicted to
mounted to explore or expand operational boundaries tend to
be reasonably distinct (1–10 ms versus
35 ms), there is some
cause more frequent disruption. In the last regard, however,
potential for overlap and accumulation of deposited energies
there has been significant recent progress in providing tokamak
in the ITER divertor and FW surfaces. Concurrent thermal and
operators with ‘online’ impending disruption prediction data magnetic energy deposition on divertor surfaces is frequently
that can, in principle, be used to either avoid disruption or observed in present experiments (see below). Furthermore, as
to soften the consequences of disruptions that do occur e.g. table 6 indicates, in ITER, the incremental time-normalized
[306]. The subject of techniques for reliable a priori disruption first-wall energy deposition from the current quench phase is
prediction is addressed in section 3.6. not completely negligible (especially if spatial peaking factors
On the converse side, there is also increasing confidence and the ex-plasma magnetic energy, not included in the table 6
that, given achievement of a well-defined set of plasma data, are taken into account) and may be sufficient to bring
operation conditions and reproducible hardware and PFC/wall the surface temperature of the ITER first wall (beryllium)
conditions, there is little evidence for any secular tendency of up to or above melting by the end of the disruption. Hence
such plasmas to disrupt as the duration of the steady-state phase magnetic energy deposition during the current quench phase
is increased. Simply put, there is no evidence for any ‘random becomes a significant incremental consideration, especially in
trigger’ mechanism that will arise to cause a disruption in high-radiation-fraction thermal quench scenarios.
what should otherwise be a sustainable plasma discharge. The
‘hybrid H-mode’ discharges obtained in DIII-D provide an Pre-disruption thermal energy loss. The IPB thermal
example of this behaviour; these plasmas exhibit stationary deposition guidelines [1] and the estimates in table 6 were (are)
conditions (including stable repetitive sawteeth, Type I ELMs based on the conservative assumption that the plasma thermal

S166
 Chapter 3: MHD stability, operational limits and disruptions

Figure 36. Schematic representation of plasma thermal and magnetic energy flows, and deposition locations during disruption. The
possibility of before-disruption (pre-disrupt) loss of Wth on an energy confinement time scale (ε factor) is indicated (see text). The PF(FW,
core) and PF(FW, div) parameters indicate spatial peaking factors (PFs) relevant to radiative deposition on the overall first-wall surface
and/or near the divertor entrance and λ is the broadening parameter of the heat load on the divertor footprint area (Adiv ).

Table 6. Energy-related parameters for ITER disruption (Scenario 2: ELMy H-mode).

Parameter Symbol Value Basis or comment
Plasma magnetic energy Wmag 395 MJ From equilibrium model; does not include energy external
to separatrix
Plasma thermal energy Wth 353 MJ For Pfus = 400 MW, with H98(y, 2) = 1.0
Plasma surface area S 683 m2 At separatrix; FW area (excluding divertor entrance) is
similar
Active divertor area (normal operation) Adiv 3.5 m2 FDR physics document
Active divertor area (disruption) Adiv, dis 25 m2 ∼7Adiv (equivalent to λTQ = λCQ = 7)
Internal Wth relaxation time τ2 0.7 ms IPB scaling (∼R 1 )
Wth loss, Wdiv deposit time tdiv 1–10 ms (1.5–15) τ2 ; see text
Current quench time (minimum) tCQ 35 ms Based upon tCQ /S = 1.67 ms m2 ; see section 3.3.1
U /(A ∗ t 0.5 ) for divertor U ∗div 144–446 MJ m−2 s−0.5 100% Wth to divertor; tdep = tdiv (range is for 1–10 ms)
U /(A ∗ t 0.5 ) for FW U ∗FW 5.2–16.3 MJ m−2 s−0.5 100% Wth to FW; tdep = tdiv ; uniform deposition
U /(A ∗ t 0.5 ) for FW U ∗mag , FW 3.1 MJ m−2 s−0.5 100% Wmag to FW; tdep = tCQ ; uniform deposition

energy at the time of disruption is equal to the thermal energy 0.4

of the ‘maximum performance’ ‘parent’ plasma. For ITER, 0.35
this thermal energy (∼350 MJ) would be that of the high-gain
Relative Probability

0.3
(Q = 10) plasma needed to realize the ITER fusion power
0.25
and energy gain (Q) goals. A high-performance steady-state
plasma is expected to have a similar thermal energy. 0.2

The assumption of disruption at full thermal energy (i.e. 0.15

ε = 1 in figure 36) is not always borne out by what is seen 0.1
in present-day ‘high-performance’ experiments that attempt 0.05
to obtain maximum plasma thermal energy. For example, in
0
JET, for most ‘types’ of disruption, the plasma thermal energy 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
at the time of the thermal quench, Wth,dis , is substantially Wt.q./Wdiamax
less than the thermal energy (Wth,max ) of the parent high-
performance plasma state that the disruption stems from [308]. Figure 37. Distribution of pre-disruption thermal energy loss in a
With the exception of disruptions in JET arising during high- 129-example sample of disruptions of JET ‘high-energy’
plasmas [308]. The ratio of thermal energy at the start of the
βN ITB discharges and of disruptions that occur from the disruption Wtq (denoted Wth,dis in the text) to maximum energy in
effects of a pure ‘hot-plasma’ VDE, wherein loss of vertical max
the discharge Wdia (Wth,max in text) is plotted [308].
control allows a full-performance plasma to drift vertically
before the disruption occurs, see section 3.3.2, there can be
substantial (up to 90%) ‘thermal collapse’ energy loss before Figure 37 shows the distribution of before-disruption
the actual disruption occurs. In contrast, ITB and VDE energy loss observed in a representative sample (129 examples)
disruptions typically exhibit 10% before-disruption thermal of disruptions occurring in JET ‘high-energy’ (Wth,max

energy loss. 4.5 MJ) discharges. Approximately 80% of the total have

S167
T.C. Hender et al

Wth,dis /Wth,max  0.5. The median energy fraction is section 3.3.1). Examples of similarly prompt disruptions are
Wth,dis /Wth,max ∼
= 0.2. Only 13% of the sample (the ITB and also seen in DIII-D in positive-shear H-mode plasmas wherein
VDE cases noted above) comprise ‘full-energy’ disruptions NTM suppression or avoidance allows operation near or above
with 0.9  Wth,dis /Wth,max  1.0. While the correlation the ideal ‘no-wall’ MHD stability (Troyon) beta-limit.
of pre-disruption thermal loss with JET plasma operation
procedures has not yet been systematically examined, many Duration of the thermal quench. The thermal quench is a
of the examples with significant loss are obtained under phenomenon in which most of the plasma thermal energy is lost
conditions where ‘soft-stop’ disruption mitigation measures by the plasma and is subsequently deposited, by conduction
(decrease of auxiliary heating and elongation and/or plasma or/and convection and radiation, on the limiter and divertor
current ramp-down) are taken upon initial detection of MHD surfaces. The duration depends on the machine size; it
instability. Given that the delay in many of the ‘high- is of the order of tens of microseconds in small tokamaks,
energy‘ examples between onset of MHD instability or other hundreds of microseconds in medium-sized tokamaks and has
disturbance and disruption can be ∼1 s, these measures are been observed to reach a few ms in JET. The duration is
likely a contributing factor to the many low-fraction examples also observed to vary significantly within a given machine,
seen in figure 37. ‘Natural’ MHD instability induced thermal apparently depending on the ‘type’ of disruption and/or the
energy loss (confinement degradation) is undoubtedly also an nature and growth rate of the triggering MHD instability. There
important factor. is also evidence that the pre-disruption magnetic shear profile
Similar magnitudes and relative time scales of pre- (positive dq/dψ versus weak or strongly reversed dq/dψ) can
disruption Wth loss are seen in DIII-D and ASDEX Upgrade, affect the rapidity of the thermal quench duration.
often owing to the effect of slowly growing or saturated The PFC energy deposition phase of the thermal quench
NTMs, or, in DIII-D, owing to the effect of slowly growing is typically preceded by one or more internal energy
(marginally stabilized) RWMs. Deterioration or loss of redistribution phases, wherein a rapid redistribution of the
H-mode confinement following approach to the Greenwald energy within the plasma occurs. Cases exhibiting one, two or
density ‘limit’ can also produce a similar ‘slow’ confinement- three or more sequential redistributions are known. References
loss thermal collapse that precedes disruption. This type [1] and [309] describe in detail a typical two-stage thermal
of ‘density limit’ disruption, which typically results from quench. The thermal energy is firstly redistributed within
excessive plasma core and edge radiation, is often described the plasma inside the q = 2 surface; after a delay τ(1−2) the
as a ‘cold-edge’ or ‘radiation limit’ disruption. The relative energy barrier within the closed flux surface region breaks and
frequency of the occurrence of cold-edge disruptions preceded the energy is redistributed between the plasma and the open
by a H-mode to L-mode transition, and a subsequent slow flux surface region within a time τ2 . It should be noted that
thermal collapse, remains to be quantified statistically. in some cases these phases effectively merge and during the
The JET and other tokamak experiences clearly initial redistribution of plasma energy there can be a significant
demonstrate that substantial plasma thermal loss—on a time power outflux from the confined plasma [308,310]. In the time
scale commensurate with the energy confinement time of interval τ2 , the energy starts being lost to the divertor. Data
the parent plasma—can occur before disruptions that are from the rudimentary multi-machine IPB disruption database
initiated by slowly growing MHD instabilities. The counter (see chapter 3 of [1]) suggest that τ2 scales with the minor
examples in JET occur in plasmas where fast-growing MHD radius of the device and that for ITER, τ2 = O(1) ms. More
instabilities cause a ‘prompt’ disruption. Similar ‘prompt recent data from ASDEX Upgrade and JET do not contradict
disruption’ examples are well known and reported in DIII-D this estimate for the relaxation or loss time.
and ASDEX Upgrade. How these ‘fast’ versus ‘slow’ energy The observed relaxation and loss times are qualitatively
loss behaviours and operational experiences will extrapolate consistent with the expected effects of global ergodization of
to ITER (wherein plasma control capabilities and control the internal plasma magnetic surfaces and/or the development
time scales may differ from those now available in present of large-scale internal ballooning mode structures (see
experiments) remains to be determined. But there does section 3.1, above). In all cases, the internal relaxation and loss
appear to be a prospect for using the natural properties times are much faster than the predictions for the time scale of
of slowly growing MHD-initiated thermal collapse, perhaps resistive reconnection. The theoretical basis for the time scale
supplemented by active ‘soft-stop’ procedures analogous to of the fast quench and internal redistribution(s) are not well
those employed in JET, to reduce ITER plasma thermal understood, and a validated basis to predict the ensuing rate of
energy levels prior to otherwise unavoidable disruptions. The plasma energy loss and particle transport to the PFC surfaces
potential benefit of this type of reduction in ITER is considered is not yet available.
below.
Plasma operation in JET with a reversed shear Time scale of PFC energy deposition. Specifying a PFC
configuration seems so far to be the exception to the experience deposition time equal to the plasma loss time gives an upper
with slowly growing instabilities: as the JET experience bound to the possible time-normalized PFC energy deposit.
shows, and is well-known in other tokamaks, disruptions Examples of such ‘loss-rate-determined’ energy deposit have
stemming from such advanced scenarios are often prompt, with been observed [1]. However, ASDEX Upgrade data show
very little pre-disruption energy loss. Thermal and current energy deposition times > τ2 , as illustrated by high-time-
quench rates in reverse shear cases are also among the fastest resolution infra-red (IR) camera data for the divertor target
examples seen in a given device (see discussion of thermal power loading during a disruption (figure 38). These data
quench times below and current quench time discussion in show that the rising phase of the heat pulse on the PFCs,

S168
 Chapter 3: MHD stability, operational limits and disruptions

total heat-pulse duration is significantly longer than the IPB

τ2 scaling (chapter 3, figure 54 of [1]) prediction of τ2 = 0.2 ms
keV
for ASDEX Upgrade. The ECE-measured central Te loss time
#17189
4 is comparable to the IPB scaling for τ2 .
The physics determining the time scale of the heat pulse
duration can likely be described by (i) conduction parallel and
3 perpendicular to the open magnetic surfaces of the SOL, which
central Te is affected by SOL turbulence and stochasticity and (ii) the
limitations imposed by the electron power flux through the
2 sheath at the plasma–wall interface. These physics elements
will certainly contribute to producing the divertor heat pulse
‘time-stretching’ (integrating) effect that is clearly evident in
1 the ASDEX Upgrade data. The strong non-linearities inherent
edge Te in a coupled SOL conduction + sheath limit model might
also explain the seemingly wide degree of variance seen in
0 plasma energy loss time and heat pulse duration data in a
power given machine. The role of high-plasma edge density in
P_max on strike- effecting both ‘thermal isolation’ (via globally reduced SOL
point 100 conduction/convection and a reduced divertor target sheath
modules flux limit) and pre-divertor radiative dissipation (to the divertor
baffles and/or divertor entrance first wall) in the SOL, of the
plasma core thermal energy, is also likely significant. These
50 % P_max 60 expectations about thermal energy transport and deposition
have been compared with B2 SOLPS code calculations for
ASDEX Upgrade [312]. Systematic study of the correlation
τr τd of plasma loss and target energy deposition times on a multi-
15% 20 machine or database basis remains as a future R&D task that
P_max will likely require improvement in Wth (r, t) and FW surface
MW energy deposition diagnostics.
1.7002 1.7006 1.701 1.7014 1.7018 Finally, there is also the matter of the actual magnitude
time (s) of τ2 . In this regard, instrumental limitations in at least some
Figure 38. Plasma temperature and divertor target power data from thermal loss data cited in the IPB (mainly based upon SXR
ASDEX Upgrade. emission) may have prevented accurate observation of the
fastest thermal loss times. For example, newly available fast
τr , coincides approximately with the fast-drop phase of the ECE measurements [308] in JET of the central plasma thermal
thermal quench, that is, with τ2 . The actual amount of energy energy loss during ‘prompt’ ITB disruptions show that τ2 in
that reaches the target in the fast-rise phase is small: most of such disruptions can be as short as 50 µs (cf τ2 = 200–500 ms
the energy arrives at the target plate in a subsequent phase reported—based on SXR data—for ‘typical’ JET plasmas in
and on a longer time scale, τd . The overall energy deposition the IPB).
versus time waveform is also ‘triangular’ rather than ‘square’: Extrapolation by the minor radius scale factor of these JET
there are time-waveform shape factors of order unity applicable ‘ITB-type’ thermal quench times to ITER yields τ2 = 100 µs
to evaluating the magnitude of the resulting time-normalized (cf 700 µs for the standard IPB scaling extrapolation). But
surface heating parameter. the ‘time-stretching’ factor inherent in setting the time scale
The waveforms in figure 38 give some indication of the of the actual divertor target heat pulse may obviate most of the
mechanisms involved. The initial central Te loss event of potentially damaging effect of such very-fast plasma thermal
duration τ2 (here resulting in a fast central energy drop of energy redistribution events. Better understanding of all of
∼80%) gives rise to a corresponding dynamic increase in edge the physics considerations involved in setting the time scale
Te , which in turn is apparently responsible for the rising phase of transport of thermal energy to the PFC surfaces is clearly
of the divertor target power and the initial part of the subsequent needed.
power decay. The central Te decay has a 20–0% ‘tail’ that
follows the edge temperature decay. The data strongly suggests Surface area and locations of heat deposition. Time-resolved
that the initial global mixing (τ2 ) event is followed by a more- power deposition during a discharge is typically measured
or-less uniform decay of plasma energy to the divertor on a time with IR cameras. Several divertor tokamaks (DIII-D, ASDEX
scale that is (in this example)
∼5τ2 . There is also an even- Upgrade, JT-60U, JET) have cameras positioned to view the
longer power decay ‘tail’ that continues after the measured primary or/and secondary divertor targets at one or two toroidal
edge and central Te are essentially zero. azimuths, and the heat flux in the divertor has been diagnosed
For ASDEX Upgrade, the measured τr is sometimes as in a variety of discharges. Routine IR monitoring of the first-
short as the time resolution of the IR system, i.e. 0.1 ms, and wall (other than the portions in direct view of the divertor
is, on the average, 0.5 ms; τd is longer than the time resolution camera) is less common though there are recent developments
of the IR system and is on the average, about 1 ms [311]. This in the IR measurements of an appreciable fraction the whole

S169
T.C. Hender et al

Table 7. SOL power and energy deposition widths in ASDEX tokamaks. In JT-60U, the power deposited during the runaway
Upgrade and JET during the thermal disruption (TQ ) and normal current termination was observed to be spatially localized
operation (SOL ); widths are quoted in the outboard mid-plane. The
(within a poloidal width of 20 cm on the inner divertor plate)
major radii quoted are the geometric radius.
and to occur in pulses of the duration of less than 0.25 ms [319]
R (m) max (cm) TQ (cm) SOL (cm) (see also section 3.4). In both ASDEX Upgrade [320] and JT-
ASDEX Upgrade 1.65 5 >5 1 60U [321], the occurrence of the n = 1 asymmetry of the halo
JET 2.96 5 5 0.5 currents is accompanied by a MHD phenomenon expelling
some particles and residual thermal energy from the plasma in
a time interval comparable to the thermal quench time.
plasma-facing surface in MAST [313], JET [314] and ASDEX During the current quench, part of the plasma magnetic
Upgrade [315], that show evidence of disruption power fluxes energy is transferred to the conducting coils and structures
outside the main divertor region. TEXTOR [316, 317] is around the plasma. The physics of mutual induction and
equipped with an IR camera monitoring the limiter. dissipation of the current in conducting structures around the
The evolution of the temperature and density profiles machine is known. Recent evaluations of the amount of energy
in the plasma SOL or, later on in the disruption, in the dissipated in the external conductors in JET [322] show that
halo region, is usually not diagnosed directly (e.g. with typically 35% of the magnetic plasma energy is dissipated
Langmuir probes or Thomson scattering) during and after in conductors outside of the plasma during disruptions. In
the thermal quench. Nevertheless the IR-measured divertor ITER, owing to the low-resistance and long (∼0.7 s) n = 0
power deposition profiles (magnetically extrapolated back to time constant of the ITER vacuum vessel, transfer of in-vessel
the outside midplane SOL) indicate that during the thermal magnetic energies to external conductors is expected to be
quench, the SOL expands considerably, relative to the similarly negligible in most fast disruptions and fast VDEs. However,
observed before-disruption ‘steady-state’ SOL width, SOL . slow disruptions and/or slow VDEs may have some in-vessel
Observation of SOL expansion factors (λTQ , see figure 36) energy transfer.
of 1–10 in limiter and divertor tokamaks is reported in the
IPB [1]; more recent data from medium- and large-scale Energy balance. In JET the thermal energy deposited in the
divertor tokamaks (see below) indicates that the range of divertor within the time scale of the thermal quench (1 ms)
SOL expansion factors is shifted to higher values: e.g. 5  is only a few % of Eth [323] and is spread on the whole
λTQ ∼20. With the exception of TEXTOR, the width of the divertor. The lack of full accountability for the plasma
power channel in the SOL during thermal quench is typically thermal energy during the thermal quench cannot be fully
larger than the field-of-view width, max , monitored (viewed) explained by (1) the effect of radiative losses upstream of the
by the IR camera (table 7). divertor, (2) asymmetries in the on-divertor distribution and (3)
In the case of JET, the energy balance calculation (amount uncertainties in the evaluation of the divertor surface power
of energy deposited within the IR camera field-of-view) fluxes. In the case of JET a large perpendicular transport
indicates that most of the plasma thermal energy is transported of energy during the thermal quench and losses beyond the
beyond the 5 cm region of the SOL ‘seen’ by the IR camera field-of-view of the IR camera were postulated to justify the
and new IR data indicates that at least some of this energy is experimental observations; more recent observations support
conducted to the walls [314]. this postulate [314].
The amount of energy deposited on the divertor during
Magnetic energy deposition. The phenomenology and the whole disruption in JET amounts to up to 10% of the total
physics basis for the current decay phase of a disruption pre-disruption energy, Etot = Eth + Emag . Between 50%
(wherein the in-vessel plasma magnetic energy is dissipated) and 80% of the total energy is found to be radiated. This
are treated in detail in section 3.3. Here only the magnetic fraction does not seem to depend on the type of disruption. ITB
energy deposition aspects will be considered. While collapses can generate the highest and most localized power
recent (since the IPB) data still generally support the densities.
recommendations given in the IPB about magnetic energy In DIII-D [324] the energy conducted to the divertor within
deposition (80–100% of Wmag more-or less uniformly radiated the whole disruption is 15–50% of the total plasma energy or
to the FW, 0–20 % locally conducted to the divertor or divertor 50–100% of the thermal energy of the pre-disruptive plasma,
baffles) [318], experience in present tokamaks shows that apparently less in the case of radiative-limit disruptions and
a significant fraction of Wmag can be (is, in at least some more in the VDE and beta-limit disruptions. Similar results
cases) deposited on the divertor targets during or following are reported in [310] for ASDEX Upgrade. During the 4 ms
completion of the thermal energy quench phase. Conversely, centred about the thermal quench time, the energy deposited on
given the present observations that a significant fraction of Wth the lower divertor is in average 90% (and can reach 200%) of
is being deposited somewhere other than in the IR-observed the thermal energy. This suggests that during this time already
portions of the divertor, deposition of the ‘missing’ fraction of a fraction of the magnetic energy may be dissipated in some
Wth on the divertor entrance baffle surfaces and/or at least a cases. The amount of energy deposited on the lower divertor
limited portion of the FW must be taking place. during the whole disruption is in average 30% (and can reach
Despite the fact that a large fraction of the plasma magnetic 45%) of Etot .
energy is radiated during current quench, temporally and
spatially localized power deposition has to be expected in Discussion, conclusions and recommendations for future
a ITER current quench, as already observed in the existing R&D. A schematic and parametric summary of the present

S170
 Chapter 3: MHD stability, operational limits and disruptions

Table 8. Single-parameter mitigation factors to avoid carbon ablation.

Mitigation Pre-disruption Pre-divertor Time SOL
method Wth reduction radiation stretching broadening
Symbol (figure 36) ε 1-fdiv, TQ (fdiv, core + fdiv ) γ ∗ τ2 λ
Reference basis 1.0 0.0 γ = 1.5 λ=7
Requirement
0.09 0.91 γ  86 λ  80

status and expectations (predictions for ITER) are embodied happen or will be possible in ITER, present data and the
in figure 36 and table 6. The most immediate change since the considerations developed above suggest that nearly melt- or
IPB lies in the 2-fold reduction in the ITER area-normalized vaporization-free divertor PFC operation may be obtainable in
and time-normalized plasma thermal and magnetic energy at least some of the proposed ITER plasma operation regimes,
loadings. With the corresponding dimensional scalings of especially in plasma operating regimes or modes where natural
the divertor deposition areas and plasma energy loss (τ2 ) time or ‘soft-stop’ before-disruption thermal energy loss will be
scales taken into account, the time-normalized Wth loading possible.
on the divertor targets drops from ∼1000 MJ m−2 s−0.5 for the We also note here that while there has been modest
ITER-EDA design to ∼450 MJ m−2 s−0.5 for ITER. This latter progress since the compilation of the ITER Physics Basis,
loading, which is calculated for a full-energy disruption (ε = there remains significant work to do in terms of our ability
1) with 100% Wth into the divertor, a 7-fold expansion of the to account for the magnitude and time history and spatial
deposition width (λTQ = 7) and with the further conservative distribution of the deposited thermal and magnetic energy,
assumption of prompt thermal deposition (tdep =∼ 1.5τ2 , with model and simulation development and testing of mitigation
τ2 following the IPB scaling), can be regarded as an upper techniques. Given the importance to ITER (and DEMO) of
bound to the expected ITER divertor energy loading. While minimizing, or eliminating, the operation-limiting effects of
this ‘upper bound’ estimate is still a factor of ∼12 (∼8) above disruptions this work remains a key area for future studies.
the ablation (melting) onset threshold for carbon (tungsten), the ITER must prove high-reliability disruption avoidance and
prospects for obtaining either reduced or negligible divertor mitigation methods, essential for DEMO (where disruptions
target erosion are now enhanced relative to the same basis power loadings problems will be more severe than for ITER).
situation for the EDA ITER. The authors of [308, 325] reach
similar conclusions about the range of possible divertor energy 3.3. Current quench dynamics
loadings in ITER; it should be noted in [325] that a 75%
reduction from the full ITER Q = 10 scenario plasma energy During the current quench phase of a disruption in a vertically
is assumed before the thermal quench (i.e. ε = 0.25). Table 8 elongated plasma, the decay of the plasma current and the
demonstrates in a parametric manner the ‘single-parameter’ ensuing motion of the plasma column induce toroidal currents
mitigation factors required to reduce the ∼450 MJ m−2 s−0.5 in the nearby toroidally conducting structures (e.g. the ITER
loading to the ∼40 MJ m−2 s−0.5 threshold for carbon ablation. torus vacuum vessel) and also drive force-free helical current
The single-parameter mitigation requirements embodied flow in the wall-contacting ex-plasma ‘halo’ region that lies
in table 8 can be compared with the corresponding observations beyond the last closed plasma flux surface. Both the induced
in present tokamaks. Pre-disruption energy loss factors as large toroidal current and the ex-plasma halo current flow act to
as 0.1 have been observed in JET. Very large scrape-off layer mediate the dynamic evolution of the current channel during
(SOL) broadenings, corresponding to λ > ∼20 and/or very the current decay phase. The components of the ex-plasma
high ‘pre-divertor’ radiation fractions (>90%) can be inferred halo current that reconnect through the vessel and the in-
from the low-divertor Wth energy depositions observed in JET. vessel conducting structures (e.g. the ITER shield blanket
Divertor heat pulse duration ‘stretching’ by a γ factor of 5–10 modules the divertor baffle and cassette modules) give rise to
is routinely observed in ASDEX Upgrade. in-vessel halo currents that produce forces on these structures.
Except for the JET observations of high fractions of In addition, the plasma current decay and motion induce locally
pre-disruption Wth loss in ‘slow’ density limit and similar circulating ‘eddy currents’ in nearby plasma-facing conducting
disruptions (and equivalent observation in ASDEX Upgrade), structures (e.g. the ITER first-wall and shield blanket modules),
none of the single-parameter mitigation observations explicitly and the interaction of these induced circulating currents with
yet matches the ITER ‘no-ablation’ requirement. On the the toroidal and poloidal fields gives rise to localized torques
other hand, a modest combination of pre-divertor radiation and overturning forces on these PFC structures.
(fdiv, core + fdiv = 0.4) moderate time-stretching (γ = 10) The magnitude and time-history of the eddy current
and doubling the SOL broadening (λ = 15) would reduce loading depends on the local rate of change of the poloidal
the ITER divertor target deposition to ∼48 MJ m−2 s−0.5 . magnetic field, Bp , and on the eddy current decay time constant
This comes close to meeting the no-ablation or no-melting (L/R time) of the affected structures. Hence understanding
thresholds (but our estimates here should not be interpreted of the range of possible plasma current quench rates, local
as constituting actual assessments of ITER divertor target dBp /dt rates and the simultaneous magnitude and direction of
response). Before-disruption energy loss would further reduce halo current flow in the affected structures is needed. Simple
the time-normalized loading. While the present understanding extrapolation of the quench time based upon a linear current
and data basis for Wth deposition and mitigation approaches decay waveform and the assumption of a static (fixed in
is insufficient to make definitive conclusions about what will space) plasma current channel may not necessarily provide

S171
T.C. Hender et al

the accurate assessment of dBp /dt needed, for example, for JT-60U 31708
3 6×1016
evaluation of the loadings on the ITER in-vessel shield Ip_Experiment
modules. Hence there has been recent attention (since the Ip_Linear fit
writing of the IPB) to the details of the time history of the Ip_Exponential fit
plasma current decay waveform and also to the eddy current Neutron
and halo current generating effects of motion of the plasma
2 4×1016
current channel during the current decay. In section 3.3.1 that

Neutron (s-1)
follows, emphasis is on global current decay; effects of current RMSE=0.157

Ip (MA)
channel motion due to a VDE, are considered separately in
section 3.3.2. Ultimately, the combined effects of current
decay and motion and the generation of poloidally flowing
1 2×1016
halo currents in in-vessel structures must be assessed in a self-
consistent manner using an integrated dynamic model (see
RMSE=0.015
section 3.5).

3.3.1. Global toroidal current decay. Evaluations detailed in

0 0
the ITER Physics Basis, chapter 3.4 [1], of the range of plasma 7.035 7.04 7.045 7.05
current decay rates observed in seven circular and elongated Time (s)
cross-section tokamaks (Alcator C-MOD, ASDEX Upgrade,
DIII-D, JET, JT-60U, TFTR and Tore Supra) resulted in the Figure 39. Waveforms for a typical fast current quench observed in
finding that the lower bound on the inferred area-normalized JT-60U. Deviation of the final portion of the decay from a pure
exponential is attributed to loss of runaway current. The
current decay time, t60 = t (80%)–t(20%), normalized by the corresponding increase in neutron rate is attributed to photo-neutron
elliptic-approximation estimate for the plasma cross-section production [327].
area, S∗ = πκa 2 , was t60 /S∗ = 0.8 ms m−2 , is independent
of plasma elongation and average toroidal current density. If 5
this area-normalized time for the inferred 60% current decay
(from 80% Ip0 to 20% Ip0 ) is directly extrapolated to the time
for 100% current decay, the resulting ‘linear-basis’ minimum 4
decay time is 0.8 ∗ 100/60 = 1.33 ms m−2 . It should be noted
that the basis for determining t60 in the IPB was not uniform
∆t / S (ms/m2)

between tokamaks and for example the JT-60U value is inferred 3

from observed maximum Ip quench rate.
The shape of the current waveform was not explicitly
considered in chapter 3.4 of [1], although anecdotal and 2
indirect evidence for many discharges having an ‘S-shaped’
decay waveform with a nearly linear intermediate current ∆t=Ip0/<dIp/dt>80=>20 except for DIII-D and JET
decay phase was cited. More recent systematic analysis 1
of JT-60U current quench waveforms shows that they can ASDEX-U JET (100 => 40)Ip0
C-MOD JT-60U
be fitted with either an exponential-like or a linear-like DIII-D (90=>10)Ip 0
function [326]. Exponential fitting is better when the area- 0
normalized quench time is small; linear fitting is better 0 1 2 3
I p0 (MA)
otherwise. Differences between linear and exponential models
for the current decay waveform (and hence the corresponding Figure 40. IPB data, replotted in terms of the linearly extrapolated
estimates of dBp /dt) affect eddy current forces in structures current decay time, t, normalized by their respective poloidal
with time constants comparable to or shorter than the current cross-section area, S [327].
decay time. Calculations of global magnetic pressures or eddy
current forces in long-time constant structures are not affected Figure 40 shows, for non-circular tokamaks only, current
by dBp /dt, since such pressures or eddy current forces depend quench data from the original 1996 international database (IPB
on the total change in poloidal field, independent of time chapter 3, section 4 of [1]), supplemented with updated JT-60U
scales. data [327] based on the actual 80–20% quench time for Ip . In
Figure 39 demonstrates the nearly exponential nature figure 40, the data are plotted to show the linear extrapolation
of the central (80–20%) segment of a typical fast JT-60U quench time, dIp /dIp /dt, calculated, except as noted for JET
current decay waveform and illustrates how approximating and DIII-D, on an 80–20% of Ip0 basis. Data for DIII-D and
the decay by a linear extrapolation, determined by taking the JET are included, corrected for the respective data basis; 90–
time difference between 80% current and 20% current, results 10% decay for DIII-D and 100–40% decay [308] for JET.
in an approximately 2-times underestimate of the peak rate For this revised data set, the lower bound on area-
of current decay. Elementary considerations show that the normalized linear quench time for 100–0% decay is
underestimating factor for an exact exponential waveform is 1.8 ms m−2 ; this lower bound is about 35% greater than the
2.3; the actual JT-60U waveform yields an ∼40/18 (= 2.2) ∼1.33 ms m2 lower bound (100–0% basis) on area-normalized
ratio of linear versus exponential decay times. linear quench time inferred from the original IPB-era data

S172
 Chapter 3: MHD stability, operational limits and disruptions

Figure 41. Plasma configuration evolution and halo current region development in a typical Alcator C-Mod disruption. The data shows that
much of the ex-plasma halo current region (inferred from magnetic reconstruction data) now misses the new, less-protruding inboard
divertor (see discussion below).

(chapter 3.4 of [1]). This increase in quench time is largely cooling of the plasma centre occurs. Such events are typically
due to the JT-60U data being based on an 80–20% Ip quench described as a ‘hot-plasma’ VDE. The plasma current centroid
time (as opposed to the maximum Ip quench rate as in the moves vertically away from its equilibrium position and the
IPB). Application of the 1.8 ms m−2 lower bound to the present moving plasma column eventually contacts a limiting surface.
ITER, with S∗ = 22 m2 yields a 40 ms linear quench time, or The direction of initial movement, even in single-null plasmas,
equivalently, an exponential decay with an 18 ms time constant. can be either towards or away from the X-point, depending on
The DIII-D data in figure 40 are based on plasma current the changes in the plasma current and pressure profiles (i.e.
decay data from an external (outside the vacuum vessel) li and βp ) as well as the initial location of the plasma current
Rogowski loop. Recent consideration [328] of decay rates centroid [330].
evaluated using an in-vessel ‘pseudo-Rogowski’ diagnostic In a VDE, the plasma continues to move into the wall,
(synthesized from a set of in-vessel magnetic probes) indicates reducing the plasma area, typically with little change in the
that the fastest DIII-D current decay data in figure 40 have S∗- total plasma current, thus reducing the edge safety factor. A
normalized decay times that fall in the range 1.4–1.7 ms m−2 . pronounced ex-plasma halo current flow also develops (see the
Further consideration of these DIII-D results and new current discussion and figure 41 below for a representative example of
quench data, to be collected on a systematic multi-machine plasma configuration evolution and halo region development
basis, under the aegis of the International Tokamak Physics during a vertically unstable disruption in Alcator C-Mod),
Activity, remains as an ongoing current quench research issue and the ‘halo’ currents flowing in the wall-contacting ex-
with the most recent analysis showing the lower bound on area plasma region reconnect through the structures that the plasma
normalized quench time should be taken as 1.7 ms m−2 [329]. comes in contact with. The resulting in-vessel currents, which
Calculations of the electromagnetic effects of induced typically flow mostly in a poloidal direction, are commonly
eddy current loads in the ITER shield blanket modules are called ‘halo currents’. With suitable instrumentation, these
described in [327]. The effects of both a 40 ms linear current in-vessel halo currents can be measured directly and the torus-
decay waveform and an 18 ms exponential decay waveform circumference sum of the measured or inferred poloidal in-
are considered. The effects of the projected in-vessel halo vessel current flow is typically taken as the measure of the total
currents are also included. The conclusion reached therein is (in-vessel) halo current, denoted symbolically in the IPB and
herein as Ih . Unless otherwise noted, reference to the poloidal
that the structural design of the blanket modules is sufficient to
component of the halo current, in the in-vessel conducting
withstand the maximum collective loading, albeit with only a
structures, should be understood.
small (∼20%) margin relative to structural design allowables.
When the plasma is in contact with the wall in-vessel halo
The authors of [327] note that a somewhat slower current
currents are also driven to compensate for the plasma toroidal
decay scenario (one with a 60–80 ms linear decay) would,
flux lost because of the plasma area reduction. The plasma
with the corresponding decrease in eddy current loading,
current stays approximately fixed during the hot plasma–wall
yield a significant increase in margin relative to the structural
contact phase because the resistive decay time of the still
allowables.
hot core plasma is long compared with the vertical motion
timescale. When the boundary safety factor decreases to a
3.3.2. Vertical instability, halo currents and mechanical forces. sufficiently low value (typically less than 2), rapid growth of
Since plasmas in typical elongated cross-section tokamaks MHD activity (n = 1) produces a fast thermal quench similar
are inherently unstable against vertical displacements, a to those observed in major disruptions. As the plasma current
sufficiently large and fast change in plasma parameters can starts to decay, toroidal currents are induced in the halo region
cause the loss of the vertical position control, leading to an to conserve the poloidal flux [331].
uncontrolled upward or downward excursion of the plasma Vertical instability can also occur following the onset of a
column. A true vertical displacement event (VDE) begins with disruptive thermal quench. In this case, the resulting vertical
a loss of vertical stability that develops before any appreciable evolution can be described as a vertical disruption, or more

S173
T.C. Hender et al

correctly, as a vertically unstable-disruption (VUD) current confirms the less-explicit understanding documented for the
quench. The resulting ‘cold-plasma’ current decay phase IPB that thermal quench onset in VDEs typically occurs for
shares the most of the attributes—including the generation of qψ ∼ 1.5–2. The ability of the plasma core to maintain gross
in-vessel halo currents and large transient vertical forces on MHD stability at qψ < 2 is also consistent with the expected
the vacuum vessel—of a ‘hot-plasma’ VDE. Figure 41 shows stabilizing effect of the current-carrying ex-plasma halo. In
magnetic reconstruction data from a typical VUD example in JET, higher halo currents and asymmetries are observed when
Alcator C-Mod. Similar examples of plasma evolution and the boundary safety factor decreases close to unity [333].
halo development during disruption or VDEs are obtained in However, events starting with a ‘thermal collapse’ usually have
ASDEX Upgrade, DIII-D, JET, JT-60U, etc. much smaller halo current fraction and hardly any asymmetry.
It is noted here parenthetically that both the IPB-era This is believed to be because in these cases, the plasma current
evaluation of the lower bound on area-normalized current decay is faster than the plasma cross section reduction and a
decay rates (chapter 3.4 of [1]), and the more recent relatively high-boundary safety factor is maintained.
reconsideration of IPB and post-IPB data for elongated The vertical force due to the plasma vertical displacement
tokamaks [327] ignore any distinction between decay rates is proportional, to first approximation, to the product of
(following onset of thermal quench) for hot-plasma VDEs the plasma current, the vertical displacement of the current
and vertically unstable cold-plasma VUDs. In the IPB centroid and gradient of the before-disruption external
comparison of lower-bound data from circular (vertically equilibrium field. Therefore a substantial reduction in vertical
stable) and elongated (vertically unstable) tokamaks, there was load on the vessel can be achieved if the vertical displacement
no observable distinction in the lower bound on minimum area- of the plasma during current quench can be minimized. The
normalized current quench times. concept of effecting plasma operation in a weak single-
null or balanced double-null plasma with a before-disruption
Halo currents: phenomenology and physics basis The basis current centroid positioned near the vertical-instability ‘neutral
and phenomenology of the generation of both ex-plasma and point’ has been advanced [334] as being advantageous to
in-vessel ‘halo currents’ is described in the IPB (chapter 3.4 minimized vertical loads. However, the electromechanical
of [1]) and in, for example, [330, 332]. Briefly put, from loads due to the plasma current decay would be still present,
a tokamak design basis point of view, the most important as well as the risk of generating runaway electrons. In
considerations for halo currents are (1) the magnitude of the addition, concerns have been expressed that variations in
in-vessel halo currents, (2) the toroidal asymmetry of these in- the before-disruption or during-disruption evolution of the
vessel halo currents and (3) the resulting jin-vessel × B forces vertical-stability-determining parameters (li and βp ) may make
that act on the vessel components. It is the sum of these it impossible to determine, for a single-null plasma, a single
local forces (plus the global vertical forces that are directly neutral point location (height) that avoids vertical instability
generated by induced toroidal vessel currents) that give rise for a range of disruption ‘types’ [330].
to the large vertical vacuum-vessel forces that are observed
in present tokamaks during VDEs and VUDs. Toroidal Halo current magnitude, asymmetries and experimental ob-
asymmetry in the halo current distribution can also give rise servations. Incorporation of new halo current measurements
to an asymmetric radial de-centering and/or tilting forces on into the existing ad hoc ITER disruption database has been
vacuum vessel systems. Finally, the magnitude and spatial ongoing since the ITER Physics Basis was published. These
(radial) extent of the ex-plasma halo current flow becomes the new data additions reflect the effect of several major hard-
dominant factor (especially in the end phase of the current ware modifications, including, in particular, ASDEX Upgrade
decay) in determining the equilibrium dynamics of both hot- operation with the Divertor II (Lyra) and Divertor IIb target
plasma VDEs and cold-plasma VUDs. More simply put, and baffling modifications, and Alcator C-Mod operation with
in this end phase, most of the remaining plasma current a newly shaped inboard divertor. The new data also include
flows in the halo region. The presence and dynamics of this new halo current measurements from JET that reflect refur-
ex-plasma halo current must be self-consistently taken into bishment of disruption-related instrumentation, and halo cur-
account in calculations of the plasma equilibrium evolution and rent measurements from the low-aspect-ratio MAST spheri-
the resulting vacuum vessel and in-vessel-component toroidal cal tokamak. The resulting updated plot of toroidal peaking
eddy currents, and forces on these component systems. The factor (TPF) versus Ihalo /Ip0 , (figure 42), includes new data
need for—and development of—self-consistent models for from JET [335], ASDEX Upgrade [320] and MAST [336].
plasma dynamics and the resulting ex-plasma and in-vessel The newly added data basically reinforces past findings and
halo currents are further addressed in section 3.5. does not exceed the previously established limit on the prod-
Recent systematic studies of halo current characteristics uct of TPF and halo current fraction. There continues to be
and phenomenology have contributed to developing the significant scatter in both Ihalo /Ip0 and TPF, which is presum-
improved physics basis understanding needed for such models. ably due to uncontrolled parameters such as the SOL resistiv-
Since the magnitude of the poloidal halo current flow depends ity (which reflects the impurity dynamics during the quench),
explicitly on the plasma edge (last closed flux surface) safety and/or un-documented differences in MHD activity in the
factor, qψ , a database for the edge safety factor at the onset of MHD-turbulent post-disruption plasma. The ability to predict
thermal quench during vertical displacement events has been (or correlate with experimental parameters), on a first-principle
compiled in JT-60U [332]. It is found that the edge safety factor basis, halo current magnitude and TPF remains elusive.
at thermal quench onset in hot-plasma VDEs varies between The new MAST and C-Mod data also illustrate some
1.5 and 2, depending on the discharge conditions. This finding machine-configuration and in-vessel-component sensitivities

S174
 Chapter 3: MHD stability, operational limits and disruptions

5 CMOD(94-95)
CMOD(96)
COMPASS-D
ASDEX-U
4 MAST
DIII-D
JET1999-
JET2002+
3
JT60_wo_gas
TPF

JT60_w_puff
TPF(0.7)
2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 Figure 44. C-Mod halo currents measured in new inboard divertor
Ih,max / I p0 are reduced by a factor of about 2 compared with the previous
scaling (dashed line) with the original inboard divertor.
Figure 42. IPB halo current data with recent data from JET and
MAST added. The upper bound on the product of Ih, max /Ip0 and ‘global’ increase of resistance (isolation) of in-vessel systems
TPF remains unchanged. Note, however, the lack of new data with
TPF > 2 [327].
from vessel ground remains to be evaluated.
In Alcator C-Mod, installation in 2002 of the new inboard
divertor, which has a much less-protruding shape than the
2 previous divertor, has resulted in significant changes in the
0 measured halo current characteristics [337]. The typical halo
-2 current scaling versus Ip scaling measured in the new divertor
kA

-4 is now about half the magnitude of the scaling measured

-6 0.1 mΩ with the previous divertor (figure 44). During the current
-8 Rib current 100 mΩ
quench, the total halo current now seen frequently changes
sign in time. The toroidal asymmetries of VDE and non-VDE
0 disruptions (VUDs) are also now quite different, whereas the
-10 toroidal asymmetries previously observed were similar. Non-
VDE disruptions now have TPF’s close to unity. The fact
V

-20
that much of the halo current region (inferred from magnetic
-30
Rib voltage reconstruction data) is now observed to miss the new limiter
-40 4043
(figure 41) provides a plausible explanation for the reduced
0.272 0.276 0.280 0.284 halo current magnitude and the transient polarity reversal.
time (sec) In JT-60U, halo current characteristics have been
measured during the runaway current phase [319]. The data
Figure 43. Halo currents and voltages measured in a rib limiter in show that the in-vessel halo current, averaged toroidally, is
MAST with low- and high-resistance connection to the vacuum
very small during the runaway current plateau phase (this
vessel [336].
absence of halo current is consistent with nearly constant total
plasma + runaway current in the plateau phase), but starts
of the halo current characteristics. On MAST, where several to increase after the termination of runaway current, which
divertor strike-point floor plates (‘rib limiters’) are connected begins when the plasma surface safety factor, qs , becomes
to vessel ground through instrumented resistors, experiments smaller than 2. Calculation of halo current using a plasma
wherein the resistor value for one of the ribs was varied equilibrium analysis code, DINA [338], in which the eddy
over a large range (0.1 m
, 0.1
, 3.3 k
) demonstrated that current in the vacuum vessel is taken into account, shows
increasing the resistance resulted in a marked reduction of the good agreement with the measured current (figure 45(A)).
halo current flowing through the rib, from a typical value of The toroidal distribution of measured halo current, shown in
∼10 kA down to just a few hundred Amps or less (figure 43). figure 45(B), indicates that the profile of the halo current has
In contrast, the voltage drop varied by no more than a factor of 3 a dominant n = 1 toroidal asymmetry with TPF ∼2, and does
[336]. The halo current flowing through the other, unmodified not change significantly with time. These observations are
ribs showed no measurable change. This implies that the the same as the previous observations in VDEs without the
disrupting plasma behaves more like a voltage source than a generation of runaway electrons [332].
constant current source, and suggests that it may be possible
to reduce the halo current in vulnerable in-vessel components Magnetic reconstruction and modelling of axisymmetric
by adjusting their resistance to vessel ground. Whether the halo currents. There has been significant progress since
overall magnitude of the halo current can be reduced by a the compilation of the IPB in developing methods that

S175
T.C. Hender et al

2.0

1.5

1.0

0.5

Z (m)
0.0

—0.5

—1.0

—1.5

—2.0

(A) 1 2
R (m)

Figure 46. JFIT reconstruction data for the end phase of a DIII-D
VDE. The darkness of the contours corresponding to each square
current element reflects the amplitude of the current in the element.
Elements located in vessel regions showing no contours have
negligible current. The solid contour in the plasma region denotes
the last closed flux surface, enclosing the ‘core’ plasma current.
Halo current lies outside this core region. Following loss of the last
closed flux surface, the halo current region rapidly diffuses to fill the
entire vessel. (Reprinted with permission from [331]. Copyright
1999, American Institute of Physics.)

separation of the core and halo currents and study of their

detailed evolution during a disruption. The data in figure 41 are
examples of a reconstruction of a C-MOD VUD using a similar
current-element method. The alternate method is DINA (G–S
equilibrium based) simulation as shown for a JT-60U plasma
(B) current termination in figure 45(A). Both methods are capable
of yielding data on halo current magnitude and width and (by
Figure 45. (A). Temporal evolution of (a) plasma current, runaway inference) effective halo region conductivity and temperature.
electrons (hard x-ray emission IHX ), calculated toroidal halo current
Ihtor , (b) surface safety factor qs , (c) poloidal halo current from Modelling of core and halo plasma current evolution
exp
Rogowski coil measurements Ih and (d) that by DINA code during DIII-D VDE’s and/or the vertical displacement phase
analysis IhDINA , during and after the termination of runaway of a major disruption has produced improved physics
current [319]. (B) Temporal evolution of the toroidal distribution of understanding of the driving mechanisms for disruption and
measured halo current shown every 0.25 ms during the phase of halo halo currents [331]. The halo safety factor is a function of
current increase (2.962 75–2.963 75 s) [319].
the competing effects of current decay and speed of motion.
Rapid core current decay relative to plasma motion (rate of
allow magnetic reconstruction of the equilibrium dynamics decrease of the core minor radius) tends to increase qhalo and
of the core plasma and halo current regions during a thus decrease poloidal halo current (Ipol ) relative to the toroidal
disruption or fast VDE. Traditional equilibrium reconstruction halo current (Itor ) (defined as a ‘Type I’ limit), whereas rapid
algorithms, based on the accurate Grad–Shafranov (G–S) motion relative to core current decay rate tends to decrease
constraints, frequently have difficulty converging during qhalo and thus an increase in Ipol relative to Itor (defined as
evolution of VDE’s, particularly when large currents are a ‘Type II’ limit). Figure 47 summarizes the measured and
driven in conducting structures. Figure 46 shows a model-simulated evolution of a ‘Type II’ DIII-D VDE . The
reconstruction—effected using the distributed-current element reconstruction produces a fit of toroidal currents alone, which
code JFIT—of the toroidal plasma and halo currents during allows an estimate of the edge safety factor (owing to the
the final phase of a DIII-D VDE [331]. Since this type of virtually complete loss of plasma stored energy producing
‘current-element-based’ code does not impose detailed force a force-free plasma). The poloidal halo current can be
balance, but rather fits both conductor and plasma currents, calculated using the force-free condition in the halo, which
its reconstruction is very robust even in the presence of implies that Ipol = Itor /qhalo . Following wall contact, toroidal
large driven vessel currents. Robust reconstruction of the current is induced and convected into the halo from the core
post-thermal quench phase of the current evolution allows plasma. Both toroidal (dIp /dt) and poloidal (d /dt) voltages

S176
 Chapter 3: MHD stability, operational limits and disruptions

#15392
2
Core plasma current (MA) JFIT 1.0 I_p (MA)
(a)
1
0.0 n=8

0
1.0
Toroidal halo current (MA) (b)
0.5 JFIT
Model

0.0 n=1
F_z (a.u.)
Poloidal halo current (MA) (c) 0.0
z_curr (m)
0.4 JFIT 0.5
Model 0.92 1.08 1.24 1.4
0.2 time (s)
TCA
0.0
Figure 48. Vertical forces on the ASDEX Upgrade vacuum vessel
1500 measured at the eight support rods (labelled n = 1–8) [320].
Toroidal halo driving voltages (V) (d)
1000
dIp/dt example of the time traces of the eight strain ASDEX Upgrade
500 dPhi/dt
signals is shown in figure 48. The high degree of azimuthal
0 symmetry is evident.
1.710 1.715 1.720 1.725 1.730 The displacement of the vessel (measured at the midplane)
t (s) has a time behaviour similar to the support-rod strain signals.
Figure 47. Comparison of measured (TCA = tile current array), The maximum vertical displacement is ∼ ±0.7 mm at the
modelled and JFIT-inferred halo current for a DIII-D VDE. midplane, while the largest net radial displacement observed
(Reprinted with permission from [331]. Copyright 1999, American was only 0.24 mm, which corresponds to a static lateral force
Institute of Physics). of 17 kN. However, since the disruption loads are applied on
a timescale much shorter than the vessel oscillation time, the
contribute to the halo current drive. The dynamic evolution instantaneous lateral forces may be much higher.
of this reconstruction-inferred poloidal halo current and the A somewhat different type of vessel behaviour during
model-simulated current are in good agreement with the value asymmetric VDEs is observed in JET. In JET asymmetric
measured by the tile current monitors. VDEs, the current and position of the plasma can become
toroidally non-uniform once the boundary safety factor
Vertical and radial forces. Vertical and radial forces on the
decreases to a critical value; this is possible when the
vacuum vessel systems in ASDEX Upgrade and JET are
plasma cross-section shrinks faster than the plasma current
routinely observed during ASDEX Upgrade disruptions and
decays [339]. The mechanical flexibility of the JET
asymmetric VDEs in JET. While the magnitudes and nature of
vessel supports, leads to sideways displacements of the
the forces and vessel responses in both cases are commensurate
vessel. Measurements of these displacements provides clear
with elementary assessments, the ASDEX Upgrade and JET
evidence of the mechanical effects of the forces acting during
experiences illustrate some of the underlying complexities and
asymmetric VDEs, against which the predictions of various
subtleties of the mechanisms responsible.
The ASDEX Upgrade vacuum vessel is equipped with interpretive models can be benchmarked [340].
diagnostics to measure mechanical forces and displacements. Analysis of the interactions among the current-carrying
The vessel is suspended through eight vertical rods (spaced systems (plasma, vessel, TF and PF coils) reveals that the
toroidally every 45◦ ), each instrumented with a strain gauge. repelling force between the plasma and the vessel is of little
The vacuum vessel is also instrumented with displacement significance in asymmetric events, even if it is the main
gauge, which measures its radial and vertical movement at symmetric repelling force. The sideways force acting on the
the equatorial midplane at four toroidal locations (every 90◦ ). vessel is shown to be mainly due to the interaction between
Over a database of 100 disruptions, vertical forces above the toroidal field and the asymmetric circulation of currents
300 kN have been observed with Ip = 0.8–1 MA. In slow in the wall [341]. The source/sink of the asymmetric wall
VDE’s (growth rates of order 100 ms), peak magnitudes have current is given by the imbalance between the influx and the
reached 500 kN. The peak forces do not scale in a simple outflow of perpendicular (poloidal) halo current at any toroidal
manner with the vertical unstable plasma force, Finst,z ∝ Ip2 or , location. For a m = 1/n = 1 macroscopic asymmetry, the
at fixed qedge , with Ip ×Bt . The upper limit of the vertical force current variation as a function of toroidal angle is expressed
is described approximately by |Fz |  250 kN/MA/T. Unlike by δIp cos φ. The continuity equation gives the current per
the large asymmetries seen on JET, the toroidal asymmetries unit length flowing vertically from the plasma into the top part
of the vertical forces on ASDEX Upgrade, as measured at the of the vessel as jv (φ) = δIp /R sin φ. Apart from a thin ring
support rods, are typically <20%, and on average ∼10%, and where the current enters/exits, the wall has zero divergence.
do not show a clear n = 1 tilt of the whole vessel. A typical Therefore the vessel current density can be expressed in terms

S177
T.C. Hender et al

of a stream function and the poloidal components of the wall conversion avoidance and runaway discharge termination that
current estimated as can be reliably effected in an ITER-class tokamak. A
comprehensive discussion of the physics basis for runaway
δIp p − 2s
js = sin φ, generation, confinement, loss and potential mitigation has
2R p been presented in the 1999 ITER Physics Basis [1]. Since
where p is the poloidal perimeter of the vessel and s a poloidal then, significant progress has been made in this subject, and
co-ordinate (varying between 0 and p). Integration over the new results and important developments in understanding are
vessel of the vertical projection of the poloidal vessel current summarized in this section.
path gives a net sideways force ∼ πBtor bδIp , where b is the
vertical major radius of the torus (full details of this calculation 3.4.1. Observations in present tokamaks. Data continues
are in [341]). to be accumulated about parametric sensitivities for runaway
Studies of plasma current quench dynamics and halo generation, confinement and loss in the present generation of
current generation mechanisms and characteristics (magnitude tokamaks. Observations reported since the writing of the IPB
relative to Ip0 and toroidal peaking factor) indicate that the have been made for example in the JET, JT-60U and Tore Supra
basic recommendations and design guidelines given in the tokamaks. Observations documenting runaway generation
ITER Physics Basis continue to apply. Some relaxation from injection of argon pellets have also been reported for
(increase) in the guideline for the lower bound to the area- DIII-D [342]. Various theoretical analyses that contribute to
understanding of these results and the corresponding estimated
normalized current decay time, to t100 /S∗ ∼ = 1.7 ms m−2 , can
effects in ITER have also been published. Taken together, the
be recommended, albeit with the caveat that a few examples
new data provide an improved understanding of the nature and
that fall slightly below this lower bound have now been seen in
possible solutions to the problem of avoiding or mitigating
DIII-D [329]. New data obtained in JET and MAST supports
runaway damage in an ITER-class tokamak.
the IPB recommendations for maximum halo current fraction
(Ih,max /Ip0 ) and (Ih,max /Ip0 ) × TPF product. The new data, Conditions for runaway generation. Conditions for disrup-
including that from Alcator C-Mod, also shows little evidence tive runaway generation and/or amplification (avalanche con-
for large halo current asymmetries, TPF >2. Progress in version) in present tokamaks continue to receive systematic
the development of magnetic reconstruction and simulation study. The previously noted tendencies reported in the IPB for
methods for assessing plasma evolution and halo region disruptive runaway generation to be more prevalent in larger-
dynamics is providing new data for calibrating predictive scale tokamaks and to also be positively correlated with higher
models for ITER. Continuing studies of the vertical and toroidal field and/or higher plasma edge safety factor have
radial forces on the ASDEX Upgrade and JET vacuum vessel been clarified by new data. Recent large-tokamak observa-
systems observed, respectively, during disruptions in ASDEX tions of the BT and edge safety factor domain for disruption
Upgrade and VDEs in JET, demonstrate basic understanding runaway electron generation [343,344] have clarified the find-
of the magnitude of the vertical loads seen in both cases, ings noted in the IPB. Runaway electrons are observed for
but less complete understanding of the radial and/or toroidal BT > 2.2 T and qeff > 2.5 in JT-60U, and for BT > 2.2 T and
asymmetric vertical forces seen in the JET vessel during VDEs q95 > 2.5 in JET (typically, in JT-60U, qeff ≈ 1.25q95 ). A
with significant toroidal asymmetries. similar
∼ 2 T threshold dependence on BT is also observed
in Tore Supra [345]. While definitive understanding of how the
BT threshold for runaway generation scales with plasma size
3.4. Runaway electrons generated by disruptions
(R and/or a) remains as a future R&D task, these observations
The production during disruptions of relativistic (runaway) and anecdotal data indicating increased tendency towards the
electrons poses a potentially serious threat to the integrity disruptive generation of runaways in ‘large’ versus ‘medium’
of ITER plasma-facing-component surfaces. ITER, like and ‘small’ tokamaks all suggest that disruption-generated run-
any high-current reactor-regime tokamak, will be inherently aways must be expected in ITER.
susceptible to efficient conversion, by Coulomb avalanche
Direct observation of runaways. Direct observations of
multiplication, of plasma current to relativistic (runaway)
disruption-generated runaway electrons in flight have been
electron current [1, 289, 290]. Such conversion, of up to 70%
obtained in JET by application of soft x-ray diagnostics [346].
of the initial plasma current, is predicted to occur following The resulting SXR data, which dominantly show line radiation
a naturally occurring disruption, an artificially induced fast of metallic impurities excited by impact of runaway electrons,
plasma shutdown or a loss-of-control vertical displacement allow the dynamic evolution of the plasma runaway content
event (VDE). The subsequent uncontrolled interaction of this to be observed directly. The JET data show that the runaway
magnitude of runaway current with PFC surfaces has the electrons are generated near the centre of vacuum vessel within
potential to produce local damage to PFC surfaces and their a small minor radius, and then subsequently move towards first
underlying substrate structures. wall (figure 49). The profile of the runaway beam is observed
Given this inherent susceptibility of ITER to runaway to be Gaussian and its diameter (FWHM) is determined to be
conversion, plus the already-documented instances of ∼37 cm (compare with an initial thermal plasma diameter of
significant runaway generation and avalanche multiplication ∼180 cm). Since the local SXR emission is proportional to
and sometimes even PFC surface damage in the present runaway current density, the q profile can be evaluated: the
(TFTR/JET/JT-60U) generation of large tokamaks, there is data indicate q ∼ 0.5 at the centre of the beam and q ∼ 3 at
urgent need to clarify the properties of ‘disruption-generated’ the edge. Such information of q profile is essential to examine
runaway electrons and to establish methods for avalanche the beam stability.

S178
 Chapter 3: MHD stability, operational limits and disruptions

j(r,t) / j(0,0)
Figure 50. Numerical simulation of the current profile during the
runaway conversion phase of a disruption in JET. As is typically
observed in experiments, the simulation indicates that the profile of
the runaway current is much more peaked than the original current
profile (shown by the line at t = 1 s) [348].
Figure 49. Observation of soft x-ray image of runaways in JET. The
downward motion (towards the divertor) is clearly seen. The
runaways are first generated 4 ms after the start of the
3.4.2. Interaction of runaway electrons with plasma facing
disruption [346]. components. The SXR images in JET show that the poloidal
width of the wall-interaction (runaway impact) region is
relatively narrow: about 10 cm [346]. Observation of discrete
Theory of runaway dynamics. Progress has been made in
pulses (bursts) of hard x-ray emission during the impact period
calculating the evolution of the runaway population and the
suggests that the runaway current channel itself is filamented.
electric field during a disruption in a self-consistent way
Similar indications (derived from IR TV data) of poloidally
[347–349]. These calculations reproduce the experimentally
localized runaway impact and localized first-wall heating
observed conversion of thermal current to runaway current and during VDEs are reported for JT-60U [319]. In the case of
thus provide a quantitative link between theory and experiment. the JT-60U data, the first-wall heating and HXR indications of
In JET, where typically half of the thermal pre-disruption runaway loss are attributed to the effects magnetic-fluctuation-
current can be converted into runaways, calculations indicate produced runaway transport (see section 3.4.3 below). In Tore
that much of the final runaway population is already being Supra, local impingement of runaway electrons on the outboard
generated by the avalanche mechanism [344]. Numerical limiter was visible in video images [351].
simulations for ITER (Ip = 15 MA) employing the same basis The runaway filamentation and impact region size
suggest that about 70% of the ITER thermal current will be observations made in JET and JT-60U support the concerns
converted to runaway current. These results are in reasonable previously identified in the IPB about the possible PFC-
accord with previous predictions, of up to 80% conversion, damaging effects of uncontrolled termination of a runaway
cited in the IPB for the 21 MA ITER design. discharge in ITER. While some uncertainty exists with regard
The present theoretical calculations also show that post- to the exact details of the material damage caused by runaway
disruption current profile becomes more peaked than the electron losses, repetitive current quenches with significant
pre-disruption current profile (figure 50) [33, 348]. Previous conversion to runaway electrons seems unacceptable in ITER.
and present indirect observations of the runaway current Simulations using the DINA code [352] indicate a runaway
channel size and/or profile (e.g. li ) and the JET SXR imaging electron energy content of ∼50 MJ if runaway electrons are
data all support this prediction. Peaking has important confined until q = 1 is reached at the edge. However, if
implications, especially in ITER, for the axisymmetric control runway electron expulsion at q ∼ 2, as observed in JT-
and equilibrium stability of the runaway beam. The reason for 60U [319, 343], is considered, the energy content will be
peaking is that the toroidal electric field diffuses into the centre lower (a likely maximum of ∼25 MJ). The duration of the
of the discharge where runaway production is most rapid. This runaway electron energy pulse in ITER is estimated to be 130–
diffusion mechanism is found to be much more important than 230 ms depending on the plasma electron temperature during
the one due to toroidal effects [350] mentioned in the IPB. the plasma limiter configuration phase of an uncontrolled VDE.
The theoretically predicted peaking probably explains why the Elementary estimates for ITER based upon 50 MJ runaway
central safety factor in JET is q ∼ 0.5. energy content, and the measured peak energy deposition of
Finally, the theoretical calculations suggest that, owing 15–65 MJ m−2 [1, 346] give a lower bound on the effective
to the high sensitivity of the runaway production mechanisms deposition area of ∼0.8 m2 . This relatively small area of
to local plasma parameters, the runaway current profile can energy deposition is consistent with expectations due to the
easily become radially filamented. Note the ‘hot spots’ small poloidal extent of the runaway electron beam and the
observed in the JET SXR data in figure 49; these and the imperfect toroidal alignment of affected wall section.
other JET observations noted below are likely indications of Owing to vertical instability of the runaway current
filamentation. channel, the runaway electron energy will be likely be

S179
T.C. Hender et al

Studies of this aspect of runaway confinement and loss,

typically at relatively low current (1 MA), have been pursued
in JT-60U. The ability of the JT-60U control system to maintain
a vertically stable plasma and divertor configuration during
the post-disruption phase [354] allows investigation of the
subsequent runaway current decay and loss phase. It is
found that the post-disruption runaway current tends to decay
slowly and smoothly, even with zero loop voltage [343]. It
is also confirmed that negative loop voltage increases the
decay rate. During the decay phase, it is theoretically found
that the observed gradual loss of runaway electrons can be
explained by the combined effects of pitch-angle scattering and
synchrotron radiation, and this damping can be more efficient
than the classical Coulomb-scattering collisional drag [355].
The characteristic decay time of runaway current in JET, 1–2 s,
is also explained by these damping mechanisms. In JT-60U,
Figure 51. Numerical simulation of the ITER first-wall temperature the temporal behaviour of the runaway current can be explained
(in ◦ C), just after energy deposition by 10 MeV runaway electrons,
with 50 MJ m−2 , deposition time = 0.1 s. From the lefthand side, by the balance of avalanche generation and slowing down due
the simulation geometry comprises 10 mm of beryllium armour, to these damping mechanisms [356].
22 mm of copper heat sink and a 10 mm inner-diameter copper The JT-60U experience demonstrates that when the
cooling water tube (lateral spacing = 28 mm). The ∼2 mm thick plasma position and shape control can be maintained after the
grey zone indicates material attaining temperature larger than the disruption and formation of a runaway current channel, it is
beryllium melting point.(Reprinted from [353]. Copyright 2003,
with permission from Elsevier). then possible to effect a slow and controlled current termination
that can act to benignly terminate runaway electrons and any
remaining plasma thermal current.
deposited on the upper or lower first-wall surface or possibly
In JT-60U, it is demonstrated that the runaway electrons
within the divertor. Hence melting of the beryllium or tungsten
that are otherwise well-confined during the after-disruption
armour, or ablation of graphite must be considered. The
phase are exhausted at an increased rate owing to the
runaway energy will be deposited on a thin surface layer
presence of the large-amplitude magnetic fluctuations that
determined by the electron stopping power and the angle of
spontaneously appear when the edge safety factor qs becomes
incidence: the deposition depth is estimated to be ∼2.5 mm for
small (2 to 3) (figure 52) [319, 357, 358]. The runaway-loss-
beryllium and carbon and ∼0.2 mm for tungsten. This energy
enhancing effect of magnetic fluctuations, already reported
deposition leads to melting of the material in both beryllium
in the IPB, is confirmed by more recent simulations which
and tungsten in the optimistic 15 MJ m−2 estimate, and melting
demonstrate that magnetic islands having widths expected
plus ablation in the pessimistic 65 MJ m−2 estimate.
during disruption cause a collisionless loss of relativistic
Similar conclusions are obtained from Monte Carlo
electrons owing to enhancement of the stochasticity of their
code simulations of PFC volumetric energy deposition and
relativistic motion. This enhancement can explain the
temperatures (figure 51) [353]. The simulations suggest that
enhanced runaway loss experimental observations reported in
several kg of molten material can be produced (and likely
JT-60U [359] and also the tendencies noted above for runaway
mobilized by gravity and J × B forces) by a single runaway
‘generation’ in large tokamaks to be inhibited at low BT and/or
interaction event. Graphite will undergo some ablation for
low-edge q. Low-edge q occurs naturally in a typical VDE
energy depositions >35 MJ m−2 . Again, while significant
and/or the end phase of a VUD; enhanced fluctuation losses
uncertainties about deposition areas and time scales and effects
seem a likely explanation for the BT and/or low-q runaway
of runaway beam filamentation and local PFC misalignments
generation thresholds observed in large tokamaks.
exist, frequent occurrence of such an uncontrolled runaway
In JT-60U, an injection of impurity neon pellets into a
impact event would appear to be unacceptable.
post-disruption runaway plasma, caused a prompt exhaust of
runaway electrons from the plasma and a reduction in runaway
3.4.3. Confinement, termination and mitigation of runaway plasma current, without any large-amplitude MHD activity
electrons. The fact that runaway electrons are well confined [356]. Since the bulk electron temperature is estimated as
in an otherwise MHD-stable tokamak discharge, even very low (∼10 eV), the pellet ablation time would be very
following disruption, is well known, and many tokamaks long (∼10 s). However, an increment in electron density was
observe persistent ‘runaway tails’ in the after-disruption observed at the pellet injection. Therefore, it is likely that
current decay waveform. In cases where the plasma/runaway the runaway electrons themselves play an important role in
current equilibrium is otherwise well controlled, in large the pellet ablation, but theoretical and modelling studies to
tokamaks multi-second post-disruption ‘runaway plasmas’ can understand this ablation process, together with the exhaust
be obtained in this manner. In many cases, the eventual loss of dynamics of runaway electrons, are needed.
runaways to the PFC surfaces appears to be governed mainly Observations shown here were found mainly in JT-60U,
by the gradual (or abrupt) decay of the plasma equilibrium. where well-confined post-disruption runaway plasmas are
In direct terms, the runaways are confined until equilibrium stably formed. Such plasmas are suitable to simulate runaway
decay or control loss ‘dumps’ them on the PFC surfaces. plasmas in ITER and to investigate mitigation techniques

S180
 Chapter 3: MHD stability, operational limits and disruptions

Runaway termination should occur, for typical tokamak parameters, when δBr /B ∼
start finish 10−3 , both in ITER and in existing experiments. Further
study in present large-tokamak experiments under conditions
3
qs where appreciable avalanche gain is otherwise evident is
qs

2
(a) required to assess the ability of magnetic fluctuation losses
1 to partially or completely offset avalanche multiplication in a
-dIp/dt (MA/s)

200
(b) 0.8 high-avalanche-gain plasma.
100

Ip (MA)
• The application of an ‘equilibrium sustenance and Ohmic
Ip
0 Ip 0.6 ramp-down’ method, supplemented by high-Z pellet injection,
to terminate ITER runaways has been studied computationally
-100 0.4 [364]. The studies demonstrate the possibility (subject to
B (T/s)

0.08 • PF system voltage, current and power limitations) to effect

B (n=1) a controlled ramp down of the currents of the poloidal
•

0.04 field coils, in a manner that applies a negative loop voltage

(c) (to reduce runaway energy) while simultaneously reducing
0 vertical elongation and maintaining vertical and radial position
4
γ (104s-1)

(d) control. The injection of a succession of small high-Z

γ pellets to enhance radiative and collisional losses during the
2
combined thermal and runaway current ramp down is also
proposed. During such a slow termination, there is time
0 to apply additional techniques (e.g. pellet injection [356] or
4
3 (e) enhancement of natural magnetic fluctuation levels [362]) for
m

2 the mitigation and termination of runaway electrons. The

m
1 possibility of mechanical intervention, i.e. a gradual insertion
IHX (a.u.) Deposited power

0 of a movable limiter, also exists.

400 Deposited power (f)
(MW) High-density mitigation. The runaway mitigation scenarios
200 and methods cited above focus on achieving a benign (or
0 acceptable) gradual termination of an already-established
runaway discharge. The estimated time scale for effecting
4000 I HX (g) such a termination in ITER is likely to be
∼10 s. An
2000 alternate approach of providing a large increase of the plasma
electron content, by a factor of
100, before or during
0
the current quench phase of a disruption or VDE, has also
2.934 2.936 2.938 2.940 2.942
been proposed [1]. The basis of this approach lies in the
Time (s) observation that at a sufficiently high electron density (free +
Figure 52. Temporal evolution of (a) surface safety factor, bound), the ratio E/Ec becomes less than unity and there is
(b) plasma current and its decay rate, (c) magnetic fluctuations with no avalanche multiplication [290]. Here E is the in-plasma
toroidal mode number n = 1, (d) growth rate γ of each spike in the toroidal electric field and Ec = (4π e3 ne ln/mc2 ) is the critical
n = 1 mode, (e) poloidal mode number m, (f ) deposited power on electrical field that governs whether or not the avalanche grows.
the inner divertor plates and (g) hard x-ray emission, during
Elementary considerations developed in the IPB show that
runaway termination in JT-60U [319].
the corresponding critical electron density, denoted here as
the Rosenbluth ‘no-avalanche’ density, nRB , is given to a
for runaway electrons. Recently, stable formation of the reasonable approximation by
post-disruption runaway plasma is reported from FTU [360]
and Tore Supra [351]. Their waveforms are very similar nRB (1020 m−3 ) ∼
= 11E(V m−1 ).
to those in JT-60U, which indicates that the physics of the
post-disruption runaway plasmas are general among different Combining this estimate with the further estimate, based upon
devices. Further inter-machine experiments and comparisons dissipation of the plasma internal flux, µ0 Rli /2, in the minimum
are strongly encouraged to obtain an even better understanding current quench time tCQ /S ∗ ∼ = 1.7 ms m−2 (see section 3.3.1),
of the post-disruption runaway plasma. leads to the ‘no-avalanche’ estimates for ITER (given in table 5
Fluctuation losses of sufficient magnitude can act to of the introduction to section 3) of E = Ec ∼ = 38 V m−1 and
reduce or eliminate avalanche gain. If electrons undergoing ∼ 22 −3
nRB = 4.2×10 m . As table 5 demonstrates, similar values
runaway acceleration are lost before they have had time to apply for the present (e.g. JET) and even larger reactor scale
produce knock-on secondaries, they do not contribute to the tokamaks (e.g. ITER-EDA).
avalanche growth [361, 362]. This has been shown both As the IPB discussion notes ‘if runaways are to be
numerically [342, 363] and analytically [363]. The exact unconditionally avoided. . . the electron density must be quite
magnetic fluctuation amplitude necessary to suppress runaway high. . . ’ and possibilities of achieving the required ∼100-fold
avalanches depends on the mode structure and spectrum of increase in density by means of the injection of single or
the fluctuations. However, the theory suggests that these multiple solid pellets and by injection of liquid jets were
dependences are fairly weak and that avalanche suppression identified. The likely need, confirmed by impurity injection

S181
T.C. Hender et al

simulations [290], for low-Z species injection (H2 /D2 or He throughout a plasma volume early in the current quench, since
or possible Be) versus medium or high-Z injection to ensure this is the critical time to suppress an amplification of seed
runaway conversion avoidance in ITER-EDA was also noted. runaway electrons from whatever source. Adding gas after the
Massive gas injection (MGI) ‘disruption mitigation’ formation of the confined runaway electron beam is much less
experiments (see section 3.6.3), effected with subsonic gas effective with respect to suppressing runaway electrons [372].
jet systems, have demonstrated the feasibility of achieving
after-injection electron densities that may have approached the
corresponding Rosenbluth no-avalanche density. Experiments 3.4.4. Summary and implications for ITER. Experiences
with massive (∼350 000 Torr L s−1 ) He injection in DIII-D with post-disruption runaway generation and parametric
[365] and subsequent experiments conducted during 2000– explorations of the conditions for runaway generation in the
2001 with similarly massive Ne and/or Ar injection present JET/JT-60U generation of divertor tokamaks, and past
(∼100 000 Torr L s−1 ) [324,366] resulted in attainment of line- experience in the TFTR limiter tokamak, all indicate the
averaged ne approaching 1021 m−3 and also reduction of the likelihood that disruptions, VDEs and even moderately fast
divertor thermal quench loading, peak halo current magnitude plasma shutdowns in ITER will generate high magnitudes of
and toroidal asymmetry (see section 3.6.3). In JT-60U, similar runaway electron current. Observations in JET and theoretical
experiments with the mixed injection comprising a small analysis confirms the important role of Coulomb avalanche
amount of noble gas (argon, krypton, and/or xenon) and a larger multiplication, even under present moderate-gain conditions.
amount of hydrogen gas verified the feasibility of injection Extrapolation to the high-gain ITER regime implies a ∼70%
schemes with the flexibility to adjust the species mix and ratio conversion of plasma current to ∼10 MeV runaway current.
of injected gases [367, 368]. Natural or uncontrolled rapid loss of this magnitude of runaway
In DIII-D, the He and Ne/Ar MGI plasma shutdown current to localized portions of the ITER first-wall or divertor
experiments resulted in fast plasma current shutdowns that did surfaces has a self-evident potential for causing damage.
not produce a detectable indication of runaways. Comparison Experience with runaway confinement in present
experiments with Ar pellet injection did produce runaways tokamaks indicates that natural loss processes (those occurring
[342]. Analysis of the He injection data showed that the at normal plasma densities and/or with typical pre-disruption
resulting MGI-initiated current quench plasma exhibited the levels of MHD fluctuation) will be slow (10–100 s) in ITER.
classical Spitzer resistivity [369, 370]. This finding indicates Hence the ability of the ITER plasma equilibrium control
that ionization and energy balance calculations of Te and Zeff system to maintain adequate control during the thermal
can be reliably used to calculate the parallel electric field for current to runaway current conversion process and during
estimating runaway electron generation and amplification. the subsequent (gradual) runaway termination phase will be
Application of the same ionization and energy balance critical. Beyond the elementary PF and plasma control system
methodology to the 2000–2001 DIII-D experiments by Whyte requirements of being able to maintain equilibrium control (see
et al suggested that, given the premise that the measured chapter 8 of this issue [373]) of the runaway column, and to
quantity of injected Ar (or Ne) was more-or-less uniformly then effect a gradual (∼10–100 s) runaway current termination,
mixed throughout the ∼20 m3 plasma volume by the end there also appears the need to take additional action(s)—e.g.
of the current quench, the resulting Ar- (or Ne-) dominated by application of a negative loop voltage and/or injection of
plasma density (nimp ∼ 2 × 1021 m−3 , equivalent to ne ∼ impurity pellets to exhaust runaways and/or by injection of
2–4 × 1022 m−3 ∼ = nRB ) would be only weakly ionized, E/Ec high-Z gas to promote enhanced collisional losses and/or by
would approach unity and the resulting runaway amplification introduction of a movable limiter and/or by artificial generation
would be small [324, 366]. This could explain the observed of enhanced MHD fluctuations—to effect a benign runaway
lack of runaway production. However, subsequent, further Ar
shutdown.
injection experiments, conducted in 2004 with an improved
Alternatively, injection of massive quantities of neutral
‘high-intensity’ gas injection system [371], clearly showed that
particles (and electrons), either prior to onset of disruption
the injected Ar does not penetrate (in neutral form) more than
(thermal quench) or immediately after thermal quench (within
a few centimetres into the plasma. Subsequent evaluation of
the injection flow data also showed that, owing to the finite 10 ms after thermal quench onset), theoretically offers a means
rise time of the injection flow, the quantity of gas delivered to to unequivocally avoid runaway generation. However, for the
the plasma by the end of the current quench was significantly technique to be fully successful, plasma electron densities in
less than the total quantity that was injected. Thus the ability excess of 1022 m−3 must be achieved before an appreciable
of MGI to collisionally suppress runaway electron avalanche current decay occurs. While present experiences with massive
remains an area of ongoing study. gas injection are modestly encouraging with regard to attaining
The observations reported in [371] also show the central high electron densities, and while the technology to implement
role of MHD fluctuations in effecting mixing and subsequent MGI hardware for ITER is straight forward, further study of the
transport of the edge-ionized neutral gas into the plasma core. underlying neutral penetration, particle/electron assimilation
Optimization of the injection system to ensure that the desired mechanisms and of the role of MHD fluctuations in effecting
quantity of gas is delivered to the plasma in a time that is mixing in an ITER-scale tokamak is required. The possibility
 tCQ (as short as 4 ms in DIII-D) is also important. These that the enhanced level of MHD fluctuations that MGI produces
injection system optimization matters are addressed further in may also act to at least partially offset the predicted Coulomb
section 3.6.3. avalanche gain also needs careful consideration.
In a high-avalanche-gain experiment such as ITER, it is Parallel development of more-penetrating neutral particle
essential that a sufficient injected electron content be present delivery means, i.e. high-density supersonic gas jets, liquid jets

S182
 Chapter 3: MHD stability, operational limits and disruptions

and/or large-scale pellet injectors with a fast multi-pellet burst from the experimental ‘database(s)’ that have been described
capability, would also seem prudent. in preceding sections. The limitation on direct extrapolation
Finally, it is important to understand that the susceptibility is due in part to the different combination of plasma geometry
of ITER (and all presently envisioned after-ITER reactor and surrounding passive structures in ITER (as contrasted with
tokamaks) to runaway conversion and PFC damage owing to geometry and structures in existing machines) and in part to the
uncontrolled runaway current termination arises solely owing differences in energy levels and underlying physics processes
to the high-current (
10 MA) nature of these devices. Hence that arise from the increase in ITER plasma size, current level
having a ‘solution’ to ITER’s runaway ‘problem’ is important and energy content.
as soon as current levels approach 10 MA, and having reliable Much of the integrated modelling or simulation of
runaway conversion avoidance or mitigation strategies will be the effects of ITER disruptions and VDEs has focused
essential for a DEMO class tokamak. on assessments of the basic plasma equilibrium dynamics
and the resulting electromagnetic consequences. A self-
3.5. Integrated modelling and simulation consistent simulation of the plasma equilibrium dynamics is
essential to obtain meaningful estimates and time histories
The preceding sections approach the phenomenology, data of induced toroidal currents in the vacuum vessel, eddy and
and physics bases for disruption and disruption effects in halo currents in the in-vessel components and estimates of
a topical manner, as if the phenomenology and effects are the location and poloidal extent of plasma–PFC interaction
separable. The same topical basis for presentation is used in regions. Simulations of the VDE and during-disruption
the IPB. However, as the discussion presented therein makes equilibrium dynamics for both present tokamaks and ITER
clear, there are cause-and-effect connections among all the have typically been developed using the DINA [338, 374]
observable attributes of disruptions. Thus interpretation of and TSC [364, 375] dynamic equilibrium codes. Here, the
disruption data from present tokamaks and predictions of what simulation results for ITER developed on the basis of the DINA
will happen in ITER ultimately require the use of some degree code are mainly described.
of an ‘integrated’ model. At the very least, the dynamic As the discussion given in section 3.3 details, the plasma
evolution of the plasma equilibrium configuration and MHD current decay time and waveform, and the ensuing dynamics
stability—from initial precursor MHD instability growth to of the vertically unstable plasma configuration evolution,
final decay of the plasma current—needs to be explicitly influence the magnitude, location and time history of induced
taken into consideration. For predictive modelling of ITER vessel currents and in-vessel-component eddy (circulating)
disruptions and VDEs, it is also necessary to incorporate or currents. In addition, the current decay rate and equilibrium
parametrize certain further aspects of the underlying plasma dynamics influence the magnitude and toroidal asymmetry of
energy balance and current composition (e.g. thermal-to- the in-vessel halo currents. While it is possible to implement
runaway current conversion in the integrated model). Finally, physical models that can calculate core and halo plasma
it is also necessary that models for ITER incorporate an temperature and resistivity (and hence current decay rates)
accurate representation of the torus vacuum vessel, in-vessel on a first-principle joule-input/radiation-loss basis, a first-
components and the ex-vessel PF coil system and, in some principle ability to predict plasma impurity content during
cases, the before- and after-disruption actions of the PF control disruption is presently lacking. Hence the past and present
system (see chapter 8 of this issue [373]). modelling of ITER disruption and/or VDEs typically proceeds
Discussion of the physics basis aspects of a complete on the empirical basis (for DINA) of adjusting the model
integrated model can be found in [1]. This subsection discusses parameters to yield the prescribed thermal energy loss and
recent advances made—in most cases, using ‘integrated’ current decay rate or waveform (see figure 53). A similar
dynamic equilibrium models—in the self-consistent numerical parametric adjustment procedure (wherein the plasma energy
simulation of ITER disruptions and VDEs and their resulting loss and resistivity models are adjusted) can be applied to
EM loading consequences. The use of a variety of obtain a prescribed thermal energy loss and/or current decay
integrated models for disruption halo current and runaway time or an approximate waveform in TSC modelling. In both
data interpretation in present experiments, or simulations of the cases, it is important to recognize that, to some extent, the
disruption mitigation methods, are described elsewhere in resulting predictions of equilibrium dynamics, halo currents
sections 3.4–3.6. Here, representative examples that contribute and eddy currents have residual sensitivities to the underlying
for reaching the conclusions noted in previous sections about modelling basis ‘input assumptions’.
the structural integrity of ITER vacuum vessel and other in- Since the DINA code (and the TSC code) provide
vessel systems are presented. Progress in incorporating more- 2D (n = 0) plasma equilibrium modelling, only toroidally
fundamental 3D MHD instability considerations into these symmetric halo currents can be evaluated. For the subsequent
types of simulations is also addressed and a brief summary EM load analyses of the structural effects of in-vessel halo
of the present capabilities and future development needs for currents, 3D effects (toroidal asymmetries) are introduced in
integrated modelling is given. an after-the-fact manner by applying a toroidal peaking factor
that is empirically derived from the halo current database (see
section 3.3.2) [1, 335].
3.5.1. ITER disruption simulations. Prediction of the
Two scenarios for disruption are simulated.
characteristics and consequences of ITER disruptions must
be performed using an integrated simulation model, since • Major disruption (MD), which starts from a thermal
the detailed behaviour of the plasma during disruptions and quench and q profile flattening when plasma is at its
VDEs cannot be directly predicted by simple extrapolation nominal position. This is followed by a fast reduction of

S183
T.C. Hender et al

MD Table 9. Major assumptions used in ITER disruption simulations.

20 5
Parameter MD VDE
(a)
−1
4 Current quench rate (time) 375 MA s 375 MA s−1
15 (b) (40 ms) (40 ms)
Z Current quench waveform Linear Linear
Current (MA)

Ip (c) Thermal quench duration 1 ms 1 ms

3
Beta drop during vertical movement 0 ≈0.3

Z (m)
(d) (before thermal quench)
10
(e) Surface q value at thermal quench 3 1.5–2
2 Decrease of li during thermal quench 0.15–0.2 0.15–0.2
(f) Te, halo /Te, core 1 1
5 (g)
1
Ihalo pol plasma equilibrium with both closed (core) and open (SOL)
magnetic surfaces. The plasma equilibrium modelling is
0 0 achieved in combination with an axisymmetric model of the
0 10 20 30 40 50 60
(a) Time (ms) external circuit (PF coils and surrounding conducting vacuum
MD vessel and in-vessel structures). Flux-surface-averaged plasma
5
energy transport equations are also solved simultaneously.
(g)
The resulting integrated model has been validated in several
(f) tokamaks [377, 378].
(a)
The DINA disruption simulation incorporates a detailed
(e) axisymmetric representation of the ITER vacuum vessel
(b) and in-vessel systems. The vacuum vessel is modelled by
Z (m)

0 (d)
a set of thin plates with relevant resistance, so that the
global L/R time can be matched with that calculated for the
actual geometry. Blanket modules are modelled by a set
(c) of axisymmetric current-carrying rings, with opposite rings
connected in the toroidal direction so that a zero net toroidal
current flow is obtained. Resistances of each pair of rings are
selected to yield an equivalent penetration time for poloidal
-5 magnetic fields through a blanket that is calibrated with a 3D
2 3 4 5 6 7 8 9 10
analysis of eddy current dynamics obtained from finite-element
(b) R (m) simulations.
Figure 53. Time evolution of (a) plasma and poloidal halo currents, During thermal quench, conservation of helicity is
(b) plasma boundary (LCFS), for the MD case in table 9. invoked to calculate the resulting redistribution of the plasma
current profile. After-thermal-quench energy transport is not
plasma current. In this case, the ex-plasma halo begins calculated and both plasma core and halo region electron
temperatures are assumed to be uniform, with a prescribed
to form immediately following the end of the thermal
value. A temperature ratio for the plasma and halo areas
quench.
can be specified (see table 9). During the current quench
• Vertical displacement event (VDE), which starts from
phase, the width of the halo region is determined by a simple
a vertical displacement due to loss of vertical position
model, benchmarked against JT-60U disruption data [379], that
control. Both downward and upward plasma vertical
is based on conservation of toroidal flux within the plasma
movements are considered. After the plasma boundary cross-section and which is taken to include the halo region
touches the wall, the value of q at the plasma boundary once vertical instability develops. Application of this flux-
decreases as the plasma cross-section area diminishes. conservation model to ITER yields a halo region evolution
When the boundary q reaches a specified value q= 1.5– that mimics what is seen in JT-60U, wherein the halo spatial
2.0 [327, 376], thermal quench takes place triggering a width is observed to increase gradually as the plasma moves
fast reduction of the plasma current. For this case of downwards during a VUD.
disruption, the halo area shows up just after the formation A somewhat different type of halo evolution is observed
of a limiter configuration. in DIII-D, where observations and simulations indicate that
the halo spatial width stays approximately constant during the
The initial plasma equilibrium is taken from the 15 MA
course of a VUD [331, 369]. While the difference between
ITER ‘reference’ inductive scenario (li = 0.85, βp = 0.7).
the two classes of halo-width observations is not large, and
Major assumptions used in the two simulation cases are
while it is likely that both types of observations are consistent
summarized in table 9.
with a common flux-conservation and diffusion physics basis
model, further physics basis understanding and development
3.5.2. DINA disruption modelling basis. For disruption and of a more-predictive halo model is required. Such a model
VDE modelling, an extended ‘disruption simulation’ version should incorporate internally self-consistent temperature and
of the basic DINA code has been developed. The extensions conductivity profiles for the halo region, to allow refined halo
provide the ability to simulate the dynamic evolution of 2D width and halo current magnitude predictions for ITER.

S184
 Chapter 3: MHD stability, operational limits and disruptions

3.5.3. Predictions for major disruptions and VDEs in 20 0

ITER. Since the impact of vertical displacements on the (a)
machine, due to eddy and halo currents, is more severe for
downward displacement (towards the X-point) than for upward
15 -1
displacement, the initial plasma position in ITER is specified
to make the plasma movement upwards. The initial position, Ip

Current (MA)
(b)
where the plasma is balanced and will move randomly upwards

Z (m)
or downwards at disruption, is known as the neutral point [334]. 10 (c) -2
The neutral point of up–down asymmetric, single-null diverted
plasmas, however, depends on the change of li and βp at
thermal quench [330], implying that no unique neutral point
5 Z -3
exists. Similarly the present initial plasma position in ITER is Ihalopol
chosen to ensure that the upward movement is obtained for a
wide range of li and βp changes. This has been confirmed
by systematic simulation studies for a range of MD cases, 0 -4
660 680 700 720
and for example, figure 53 shows the time evolution of the
plasma and poloidal halo currents and the plasma boundary (A) Time (ms)
(last closed flux surface) for an MD case simulated by the DINA 1
VDE downward
code. It is found that for the changes of li and βp at thermal 4 (a)
quench prescribed in table 9, the plasma moves upwards after 15
0 3
the thermal quench. The maximum in-vessel (poloidal) halo
(b)
current reaches about 1.5 MA (∼10% of Ip0 ).
Figure 54 shows the time evolution of the plasma and -1 2 16
poloidal halo currents, plasma vertical position and (b) plasma
(c)
core (LCFS) and halo boundaries for a downward VDE case.
Z (m)

-2 1 17
A small initial downward perturbation is provided at t = 0,
18
and the plasma moves downwards, without vertical control,
-3
before the thermal quench occurs at t = 670 ms. The peak
halo current for this VDE reaches about 3 MA (∼20% Ip0 ).
-4

3.5.4. Calculation of non-axisymmetric halo current effects

in ITER by M3D code. Full three-dimensional simulations -5
3 4 5 6 7 8 9
of VDEs in ITER have been carried out using the M3D
(B) R (m)
code [380, 381]. Scaling of the VDE vertical growth rate
proportional to wall resistivity has been verified. Simulations Figure 54. Time evolutions of (A) plasma and poloidal halo
have been done of disruptions caused by large-inversion-radius currents, vertical position (Z) and (B) plasma LCFS (closed ellipses
internal kink modes, as well as by the non-linear growth of labelled ‘(a)–(c)’) and halo boundaries (open curves labelled
resistive wall modes. ‘(a)–(c)’) for the downward VDE case [327].
The M3D code includes resistive wall boundary
conditions, which match the solution inside the resistive wall calculated toroidal peaking factor (TPF) for the poloidal halo
to the exterior vacuum solution. The exterior problem is current can be as large as 3 but typically tends to about 2. This
solved with a Green’s function method, using the GRIN code range of TPF matches the observations for present experiments
[382]. The M3D code has a time-dependent self-consistent in the halo current database (see section 3.3.2).
resistivity model and realistic ITER geometry including the Simulations have also been made of disruptions caused
magnetic separatrix. Thermal conductivity in the presence by resistive wall modes, which, in turn, are triggered by the
of the separatrix provides an adequate temperature contrast application of static non-axisymmetric magnetic error fields.
between the core and the halo region (the open field line region A parallel viscosity model has been introduced in M3D to
outside the separatrix). provide the dissipative coupling needed to reproduce resistive
The M3D simulation capabilities allow modelling of both wall mode physics. The model conserves toroidal angular
the disruption and VDE phases of a VUD, where disruption momentum and is consistent with equilibria with toroidal flow
causes a thermal quench, which in turn causes a current quench, in which the equilibrium angular frequency is a flux function.
vertical instability and halo current development. The example When the external magnetic perturbations are applied, the
in figure 55 shows a VUD initiated by an unstable large- parallel viscosity causes a damping of the toroidal flow, growth
inversion-radius internal kink mode. When the internal kink of the plasma RWM and onset of disruption.
instability becomes sufficiently non-linear, toroidal coupling to The example simulations presented here illustrate the
other modes causes disruption. The plasma cools because of potential of the M3D model, with incorporation of appropriate
transport along stochastic field lines. This raises the resistivity ‘auxiliary’ MHD and energy transport physics models, to
and dissipates the current. The current decay is accompanied simulate the full sequence of events for ITER-like VUDs and
by a VDE and halo current development. The VDE growth also hot-plasma VDEs. Validation of the overall simulation
rate is at least twice as fast in the presence of disruption. The capabilities and/or auxiliary models, by comparison with

S185
T.C. Hender et al

3.6. Disruption avoidance, prediction and mitigation

The importance for ITER of avoiding, wherever possible,
the occurrence of disruption or loss-of-equilibrium-control
events that lead to disruption is universally acknowledged.
Similarly, the need for mitigation strategies that minimize
or eliminate the adverse consequences of disruptions is well
known. Finally, there are operational reasons—mostly related
to protecting the integrity and continued function of PFC
components—why a fast plasma power shutdown capability
is needed. The combination of these considerations leads to a
compelling need to develop disruption prediction, avoidance
and mitigation strategies that can be effected with high
reliability, while at the same time allowing ITER scientists
to conduct exploratory plasma development and optimization
studies in burning plasma regimes where the risk of disruption
will be appreciable.
The needs and considerations noted above have been
identified and discussed in the ITER Physics Basis [1].
In the three sections that follow, recent progress in (i)
Figure 55. An example of M3D calculation for a VUD caused by a disruption prediction (section 3.6.1), (ii) disruption avoidance
large-inversion-radius internal kink mode. Contours of poloidal flux (section 3.6.2) and (iii) disruption mitigation and fast plasma
at two stages of the VDE development are shown [383].
shutdown (section 3.6.3) are summarized. There is overlap
among these topics and some of the physics basis findings
specific VUD and VDE data from present experiments, already cited in the previous sections of this chapter. Overall
remains a future R&D task. implications for ITER and needs for further research are
summarized in the conclusion of section 3.6.3.
3.5.5. Summary and future needs. The development
and application of dynamic 2D plasma equilibrium models
3.6.1. Disruption prediction. Having a ‘real-time’ capability
for internally self-consistent disruption and VDE data
to predict impending disruption is acknowledged as critical
interpretation, and simulation of ITER disruption and
to being able to operate a tokamak with adequate disruption
VDE scenarios, has reached a reasonably mature state.
protection (to the extent that subsequent resumption of plasma
Modelling/simulation accuracies (self-consistency) seem
operation in a timely manner will not be compromised).
adequate for purposes of providing ITER vacuum vessel and
Here ‘real-time’ denotes the basing of disruption prediction
in-vessel component electromagnetic loading data. While the
basic equilibrium evolution dynamics aspects of the presently- on real-time plasma and device diagnostic data, to provide
available models have been well validated, validation of some sufficient advance indication of impending disruption so that
of the other modelling elements (e.g. halo current dynamics avoidance and/or pre-emptive mitigation measures can be
and core and halo region energy balance) is not as complete taken. The specifics of what constitutes ‘sufficient’ depends
or universal. Hence predictions from 2D models must be on the avoidance and/or mitigation techniques to be employed
accompanied by due recognition of the ranges of outcomes and also on the inherent time scale of the tokamak in question.
that uncertainties in the underlying disruption physics basis, These specifics and requirements for real-time prediction in
or auxiliary modelling element basis, introduce. Needs for present tokamaks and ITER are discussed below.
sensitivity studies to investigate the range of outcomes and for The implementation of a reliable real-time disruption
improvements upon the present empirical or ad hoc nature of prediction or an advance warning method is a challenging
some of the auxiliary modelling bases are also obvious. task that has been investigated in various tokamaks over the
The emerging development of full 3D MHD dynamic last decade. In many experiments, ‘deterministic’ prediction
equilibrium modelling capabilities is promising and offers (warning) methods based upon monitoring of one or two key
prospects for a ‘fully self-consistent simulation’ of ITER indications of impending disruption—for example, locking of
disruption scenarios and consequences. The application of an initially rotating tearing mode and/or bolometric indication
3D dynamic modelling to the interpretation of disruption data of a global plasma energy balance deficit—are deployed and
in present tokamaks will likely aid in sorting out some of the in some instances used in a real-time manner to initiate
presently obscure internal cause-and-effect dynamics of the remedial or pre-emptive actions (see section 3.6.2). While
precursor-growth to thermal-quench phase of disruptions and such ‘disruption avoidance’ methods are often effective for
perhaps also help in clarifying the cause(s) for the wide range their intended application, they do not always reliably predict
of data scatter in current quench and halo current databases. the impending occurrence of disruption over a wide range
Finally, as for the 2D modelling, incorporation of adequate of plasma parameters or for all phases of a discharge.
(and validated) auxiliary modelling elements into the 3D Such methods may also produce undesirable ‘false-alarm’
models will be needed before an ‘ITER disruption simulation’ indications of pending disruption in cases where the plasma
can contain sufficiently realistic physics to make quantitative would not actually disrupt. In situations where alarm
predictions. indications are coupled to pre-emptive action, too-frequent

S186
 Chapter 3: MHD stability, operational limits and disruptions

false alarms can thwart progress in the development of high- #13548

performance plasma operation regimes.
0.8
To address some of the perceived shortcomings of Plasma current (MA)
deterministic prediction, the neural network (NN) technique— NBI power (MW)
0
5
which draws upon a wider range of input data to Gas flux (a.u.)
develop a single disruption predictor output—has been 0
applied. A significant NN development and test effort
Energy (kJ) 500
has been implemented on present experiments, including
ADITYA, ASDEX Upgrade, DIII-D, JET, JT-60U and TEXT. 0
Performance of a NN is typically quantified in terms of Density (10 20 m -3 )
the success rate, SR, the fraction of instances where the 0.5
0
NN successfully predicts the occurrence of disruption or, 1
Time for disruption (s)
alternately, the failure rate (FR) or the missed alarm (MA)
rate (FR = 1 − SR), wherein the NN fails to detect an
impending disruption. The false alarm (FA) rate, where the Alarm level (50 ms)
.
NN indication of pending disruption proves to be incorrect, is . 0
a second important performance attribute. 0.6 1.4 2.2 3.0 3.8
time (s)
Neutral networks require ‘training’: before-deployment
development of the weighting coefficients that control the Figure 56. Neural network prediction for an ASDEX Upgrade
summation of the input signals. This training requires a density limit disruption that follows NBI turn-off [386].
multi-discharge data set—time histories of the relevant input
parameters—for discharges that disrupt. The training is
specific to the NN input data set being utilized and also, at least
to some extent, to the plasma operation mode and operation
attributes of the particular tokamak that the NN is being
implemented on. Once the NN is trained, its performance can
be validated ‘off-line’ against a second (independent) already-
archived data set or tested, either ‘off-line’ or ‘on-line’, against
newly-acquired or real-time data.
NN development work performed at DIII-D used a NN
to predict the value of the βN at the disruption starting
from a 33-input set of data acquired during NBI-heated
discharges [384]. The NN was trained to predict, some tens of
milliseconds in advance, the maximum βN that a given plasma
would reach at the time of disruption. In an off-line validation
test, the NN had a
90% SR in accurately predicting βN at
disruption, but the FA rate was 20% for non-disruptive cases.
(ms)
The same type of ‘simple’ NN method was tried to predict the
density limit disruption in the ADITYA Tokamak [385]. Figure 57. JT-60U: Prediction success rates of the major disruption
In ASDEX Upgrade, a NN predictor was used to trigger caused by the density limit (206 shots), the plasma current ramp
killer pellet injection as a mitigation action for density limit down (31 shots), the low density locked modes (LDLM) (34 shots)
disruptions [386]. The NN output, developed from 13 input and the β-limit (29 shots) [387].
parameters sampled every 2.5 ms, predicted the time interval,
tNN , before disruption. An alarm action was activated for from 9 years of JT-60U operation. The SR for predicting
tNN  50 ms. A database from 99 disruptive shots and 386 disruptions—other than those caused by density limit, high
non-disruptive shots was used for training. In an off-line (open- li during current ramp down or locked modes at low density—
loop, without pellet injection) validation, an 85% SR (55 of 65 with 10 ms advance warning was 97–98%. The FA rate for
disruptive shots in the validation set) was achieved, with a non-disruptive shots was 2.1%. Careful selection of input
1% FA rate for 500 non-disruptive shots. Figure 56 shows parameters and the two-step training method reduced the FA
an example of open-loop alarm indication. The disruption is rate, resulting in a considerable improvement of the overall
predicted more than 20 ms before occurrence. prediction success rate.
In JT-60U, a NN-determined ‘stability level’, calculated The effects of the type (cause) of disruption on the SR are
from 9 input parameters, sampled every 2 ms, is used to predict shown in figure 57. For low-density locked mode disruptions,
the occurrence of disruption [387]. The NN was trained the SR is a satisfactory 100 ms before disruption, despite the
in two steps, first with 12 disruptive and 6 non-disruptive fact that the locked mode data was not included in the training
shots (Step 1) and second (Step 2), with the NN output data set. In contrast, the prediction of the βN limit disruption
data for 12 disruptive shots modified according to the output has a much lower SR, probably because there are no clear
stability levels from the NN trained in Step 1. The resulting precursor indications in the input set that occurs more than
optimized ‘neural net disruption predictor’ was tested against 20 ms before disruption. Recent investigations in JT-60U
300 disruptive shots and 1008 non-disruptive shots selected have shown that training with non-disruptive shots and careful

S187
T.C. Hender et al

Table 10. JET NN performances. tp is the time before the 2.6
disruption. The numbers in brackets indicate the total number of the
shots for training, validation and performance test [389]. 2.4
Plasma current (A)
tp (ms) Network Output
2.2

Plasma current (A) (106)

40 60 80 100 120 140 160 180 200
2.0

Network Output
1.0
Training set
MA (86) 2 3 4 3 4 5 8 9 17 1.8
FA (400) 0 0 0 0 0 0 0 0 0
Threshold
Validation set 1.6
MA (35) 0 0 1 1 2 3 4 5 5 0.5
FA (246) 0 0 0 1 1 1 1 1 2
1.4
Test set
MA (62) 9 11 13 10 13 17 18 21 22

JG02.384-7c
FA (132) 0 0 0 0 0 0 0 0 0 0

18.0 18.5 19.0 19.5 20.0 20.5 21.0 21.5 22.0

Time (s)

adjustment of the Step 2 output (β-limit) data can improve the Figure 58. False alarm generated in JET shot 46314. Alarm
beta limit prediction and reduce the FA rate [388]. indication occurs at 19.7, 21.1 and 21.2 s [389].
Various NN optimizations have been applied for the
prediction of disruptions in JET [389]. The data sampling input signals with high reliability, adequate time resolution
interval is 20 ms. The performances of the best NN and long-term calibration stability. Signals used to date
optimizations are shown in table 10. The FA rates are for NN prediction have typically been generated for other
negligible for both the validation and the test sets over the 20– purposes and have some shortcomings for NN testing and
200 ms time window considered. The number of MAs is larger deployment. In addition to ‘quality’ considerations, it is
but comprises approximately 11% of the entire validation and essential that NN input data include signals that provide
test data set (97 shots) at 100 ms before the disruption. precursor indication of pending disruption. For some classes
For the best-performing NN configuration considered by and/or overt causes of disruption (e.g. density limit or cold-
Cannas et al the MA rates are 10–20% (SR = 80–90%) for the edge disruption, see section 3.1), ample precursor indication
test data set, whereas the FA is low (zero). This balance of the is present in multiple signals. For other cases (e.g. fast-
SR versus FA rate arises owing to the penalty applied, during growing internal MHD triggered β-limit or pressure-gradient-
the NN training process, to the occurrence of a FA, which, limit disruptions), signals with sufficient precursor content are
in the intended application, would invoke a ‘soft landing’ more difficult to obtain.
of the plasma current (see section 3.6.2), thereby preventing There are also other critical points that are intrinsic
completion of the experiment. A different choice in the penalty to the NN approach. These points include limitations on
of the NN output signal would achieve a different balance extrapolation capability, the need during initial training for
between the SR and FA performance. a sufficiently large and comprehensive database of disruptive
For JET disruptions caused by tearing instability or discharges and the machine-specific nature of the NN design.
the density limit, the NN prediction provides satisfactory The extrapolation capability of NNs is intrinsically poor in
reliability and a low FA rate for look-ahead times up to a few situations where expanded or new operation regimes are
hundreds of milliseconds before the disruption occurrence. A encountered. In present experiments, plasma operation
similarly satisfactory result is obtained in ASDEX Upgrade parameters naturally evolve with hardware improvement and
and JT-60U. Better results with regard to high SR are obtained new abilities to produce plasmas with higher performances.
if the operations-regime and disruption-type space parameters Hence the training of NNs requires updating as the plasma
used for the training and validation of the NN are well covered operating domain and performance capabilities are expanded.
and representative of the parameters that are encountered upon This requires having a database that covers the full operational
NN deployment. On the other hand, further NN development is parameter space of the machine, as was tried for example
necessary to reduce the number of FAs generated: in real-time in JT-60U for 13 years of operation [387]. Furthermore,
deployment NNs, unexpected FAs (figure 58) are sometimes for developing a machine-independent disruption prediction
encountered. method applicable to a next-step device, a more comprehensive
database that covers a variety of past and presently-operating
Discussion and future work. The experience gained with tokamaks in a more ‘dimensionless’ manner appears essential.
NN disruption prediction methods has identified a number of Finally, given the data quality and stability concerns noted
issues that call, on the one hand, for a dedicated effort in the above, quality and calibration stability of the data that is to
generation of the input signals and, on the other hand, for a comprise this multi-machine database will need to be given
deeper understanding of disruption dynamics and precursor careful consideration.
identification. Recently, a first-of-a-kind cross-machine prediction of
Regarding the first aspect, any kind of disruption disruption occurrence has been undertaken, using data from
prediction means, NN or other, requires a dedicated set of JET and ASDEX Upgrade, with a NN predictor trained

S188
 Chapter 3: MHD stability, operational limits and disruptions

on seven normalized dimensionless plasma parameters plus This section continues the discussion of disruption
normalized time. The NN was trained with JET data and avoidance found in the ITER Physics Basis [1]. Examples of
tested on ASDEX Upgrade data, with a resulting SR of 67%; experimentally implemented disruption avoidance techniques
conversely a NN trained with ASDEX Upgrade data and tested will be presented. As the discussion presented in section 3.6.1
on JET data showed a 69% SR [390]. These initial inter- tries to make clear, the emphasis here is on the ‘passive’
machine NN development results are an encouraging first step or ‘single-parameter’ deterministic and active control aspects
and indicate a possible way to overcome some of the concerns of disruption avoidance, rather than on the ‘multi-parameter’
about NN portability and the need to have data from disruptive prediction means that can also provide indication of pending
operation of a future device (ITER) before the NN predictor disruption.
for disruption-onset prediction can be developed. Disruption avoidance techniques employed in present
Another realization that has emerged from present tokamaks (and envisioned for ITER) are divided into two
studies is that an integrated set of NNs, each optimized to categories:
predict a given ‘type’ (cause) of disruption, will probably (i) machine preparation and design of ‘safe-pulse’ schemes
be required. Hence classification and characterization of and
observable precursors and onset processes for each ‘type’ (ii) real-time intervention with single or multiple feedback
of disruption will be needed (see section 3.1), and during- control loops or algorithms—often based upon physical
operation refinement of ITER NN predictions will undoubtedly models—to control specific plasma regimes and/or
be required. Nevertheless, in a next-step (after ITER) reactor recover from failure states [393]. Recovery procedures
tokamak, where the plasma configuration and operating mode can include effecting ‘soft landing’ or ‘soft-stop’ measures
will necessarily be well defined and not subject to major that attempt to end the plasma discharge without
evolution, neural net prediction of impending disruption will be producing a full-current or full-thermal-energy disruption.
a promising, but not exclusive, means available for activating
disruption avoidance (and ultimately mitigation) schemes. Faster-acting plasma shutdown and machine protection
processes (sometimes called ‘hard stop’ action(s)) intended
to be applied in situations where disruption appears
3.6.2. Disruption avoidance. The number of disruptions in
to be otherwise unavoidable are separately discussed in
ITER must be kept as low as possible. In principle, disruptions
section 3.6.3.
can be avoided during operation by an accurate construction
of the machine itself, careful selection of the plasma operating Passive avoidance. The need for thorough and consistent pre-
space and discharge development procedures (here ‘discharge pulse conditioning of the plasma-facing wall (by dc or HF glow
development’ denotes the plasma ‘start-up’ procedure that discharge cleaning [394]) and by the periodic deposition of
produces the final stationary plasma magnetic and kinetic first-wall coating materials (e.g. boron) is well known. Wall
equilibrium state) and deployment of a sophisticated plasma conditioning has a distinct impact on the plasma impurity
control system (see chapter 8 in this issue [373]). Careful content and therefore on the density limit. Wall conditions
alignment of the magnetic coils, for example, will reduce the also affect the plasma breakdown and initial current channel
tendency of MHD modes to lock owing to the presence of finite development phase of the discharge and hence ‘start-up’
non-axisymmetric magnetic field errors [1, 391, 392]. Precise reliability (see chapter 8 of this issue [373]).
adjustment of first-wall elements will reduce the probability The importance of avoiding mode locking during the
of local PFC surface overheating and impurity production. plasma start-up and initial low-density Ohmic flattop phase
Precise control of the magnetic and kinetic evolution of the is also well known. Disruption-avoiding ‘pulse designs’
plasma configuration will prevent unwanted wall-interaction that prevent mode locking during pulse ramp up have been
events and help ensure that the plasma development and demonstrated on TCV [395]. The operation space for locking
equilibrium state successfully avoids the various ‘causes’ of of MHD modes was statistically investigated in terms of plasma
disruption that apply. Finally, given the need for reliable density, safety factor, shaping (elongation and triangularity)
and effective plasma control, it is self-evident that to effect and vertical position. A clear tendency to develop mode
disruption avoidance, all the operation systems involved locking is found for low density and low plasma shaping.
(hardware and software) must be highly reliable and fail-safe. Modification of the shaping in the pulse ramp-up phase
The selection of ITER-applicable operation scenarios with successfully avoids mode locking (figure 59). Findings [396]
low-disruption rates is a further and inherently problematic about the need for low levels of non-axisymmetric error field
issue. Burning plasma operation in ITER will require for locked mode avoidance in low-density DIII-D Ohmic
operation in regimes with small margins against each of the plasmas and in higher-density high-β NBI-heated plasmas
three major plasma operation limits (see section 3.1) and under [392] have already been reported in the IPB.
conditions where the need for external stabilization of NTM
and/or RWM MHD instabilities and active control of divertor Disruption avoidance strategies. In present experiments,
attachment (power flux), plasma impurity content and fusion so-called ‘soft-landing’ or ‘soft-stop’ procedures comprise
power is anticipated. So achieving low disruption rates in the most-frequently applied action to avoid occurrence of
ITER will require a combination of precise and reliable plasma impending disruption. Here, following detection of an
control, reliable operation of external stabilization systems and impending disruption (see previous section) or violation of
careful development of plasma scenarios that successfully skirt one or more of various plasma operation ‘monitor limits’, the
the multiple ‘causes’ of disruption that lie along the plasma plasma current is immediately ramped down in an externally
‘start-up’ trajectory. controlled pre-defined way [393]. This action does not

S189
T.C. Hender et al

9018 @ 0.1 s
6361 @ 0.110 s Ip (+) Bt (-)
Ip (+) Bt (+) ) )

1 1.5

0.8
Distance to LM

Distance to LM
1
0.6

0.4
0.5
9018 @ 0.18 s
0.2 Ip (+) Bt (-)

0 6361 0 9018
0.05 0.06 0.07 0.08 0.09 0.1 0.05 0.1 0.15 0.2 0.25

5 5
4.5 4.5
4 4

Qedge
Qedge

3.5 3.5
3 3
2.5 2.5
2 2
0.05 0.06 0.07 0.08 0.09 0.1 0.05 0.1 0.15 0.2 0.25
Time [s] Time [s]

Figure 59. Time evolution of the distance to the l-closest locking modes (LM, see [395] for details), l = 1, 3, 5, shown together with the
evolution of q-edge during pulse ramp up. In a standard case (on the left) modes lock at ∼0.11 s (as the distance to LM tends to zero). On
the right, adjusted plasma shaping prevents locking and ensuing disruption [395].

always avoid disruption, since the typical plasma evolution Avoidance of Disruptions #88265
time scales, such as the energy or particle confinement times, NBIpuls related to mode activity
8
are significantly shorter than ramp down times. However,
NBI1
depending on the monitor limit and ramp-down parameter
settings, ramp down can often reduce the severity of the
disruption or VDE that ultimately occurs. In JET, for example, 6
the soft-landing procedure can be initiated by any of several
Signal [arb. unit]

‘protection’ monitor signals that are available. Monitored

plasma characteristics may include the MHD activity level,
radiated power fraction, x-ray radiation level and plasma 4

density relative to the Greenwald limit. EC 140 GHz HFS

Soft-stop actions can reliably effect machine protection
LFS
and reduce disruption severity (see discussion of pre-disruption 2 EC 107 GHz
thermal energy loss in section 3.2). However, initiating soft- TTLSignal

stop action can prematurely terminate pulses which are, in

principle, ‘repairable’. For example, many of the common Mode Onset

conditions that lead to disruption—enhanced MHD activity 0

0 0.2 0.4 0.6 0.8 1
and mode locking or strong divertor detachment—can be Time [s]
counteracted with intervention or ‘repair’ measures. One of
the more ordinary reactions is to reduce, or stop, external gas Figure 60. Time traces of high-field-side (HFS) and low-field-side
(LFS) ECE channels. Detection of a high correlation coefficient
or pellet injection if the plasma density becomes too high. initiates tangential NBI to speed up toroidal rotation and mode
A more elaborate technique to forestall mode locking growth is reversed (TTL = transistor transistor logic signal derived
has been developed on TEXTOR [397]. A real-time cross- from the ECE which triggers the NBI) (reprinted from [398].
correlation method is utilized to detect growing m = 2, Copyright 2001, with permission from Elsevier).
n = 1 modes. The ECE channels measure the electron
temperatures close to the q = 2 surface, on both the high- NBI-induced plasma rotation in JT-60U [401], control of the
and the low-magnetic field sides of the plasma column. If plasma stored energy in advanced tokamak plasmas in JT-
the cross-correlation coefficient exceeds a pre-programmed 60U [402] and DIII-D [254], control of resistive wall modes in
value, the injection of a tangential neutral beam is triggered, DIII-D [190], heating power control of NTM onset or peaking
leading to an acceleration of the toroidal plasma rotation and of the pressure profile in JET [403], control of gas injection
reversal of mode growth (figure 60). This action not only by monitoring the edge electron temperature via Langmuir
prevents disruption but can also restore the enhanced plasma probes in TORE SUPRA [404] and increase of NBI heating
confinement regime existent before the onset of the mode power and cessation of gas fuelling upon detection of ‘deep
growth. divertor detachment’ in ASDEX Upgrade [393]. In addition to
Other examples of real-time repair actions include these already implemented examples, emerging developments
stabilization of (2,1) tearing modes by ECRH injection in in ‘active’ MHD spectroscopy allow direct measurement of
JFT-2M [399] and RTP [400], control of mode locking by the damping rate of stable MHD modes [405,406], and thus in

S190
 Chapter 3: MHD stability, operational limits and disruptions

#12067
Ip ne
0

a.u.
gas flux PNBI
pulse repair
0

{
{
{
{
DDD DDD DDD DDD DDD
out in
DDD alarm limit Dα / Dα

Neural Network - time for disruption estimation

0.8
pulse termination
[s]

0.4
deactivated
disruption alarm impurity injection
0
1.6 2.4 3.2 4 4.8
Time [s]

Figure 61. Competition of two independent machine protection processes (a pulse repair and a pulse termination reflex, see text). This
example illustrates the need of a real-time supervisor to assure a clear control and protection hierarchy (reprinted from [408]. Copyright
2003, with permission from Elsevier).

principle provide a real-time method to measure the approach for methods and/or technologies that remain to be realized
to a stability limit. and tested in a tokamak. As the conclusion of this section
The various types of discharge repair actions noted are indicates, given the present uncertainties about how presently-
initialized via a discharge pulse control system and normally tested and proposed methods will (can) work in ITER and
act independently from each other and/or the standard pulse given the already-noted fact that the conflicting requirements
programme control logic. This can lead to situations where for disruption mitigation in ITER may force choices as to
two or more repair actions interfere. This is demonstrated mitigation priorities, continuing pursuit of a range of options
in an example from ASDEX Upgrade in figure 61. The and technologies is justified.
pronounced increase of the density in the Ohmic phase initiated
the deep divertor detachment (DDD) repair action [393], which Methods of disruption mitigation. In certain types of
interrupts the standard pulse programme, increases the level of disruptions or loss-of-control VDEs, plasma control actions
auxiliary heating power and closes all gas injection valves to taken immediately upon onset may be able to at least
restore stable plasma operation conditions. The DDD action is partial mitigate some consequential effects, especially those
successful in repairing the discharge and preventing disruption. associated with the current quench and vertical instability
However, the estimated time for disruption, calculated in phases of such events. For example, experiments in JT-60U
parallel by a real-time neural network, indicates a need to have demonstrated the possibility of mitigating vertical plasma
take mitigating action (neon impurity injection) to avoid motion in a major disruption by executing a rapid shift of
excessive disruption-produced forces on the vacuum vessel. the plasma vertical position immediately after the thermal
quench is detected [334]. Experiments in several devices (e.g.
In the example shown, the injection-triggering output of the
[324,334]) have shown that early injection of impurity species
neural network was inhibited. If the output had not been
into a developing VDE reduces peak halo currents more than
disabled it would have triggered a (mitigated) disruption. This
later injection. While the relatively long time scale (∼0.5 s)
example shows clearly that a real-time supervisor is needed to
for penetration of plasma equilibrium control fields through
ensure an unequivocal plasma control and machine protection
the ITER vacuum vessel may limit external control response
hierarchy. This could be guaranteed, e.g. by a plasma regime
to a VUD, the concept of using ‘slow’ impurity injection into a
identification algorithm [407], which dependently allocates
developing VDE to reduce halo currents appears to have merit.
a pre-defined protection action hierarchy on the basis of
External control response to disruptions or VDEs that produce
well-identified plasma and plasma operation states.
appreciable runaway current conversion is also critical to being
able to sustain control of the resulting current channel until the
3.6.3. Disruption mitigation. The rationale and physics basis runaway content can be benignly terminated (see section 3.4
reasons for why disruption mitigation will be essential for the and also chapter 8 of this issue [373]).
reliable operation of an ITER-class tokamak have already been Mitigation of many of the most immediately-damaging
extensively developed in the ITER Physics Basis [1] and in disruption effects (including heat flux conducted along field
the preceding sections of this chapter. The fact that disruption lines, halo current loads and runaway electrons) requires rapid
mitigation also introduces sometimes-conflicting requirements radiation of the plasma thermal and magnetic energies (as
on the method(s) to be deployed has also been developed discussed below) coupled with the timely delivery of large
in the IPB and in preceding sections. What follows below numbers of electrons to the plasma volume during the current
comprises a presentation of available methods and options for decay. As section 3.4 details, massive electron delivery
disruption mitigation in ITER. Some of these methods have immediately following the onset of the ITER current quench
been tested in present experiments; others constitute proposals is likely critical to avoiding avalanche conversion of thermal

S191
T.C. Hender et al

current to runaway current and the subsequent problems of

then having to (very carefully) terminate the runaway current.
Both these requirements—for radiating impurity delivery and
for electron delivery—can be satisfied by injection of impurity
atoms. Noble gas impurities are favoured owing to their
otherwise benign interaction with in-vessel surfaces and ease of
subsequent removal. The electron delivery and removal ease
requirements can also be satisfied by injection of hydrogen
or deuterium (although H2 injection in ITER will introduce
a non-reactive species that can compromise subsequent DT
reactivity). Several methods of delivery are possible, all of
which are conceptually capable, in an ITER embodiment, of
delivering, within a few milliseconds, the large quantity (∼1025
atoms or molecules) of impurities H2 or D2 needed.

Pellet injection. One obvious method for rapidly inserting Figure 62. Plasma current, size and position waveforms for Ar killer
impurities into a tokamak plasma is injection of impurity pellet and Ar massive gas injection in DIII-D. A ∼0.6 MA runaway
current ‘tail’ develops following the Ar KP injection (dotted curve).
‘killer’ pellets. Injected species used in killer pellet (KP) Adapted from data originally presented in [365] and in [366].
tests in present experiments have included cryogenic H2 , D2 ,
Ne, Ar, Kr-doped cryogenic D2 and Ag-doped polyethylene.
(4π e3 ne ln/mc2 ) is the critical electrical field that governs
The efficacy of the KP approach in reducing divertor energy
whether or not the avalanche grows (see section 3.4 and the
deposition and halo current magnitude and toroidal asymmetry
massive gas injection discussion that follows below).
has been demonstrated in many devices. Reductions of
25–95% in thermal flux conducted to the divertor, 50–75% Massive gas injection. Experiments conducted since the
in peak halo current magnitudes and reduction in halo current writing of the 1999 IPB [1] have shown that injection of
toroidal peaking factors (TPFs) to unity (e.g. [332, 365, 409, massive amounts of impurity and/or H2 or D2 neutral gas
410]) have been demonstrated. Modelling has also contributed into tokamak plasmas can provide the beneficial disruption
to improved understanding of disruption mitigation by KP mitigation effects of impurity killer pellets but with a reduced
injection [364]. Various possible pure and mixed-species or negligible tendency to produce runaways. Here the
pellets have been investigated in these modelling studies, terminology massive gas injection (MGI) is used to denote
including D2 , He, CH4 , Ag-doped H, Kr, Xe, Ne and Ar. disruption mitigation methods that employ a gas injection
Injection of low-Z materials such as D2 and He is found to system (typically a small, high-pressure gas reservoir coupled
be less effective in mitigation than the high-Z radiators such to a fast-opening valve) to produce a subsonic or trans-sonic gas
as Ne and Ar (e.g. [366]). jet that is capable of delivering a large quantity of neutral gas,
Killer pellet injection velocities are typically 1 km s−1 . typically more than 1022 atoms (∼0.1 atm litre = 100 Pa m3 )
As with standard plasma fuelling by injection of D2 pellets, the to the torus vacuum vessel within 10 ms. Table 11 shows
penetration depth for KPs based on present-day single-stage the injection parameters and species used in the various DIII-
gas gun or centrifuge technologies will be much smaller than D MGI studies ( [324, 365, 366, 371]) conducted during the
the ITER plasma minor radius. Both low-field-side (LFS) and period 1997–2005. Injection rates for ASDEX Upgrade and
high-field-side (HFS) injections are possible. Higher velocities JT-60U experiments are ∼3 and ∼30 times lower, respectively.
and deeper initial penetration are attainable with LFS injection, Automatic initiation of neon MGI taken up on warning of
but polarization drift (see e.g. [411]) tends to inhibit further impending disruption is now routinely used to limit vertical
inward transport of the injected impurities. The velocity force on the vacuum vessel in ASDEX Upgrade [414] (see
limitations imposed by HFS guide tubes limit usable HFS section 3.6.2 above).
velocity, but penetration, even with reduced velocity is deeper Experiments with noble gas MGI on many devices (e.g.
owing to polarization drift and resulting inward transport of the [362, 365–367, 371, 414, 415]) have demonstrated the highest
injected impurities. Anomalous pellet material transport to the level of mitigation of thermal and electromagnetic loads
plasma centre has also been observed with LFS injection [412]. observed in tokamaks, both in major disruptions and VDEs.
The principle shortcoming of cryogenic pellet injection Measurements in DIII-D show that the propagation of the
is the frequently-observed production of runaway electrons cooling front to the plasma centre occurs with a radial
(figure 62), which arises owing to the high local electric propagation velocity that is commensurate with the room-
field that pellet injection and subsequent ablation produces temperature sound speed of the injected species (He, Ne or
[365,413]. While the details of how the runaways are produced Ar) and bolometric data show essentially complete radiation
involves careful consideration of the local impurity deposition of the plasma stored energy [324]. In ASDEX Upgrade and
effect and modification of the electron distribution function, DIII-D, halo current magnitudes are typically reduced by more
the inherent propensity of KP injection to produce runaways than 50%, and toroidal peaking factors are reduced to unity,
can be understood from the simple fact that the number of resulting in a typical reduction in electromagnetic forces of
impurity electrons added by typical cryogenic pellets (e.g. in more than 75% [331, 414]. The reduction in halo current
DIII-D, ∼1–2 mm diameter Ar pellets containing ∼5 × 1020 magnitude and TPF obtained in MGI experiments is consistent
atoms) is insufficient to maintain E  Ec , where Ec = with the fast current quench, reduced vertical instability and

S192
 Chapter 3: MHD stability, operational limits and disruptions
a
Table 11. Gas injection capabilities for DIII-D MGI experiments (1997-2005) .
Total injected
quantity (10 ms)
Flow Flow (1022 atoms or
Gas (105 Torr L s−1 ) (104 Pa m3 s−1 ) molecules)
96% H2 + 4% Ar ∼ 2.5 ∼ 3.3 ∼ 8.4
D2 3.7 4.9 12.4
He 3.5 4.6 11.6
Ne 1.3 1.7 4.59
Ar 0.93 1.22 3.28
a
Experimentally measured equilibrium flow rates, typically obtained for
70 atm reservoir pressure. Some variation in valve and/or injection
configuration has occurred over the data collection period.

Table 12. KPRAD model calculations and experiment data for DIII-D disruption mitigation experiments (data adapted from [366]). Here
nAr is the injected argon density, Z is the average charge state, Te is electron temperature and τCQ the L/R current decay time.
Model Experiment
−3
Mitigation means nAr (m ) Z Te (eV) E/Ec τCQ (ms) Z τCQ (ms)
Ar pellet 2 × 10 19
5.9 7.5 312 2.0 5.7 ± 1 2.1 ± 0.1
Ar MGI 2 × 1021 0.45 1.46 3.8 1.38 0.3 ± 0.1 1.8 ± 0.1

cooling/symmetrization of the ex-plasma halo region that is

TQ
also obtained in naturally occurring fast-current-quench ‘radial vacuum edge cold
transit fill front CQ
disruptions’ (see section 3.3.2).
The high electron densities that MGI is capable 0.4 (a) jet trigger
[a.u.]

0
of producing can approach the magnitudes necessary 0.4
for collisional suppression of runaway avalanching (see 0
(b) jet photodiode
section 3.4). Table 12 compares the impurity densities 0.4
(c) edge ece
[keV]

and impurity ionization states (volume-average plasma Zeff ) 0

estimated by Whyte et al [366] to have been obtained for 2 (d) core ece
otherwise comparable DIII-D argon killer pellet and MGI 0
[MA] [a.u.] [T/s]

examples (e.g. the cases illustrated in figure 62). The 50 (e) Bdot
0
calculations were effected with a self-consistent 0-D KPRAD 2 (f ) Prad
impurity radiation model. The estimated and measured 0
plasma Zeff and current quench times (exponential decay time 1 (g) Ip
constant, τ ) are found to be in good agreement, lending support 0
to the hypothesis that the temperature and ionization state of the 3000 3005 3010 3015 3020 3025
t TQ
after-injection killer pellet and MGI plasmas are determined, time [ms] #117476

as they often are for natural disruptions, by radiation-loss

Figure 63. Fast plasma shutdown sequence in DIII-D (Ar
versus Ohmic-input power balance considerations. Owing MGI) [371] showing (a) the gas jet solenoid current, (b) visible
to their high impurity density, MGI plasmas attain a low-Te emission, (c) second harmonic edge ECE, (d) core ECE,
equilibrium state, where the Ar impurity is only partial ionized (e) magnetic fluctuations, (f) radiated power and (g) plasma current.
and where all the injected electrons are presumably present
(mainly in bound form) to contribute to collisional suppression
of runaway growth. energy radiation are achieved within ∼2 ms of initiation of
Further DIII-D MGI experiments conducted with a significant impurity delivery; this efficient radiation of thermal
modified injection system designed to provide higher jet energy appears to be facilitated through a combination of rapid
pressure at the plasma surface [371] have shown that the direct impurity-ion mixing and fast heat transport, both driven by the
penetration of injected Ar gas into stationary H-mode plasmas onset of large-scale MHD activity that follows the initial arrival
is limited to a few cm. Fast-gated camera images in Ar-I of gas at the plasma surface. Figure 63 shows the sequence
light indicate that the neutral deposition is initially localized of events that is observed in all DIII-D MGI examples. Initial
near the injection location and that there is no indication arrival of gas at the plasma edge (precursor flow  equilibrium
of singly ionized Ar penetration far into the pedestal edge flow [416]) is followed by the onset of edge cooling and a
gradient region. A similar lack of direct penetration is observed slight drop in plasma current (indicative of a narrowing of
for low-temperature Ohmic plasmas and during the after- the current profile), abrupt increase in MHD activity (2/1
thermal-quench phase of MGI (where the estimated plasma and other n = 1 modes), enhanced volumetric radiation and
temperature has dropped to ∼2 eV). Nevertheless, despite this onset of thermal quench and a disruption-like positive plasma
lack of observable direct neutral penetration, fast bolometry current spike that follows the thermal quench onset. A rapid
indicates that high (∼90%) fractions of volumetric thermal current quench that is usually runaway-free follows. A non-

S193
T.C. Hender et al

Table 13. Parameters for a single-species ITER MGI system.

Value or requirement Units D2 Ar Basis or assumption

−1
Sound speed (vs ) ms 935 320 300 K
Time to reach plasma surface ms 3.1 (5.3) 9.4 (15.6) 3 m (5 m) to plasma; propagation at vs
Additional delay ms 2 2 Valve trigger and opening delay, etc
Minimum look-ahead time for ms 5.1 (7.3) 11.4 (17.6) Sum of propagation and delay times,
pre-emptive action assumes deployment before natural
TQ onset is required
NRB Atoms or molecules 1.75 × 1025 1.94 × 1024 To achieve ne = nRB , for 830 m3 plasma volume,
100% assimilation
Delivery time (tdel ) ms 9 9 4∗ tdel = tCQ = 36 ms; tCQ /S ∗ = 1.7 ms m−2
Required average flow rate s−1 1.94 × 1027 2.16 × 1026 Assumes prompt rise
Required average flow rate Pa m3 s−1 7.2 × 106 8.0 × 105 Assumes prompt rise
Required average flow rate Torr L s−1 5.5 × 107 6.1 × 105 Assumes prompt rise

ideal MHD numerical simulation (NIMROD) has been used jet characteristics to prevent ‘shine through’ and damage to
to recreate similar experimental features of MGI mitigation on PFCs from direct jet impact. Although the full technology
Alcator C-Mod [417]. Forced plasma cooling at the plasma for cryogenic liquid jet injection has not been demonstrated,
edge was found to trigger robust MHD modes, which then led prototypes of room-temperature liquid (water) injectors with
to development of stochasticity and the loss of internal plasma relevant pressures and operating parameters have been
energy by heat conduction into the cold boundary plasma, produced and demonstrated (e.g. [419]).
without any impurity particle penetration deep into the core
plasma. This result verifies the important coupled role of Other concepts. Mitigation methods based upon multiple
plasma cooling and MHD, in disruption mitigation. HFS-launch pellet injection (a sequential burst of moderate-
Modifications to the DIII-D injection system made over sized pellets, intended to facilitate incremental penetration,
the course of the latest set of experiments that changed the [364]) and hypervelocity injection of dense compact plasma
rate of rise of initial gas delivery were found to shorten the toroids or spheromaks [420] have been proposed for ITER
overall time scale of the sequence of events that leads up to and reactor application. Both methods offer a potential
thermal quench onset, but were found to otherwise have little solution to the problem of effecting direct and immediate
effect on the subsequent current decay phase. The sequence delivery of radiating impurities to the plasma core. Testing of
of events and characteristics observed following MGI in DIII- either concept in present tokamaks and development of ITER-
D closely resemble those observed for natural ‘high-density’ applicable injection technologies remain as future research
fast disruptions. However, the MGI-initiated ‘disruption’ tasks.
exhibits (more) benign divertor energy loading and halo current
characteristics and favourable runaway electron mitigation Application to ITER. There are multiple challenges in
characteristics relative to otherwise comparable killer pellet developing reliable disruption mitigation methods for ITER.
initiated fast shutdown (note that the KP shutdown shown in Successful scenarios must bring together real-time stability
figure 62 does not show a current spike or other evidence of analysis and timely disruption prediction, reliable and timely
internal reconnection). triggering of the mitigation system and mitigation means
A detailed physics understanding of the mechanisms for that provide unequivocal runaway electron (RE) suppression.
gas jet penetration and ion mixing in the thermal quench Above all, the underlying principle must be that the mitigation
and current quench phases of MGI is not yet available but scheme ‘does less harm’, both with regard to recovery of
is a topic of present research. Gas injection also offers a plasma operations and maintaining the viability of plasma-
possible means for raising impurity densities during the CQ facing components than allowing unmitigated occurrence
itself. Other techniques, e.g. high-velocity pellets or liquid of disruption. Experimental and modelling advances in
jets, are predicted to simply pass through the cold after- detection (3.6.1), passive avoidance and precision/intelligent
TQ plasma without appreciably further raising the density or control (section 3.6.2) and mitigation means that can act
slowing runaway electrons. As the studies of methods for ITER to inhibit runaway conversion point the way to an ITER
indicate, meeting the competing requirements of mitigation disruption mitigation scheme based on massive impurity
for the various potentially damaging disruption effects may and/or hydrogenic injection. The single-species MGI approach
require combined or separate injection of different species of described above, which so far has been successful in mitigating
impurities at different points in the shutdown process. most of the consequences of disruptions in present tokamaks,
offers promise as an ITER disruption mitigation means.
Liquid jets. It has been proposed that one method of Table 13 gives a set of elementary ‘design basis requirements’
achieving sufficient impurity penetration for disruption for a single-species MGI system for ITER. Possibilities
mitigation in a reactor-grade plasma is to use cryogenically for either low-Z (D2 ) and moderate-Z (Ar) injections are
cooled liquid jets. Calculations suggest that this method may shown; requirements for intermediate-Z species (or mixtures
offer the capability of depositing sufficiently large quantities of higher-Z noble gases with D2 ) lie between these extremes.
of impurities well within the plasma [418]. Penetration Moderate-Z (Ar) injection reduces the requirement for
may actually be so efficient as to require careful control of injection quantity and flow rate because more electrons arrive

S194
 Chapter 3: MHD stability, operational limits and disruptions

for each gas atom [366] but requires a longer look-ahead time injection disruption mitigation in ITER remains open: as
for pre-emptive deployment. Low-Z (D2 ) injection increases the previous discussions have shown, high-pressure gas jets
the gas quantity and flow requirements but appreciably reduces (MGI), liquid jets and single and multiple (sequentially
the look-ahead time requirements. The estimates in table 13 injected) cryogenic pellets all remain as possible options.
assume that the MGI system valve (or valves) will, for reasons At a minimum, any option or combination of options
of nuclear radiation resistance and service access, need to be selected must satisfy a fundamental requirement of being
located between 3 and 5 m from the plasma surface. able to deliver the impurities required within a time shorter
There are also further issues in choosing between than the delay between detection of impending disruption
moderate-Z and low-Z injection: the gas delivery rate for a and onset of a natural disruption TQ. Since most of the
given valve and ‘jet tube’ design (see e.g. [371]) scales as vs proposed delivery systems are limited to helium-sound-speed
(the gas velocity), while the gas flow rise time (the time for the (103 m s−1 ) injection velocities, if the impurity injector and
gas flow through the ‘jet tube’ to reach full equilibrium value) associated hardware are to be located outside the ITER toroidal
scales as vs−1 . Hence the system flow capability for D2 is field coil (on an equatorial port), the minimum transit time to
∼3 times higher than the corresponding Ar capability and rise r/a = 0.5 in the plasma will be
6 ms. The combination
time will be ∼3 times shorter. To first approximation, these gas of this minimum propagation time with a 2–4 ms action time
delivery factors offset the 9 times increase in electron delivery (opening of a valve, etc) results in the need for an 8–10 ms
rate that Ar affords relative to D2 . Also, given that the measured look-ahead time, plus prompt action of the disruption detection
rise-time for Ar for the 1.3 m long DIII-D MGI system used (prediction) and control authorization software. While the
for the reference [371] experiments is ∼10 ms [416], rise- detection of many types of ITER disruptions more than 10 ms
time limitations imposed by a 3–5 m long ITER system may before occurrence appears feasible (sections 3.6.1 and 3.6.2),
adversely impact the prompt electron delivery efficacy for the detection of impending fast β-limit disruptions within 6 ms
moderate-Z or high-Z injection. may (owing to diagnostic limitations or fast MHD growth rates)
The parameters in table 13 are based on a requirement be less feasible.
to achieve the Rosenbluth no-avalanche density within tCQ /4 In this case, to minimize delay time, the injector hardware
and incorporate no explicit assessment of the radiation cooling must be placed as close as possible to the plasma surface
effectiveness of the injected impurities in mitigating the (so that propagation times approach 1 ms), mechanical and
divertor thermal energy loading or in achieving sufficient after- electronic system activation delays must be minimized and
TQ radiation to obtain a prompt current shutdown (with low high velocity injection schemes are favoured. Considerations
halo current). While the tCQ /4 specification is somewhat of simplicity, volume and reliability argue for a small-volume
arbitrary, its intent is to ensure that no significant runaway high-pressure gas reservoir and a fast-acting valve with a
electron avalanching occurs in the early phases of the current minimal nozzle or jet tube. Compact and/or magnetic-field-
quench, when the driving electric field is highest. And while tolerant systems of this nature have already been implemented
there is no significant question about the ability of the quantities in ASDEX Upgrade [386] and TEXTOR [415]. However
of injected Ar (or Ne) specified in table 13 to efficiently radiate technology development is required to design an ITER-
the full 350 MJ of ITER thermal energy within a few ms, qualified system that is also compatible with both the high
there are concerns, first identified in the IPB, that too-effective magnetic-field and nuclear radiation and tritium environments
radiation during the initial edge-cooling onset phase of MGI and is otherwise capable of meeting the short activation time,
(before internal reconnection and TQ develop) can lead to fast rise time and total gas quantity requirements.
excessive local or global energy loading on ITER first-wall Recent observations of MGI gas jet penetration in DIII-
surfaces. D [371] and models [421], developed to explain the lack of
Here elementary calculations show that the minimum direct penetration even for low-field, low Te plasmas, indicate
deposition time for uniform radiation of the full Wth to a that the local jet ram pressure (ρv2 ) at the plasma surface
beryllium first-wall should be
∼0.6 ms if surface melting is must approach the local magnetic field pressure (∼ B 2 /2µ0 )
to be avoided. Local peaking factors or pre-heating of the first- for effective penetration. If this requirement applies for
wall surface will increase this minimum requirement. While ITER, then gas jet surface pressures of 10 MPa (100 bar)
some tailoring of the initial radiation cooling attributes of a will be required for direct penetration. It is unlikely that a
D2 -dominated system may be possible by judicious admixture compact/fast-acting gas jet system capable of generating this
of higher-Z noble gases [367], present physics understanding pressure at the plasma is possible. MGI in ITER will therefore
of the combination of gas jet rise-time characteristics, neutral need to depend on surface delivery, with subsequent MHD
gas penetration, impurity ionization and transport along flux mixing of the surface-deposited impurities (as appears to be
surfaces and overall radiation dynamics is insufficient to the case in present experiments).
make detailed estimates of exactly what will happen to the Pressures of ∼100 bar are more easily attained with liquid
first-wall surface from ITER MGI deployment. Present jets and calculations have shown that liquid He and/or liquid
physics understanding is also insufficient to determine whether H2 /D2 jets can penetrate into the ITER core [413, 418].
injection of the quantities of gas specified in table 13 will, in While further assessments of the technological feasibility of
fact, result in effective suppression of runaway avalanching in cryogenic liquid jets are needed and while concerns remain
the plasma core during the ensuing current decay. about jet fragmentation (owing to hydrodynamic stability),
Given these uncertainties about application of MGI once the jet stream enters the plasma, liquid jets continue
for thermal energy and runaway mitigation, the choice of to offer a potentially attractive means for rapidly delivering
the optimal species and delivery technique(s) for impurity impurities to near the centre of the ITER plasma.

S195
T.C. Hender et al

Finally, the calculations of Kuteev et al [422] have shown a role in disruption development. The emergence of full
the conceptual (ablation physics) feasibility of achieving 3D MHD dynamic equilibrium modelling capabilities offers
deep penetration with the LFS launch of high-velocity (3– prospects for quantitative predictions of ITER disruption
5 km s−1 ) cryogenic noble gas pellets with diameters ∼10 mm. scenarios and consequences. The application of these models
Velocities of this magnitude can be achieved with two-stage to interpretation of data in present tokamaks will likely aid
gas-gun launchers, albeit with a need for appreciably more in sorting out some of the presently obscure internal cause-
look-ahead time than is required for a single-stage pellet gun and-effect dynamics of the precursor-growth and thermal-
or gas jet and a need to reinitialize the launcher system after quench-onset phases of disruption. Three-dimensional MHD
each use. Again, further technology assessments are needed simulations will perhaps also aid in clarifying the cause(s) for
and there is no present experience with the interaction of such the wide range of data scatter in the existing current quench
large-scale hypervelocity pellets with a correspondingly large- and halo current databases.
scale, high-temperature and high-density tokamak plasma. Data quantifying the range of physics and hard-
The considerations noted above as to the possible need to ware/operations ‘causes’ for disruptions demonstrate that all
‘tailor’ the pellet species and spatial and/or time-profile of disruptions have identifiable physics and/or operations causes
the delivered impurity (or impurities) to obtain a first-wall and that there are often clear distinctions in the precursor
acceptable radiation history also apply. and development phases of disruptions initiated by a plasma
Taking the above considerations together, there is edge energy balance deficit (‘cold-edge’ or density-limit-onset
some basis for cautious optimism about the development disruption) versus those initiated by rapidly growing internal
of disruption mitigation schemes that will be applicable ideal-MHD instabilities (beta-limit or ITB pressure-gradient
to ITER. The need for effective mitigation, including disruptions). There are also ‘thermal-collapse’ disruptions,
mitigation of runaway electron conversion, is unequivocal, initiated by the onset of slowly-growing resistive MHD insta-
and runaway mitigation will be needed once high-current bilities (e.g. NTMs), that share many similarities with cold-
plasma commissioning operations commence in the hydrogen edge disruptions. Cold-edge and thermal-collapse disruptions
phase. With regard to urgent research needs in present have long development phases, with ample before-onset diag-
experiments, the most pressing areas for research appear to nostic indication(s) of impending disruption; ideal-instability-
be related to detection, gas jet penetration and subsequent onset disruptions have much shorter development phases and
impurity mixing throughout the plasma volume and strategies only limited external indications that disruption is imminent.
to avoid or minimize radiative melting of the first-wall. While The difference in onset characteristics impacts the types and
gas injection systems appear to offer many advantages for reliability of predictive methods that can be used to take timely
eventual utilization in ITER as primary and back-up means before-disruption avoidance or mitigation action. Both single-
for disruption and/or runaway electron conversion avoidance, parameter deterministic indicators and multi-parameter neural
concerns about the ability of gas jets to effect direct control over net methods have been successful in reliably predicting many,
impurity delivery and in-plasma distribution during the onset but not all, types of disruption onset in present tokamaks, and
phase of mitigation suggest that alternative liquid jet, multi- both methods hold promise for ITER. Concerns do exist, how-
pellet and hypervelocity pellet injection approaches should ever, about the a priori need for data from disruptive ITER
also be pursued. plasmas to ‘train’ an ITER neural network, though there has
been some initial progress on this issue. For both methods,
3.7. Summary and R&D needs achieving an acceptable balance between prediction reliability
and ‘false-alarm’ rates is also an issue.
There has been substantial new data and progress in Statistical studies of the physics and hardware causes and
understanding the physics basis for disruptions and their ‘types’ of disruption continue to support the understanding
consequential effects in an ITER-class tokamak. In addition, documented in the IPB that the overall frequency of disruptions
the size and plasma energy content of ITER have been in tokamaks conducting a wide-ranging programme of
reduced (relative to the ITER-EDA design), resulting in some plasma development and exploration studies is about 10%.
relaxation of the divertor PFC surface energy loadings. There Disruptivity (frequency of disruption per pulse) tends to be
has also been substantial progress in developing disruption higher during exploratory phases or campaigns to develop
consequence mitigation schemes and in improving prediction higher-performance plasmas, but higher rates of disruption
and avoidance strategies to minimize the number of disruptions are not always correlated only with proximity to one or
that will occur during ITER operation. more of the three traditional operations limits. There is also
The overall understanding—summarized in the ITER emerging statistical evidence that once the prescription and
Physics Basis [1]—of the MHD instability processes and control means for achieving a stationary high-performance
causative factors that contribute to onset of disruption plasma state is obtained, there is no further tendency for
remain largely unchanged. Conventional pictures of such plasmas to disrupt as the duration of the stationary
the internal triggering mechanism for major disruption, phase is extended. The reproducibility of the plasma ‘start-up
based on magnetic reconnection and non-linear growth of phase’ (which here denotes the entire approach to stationary
helical instabilities (cold bubble ingress) in the low-shear conditions) can, with adequate pre-discharge wall conditioning
central region, are now being supplemented by internal and hardware/control reliability, approach 100%. These
ballooning/interchange models that better explain the observed findings support the expectation that a combination of ‘passive’
insensitivity of internal thermal equilibriation times to initial and ‘precision control’ disruption avoidance strategies coupled
plasma temperature. It is likely that both mechanisms play with well-qualified prediction capabilities will be successful in

S196
 Chapter 3: MHD stability, operational limits and disruptions

contributing to the achievement of low disruptivity in ITER, by large-amplitude magnetic fluctuations and (ii) increase of
even in exploratory regimes. the plasma electron density (free + bound) to the Rosenbluth
Expectations for avoiding disruptions and/or being able no-avalanche density, ∼1022 m−3 , within a few ms after
to successfully mitigate many of their effects notwithstanding, the onset of current quench. Both of these mechanisms
present data continues to support the need to design the present challenges to implement in ITER. For fluctuations, the
ITER vacuum vessel and in-vessel component structures to challenges lie in how to rapidly apply externally generated
withstand the electromagnetic loading effects of a credible helical fields or in how to induce natural (self-generated)
number of unmitigated ‘limit-case’ disruptions and VDEs. MHD fluctuations, that are capable of effecting the prompt
Newly-acquired and re-interpreted data on the current decay (∼103 s−1 ) loss rates needed to offset avalanche gain. For
time scaling indicates that the lower bound on area-normalized electron density increase, the challenges lie in being able
current decay time is approximately 1.7 ms m−2 . New data on to reach the Rosenbluth density before appreciable current
halo current magnitude and toroidal asymmetry (TPF) confirm decay develops: this will require the delivery of ∼3 × 1025
the continued use of the IPB-identified guidelines of Ih /Ip0  electrons (= 1.5 × 1025 H2 or 1.8 × 1024 Ar) to the
0.4 and Ih /Ip0 ∗ TPF  0.75, albeit with increasing evidence ITER plasma in 10 ms. Options for achieving these high
that TPF > 2 can be excluded as a design basis. Emerging delivery rates include sequential multiple pellet injection,
3D MHD code modelling capabilities also hold the promise of high-velocity liquid jets and massive gas injection (MGI).
being able to make first-principles-based simulations of halo Of these three options, only MGI has so far been tested in
current and 3D MHD effects in ITER disruptions and VDEs. present tokamaks at ITER-equivalent injection rates. While the
New and more comprehensive data on the internal ability of MGI (and also small ‘killer pellets’) to successfully
redistribution and subsequent loss of plasma thermal energy reduce divertor energy deposition and halo current magnitude
(Wth ) to divertor and other in-vessel PFC surfaces—in and asymmetry in present-tokamak testing is unequivocal,
combination with the twofold reduction in the ITER thermal improvement of present injection system flow rate and rise
energy loading ratio—has significantly improved prospects for time capabilities and tests in a large, high-current tokamak
avoiding disruption-produced thermal erosion of divertor PFC are needed before a definitive demonstration of reaching the
surfaces. Thermal energy loss and deposition observations in Rosenbluth density for runaway prevention within a fraction
present tokamaks show that while internal Wth redistribution of the plasma current decay can be demonstrated. Physics and
times for ITER are still projected to be ∼1 ms, the timescale extrapolation basis questions also apply to how the presently
for divertor surface thermal energy deposit will be significantly observed role of self-generated MHD fluctuations in promoting
longer, perhaps as long as 10 ms. Significant broadening rapid ‘MHD mixing’ of the edge-ionized atoms that MGI
of the divertor energy deposition area, by a factor of about produces will extrapolate to ITER and also, how, in situations
10, is also now expected, and there is increasing present- where avalanche gain is high, self-generated fluctuations affect
experiment evidence for ‘pre-divertor’ convective and/or runaway multiplication.
radiative deposition of a significant fraction of the plasma Finally, there has been continuing progress in improving
thermal energy on the divertor entrance and first-wall surface the scope and application of integrated disruption models.
areas. Finally, accumulating evidence in present tokamaks Two-dimensional dynamic equilibrium codes with various
suggests that significant before-disruption thermal energy loss degrees of auxiliary models specialized for disruption and
may be expected in at least some classes of ITER disruptions VDE modelling are now available and have been used
(e.g. those with slowly developing initial phases). While to make self-consistent estimates of ITER electromagnetic
there is presently insufficient understanding to quantify how loadings. However, further experimental validation, better
these various Wth -mitigation factors will combine in ITER, the self-consistent integration of the auxiliary models, expansion
prospects of obtaining reduced or zero divertor PFC erosion of dynamic modelling to include full 3D plasmas and 3D
during many ITER disruptions is now significantly enhanced vessel and in-vessel component representations are needed to
relative to previous prospects arising from the 1999 ITER establish comprehensive disruption scenarios and evaluations
Physics Basis [1]. of disruption mitigation in ITER.
Runaway electron conversion continues to be serious In summary there has been steady progress in
concern for ITER. Already, in JET, half of the thermal pre- understanding disruptions and their consequential effects.
disruption current can be converted into runaway current, with Of particular note is progress in developing massive gas
much of the runaway population generated by the Coulomb injection techniques for ameliorating disruption consequences.
There does remain a requirement to develop better disruption
avalanche multiplication process. Avalanche gain in ITER will
prediction techniques, to better understand energy deposition
be much higher and simulations show that, without additional
on the first wall and divertor, to develop a truly predictive
losses, a substantial fraction (70–80%) of the thermal current
modelling capability for ITER and to fully understand
will be converted into runaway current following disruption or
mechanisms of how injected gas penetrates into the plasma
onset of the current quench phase of a VDE. Other simulations
and the capacity of such techniques to inhibit runaway electron
of the effect of the runaway energy deposition expected
formation.
on the ITER first-wall or divertor surfaces indicate that an
uncontrolled interaction of this magnitude of runaway current
has the potential to produce PFC surface and substrate melting References
and erosion.
[1] ITER Physics Basis 1999 Nucl. Fusion 39 2137
Two mechanisms to forestall runaway conversion in ITER [2] Gormezano C. et al 2007 Progress in the ITER Physics Basis
are identified: (i) enhanced prompt runaway loss produced Nucl. Fusion 47 S285–S336

S197
T.C. Hender et al

[3] Loarte A. et al 2007 Progress in the ITER Physics Basis [43] Porcelli F. et al 2004 Nucl. Fusion 44 362
Nucl. Fusion 47 S203–S263 [44] Bussac M.N. et al 1975 Phys. Rev. Lett. 35 1638
[4] Sauter O. et al 2002 Phys. Rev. Lett. 88 105001-1 [45] Kruskal M.D. and Oberman C.R. 1958 Phys. Fluids 1 275
[5] Nave M.F.F. et al 1995 Nucl. Fusion 35 409 [46] Ward D.J. and Jardin S.C. 1989 Nucl. Fusion 29 905
[6] Porcelli F., Boucher D. and Rosenbluth M. 1996 Plasma [47] Jardin S.C., Bell M.G. and Pomphrey N. 1993 Nucl. Fusion
Phys. Control. Fusion 38 2163 33 371
[7] Kadomtsev B.B. 1975 Fiz. Plazmy 1 710 [48] Bateman G. et al 1998 Phys. Plasmas 5 2355
Kadomtsev B.B. 1976 Sov. J. Plasma Phys. 1 389 [49] Waltz R. et al 1997 Phys. Plasmas 4 2482
[8] Ottaviani M., Porcelli F. and Grasso D. 2004 Phys. Rev. Lett. [50] Güenter S. et al 1998 Nucl. Fusion 38 1431
93 075001 [51] Maraschek M. et al 2003 Plasma Phys. Control. Fusion
[9] Gimblett C.G. and Hastie R.J. 1994 Plasma Phys. Control. 45 1369
Fusion 36 1439 [52] La Haye R.J. and Sauter O. 1998 Nucl. Fusion 38 987
[10] Sauter O. et al 1999 Proc. Joint Varenna-Lausanne Int. [53] Buttery R.J. et al 2003 Nucl Fusion 43 69
Workshop on Theory of Fusion Plasmas, 18th Int. School [54] Sauter O. et al 2002 Plasma Phys. Control. Fusion 44 1999
of Plasma Physics (Varenna, 1998) ed J.W. Connor et al [55] Isayama A. et al 2001 Nucl. Fusion 41 761
ISSP-18 (Bologna: Editrice Compositori) p 403 [56] Kislov D.A. et al 2001 Nucl. Fusion 41 1619
[11] Lazzaro E. et al Proc. 26th EPS Conf. on Controlled Fusion [57] La Haye R.J. et al 2000 Phys. Plasmas 7 3349
and Plasma Physics (Maastricht, The Netherlands, 1999) [58] Buttery R.J. et al Proc. 20th Int. Conf. on Fusion Energy 2004
vol 23J (ECA) p 381 (Vilamoura, Portugal, 2004) (Vienna: IAEA) CD-ROM
[12] Angioni C. et al 2000 Theory of Fusion Plasmas: Proc. Joint EX/7-1 and http://www.naweb.iaea.org/napc/physics/
Varenna–Lausanne Int. Workshop, 19th Int. School of fec/fec2004/index.html
Plasma Physics (Varenna, 2000) (Bologna: Editrice [59] Glasser A.H. et al 1976 Phys. Fluids 19 567
Compositori) pp 73–86 [60] Connor J.W et al 2001 Phys. Plasmas 8 2835
[13] Mantsinen M.J. et al 2002 Plasma Phys. Control. Fusion [61] Fitzpatrick R. 1995 Phys. Plasmas 2 825
44 1521–42 [62] Mikhailovskii A.B. 2003 Contrib. Plasma Phys. 43 125
[14] Angioni C. et al 2003 Nucl. Fusion 43 455 [63] Reimerdes H. et al 2002 Phys. Rev. Lett. 88 105005-1
[15] Angioni C. et al 2002 Plasma Phys. Control. Fusion [64] Brennan D.P. et al 2003 Phys. Plasmas 10 1643
44 205 [65] Gude A., Guenter S. and Sesnic S. et al 1999 Nucl. Fusion
[16] Angioni C. et al 2002 Proc. 29th Conf. on Plasma Physics. 39 127
and Controlled Fusion (Montreux, Switzerland, 2002) [66] Frederickson E. et al 2002 Phys. Plasma 9 548
vol 26B (ECA) P-1.118 [67] Gorelenkov N.N. et al 1996 Phys. Plasmas 3 3379
[17] Goodman T.P. et al 2003 Nucl Fusion 43 1619 [68] Konovalov S.V. et al 2002 Plasma Phys. Control. Fusion
[18] Goodman T.P. 1999 Proc. 26th EPS Conf. in Controlled 44 L51
Fusion and Plasma Physics (Maastricht, The Netherlands [69] Poli E. et al 2003 Plasma Phys. Control. Fusion 45 71
1999) vol 23 (ECA) p 1101 P3.040 [70] Buttery R.J. et al 2002 Phys. Rev. Lett. 88 125005
[19] Henderson M.A. et al 2001 Fusion Eng. Des. 53 241 [71] Lütjens H. et al 2002 Phys. Plasmas 9 4837
[20] Mück A. et al 2003 Proc. 30th EPS Conf. on Controlled [72] Waelbroeck F.L. et al 2001 Phys. Rev. Lett. 87 215003-1
Fusion and Plasma Physics (St. Petersburg, Russia, 2003) [73] Smolyakov A.I. et al 2004 Plasma Phys. Control. Fusion
vol 27A (ECA) P-1.131 and http://eps2003.ioffe.ru/ 46 L1-L6
[21] Zohm H. et al 2003 Nucl. Fusion 43 1570 [74] Smolyakov A.I. and Lazzaro E. 2004 Phys. Plasmas 11 4353
[22] Zohm H. et al 2006 Proc. 21st Int. Conf. on Fusion Energy [75] La Haye R.J. et al 2003 Phys. Plasmas 10 3644
Research 2006 (Chengdu, China) (Vienna: IAEA) [76] Ozeki T. et al 2003 Plasma Phys. Control. Fusion 45 645
CD-ROM EX/4-1Rb [77] Bergmann A. et al 2002 Proc. 19th Int. Conf. on Fusion
[23] Isayama A. et al 2002 J. Plasma Fusion Res. Series 5 324 Energy 2002 (Lyon, 2002) (Vienna: IAEA) CD-ROM
[24] Pinsker R.I. et al 2003 Bull. Am. Phys. Soc paper GO1.004 TH/P1-01 and http://www.iaea.org/programmes/ripc/
http://flux.aps.org/meetings/YR03/DPP03/baps/abs/ physics/fec2002/html/fec2002.html
S930004.html [78] Nave M.F.F. et al 2003 Nucl. Fusion 43 179
[25] Hanada K. et al 1992 Phys. Fluids B 4 2675 [79] Itoh S.-I. et al 2004 Plasma Phys. Control. Fusion 46 123
[26] Westerhof E. et al 2003 Nucl. Fusion 43 1371 [80] Pustovitov V.D. 2005 Nucl. Fusion 45 245
[27] Ikeda Y. et al 2002 Nucl. Fusion 42 375 [81] Buttery R.J. et al 2001 Proc. 28th EPS Conf. on Plasma
[28] Mück A. et al 2002 Proc. 29th EPS Conf. on Controlled Physics (Funchal, Portugal, 2001) vol 25A (ECA)
Fusion and Plasma Physics (Montreux, Switzerland, 2002) pp 1813–6 paper P-5.011 and http://www.cfn.ist.utl.pt/
vol 26B (ECA) P-1.037 EPS2001/CD/pdfs/P5.011.pdf
[29] Pietrzyk Z.A. et al 1999 Nucl. Fusion 39 587 [82] Popov A.M. et al 2002 Phys. Plasmas 9 4205
[30] Isayama A. et al 2001 Fusion Eng. Des. 53 129 [83] Hender T.C. et al 2004 Nucl Fusion 44 788
[31] Petty C.C. et al 2003 Nucl. Fusion 43 700 [84] Hender T.C. et al Proc. 18th Int. Conf. on Fusion Energy 2000
[32] Pochelon A. et al 2001 Proc. 28th EPS Conf. on Controlled (Sorrento, 2000) (Vienna: IAEA) CD-ROM EX/P3-02 and
Fusion and Plasma Physics (Funchal, 2001) (ECA) http://www.iaea.org/programmes/ripc/physics/fec2000/
vol 25A p 1805 html/model.htm
[33] Eriksson L.-G. et al 2004 Phys. Rev Lett. 92 235004 [85] Kruger S.E. et al 1998 Phys. Plasmas 5 455
[34] Tobita K. et al 2000 JAERI-Conf. 2000-004 JAERI (Naka, [86] Mikhailovskii A.B. et al 2000 Plasma Phys. Rep. 26 375
Japan) [87] Marchenko V.S. and Lutsenko V.V. 2001 Phys. Plasmas
[35] Kramer G.J. et al 2000 Nucl. Fusion 40 1383 8 4834
[36] Tobita K. et al 2002 Fusion Sci. Technol. 42 315 [88] Zohm H. et al 2001 Phys. Plasmas 8 2009
[37] Graves J.P. 2004 Phys. Rev Lett. 92 185003 [89] Morris A.W. et al 1992 Proc. 19th EPS Conf. on Controlled
[38] Günter S. et al 2003 Nucl. Fusion 43 161 Fusion and Plasma Physics (Innsbruck, Austria, 1991)
[39] Westerhof E. et al 2002 Nucl. Fusion 42 1324 vol 16C-I (ECA) p 423
[40] Eriksson L.-G. et al 1998 Phys. Rev Lett. 81 1231 [90] Warrick C.D. et al 2000 Phys. Rev. Lett. 85 574
[41] Mayoral M-L. et al 2002 Proc. 29th EPS Conf. on Plasma [91] Zohm H. et al 1999 Nucl. Fusion 39 577
Physics. and Controlled. Fusion (Montreux, Switzerland, [92] Isayama A. et al 2000 Plasma Phys. Control. Fusion 42 L37
2002) vol 26B (ECA) P-1.026 [93] La Haye R.J. et al 2002 Phys. Plasmas 9 2051
[42] Kolesnichenko Ya.I. et al 1992 Phys. Rev. Lett. 68 3881 [94] Petty C.C. et al 2004 Nucl. Fusion 44 243

S198
 Chapter 3: MHD stability, operational limits and disruptions

[95] Doyle E.J. et al 2007 Progress in the ITER Physics Basis [145] Bondeson A. et al 2003 Plasma Phys. Control. Fusion
Nucl. Fusion 42 S18–S127 45 A253
[96] Zohm H. et al 2001 Nucl. Fusion 41 197 [146] Liu Y.Q. et al 2004 Nucl. Fusion 44 232
[97] Maraschek M. et al 2004 Nucl. Fusion 45 1369 [147] Kurita G. et al 2003 Nucl. Fusion 43 949
[98] Gantenbein G. et al 2000 Phys. Rev. Lett. 85 1242 [148] Boozer A.H. and Maslovsky D. 2004 Proc. 31st EPS Conf.
[99] Isayama A. et al 2003 Nucl. Fusion 43 1272 on Plasma Physics (London, UK, 2004) vol 28B (ECA)
[100] Lazzaro E. et al 2005 6th Int. Workshop on Strong P-2.159
Microwaves in Plasmas (Nizhny Novgorod) [149] Okabayashi M. et al 2002 J. Plasma Fusion Res. Ser. 5 42
[101] Berrino J. et al 2006 IEEE Trans. Nucl. Sci. 53 1009 [150] Okabayashi M. et al 2002 Plasma Phys. Control. Fusion
[102] Nagasaki K. et al 2003 Nucl. Fusion 43 L7 44 B339
[103] Isayama A. et al 2005 Phys. Plasmas 12 056117 [151] Gryaznevich M.P. et al 2003 APS (Albuquerque, New
[104] Buttery R.J. et al 2004 Nucl. Fusion 44 678 Mexico, 2003) paper RP1.036 and http://www.aps.org/
[105] Wolf R.C. et al 1999 Plasma Phys. Control. Fusion 41 B93 meet/DPP03/baps/abs/S2080036.html
[106] La Haye R.J., Rice B.W. and Strait E.J. 2000 Nucl. Fusion [152] Liu YQ. et al Proc. 20th Int. Conf. on Fusion Energy 2004
40 53 (Vilamoura, Portugal, 2004) (Vienna: IAEA) CD-ROM
[107] Campbell D.J. et al 1988 Phys. Rev. Lett. 60 2148 TH2-1 and http://www.naweb.iaea.org/napc/physics/
[108] Maget P. et al 2002 Theory of Fusion Plasmas Proc. Joint fec/fec2004/index.html
Varenna–Lausanne Int. Workshop (Varenna 2002) ed [153] Garofalo A.M., Jensen T.H. and Strait E.J. 2003 Phys.
J.W. Connor et al (Bologna: Editrice Compositori) p 363 Plasmas 10 4776
ISPP-20 [154] Garofalo A.M. et al 2006 Proc. 21st Int. Conf. on Fusion
[109] Stober J. et al 2002 Plasma Phys. Control. Fusion 44 A159 Energy Research 2006 (Chengdu, China, 2006) (Vienna:
[110] Angioni C. et al 2003 Phys. Plasmas 10 3225 IAEA) CD-ROM EX/7-1Ra
[111] Garbet X. et al 2003 Phys. Rev. Lett. 91 035001 [155] Takechi M. et al 2006 Proc. 21st Int. Conf. on Fusion Energy
[112] Angioni C. et al 2004 Nucl. Fusion 44 827 Research 2006 (Chengdu, China, 2006) (Vienna: IAEA)
[113] Gude A. et al 2002 Nucl. Fusion 42 833 CD-ROM EX/7-1Rb
[114] Günter S. et al 2001 Phys. Rev. Lett. 87 275001 [156] Polevoi A.R. et al 2002 J. Plasma Fusion Res. Ser. 5 82
[115] Günter S. et al 2004 Nucl. Fusion 44 524 [157] Liu Y.Q. et al 2000 Phys. Plasmas 7 3681
[116] Sips A.C.C. et al 2002 Plasma Phys. Control. Fusion 44 B69 [158] Bondeson A. et al 2002 Nucl. Fusion 42 768
[117] Raju D., Sauter O. and Lister J.B. 2003 Plasma Phys. [159] Okabayashi M., Pomphrey N. and Hatcher R.E. 1998 Nucl.
Control. Fusion 45 369 Fusion 38 1607
[118] Yu Q., Günter S. and Lackner K. 2000 Phys. Rev. Lett 85 2949 [160] Pustovitov V.D. 2001 Plasma Phys. Rep. 27 195
[119] St. John H.E. et al 1995 Proc. 15th Int. Conf. on Plasma [161] Strait E.J. et al 2003 Nucl. Fusion 43 430
Physics and Controlled Nuclear Fusion Research 1994 [162] Okabayashi M. et al 2001 Phys. Plasmas 8 2071 075001
(Seville, 1994) vol 3 p 603 (Vienna: IAEA) [163] Pustovitov V.D. 2002 J. Plasma Fusion Res. Ser. 5 278
[120] Lao L.L. et al 1990 Nucl. Fusion 30 1035 [164] Bialek J. et al 2001 Phys. Plasmas 8 2170
[121] Matsuda K. 1989 IEEE Trans. Plasma Sci. PS-17 6 [165] Chu M.S. et al 2003 Nucl. Fusion 43 441
[122] Prater R. et al 2003 Nucl. Fusion 43 1128 [166] Fitzpatrick R. 2001 Phys. Plasmas 8 871
[123] La Haye R.J. et al 2006 Nucl. Fusion 46 451 [167] Gribov Y. and Pustovitov V.D. 2002 Proc. 19th Int. Conf. on
[124] Harvey R.W. and Perkins F.W. 2001 Nucl. Fusion 41 1847 Fusion Energy 2002 (Lyon, France, 2002) (Vienna: IAEA),
[125] Hamamatsu K. and Fukuyama A. 2000 Plasma Phys. CD-ROM CT/P-12 and http://www.iaea.org/programmes/
Control. Fusion 42 1309 ripc/physics/fec2002/html/fec2002.htm
[126] Pustovitov V.D. et al 2000 Proc. 18th Int. Conf. on Fusion [168] Liu Y.Q. and Bondeson A. 2002 Plasma Phys. Control.
Energy 2000 (Sorrento, Italy, 2000) (Vienna: IAEA) Fusion 44 L21
CD-ROM ITERP/07 and http://www.iaea.org/ [169] Chu M.S. et al 2004 Phys. Plasmas 11 2497
programmes/ripc/physics/fec2000/html/nodel.htm [170] Liu Y.Q. et al 2004 Nucl. Fusion 44 77
[127] Hayashi N. et al 2004 Nucl. Fusion 44 477 [171] Medvedev S.Yu. and Pustovitov V.D. 2003 Plasma Phys.
[128] Hayashi N. et al 2004 J. Plasma Fusion Res. 80 605 Rep. 29 1009
[129] Hegna C.C. and Callen J.D. 1997 Phys. Plasmas 4 2940 [172] Okabayashi M. et al 1996 Nucl. Fusion 36 1167
[130] Strait E.J. et al 1995 Phys. Rev. Lett. 74 2483 [173] Sabbagh S.A. et al 2002 Phys. Plasmas 9 2085
[131] Taylor T.S. et al 1995 Phys. Plasmas 2 2390 [174] Takeji S. et al 2002 Nucl. Fusion 42 5
[132] Bondeson A. and Ward D.J. 1994 Phys. Rev. Lett. 72 2709 [175] Pinches S.D. et al 2003 Proc. 30th EPS Conf. on Controlled
[133] Chu M.S. et al 1995 Phys. Plasmas 2 2236 Fusion and Plasma Physics (St Petersburg, Russia, 2003)
[134] Bondeson A. and Chu M.S. 1996 Phys. Plasmas 3 3013 vol 27A (ECA) P-1.93
[135] Kuvshinov B.N. Mikhailovskii A.B. 1998 Plasma Phys. Rep. [176] Reimerdes H. et al 2005 Proc. 32nd EPS Conf. on Plasma
24 623 Physics and Controlled Fusion (Tarragona, Spain, 2005)
[136] Garofalo A.M. et al 2000 Proc. 18th Int. Conf. on Fusion paper P5.056
Energy 2000 (Sorrento, Italy, 2000) (Vienna: IAEA) [177] Garofalo A.M. et al 2001 Nucl. Fusion 41 1171
CD-ROM EXP3-01 and http://www.iaea.org/programmes/ [178] Boozer A.H. 2003 Phys. Plasmas 10 1458
ripc/physics/fec2000/html/nodel.htm [179] Pustovitov V.D. 2003 Proc. 30th EPS Conf. on Controlled
[137] Boozer A.H. 2001 Phys. Rev. Lett. 86 5059 Fusion and Plasma Physics (St. Petersburg, Russia, 2003)
[138] Garofalo A.M., Jensen T.H. and Strait E.J. 2002 Phys. vol 27A (ECA) P-4.167 and http://eps2003.ioffe.ru/
Plasmas 9 4573 [180] Pustovitov V.D. 2002 Plasma Phys. Control. Fusion 44 295
[139] Garofalo A.M. et al 2002 Phys. Rev. Lett. 89 235001 [181] Garofalo A.M., et al 1999 Phys. Rev. Lett. 82 3811
[140] La Haye R.J. et al 2005 Nucl. Fusion 44 1197 [182] Takeji S. et al 2002 J. Plasma Fusion Res. 78 447
[141] Hender T.C. et al 2004 Proc. 20th Int. Conf. on Fusion Energy [183] Garofalo A.M. et al 2002 Phys. Plasmas 9 1997
(Vilamoura, Portugal, 2004) (Vienna: IAEA) CD-ROM [184] Shilov M. et al 2004 Phys. Plasmas 11 2573
EX/P2-22 and http://www.naweb.iaea.org/napc/physics/ [185] Sabbagh S.A. et al 2000 Proc. 20th Int. Conf. on Fusion
fec/fec2004/index.html Energy 2004 (Vilamoura, Portugal, 2004) (Vienna: IAEA)
[142] Reimerdes H. et al 2006 Phys. Plasmas 13 056107 CD-ROM EX/3-2 and http://www.naweb.iaea.org/napc/
[143] Pustovitov V.D. 2003 JETP Lett. 78 281 physics/fec/fec2004/index.html
[144] Gregoratto D. et al 2001 Plasma Phys. Control. Fusion [186] Garofalo A.M., La Haye R.J. and Scoville J.T. 2002 Nucl.
43 1425 Fusion 42 1335

S199
T.C. Hender et al

[187] Wade M.R. et al 2003 Nucl. Fusion 43 634 (Albuquerque, USA, 2003) Bull. Am. Phys. Soc. 48 264
[188] Anderson P.M. et al 2003 Fusion Eng. Des. 66–68 791 paper QP1.074 and http://www.aps.org/meet/DPP03/
[189] Jackson G.L. et al 2003 Proc. 30th EPS Conf. on Controlled baps/abs/S2080036.html
Fusion and Plasma Physics (St Petersburg, Russia, 2003) [227] Scoville J.T. et al 2004 46th Annual Meeting of the Division
vol 27A (ECA) P-4.47 of Plasma Physics, American Physical Society (Savannah,
[190] Strait E.J. et al 2004 Phys. Plasmas 11 2505 USA, 2004) paper NP1.008
[191] Menard J.E. et al 2003 Nucl. Fusion 43 330 [228] Luxon J.L. et al 2003 Nucl. Fusion 43 1813
[192] Sabbagh S.A. et al 2004 Nucl. Fusion 44 560 [229] Schaffer M.J et al 2003 45th Annual Meeting of the Division
[193] Tsuzuki K. et al 2004 Proc. 20th Int. Conf. on Fusion Energy of Plasma Physics, American Physical Society
2004 (Vilamoura, Portugal, 2004) (Albuquerque, USA, 2003) Bull. Am. Phys. Soc. 48 264,
[194] Kurita G. et al 2004 Proc. 20th Int. Conf. on Fusion Energy poster QP1.035 and http://www.aps.org/meet/DPP03/baps/
2004 (Vilamoura, Portugal, 2004) (Vienna: IAEA) abs/S2080036.html
CD-ROM FT/P7-7 and http://www.naweb.iaea.org/napc/ [230] Scoville J.T. and La Haye R.J. 2003 Nucl. Fusion 43 250
physics/fec/fec2004/index.html [231] Isei N. et al 2001 Fusion Technol. 39 1101
[195] Ward D.J. and Bondeson A. 1995 Phys. Plasmas 2 1570 [232] Brennan D.P. et al 2002 Phys. Plasmas 9 2998
[196] Bondeson A. et al 2002 Phys. Plasmas 9 2044 [233] Amoskov V. et al 2004 Plasma Devices Oper. 12 285
[197] Cates C. et al 2000 Phys. Plasmas 7 3133 [234] Amoskov V. et al 2005 Plasma Devices Oper. 13 87
[198] Maurer D.A. et al Proc. 19th Int. Conf. on Fusion Energy [235] Lin-Liu Y.R. et al 1999 Phys. Plasmas 6 3934
2002 (Lyon, France, 2002) (Vienna: IAEA) CD-ROM [236] Gruber O. et al 1999 Phys. Rev. Lett. 83 1787
TH/P3-13 and http://www.iaea.org/programmes/ripc/ [237] Kamada Y. et al 1999 Nucl. Fusion 39 1845
physics/fec2002/html/fec2002.htm [238] Rice B.W. et al 1999 Nucl. Fusion 39 1855
[199] Bondeson A. et al 2001 Nucl. Fusion 41 455 [239] Wade M.R. et al 2001 Phys. Plasmas 8 2208
[200] Boozer A.H. 1998 Phys. Plasmas 5 3350 [240] Joffrin E. et al 2005 Nucl. Fusion 45 626
[201] Reimerdes H. et al 2004 Phys. Rev. Lett. 93 135002 [241] Wade M.R. et al 2005 Nucl. Fusion 45 407
[202] Reimerdes H. et al 2004 Proc. 20th Int. Conf. on Fusion [242] Staebler A. et al 2005 Nucl. Fusion 45 617
Energy 2004 (Vilamoura, Portugal, 2004) (Vienna: IAEA) [243] Lao L.L. et al 1996 Phys. Plasmas 3 1951
CD-ROM EX/3-1Rb and http://www.naweb.iaea.org/napc/ [244] Takeji S. et al 1997 Phys. Plasmas 4 4283
physics/fec//fec2004/index.html [245] Huysmans G.T.A. et al 1999 Nucl. Fusion 39 1489
[203] Polevoi A.R. et al Proc. 19th Int. Conf. on Fusion Energy [246] Günter S. et al 2001 Nucl. Fusion 41 1283
2002 (Lyon, France, 2002) (Vienna: IAEA) CD-ROM [247] Manickam J. et al 1999 Nucl. Fusion 39 1819
CT/P-08 and http://www.iaea.org/programmes/ripc/ [248] Turnbull A.D. et al 1998 Nucl. Fusion 38 1467
physics/fec2002/html/fec2002.htm [249] Sarazin Y. et al 2002 Plasma Phys. Control. Fusion 44 2445
[204] Navratil G.A. et al 2003 RWM Control in FIRE and ITER [250] Burrell K.H. et al 2001 Phys. Plasmas 8 2153
Workshop on Active Control of MHD stability (University [251] Doyle E.J. et al 2002 Nucl. Fusion 42 333
of Texas-Austin) [252] Greenfield C.M. et al 2002 Plasma Phys. Control. Fusion
[205] Pustovitov V.D. and Medvedev S. Yu. 2003 Modeling of 44 A123
feedback stabilization of RWM in the T-15M tokamak Web [253] Menard J.E. et al 2004 Phys. Plasmas 11 639
Proc. EPS 2003 (St Petersburg, July 2003) paper P-3.136 [254] Luce T.C. et al 2001 Nucl. Fusion 41 1585
and http://eps2003.ioffe.ru/ [255] Ferron J.R. et al 2005 Phys. Plasmas 12 056126
[206] Liu Y.Q. et al 2005 Nucl. Fusion 45 1131 [256] Buttery R.J. et al 2004 Nucl. Fusion 44 1027
[207] Medvedev S.Yu. 2003 private communication, (Moscow [257] Takeji S. et al 2000 J. Plasma Fusion Res. 76 575
State University) [258] Hender T.C. et al 2002 Plasma Phys. Control. Fusion
[208] Bialek J. 2005 private communication (Columbia University, 44 1143
New York) [259] Turnbull A.D. et al 2002 Nucl. Fusion 42 917
[209] Manickam J. 2003 private communication (PPPL, USA) [260] Günter S. et al 2000 Nucl. Fusion 40 1541
[210] Buttery R.J. et al 2000 Nucl. Fusion 40 807 [261] Callen J.D. et al 1999 Phys. Plasmas 6 2963
[211] La Haye R.J. et al 1992 Phys. Fluids B 4 2098 [262] Turnbull A.D. et al 1997 Proc. 16th Int. Conf. on Fusion
[212] Buttery R.J. et al 1999 Nucl. Fusion 39 1827 Energy 1996 (Montreal, Canada, 1996) vol 2 (Vienna:
[213] Scoville J.T. and La Haye R.J. 1997 Bull. Am. Phys. Soc. 42 IAEA) p 509
1979 [263] Ishii Y. et al 1998 Plasma Phys. Control. Fusion
[214] Buttery R.J. et al 1997 Proc. 24th EPS Conf. on Controlled 40 1607
Fusion and Plasma Physics (Berchtesgaden, Germany, [264] Ozeki T. et al 1995 Nucl. Fusion 35 861
1997) vol 21A 265 [265] Makowski M.A. et al 2003 Proc. 30th EPS Conf. on
[215] Scoville J.T. and La Haye R.J. 1996 Bull. Am. Phys. Soc. 41 Controlled Fusion and Plasma Physics (St Petersburg,
1570 Russia, 2003) vol P-2 p 113 and http://eps2003.ioffe.ru/
[216] Wolfe S. et al 2005 Phys. Plasmas 12 056110 [266] Greenfield C.M. et al 2003 Proc. 30th EPS Conf. on
[217] Koslowski H.R. et al 2004 Proc. 31st EPS Conf. on Plasma Controlled Fusion and Plasma Physics (St Petersburg,
Physics (London, UK, 2004) paper P1.124 Russia, 2003) vol P-4 p 92 and http://eps2003.ioffe.ru/
[218] Lazzaro E. et al 2002 Phys. Plasmas 9 3906 [267] Kessel C.E. et al 2003 Proc. 30th EPS Conf. on Controlled
[219] Fitzpatrick R. 1993 Nucl. Fusion 33 1049 Fusion and Plasma Physics (St Petersburg, Russia, 2003)
[220] Lazzaro E. and Zanca P. 2003 Phys. Plasmas 10 2399 vol P-4 p 44 and http://eps2003.ioffe.ru/
[221] Hender T.C. et al 2002 Proc. 19th Int. Conf. on Fusion [268] Ishii Y. et al 2000 Phys. Plasmas 7 4477
Energy 2002 (Lyon, France, 2002) (Vienna: IAEA) [269] Popov A.M. et al 2001 Phys. Plasmas 8 3605
CD-ROM EX/S1-2 and http://www.iaea.org/programmes/ [270] Fujita T. et al 2001 Phys. Rev. Lett. 87 245001
ripc/physics/fec2002/html/fec2002.htm [271] Hawkes N.C. et al 2001 Phys. Rev. Lett. 87 115001
[222] Shaing K.C. 2003 Phys. Plasmas 10 1443 [272] Chu M.S. and Parks P. 2002 Phys. Plasmas 9 5036
[223] La Haye R.J et al 1992 Nucl. Fusion 32 2119 [273] Takizuka T. 2002 J. Plasma Fusion Res. 78 1282
[224] Zhu W. et al 2006, Phys Rev Lett 96 225002 [274] Martynov A.A. et al 2003 Phys. Rev. Lett. 91 085004
[225] Fitzpatrick R. and Hender T.C. 1991 Phys. Fluids B [275] Huysmans G.T.A. et al 2001 Phys. Rev. Lett. 87 245002
3 644 [276] Huysmans G.T.A. et al 2002 Proc. 29th EPS Conf. on
[226] Scoville J.T. et al 2003 45th Annual Meeting of the Division Plasma Physics and Controlled Fusion (Montreux,
of Plasma Physics, American Physical Society Switzerland, 2002) pp 17–21

S200
 Chapter 3: MHD stability, operational limits and disruptions

[277] Stratton B.C. et al 2002 Plasma Phys. Control. Fusion [320] Pautasso G. and Gruber O. Fusion Sci. Technol. 44 (2003) 716
44 1127 [321] Neyatani Y. et al Proc. 17th Int. Conf. on Fusion Energy 1998
[278] Ozeki T. et al 2003 Proc. 30th EPS Conf. on Controlled (Yokohama, Japan, 1998) (Vienna: IAEA) CD-ROM file
Fusion and Plasma Physics (St Petersburg, Russia, 2003) EXP3/11 and http://www.iaea.org/programmes/ripc/
vol P-2 pp 111 and http://eps2003.ioffe.ru/ physics/start.htm
[279] Takeji S. et al 2002 Fusion Sci. Technol. 42 278 [322] Andrew P. et al 2005 J. Nucl. Mater. 337–339 99
[280] Shimomura Y. et al 2001 Plasma Phys. Control. Fusion [323] Andrew P. et al 2003 Proc. 30th EPS Conf. on Controlled
43 A385 Fusion and Plasma Physics (St Petersburg, Russia, 2003)
[281] Holties H.A. et al 1996 Nucl. Fusion 36 973 vol 27A P-1.108 and http://eps2003.ioffe.ru/
[282] Bondeson A. et al 1997 Nucl. Fusion 37 1419 [324] Whyte D.G. et al 2003 J. Nucl. Mater. 313–316 1239
[283] Pereverzev G.V. et al 1991 Report IPP5/42 [325] Loarte A. et al 2004 Proc. 20th Int. Conf. on Fusion Energy
[284] Degtyarev L. et al 1997 Comput. Phys. Commun. 103 10 2004 (Vilamoura, Portugal, 2004) (Vienna: IAEA)
[285] Polevoi A.R. et al Proc. 20th Int. Conf. on Fusion Energy CD-ROM IT/P3-34 and http://www.naweb.iaea.org/napc/
2004 (Vilamoura, Portugal, 2004) (Vienna: IAEA) physics/fec/fec2004/index.html
CD-ROM IT/P3-28 and http://www.naweb.iaea.org/napc/ [326] Sugihara M. et al 2003 Plasma Fusion Res. 79 706
physics/fec/fec2004/index.html [327] Sugihara M. et al 2004 Proc. 20th Int. Conf. on Fusion
[286] Shimada M. et al 2005 Nucl. Fusion 44 350 Energy 2004 (Vilamoura, Portugal, 2004) (Vienna: IAEA)
[287] Parker R.R 2000 Nucl. Fusion 40 473 CD-ROM IT/P3-29 and http://www-naweb.iaea.org/napc/
[288] Aymar R. et al 2002 Proc. 19th Int. Conf. on Fusion Energy physics/fec/fec2004/datasets/index.html
2002 (Lyon, France, 2002) (Vienna: IAEA) CD-ROM [328] Hyatt A. 2004 private communication (San Diego: General
OV/1-1 and http://www.iaea.org/programmes/ripc/physics/ Atomics)
fec2002/html/fec2002.htm [329] Wesley J. et al 2006 Proc. 21st Int. Conf. on Fusion Energy
[289] Sokolov Yu. A. 1979 JETP Lett. 29 244–6 Research 2006 (Chengdu, China, 2006) (Vienna: IAEA)
[290] Rosenbluth M.N. and Putvinski S.V. 1997 Nucl. Fusion CD-ROM IT/P1-21
37 1355 [330] Nakamura Y. et al 2002 Plasma Phys. Control. Fusion
[291] Plyusnin V.V. et al 2006 Nucl. Fusion 46 277 Plasma Physics 44 1471
(Funchal, Portugal, 2001) vol 25A (ECA) pp 1805 [331] Humphreys D.A. and Kellman A.G. 1999 Phys. Plasmas
[292] Kadomtsev B.B. 1975 Plasma Phys. (Russ) 1 710 6 2742
[293] Kadomtsev B.B. and Pogutse O. 1974 Sov. Phys.–JETP 38 [332] Neyatani Y. et al 1999 Nucl. Fusion 39 559
283 [333] Riccardo V. et al 2004 Plasma Phys. Control. Fusion 46 925
[294] Wesson J.A. et al 1997 Phys. Rev. Lett. 79 5018 [334] Yoshino R., Nakamura Y. and Neyatani Y. 1996 Nucl. Fusion
[295] Taylor P.L. et al 1996 Phys. Rev. Lett. 76 916 36 295
[296] Helander P. et al 2002 Phys. Rev. Lett. 89 235002 [335] Riccardo V. et al 2003 Plasma Phys. Control. Fusion
[297] Mirnov S. et al 1998 Phys. Plasmas 5 3950 45 A269
[298] Mirnov S.V. 2001 Proc. 28th EPS Conf. on Controlled Fusion [336] Counsell G.F. et al 2003 Nucl. Fusion 43 1197
and Plasma Physics (Funchal, Portugal, 2001) vol 25A [337] Irby J. et al 2006 Fusion Sci. Technol. 51 460
(ECA) pp 1473–6 [338] Khayrutdinov R.R. and Lukash V.E. 1993 Comput. Phys.
[299] Kleva R.G. and Guzdar P.N. 2001 Phys. Plasmas 8 103 Commun. 109 193
[300] Kruger S.E. et al 2004 Computer Phys. Commun. 164 34 [339] Riccardo V., Walker S. and Noll P. 2000 Plasma Phys.
[301] Cowley S.C. et al 2003 Plasma Phys. Control. Fusion 45 A31 Control. Fusion 42 29
[302] Salzedas F. et al 2003 Phys. Plasmas 9 3402 [340] Riccardo V. Walker S. and Noll P. 2000 Fusion Eng. Des.
[303] Salzedas F. et al 2002 Proc. 29th EPS Conf. on Controlled 47 389
Fusion and Plasma Physics (Montreux, Switzerland, 2002) [341] Riccardo V., Noll P. and Walker S.P. 2000 Nucl. Fusion
P-1.039 40 1805
[304] Salzedas F. et al 2003 Proc. 30th EPS Conf. Controlled [342] Harvey R.W. et al 2000 Phys. Plasmas 7 4590
Fusion and Plasma Physics (St Petersburg, Russia 2003) [343] Yoshino R., Tokuda S. and Kawano Y. 1999 Nucl. Fusion
vol 27A P-2.95 and http://eps2003.ioffe.ru/ 39 151
[305] Hyatt A.W. et al 2001 Bull. Am. Phys Soc. 45 300 [344] Gill R.D. et al 2002 Nucl. Fusion 42 1039
[306] Pautasso G. et al 2001 J. Nucl. Mater. 290–293 1045 [345] Martin G. 2000 Proc. 6th IAEA TCM on Energetic Particles
[307] 2001 ITER Final Design Report complied by the ITER in Magnetic Confinement Systems (Naka, Japan, 2000)
Director (http://www.naka.jaea.go.jp/ITER/FDR/) JAERI-Conf 2000-004, 1
[308] Riccardo V. et al 2005 Nucl. Fusion 45 1427 [346] Gill R.D. et al 2000 Nucl. Fusion 40 163
[309] Schuller F.C. et al 1995 Plasma Phys. Control. Fusion [347] Helander P., Eriksson L.-G. and Andersson F. 2002 Plasma
37 A135 Phys. Control. Fusion 44 B247
[310] Pautasso G. et al 2003 Proc. 30th EPS Conf. on Plasma [348] Helander P. et al 2003 Runaway electrons and current
Physics (St Petersburg, Russia, 2003) vol 27A P-1.135 and dynamics during tokamak disruptions 8th IAEA Technical
http://eps2003.ioffe.ru/ Meeting on Energetic Particles in Magnetic Confinement
[311] Pautasso G. et al 2004 Proc. 31st EPS Conf. on Plasma Systems (San Diego, 6–8 October 2003)
Physics (London, UK, 2004) 28G P-4.132 [349] Eriksson L.-G. and Helander P. 2003 Comput. Phys.
[312] Konz C. et al 2005 Proc. 32nd EPS Conf. on Plasma Physics Commun. 154 175
and Controlled Fusion, (Tarragona, Spain, 2005) vol 29C [350] Schittenhelm M. 1997 Proc. 24th Conf. on Controlled Fusion
pp O2-005 and Plasma Physics (Berchtesgarden, Germany, 1997)
[313] Counsell G.F. et al 2005 Nucl. Fusion 45 S157 vol 21A, Part III, (Geneva: European Physical Society)
[314] Andrew P. et al 2006 Main chamber power loads during p 985
disruptions Preprint EFD-C(06)02/08 [351] Martin G. et al 2004 Proc. 20th Int. Conf. on Fusion Energy
[315] Herrmann A. et al 2004 Plasma Phys. Control. Fusion 46 971 (Vilamoura, Portugal, 2004) (Vienna: IAEA) CD-ROM
[316] Ciotti M. et al 1999 J. Nucl. Mater. 266–269 1023 EX/10-6Rc and http://www-naweb.iaea.org/napc/
[317] Finken K.H. et al 2001 J. Nucl. Mater. 290–293 1064 physics/fec/fec2004/datasets/index.html
[318] Wesley J. et al 1998 IEEE Proc. 17th Symp. Fusion [352] Lukash V.E. and Khayrutdinov R.R. 2000 Proc. 6th IAEA
Engineering (San Diego, USA, 1997) vol 1 (Piscataway: TCM on Energetic Particles in Magnetic Confinement
IEEE) p 483 Systems (Naka, Japan, 2000) JAERI-Conf 2000-004, 13
[319] Tamai H. et al 2002 Nucl. Fusion 42 290 [353] Maddaluno G. et al 2003 J. Nucl. Mater. 313–316 651

S201
T.C. Hender et al

[354] Yoshino R., Nakamura Y. and Neyatani Y. 1997 Nucl. Fusion [389] Cannas B. et al 2004 Nucl. Fusion 44 68
37 1161 [390] Windsor C.G. et al 2005 Nucl. Fusion 45 337
[355] Andersson F., Helander P. and Eriksson L.-G. 2001 Phys. [391] Fishpool G.M. and Haynes P.S. 1994 Nucl. Fusion 34 109
Plasmas 8 5221 [392] La Haye R.J. et al 1997 General Atomics Report GA-A22468
[356] Kawano Y. et al 2005 J. Plasma Fusion Res. 81 593 [393] Mertens V. et al 2003 Fusion Sci. Technol. 44 593
[357] Yoshino R. and Tokuda S. 2000 Nucl. Fusion 40 1293 [394] Esser H.G. et al 1997 J. Nucl. Mater. 241–243 861
[358] Kawano Y. et al 2005 J. Plasma Fusion Res. 81 743 [395] Martin Y. et al 1998 Proc. 25th EPS Conf. on Controlled
[359] Tokuda S. and Yoshino R. 1999 Nucl. Fusion 39 1123 Fusion and Plasma Physics (Prague, Czech Republic,
[360] Martin-Solis J.R. et al 2006 33rd EPS Conf. Plasma Physics 1998) P3.017
(Rome, Italy, 2006) P5.078 [396] Scoville J. et al 1991 Nucl. Fusion 31 875
[361] Kawano Y. et al 1997 Proc. 16th Int. Conf. on Fusion Energy [397] Kraemer-Flecken A. et al 2003 Nucl. Fusion 43 1437
1996 (Montreal, Canada, 1996) vol 1 (Vienna: IAEA) [398] Kraemer-Flecken A. et al 2001 Fusion Eng. Des. 56–57 773
p 345 [399] Hoshino K. et al 1995 Proc. 15th Int. Conf. on Plasma
[362] Kawano Y. et al 2002 Fusion Sci. Technol. 42 298 Physics and Controlled Nuclear Fusion Research 1994
[363] Helander P., Eriksson L.-G. and Andersson F. 2000 Phys. (Seville, 1994) (Vienna: IAEA) vol 1 p 697
Plasmas 7 4106 [400] Salzedas F. et al 1999 Proc. 26th EPS Conf. on Controlled
[364] Jardin S.C. et al 2000 Nucl. Fusion 40 923 Fusion and Plasma Physics (Maastricht, The Netherlands,
[365] Taylor P.L. et al 1999 Phys. Plasmas 6 1872 1999) vol 23J p 625
[366] Whyte D.G. et al 2002 Phys. Rev. Lett. 89 055001 [401] Yoshino R. et al 1994 J. Plasma Fusion Res. 70 1081
[367] Bakhtiari M. et al 2002 Nucl. Fusion 42 1197 [402] Oikawa T. et al 2004 Fusion Eng. Des. 70 175
[368] Bakhtiari M. et al 2005 Nucl. Fusion 45 318 [403] Joffrin E. et al 2003 Plasma Phys. Control. Fusion
[369] Humphreys D.A. and Whyte D.G. 2000 Phys. Plasmas 45 A367
7 4057 [404] Bucalossi J. et al 2001 J. Nucl. Mater. 290–293 566
[370] Whyte D.G., Humphreys D.A. and Taylor P.L. 2000 Phys. [405] Reimerdes H. et al 2003 Proc. 30th EPS Conf. on Controlled
Plasmas 7 4052 Fusion and Plasma Physics (St Petersburg, Russia, 2003)
[371] Hollmann E.M. et al 2005 Nucl. Fusion 45 1046 vol 27A P-4.45 and http://eps2003.ioffe.ru/
[372] Kawano Y. et al 2001 Study on characteristics of runway [406] Testa D. et al 2000 Proc. 27th EPS Conf. on Controlled
electrons in JT-60U 18th JSPF Meeting (Fukuoka, 27–30 Fusion and Plasma Physics (Budapest, Hungary, 2000)
November 2001) vol 24B p 1429 and (http://202.127.204.25/gjhy/
[373] Gribov Y. et al 2007 Progress in the ITER Physics Basis EPS/27th(2000)/pdf/p4 044.pdf) (2004) (Vienna: IAEA)
Nucl. Fusion 47 S385–S403 CD-ROM EX/P2-33 and http://www-naweb.iaea.org/
[374] Lukash V.E. and Khayrutdinov R.R. 1996 Plasma Phys. Rep. napc/physics/fec/fec2004/index.html
22 91 [407] Franzen P. et al 1998 Fusion Technol. 33 84
[375] Jardin S.C. et al 1986 Comput. Phys. Commun. 66 481 [408] Mertens V. et al 2003 Fusion Eng. Des.
[376] Sugihara M. et al 2004 Plasma Phys. Control. Fusion 66–68 119
46 1581 [409] Granetz R.S. et al 1996 Proc. 16th Int. Conf. on Fusion
[377] Humphreys D.A. et al 1997 General Atomics Report Energy 1996 (Montreal, Canada, 1996) vol 1 (Vienna:
GA-22692 IAEA) p 757
[378] Khayrutdinov R.R. et al 2001 Plasma Phys. Control. Fusion [410] Pautasso G. et al 1996 Nucl. Fusion 36 1291
43 321 [411] Strauss H.R. and Park W. 1998 Phys. Plasmas 5 2676
[379] Lukash V.E. 2002 Validation of DINA halo area expansion [412] Whyte D.G. et al 1998 Phys. Rev. Lett. 81 4392
model against JT-60U disruption data 2nd Meeting of the [413] Rosenbluth M.N., Putvinskij S.V. and Parks P.B. 1997 Nucl.
ITPA Topical Group on MHD, Disruptions and Control Fusion 37 955
(Garching, Germany, October 2002) [414] Pautasso G. et al 2002 Proc. 29th EPS Conf. on Plasma
[380] Park W. et al 1999 Phys. Plasmas 6 1796 Physics and Controlled Fusion (Montreux, Switzerland,
[381] Paccagnella R. et al 2005 Fusion Eng. Des. 75–79 589 2002) vol 26B, P2.051
[382] Pletzer A. http://w3.pppl.gov/rib/repositories/NTCC/ [415] Finken K.H. et al 2001 Nucl. Fusion 41 1651
catalog/Asset/grin.html [416] Jernigan T.C. et al 2005 private communication (ORNL
[383] Paccagnella R. et al 2003 Halo current simulation for USA)
tokamak plasmas 45th APS Division Plasma Physics [417] Izzo V.A. 2006 Nucl. Fusion 46 541
Meeting (Albuquerque, USA, 2003) paper QP1.079 and [418] Parks P.B. et al 1997 Fusion Technol. 35 267
http://www.aps.org/meet/DPP03/baps/abs/S2080036.html [419] Summers D.A. 1995 Waterjetting Technology (London:
[384] Wroblewski D., Jahns G.L. and Leuer J.A. 1997 Nucl. Fusion Chapman & Hall)
37 725 [420] Nagata M. et al 2005 Nucl. Fusion 45 1056
[385] Sengupta A. and Ranjan P. 2001 Nucl. Fusion 41 487 [421] Parks P.B. 2005 Gas jet pressure requirement for direct
[386] Pautasso G. et al 2002 Nucl. Fusion 42 100 penetration, private communication (General Atomics,
[387] Yoshino R. 2003 Nucl. Fusion 43 1771 San Diego, USA)
[388] Yoshino R. 2005 Nucl. Fusion 45 1232 [422] Kuteev B.V., Sergeev Yu V. and Sudo S. 1995 Nucl. Fusion
35 1167

S202