PAPER Related content

- Three-dimensional distortions of the

Nonlinear simulation of ELM dynamics in the tokamak plasma boundary: boundary
displacements in the presence of resonant
presence of resonant magnetic perturbations magnetic perturbations
I.T. Chapman, M. Becoulet, T. Bird et al.

- Increase of turbulence and transport with

To cite this article: D. Chandra et al 2017 Nucl. Fusion 57 076001 resonant magnetic perturbations in ELM-
suppressed plasmas on DIII-D
G.R. McKee, Z. Yan, C. Holland et al.

- Gyrofluid computation of magnetic

perturbation effects on turbulence and
View the article online for updates and enhancements. edge localized bursts
J. Peer, A. Kendl, T.T. Ribeiro et al.

Recent citations
- Visco-resistive MHD study of internal kink
(m = 1) modes
J. Mendonca et al

This content was downloaded from IP address 103.49.147.111 on 24/04/2018 at 10:21

 International Atomic Energy Agency Nuclear Fusion

Nucl. Fusion 57 (2017) 076001 (7pp) https://doi.org/10.1088/1741-4326/aa6dca

Nonlinear simulation of ELM dynamics

in the presence of resonant magnetic
perturbations
D. Chandra1, A. Thyagaraja2, A. Sen1 and P. Kaw1
1
  Institute for Plasma Research, Bhat, Gandhinagar 382428, India
2
  Astrophysics Group, University of Bristol, Bristol, BS8 1TL, United Kingdom

E-mail: debasis@ipr.res.in

Received 14 December 2016, revised 17 March 2017

Accepted for publication 19 April 2017
Published 11 May 2017

Abstract
We report on nonlinear simulation studies on the dynamical behaviour of ELMs under
the influence of resonant magnetic perturbations (RMPs) using a two-fluid initial value
electromagnetic nonlinear global code (CUTIE). To simulate ELMs we introduce a particle
source in the confinement region and a particle sink in the edge region. To study ELM control
using RMPs we have applied an n  =  2 static external magnetic perturbation at the edge
and made detailed parametric studies under varying conditions for the machine and plasma
parameters typical of COMPASS-D. Our results show that ELM mitigation is possible for
RMP powers beyond a specific threshold. The results also provide valuable insights into the
RMP induced modifications of the complex nonlinear dynamics of the ELMs, in particular on
the redistribution of mode energy and the cascading of energy to shorter scale lengths. We also
observe a hysteresis in states as we increase the amplitude of RMPs and then decrease it to the
same value.

Keywords: ELMs, RMPs, simulations

(Some figures may appear in colour only in the online journal)

1. Introduction open issues associated with ELM simulations include cap-

The occurrence of edge localized modes (ELMs) during the
H mode operation of a tokamak remains a serious concern for
future devices like ITER because of their potential deleterious
effects on the plasma facing components and divertor plates.
These finite-amplitude perturbations of the plasma edge region
which originate due to MHD instabilities driven by steep
pressure gradients or edge plasma currents present during an
H-mode have therefore been the subject of much theoretical and
experimental studies over the past several years [1–9]. While
the linear characteristics of these modes have been extensively
modelled and attributed to the excitation of peeling/ballooning
modes, their nonlinear dynamics is quite complex and far from
being fully understood. Large scale numerical simulations on
a number of nonlinear MHD and two fluid codes are there-
fore currently being carried out to explore various facets of the
nonlinear behaviour of ELMs [10–12]. Some of the important
turing their multicycle nature as well as determining the factors
that influence their sizes and frequencies. A further challenge
is in understanding the physical mechanisms underlying the
suppression/mitigation of ELMs by application of RMPs as
observed in several experiments. In this work, we report on
nonlinear simulation studies aimed at exploring the dynamical
behaviour of ELMs under the influence of RMPs.
The paper is organized as follows. In the next section 
(section 2) the physical basis of the simulation and the two-
fluid model are described. Section  3 presents our nonlinear
simulation results in the presence of RMP amplitudes held
constant in time. In section 4 the spectral properties of edge
turbulence and possible mechanisms of RMP-ELM interac-
tion in our simulations are discussed. In section 5 we discuss
the plasma state hysteresis with programmed time variations
of large amplitude RMPs. Finally we provide some conclu-
sions and a brief discussion of open questions in section 6.

1741-4326/17/076001+7$33.00 1 © 2017 IAEA, Vienna  Printed in the UK

Nucl. Fusion 57 (2017) 076001 D. Chandra et al

2.  Physical basis of the simulations it also successfully modelled the L-H transition behaviour in
and the two-fluid model COMPASS-D [15].
We reproduce the equations  used in CUTIE to make
Our simulations have been carried out using the Culham trans- this paper self-contained (more details are available in the
porter of ions and electrons (CUTIE) code [13–15]. CUTIE references cited, in particular in [13, 15])
is a two-fluid electromagnetic nonlinear global code which
has been used in a variety of simulations, both of transport ∂n
+ ∇ . (n v) = Sp
(1)
processes [16–19] in a range of tokamaks and also in model- ∂t
ling visco-resistive, single-fluid MHD in both linear and non-
linear regimes [20]. The full set of model fluid equations [15] dv j×B
m in
(2) = −∇p + + Feff
are solved for the so-called mesoscale, an intermediate scale dt c
between the device size and the ion gyro radius, incorporating
approximations for the underlying classical and neoclassical 3 dpe,i
(3) + pe,i ∇ . ve,i = −∇ . q e,i + Pe,i
transport effects. The code is based on a periodic cylinder 2 dt
model of the tokamak geometry where the toroidal curvature
of the magnetic field lines is taken into account to first order v ×B ∇p
E+ e
(4) = − e + Re
in the inverse aspect ratio. The poloidal mode numbers are c en
coupled through additional curvature terms, each to its side-
bands. The code has been used earlier to simulate ELMs, in 4π j
∇×B =
(5) .
the absence of RMPs, by creating a sharp density gradient c
at the edge through a careful combination of a source and a The following definitions apply: (m i + m e )v = m i vi + m e ve;
sink [15]. A simple particle (and energy) sink is introduced B = ∇ψ × b t + B0 bt, where b t is the unit vector in the ‘toroidal’
in the region r/a  >  0.95 and a particle source is placed in the direction. vi,e, q i,e are the ion and electron fluid velocities and
confinement region r/a  <  0.95. This combination produces a 1 ∂ψ
strong density gradient at the edge, somewhat inside of the heat fluxes respectively. The electric field E = − c ∂t b t − ∇φ,
edge of the sink, where the transport barrier forms, thereby and the plasma pressure, p = pe + pi = n(Te + Ti ). Sp is the
producing a mock-up of what happens in an X-point geometry net particle source; Feff represents the total viscous stress com-
in real experiments. Thus our plasma region is from r/a  =  0 prising neoclassical and turbulent viscosity; Pei is the net energy
to r/a  =  0.95. The vacuum/SOL edge region i.e. r/a  =  0.95 source that includes Ohmic heating, auxiliary heating and
to 1 has very low fixed density, temperature, and current. At electron-ion equilibration; R e represents the effects from neo-
r/a  =  1 i.e. at the ‘wall’, all fluctuating quantities are set to classical and turbulent resistivity. The reduction of these two-
zero. No boundary conditions are imposed at r/a  =  0.95. The fluid, quasi-neutral, nonlinear, driven-dissipative equations of
edge source is used for fueling the discharge while the line- motion to cylindrical geometry and the appropriate sources
averaged density is controlled by a feedback source with the and boundary conditions are available in the cited references.
profile Sp = S (r /a )[n¯ target − n¯] /τr. This feedback source pro- It should be stressed that the treatment of the source and sink
vides particles whenever the line-average density n becomes are completely independent of the RMP and remains the same
less than the target density n̄ target. Here S(r) is a fixed profile in the code, with or without RMPs. In particular, the very same
peaking at the edge and τr is a fueling time which is fixed prescriptions were used in the earlier work [15] where many
typically at 100 μs (see [15] for further details). It should features of RMP-free ELMing states were simulated.
be noted that these parameters have been chosen considering For our present studies on ELM control using RMPs we
the COMPASS-D conditions. The dynamics of ELMs may have applied an n  =  2 static external magnetic perturbation
be sensitive to the nature of the source/sink terms and may at the edge and made detailed parametric studies under var-
change if we take different forms of these. However as stated ying conditions. The model we use allows us to prescribe the
before, our aim in this paper is to study the relative changes values of the radial magnetic perturbations on r  =  a, which,
in the ELM behavior of [15] when RMPs are imposed. So in the simulation represents the wall. Thus, this represents a
we have strictly adhered to the sink/source model adopted boundary condition on the perturbed poloidal flux function
in [15]. δψ(r , θ, z /R, t ). It therefore induces a potential, static magnetic
A major strength of the code is that it incorporates turbu- perturbation for r  <  a and m  >  0; n  =  2. The amplitude of
lent transport effects and allows the development of profile- this and the m-dependence of this RMP field are given by the
turbulence interactions, thereby enabling a self-consistent following formula:
description of the evolution of ELM’s in the sense that MHD KaBpol 1
and transport are linked and the profile-turbulence interac- ψ RMP (a, θ, ζ ) = (
(6) ) ∑[ 2 ] cos(mθ + 2ζ )
tion is present. The source and sink terms do not form part of
B0 m≠0 m + 1

the self consistent model but have been introduced externally
to create appropriate conditions for the excitation of ELMs.
The source models the gas-puff whilst the sink represents
the loss effects in the SOL. CUTIE was one of the earliest
codes to observe repetitive edge relaxation oscillations and
where θ and ζ = z /R represent the poloidal and ‘toroidal’ coor-
dinates and K  BRMP /B0∣r = a is an amplitude parameter. In our
simulations m ranges from 1 to 32. The perturbed (potential)
field decays from the edge to the interior, and naturally inter-
acts with the turbulent modes, especially close to the plasma

2
Nucl. Fusion 57 (2017) 076001
Figure 1.  Wave forms of (dTi /T0i ) 2 at r/a  =  0.9 for zero (red spikes)
and for a non-zero value of the RMP amplitude (blue low amplitude
oscillations).
edge. The perturbation is even in n and relatively ‘flat in m’
reflecting the fact that the RMP coils in actual machines are
rather localized in poloidal angle.
We are aware that our model is qualitative and not
intended to simulate any experiment in all details, but to pro-
vide insights relating to dynamical behaviour of ELMs in
the presence of RMP-like field perturbations at the vacuum
vessel wall. We are particularly interested in trying to under-
stand the differences in ELM dynamics between our earlier,
RMP-free ELMs simulated by CUTIE [13] and the new sim-
ulations in which the RMP-field amplitude is varied over a
range of experimentally reasonable values, keeping all other
conditions exactly the same. It is our purpose in this work to
document the interesting results obtained in this comparative
study.
3.  Simulations of ELMs in presence of RMPs
Typical wave-forms of ion temperature fluctuations at r/a  =  0.9
obtained in CUTIE simulations with and without RMPs
are shown in figure  1 for the machine and plasma param­
eters typical of COMPASS-D with R  =  0.56 m, a  =  0.17 m,
Btor = 2.07 T, Ip = 242 KA and an electron cyclotron heating
power of PECH = 0.5 MW. The curve in red (tall spikes) is for
zero RMP amplitude, while the one in blue (small amplitude
oscillations) is for an amplitude K = 3 × 10−3. It is seen that
having the RMP on induces a dramatic change in the behaviour
of fluctuations, virtually suppressing the ELMs. The details of
the dynamics possibly responsible for this will be presented in
section 4.
As discussed in earlier work [15] the modes responsible
for ELMs in CUTIE simulations for ELMs have a typical bal-
looning character. This can be seen in the contour plots of the
ion temperature perturbations in figure  2. The predominant
toroidal mode numbers are between n ∼ 2–5 and the poloidal
mode numbers m ∼ 10–20.
The effects of RMP amplitudes on particle and energy con-
finement are interesting and non-trivial. The ‘particle confine-
ment time’ τp is defined at any instant in CUTIE thus:
3
D. Chandra et al
Ntot (t )
τp = .
N˙ (t )
Here Ntot (t ) is the total number of ions (and electrons) within
the plasma volume, and N˙ (t ) is the number injected per second
by the net source at that time. Similarly, we define,
W (t )
τE =
Ptot
where W is the total plasma internal energy and Ptot is the total
power (auxiliary plus Ohmic).
In figure 3(a) we compare the evolution in time of β (%) for
two cases: one with no RMP (red solid curve) and with RMP
amplitude (blue dotted curve) of 4 × 10−3 starting from the
same ELM-free state at t ∼ 263 ms. We note that the two traces
coincide and rise until t ∼ 267 ms. It is seen that the case with the
RMP overshoots this value and then both curves proceed with
some dynamical variations and reach a quasi-stationary state
at t ∼ 300 ms with β RMP ∼ 0.76 larger than β no − RMP ∼ 0.72.
In figure 3(b) we show the corresponding time variation of τE
over the same path. This clearly reflects the behaviour of β
and shows that τE − RMP > τE − no − RMP. Figure 3(c) displays the
corresponding behaviour of 1/τp. It is seen that this quantity
fluctuates much more than τE because the feedback control
on the line averaged density makes the net source vary in
time. Interestingly τp − RMP is seen to be consistently less than
τp − no − RMP indicating that the particle transport is enhanced by
the application of RMPs unlike energy transport.
4.  Spectral properties of edge turbulence
and possible mechanisms of RMP-ELM
interaction in CUTIE simulations
In this section  we consider the possible mechanisms which
are responsible for the effects of RMP fields on the nonlinear
dynamics of ELMs. First we present some interesting details
obtained from our nonlinear simulations of the power spectra
in both temporal and spatial dimensions. We have already pre-
sented the wave-forms of the ion temperature fluctuations at
r/a  =  0.9 over a 2 ms period (see figure 1) for the case of no
RMPs and a typical value of RMP amplitude. In figure 4 we
show the frequency spectra of the time series pertaining to the
wave-forms in the same location over the same period: the
sampling rate used allows us to have a maximum frequency
of 1 MHz, whilst the 2 ms epoch means that the resolution is
0.5 Hz. In figure  4(a) are shown the power spectra for ‘no
RMP’ and RMP amplitudes, (2, 3, 4) × 10−3.
It is clear that there are some remarkable features to these
traces: firstly we note that when no RMP is applied, the power
is very small for frequencies greater than 400 kHz. The spec-
trum for 2 × 10−3 amplitude follows the same trend, but dips
down considerably below the no RMP case beyond 200 kHz.
A further increase of the RMP amplitude results in rather dra-
matic spectral changes: at 3 × 10−3, there is a very consid-
erable reduction in power for frequencies less than 300 kHz.
However, we see that the power spectrum develops new high
Nucl. Fusion 57 (2017) 076001 D. Chandra et al

Figure 2.  Contours of ion temperature fluctuations before (a) and at the ELM crash (b).

Figure 3.  Effect of RMPs on time evolution of plasma β (a), the energy confinement time (b) and the particle confinement time (c). The
solid curve (red) represents the no RMP case and the dotted curve (blue) represents the evolution in the presence of RMP.

frequency features. Firstly, there is a sharp spectral peak near
600 kHz with a maximum which is only slightly smaller than
the zero frequency peak. There is also a second peak around
900 kHz. There is also a sharp drop in power at 840 kHz.
An increase of the RMP amplitude to 4 × 10−3 results in a
‘reversal’ of the spectrum at frequencies above 200 kHz but
below 400 kHz. The peak at 600 kHz disappears completely.
However, the power increases almost continuously to pro-
duce a rather broad peak stretching from 750 to 900 kHz. A
much smaller peak is also seen at the very highest frequen-
cies. Figure 4(b) shows the detail of the 0–25 kHz part of the
spectra. It is evident that the power at zero frequency in this
location is very low in the absence of RMP. As the amplitude
is raised, there is an increase, indicating a local, steady max-
imum (‘bump’) in the ion temperature fluctuation profile. As
the RMP amplitude is raised in stages, this peak also moves up
to 0.01. Nevertheless, it is clear that at 3 × 10−3, the low fre-
quency power is significantly lower and then starts increasing.
Figure 4(c) shows some interesting features of the com-
plicated structural characteristics of RMP-modified edge
fluctuations. We plot the ratio of the integrated power in the
0–200 kHz range to the total power. We also plot the total
power in arbitrary units. The figure  demonstrates that there
is a significant transfer of fluctuation energy from this ‘low
frequency’ range to the 200 kHz to 1 MHz range, as the RMP
amplitudes are raised from 0 to 4 × 10−3. The figure  also

4
Nucl. Fusion 57 (2017) 076001
Figure 4.  Power spectra for (a)‘no RMP’ and RMP amplitudes of (2, 3, 4) × 10−3 as distinguished by different colors (lines styles) shown
in the legend within the figure. (b) An amplified plot of the low frequency portion of the power spectra given in (a). In (c) the filled (red)
circles represent values of Plow /Ptotal while the filled (blue) squares are Ptotal in arbitrary units.
shows that the total power in the fluctuations remain nearly
constant over the range studied.
This figure  presents evidence that the RMP fields influ-
ence the edge fluctuations in important ways: Firstly, they
lead to an interesting zero frequency ‘bump’ (possibly a
‘locked’ mode) linked to the strongest RMP perturbation
at the outboard mid-plane, as also shown directly by other
diagnostics such as those in figure 2 when RMPs are active.
Secondly, they appear to transfer turbulence energy from the
low frequencies to much higher frequencies which tend to
then mitigate the ELM amplitude by reducing the density gra-
dients and removing the free energy available to drive them.
The higher frequencies, as observed in our simulations are of
the order of VAlfven /qR (and harmonics thereof) that fall in the
range of Alfven gap modes and/or shear Alfven frequencies.
Next we present some results pertaining to the nonlinear
dynamics of ELMs in the presence of RMPs. For this we
compare the poloidal mode number spectral plots of ELMs
in the absence and presence of RMPs. These are shown in
figures 5(a) and (b) respectively. We note that in the absence
of the RMP the spectrum is quite peaked around a mode
number of 10 and spatially localized in the region r /a ∼ 0.9
to 0.95. In the presence of the RMPs the spectrum gets con-
siderably broadened in mode space and also shifts spatially in
the inward direction. Nonlinear and linear toroidal coupling
of the low and high-k modes in the presence of RMPs may be
playing a role in causing this redistribution of mode energy
and the direct cascading of energy to shorter scale lengths.
The presence of higher m modes due to interaction with the
5
D. Chandra et al
RMPs also leads to a deeper penetration to occur close to the
edge.
5.  Plasma state hysteresis with programmed time
variations of large amplitude RMPs
In figures 6(a) and (b) we show our results for the changes in
β and energy confinement when the RMP amplitude is gradu-
ally changed. In figure 6(c) we show the time variation of the
imposed RMP field. Each point in the graph represents the
starting amplitude value of the RMP field at a particular time.
That amplitude is held constant for a period of 4 ms after which
it makes a rapid but smooth transit (over a time scale of 100 μs
as shown in the inset figure) to a higher (or lower) value of the
amplitude. For these runs we have used a momentum source in
the code which drives an axial core flow of 20 kms s−1 corre­
sponding to an Alfven mach number of 5 × 10−3. This was
done to study the effects of RMPs and ELMs on the flow pro-
file which is not discussed here, since the momentum source
was positioned at the core near the magnetic axis and the flow
is not significant in the edge region. The figure  shows that
when the amplitude is progressively raised to 1.7 times the
initial value of BRMP /B0 ∼ 7 × 10−3 during the time interval
of 393 ms to 417 ms there is a marked increase of plasma β
and energy confinement for RMP amplitudes beyond a spe-
cific threshold value (BRMP /B0 ∼ 10−2). However for much
higher RMP amplitudes, the confinement starts to degrade.
If we now decrease the RMP amplitude to earlier values e.g.
corre­sponding to those at t  =  420 ms the confinement does
Nucl. Fusion 57 (2017) 076001 D. Chandra et al

Figure 5.  Spectrum plot during ELMs without RMPs (a) and with RMPs (b).

Figure 6.  Dynamic effects of higher values of RMP amplitude on the time evolution of beta (a) and the energy confinement time (b). The
solid curves (red) show the evolution in the presence of an RMP that is programmed to first increase in amplitude and is then brought back
to its initial value. The dotted curves (blue) represent the evolution when the RMP amplitude is held at a constant value in time. In (c) the
time variation of the amplitude of BRMP /B0 has been plotted. The amplitude at a given time is held constant for a period of 4 ms after which
it makes a rapid but smooth transit (over a time scale of 100 μs, as shown in the inset figure) to a higher (or lower) value.

not necessarily revert to the same state, indicating the pres-
ence of hysteresis. The gentle time variation of the plasma
beta and energy confinement time in the presence of a con-
stant RMP and constant sources (blue curves in figure 6) are
related to the internal dynamics of the system governed by
profile-turbulence interactions. This variation (of the order
1.5% of the time averaged mean) is independent of the RMP
ampl­itude. The point of figure 6 is to demonstrate the differ-
ence of the system response (as represented by β and τE) to the
programmed RMP amplitude variation in time as described
in figure 6(c). As the red curves show, the final values of beta
and energy confinement time are significantly higher than the
constant RMP case, namely of the order of 10% of the time
averaged mean.
6.  Summary and discussions
To summarize, we have carried out numerical studies to examine
the influence of resonant magnetic perturbations on the edge
localized modes. Our simulations which utilize a particle source
in the confinement region and a particle sink in the edge SOL
region in the CUTIE code provide a simplified model description
of ELMs in the form of repeated bursts of fluctuations at the edge.
The application of RMPs is found to reduce the amplitude of these
bursts and to consequently improve the plasma beta and energy
confinement time of the system. However the particle confine-
ment time is found to get degraded. Our simulation results provide
some valuable insights into the RMP induced modifications of the
complex nonlinear dynamics of the ELMs such as the broadening

6
Nucl. Fusion 57 (2017) 076001
of the energy spectrum due to redistribution of the mode energy
and the direct cascading of energy to shorter scale lengths due
to nonlinear and linear toroidal coupling of the low and high-k
modes in the presence of RMPs. This effect can be qualitatively
understood in terms of the nonlinear modulational instability [21]
in three and four wave interactions. The RMP fields act by linear
coupling to provide a link between the low-k MHD-type insta-
bilities to much higher wave number stable modes (e.g. Alfven
continuum and gap modes possibly induced by RMPs) which can
effectively dissipate/disperse the substantial field energy released
by the low-k modes. The dynamical details of these processes
remain to be investigated in detail but they do appear to provide a
robust qualitative account of the results we have presented in this
work. It should be noted that a similar significant change in the
convective energy transport rate in the presence of RMPs has been
observed in simulations involving the interaction of RMPs and
unstable resistive ballooning modes (RBMs) in [22]. The basic
mechanism attributed for that process is nonlinear mode coupling
of fluctuations through side bands leading to a suppression of
bursts of RBM activity. In the present simulation study, we have
not considered the presence of an equilibrium plasma rotation
that can impact the penetration of RMPs through flow screening
effects [23]. The resultant modifications in the RMP-ELM inter-
action process remains an interesting and important open problem
that we hope to study in future.
D. Chandra et al
References
[1] Zohm H. 1996 Plasma Phys. Control. Fusion 38 105
[2] Snyder P.B., Wilson H.R., Osborne T.H. and Leonard A.W.
2004 Plasma Phys. Control. Fusion 46 A131
[3] Hender T.C. et al 2007 Nucl. Fusion 47 S128
[4] Hawryluk R.J. et al 2009 Nucl. Fusion 49 065012
[5] Evans T.E. et al 2008 Nucl. Fusion 48 024022
[6] Fenstermacher M.E. et al 2008 Phys. Plasmas 15 056122
[7] Kirk A. et al 2010 Nucl. Fusion 50 034008
[8] Jeon Y.M. et al 2012 Phys. Rev. Lett. 109 035004
[9] Kim J. et al 2017 Nucl. Fusion 57 022001
[10] Becoulet M. et al 2014 Phys. Rev. Lett. 113 115001
[11] Orain F. et al 2015 Phys. Rev. Lett. 114 035001
[12] Xu X.Q. et al 2016 Phys. Plasmas 23 055901
[13] Thyagaraja A. 2000 Plasma Phys. Control. Fusion 42 B255
[14] Thyagaraja A. 2005 Phys. Plasmas 12 090907
[15] Thyagaraja A., Valovic M. and Knight P.J. 2010 Phys. Plasmas
17 042507
[16] de Baar M.R. et al 2005 Phys. Rev. Lett. 94 035002
[17] Mantica P. et al 2006 Phys. Rev. Lett. 96 095002
[18] Garzotti L. et al 2006 Nucl. Fusion 46 73
[19] Valovic M. et al 2008 Nucl. Fusion 48 075006
[20] Chandra D. et al 2015 Nucl. Fusion 55 053016
[21] McCarthy D.R., Lashmore-Davies C.N. and Thyagaraja A.
2004 Phys. Rev. Lett. 93 65004
[22] Nishimura S. and Yagi M. 2011 Plasma Fusion Res.
6 2403119
[23] Becoulet M. et al 2012 Nucl. Fusion 52 054003