gas pu imaging studies of tokamak edge physics in the
TRANSCRIPT
Gas Puff Imaging Studies of Tokamak Edge Physics in the
National Spherical Torus Experiment
by
Yancey Sechrest
B.S., University of Arizona, 2007
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Department of Physics
2014
This thesis entitled:Gas Puff Imaging Studies of Tokamak Edge Physics in the National Spherical Torus Experiment
written by Yancey Sechresthas been approved for the Department of Physics
Prof. Tobin Munsat
Reader #2:
Reader #3:
Date
The final copy of this thesis has been examined by the signatories, and we find that both thecontent and the form meet acceptable presentation standards of scholarly work in the above
mentioned discipline.
iii
Sechrest, Yancey (Ph.D., Plasma Physics)
Gas Puff Imaging Studies of Tokamak Edge Physics in the National Spherical Torus Experiment
Thesis directed by Prof. Tobin Munsat
In order to be viable, Next-step fusion devices must overcome two pressing problems: they
must be able to achieve high levels of confinement while also handling potentially damaging heat
loads on material surfaces. The study of plasma edge physics promises solutions to both problems
because the plasma edge, being the boundary between confined and unconfined regions, plays a
key role in determining the global confinement and the plasma interaction with material surfaces
(e.g. edge transport barriers, pedestal evolution, and edge localized modes). However, the steep
gradients in density and temperature in the plasma edge that drive strong fluctuations in plasma
parameters require measurements of fluctuations with high spatial and temporal resolution. By
measuring drift scale (kyρs < 2) fluctuations for frequencies less than ∼ 200 kHz, Gas Puff Imaging
(GPI) meets these requirements while providing two-dimensional coverage at a large number of
measurement locations. This dissertation presents GPI studies of transitions from low to high
confinement regimes (L-H transitions) and Edge Localized Modes (ELMs). In 2010, a study of
L-H transitions with the GPI diagnostic revealed quasi-periodic reductions in the scrape-off-layer
turbulence levels during the 30 ms preceding the transition. The two-dimensional flow fields for
these “quiet-periods”, estimated from the GPI data by a pattern-matching velocimetry technique,
exhibit intriguing similarity with the Drift Wave - Zonal Flow paradigm, a leading candidate in
explaining L-H transitions. Following this study, a survey of GPI data from RF heated H-mode
plasmas near the L-H power threshold identified short-lived, coherent oscillations in edge emission
preceding the ELM crash. These observations provide detailed two-dimensional dynamics of the
growth, filamentation, and crash of the ELM event, which could improve our understanding through
comparison with nonlinear simulation. Cross diagnostic comparisons of GPI and Beam Emission
Spectroscopy measurements of edge fluctuations are also presented.
Dedication
To my lifeline, my family.
v
Acknowledgements
Two people deserve special recognition for their guidance and support. Chief among these
two is Tobin who was always patient and supportive. He encouraged me to explore and learn even
if it meant a slip in my productivity. Stewart Zweben also deserves special praise. His insight and
critical feedback have greatly influenced my critical thinking and writing, and I am very grateful
for his guidance.
During my work I have had the great opportunity to collaborate with many good scientists
who I admire. Among them are Ricky Maqueda, Devon Battaglia, and Dave Smith. Also, Jim
Myra, Dan D’ippolito, and Dave Russel of the Lodestar Research group. Our work would not have
been possible without the support of the whole NSTX team.
Grad school would have been nigh unbearable without the support of my close friends.
Whether watching movies at Steve and Carrie’s, playing Tetris Attack with Kevin, Climbing with
Travis and Andy, losing at Softball with Eric and the other Zeroes, or geeking out about all things
nerd with J.R., I can honestly say I wouldn’t have finished without all of you. Sami, Adam and
Carl also provided helpful discussion, or much needed stress-relief via lab basketball.
As noted in the dedication, my family has supplied me with a never-ending source of love
and support. Mom and Dad, you’ve always been willing to go to bat for me, and I’m incredibly
thankful for the opportunities you’ve provided for me. Tuck, thanks for always being there to chat,
whether about games or quarter-life crises. I’ll be ready to do this all again for your defense.
Finally, Tory. You’ve done so much to give me perspective, and keep me sane and grounded.
Thanks for putting up with me and my idiosyncracies.
Contents
Chapter
1 Introduction 1
1.1 Global Energy Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Nuclear Fusion and its Potential as an Energy Source . . . . . . . . . . . . . . . . . 2
1.3 Requirements for a Burning Plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Magnetic Confinement and Force Balanced Equilibria . . . . . . . . . . . . . . . . . 6
1.5 The Tokamak Reactor Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 The Spherical Torus Concept and NSTX . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Overview of Presented Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 The Gas Puff Imaging Diagnostic 12
2.1 The NSTX GPI System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Atomic Physics Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Direct Comparison of GPI and BES Measurements of Edge Fluctuations in NSTX 20
3.1 The Beam Emission Spectroscopy Diagnostic . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Plasma Conditions and NSTX Operation . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Relative Diagnostic Positioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Fluctuation Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5 Fluctuation Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.6 Cross-Diagnostic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
vii
3.7 Characteristic Time and Length Estimates . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Gas Puff Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.9 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Velocimetry 44
4.1 Time Delay Estimation for Motion Estimation . . . . . . . . . . . . . . . . . . . . . 44
4.2 The Optical Flow velocity estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Pattern Matching velocity estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 The Hybrid Optical Flow and Pattern Matching Velocimetry Algorithm (HOP-V) . 51
4.5 A Note On Post-Processing Techniques: a Navier-Stokes Inspired Smoothing Algorithm 52
5 Measurement of 2D flows in the edge and SOL preceding L-H transitions 56
5.1 The L-H transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 The Drift-Wave Zonal Flow Paradigm . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 The Predator-Prey model of the L-H transition . . . . . . . . . . . . . . . . . . . . . 67
5.4 GPI Observations of Flows preceding L-H transitions . . . . . . . . . . . . . . . . . . 69
5.5 Time averaged-flow profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.6 turbulence “Quiet-Periods” and quasi-periodic velocity fluctuations . . . . . . . . . . 72
5.7 Spatial Structure of 3 kHz mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.8 Flow shear and Reynolds stress calculations . . . . . . . . . . . . . . . . . . . . . . . 81
5.9 Quiet-periods as Limit Cycle Oscillations . . . . . . . . . . . . . . . . . . . . . . . . 86
6 Precursor Fluctuations During Small ELMs in NSTX 90
6.1 MHD stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 An Overview of ELMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3 NSTX observations of ELM precursors . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4 Operational Parameters and Plasma Conditions . . . . . . . . . . . . . . . . . . . . . 97
6.5 Precursor Oscillations in GPI Intensity . . . . . . . . . . . . . . . . . . . . . . . . . . 99
viii
6.6 Wavenumber and Frequency Characterization of the Precursor Mode . . . . . . . . . 103
6.7 Quantification of Edge Deformation During Precursor Evolution . . . . . . . . . . . 105
6.8 Magnetic Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.9 Pedestal Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.10 Concluding Remarks on the Nature of the Precursor Mode . . . . . . . . . . . . . . . 111
7 Directions for Future Work 112
Bibliography 114
Tables
Table
3.1 Shot list for this study containing the shot number, toroidal field at the magnetic
axis, plasma current, neutral beam heating power, average density, and GPI puff
timing. Values are taken at the time of the gas puff. . . . . . . . . . . . . . . . . . . 23
3.2 Comparison of poloidal correlation lengths, decorrelation times, and velocities esti-
mated from GPI and BES correlation functions. . . . . . . . . . . . . . . . . . . . . . 38
3.3 Poloidal correlation length estimates from 60-0 ms before, and 30-90 ms after the
gas puff trigger. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.1 Shot database for this study including the shot number, timeframe of interest,
toroidal magnetic field, plasma current, and RF heating power . . . . . . . . . . . . 98
Figures
Figure
1.1 Left: Diagram of NSTX design showing toroidal field (TF) coils, poloidal field (PF)
coils, center stack containing ohmic heating solenoid, and carbon wall tiles. Right:
Equilibrium reconstruction showing contours of poloidal flux. . . . . . . . . . . . . . 11
2.1 (a) Schematic of GPI system with view from machine center stack looking out.
(b) positioning of GPI view in R and Z with flux contours for typical operational
parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Diagram of GPI optical system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Cross-correlation map for each pixel with reference pixel at [17,40] for L-mode tur-
bulence. Green trace indicates field line trajectory through GPI gas cloud. . . . . . . 15
2.4 CR calculations of exponent for ne, α, and exponent for Te, β, used in estimation of
GPI Dα signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1 Comparison of Thomson scattering profiles of ne and Te with GPI emission profile
(blue curve) and R=140 cm BES radial array channels (black and red dashed lines).
The red dashed lines indicate the position of the inner and outer BES poloidal arrays. 23
xi
3.2 (a) R = 140 cm BES channel positions (red, diamonds) and GPI viewing area (blue
polygon) plotted over contours of poloidal flux for NSTX shot 141254. Flux surfaces
are labeled by their midplane r/a value, and the separatrix is indicated by the solid
black line. BES channel positions and the corners of the GPI view are plotted using
cylindrical coordinates. (b) BES inner poloidal array (red diamonds) plotted in
toroidal coordinates with GPI points (blue) for similar flux value. Magnetic field
line traces are plotted as dashed lines, and the dot-dashed line traces the generalized
poloidal direction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Traces of (a) GPI raw (gray) and 〈I〉 (blue), and (b) δI. Plots (c) and (d) are BES
traces. All traces taken at the 0.85 ΨN location. . . . . . . . . . . . . . . . . . . . . 26
3.4 Traces of δIRMS/〈I〉 for BES (red) and GPI (blue) for ΨN = 0.85. . . . . . . . . . . 27
3.5 PDFs of (a) GPI and (b) BES intensity from 10 ms period. Dashed Lines indicate
Gaussian PDFs with similar mean and variance. . . . . . . . . . . . . . . . . . . . . 28
3.6 Scatterplot of skewness and kurtosis values for BES (filled triangles) and GPI (filled
squares) at r − rsep ≈ 2.9 cm, and BES (open triangles) and GPI (open squares) at
r − rsep ≈ 0.7 cm. The dashed line represents a second degree polynomial fit to the
GPI data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.7 Contour plot of time-lagged cross-correlation between GPI and BES signals versus
time. Peak correlation value persists in time, but constant linear drift is present. . . 32
3.8 Traces of (a) time-lagged cross-correlation between GPI and BES, and (b) time traces
of GPI and BES intensity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.9 Plots (a) and (b) are traces of cross-correlation vs. radius for GPI pixels with a BES
reference channel. Plots (c) and (d) are contour plots of the cross-correlation vs.
radius and time-lag. The black dashed line indicates the separatrix location, and the
dashed red line indicates the radial location of the BES reference channel. . . . . . . 33
3.10 Traces of (a) cross-coherence, (b) cross-spectral density, (c) cross-phase, and (d)
phase uncertainty for a BES coord and the closes GPI pixel in the R-Z plane. . . . . 34
xii
3.11 Plots of: the Cross-Coherence between poloidaly separated channels for (a) GPI and
(c) BES, and Autopower spectra for GPI (b) and BES (d). . . . . . . . . . . . . . . 36
3.12 plots of: (a) and (e) Time-lagged cross-correlations (solid) with envelope functions
(dashed), (b) and (f) zero-lag envelope peak correlation versus poloidal separation,
(c) and (g) envelope peak correlation versus time-lag to peak, and (d) and (h) time-
lag to envelope peak correlation versus poloidal separation. Values for poloidal cor-
relation length, and decorrelation time represent the 1/e length for a Gaussian fit to
the corresponding plot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.13 Scatterplots comparing (a) poloidal correlation lengths estimates, (b) decorrelation
time estimates, and (c) TDE velocity estimates. . . . . . . . . . . . . . . . . . . . . 37
3.14 Time traces of low-pass filtered BES intensity for varying (a) Xsep = r − rsep and
(b) Z. Times are relative to the GPI gas puff timing. . . . . . . . . . . . . . . . . . . 39
3.15 Continous Wavelet Transforms of BES fluctuations normalized to 100 Hz low-pass
filter for shots 138845 (a) and 141249 (b). Time traces of the average GPI intensity
is plotted above the CWTs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.16 (a) and (c): Block-averaged autopower spectra of fluctuations normalized to mean.
(b) and (d): coherence for BES channels separted by 4.8 cm. Black traces are
spectra for 60-0 ms before the gas puff trigger, and red traces are spectra for 30-90
ms following the trigger. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Pattern Matching vector fields for (a) unsmoothed and (b) smoothed. Profiles of vx
and vy for unsmoothed, (c) and (d), and smoothed, (e) and (f). Red traces are the
imposed velocity field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1 Traces of (a) plasma current and injected neutral beam power, (b) diverter Dα light,
(c) Energy confinement times, and (d) plasma stored energy. Black traces are for
an ohmic L-mode shot, and red traces are for a neutral beam driven H-mode. L-H
transition timing indicated by dashed red line. . . . . . . . . . . . . . . . . . . . . . 58
xiii
5.2 Thomson scattering profiles of electron temperature (top) and electron density (bot-
tom). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.3 Traces of (a) fraction of GPI light in SOL, (b) GPI raw signal near separatrix, and
(c) normalized GPI fluctuation level. . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Conceptual picture of the drift wave zonal flow paradigm. Gradients drive insta-
bilities to turbulent state. Turbulence drive anomalous transport via fluctuations.
Turbulence also self-generates zonal flow via Reynolds stresses. Zonal flows regulate
turbulence via shear suppression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.5 Illustration of the drift wave mechanism. The background density gradient is in the
−x direction, and the magnetic field is in z. E × B circulation around potential
perturbation (blue contours) pushes high density from −x and low density from +x
producing a propagation of the perturbation in +y. . . . . . . . . . . . . . . . . . . 63
5.6 Conceptual picture of limit cycle oscillation process. System evolves between a high
turbulence, low flow state and a high flow, low turbulence state. If the input power
is sufficient to steepen the density gradient during a period of suppressed turbulence,
then a mean flow shear develops and a transition to H-mode is observed. . . . . . . . 69
5.7 (a) Time averaged velocity field superimposed on time averaged GPI intensity con-
tours (gray contours). The separatrix is indicated by the dot-dashed line, and the
cropped field of view is indicated by the dashed box. The maximum time-averaged
velocity magnitude for the cropped region is 1.5 km/s, while the maximum instan-
taneous velocity magnitude is 7.4 km/s. The +x direction is radially outward. (b)
radial profiles of time averaged poloidal flow several shots for 10.5 µs preceding the
L-H transition. RMS values of the fluctuating velocity are shown for shot 135042. . . 71
5.8 Normalized GPI intensity (top), poloidal velocity (middle), and radial velocity (bot-
tom) ∼1 cm inside the separatrix. All traces represent poloidally averaged quantities
from shot 135042. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
xiv
5.9 Traces from shot 135042 of Fsol filtered (a), and poloidally averaged poloidal velocity
inside (b), at (c), and outside (d) the separatrix. Fsol filtered has had the low
frequency components above zero frequency and below 1 kHz removed. Gray bars
indicate time periods where Fsol filtered is below 0.16. . . . . . . . . . . . . . . . . 74
5.10 Spectrogram of Fsol for the L-mode portion (t ≈ 0.215 − 0.245) of shot 135042
plotted with a linear color scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.11 Map of phase differences of ∼2.4 kHz fluctuations for: (a) GPI signal with reference
GPI signal, (b) vpol with reference vpol, (c) GPI signal with vpol. Reference signals
are at x ≈ 0cm, y ≈ 11 cm. Plots (a) and (c) have had the phase discontinuity
remapped so that the contour plots appear smooth. Plots cover the time range
t ≈ 0.241− 0.243 ms of shot 135042. The dashed-line indicates the separatrix. . . . . 77
5.12 Phase of ∼3 kHz mode plotted vs spatial cordinates for GPI signal and poloidal
velocity for periods of shots 135042-135045 with “rotating mode” visible in band-
pass filtered. Time periods are 512 frames (1.8 msec) and begin at: 135042(black)
t=0.241, 135043(gold) t=0.238, 135044(green) t=0.230, 135045(red) t=0.236 s. . . . 78
5.13 Coherence and phase plots versus frequency for poloidal velocity signals separated
by ∼12 cm poloidally at ∼1 cm inside the separatrix. . . . . . . . . . . . . . . . . . 80
5.14 Plot shows coherence of poloidal velocity vs. y separation for 2.8 kHz mode (�)
and background turbulence between 20-30 kHz (4) for poloidal velocity of shot
135042. Velocities are measured at ∼1 cm inside the separatrix. Solid lines are fits
to exponential decays with correlation lengths of 56 cm for the 2.8 kHz mode and 4
cm for the background turbulence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.15 Autopower spectra of poloidal velocity at 3 radial positions for t ≈ 0.240 − 0.243 s
of shot 135042. The feature at 2.5 kHz does not appear to shift in frequency. . . . . 81
xv
5.16 Traces from shot 135042 of Fsol, poloidal velocity (km/s), shearing rate (MHz), and
the Reynolds stress (km2/s2). Poloidal velocity and Reynolds stress are taken at ∼1
cm inside the separatrix, and the shearing rate is taken at the separatrix. Traces
have been smoothed with a 3 point boxcar average. . . . . . . . . . . . . . . . . . . . 83
5.17 Poloidal velocity profiles for ∼30 ms of shot 135042 are plotted in (a). The averages
are over times where Fsol is greater than its mean value (dashed - bursty) and less
than its mean value (solid - quiet). “Error bars” indicate ±1 σ(standard deviation)
about the average value for the low Fsol case, and indicate the level of fluctuation.
Standard deviations are shown in (b), and skewness is shown in (c). . . . . . . . . . 84
5.18 Traces of Fsol, poloidal velocity (km/s), shearing rate (ξ), and the Reynolds shear
stress. Traces are from a SOLT turbulence simulation illustrating the bursty regime.
The frequency of bursts is 3.6 kHz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.19 Radial profiles of the Reynolds shear stress averaged over the poloidal direction and
10 ms directly preceding the L-H transition for 4 shots. Error bars indicate estimated
level of uncertainty assuming uncertainties in the velocity measurments of 0.5 km/s. 86
5.20 Radial profiles of the average poloidal velocity, and the amplitude of the 3 kHz flow
feature during the quiet period oscillations. . . . . . . . . . . . . . . . . . . . . . . . 88
6.1 Time traces of RF heating power PRF , plasma current IP , Dα light, and line-
integrated density N for typical shot from this study. Neutral beams are also used
early in the shot for plasma conditioning. The shaded region indicates the time
period for traces plotted in Figure 6.3. . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2 Multiframe image stills of an ELM event with precursor intensity fluctuations from
shot 141918. The time between frames is ∼7.5 µs. Distinct mode structure can be
seen in precursor oscillations leading to the ejection of the filament in the last two
frames. The approximate location of the separatrix is indicated by the dashed line. . 100
xvi
6.3 Time traces of (a) scrape-off layer fraction FSOL and (b) integrated edge intensity
Iedge corresponding to shaded timeperiod in Fig. 6.1. Traces show edge intensity
fluctuations preceding an ELM at 0.2425s, and an ELM-induced back transition at
0.245s. Low level fluctuations can also be seen near 0.244s. . . . . . . . . . . . . . . 101
6.4 Two-dimensional slices through GPI data with one cut at y=15.5 cm (a), and one
at x=10.5 cm (b). Precursor fluctuations are very distinct brightness pulses which
appear to move upward, or in the electron diamagnetic direction indicated by the tilt
of structures in (b). Structures are also seen to drift radially outward as indicated
by +x motion in (a). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.5 Traces of (a) Average perpendicular velocities, and (b) FSOL for many ELM events
from the shots database. Velocities are measured ∼2 cm inside the separatrix. Tim-
ings are relative to the peak FSOL for each event. . . . . . . . . . . . . . . . . . . . . 102
6.6 Wavelet scalogram of integrated edge intensity, Iedge accompanied by the time trace
of the scrape-off layer fraction for shot 141919. The power spectrum shows significant
power at the 20 kHz scale during the ELM precursor fluctuations. The shaded region
in the wavelet scalogram indicates where edge effects become important, and the
white contour indicates the 95% significance level. . . . . . . . . . . . . . . . . . . . 104
6.7 Scatter plot of frequency of precursor fluctuations against the perpendicular wavenum-
ber derived from GPI intensity fluctuations. Red triangles are events that lead to
an ELM or back-transition, while blue squares are edge intensity fluctuations that
do not lead to an ELM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.8 Example (a) image frame from shot 141917 and (b) xedge function overlayed on
intensity contours. Maximum radial excursion relative to EFIT separatrix location
is plotted in (c), and the edge curvature, κ corresponding to this point is plotted in
(d). The Dashed line indicates the time point of image (a). . . . . . . . . . . . . . . 105
xvii
6.9 Scatterplot of minimum edge curvature, κmin and maximum radial excursion rela-
tive to the EFIT separatrix location for several ELM precursor events (diamond).
Intensity fluctuations not leading to an ELM are also included (square). . . . . . . . 107
6.10 Time traces of (a) Iedge and (b) low-pass filtered magnetic signals from shot 141917.
The black trace has been lowpass filtered at 200 kHz, and the red trace has been
bandpass filtered around 20 kHz. Magnetics traces are strongly correlated with
fluctuations in edge intensity as shown in the lagged correlation plot (c). Absolute
values of the correlation coefficient reach 0.8 for periods during the precursor activity.109
6.11 Pedestal parameters for height hped and width ∆ped are extracted from a modified
tanh fit to the electron pressure profile (inset) and a comparison of several shots is
presented. Squares are 0.6-1.0 MW RF shots, triangles are Ohmic H-mode shots,
stars are ∼ 1.0 MW NBI shots, and the diamond is a 4 MW NBI heated shot. . . . . 110
Chapter 1
Introduction
1.1 Global Energy Challenges
As populations increase and industrialization spreads, the energy needs of the world will
continue to rise. Providing for the world’s energy needs over the next century poses a number
of serious challenges. Since the industrial revolution, much of the world’s energy production has
come from the burning of fossil fuels (e.g. oil, coal, and natural gas) [21], and the burning of these
fuels has led to the release of massive quantities of greenhouse gases (GHG). The increase in GHG
emission has been correlated with the rise of global mean temperatures, and the current scientific
consensus is that anthropogenic GHG is driving significant changes in the global climate [66]. These
changes in climate have wide ranging effects including impacts on water supply and crop yield, and
harsher impacts of natural dissaters on natural and human systems [67].
In addition to the environmental impacts, it must be recognized that fossil fuels are a finite
resource, and many concerns surround the remaining reservoirs of these resources, their accessibility,
and their ability to meet future energy needs [41]. While predictions of the depletion time of fossil
fuel reservoirs are uncertain, one prediction is that oil and gas resources will be depleted in 40
years, and coal resources will be depleted on the order of 100 years [125]. As fossil fuels become
more scarce and difficult to extract, other energy generation methods will need to be tapped. These
include nuclear fission, biofuels, and renewable sources such as hydroelectric, wind and solar [21].
Scientific innovation will be required to meet the energy needs of the future in a sustainable way
with minimal environmental impact.
2
1.2 Nuclear Fusion and its Potential as an Energy Source
Fusion is a nuclear reaction in which two atomic nuclei collide at high energy and fuse to
create a single, heavier nucleus in addition to other energetic products. The cross-section for a
fusion reaction depends on the energy, or temperature, of the reactants. The fusion reactions
with cross-sections that peak at the lowest energies, and consequently are easiest to achieve in lab
experiments, are:
D + T = 4He(3.52MeV ) + n(14.06MeV ), (1.1)
D +D = 3He(0.82MeV ) + n(2.45MeV ), (1.2)
D +D = T (1.01MeV ) + p(3.03MeV ), (1.3)
D +3 He = 4He(3.67MeV ) + p(14.67MeV ). (1.4)
The cross-section for Deuterium-Tritium (D − T ) fusion is much greater than the other processes
at experimentally viable energies in the range 10-100 keV, and peaks around 100 keV or 1.16
trillion Kelvin. For this reason, first generation fusion reactors are expected to operate using D−T
fusion, and so it is currently the most important reaction. Tritium is radioactive and a controlled
substance, however, so research experiments often use D −D fuel (i.e. pure deuterium).
Fusion energy production promises a number of benefits over other energy solutions. For
comparison, a fission reaction involving U −235 yields 0.86 MeV per nucleon, and a D−T reaction
produces 3.52 MeV per nucleon. In terms of specific energies, Coal has a specific energy of ∼30
MJ/kg, fission with U−235 has a specific energy of 8.3×107 MJ/kg, and D−T fusion has a specific
energy of 3.4× 108 MJ/kg. In addition to its high specific energy, the deuterium in D− T fuel is a
light element and naturally abundant. Tritium is radioactive, and more difficult to obtain though
options do exist. The most appealing option is for the fusion reactor to breed the required tritium
fuel in a lithium blanket surrounding the device as lithium is relatively abundant. Other sources of
tritium include heavy water fission reactors. Compared to other fuels used for energy production,
D − T fuel has a high energy density, and is relatively abundant source of energy, assuming the
3
issues surrounding tritium supply can be resolved.
While a fusion reactor will become radioactive due to neutron bombardment of construction
materials, the disposal of the low level radioactive waste from a fusion reactor is expected to be
simpler and significantly cheaper than disposal of high level radioactive waste from a fission reactor
[124]. The bulk of the radioactive waste from the fusion reactor can be disposed of by shallow
land burial, and radioactivity of all materials decays to below safety limits on the time scale of 100
years or less. Still, the cost of waste disposal must be included in assessments of the economics of
a possible fusion reactor.
The idealized picture of fusion energy is incredibly appealing, especially in light of the envi-
ronmental and resource related challenges currently on the global energy horizon. Fusion promises
energy free of GHG emissions from a relatively abundant, high energy density fuel source. The only
drawback being the production of easily managed, short-lived radioactive waste. For this reason,
the study of fusion has continued to captivate scientists for the better part of a century, though at
times a working fusion reactor seems to resemble Ahab’s white whale.
1.3 Requirements for a Burning Plasma
To fuse, two particles must collide with sufficient energy to overcome the Coulomb repulsion
barrier. As discussed in the previous section, this requires a great deal of energy. To drive an
appreciable number of fusion reactions, the population of plasma ions must be heated to very high
temperatures (T ≥ 10 keV). Intuitively, we can expect that the yield of fusion energy will depend
on the density of plasma ions and the plasma temperature. The actual rate of fusion reactions is
given by [158]
R =
∫ ∫σ(v′)v′f1(v2)f2(v2)d
3v1d3v2, (1.5)
where v′ is the relative velocity, σ(v) is the cross-section for the fusion reaction, f is the distribu-
tion function, and subscript 1, 2 refers to ion species. Using Maxwellian distributions for fi and
4
performing some simplifications this becomes
R = 4πn1n2( µ
2πT
)3/2 ∫σ(v′)v′3 exp
(− µv′2
2T
)dv′. (1.6)
This has the formR = n1n2〈σv〉, so the thermonuclear power per unit volume can then be expressed
as
pTn = n1n2〈σv〉E (1.7)
where E is the energy released per reaction. For a given n = n1 + n2, the power is maximized for
n1 = n2 = 12n, so the power can be written as
pTn =1
4n2〈σv〉E . (1.8)
In D − T fusion, the resulting neutron carries 4/5 of the released energy, and the helium
nucleus, or α particle, carries the remaining 1/5. The neutrons are neutral particles, and are not
affected by electric and magnetic forces. Therefore they easily escape the plasma, and, ideally,
deposit their energy into a coolant used for heat exchange and energy production. The alpha
particle, though, is a charged product, and so will be confined with the plasma where it transfers
its energy to the plasma ions through collisions. Thus the alpha particle will lead to plasma heating
per unit volume given by
pα =1
4n2〈σv〉Eα, (1.9)
and the total α particle heating is given by
Pα =
∫pαd
3x (1.10)
In steady-state operation, the power loss PL is balanced by external heating PH and Pα
giving
PL = PH + Pα. (1.11)
The power loss is defined as
PL = −dWdt
=W
τE(1.12)
5
where
W =
∫3nTd3x (1.13)
is the plasma stored energy and τE is the energy confinement time. If the α heating is large enough
to completely offset the power loss PL then the external heating systems can be turned off, and
the fusion reaction will be self-sustaining. This is called ignition, and the plasma is said to be a
burning plasma. A rough estimate of the requirements for ignition may be obtained from the above
definitions by assuming a constant density and temperature. The result is
nτE >12
〈σv〉T
Eα. (1.14)
In the temperature range 10-20 keV this can be approximated as
nTτE > 3× 1021m−3keV s. (1.15)
The triple product nTτE is revealed as an important metric of plasma performance. This makes
sense intuitively because this relation states that a burning plasma requires a sufficient energy
density nT be maintained for an adequate time τE so that many fusion reactions can occur.
Another important quantity is the plasma Q factor, the ratio of thermonuclear power pro-
duced to the externally supplied heating power
Q =14n
2〈σv〉EVPH
. (1.16)
Important values of Q are:
• Q = 1: heating power is equal to thermonuclear power generated, break even point ignoring
electrical efficiencies.
• Q = 5: Alpha particle heating of D − T fusion equals supplied heating power.
• Q ∼ 5: D−T Break even assuming electricity generation efficiency of 33%, heating efficiency
of 75%.
• Q ∼ 20: Reactor relevant Q values. Assumes above efficiencies and 25% power recirculation.
6
• Q→∞: Burning Plasma. No External heating required.
The Japanese tokamak JT-60 currently holds the world record for achieving Q=1.25, but this
number is extrapolated to D − T fusion yield based on an experimental result using D − D fuel
[48]. The International Thermonuclear Experimental Reactor (ITER) currently under construction
in Cadarache, France is expected to achieve Q > 5 [104].
1.4 Magnetic Confinement and Force Balanced Equilibria
As we saw in the last section, a viable fusion reactor must be able to confine a dense plasma
long enough to heat the plasma to sufficient temperatures for fusion reaction to take place. From
basic electromagnetism, we know that charged particles moving in a magnetic field experience a
force F = qv×B. The direction of motion and the resulting force are always perpendicular, so the
particles will undergo circular motion in a plane perpendicular to B described by the equation
ma⊥ = mv2⊥rL
= qv⊥B. (1.17)
The radius of the orbit is then rL = mv⊥/(qB). The particles motion in the parallel direction is
unaffected. The conclusion from this is that charged particles in a magnetic field will be confined
to the magnetic field lines. This simple concept is the basis for the field of magnetic confinement
fusion.
The simplest description of a plasma is given by the magnetohydrodynamic (MHD) model
that treats the plasma as a single conducting fluid. The plasma is assumed to be neutral, so the net
charge is zero. Furthermore, fluid motions are taken to be slow compared to characteristic times
scales of the plasma so that the displacement current is neglected. The MHD set of equations are:
∂ρm∂t
+∇ · (ρmU) = 0 (1.18)
ρm∂U
∂t+ ρmU · ∇U = J×B−∇P (1.19)
J = σ(E + U×B) (1.20)
d
dt(Pρ−γm ) = 0. (1.21)
7
When combined with Maxwell’s equations this yields a complete set of equations describing the
plasma.
The condition for static plasma equilibrium is given by Eqn. 1.19. For an equilibrium with
zero flow this gives the equation for a force-balanced equilibrium
J×B = ∇P. (1.22)
From this it can be seen that both
B · ∇P = 0 (1.23)
and
J · ∇P = 0, (1.24)
so the current and magnetic field lie in surfaces of constant pressure. Hopf’s Theorem from topology
tells us that the geometric form of the closed constant pressure surface that satisfies the above con-
straints must be a torus. Thus plasma equilibria and confinement devices with closed surfaces are
inherently toroidal objects; from self-organized systems like the spheromak and field-reversed con-
figuration, to externally imposed equilibria such as the tokamak, and reversed field pinch. Toroidal
equilibria may be described using cylindrical coordinates with the center axis of the torus chosen
as the z-axis, and φ then corresponds to the long dimension around the torus or toroidal direction.
The short dimension around the torus is called the poloidal dimension. The radius from the z-axis
to the torus center line is the major radius, and the radius from the torus centerline to the plasma
boundary is the minor radius.
For axisymmetric equilibria derivatives with respect to φ are zero, so B = ∇×A gives
Br = −∂Aφ∂z
(1.25)
Bz =1
r
∂
∂r(rAφ). (1.26)
Thus Br and Bz can be expressed in terms of a single function ψ = rAφ. The function ψ is related
to the poloidal magnetic flux in the torus. The requirement that ∇ ·B then gives
Br∂ψ
∂r+Bz
∂ψ
∂z= B · ∇ψ = 0, (1.27)
8
and B is found to lie in surfaces of constant ψ. It follows then that the plasma pressure may be
expressed as a function of ψ. Similarly, the φ component of the force balance equation gives
B · ∇F = 0, (1.28)
where the function F = rBφ is related to the poloidal current flux.
The above relations can be used to derive a single equation that, upon solution, gives the form
of the axisymmetric magnetostatic equilibrium. This equation is derived from the radial component
of the force balance equation
JφBz − JzBφ =∂P
∂r. (1.29)
Using Ampere’s law and the definitions of ψ and F this can be expressed as
r∂
∂r
1
r
∂ψ
∂r+∂2ψ
∂z2= −µ0r2
d
dψP (ψ)− 1
2
d
dψF (ψ)2. (1.30)
This is the Grad-Shafranov equation, and its solution gives the equilibrium flux function ψ in
terms of the two free functions P (ψ) and F (ψ). Solutions are usually computed numerically by
first specifying the two free functions. An important class of solutions are those with constant ψ
surfaces forming nested toroidal surfaces as these solutions describe a plasma equilibrium that may
be magnetically confined and isolated from material surfaces. Example equilibrium flux surfaces
for the low aspect ratio spherical torus NSTX is shown in Fig. 1.1.
1.5 The Tokamak Reactor Concept
In the previous section we outlined some of the requirements for an axisymmetric plasma
equilibrium. Now we can put these pieces together to conceptualize a device that could in theory
confine a plasma in stable equilibrium. First, remember that Hopf’s theorem tells us that the
equilibrium should be toroidal, so we will begin by imposing a toroidal field using a set of external
magnet coils. For magnetic coils with a straight leg on the inside of the torus, the toroidal field
can be obtained from the integral form of Ampere’s law, and is given simply by Bφ = µ0NI/2πr
where N is the number of coils and I is the current per coil. Conveniently, this simplifies the
9
Grad-Shafranov equation since
F = rBφ = constant. (1.31)
Using the toroidal component of Ampere’s law and the Grad-Shafranov equation with ∂F/∂ψ = 0,
we see that a toroidal current
Jφ = rdP
dψ(1.32)
is required to provide equilibrium. One solution is to drive a toroidal current in the plasma in-
ductively by driving a time dependent magnetic flux with a solenoid located on the z-axis. This
confinement scheme, characterized by a strong toroidal field imposed by external magnet coils and
a toroidal current carried by the plasma, is the Tokamak. In practice, several sets of poloidal field
coils are employed in addition to the toroidal field coils for equilibrium shaping, and a vertical field
coil is used to counterbalance the hoop force.
The tokamak equilibrium can be understood conceptually by examining the single particle
drifts. For a toroidal device, the imposed toroidal field generally falls off as Bφ ∝ 1/r and the
radius of curvature RC points in −r. The gradient drift,
vg =w⊥qB2
b×∇B, (1.33)
and the curvature drift,
vC =2w‖
qBRCb× RC (1.34)
are both in z. Here w⊥ and w‖ are the perpendicular and parallel kinetic energy. Driving a toroidal
current in the plasma produces a poloidal field, so the magnetic field lines become helical and lie
in nested toroidal surfaces. As the particle travels around the tokamak it will spend some of the
time drifting inward, toward the center line of the torus, and some of the time drifting outward,
away from the center line. The net result is that particles are confined about the starting surface.
1.6 The Spherical Torus Concept and NSTX
The low aspect ratio (ratio of major radius to minor radius), spherical torus concept was orig-
inally explored by Peng and Strickler [113]. The compact design offers a number of compelling
10
advantages including reduced construction costs and equilibria characterized by high natural elon-
gation, large plasma current, and strong paramagnetism [113]. The low aspect ratio design requires
the removal of neutron shielding protecting the toroidal field coils at the center stack, however, and
as a consequence normal conducting copper magnets become necessary. Additionally, engineering
considerations lead to a reduced toroidal field relative to the conventional tokamak design. To be
feasible then, the spherical torus must operate at high β ≡ 2µ0〈p〉/B2v where Bv is the vacuum
toroidal magnetic field at the plasma geometric center and 〈p〉 is the volume averaged pressure.
The plasma β is loosely related to the ratio of thermonuclear power output to energy expended
establishing the magnetic field [158]. Fortunately, equilibrium and stability studies have indicated
that the low aspect ratio design yields improved confinement, and optimized, high-β equilibria are
found with low aspect ratio, high plasma-generated bootstrap current fraction, and high elongation
[100, 73].
The National Spherical Torus Experiment (NSTX) [109, 76] is a low aspect ratio R/a = 1.27
tokamak experiment. The major radius of the device is R = 0.85m and the minor radius is a =
0.67m. The machine can operate at toroidal magnetic fields up BT = 0.55 T with plasma currents
of Ip = 1.4 MA. An engineering schematic of the device and a typical equilibrium reconstruction
is included in Fig. 1.1. Up to 7 MW of neutral beam injection (NBI) heating power and 6 MW
of high-harmonic fast wave (HHFW) heating at 30 MHz is available for auxiliary heating. Central
temperatures of ∼ 1 keV, average densities of ∼ 6.0e19 m−3, and pulse lengths of up to ∼ 1 s
have been achieved. Most critically, experiments with active suppression of resistive wall mode
instabilities have demonstrated high-β operation achieving β = 39% [118].
1.7 Overview of Presented Work
This dissertation details studies of tokamak edge physics in NSTX using the Gas Puff Imaging
diagnostic. The GPI diagnostic is described in Chapter 2, and Chapter 3 presents the characteriza-
tion of edge turbulence and fluctuations with GPI and their comparison with Beam Emission Spec-
troscopy measurements. Chapter 4 details several imaging-based velocity extraction techniques,
11
Figure 1.1: Left: Diagram of NSTX design showing toroidal field (TF) coils, poloidal field (PF)coils, center stack containing ohmic heating solenoid, and carbon wall tiles. Right: Equilibriumreconstruction showing contours of poloidal flux.
and Chapter 5 examines turbulent flows in the edge preceding the transition to high-confinement
operation using these techniques. Finally, Chapter 6 presents GPI observations of Edge Localized
Mode dynamics including the characterization of a coherent precursor in GPI intensity.
Chapter 2
The Gas Puff Imaging Diagnostic
Gas Puff Imaging (GPI) is a diagnostic capable of imaging edge turbulence in fusion devices.
During operation of the diagnostic, a neutral gas puff is discharged into the edge of the plasma and
the visible light emission from collisional excitation of the neutral gas with the plasma is imaged
on timescales faster than the turbulence autocorrelation time (typically . 10µs). This allows for
the 2D field of turbulent fluctuations in the edge of the device to be imaged, and the motion of
the turbulent structures to be tracked. GPI has been used on a number of fusion experiments
including NSTX, Alcator C-mod, TEXTOR, RFX-mod and EAST [97, 169, 127, 1, 90]. GPI has
been used to study edge turbulence and blobs [2, 27, 97, 169, 87, 106], turbulent velocity fields
[147, 105, 167, 123, 26], ELMs [96, 122], and L-H transitions [167, 161]. This chapter details the
design and operation of the NSTX GPI diagnostic and the atomic physics basis for GPI.
2.1 The NSTX GPI System
The Gas Puff Imaging system on NSTX [97, 170] uses a fast-framing camera to capture
images of line-filtered emission from the collisional excitation of a neutral gas by the background
plasma. A schematic of the diagnostic can be seen in Fig. 2.1(a), and the position of the view in
machine R and Z is shown in Fig. 2.1(b). The gas puffing is delivered by a manifold mounted on the
outer wall behind the RF limiter shadow 20 cm (∼28 ◦ poloidally) above the outboard midplane.
Gas is injected through thirty 1 mm diameter holes spaced evenly along the 29 cm length of the gas
manifold. The gas injection system is tilted with respect to the horizontal to produce a gas cloud
13
that is elongated along the direction of the local magnetic field at the edge (∼35-50◦ pitch angle).
Deuterium, Helium, and Argon have been used for the neutral gas species. Deuterium emission
competes with background thermal emission of the plasma, so Helium and Argon typically have
higher signal to background ratios. During an observation period, gas flow rates of 6×1021 atoms/s
are maintained for ∼ 50 ms.
Figure 2.1: (a) Schematic of GPI system with view from machine center stack looking out. (b)positioning of GPI view in R and Z with flux contours for typical operational parameters.
A diagram of the optical system can be seen in Fig. 2.2. The gas cloud is viewed through a
re-entrant window approximately 70 cm away. Light from the gas cloud is imaged onto a 400 pixel
× 400 pixel coherent fiber optic bundle by a 1 in. F/1.3 lens with a 25mm focal length. The image
from the fiber bundle is then imaged onto a fast-framing camera using a 50mm focal length, 1 in.
F/1.3 lens on the fiber bundle and a 8 mm, 0.5 in. F/1 lens on the camera. An interference filter
is used to select spectral line emission for the chosen neutral species. Typically Deuterium is used
for the gas species, and the Balmer-α (Dα : n = 3→ 2) line at 656 nm is imaged.
Many cameras have been used over the life of the GPI diagnostic system, but the current
iteration, which recorded most of the data discussed in this thesis, uses a Vision Research Phan-
tom v710 fast-framing camera. The camera uses a CMOS sensor capable of capturing images at
14
Figure 2.2: Diagram of GPI optical system
1280×800 resolution with framerate that depends on pixel count. For the GPI system, Images are
captured at a framerate 400 kHz with a 64×90 resolution and 2.2 µs exposure time. The sensor has
a readout noise of 29 electrons per pixel at 30.2◦C, and a 23200 electron full well capacity. Typical
turbulence autocorrelation times in the edge of NSTX are on the order of ∼10 µs, so turbulence
time scales are well resolved by the time per frame of 2.5 µs of the camera. For data discussed in
Chp. 5, images from two Phantom v7.3 cameras were interleaved to achieve a 285 kHz framerate
at 64×64 resolution and 3 µs exposure times.
Empirical studies have demonstrated that edge turbulence correlations are significantly longer
along the magnetic field than across[97, 170]. Thus turbulent structures are filamentary in nature
with 3-5 cm cross-sections in the cross-field directions and several 10’s of centimeters along the
field. To maximize the cross-field resolution, the viewing direction is tilted by 40◦ with respect to
horizontal so that the view is approximately aligned with the local direction of the magnetic field at
the intersection with the gas puff during standard operation. The x and y coordinates of the camera
image are approximately perpendicular and tangent to surfaces of constant flux, respectively. Thus,
the image x coordinate is approximately the radial coordinate, and the y coordinate is then the
generalized poloidal coordinate.
The pixel resolution of the optical system at the location of the gas puff is ∼4 mm. The gas
cloud does extend ∼24 cm along the line of sight [97], however, so the resolution is degraded by
15
the coupling of the elongated structures with the magnetic field line curvature and misalignment
of the optical system with the magnetic field. To illustrate the smearing of pixels, the cross-
correlation map of GPI data for typical L-mode turbulence is shown in Fig. 2.3 with the projection
of a magnetic field line trajectory through the gas cloud. Field line curvature reduces the radial
resolution to ∼ 3 cm. Misalignment of the line of sight with the magnetic field pitch angle can
reduce the poloidal resolution by up to ∼2 cm, but here the poloidal resolution is ∼ 0.5 cm. For
typical NSTX edge parameters, the GPI light emission is localized to the region with Te, Ti . 0.3
keV, and ion sound gyroradius ρs . 1 cm. Thus, fluctuations that satisfy 〈kpol〉ρs . 2.0, including
typical drift turbulence scales [151] of 〈kpol〉ρs ≈ 0.1− 1, are well resolved by the GPI diagnostic.
Figure 2.3: Cross-correlation map for each pixel with reference pixel at [17,40] for L-mode turbu-lence. Green trace indicates field line trajectory through GPI gas cloud.
2.2 Atomic Physics Basis
The system under consideration is a neutral deuterium gas immersed in a plasma character-
ized by electron density ne, ion density ni, and electron temperature Te. Our interest is to describe
the dependence of the neutral Dα emission on the plasma parameters. The present description will
16
loosely follow the presentation in the DEGAS neutral transport code user manual [139]. The Dα
emission rate will be given simply by S = N3A23, where N3 is the population of the n = 3 excited
state of deuterium and A23 is the Einstein coefficient for spontaneous decay from level n = 3 to
n = 2. Neutral atoms in a plasma undergo a wide array of collisional and radiative transitions that
distribute its population over a great many excited states, so the above equation for the deuterium
emission is deceptively simple. The time evolution of a given atomic state under the influence of
these various interactions is given by the Collisional Radiative (CR) model:
dNm
dt= −{
∑n<m
Kd,nmne +∑n>m
Ke,nmne +Ki,mne +∑n<m
Anm}Nm
+∑n>m
Kd,mnNnne +∑n<m
Ke,mnNnne +∑n>m
AmnNn
+{βrad,m + βdia,m +Kr,mne}neni. (2.1)
The interactions included in the model and the meaning of each term is as follows:
• Collisional dexcitation from level m to n by electron impact at a rate Kd,nm
• Collisional excitation from level m to n by electron impact at a rate Ke,nm
• Ionization from excited state m by electron impact Ki,m
• Spontaneous radiative decay from state m to n at a rate Anm
• Radiative recombination to state m at a rate βrad,m
• Diaelectronic recombination to state m at a rate βdia,m
• Three-body recombination to state m at a rate Kr,m
In words, the model accounts for the collisional and radiative transitions out of state m given by the
negative terms proportional to Nm, the collisional and radiative transitions into state m from all
other states given by the terms proportional to Nn, and the recombination of plasma ions into state
m given by the final 3 terms. For the temperatures and densities relevant to the Tokamak edge, the
electron collision processes dominate, so all collisional terms in the model are mediated by electrons.
17
This model was originally developed for hydrogen by Bates, Kingston and McWhirter [6], and
many calculations have been perform since [5, 99, 34, 35, 46, 74, 15]. Alternatively, this system of
equations can be written compactly in matrix form as
n = Mn + Γ, (2.2)
where the transitions due to collisions, radiation, and ionization are combined into the matrix M,
and the recombination terms have been gathered into the vector Γ.
Equation 2.1 defines a large (technically infinite) system of coupled linear differential equa-
tions. A number of assumptions are leveraged to make the problem tractable. First, it’s typically
assumed that the population for states above some sufficiently highly excited atomic state is neg-
ligible so that the summation can be truncated. Second, a quasi-steady state approximation is
invoked for states above the n = 1 state which allows for their time derivative to be set to zero.
The reasoning behind this is that the timescales for the evolution of the excited states are much
faster than the evolution of the ground state, so that these states can be assumed to reach equi-
librium instantaneously. These assumptions reduce the system to a countable number of coupled
linear algebraic equations, and one differential equation for the time evolution of the ground state.
To solve the system of equations, we truncate the system at Q+ 1 states, and then separate
the n = 1 state from the Q excited states and treat it explicitly. For the excited states, eqn. 2.2 is
rewritten to yield,
0 = MQnQ + M1n1 + ΓQ. (2.3)
This system of equations can then be solved (e.g. by inversion of MQ) to give the population of
the excited states in terms of the ground state population n1 and the plasma parameters. This is
usually written in the form,
nq = C1,qn1 + Ci,qni, (2.4)
where C1,q is the coupling to the ground state and Ci,q is the coupling to the “continuum”.
18
The solutions for the Q excited states may then be substituted into the time evolution
equation for the n = 1 state to yield,
dn1dt
= −n1Seff + neniReff , (2.5)
with the effective ionization rate Seff and effective recombination rate Reff defined as,
Seff =∑m≥1
C1,mKi,m, (2.6)
Reff = −∑m≥2
Ci,mKi,m +∑m≥1
(βrad,m + βdia,m +Kr,mne). (2.7)
Transport terms, source terms, and etc. can be added in to eqn. 2.5 depending on simulation
needs.
The procedure for estimating the Dα emission is first to collect atomic data for the rates
of collisional and radiative transitions, ionization, and recombination as function of ne and Te.
Calculations of these terms are discussed in Burgess and Summers 1976 [15], and data has
been tabulated by Janev and Smith 1993 [72]. Then, solving the excited state equations with
n1 = 1 and ni = 0 gives C1,q, and solving the equations with n1 = 0 and ni = 1 gives Ci,q.
With these quantities, we can obtain the excited state population as a function of n1 and ni, and
multiplying by the rate of spontaneous emission A23 gives the Dα emission rate in terms of these
quantities. Recombination is typically negligible at temperatures relevant to the tokamak edge, so
ignoring the ni terms and substituting the approximate ne and Te dependence of the C1,q term
yields the GPI Dα emission rate,
S = n1nαe T
βe , (2.8)
where α and β are obtained from the CR model calculations and depend on the local values of ne
and Te. These values have been tabulated [20] for densities and temperatures relevant for the edge
of NSTX, and plots of their dependence on plasma parameters are presented in Fig. 2.4. Typical
plasma parameters near the peak of the GPI emission profile are ne ≈ 1013 cm−3 and Te ≈ 30 eV,
thus GPI Dα emission has the approximate form S ≈ n1n0.7e T 0.3e .
19
Figure 2.4: CR calculations of exponent for ne, α, and exponent for Te, β, used in estimation ofGPI Dα signal.
The simple CR model discussed above can be extended to account for molecular interactions,
or the presence of meta-stable states. Details on the inclusion of molecular interactions and their
rates can be found in Stotler 1996 [140]. CR models for helium including the treatment of the 21S
and 23S meta-stable states is discussed in included references [45, 50]. In addition, a more formal
description of CR models and a procedure for determining their validity is described in Greenland
1998 [51].
Experimental observations with Gas Puff Imaging have been compared to the Monte Carlo
neutral transport code DEGAS 2 which also computes a CR model for the given gas puff species
[142, 141, 138, 20]. Good agreement is found between time average profiles of GPI emission and the
simulation results. Also, absolute light emission values have been compared for deuterium puffs,
and again good agreement is found [20].
Chapter 3
Direct Comparison of GPI and BES Measurements of Edge Fluctuations in
NSTX
Understanding the physics of the edge and pedestal regions of tokamak plasmas is essential to
the operation of high performance fusion devices. Turbulence in the edge enhances transport above
neoclassical levels which severely limits the achievable confinement. Experiments have observed that
during the transition to high confinement (H-mode) an edge transport barrier (ETB) is formed,
and a suppression of the transport is achieved [149, 103, 152]. The reduced transport at the edge
leads to a buildup of pressure with steep gradients in temperature and density. This yields a
“pedestal” in the density and temperature profiles that the core profiles sit atop, thus the height
and width of the pedestal significantly impacts the achievable performance of the device. The steep
gradients of the pedestal provide a source of free-energy for MHD instabilities localized to the edge,
accordingly called Edge Localized Modes (ELMs), that can exhaust large amounts of stored energy
and particles. These ELMs can damage plasma facing components, therefore controlling ELMs
is a major concern for current and next-step fusion devices [42]. The nature of microturbulence
that drives transport in the edge, the development of the ETB during the L-H transition and
subsequent suppression of turbulence, the evolution of the pedestal, and the dynamics of ELMs
all underscore the importance of understanding this region of the plasma, and the need for highly
resolved diagnostic information.
In this chapter, we examine two diagnostics that meet this need, Gas Puff Imaging (GPI) and
Beam Emission Spectroscopy (BES). Measurements of edge turbulence made with BES and GPI
21
are directly compared in MHD quiescent H-mode operation that is free of large ELMs. This study
tests our understanding of the operation of, and interpretation of data from these two diagnostics.
The goal of this comparison is to provide experimental verification that the measurements agree
where modeling of diagnostic response suggests they should, and that observations are consistent
between the two diagnostics. In addition to increased confidence in the measurements that this
provides, this study lays the groundwork to use the diagnostics more collaboratively in the future.
Furthermore, the relatively close proximity of the BES viewing volume to the GPI gas manifold
allows the local effects of the GPI neutral gas puff on the plasma to be quantified.
3.1 The Beam Emission Spectroscopy Diagnostic
Similar to the GPI diagnostic, the Beam Emission Spectroscopy diagnostic measures light
emission from the collisional excitation of neutral atoms. Where the GPI diagnostic uses a gas puff
to localize the observation, the BES diagnostic images Dα line emission that is localized to the
intersection of optical sight lines with the neutral heating beam. The velocity of beam neutrals in
combination with the viewing geometry produces a Doppler shift which isolates the beam Dα line
from thermal emission from the bulk plasma. The Dα emission is collected by a high-throughput
optical assembly and imaged onto a set of optical fiber bundles. The fiber bundles transmit the light
to a set of collimating lenses and transmission filters that selectively pass the doppler shifted Dα
emission. Finally, PIN photodiodes measure the Dα light intensity. The dependence of measured
emission on plasma parameters has been studied (see Hutchinson,et al. [64] and references
therein for review), and for typical parameters in the NSTX pedestal the light emission can be
approximated by δIBES/〈IBES〉 ≈ 0.5δne/〈ne〉 for modest fluctuation levels, δne/〈ne〉 ≤ 10%
[64, 44]. The sensitivity of light emission to other parameters, including temperature, is found to
be much weaker than the density dependence.
The NSTX BES system, discussed in detail in references [128, 129], consists of two separate
arrays of viewing channels with one view centered at R = 130 cm (r/a ≈ 0.45) and another view at
R = 140 cm (r/a ≈ 0.85). For the current study we make use of only the outer 28 channel array,
22
and an illustration of the channel positions is presented in Fig. 3.2. The radial array provides
coverage from r/a ≈ 0.4 to well into the scrape-off layer, and two poloidal arrays are positioned at
r/a ≈ 0.85 and in the scrape-off layer. The optical view is aligned with the local magnetic field
at the position of the heating beam, an angle of 37◦ with respect to horizontal, to optimize cross-
field resolution. Spot-sizes at the neutral beam cross-section are 2-3 cm, and detailed point-spread
function calculations indicate minor image distortions from field line misalignment and atomic state
lifetimes. Thus, turbulent fluctuations with k⊥ρs ≤ 1.5 are well resolved. Data is acquired at a
2 MHz sampling rate, and frequency-compensating, wideband preamplifiers provide photon-noise
limited measurements at frequencies up to 400 kHz.
3.2 Plasma Conditions and NSTX Operation
The shots selected for this study are from a subset of the NSTX 2010 run campaign in which
both the GPI and BES diagnostics were operational, and each shot is chosen to be MHD quiescent
and free of large ELM events during the GPI observing period. Table 3.1 details the list of chosen
shots and shot parameters. The shots are selected from 2 different experiments carried out on
different run days. Typical ne and Te profiles for these shots are plotted in Fig. 3.1 along with
the GPI emission profiles and the R=140 cm BES radial array channel locations. The GPI light
emission is localized around the last closed flux surface (LCFS), and the light emission decays to
its half maximum value near the normalized flux value, ΨN of 0.8. This position corresponds to
densities of 2 − 5 × 1013 cm−3 and temperatures of 100 − 200 eV. The R=140 cm BES array has
radial views located at normalized flux values between ∼ 0.3 − 1.2, spanning the SOL, gradient
region, and pedestal top. This study focuses on the near-SOL and gradient region to compare with
the GPI measurements.
3.3 Relative Diagnostic Positioning
A schematic representation of the BES and GPI views is illustrated in Fig. 3.2(a) and (b).
Fig. 3.2(a) shows the relative position of the GPI view (blue polygon) and the BES channels (red
23
Table 3.1: Shot list for this study containing the shot number, toroidal field at the magnetic axis,plasma current, neutral beam heating power, average density, and GPI puff timing. Values aretaken at the time of the gas puff.
Shot BT [T] Ip [MA] PNBI [MW] 〈n〉 [cm−3] tGPI [s]
138845 0.34 0.8 3.8 5.9× 1013 0.58138846 0.34 0.8 3.8 5.9× 1013 0.58138847 0.33 0.8 3.8 6.0× 1013 0.58141249 0.36 0.7 2.9 3.3× 1013 0.48141254 0.37 0.8 2.9 3.4× 1013 0.48
Figure 3.1: Comparison of Thomson scattering profiles of ne and Te with GPI emission profile (bluecurve) and R=140 cm BES radial array channels (black and red dashed lines). The red dashedlines indicate the position of the inner and outer BES poloidal arrays.
diamonds) in the (R, Z) plane. BES channels and corners of the GPI view are plotted using their
cylindrical R and Z coordinates, ignoring the different toroidal angles. Flux surfaces are indicated
by dashed contours and labeled with their midplane r/a value, and the separatrix is indicated by
the solid black contour. In Fig. 3.2(b), The BES inner poloidal array channels (red diamonds)
are plotted in toroidal and machine-poloidal coordinates along with GPI pixels (blue points) with
similar values of poloidal flux. Magnetic field line traces in these coordinates are indicated by
dashed lines, and a trace of the generalized poloidal coordinate passing through the GPI points is
24
indicated by the dot-dashed line.
Figure 3.2: (a) R = 140 cm BES channel positions (red, diamonds) and GPI viewing area (bluepolygon) plotted over contours of poloidal flux for NSTX shot 141254. Flux surfaces are labeledby their midplane r/a value, and the separatrix is indicated by the solid black line. BES channelpositions and the corners of the GPI view are plotted using cylindrical coordinates. (b) BES innerpoloidal array (red diamonds) plotted in toroidal coordinates with GPI points (blue) for similarflux value. Magnetic field line traces are plotted as dashed lines, and the dot-dashed line traces thegeneralized poloidal direction.
The GPI view is centered 28◦ above the outboard midplane, and the radial extent of the view
typically captures the edge region, separatrix, and SOL. The R=140 cm BES array is positioned
at 12◦ above the outboard midplane and provides similar radial coverage. The separation between
the lower-left corner of the GPI view and the intersection of the BES radial and inner poloidal
array, as seen in Fig. 3.2(b), is 16◦ in the toroidal direction and 3◦ in the poloidal direction.
The physical distance between these two points is 35 cm, and the distance along the generalized
poloidal direction from the GPI corner to the intersection with the field line passing through this
BES channel is ∼ 20 cm.
The comparisons presented in Sections 3.4, 3.6, and 3.5 use the BES channel at the intersec-
tion of the radial and inner poloidal array as reference for shots 141249 and 141254. In R and Z,
this channel is located just inside of the inner edge of the GPI view ∼ 2 cm above the bottom of the
GPI view. For shots 138845-47, the BES channel one channel inside from the intersection of the
radial array with the outer poloidal array is used for comparison. This channel lies ∼ 4 cm out from
25
the inner edge and ∼ 1 cm above the bottom edge of the GPI view. GPI pixels used in comparisons
are chosen to overlap in R and Z with the BES reference point to within 1 cm. This ensures that
the comparison is done for similar flux surfaces, but field lines passing through the measurement
locations are still separated by ∼ 20 cm in the direction perpendicular to the magnetic field due to
the toroidal separation. For the calculation of poloidal correlation lengths presented in Section 3.7
shots 141249 and 141254 use the BES inner poloidal array, and shots 138845-47 use the outer BES
poloidal array. GPI correlation lengths are calculated for poloidal separations about the center
of the GPI view at the radial position of closest flux surface overlap with the given BES poloidal
array.
3.4 Fluctuation Levels
Based on the collisional radiative model, GPI emission is given by IGPI ∝ nαe Tβe , and α and
β are tabulated for a given values of ne and Te [142]. Near the 0.85 normalized flux position, the
plasma conditions are ne = 2 − 5 × 1013 cm−3 and Te = 100 − 200 eV. Under these conditions,
GPI exponents are α = 0.7 − 0.6 and β = 0.12 − 0.03, thus for small fluctuations the density
dependence is at least 5 times stronger than the temperature dependence. Similarly, BES emission
varies with plasma density to the 0.6− 0.5 power, and temperature dependence of the emission is
typically negligible. Therefore, it is expected that both diagnostics predominantly measure density
fluctuations, and, for small fluctuation amplitude, the ratio of GPI to BES RMS fluctuation levels
normalized to mean is expected to be ∼ 1.2.
The BES light emission is subject to low-frequency, < 4 kHz, fluctuations in the neutral
heating beams. In the analysis presented in this section, we account for this slow oscillation by
applying a Gaussian convolution filter (GCF) with a 4 kHz e−1 frequency cutoff to time series of
both diagnostics. The slowly varying “mean” intensity, 〈I〉 is then defined as the GCF of the raw
trace, and the fluctuating signal is defined as δI = I−〈I〉. By extension then, the RMS fluctuation
26
level is given by,
δIrms/〈I〉 =〈(I − 〈I〉)2〉1/2
〈I〉, (3.1)
where 〈·〉 represents the GCF with 4 kHz frequency cutoff. Time traces of raw (I), mean (〈I〉).
The GCF is strictly positive-valued in the time domain, which ensures that the resulting RMS
fluctuations are real-valued at all times. Fluctuating (δI) intensity for both GPI and BES are
plotted in Fig. 3.3. GPI RMS fluctuation levels are plotted in Fig. 3.4. Generally for shots in this
database, GPI RMS fluctuation levels are ∼ 18% of the mean which is ∼ 9 times greater than the
BES values of ∼ 2% at this location. This large discrepancy in fluctuation levels is well above the
expected ratio of 1.2.
Figure 3.3: Traces of (a) GPI raw (gray) and 〈I〉 (blue), and (b) δI. Plots (c) and (d) are BEStraces. All traces taken at the 0.85 ΨN location.
The large discrepancy in RMS fluctuation levels suggests that either BES is underestimating,
or GPI is overestimating the fluctuations at this location. Temperature dependence is not expected
to be important, as discussed above, and both diagnostics have similar radial localization. The
poloidal resolutions are different, which could lead to a difference in measured fluctuation levels
due to an averaging over small scales, but applying a spatial filter to the GPI observations cannot
account for the observed discrepancy. Large plasma fluctuations can alter the neutral gas density of
27
the GPI gas cloud, and this may induce fluctuations in light emission at smaller radii. This effect is
commonly referred to as “shadowing”, and a simple discusion of the effect can be found in [142]. To
account for the observed discrepancy, the neutral density perturbation at the measurement location
would need to be of order 10%. Shadowing is a complex effect involving full 3D plus time dynamics,
and efforts to better quantify the effect are currently underway. For now, shadowing remains as a
possible explanation.
Figure 3.4: Traces of δIRMS/〈I〉 for BES (red) and GPI (blue) for ΨN = 0.85.
The NSTX fast scanning probe [11] is capable of making measurements of edge fluctuations
in NSTX H-mode plasmas for ψN > 0.9. Probe data from Type-III ELMing H-mode plasmas with
Bt = 0.45T , 1.3 MW of NBI power, and 0.8 MA plasma current have been presented in [12]. For
these shots, it was observed that normalized density fluctuations increased from ∼ .20 at ΨN = 0.9
to a peak of ∼ 1.50 at Ψn = 1.0. In the SOL, Normalized fluctuation levels were relatively flat
at a value near 1.0. These H-modes are significantly different than those presented in this paper.
Most notably these shots were at higher Bt, lower NBI power, and Type-III ELMs were present.
Therefore, a comparison with results presented here is not useful. However, future experiments
could be designed to compare normalized fluctuation levels between all three diagnostics, and this
may help resolve the observed discrepancy between the GPI and BES fluctuation levels.
28
3.5 Fluctuation Statistics
Fluctuation statistics (i.e. PDFs and their moments) provide a well defined and experimen-
tally accessible way of characterizing turbulence. The statistical description of plasma turbulence
seeks to explain these quantities and gain insight into the underlying physical mechanisms in the
process. Much effort in the study of turbulence has focused on the understanding and prediction
of the PDFs of turbulent quantities and the search for a universal distribution or universal features
of turbulent distributions (for a review of the statistical description of plasmas see Krommes[85]).
Furthermore, observations of non-Gaussian statistics and intermittency have led to a number of
realizations about the importance of coherent structures and the nature of turbulent transport in
the edge and SOL regions (e.g. [107, 13, 14]).
Figure 3.5: PDFs of (a) GPI and (b) BES intensity from 10 ms period. Dashed Lines indicateGaussian PDFs with similar mean and variance.
As mentioned previously, BES emission is subject to neutral beam oscillations below 4 kHz,
so PDFs are created from high-pass filtered, 10 ms time traces for both GPI intensity and BES
intensity. The PDFs and their normalized skewness (3rd moment) and normalized excess kurtosis
values (4th moment) are shown in Fig. 3.5(a) and (b). Moments are normalized to the relevant
power of the variance, and the normalized excess kurtosis is the difference of the normalized 4th
moment of the PDF from Gaussian statistics. The first two moments of the PDF, mean and
variance, are largely meaningless here due to the filtering of the low-frequency contributions and
differing intensity scales, but the desired information is captured by δIrms/〈I〉 as seen in Fig. 3.4.
29
Skewness is a measure of the asymmetry of a distribution, while kurtosis is a measure of the relative
weight of the tails. Deviations from a Gaussian PDF (represented by the dashed line in Fig. 3.5) are
indicated by deviations of the normalized skewness from S = 0 and the normalized excess kurtosis
from K = 0. For this shot, the GPI distribution is slightly skewed toward positive values, and a
small positive kurtosis indicates more weight is present in the tails relative to a normal distribution.
The BES distribution has negligible skew and small positive kurtosis.
The skewness and kurtosis values for a given PDF can be used as a reduced description
of the full PDF. This simplification enables easy comparison of PDFs for different shots, plasma
conditions, or radial locations, so that trends in the fluctuation PDFs may be identified (e.g. see
analysis in Labit et al. [88]). Figure 3.6 plots skewness and kurtosis values calculated from PDFs
of ∼ 10 ms time blocks of intensity fluctuations that have been high-pass filtered at 4 kHz. Five
successive time blocks from each shot are analyzed for t = 0− 50 ms after the GPI emission peak.
Skewness and kurtosis values from 2.9 cm (filled symbols) and 0.7 cm (open symbols) inside the
separatrix are both included in the figure.
Figure 3.6: Scatterplot of skewness and kurtosis values for BES (filled triangles) and GPI (filledsquares) at r − rsep ≈ 2.9 cm, and BES (open triangles) and GPI (open squares) at r − rsep ≈ 0.7cm. The dashed line represents a second degree polynomial fit to the GPI data.
Near r − rsep ≈ −2.9 cm, both GPI and BES intensity may be approximated by I ≈ Cnαe ,
30
where α ≈ 0.6 for GPI and α ≈ 0.5 for BES. Therefore, it is expected that the measured PDFs
would be similar at this location, and rough agreement is apparent in Fig. 3.6. Averaging values
for the 6 shots yields GPI skewness of S = 0.24 ± 0.25 and kurtosis of K = 0.38 ± 0.44, and BES
skewness of S = −0.01 ± 0.13 and kurtosis of K = 0.11 ± 0.34. Values are close to a normal
distribution within variances for both diagnostics, but GPI does show a shift in the average values
toward small positive skewness and kurtosis. Raising a random variable with a skewed distribution
to a fractional power will shift the skew to negative values though, so it is possible that small
differences could be explained by the difference in dependence on plasma parameters.
GPI skewness values at both radial locations vary by about ±0.5, and points appear to
follow a quadratic trend when kurtosis is plotted vs. skewness. Fitting the data to a second degree
polynomial yields the dashed line plotted in Fig. 3.6. Similar quadratic trends have been observed
in GPI data of edge plasma fluctuations across a number of machines [120] and in probe data from
TORPEX [88]. In addition, analytic models attempting to illuminate the physical mechanisms
determining this trend have been explored [86]. Realizability constraints, derived by invoking the
positive semidefiniteness of the mean of any nonnegative quantity, yield the relation [85] K ≥ S2−2,
so a quadratic trend is perhaps not unexpected. BES values, however, exhibit a high degree of
scatter in skewness and kurtosis for the outer radial location, and they do not clearly exhibit a
quadratic relationship between S and K.
The previous observations using GPI by Sattin et al. [120] measured skewness and kurtosis
values across a wide range of radial locations from 6 cm inside the separatrix to 6 cm into the SOL,
and a simple model was then put forward to describe the observed PDFs as the sum of two inde-
pendent basis functions. Both the sum of two gamma distributions and two beta distributions were
explored. The correspondence of GPI measurements and probe measurements was not addressed,
though, thus GPI measurements were effectively assumed to measure density fluctuations.
Caution should be taken, though, as interpreting the observed trends in GPI data is quite
difficult, for a number of reasons. The functional form of the emission determined by the atomic
physics necessarily implies a complicated dependence of the distribution of the intensity on the
31
distributions of underlying plasma parameters, both ne and Te. Furthermore, the functional form
of the GPI intensity depends on the plasma parameters which change significantly across the edge
and SOL regions, and order unity fluctuations are likely to complicate the interpretation of both
GPI and BES diagnostics near the separatrix. Finally, it is unlikely that ne and Te fluctuations can
be considered to be independent, thus understanding the trends of moments and their relation to
the underlying plasma fluctuations would likely require a focused effort to model the edge plasma
with synthetic GPI and BES diagnostics.
3.6 Cross-Diagnostic Analysis
As discussed in previous sections, both the GPI and BES diagnostics should predominantly
measure density fluctuations near the 0.85 normalized flux position. In this section, we employ cross-
correlation and cross-spectral analysis to estimate the similarity between time traces and spectral
content of the signals. Due to technical issues with the BES data acquisition for these shots, there
exists an unknown time offset between the GPI and BES time records, and, compounding this
difficulty, a constant drift is present in the BES clock time kept by one diagnostic relative to the
other. Fortunately, this drift is readily visible in the time-lagged cross-correlation function versus
time, shown in Fig. 3.7, and can therefore be easily corrected. Time traces have been high-pass
filtered to removed the < 4 kHz fluctuations which are problematic due to beam fluctuations.
Finding the maximum correlation value for each time point yields a 1D function for the lag-time
to peak correlation versus time. Then, a clock-correcting offset and dilation factor can be obtained
by performing a linear fit of this function, and we utilized this dilation factor to align the BES time
base with GPI in the analysis contained in this section. An unresolved absolute time offset between
the two signals still exists, however, and this offset limits the potential use of the correlation analysis
in some applications. Note that this peak correlation value is persistent in time, and that the peak
correlation value of R ≈ 0.5 is significanly above the background in this case.
With the correction applied to the BES time base, Fig. 3.8(a) shows the traces of time-
lagged cross-correlation, and time traces of intensity fluctuations for a GPI pixel and a BES channel
32
Figure 3.7: Contour plot of time-lagged cross-correlation between GPI and BES signals versus time.Peak correlation value persists in time, but constant linear drift is present.
normalized to standard deviations are compared in Fig. 3.8 (b). For shots listed in Table 3.1, peak
correlation values often exceed R = 0.6, and a high degree of similarity is visually apparent in
traces of the normalized fluctuations.
Figure 3.8: Traces of (a) time-lagged cross-correlation between GPI and BES, and (b) time tracesof GPI and BES intensity.
The cross-correlation vs. radius for GPI pixels correlated with a single BES channel is plotted
in Fig. 3.9(a) and (b), and contour plots of the time-lagged cross-correlation vs. radius are included
in Fig. 3.9(c) and (d). The x-axis for these plots is the difference of the radial location of the GPI
pixels mapped to the midplane and the midplane outer separatrix radius. The r− rsep location for
33
the BES reference channel is illustrated by the dashed red line. The radial array of pixels ∼ 2 cm
above the bottom of the GPI view is used for this comparison.
The correlation functions in Figures 3.9(a)-(d) exhibit a number of interesting features. First,
the cross-correlation features for all shots are significantly radially extended, and strong correlation
exists up to ∼4 cm into the SOL. Second, shots in the 141254 series generally exhibit a roughly
linear time shift in the peak correlation time vs. radius. This is due to the radially extended
wavefronts for this structure being tilted in the radial vs. poloidal plane. Shots in the 138845 series
exhibit distinctly different behavior. For these shots, the peak correlation time is roughly constant
inside of r − rsep ≈ −2 cm. At r − rsep = −1 cm the maximum correlation values are slightly
suppressed, and outside of this location a time shift or anti-correlation is observed. This may be
indicative of a strong sheared flow present at r − rsep = −1 cm. It should be noted, however, that
the GPI light falls off rapidly inside of r − rsep ≈ −4 cm.
Figure 3.9: Plots (a) and (b) are traces of cross-correlation vs. radius for GPI pixels with a BESreference channel. Plots (c) and (d) are contour plots of the cross-correlation vs. radius and time-lag. The black dashed line indicates the separatrix location, and the dashed red line indicates theradial location of the BES reference channel.
The cross-spectral density may be used to provide a measure of the shared frequency content
34
between the BES and GPI signals. Traces of (a) cross-coherence, (b) cross-spectral density, (c)
cross-phase, and (d) phase uncertainty comparing a BES channel and a nearby GPI pixel in the
R-Z plane are shown in Fig. 3.10. Similarities are seen for the low frequencies in the auto and cross
spectral densities plotted in Fig. 3.10(b), and a peak near 10 kHz is observed by both GPI and
BES. The BES auto spectral density shows a second feature near 80 kHz that is not seen in the
GPI spectrum. This 80 kHz feature is seen only in this shot, and it is likely dominated by noise
due to the higher noise floor in the GPI spectrum.
Cross-coherence measures the constancy of the relative phase of similar frequency components
between signals. The cross-coherence spectrum in Fig. 3.10(a) demonstrates that a well-defined
phase relationship exists between the two diagnostics for the 10 kHz frequency band, and similar
peaks in the cross-coherence spectrum are seen in all shots included in Table 3.1. GPI observations
show intensity fluctuations propagating downward (-y, ion diamagnetic drift direction) through the
view, therefore the strong cross-coherence suggest that fluctuations, which are likely extended along
field lines, propagate coherently between the GPI and BES views. This corresponds to a distance
of ∼ 20 cm in the generalized poloidal direction perpendicular to the magnetic field.
Figure 3.10: Traces of (a) cross-coherence, (b) cross-spectral density, (c) cross-phase, and (d) phaseuncertainty for a BES coord and the closes GPI pixel in the R-Z plane.
35
3.7 Characteristic Time and Length Estimates
Estimates of the poloidal correlation length and the decorrelation time provide a characteristic
length scale and a characteristic lifetime for the fluctuations, respectively. Aside from characterizing
the fluctuations, the correlation lengths, decorrelation times, and scalings of these quantities with
other parameters may offer insight into the underlying instability driving the turbulence (e.g see
references [130, 131]). We do not address the scaling here, but instead compare the estimates
produced by the BES and GPI diagnostics. Auto-spectral density functions and cross-coherence
functions between pixels separated by 2.4 cm, 4.8 cm, and 7.2 cm are plotted in Fig. 3.11. As
discussed previously, peaks near 10 kHz and 80 kHz are present in the BES coherence spectrum,
but only the 10 kHz peak is present in the GPI spectrum. This 80 kHz feature is only present
in shot 141254, but a strong coherence feature at low frequencies, between about 0-20 kHz, is
present in all of the shots. Therefore, for the following analysis a band-pass filter is used with a
low frequency cutoff of flc = 4 kHz and a high frequency cutoff of fhc = 50 kHz. The low frequency
cuttoff is chosen to eliminate contamination of beam fluctuations in the BES signal, and the high
frequency cutoff is placed at the point where the first coherence peak crosses the√N noise floor,
where N is the number of time blocks used for the coherence calculation. This method of estimating
decorrelation times is similar to the generalized cross-correlation method of time-delay estimation
[84], but an ad-hoc filter based on the coherence spectrum is used here.
After applying this filter, time-lagged cross-correlations are calculated using 4 ms long time
blocks for varying pixel separations, and envelope functions are calculated using the Hilbert trans-
form. Individual estimates of the correlation function are then averaged over 60 ms. Block-averaged
correlation functions for GPI and BES are shown in Fig. 3.12(a) and (e) respectively, and envelope
functions calculated with the Hilbert transform are overlaid. Poloidal correlation lengths can be
estimated by a Gaussian fit to the zero-lag envelope correlation versus pixel separation shown in
Fig. 3.12(b) and (f). Similarly, decorrelation times can be estimated by a Gaussian fit to the
envelope peak correlation versus time to peak correlation shown in Fig. 3.12(c) and (g). Finally, a
36
Figure 3.11: Plots of: the Cross-Coherence between poloidaly separated channels for (a) GPI and(c) BES, and Autopower spectra for GPI (b) and BES (d).
velocity for structures can be estimated by Time-Delay Estimation (TDE) from the plot of time to
peak correlation versus poloidal separation shown in Fig. 3.12(d) and (h). Structures are observed
to propagate downward (-y, ion diamagnetic drift direction) in the GPI view, and the magnitude
of the TDE velocity is estimated from Fig. 3.12(d) and (h).
Estimates of poloidal correlation lengths, decorrelation times, and TDE velocities for several
shots are compared in Fig. 3.13(a) and (b), respectively, and results are tabulated in Table 3.2.
Dashed lines in Fig. 3.13(a) and (b) indicate perfect agreement and where the absolute difference
equals 40% of the mean. All values show good agreement at the ±40% level, though BES velocity
estimates derived from the correlation functions are consistently lower than GPI velocities. All
velocities are in the ion diamagnetic drift direction (-y in the GPI view).
Poloidal correlation length estimates have also been performed using the full extent of the
GPI view, and this method is able to resolve the first anti-node of the spatial correlation function.
The results of the calculation using the full view strongly suggest that the 4-point estimate used
in Fig. 3.12 and 3.13 significantly underestimate the correlation length. During the 2010 run
37
Figure 3.12: plots of: (a) and (e) Time-lagged cross-correlations (solid) with envelope functions(dashed), (b) and (f) zero-lag envelope peak correlation versus poloidal separation, (c) and (g)envelope peak correlation versus time-lag to peak, and (d) and (h) time-lag to envelope peakcorrelation versus poloidal separation. Values for poloidal correlation length, and decorrelationtime represent the 1/e length for a Gaussian fit to the corresponding plot.
Figure 3.13: Scatterplots comparing (a) poloidal correlation lengths estimates, (b) decorrelationtime estimates, and (c) TDE velocity estimates.
campaign, it was discovered that BES poloidal arrays did not have the poloidal coverage to resolve
the first anti-node, and, due to data acquisition constraints, only 4-channels of the BES poloidal
array were typically available. Therefore, the 4-point estimate is used for comparison here, and
this estimate does appear to be consistent between the two diagnostics. The upgraded BES view
38
Table 3.2: Comparison of poloidal correlation lengths, decorrelation times, and velocities estimatedfrom GPI and BES correlation functions.
BES GPI BES GPI BES GPIShot LPOL [cm] LPOL [cm] τc [µs] τc [µs] vg [km/s] vg [km/s]
141249 11.1 ± 1.1 16.4 ± 1.0 22.6 ± 2.2 27.4 ± 0.1 4.3 ± 1.0 7.5 ± 0.9141254 13.3 ± 1.9 15.0 ± 0.7 29.4 ± 4.8 23.6 ± 1.8 5.1 ± 1.4 7.1 ± 0.8138845 18.1 ± 5.4 14.9 ± 0.4 42.0 ± 0.1 41.4 ± 3.1 4.0 ± 0.9 5.7 ± 0.4138846 19.0 ± 8.1 14.9 ± 0.7 31.9 ± 0.8 34.1 ± 2.7 5.3 ± 1.7 6.2 ± 0.5138847 14.4 ± 4.9 14.2 ± 0.6 30.3 ± 10.0 36.4 ± 2.4 4.2 ± 1.4 5.9 ± 0.6
for NSTX-U has been designed to address this issue, and should be able to resolve anti-nodes of
the correlation function.
3.8 Gas Puff Effects
During the GPI observation period ∼ 3 × 1020 neutral deuterium atoms are puffed into the
edge, and the total electron content of the plasma before the GPI pulse is ∼ 6× 1020 electrons [20].
It is estimated that ∼ 20% (6 × 1019) of the gas puff atoms become ionized inside the separatrix,
but due to losses by edge particle diffusion the global density increase is expected to be ≤ 3% [168].
Previous work [168] has explored the possible perturbing effects of the GPI gas puff on the edge
plasma and turbulence, and found that edge electron density and temperature changed by ≤ 10%
at the Thomson scattering location far from the GPI puff preceding and up to the peak in the GPI
emission. Furthermore, edge turbulence quantities as measured by the GPI diagnostic did not show
significant variation during the gas puff. Still, the effect of the gas puff on the local density and
temperarture in the gas cloud could not be measured. A cursory look at the BES measurements
was included in the previous study, and it was found that the mean BES Dα emission increased
and closely followed the GPI Dα emission during the GPI puff. Here we provide a more detailed
examination of the GPI puff effects on the BES fluctuation spectra and poloidal correlation length.
The close proximity (∼ 40 cm) of the BES sight-lines to the GPI gas cloud provides us with
an opportunity to examine the local effects of the gas cloud on the edge turbulence. The low-pass
39
filtered, mean BES intensity is seen to increase concurrently with the GPI gas puff, and the increase
at different radial and poloidal locations is captured by Fig. 3.14. The largest increase is seen at
Xsep = r − rsep = −0.8 cm and Z = 7.6 which corresponds to the bottom of the GPI view. Data
from BES channels above Z = 7.6, closer to the center of the gas puff, is not available in these
shots. The increase in signal decays at locations further from the gas cloud (decreasing R and Z),
and typically the effect of the gas puff on the mean BES signal is negligible inside of Xsep ≈ −12
cm.
Figure 3.14: Time traces of low-pass filtered BES intensity for varying (a) Xsep = r − rsep and (b)Z. Times are relative to the GPI gas puff timing.
The effect of the gas puff on the fluctuations is illustrated by Fig. 3.15(a) and (b) which show
Continous Wavelet Transforms (CWT) for δI/I, the BES fluctuations normalized to the 200 Hz
low-pass filtered trace. Time traces of frame-averaged GPI emission are plotted above the CWT.
The low-pass filter is used to capture only the slow variation due to the GPI puff. Beam fluctuations
manifest as a coherent feature near 900 Hz in both figures, and weaker, less coherent harmonics can
be observed up to ∼ 4 kHz in average spectra for long time blocks. In Fig. 3.15(a) the amplitude
of the 1-20 kHz fluctuation band increases with the GPI emission. Shots 138845-47 each share this
behavior, but in shots 141249 and 141254 any increase in amplitude in this band is less clear due
to a feature centered at 10 kHz with a ∼ 10 kHz width that appears to be insensitive to the GPI
puff.
Block-averaged BES frequency spectra for δI/I before and during the gas puff, plotted in
40
Figure 3.15: Continous Wavelet Transforms of BES fluctuations normalized to 100 Hz low-passfilter for shots 138845 (a) and 141249 (b). Time traces of the average GPI intensity is plottedabove the CWTs.
Fig. 3.16(a) and (c), illustrate this increase in fluctuations more clearly. Here, autospectral power
estimates are obtained by averaging spectra from 23 time blocks ∼2.6 ms in length. Time blocks
are taken at 60-0 ms before, and 30-90 ms following the gas puff trigger. Fluctuations in the 1-10
kHz band increase signifcantly during the gas puff for Shots 138845-47, and a smaller increase is
seen only below ∼4 kHz in Shots 141249 and 141254. Coherence spectra (using raw traces without
normalization) for BES channels separated by 4.8 cm in the poloidal direction are plotted in Fig.
3.16(b) and (d). Before the gas puff, a strong coherence peak is seen between 10-20 kHz for shots
138845-47, and during the gas puff the coherence values for this feature are reduced by ∼0.2. In
contrast, Shots 141249 and 141254 show only minor changes in the coherence spectrum with the
gas puffing, and the feature near 10 kHz appears to shift to slightly higher frequency.
Poloidal correlation lengths offer another tool to assess possible changes in the turbulence
due to the GPI neutral puff. Correlation lengths are estimated by the same method used in the
previous section for 60-0 ms preceding, and 30-90 ms following the gas puff trigger. The results are
tabulated in Fig. 3.3. Poloidal correlation lengths for all shots become shorter during the gas puff,
although only two shots are different from zero within uncertainties.
The large, ∼300% increase in BES mean intensity levels during the gas puffing is a striking
feature of the analysis in Section 3.8. To better understand the effects of the gas puff on the
BES measurements, DEGAS2 simulations [141] have been performed, and it is found that gas
41
Figure 3.16: (a) and (c): Block-averaged autopower spectra of fluctuations normalized to mean.(b) and (d): coherence for BES channels separted by 4.8 cm. Black traces are spectra for 60-0 msbefore the gas puff trigger, and red traces are spectra for 30-90 ms following the trigger.
Table 3.3: Poloidal correlation length estimates from 60-0 ms before, and 30-90 ms after the gaspuff trigger.
Shot LPOL pre-puff LPOL post-puff Diff.
138845 19.7 ± 3.3 [cm] 17.7 ± 4.6 [cm] -2.0 ± 5.7 [cm]138846 20.5 ± 2.6 17.3 ± 6.7 -3.1 ± 7.2138847 19.5 ± 2.6 13.3 ± 3.5 -6.2 ± 4.3141249 16.2 ± 2.6 10.5 ± 1.1 -5.6 ± 2.8141254 14.5 ± 1.0 13.2 ± 1.8 -1.4 ± 2.1
puffing increases the neutral deuterium content at the intersection of the BES view with the neutral
beams by ∼ 1e16/m3. This is 3 orders of magnitude less than the plasma density at this location,
therefore it is not expected that neutral-neutral collisions will play a significant role. Furthermore,
the increase in local plasma density is presumably of this order or less, so it is unlikely that the
increased emission is attributable to an increased plasma density.
The gas puff does lead to an increase in Dα signal on the BES channels, but the BES system is
designed to use the doppler shift and interference filters to reject most of this background emission.
42
Simulations estimate the thermal Dα signal on the BES channels to be . 6.6e19 photons/(m2 sr s).
The BES etendue is 2.3 mm2 sr, optical fiber losses are 42%, and attenuation by the interference
filter is 5e−4% at the unshifted Dα wavelength. Thus BES is expected to measure 3.2e6 photons/s,
or a 65 µV signal output. This is several orders of magnitude less than the ∼1 V signal levels
typically seen, and clearly does not account for the increase seen in the mean BES signal.
3.9 Summary of Results
Comparisons of BES and GPI measurements of edge fluctuations in NSTX have been pre-
sented. The BES and GPI diagnostic views share coverage over the range of normalized flux from
Ψn ≥ 0.8 into the SOL at similar poloidal angles, but are separated in toroidal angle by ∼ 16◦.
Near Ψn = 0.85 both diagnostics are expected to predominantly measure density fluctuations. The
measured fluctuations are therefore expected to have similar characteristics, and many similarities
are observed. Direct comparison of GPI and BES fluctuation measurements on field lines separated
by ∼ 14 cm in the direction perpendicular to B yields strong correlations, R > 0.6, and strong
cross-coherence between ∼ 5−15 kHz. Fluctuation PDFs and their moments show good agreement
at Ψn = 0.85, and both GPI and BES distributions are close to Gaussian at this location. Cor-
relation lengths, decorrelation times, and TDE velocity estimates all show good agreement within
±40%. It should be emphasized that, given the physical separation of the measurement locations,
the observed agreement is very strong.
There are some important differences, however. Measured GPI fluctuation levels normalized
to the mean are a factor of 5-6× greater than the measured BES fluctuation levels. This discrepancy
is unresolved, but could be explained by either an underestimate of the fluctuation level by BES at
this location or an overestimate by GPI. It is possible that large fluctuations in the plasma density
near the separatix may induce fluctuations in the neutral density, thus an increased fluctuation
level would be measured by GPI at smaller radii due to the modulated neutral density. In addition
to the dissimilarity of fluctuation levels, differences in the fluctuation PDFs are present at larger
radii, but this is not unexpected due to large-fluctuation amplitudes and increased temperature
43
dependence of the light emission. Also, comparing the 4-point estimate of the correlation length to a
calculation utilizing the full GPI spatial information revealed that the 4-point estimate consistently
underestimates the poloidal correlation length.
Finally, effects of the neutral deuterium puff used for GPI on the BES mean and fluctuations
are quantified. BES mean intensity levels are seen to increase by as much as 300% near the
separatrix position. Sensitivity of the BES mean to the neutral puff decreases with distance of the
measurement location from the center of the gas cloud. In addition, CWTs and frequency spectra
for fluctuations normalized to the 200 Hz low-pass filtered BES signal show an increase in relative
fluctuation levels in the 1-10 kHz band, but only one shot shows a significant change in the poloidal
correlation measured before and during the gas puff.
Chapter 4
Velocimetry
The estimation of motion is a very general problem arising in computer vision, image analysis,
and signal processing. A great many techniques, spanning a diverse range of fields, have been
developed to handle this problem. This chapter examines the common techniques of Time Delay
Estimation, Optical Flow, and Pattern Matching for application to velocity extraction from image
sequences. Each of these techniques has been employed extensively in analyzing Imaging data of
plasma systems.
4.1 Time Delay Estimation for Motion Estimation
The estimation of time delay of signal arrival between spatially separated measurements,
or simply time delay estimation (TDE), is a common problem encountered in signal processing.
It has played an important role in a diverse array of fields including radar and sonar detection,
oceanography, seismology, and speech localization and tracking. TDE techniques have also been
used extensively In the field of plasma physics to measure turbulent velocities [147, 71, 70, 164, 60,
56, 26, 28]. This section describes the basic TDE technique.
The simplest conception of TDE is a two-point, one dimensional velocity estimation tech-
nique. Consider two zero-mean time traces, X(t) and Y (t) measured at locations separated by a
distance ∆x. This could be two probes, two pixels from an image, or any spatially separated mea-
surements. The full time traces are made up of N time points Xi, Yi with a time between samples
dt. To estimate the velocity at time tn = ndt, the full arrays are broken into smaller time blocks
45
Nb in size starting at index n. From these subarrays, the cross-correlation function, RXY (τ = ldt)
is calculated by shifting one subarray relative to the other, and it is given by,
RXY (τ) =1
Nb
n+Nb−1∑i=n
Xi ∗ Yi+l. (4.1)
It’s assumed here that enough data points exist in the full arrays for an unbiased estimate of RXY
can be calculated at the maximum lag value. The correlation coefficient function is the normalized
correlation function given by,
ρXY (τ) =RXY (τ)√
RXX(0)RY Y (0), (4.2)
and is bounded between −1 < ρ < 1. For a given signal pulse propagating between the two
measurement locations, the peak of the correlation coefficient function occurs at a time lag of τpeak,
and the velocity estimate v(tn) = ∆x/τpeak. The cross-spectral density function is the Fourier
transform pair of the cross-correlation function, so correlation functions can be calculated in the
frequency domain as an alternative to the above description. Common extensions of this procedure
can be made to account for motion in 2D, to include 3 or more points in the estimate, and to
estimate sub-dt time delays.
A number of complications may arise for real applications. First, if the signal pulse expe-
riences significant dispersion during propagation then the peak of the cross-correlation coefficient
function will not coincide with the peak of the envelope function introducing an error in the veloc-
ity estimate. This issue is discussed in some detail in Bendat and Piersol [9]. Second, spatially
extended structures moving orthogonally to the separation direction can yield a non-zero velocity.
For example, a plane wave propagating in the direction orthogonal to the separation with a velocity
v and wave fronts aligned at an angle θ with respect to the propagation direction will be measured
moving at a velocity vo = v tan(θ) in the orthogonal direction. The impact of spatial structure on
TDE and possible solutions have been discussed in [43].
Simple TDE techniques have been implemented for use with NSTX GPI data, however their
use is limited by several issues. First, TDE techniques necessarily produce velocity measurements at
a fraction of the original time resolution, and the GPI time resolution of 2.5 µs is currently enough
46
to only marginally resolves the turbulence dynamics with associated ∼10 µs autocorrelation times.
Therefore, TDE velocimetry applied to the NSTX GPI data is only capable of extracting the slowly
evolving, mean velocity components. Second, correlation structures of edge turbulence are often
spatially extended and tilted or sheared with respect to the propagation direction, so significant
errors are introduced as discussed in the previous paragraph. The addition of a fast photomultiplier
array, as has been done in Alcator C-mod, could increase the time resolution to the point where
TDE techniques become more feasible, and the implementation of more advanced, multi-point TDE
algorithms could overcome the errors introduced by the spatial structure of the turbulence.
4.2 The Optical Flow velocity estimation
The optical flow approach to velocity estimation is derived from the assertion that brightness,
I(x, t) is conserved from frame to frame. The original implementation was described by Horn and
Schunk[61], but a great many extensions have been proposed since[4, 136]. In analogy to the
continuity equation in fluid mechanics, the constraint equation can be expressed as
∂tI +∇ · [vI] = 0. (4.3)
Typically, the more rigid constraint of constant brightness is imposed to give
∂tI + v · ∇I = 0, (4.4)
∂tI + u∂xI + v∂yI = 0. (4.5)
This is equivalent to the continuity equation for the case of divergence free flow. The intensity field
is discretized on the image pixel grid giving one such equation for each pixel, and also two unknown
velocity components for each pixel. Following the procedure of Srinivasan and Chellappa [136],
the velocity field is decomposed into a set of K basis functions φk so that
u =∑k
ukφk , v =∑k
vkφk. (4.6)
Substituting eq. 4.6 into eq. 4.5 gives
∂tIp +∑k
ukφk∂xIp +∑k
vkφk∂yIp = 0, (4.7)
47
where the pixel index p has been included here explicitly. If error is allowed for in the above
equation by introducing an error term εp to the right hand side then the velocity components may
be solved for by a least squares minimization. The procedure is to construct the sum over all pixels
of the squared errors and then set each of the partials of this function with respect to the individual
uk and vk to zero. This process yields 2K equations
0 =∑p
∂tIpφk′∂xIp +∑k
ukφkφk′∑p
∂xIp∂xIp +∑k
vkφkφk′∑p
∂yIp∂xIp
0 =∑p
∂tIpφk′∂yIp +∑k
ukφkφk′∑p
∂xIp∂yIp +∑k
vkφkφk′∑p
∂yIp∂yIp (4.8)
Solving the above equations yields the coefficients uk and vk, and then eq. 4.6 can be used to
recover the full velocity field.
Writing the above in matrix form can reduce the least squares solution to a relatively simple
matrix problem. Equation 4.7 can be expressed in matrix form by using the following definitions,
Apk =
φk∂xIp , k even
φk∂yIp , k odd
and,
xk =
uk , k even
vk , k odd
With these definitions, the equation can be rewritten as
Ax = b, (4.9)
where the time derivatives have been collected into b. It can be shown that the least squares
solution, eq. 4.8 can be rewritten as
ATAx = ATb. (4.10)
Then the solution is provided by inverting ATA,
x = (ATA)−1ATb. (4.11)
48
The procedure to solve for the velocity field is to first construct the arrays A, and b from
the partial derivatives of the intensity field and the chosen basis functions φk. Then, the velocity
coefficients are found by eq. 4.11, and the full velocity field is recovered by eq. 4.6. This process is
then repeated for each frame in the image sequence to give a velocity map v(x, y, t) for each image
at the original image resolution.
The implementation used to examine NSTX GPI data is discussed in Munsat, et al. [105].
The 2D basis functions are taken to be separable, i.e.
φk(x, y) = φi(x)φj(y), (4.12)
with k = Nx ∗ j + i where Nx is the image x dimension. The chosen set of 1D basis functions are
the complete set of discrete second-order coiflet functions. This is a complete set of orthogonal
functions on the discrete pixel grid that are defined as scaled and translated transformations of the
‘mother’ wavelet.
Optical flow estimation schemes experience two major limitations. First, the derived velocity
field cannot measure flow along intensity contours. This can be seen in the original definition of
the constraint equation, eq. 4.5 where the velocity is dotted into the intensity gradient so that only
the components in the direction of the intensity gradient are constrained. This is more commonly
referred to as the ”aperture problem”, and places severe constraints on the extracted velocity field.
One attempt at overcoming this problem is to use spatial smoothing or some other neighborhood
function that uses nearby information to correct this problem. This can help ameliorate the issue,
but the problem can still persists for scales larger than the chosen neighborhood. At a more
fundamental level, there is no information in the intensity field by itself that can be used to
disambiguate motion along contours of intensity. To resolve this kind of motion requires extra
information, further assumptions, or more constraints on the problem. Second, If the motion of
structures is not adequately time resolved the structures can become “lost” by the algorithm. This
is known as temporal aliasing, and effectively imposes an upper bound on the extracted velocity
above which the algorithm gives inaccurate results. Tests have shown that structures that move
49
less than half their spatial width are well tracked by the algorithm[105].
4.3 Pattern Matching velocity estimation
Pattern matching, block matching, or region based matching algorithms [4, 3] have been
employed extensively in the computer vision community for velocity estimation of image sequences,
and this method is frequently used to perform motion estimation for video compression. The
technique used for velocity estimation of NSTX GPI data is adapted from Wildes, et al. [159]
with some modification, and the full details of the method and tests of its performance may be
found in Munsat, et al. [105]. The general concept of the technique is to use a block-matching
algorithm to find the displacement vector that minimizes the error between a subsection of an
image and a subsection of the subsequent image for a specified error functional. In this manner, a
two-dimensional velocity field is obtained for each image in the sequence. The original algorithm
was developed by Munsat, et al. [105], but a number of upgrades and extensions have been
included as part of work performed for this thesis.
For a given image sequence I(x, y, t), the algorithm begins by segmenting the first image into
K subsections (tiles) positioned at (Xk, Yk). Then, for each tile in the image a two-dimensional
search of the subsequent image is performed that seeks to minimize the function
Ferr = FID + λFS , (4.13)
where FID is the absolute intensity difference between image subdivisions, FS represents a velocity
field smoothness constraint, and λ is a free parameter that sets the relative weighting between the
two functions. The intensity difference contribution for the kth tile is
FID =
∫tile|I(Xk, Yk, t)− I(Xk + ∆x, Yk + ∆y, t+ ∆t)|dxdy, (4.14)
where d = ∆xx+ ∆yy is the displacement vector, ∆t is the time between frames, and the integral
is over the tile area. The smoothness contribution is
FS = (1− exp(−τc2s)), (4.15)
50
where
cs =
√dvxdx
2
+dvxdy
2
+dvydx
2
+dvydy
2
, (4.16)
and the velocity derivatives are calculated at (Xk, Yk). The velocity derivative tolerance is set by
the free parameter τ . Typically, the values of λ and τ are chosen such that FS is comparable to
twice the standard deviation of I(x, y, t) multiplied by the tile area when c2s exceeds the tolerance
1/τ . The displacement vector d that minimizes the combined error yields a velocity estimate for
tile k of v(Xk, Yk, t) = d/∆t, and this minimization process is repeated for each tile and each image
to yield a velocity field v(x, y, t).
The choice of error functional described above carries with it a number of consequences.
First, the use of the absolute intensity difference effectively assumes that, to a good approximation,
brightness is conserved between frames. Second, by imposing a smoothness constraint on the
velocity field, it is assumed that the actual flow field does not have velocity gradients that exceed
∼ 1/√τ . In practice it is found that imposing a ?soft penalty? on the smoothness of the velocity
field allows for some abrupt variation, but the occurrence of unphysical or spurious vectors is
reduced. Finally, the smoothness constraint requires knowledge of the local velocity field in the
neighborhood of the kth tile. An initial guess for the velocity field may be provided in any number of
ways, but, fundamentally, accurate velocity information is not known prior to the error minimization
process. To resolve this issue, an iterative, multi-resolution technique is employed to estimate the
velocity field, where the first iteration carries out the minimization process using only the intensity
difference component. The workflow can be summarized as follows,
(1) Segment image into tiles.
(2) For each tile, find displacement that minimizes FID.
(3) Repeat step 2 for each frame.
(4) For each tile, find displacement that minimizes Ferr.
(5) Repeat step 4 for each frame
51
(6) Repeat steps 4-5 for N iterations
(7) Subdivide Tiles.
(8) Repeat steps 4-7 until final resolution is reached.
Upon completion of this procedure a two-dimensional velocity field is obtained for each frame in the
image sequence at a specified final resolution that is some fraction of the original image resolution.
4.4 The Hybrid Optical Flow and Pattern Matching Velocimetry Algorithm
(HOP-V)
The HOP-V algorithm described by Munsat, et al. [105] seeks to combine the optical
flow and pattern matching techniques to overcome some of the limitations of each. The pattern
matching algorithm uses the spatial structure of the intensity field in a neighborhood around the
chosen pixel in estimating the velocity. Often for real image data the intensity gradient is not one-
dimensional, so the pattern matching method is able to overcome the aperture problem. However,
the pattern matching technique needs an initial guess to be provided for the velocity field. The
optical flow solution can fill this need by providing a reasonable first guess to be used by the
pattern matching algorithm. The HOP-V algorithm proposes to apply the optical flow and pattern
matching techniques to image data in series to achieve the best results. Details of the performance
of this technique can be found in Munsat, et al. [105].
Chapter 5 will examine the application of the HOP-V algorithm to GPI data of an oscillatory
L-mode like state preceding the L-H transition in NSTX. In theory, one could use the optical flow
result, and some assumption on its validity, to reduce the extent and convergence time of the
pattern matching 2D error minimization. In practice, however, it is found that the application of
the optical flow and pattern matching techniques in series is extremely time consuming, and results
from the combined technique do not appear to out perform the pattern matching technique seeded
with a random velocity field. Still, a more sophisticated wedding of the two techniques could out
perform the individual application of either. Regardless, a random velocity seeding is performed
52
for the results discussed in chapter 5, and the optical flow technique is not used.
4.5 A Note On Post-Processing Techniques: a Navier-Stokes Inspired
Smoothing Algorithm
The velocity estimation routines discussed above generally assume that translation is the only
allowed transformation of the intensity. For example, the optical flow algorithm assumes constant
brightness, DI(x, t)/Dt = 0. This is a good assumption for the case of rigid body motion that
these routines were originally developed for, but this can be inappropriate for the velocity fields of
neutral fluids and plasmas. One method to make the extracted velocity fields more ”fluid-like” is to
apply a Navier-Stokes inspired iterative smoothing algorithm to the output of the above mentioned
velocity extraction routines. An adaptation of the original technique developed by Doshi and
Bors [32, 33] has been implemented, and a brief discussion is included here.
The Navier-Stokes equation for an incompressible Newtonian fluid is
∂tv + v · ∇v = −∇p+D∇2v + f (4.17)
where p is the pressure, D is the diffusion coefficient, and f is the “body forces” present in the
system. The smoothing routine attempts to approximate fluid like evolution of the velocity field in
a 3 step process iterative process. First, an anisotropic diffusion process is applied to the velocity
field. Second, velocity information is advected forward and backward in time. Finally, the original
velocity field is updated with the smoothed vector by a weighted averaging,
vi(k + 1) = (1− ε)vi(k) + εvSi (k) (4.18)
The diffusion step employs an anisotropic kernel that smooths the velocity field mainly along
the direction of the local flow, and attempts to preserve velocity gradients. This is done using
components of the local Hessian matrix of the flow,
S =
∂xxu ∂xyu
∂yxv ∂yyv
53
The diffusion is applied using a gaussian kernel with the covariance matrix S. The smoothed
velocity field is given by
vDc (k + 1) =
∑xi
vi(k) exp[−(xi − xc)TS−1c (xi − xc)]∑xi
exp[−(xi − xc)TS−1c (xi − xc)(4.19)
where vDc (k+1) is the diffused velocity at position xc for iteration k+1, Sc is the covariance matrix
computed at position xc, and vi(k) is the velocity vector at position xi in the neighborhood N(xc).
Following the diffusion, a median filter is applied to the velocity field to improve the performance
of the diffusion in the presence of outliers.
The advection process is approximated by a weighted averaging of the velocity v(x, t) and
v(x+vdt, t+dt). This is preformed both forward in time and backwards in time. For each velocity
vi(xn, ym, t) = {ui(xn, ym, t), vi(xn, ym, t)} in the vector field at time t and iteration i, the velocity
at time t+ 1 for the forward in time advection process is given by
ui+1(xn′ , ym′ , t+ 1) =wnm
2ui(xn, ym, t) + (1− wnm
2)ui(xn′ , ym′ , t)
vi+1(xn′ , ym′ , t+ 1) =wnm
2vi(xn, ym, t) + (1− wnm
2)vi(xn′ , ym′ , t) (4.20)
where the weights are given by
wnm = exp[−√
(xn′ − xc)2 + (ym′ − yc)2], (4.21)
and (xc, yc) = (xn, ym) +vi(xn, ym, t)dt. The averaging is preformed in a 2×2 grid point neighbor-
hood around (xc, yc). Admittedly, this is somewhat of a kludge, and an improvement would be to
implement a second-order accurate integration scheme for the numerical advection (e.g. See Rood
[114])
Results of the application of this Navier-Stokes post-processing to output of a pattern-
matching velocity extraction method for a synthetic sheared-flow test case can be seen in 4.1.
The velocity field smoothness constraint was removed for this test, and as a result some noticeable
velocity discontinuities exist in the resulting unsmoothed velocity field in Fig. 4.1(a). Also, signifi-
cant edge effects can be seen at the top and bottom of the frame where the structures leave the field
54
of view. After application of the Navier-Stokes smoothing the discontinuities have been suppressed,
and the edge effects have been lessened as seen in Fig. 4.1(b). Profiles for the unsmoothed vx and
vy and smoothed vx and vy can be seen in Fig. 4.1(c) and (d), and Fig. 4.1(e) and (f) respectively.
The affect of the smoothing is to decrease the overall variation in the flow, but smoothing also leads
to a slight underestimation of the peak velocity. Smoothing also pulls the average velocity up at
the edge of the view.
The results of the Navier-Stokes smoothing are quite promising, though some errors are
introduced near maxima and minima of the flow that are common to all smoothing operations. The
algorithm is an iterative adjustment of the original flow field requiring a number of calculations
to be done for each velocity vector, so application of this method to real data can be quite time
consuming. In practice it is found that the time costs of applying this post-processing to the velocity
fields is generally not worth the marginal gain in accuracy compared to the HOP-V routine with the
added smoothness constraint. Still, if a fast algorithm with an improved treatment of the advection
could be developed then this technique could be quite useful for denoising or correction of aperture
effects of velocity fields extracted by optical flow or other velocimetry techniques.
55
Figure 4.1: Pattern Matching vector fields for (a) unsmoothed and (b) smoothed. Profiles of vx andvy for unsmoothed, (c) and (d), and smoothed, (e) and (f). Red traces are the imposed velocityfield.
Chapter 5
Measurement of 2D flows in the edge and SOL preceding L-H transitions
Since its discovery more than 30 years ago [156], the high confinement mode, or H-mode,
and the associated transition from the lower confinement L-mode has been studied extensively as
evidenced by the great many experimental and theoretical reviews [149, 17, 52, 23, 63, 68, 137, 155,
49]. In this chapter, a GPI study of L-H transition dynamics on NSTX is presented. In this study,
flows and turbulence behavior in the edge of NSTX preceding the L-H transition is characterized,
and observations of oscillations between a low and high turbulence state preceding the transition
are presented. First, however, a brief overview of the L-H transition accented by observations of
L-H transitions in NSTX is presented, and a brief discussion of theories of the transition is included.
Following this, the drift wave zonal flow paradigm and associated predator-prey model of the L-
H transition are discussed in detail. Finally, the experimental observations are evaluated in the
context of the predator-prey model.
5.1 The L-H transition
With the application of auxiliary heating by neutral beams or RF waves, a transition is
often observed in magnetic confinement devices from a state of low confinement to a state of high
confinement where plasma fluctuations and transport are drastically reduced at the edge of the
device. Originally observed in a divertor tokamak [156], the improved confinement regime has been
realized on a wide array of magnetic confinement devices including divertor and limiter tokamaks,
stellarators, and even in a mirror machine with limiter biasing [119]. At the transition to H-mode
57
an abrupt drop (sub ∼100 µs) in the light emission from recycling neutrals generated by plasma
material interactions is observed, and plasma density begins to increase. These two observations
imply an increase in confinement, and provide an indicator for the transition. A transport barrier
is formed just inside the separatrix with a ∼1 cm width that reduces edge fluctuation levels and
particle, momentum, and energy transport. As a result edge pressure gradients steepen, energy
confinement times improve by a factor of ∼2, and subsequent increases in plasma β and stored
energy are seen.
Traces of operational and plasma parameters for 2 different NSTX shots are plotted in Figure
5.1. One shot, 141740, is an ohmic L-mode shot, and the other, 138692, is a neutral beam heated
H-mode. The L-H transition occurs at 0.166 s as indicated by the abrupt drop in the divertor Dα
emission. Following the transition to H-mode, the energy confinement time increases steadily, and
the density and temperature rise leading to an significant increase in stored energy. The energy
confinement time for the H-mode case is comparable to the Ohmic case, but still greater than
typical auxiliary heated L-mode energy confinement times. The difference in the L and H-mode
density and temperature profiles are illustrated by Fig. 5.2. Steep gradients in both density and
temperature are developed at the edge during H-mode indicating the presence of an edge transport
barrier. In NSTX, the H-mode density profile tends to become hollow with a peak near 85% of the
minor radius.
GPI measurements exhibit a significant drop in fluctuations at the L-H transition as seen
in Fig. 5.3. The fraction of GPI light contained in the SOL, FSOL is a clear indicator of the
transition. Average values of FSOL drop by ∼7% at the transition, and fluctuations in FSOL are
significantly reduced. This implies a significant drop in transport into the SOL and an improvement
in confinement. Raw GPI signals inside the separatrix increase in mean value at the transition,
and the normalized fluctuation level drops significantly.
A number of theories have been put forth to explain the creation of the edge transport
barrier and thus the transition to H-mode. One such class of theories argue for a change in plasma
conditions leading to the stabilization of certain instabilities at the edge as the mechanism for
58
Figure 5.1: Traces of (a) plasma current and injected neutral beam power, (b) diverter Dα light,(c) Energy confinement times, and (d) plasma stored energy. Black traces are for an ohmic L-modeshot, and red traces are for a neutral beam driven H-mode. L-H transition timing indicated bydashed red line.
producing an H-mode. Theories involving the stabilization of Ideal MHD ballooning and peeling
modes, drift resistive ballooning modes, collisional drift waves, or drift-Alven waves have been
developed, and each found varying success in predicting transitions on various machines [49, 23].
However, none of these theories were able to consistently explain experimental results across all
machines and all transition types (e.g. neutral beam, sawtooth, or pellet injection triggered).
Another class of theories that has found considerable success in describing the H-mode tran-
sition propose the suppression of turbulence and transport via E×B shearing [126, 10, 69, 148, 23].
Theoretical work showed that a shear in the E × B velocity can act to reduce the radial corre-
lation length of turbulence or de-phase the density and potential fluctuations, thus reducing the
59
Figure 5.2: Thomson scattering profiles of electron temperature (top) and electron density (bot-tom).
Figure 5.3: Traces of (a) fraction of GPI light in SOL, (b) GPI raw signal near separatrix, and (c)normalized GPI fluctuation level.
turbulence and transport. A sufficient requirement for the quenching of turbulence by flow shear
is expressed as ωE×B ≥ γMAX , where ωE×B is the shearing rate and γMAX is the maximum linear
60
growth rate [157, 16].
Experimental observations have confirmed the development of a layer of Er shear at the edge
of the plasma with a ∼1 cm width at the H-mode transition, and this shear layer is correlated
with the reduction in turbulent fluctuations, transport, and radial correlation lengths as expected
[49, 155, 17]. Furthermore, it has been shown that changes in Er precede changes in the density
and temperature profiles [53], and transitions have even been induced by biasing of the edge with
probes [145, 17]. Taken together, these observations point to the development of Er shear as the
driving mechanism for the transition.
Even still, with all the experimental and theoretical work that has been performed on the
H-mode transition, clear evidence which could determine the mechanism that creates the Er shear
layer and triggers the transition into H-mode has escaped us. A number of mechanisms have
been proposed to explain the creation of a sheared flow layer at the edge including ion orbit loss,
neoclassical poloidal flows, Stringer spinup of poloidal flow, and zonal flows driven by turbulent
Reynolds stress [49, 23]. Other theories include Bifurcation theories and phase-transition models
[23]. Very recently, many experimental studies have indicated the importance of the interaction
between Zonal flows (m=0,n=0 potential structures) and turbulence in the L-H transition [39, 40,
25, 98, 121, 153, 162, 95], and several studies suggest nonlinear transfer of energy from turbulence
to the Zonal flow may provide an H-mode trigger mechanism [153, 95, 26]. In addition, the NSTX
GPI study of flows preceding the L-H transition discussed in this chapter presents a number of
results with qualitative similarity to the limit cycle oscillation of Predator-Prey models of the L-H
transition [80, 81]. In light of these results, the following section will give a closer look at models
of edge transport barrier formation including zonal flows.
5.2 The Drift-Wave Zonal Flow Paradigm
The most complete theory of transport in magnetically confined plasma devices is the neo-
classical theory which describes the diffusion of particles, heat, and momentum due to coulomb
collisions and geometrical effects. In magnetic confinement devices, experimentally measured trans-
61
port often significantly exceeds the neoclassical predictions, thus this enhanced transport is termed
‘anomalous’ transport. Experiments have indicated that anomalous transport is due to turbulent
transport driven by plasma instabilities. The interactions between the plasma equilibrium, micro-
instabilities, and anomalous transport is encapsulated in the Drift Wave paradigm (blue boxes of
Fig. 5.4). More recent work in geophysical and planetary flows (see Busse 1994 [18]) and plasma
turbulence [30] has highlighted the importance of the interaction of turbulence with mesoscale
flows. These Zonal flows are self-generated by the turbulence, and can interact with, and even reg-
ulate, the turbulence via shearing. This motivated the shift to a Drift Wave Zonal Flow (DWZF)
paradigm that describes a self-regulating turbulence system. This section seeks to motivate each
of the interactions included in the DWZF paradigm which will be important for the Predator-Prey
model of the LH transition described in the following section.
Figure 5.4: Conceptual picture of the drift wave zonal flow paradigm. Gradients drive instabilitiesto turbulent state. Turbulence drive anomalous transport via fluctuations. Turbulence also self-generates zonal flow via Reynolds stresses. Zonal flows regulate turbulence via shear suppression.
The drift wave is a class of instabilities that tap into the free energy provided by gradients in
plasma pressure. As gradients are inherent to magnetically confined plasmas, the drift wave is one
of the most fundamental instabilities. Experimental evidence also points to these low frequency
modes as a major driver of transport in tokamaks and other magnetic confinement schemes. The
discussion of drift waves included here loosely follows that in Bellan [8]. The basic mechanism can
be illustrated by a simple example with a density gradient in the x direction, magnetic field in z,
and the plasma is taken to be collisionless. Ion temperature is assumed to be less than electron
62
temperature, so that ion pressure forces may be ignored. Electron thermal velocity is taken to be
much greater than parallel phase velocity of the perturbation (ω/kz � vTe) so that the electron
parallel momentum balance yields a Boltzmann response
ne1ne0
= −qeφ1Te
. (5.1)
Since ω � ωci and ions are assumed to be cold, the lowest-order perpendicular ion motion given
by the ion momentum balance equation is simply E×B drift,
Ui1 =−∇φ1 ×B
B2. (5.2)
Invoking quasi-neutrality gives ne1 = ni1. Then substituting Eqn. 5.1 and 5.2 into the linearized
ion continuity equation and Fourier transforming yields
ω = −kyTeeB
∂xni0ni0
= kyvde. (5.3)
Thus the presence of a pressure gradient in a magnetized plasma provides for the propagation of a
wave of coupled density and potential perturbations in the electron diamagnetic drift direction, vde.
This simplified drift wave mechanism is illustrated in Fig. 5.5. E×B circulation around potential
perturbations carries plasma from low density areas to high density areas and vice versa. The net
result is that the perturbation propagates in +y.
The inclusion of collisional drag between electrons and ions in the momentum balance equa-
tions results in a phase shift between the density and potential responses given by
ne1ne0
= −eφ1Te
−ikyvde + k2zTe/νeime
−iω + k2zTe/νeime. (5.4)
This effectively shifts the contours of the density and potential perturbations in Fig. 5.5 relative
to each other, so that the E × B motion about the potential structure reinforces the density
perturbation. Thus the wave is destabilized.
If a successive approximation of the ion perpendicular motion is used in the derivation of
collisionless drift waves sketched above then the next order is found to include the ion polarization
63
Figure 5.5: Illustration of the drift wave mechanism. The background density gradient is in the−x direction, and the magnetic field is in z. E×B circulation around potential perturbation (bluecontours) pushes high density from −x and low density from +x producing a propagation of theperturbation in +y.
drift
Ui1 =−∇φ1 ×B
B2− 1
ωciB
d
dt∇φ1 (5.5)
with d/dt being the convective derivative given by d/dt = ∂t − U · ∇. Retaining this term in
the derivation and keeping quasi-linear combinations of first order fluctuating terms yields the
Hasegawa-Mima equation [57], the simplest drift wave model that includes a nonlinear interaction.
This nonlinearity originates from the action of the convective derivative on the E×B motion. The
result is a single field equation for the nonlinear evolution of wave amplitudes
∂tφk + iωkφk =1
2
∑k=k1+k2
Nk,k1,k2φk1φk2 (5.6)
with
Nk,k1,k2 =1
1 + k2(k1 × k2 · z)(k22 − k21), (5.7)
and
ωk =ky
1 + k2. (5.8)
Here, quantities have been normalized in the following way, φ = Lneφ/ρsTe, x = x/ρs, t =
ρsωcit/Ln, and L−n 1 = ∂x lnn0. Multiplying Eqn. 5.6 by φ∗k gives an equation for the evolution of
64
wave energy
1
2∂t|φk|2 + iωk|φk|2 =
1
2
∑k=k1+k2
Nk,k1,k2φk1φk2φ∗k. (5.9)
The sum on the right describes the nonlinear three wave coupling process. This process is ubiquitous
in plasma turbulence due to the inherent nonlinearity of the convective derivative, and it describes
the spreading of turbulence in wavenumber space. Instabilities driven in some range of unstable
wavenumbers will transfer energy to other, linearly stable modes in a cascade mediated by three
wave interactions. Effectively, this process ensures that instabilities with sufficiently strong driving
will produce a turbulent spectrum of fluctuations. Many extensions to this simple picture exist
(e.g the two field Hasegawa Wakatani model [58]), but the simple models serve to illustrate the
fundamental characteristics. Taken together, the drift-wave mechanism and the three wave coupling
process describes the formation of a gradient driven turbulence state in magnetically confined
plasmas.
Next, we will look at the implications of the turbulent state on the fluxes. Neoclassical
theory considers plasma variables n, p, T , E, B, etc. to be stationary, however experiments observe
significant fluctuating components of these quantities. These fluctuations can drive transport which
is not captured by the neoclassical theory. As a demonstration of this for electrostatic fluctuations,
consider the fluid moment equation for ion momentum:
min∂tU +minU · ∇U = neE + neU×B−∇p+ R (5.10)
where R is the collisional friction term, and other quantities have their usual meanings. We take
all quantities save B to be fluctuating quantities with fluctuation A defined as A = 〈A〉+ A and 〈·〉
denotes the ensemble average. Upon dropping the kinetic terms on the left hand side, performing
an ensemble averaging of the equation, and rearranging, the quasi-linear perpendicular particle
flux, Γ⊥ = nU⊥ is found to be [115]
〈Γ⊥〉 = − 1
eB2∇〈p〉 ×B +
1
B2〈n〉〈E〉 ×B +
1
B2〈nE〉 ×B +
1
B2R×B (5.11)
The first two terms on the right hand side describe the within flux surface diamagnetic and equi-
librium E × B particle flux contributions. The third term gives the fluctuation driven E × B
65
particle flux, and the final term describes the particle flux driven by collision processes. This sim-
ple demonstration reveals that particle flux can be driven by electrostatic fluctuations if the density
and electric field perturbations are correlated. The particle flux can easily be calculated from the
perturbed density and potential found in the collisional drift wave model described above giving [8]
Γx =k2yρ
2s
1 + k2yρ2s
k2y|φ1|2
2B2
νeime
k2zTe
dn0dx
. (5.12)
This flux is proportional to the density gradient, so the net effect is an enhanced diffusion driven
by drift wave fluctuations.
The remaining interactions in the DWZF paradigm requiring motivation are the interactions
between the turbulence and the zonal flow. In the context of toroidal plasmas, the zonal flow (see
Fujisawa 2009 [47] for review) is a potential structure which is constant on a flux surface, so
it has a toroidal mode number n = 0 and poloidal mode number m = 0. It has a finite radial
wavenumber which lies between the turbulence scales, on the order of the ion gyroradius, and the
system scales described by the minor radius of the device. This scale range can be expressed as
λr ≈ 10 − 50ρi. The zonal potential structure gives rise to radial electric which drives a poloidal
flow in the plasma. Since the zonal flow is a poloidal flow that is constant on a flux surface, it will
drive a pressure perturbation due to the poloidal asymmetry of the toroidal geometry. Essentially
more plasma exists on the outside of the torus than the inside due to the geometry. This pressure
perturbation may be relieved by driving a toroidal flow, or driving an oscillating flow that couples
to a pressure perturbation. The former is the stationary or zero mean frequency zonal flow, and the
latter is the geodesic acoustic mode. In addition to the previously mentioned characteristics, the
geodesic acoustic mode is associated with an m=1 density perturbation and is predicted to oscillate
at frequency between 4-12 kHz in the NSTX edge [167]. Zonal flows have also been discovered in
neutral fluid systems, and are seen, for instance, in the atmospheres of the major planets in our
solar system (e.g. Jupiter’s stripes) [18].
Now, let’s examine the self-generation of the zonal flow by the turbulence. In a sense, this
process has already been described by the three wave coupling process described in the Hasegawa
66
Mima model (Eqn. 5.6). In 2D turbulence, the cascade of wave energy due to three wave processes
is often such that wave energy is transported to larger scales, and eventually this energy condenses
at the lowest wavenumbers. The zonal flow generation occurs by a modulational instability, a non-
local process in wavenumber space involving the coupling of two high-k drift waves into a low-k zonal
mode [30, 47]. Alternatively, the generation of large scale structured flows can be understood as
the result of work done by Reynolds stresses. Averaging the poloidal momentum balance equation
over flux surface yields the equation for evolution of the zonal flow [31, 148]
∂t〈Uθ〉 = ∂r〈uruθ〉 − µθ〈Uθ〉 (5.13)
where µθ is the poloidal flow damping rate, and 〈·〉 denotes a flux surface averaging. The first term
on the right hand side clearly represents a generation of zonal flow by the divergence of turbulent
driven momentum flux, the Reynolds stress.
The back reaction of the zonal flow on the turbulence via flow shearing can be demonstrated
by a simple dimensional argument [148]. Consider an eddy in a background sheared flow uy(x).
The shear flow will stretch the eddy by a distance ∆ys in a time τs given by
∆ys = τs∆xe∂xuy. (5.14)
The factor ∆xe∂xuy gives the difference in y velocity across the eddy. During this time, the eddy
will advect a fluid element a distance
∆xs = τsux = ∆xeτsτe
(5.15)
where ∆xe and τe are the eddy x dimension and eddy lifetime, respectively. If the shear is strong
(τs < τe) then ∆ys, the shear distortion, can exceed the turbulence coherence length, and the eddy
will lose coherence due to interaction other turbulent flows. The eddy lifetime is then set by τs,
and Eqn. 5.15 shows that the shear-wise eddy dimension is reduced. Thus the turbulence becomes
decorrelated, the correlation length across the flow is shortened, and subsequently the transport
is reduced. Furthermore, it has been shown [10] that this shear interaction can also reduce the
amplitude of the turbulent fluctuations.
67
The discussion in this section introduced the Drift Wave Zonal Flow paradigm, and provided
motivation for the various interactions illustrated in Fig. 5.4. In the next section we will examine
a theory for the L-H transition based on these interactions.
5.3 The Predator-Prey model of the L-H transition
Motivated by the possibility for Zonal Flows to regulate the turbulence amplitude and trans-
port, a simple 0D model of the L-H transition was proposed by Kim and Diamond [80, 81]. The
model system evolves equations for the Turbulence amplitude ε, the zonal flow shear V ′ZF , and
the gradient of the density profile N = −(Ln/n)∂rn. Since the zonal flow generation is directly
dependent on the turbulence amplitude, a sustained state of reduced turbulence (i.e. an H-mode)
requires the inclusion of a mean flow shear. The model incorporates this by considering a reduced
momentum balance equation given by
〈V 〉 = − 1
eBz
1
n
dpidr. (5.16)
This assumption neglects the affects of toroidal and zonal flows in the momentum balance, and
furthermore does not address the question of the generation of an Er during the H-mode. In
this regard, the model serves only to demonstrate how Zonal Flows may affect a transition. A
more sophisticated treatment of the momentum balance is required to fully understand the edge
transport barrier formation. It should be kept in mind that the model is only a 0D model, and
thus only a qualitative description of the L-H transition dynamics is offered.
The model equations are as follows:
∂tε = εN − a1ε2 − a2V′2ε− a3V
′2ZF ε, (5.17)
∂tV′ZF = b1
εV ′ZF1 + b2V
′2− b3V ′ZF , (5.18)
∂N = −c1εN − c2N +Q, (5.19)
V ′ = dN2. (5.20)
The evolution of the turbulence amplitude (Eqn. 5.17) is given by, from left to right, linear
68
growth driven by the density gradient, nonlinear saturation, suppression by mean flow shear, and
suppression by zonal flow shear. The evolution of the zonal flow shear (Eqn. 5.18) is given by
turbulent generation of zonal flow by Reynolds stress, and zonal flow damping. The factor 1/(1 +
b2V′2) represents the suppression of zonal flow drive by mean flow shear as described in Kim and
Diamond [80]. The density evolution (Eqn. 5.19) is affected by turbulent diffusion, neoclassical
diffusion, and auxiliary heating input. Finally, Eqn. 5.20 closes the system of equations by using
Eqn. 5.16 to express the mean flow shear in terms of the diamagnetic flow. In addition to the
simplified motivations provided in the previous section, this set of equations and the interactions
included in them can be justified by a wave kinetic treatment of the problem [94].
The evolution of the system toward an H-mode state can be studied by varying the input
power Q as a function of time. The evolution exhibits three main phases. First, at low input
power drift waves are driven unstable, and ε grows linearly. Second, the zonal flow begins to grow
when the turbulent drive overcomes the damping. As the flow grows it begins to interact with the
turbulence, and a limit cycle phase is observed. Finally, at high input power the density profile
grows, driving a mean flow shear, and the turbulence is suppressed as the system transitions into
an H-mode like state.
The limit cycle dynamics and the transition to H-mode can be described by conceptual picture
given by Fig. 5.6. The limit cycle evolves in the following way. Gradient driven instabilities drive
growth of turbulent fluctuation level. When the zonal flow damping is overcome by the turbulent
drive, zonal flow amplitude begins to grow. Once the zonal flow is strong enough it begins to
suppress the turbulence via shearing. As the turbulence decays, the zonal flow drive decreases
accordingly. Eventually the zonal flows are damped away, and the turbulence begins to grow again.
A transition to a quiescent, H-mode state is possible if the input power is strong enough to develop
a mean flow shear during the period of reduced turbulence.
Many experimental studies in recent history have observed qualitatively similar dynamics
to this limit cycle oscillation [39, 40, 25, 98, 121, 153, 162, 95]. Most notably, perhaps, is the
recent work by Schmitz which details the 1D plus time dynamics of the turbulence amplitude and
69
Figure 5.6: Conceptual picture of limit cycle oscillation process. System evolves between a highturbulence, low flow state and a high flow, low turbulence state. If the input power is sufficientto steepen the density gradient during a period of suppressed turbulence, then a mean flow sheardevelops and a transition to H-mode is observed.
radial electric field at the edge using Doppler Backscattering. The promising similarity between the
experimental results and the simple model has motivated the extension of the 0D model to a 1D
plus time dynamical model [101]. In the next section, we’ll examine a set of experiments on NSTX
that exhibit qualitative similarity to limit cycle oscillations described by the simple 0D model.
5.4 GPI Observations of Flows preceding L-H transitions
The interaction between drift wave turbulence and zonal flows make it clear that flow orga-
nization can play an important role in magnetic confinement devices. Recently, Zweben, et al.
[167] reported on the observation of ’quiet periods’ in the National Spherical Torus Experiment
(NSTX)[77] edge prior to the L-H transition. These quiet periods were observed with the GPI
diagnostic, and they found that the GPI Dα light emission during these periods resembled that of
H-mode. The quiet periods occurred at a frequency of ∼3 kHz, and were correlated with changes in
the direction of the poloidal flow. They also found that the dimensionless poloidal shearing values
70
were correlated with the quiet periods.
In this chapter, we analyze the same database of shots as Zweben, et al. [167]. However,
we use the hybrid optical flow and pattern matching velocimetry (HOP-V) technique [105] to derive
2D velocity fields using GPI data from the NSTX edge region. This technique yields time resolved
velocity fields with a spatial resolution of ∼1.5 cm, at 16×16 spatial points, and time resolution of
3.5 µs. Using the derived velocity fields and GPI data we report on three main results: (1) detailed
characterization of space and time evolution of zonal flow features, (2) calculation of turbulence
shear stress and its radial profiles, (3) relationship between zonal flow and turbulent bursts.
5.5 Time averaged-flow profiles
GPI data for each of the shots discussed in this chapter (#135021-23, #135041-46) have an
image size of 64×64 pixels, and the HOP-V algorithm calculates velocities for tiles of 4×4 pixels.
This produces a 16×16 point velocity field at the temporal resolution of the image sequence as
illustrated by Fig. 5.7a. In the Figure the time averaged velocity field for a period of approxi-
mately 10 milliseconds during L-mode operation is superimposed upon time averaged contours of
GPI intensity. The GPI intensity falls off near the top and bottom of the view due to optical
vignetting and the limited size of the GPI emission cloud. The velocity field has been cropped to
an approximately 10.4 cm × 14.8 cm (7×10 points) section in the center of the GPI field of view
to avoid the region of reduced signal, and to avoid edge effects introduced by structures that are
not tracked as they travel out of the GPI view. Also, due to the emission profile of the GPI gas
puff, the intensity fluctuations outside this boxed region are not high enough to consistently yield
strong correlations for the HOP-V algorithm, and as a result velocity data from these regions is
often not reliable. All analysis (poloidal averaging, etc.) in this work will be limited to the boxed
region of Fig. 5.7a. The approximate location of the separatrix (dot-dashed line) is determined
by the NSTX-standard model LDRFIT, and is uncertain up to ±2 cm. The exact location of the
separatrix is not, however, a critical aspect of the analysis. The radial position is displayed relative
to the separatrix position.
71
Figure 5.7: (a) Time averaged velocity field superimposed on time averaged GPI intensity contours(gray contours). The separatrix is indicated by the dot-dashed line, and the cropped field of viewis indicated by the dashed box. The maximum time-averaged velocity magnitude for the croppedregion is 1.5 km/s, while the maximum instantaneous velocity magnitude is 7.4 km/s. The +xdirection is radially outward. (b) radial profiles of time averaged poloidal flow several shots for 10.5µs preceding the L-H transition. RMS values of the fluctuating velocity are shown for shot 135042.
(a)
(b)
radial
poloidal
Figure 5.7 serves to illustrate some general characteristics of the observed flow. In movies of
the GPI signal (viewable in the multimedia section of Zweben, et al. [167]), intensity structures
(“blobs”) are typically seen to move with positive y-velocity (upward, in the electron diamagnetic
direction) in the inner region (x ≈ -5-0 cm), while slowly moving radially outward. As the structures
cross the separatrix and move into the scrape off layer (SOL), their poloidal velocity reverses
direction, and they move downward (in the ion diamagnetic direction). Figure 5.7b shows a similar
72
pattern with negative average velocities outside the separatrix, while flow reversals tend to average
to zero near the separatrix. Average velocities are on the order of 1.5 km/s poloidally and .5
km/s radially, and instantaneous velocities can be significantly higher, as discussed below. Several
centimeters inside the separatrix, the average poloidal velocities are positive. Thus a non-zero time-
averaged poloidal flow sheared is present at the separatrix. A similar pattern in the time-averaged
flow field is seen in shots 135041-135046, and similarities can be seen in Figure 5.7b. In addition
most all shots show a clear downward average flow outside the separatrix. ‘Error bars’ in Figure
5.7b indicate RMS levels of the fluctuating poloidal velocity for shot 135042, so it can be seen that
the velocity fluctuations are on the order of the mean flow values.
5.6 turbulence “Quiet-Periods” and quasi-periodic velocity fluctuations
Figure 5.8: Normalized GPI intensity (top), poloidal velocity (middle), and radial velocity (bottom)∼1 cm inside the separatrix. All traces represent poloidally averaged quantities from shot 135042.
To illustrate temporal behavior, time traces for normalized GPI intensity, poloidal velocity,
and radial velocity are shown in Fig. 5.8 (shot 135042, t ≈ 0.241-0.243 s). In the Figure, each
trace represents a poloidal average, localized radially approximately 1 cm inside the separatrix,
73
which is near the time-averaged GPI intensity maximum. Gray outlines of Vp and Vr indicate the
level of uncertainty in these values assuming random, statistically independent uncertainties in the
individual velocity measurements of 0.5 km/s (∼0.5 pix/frame). In the averages shown, the level
of uncertainty is ∼0.15 km/s.
The GPI intensity is normalized by subtracting from each pixel its mean value taken over the
full 17000-frame exposure and then dividing by the mean, so a value of 1.0 indicates a fluctuation
which is on the order of the mean value. Typically, poloidal velocities are seen to range between
±10 km/s. Radial velocities are generally on the order of a few kilometers per second, and can
approach the minimum threshold of measurable velocities of ∼0.5 km/s. As mentioned previously,
the poloidal velocity changes from being predominantly positive inside the separatrix to negative
in the SOL. Near the separatrix, the poloidal flow is seen to fluctuate around zero with periodic
reversals.
Time traces for the GPI intensity and poloidal flow quantities, like those in Fig. 5.8, of all
shots examined here exhibit distinct quasi-periodic oscillations/bursts at a dominant frequency of
3 kHz. To compare the poloidal flow oscillation with the behavior of the turbulence, we define The
quantity Fsol as the fraction of the GPI Dα intensity contained in the SOL to total image intensity,
and a time trace of Fsol is plotted in Fig. 5.9a along with poloidal velocity traces at three radial
locations (chosen inside, at, and outside the separatrix, shown in Fig. 5.9b,c,d respectively). A
spike in Fsol corresponds to intensity structures (blobs) passing into the SOL. In this way Fsol can
be used as a proxy for turbulent transport at the edge. Periods of low Fsol indicate little activity
in the SOL and no or few occurrences of blobs, similar to GPI observations of H-mode plasmas.
These periods are described as “Quiet-periods” [167], and are indicated in Fig. 5.9 by the vertical
gray bars. Periods of high Fsol indicate increased SOL turbulence activity and frequent creation
and ejection of blobs.
Low frequency components of Fsol between 0-1 kHz have been removed from the filtered
Fsol trace in Fig. 5.9a. These frequencies are dominated by the slow rise and fall of the total
GPI emission due to the dynamics of the gas puffing. The gray bars indicate where the filtered
74
Figure 5.9: Traces from shot 135042 of Fsol filtered (a), and poloidally averaged poloidal velocityinside (b), at (c), and outside (d) the separatrix. Fsol filtered has had the low frequency componentsabove zero frequency and below 1 kHz removed. Gray bars indicate time periods where Fsol filteredis below 0.16.
75
Fsol is below 0.16 (arbitrarily chosen as a rough guide to distinguish “high” vs. “low”). Periodic
behavior in Fsol is easily discernible; the large scale oscillations between t = 0.241− 0.243 sec have
a frequency of ∼3 kHz. However, this periodicity in Fsol is somewhat irregular and appears to be
intermittent.
The intermittency of this ∼3 kHz feature is demonstrated by the spectrogram of Fsol shown in
Fig. 5.10. The spectrogram covers the entire recorded L-mode portion of shot 135042 with a window
of 256 time points (∼0.9 ms), giving roughly 1.1 kHz resolution. In constructing the spectrogram, a
sliding fast-Fourier-transform is used with Hanning windowing and 75% overlap. The spectrogram
shows a dominant mode that appears sporadically near 3 kHz. This mode fluctuates between 2-4
kHz, and tends to have relatively stationary durations of ∼1-2 ms. These characteristics are directly
apparent in Fig. 5.9a as well. Periodic bursts in Fsol and poloidal velocity at about 2.5 kHz start
near 0.2410 s and are clearly observable until 0.2420 s, then after 0.2425 s this periodicity revives.
During these time periods, Large scale fluctuations of the poloidal velocity appear to be about 180◦
out of phase with Fsol.
Figure 5.10: Spectrogram of Fsol for the L-mode portion (t ≈ 0.215−0.245) of shot 135042 plottedwith a linear color scale.
Shot 135021, and 135041-44 each have a feature in the spectrogram of Fsol with similar
qualities to shot 135042. This feature is generally centered between 2 and 5 kHz, and fluctuates
within a ∼2 kHz range around this center value. The feature is stationary for durations of 1-2
ms such as with shot 135042. The other shots (135022-23, and 135045-46) show a similar feature,
76
however it is less distinct in these shots and often does not persist through the entire shot.
The measured poloidal velocities shown in Fig. 5.9 also exhibit strong correlation with Fsol.
Periods of low Fsol (quiet periods) are correlated with positive peaks in poloidal velocity inside
the separatrix, and peaks in Fsol correspond to periods of nearly zero flow. Outside the separatrix
the reverse is true, with poloidal velocities tending toward zero during quiet periods and peaking
negative during turbulent bursts. The correlation coefficient at each of the three locations chosen
between Fsol and Vp for the time period shown is ∼0.6.
Correlation values for HOP-V tiles between successive images (which measure the “tracking
quality” of the velocimetry algorithm) for velocity vectors in the SOL tend to decrease during quiet
periods, but a majority of the vectors remain above an 80% correlation threshold. Therefore, it
appears that that the SOL velocities are not simply an artifact of low GPI signal in the SOL.
Reversals in poloidal velocity can be seen at the separatrix, with positive velocities observed
during quiet periods, switching sign as Fsol rises. The correlation between these quantities suggests
a relationship between the poloidal flow and the turbulent bursts. This may indicate that some
feature of poloidal flows acts to control the turbulence, or that the flow is generated by the release
of blobs. A causal relationship is, however, unclear at this point, and begs further inquiry.
5.7 Spatial Structure of 3 kHz mode
The spatial structure of the ∼3 kHz mode may be analyzed by first applying a bandpass
filter to the GPI signal around this dominant frequency. For t ≈ 0.241−0.243 s of shot 135042, the
filter is used to select a single mode near 2.4 kHz, corresponding to the frequency of peak spectral
power for GPI intensity ∼1 cm inside the separatrix. With the applied filter, a rotating mode is
clearly observed which exhibits long poloidal wavelength and radial size greater than 4 cm. The
mode appears in the images as an oscillating band centered 0.5 cm inside the separatrix with a
radial width of ∼2 cm that spans the full GPI view in the poloidal direction.
An alternative characterization of the spatial structure of this mode is illustrated by the
X-Y phase maps shown in Fig. 5.11. Figure 5.11a(5.11b) shows the X-Y map of the cross-phase
77
Figure 5.11: Map of phase differences of ∼2.4 kHz fluctuations for: (a) GPI signal with referenceGPI signal, (b) vpol with reference vpol, (c) GPI signal with vpol. Reference signals are at x ≈ 0cm,y ≈ 11 cm. Plots (a) and (c) have had the phase discontinuity remapped so that the contour plotsappear smooth. Plots cover the time range t ≈ 0.241 − 0.243 ms of shot 135042. The dashed-lineindicates the separatrix.
calculated from the cross-spectral density function of raw GPI signal (poloidal velocity) at each
point with a reference signal. Figure 5.11c shows the cross-phase of GPI signal with poloidal
velocity. The GPI fluctuations (Fig. 5.11a) undergo a phase shift of nearly π between locations
inside and outside the separatrix. However, in the poloidal direction at x ≈ −1 cm the phase
changes approximately linearly with position, and over the full GPI range it changes by ∼ π2 . The
poloidal wavenumber, kp can be obtained by a linear fit to the phase shift vs. y, resulting in
kp = 7.7 m−1 at x ≈ −1 cm. The poloidal velocity fluctuations (Fig. 5.11b) at x ≈ −1 cm have
a significantly longer (but not infinite) wavelength in the poloidal direction, with kp = 3.4 m−1.
The phase of the poloidal velocity also does not show the strong radial dependence seen in the GPI
intensity.
Figure 5.12 plots the phase of the ∼3 kHz oscilations versus radius and poloidal position for
periods of shots 135042-135045 where a rotating mode is visible in the bandpass filtered data. For
the radial dependence, the data is averaged poloidally before the phase is calculated, and the phase
versus y is taken ∼1 cm inside the separatrix. The phase of the GPI signal for these periods changes
rapidly from x=-2 to x=0 cm, while in the poloidal direction the phase has an approximately linear
dependence on position. From this linear dependence it can be seen that the poloidal wavelengths
78
Figure 5.12: Phase of ∼3 kHz mode plotted vs spatial cordinates for GPI signal and poloidalvelocity for periods of shots 135042-135045 with “rotating mode” visible in bandpass filtered. Timeperiods are 512 frames (1.8 msec) and begin at: 135042(black) t=0.241, 135043(gold) t=0.238,135044(green) t=0.230, 135045(red) t=0.236 s.
79
are long, with poloidal wavenumbers on the order of 10 m−1. As a function of radius, the phase of
the poloidal velocity remains small, but becomes more negative with increasing radius. This means
that the 3 kHz flow oscillations in the SOL lead the flow inside the separatrix, and this lag time
is ∼15 µs. For the poloidal direction, wavenumbers for the poloidal velocity are typically smaller
than the GPI signal by about a factor of 2.
Poloidal correlation lengths for the ∼3 kHz feature can be estimated from the coherence
function between two signals. Figure 5.13 shows the coherence and phase versus frequency for
poloidal velocity signals of shots 135042-44. The coherence function for two points separated by
∼12 cm in the poloidal direction is calculated from the cross-spectral and autospectral density
functions as outlined in section 11.6 of Bendat and Piersol [9]. The cross-spectral and autospectral
density functions are estimated by ensemble averaging estimates from 14 time blocks of 512 time
points (1.8 msec) each. Each shot shows a mode near 2.8 kHz that maintains high coherence and
small phase difference over ∼12 cm. All shots analyzed exhibit a peak in the coherence function in
the range of 2-4 kHz with phase differences less than π4 .
Figure 5.14 shows the coherence as a function of poloidal separation for poloidal velocity
signals. Coherence functions are estimated for increasing poloidal separations, and the Figure
compares the coherence of the 2.8 kHz mode to the average coherence value for frequencies between
20-30 kHz. The 2.8 kHz mode has a correlation length of 56 cm which is significantly longer than
the ∼4 cm correlation length for the 20-30 kHz band. Figures 5.14 and 5.13 suggests that this
poloidal flow oscillation near 3 kHz is a large scale, spatially coherent oscillation.
Figure 5.15 illustrates the radial dependence of the poloidal velocity power spectrum. The
spectra for the mean-subtracted poloidal velocity between t ≈ 0.240 − 0.243 ms are first norm-
squared, and then averaged poloidally. The figure shows the spectrum at three radial locations
at x=-3.9,0.6, and 3.5 cm. A strong feature near 2.5 kHz is visible at the same frequency in the
spectra at -3.9 and 0.6 cm, suggesting that the frequency of the mode does not depend on radius
and hence temperature profiles. In addition, This feature in the poloidal velocity spectra is seen to
extend a few centimeters into the SOL, but at 3.5 cm the feature is significantly reduced.
80
Figure 5.13: Coherence and phase plots versus frequency for poloidal velocity signals separated by∼12 cm poloidally at ∼1 cm inside the separatrix.
Figure 5.14: Plot shows coherence of poloidal velocity vs. y separation for 2.8 kHz mode (�) andbackground turbulence between 20-30 kHz (4) for poloidal velocity of shot 135042. Velocities aremeasured at ∼1 cm inside the separatrix. Solid lines are fits to exponential decays with correlationlengths of 56 cm for the 2.8 kHz mode and 4 cm for the background turbulence.
81
Figure 5.15: Autopower spectra of poloidal velocity at 3 radial positions for t ≈ 0.240− 0.243 s ofshot 135042. The feature at 2.5 kHz does not appear to shift in frequency.
In summary, a strong feature near 3 kHz is observed in both the GPI and poloidal velocity
spectra, and this feature exhibits a long poloidal wavelength in each case. In the radial direction, the
3 kHz flow oscillations outside the separatrix have a small negative phase shift relative to oscillations
inside the separatrix. This phase shift indicates that the 3 kHz flow in the SOL slightly leads the
flow inside the separatrix. Poloidal velocity fluctuations at 3 kHz also have a long correlation
length that is an order of magnitude greater than the turbulence correlation length. In addition,
the poloidal velocity fluctuations show no frequency variation between the separatrix and the inside
edge of the view (x ≈ −4 − 0), but the amplitude of the fluctuation decreases significantly inside
the SOL.
5.8 Flow shear and Reynolds stress calculations
Turbulent flow properties, such as the shearing rate, 〈dVp/dr〉, and the Reynolds stress,
〈δVpδVr〉, can be derived from the two dimensional flow field produced by the HOP-V algorithm.
Here, δV = V − 〈V 〉 and 〈· · · 〉 denotes a poloidal average unless otherwise noted. These quantities
are important to the evolution of the turbulent system, and may partly control the strength of
the turbulence and the transport of scalar quantities. The radial derivative of the Reynolds stress
82
term is believed to drive the production of zonal flows via nonlinear energy transfer from drift wave
turbulence [31], thus leading to the suppression of drift wave turbulence and transport. Shearing of
poloidal velocity is also thought to control turbulent transport via vortex stretching [10, 148, 30].
To explore these relationships, Fsol is again plotted with poloidal velocity in Fig. 5.16, but
also included here are the shearing rate and the turbulent Reynolds stress derived from the velocity
field. Example traces of the Reynolds stress and poloidal velocity are taken ∼1 cm inside the
separatrix to illustrate the time dependence of these quantities. Due to the shearing of eddies, one
expects that flow shear localized to the area near the separatrix might suppress transport into the
SOL [148], thus Fig. 5.16 plots the shearing rate at the location of the separatrix. The plot covers
600 frames (∼2 ms) of shot 135042 approximately 3 ms before the L-H transition. In addition
to the correlated behavior of poloidal velocity with Fsol as described above, peaks in Fsol are
correlated with periods when the shearing rate is near zero, and quiet periods are coincident with
negative spikes in the shearing rate. For the time period shown, Fsol and the shearing rate are
well correlated with a correlation coefficient of 0.53. This is consistent with a scenario in which the
turbulence level is limited by the local shearing rate, as seen in recent simulations of the NSTX
edge (Figures 4 and 7 of Reference) [116]. Almost all of the shots analyzed have short periods of ∼2
ms where the correlation coefficient between Fsol and the shearing rate is greater than 0.4. Over
longer time periods, however, the correlation values are much weaker.
To better understand changes in the poloidal flow profile during the quiet-periods, we employ
a conditional averaging technique to produce averaged radial profiles of the poloidal flow during
periods of low Fsol and high Fsol. Figure 5.17 illustrates the differences in the conditionally
averaged poloidal flow profiles for ∼30 ms of shot 135042. In Fig. 5.17a the dashed(solid) line
is the poloidal flow profile averaged over times when Fsol is above(below) its mean value. Shots
#135021-23 and #135041-45 all show a shift in the mean flow profile similar to Fig. 5.17a. The
shape of the profile doesn’t appear to change significantly, indicating that the poloidal flow shear is
similar in both cases. However, the entire profile is seen to shift ∼1 km/s in the positive direction
during periods of low Fsol. The “error bars” in figure 5.17a indicate ±1 standard deviation for
83
Figure 5.16: Traces from shot 135042 of Fsol, poloidal velocity (km/s), shearing rate (MHz), andthe Reynolds stress (km2/s2). Poloidal velocity and Reynolds stress are taken at ∼1 cm inside theseparatrix, and the shearing rate is taken at the separatrix. Traces have been smoothed with a 3point boxcar average.
the low Fsol average. These indicate the level of fluctuation about the mean value, and do not
necessarily correspond to an uncertainty of the measured value. Also, The standard deviations
(Fig. 5.17b) inside the separatrix are ∼30-50% greater during these periods. Therefore, during
quiet periods we observe the mean flow becoming more positive, and, inside the separatrix, the
fluctuations in the flow have a greater amplitude. The skewness profiles (Fig. 5.17c) typically
do not show a significant difference during quiet periods, but some general characteristics may be
observed. Inside the separatrix the skewness is positive, while outside the separatrix it becomes
negative. Thus, inside the separatrix the distribution of measured poloidal velocities has a longer
tail in the direction of more positive velocities, while the more negative tail of the distribution
is longer outside the separatrix. This is consistent with our interpretation of Fig. 5.9 where we
observed strong positive peaks in poloidal flow velocity during quiet times and strong negative
peaks during bursts.
84
Figure 5.17: Poloidal velocity profiles for ∼30 ms of shot 135042 are plotted in (a). The averagesare over times where Fsol is greater than its mean value (dashed - bursty) and less than its meanvalue (solid - quiet). “Error bars” indicate ±1 σ(standard deviation) about the average value forthe low Fsol case, and indicate the level of fluctuation. Standard deviations are shown in (b), andskewness is shown in (c).
Recent SOLT (“Scrape-Off Layer Turbulence”) simulations [116] using parameters consistent
with NSTX L-mode discharges found a parameter regime with large intermittent spikes of particle
transport accompanied by quiescent periods of low particle transport. It was found that during the
quiescent periods the shearing rate had a negative value, but crosses zero shear during bursts. For
qualitative comparison, we include traces of Fsol, poloidal velocity, shearing rate (ξ), and Reynolds
shear stresses from a SOLT simulation in the bursty regime (Fig. 5.18). Similar to figure 5.16, we
see periodic bursts of Fsol at a frequency of 3.6 kHz accompanied by periods of low Fsol, quiet
periods. Fsol exhibits strong correlation with poloidal velocity, and bursts are well correlated with
the shearing rate, ξ approaching zero.
As described above, the Reynolds stress and it’s radial gradient play an important role in
the poloidal momentum balance, and regions of finite radial gradient of turbulent reynolds stress
can generate mean poloidal flow. Profiles of the Reynolds stress can be optioned from the 2-D flow
fields produced by the HOP-V algorithm. Here, Figure 5.19 plots radial profiles of the Reynolds
shear stresses for four shots averaged over 10 ms preceding the L-H transition. Interestingly, each
85
Figure 5.18: Traces of Fsol, poloidal velocity (km/s), shearing rate (ξ), and the Reynolds shearstress. Traces are from a SOLT turbulence simulation illustrating the bursty regime. The frequencyof bursts is 3.6 kHz.
profile has a local maximum near x = −1 cm, negative slope between -1 cm and 2 cm, and a section
of positive slope between -2 cm and -1 cm. The region of positive slope inside the separatrix is
coincident with the region of poloidal flow shear seen in Fig. 5.7. Error bars have been added to
the profile of shot 135042 assuming random, statistically independent uncertainties in the velocity
measurements of 0.5 km/s (∼0.5 pix/frame). This result suggests that turbulent reynolds stress is
acting to drive a net positive poloidal flow between x = −2 and x = −1 cm inside the separatrix,
and a negative poloidal flow in the SOL near x = +2 cm.
This conclusion should be taken with some skepticism, though. First, the above measure-
ments employ both an averaging over the limited GPI poloidal view and a time averaging procedure.
This assumes Taylor’s hypothesis, i.e turbulent fluctuations are assumed to be frozen into the flow
and swept past the measurement location so that space and time may be interchanged. Ideally,
one would conduct a “zonal” average of the Reynolds stress over a flux surface. Obviously this is
prohibitively difficult given diagnostic constraints, but it is not clear that the averaging performed
86
produces an adequate approximation of the flux surface average. Secondly, validation of Reynolds
stress from the HOP-V technique has not been performed. Comparison with probe measurements
of the Reynolds stress is an obvious path toward validation, but probe measurements, which are
very difficult to make in the edge of the Tokamak, are not available here. Furthermore, preliminary
results of work done on the linear plasma device CSDX to compare probe and imaging measure-
ments suggested that Reynolds stress measurements did not agree. This is still an open question
and an area of interest for future work.
Figure 5.19: Radial profiles of the Reynolds shear stress averaged over the poloidal direction and10 ms directly preceding the L-H transition for 4 shots. Error bars indicate estimated level ofuncertainty assuming uncertainties in the velocity measurments of 0.5 km/s.
5.9 Quiet-periods as Limit Cycle Oscillations
The behavior seen in Fig. 5.9 is in many ways qualitatively similar to the Limit Cycle
Oscillation (LCO) of predator-prey model of the L-H transition. The turbulence activity is seen to
be well correlated with the changes in poloidal flow, and the phase lag between the two quantities
is 180◦. It is important to note that the exact phase relationship is sensitive to the definition of
“turbulence amplitude” used. Here we use FSOL as a proxy for the turbulent activity. The results
presented above can be used to further evaluate the similarity with the predator-prey model. Of
87
particular interest are two questions: Is the observed 3 kHz flow a low frequency zonal flow or
GAM? Does the relationship between the turbulence and shear support a flow shear suppression
scenario? This section will attempt to address these questions.
As mentioned previously, zonal flows have a few identifying characteristics: n = m = 0
potential (flow) structure, and finite radial wavenumber that lies between the turbulence scales
and the system size described by the minor radius. This can be expressed as λr ≈ 10 − 50ρi.
Additionally, GAMs may be identified by their m=1 density perturbation and are predicted to
have a frequency of between 4-12 kHz in the NSTX edge [167]. The poloidal structure of the mode
can be inferred from Fig. 5.12 and compared with expectations for a zonal flow. The linear trend
of the 3 kHz poloidal velocity phase versus poloidal position indicates that the poloidal wavelength
of the mode is ∼ 1.2 m. While this is quite long, and indeed much longer than the length of the
GPI view, it is inconsistent with an m=0 mode. However, the pattern matching velocity algorithm
tracks changes in density contours, thus, if the 3 kHz feature is a GAM, the phase velocity of
the m=1 density perturbation could impact the poloidal structure of the flow measured by the
velocimetry algorithm.
The radial structure of the 3 kHz feature is plotted in Fig. 5.20 and Fig. 5.12. Two
observations can be made. First, The amplitude of the 3 kHz feature is non-zero in the SOL. This
would appear to be inconsistent with the zonal flow picture, because it is expected that zonal flow
would be heavily damped in this region. Furthermore, the radial profile of the amplitude and the
phase of the mode versus radius indicates that the radial scale of the mode is comparable to the
minor radius. Again, this is inconsistent with what is expected for a zonal flow.
The relationship between the shear and the turbulence was explored by Fig. 5.16 and Fig.
5.17. It was found that for some time periods a strong correlation between the shear and the
turbulence can be found (e.g. traces shown in Fig. 5.16), but over longer time periods, like that
considered in the conditional averaging of Fig. 5.17, the connection between the shear and the
turbulence is unclear. This relationship was examined further in [167], and similar results to those
discussed here were reported (i.e. some correlation between shear and turbulence observed). It
88
Figure 5.20: Radial profiles of the average poloidal velocity, and the amplitude of the 3 kHz flowfeature during the quiet period oscillations.
is possible, and indeed likely, that a more complex relationship exists between the shear and the
turbulence. Based on the simple dimensional argument of shear suppression presented above, one
would expect that turbulence would be affected when S = |dV/dr|(Lr/Lp)τ > 1. This would be
consistent with results presented in [167], however a detailed statistical analysis showing a significant
shift in turbulence amplitude or FSOL when the shear criterion is met was not presented.
In conclusion, a number of inconsistencies with the DWZF picture and the observations of
the quiet period oscillations exist despite good qualitative similarity on the surface. Still, the sim-
ilarities motivate further study. A future study would ideally be able to resolve the full 1D plus
time dynamics of the potential and turbulence present in the edge region. Also, long range corre-
lations or toroidally and poloidally separated potential measurements are needed to unequivocally
identify the zonal flow. Beyond this, measuring the nonlinear transfer of turbulent energy into the
zonal flow, and decoupling the mean shear and zonal shear effects on turbulence in experimental
observations are important pieces of understanding the L-H transition dynamics. Additionally, a
better understanding of the physical mechanism leading to the edge radial electric field well during
the transition is needed. As mentioned previously, experimental studies have made progress on
these points in the past several years [153, 95, 26], and theoretical work on an improved 1D model
89
of the L-H transition based on the initial predator-prey study is in progress [101].
Chapter 6
Precursor Fluctuations During Small ELMs in NSTX
The achievement of a high confinement mode, or H-mode, in ITER is widely considered
necessary to reach the operational goals of the project: namely, a burning plasma with a ratio
of fusion power generated to input power of 10 [37, 154]. The improved confinement in H-mode
operation is partly attributed to the development of an edge transport barrier (ETB) that restricts
the transport of particles and heat into the scrape-off layer (SOL), thus reducing losses to the wall
along open field lines [49]. The ETB then builds up a region of steep gradients in temperature and
density near the plasma edge known as the H-mode pedestal. Gradients in the edge pressure or
edge current profiles provide a source of free energy for a number of disruptive instabilities known as
edge localized modes, or ELMs [166]. ELMs are often explosive events that eject significant energy
into the SOL; upon contacting the plasma facing components, these energy bursts can exceed the
heat load limits of the material components and cause significant damage. Current projections
predict that ITER components will not be able to tolerate the impulsive energy densities of the
largest ELM events [165]. Therefore, it is clear that a thorough understanding of ELM physics and
a strategy for avoiding and/or mitigating ELMs is necessary for next-step fusion devices to succeed.
This chapter details a GPI study of edge localized mode (ELM) dynamics of the growth and
collapse period, and the characterization of a coherent precursor oscillation is presented. First, a
brief overview of ELM physics and observations is presented.
91
6.1 MHD stability
MHD instabilities are primary candidates for the driving mechanism behind ELMs because
of the ELM’s global impact on the plasma equilibrium, magnetic fluctuation signature, and short
timescales for ELM growth. To study the stability of an equilibrium in ideal MHD, the linearized
MHD momentum equation can be recast as an equation for the displacement of the plasma from
equilibrium
ρ∂2
∂t2ξ = F(ξ), (6.1)
where F(ξ) is the linearized force operator
F(ξ) = ∇(γp0∇ · ξ + ξ · ∇p0)−B1 × J0 −B0 × (∇×B1). (6.2)
The perturbed magnetic field is given by the frozen flux condition B1 = ∇× (ξ×B0). The change
in potential energy of the system is then given by
δW = −1
2
∫ξ · F(ξ)d3x. (6.3)
If δW > 0 for all displacements ξ then the plasma is stable, but the plasma is unstable if δW < 0
for any possible displacement. The displacement may be solved for by either solving the eigenmode
equation
ρω2nξn = −F(ξn), (6.4)
or by minimizing the Lagrangian
L = δK − δW, (6.5)
where δK =∫
12ρ|ξ|
2d3x. In practice this is exceedingly difficult to do analytically, and often is
quite difficult even numerically.
The change in potential energy δW can be expressed as the sum of two distinct contributions
δW = δWF + δWS , (6.6)
92
where δWF and δWS represent the contribution from the plasma volume, and surface. After
performing some non-trivial manipulations [8], these contributions can be expressed as
δWF =1
2µ0
∫Pγµ0p0(∇ · ξ)2 +B2
1⊥ +B20 [2ξ⊥ · κ + (∇ · ξ⊥)]2
−B0 · (B1⊥ × ξ⊥)µ0J0‖
B0− 2µ0(ξ⊥ · ∇p0)(ξ⊥ · κ)d3r, (6.7)
δWS =1
2µ0
∫S
(µ0p1 + B0 ·B1)ξ⊥ · ds, (6.8)
where the subscript ⊥,‖ indicates components perpendicular and parallel to the magnetic field,
and κ = b0 · ∇b0 is the local curvature of the equilibrium field. This form is useful for illustrating
the meaning of the various terms, though somewhat cumbersome to work with. The first three
terms are strictly positive, and so are stabilizing. The first term gives the stabilizing effect of
compressing a plasma with equilibrium pressure p0. The second and third terms are stabilizing
effects due an increase in magnetic field strength and the compression or bending of field lines. The
final two terms may be negative, and thus allow for instability. The fourth term is destabilizing
when a plasma displacement generates a magnetic force that reinforces the displacement, and is
driven by the equilibrium current. The final term gives the destabilization by the interaction of the
equilibrium pressure gradient with the magnetic field curvature. For a toroidal equilibrium, this
term is stabilizing on the inside of the torus and destabilizing on the outside where the radius of
curvature and pressure gradient point in the same direction. The stabilizing terms are minimized
by an incompressible perturbation, so these are often found to be the most unstable modes.
To examine the current driven instabilities more closely, we take the displacement to be
incompressible and ignore pressure effects so that the change in potential energy is given by
δW =1
2µ0
∫PB2
1 −B0 · (B1⊥ × ξ⊥)µ0J0‖
B0d3r +
∫vac
B2vac
2µ0d3r. (6.9)
In the large aspect ratio tokamak limit with circular cross-section [158], δW takes the form
δW = πR
∫ a
0
B21
µ0− Jz0(Br1ξθ −Bθ1ξr)dθrdr + πR
∫ b
a
B2vac
µ0dθrdr, (6.10)
where r is the minor radius, z = Rφ is the toroidal coordinate, and θ is the poloidal coordinate.
The plasma-vacuum boundary is at r = a, and a perfectly conducting wall is placed at r = b. The
93
θ and φ dependence may be decomposed into Fourier modes ξ = ξ exp[imθ− inθ]. Then, expanding
δW with ∇ · ξ = 0 and making use of the frozen flux condition to express B1 in terms of ξ gives,
after much simplification [158],
δW =π2B2
φ
µ0R
∫ a
0
[(rdξ
dr
)2+ (m2 − 1)ξ2
]( nm− 1
q
)2rdr
+π2B2
φ
µ0R
[ 2
qa
( nm− 1
qa
)+ (1 +mλ)
( nm− 1
qa
)2]a2ξ2a, (6.11)
where
q =# of toroidal orbits
1 poloidal orbit=rBφRBθ
(6.12)
is the safety factor and
λ =1 + (a/b)2m
1− (a/b)2m. (6.13)
Given Eqn. 6.11, we can draw a number of conclusions about the stability of this mode. First,
the integral over the plasma volume is strictly positive, so a plasma terminated by a conductor at
r = a will be stable to all perturbations of this type. Second, any mode with m/n < qa will be
stable. Typical q profiles are increasing functions of r, so this includes all modes with a resonant
surface inside the plasma. Finally, modes with m/n > qa can be unstable. The unstable modes
should be localized near the edge to minimize the stabilizing effects on the plasma interior, and
have resonant qr = m/n surface just outside the plasma (qr > qa) so that the surface contribution
to δW is negative. These modes are named the external kink or peeling instability due to their
helical nature and tendency to evolve in such a way as to “peel” off the outer flux surfaces of the
plasma [149].
Another important class of ideal MHD instabilities are the ballooning instabilities. These
instabilities are driven by the (ξ⊥ · ∇p0)(ξ⊥ · κ) term, and so are strongest in the bad curvature
region on the outside of the torus. Since the curvature is an essential feature of the instability,
a cylindrical treatment like that used for the kink instability is typically inadequate. For an
incompressible mode, a rough estimate of instability is given by balancing the stabilizing magnetic
94
energy term,
δWs =B2
1
µ0=
1
µ0|∇ × ξ ×B0|2 ≈
k2‖B2φξ
2
µ0, (6.14)
with the destabilizing pressure driven term,
δWd ≈ −1
Rc
dp0dr
ξ2. (6.15)
Setting these equal and taking k‖ = 1/qR, the distance along a field line from the inside to the
outside of the torus, and taking the radius of curvature to be approximately the major radius,
Rc ∼ R0, gives
−dp0dr∼
B2φ
µ0q2R0. (6.16)
A more detailed analysis [158, 22] of the high-n instability finds that magnetic shear, s = d ln q/d ln r
is stabilizing, and the stability boundary at moderate levels of shear can be approximated as
s = 1.67α where the normalized pressure gradient is given by
α = −2µ0Rq2
B2
dp
dr. (6.17)
The above discussion of ideal MHD instabilities finds that low-n kink modes are destabilized at
low edge pressure and high edge current, and high-n ballooning modes are destabilized at high edge
pressure and low edge current. It’s also possible for peeling-type instabilities and ballooning-type
instabilities to couple [59, 135], and the resulting intermediate-n instability appears at high edge
pressure and high edge current. This combined peeling-ballooning model of ideal MHD stability
[160] has been shown to be quite useful for understanding ELMs as we’ll discuss in the next section.
Beyond the Ideal treatment of these instabilities, non-ideal effects, such as resistivity, can also have
significant impact on the stability, and may be important for describing certain ELMing regimes.
For real cases, the stability of a given experimentally determined plasma equilibrium is quite difficult
to ascertain, and depends on the plasma magnetic equilibrium as well as profiles of pressure and
current.
95
6.2 An Overview of ELMs
A great many experimental and theoretical reviews of ELMs exist in literature [166, 143, 7,
110, 91, 75]. The focus of this section is to give a brief overview of ELM experiments, so the results
discussed later in this chapter may be put in context.
Experimental observations of ELMs in H-mode plasmas across many devices has lead to the
classification of several distinct ELM types [166]. These are:
• Type-I (Giant) ELMs seen in H-modes with high auxiliary heating and plasmas with
steep edge pressure gradients,
• Type-II (Grassy) ELMs seen in strongly shaped H-mode plasmas,
• Type-III ELMs seen in close proximity to the L-H transition and for low auxiliary heating.
Beyond these general classifications, a few machine specific ELM types have been observed (e.g.
Type-V ELMs of NSTX [92]).
Type-I ELMs can most readily be identified by their increase in frequency with heating power
[166]. This observation is consistent with a pressure or pressure gradient driven mode, like the ideal
MHD ballooning mode discussed above, limiting the pedestal growth. This is further reinforced
by the observation of a constant pressure limit in measurements of edge temperature and density
during Type-I ELMing regimes [143]. Magnetic fluctuations are also seen to increase during the
ELM event, and precursors have been observed 100-400 µs preceding the ELM in edge density,
temperature, and magnetics [110]. The precursor observations do not appear to be consistent
across all machines, however. ELM power loss is typically 5-15% of the pre-ELM pedestal stored
energy for Type-I ELMs [75].
Type-II ELMs are a class of high-frequency, small ELM originally observed in DIII-D and JT-
60U [143]. These ELMs typically occur at high edge pressure gradient, with high edge safety factor,
and strong elongation and triangularity. Because of these features, type-II ELMs are associated
with the second stability region of ideal MHD theory [166].
96
Type-III ELMs are seen soon after the transition to H-mode, while the edge temperature
and density are still relatively low compared to conditions for type-I ELMs [143]. These ELMs are
seen to decrease in frequency with increased heating power, and eventually disappear at sufficiently
high edge temperatures. Edge pressure gradients for type-III ELMs are not near typical ballooning
limits. These observations suggests that resistive effects may be important for these ELMs [75].
ELM power loss is typically < 5% of the pre-ELM pedestal stored energy. Magnetic precursors are
consistently seen preceding type-III ELMs. These precursors exhibit intermediate toroidal mode
numbers, n = 5− 10, and f ≈ 70kHz [166].
In recent years, the development of fast edge diagnostics has produced detailed descriptions
of the full ELM cycle including ELM filament structure [111, 108, 89]. One-dimensional imaging
systems viewing visible light fluctuations in the plasma edge have also been used to observe ELM
fine-structure on ASDEX [38] and ELM precursors on Alcator C-MOD [146], and two-dimensional
imaging systems have been used on MAST to examine ELM filament structure [83, 82]. In addition,
ELM filament structure has been examined using GPI observations in several ELMing regimes
[96, 93], and measurements of NBI heated H-modes with Type-III ELMs captured the birth of
filamentary structures in two-dimensions over ∼ 50 µs [96].
The theory of coupled ideal MHD peeling-ballooning modes has been very successful in de-
scribing observed stability boundaries for large, type-I ELMs [24, 134, 135, 132]. Peeling-Ballooning
stability calculations have shown the edge pedestal to be unstable to coupled, intermediate-n modes
at high edge pressure gradient and high edge current in DIII-D, Asdex Upgrade, JET, JT60-U, and
Alcator C-mod [132, 54]. At low aspect ratio, NSTX pedestals have been shown to be unstable
primarily to the kink/peeling instability [19] at low-n, and the pedestals typically exist far from the
high-pressure ballooning mode boundary. In addition to these linear stability studies, nonlinear
simulations of ELMs have also been carried out [36, 133, 65, 163, 102, 79, 112, 78].
97
6.3 NSTX observations of ELM precursors
In the remainder of the chapter, we’ll examine results from two-dimensional imaging of pre-
cursor modes preceding small ELM events in NSTX. Gas Puff Imaging (GPI) of visible light fluc-
tuations near the last closed flux surface has revealed precursor edge intensity fluctuations that are
wave-like in nature. These edge oscillations are seen to grow in amplitude preceding ELM events
and H-L back transitions, but they are also observed intermittently at low amplitudes throughout
H-mode operation. A magnetic signature is also observed that is concurrent and strongly correlated
with the edge intensity oscillations. The nonlinear evolution of the precursor mode and ELM crash
have been imaged in the plane perpendicular to the magnetic field using the GPI diagnostic, and
the two-dimensional structure and dynamics are analyzed and presented here. Precursor modes
were imaged primarily in near-threshold RF-heated H-modes and some Ohmic heated cases, but
they were absent in the similar, near-threshold NBI heated cases studied.
6.4 Operational Parameters and Plasma Conditions
The experimental observations discussed in this chapter were obtained during the 2010 run
campaign of NSTX. Discharges are deuterium plasmas with on-axis toroidal field Bt = 0.45 T and
a plasma current Ip = 900 kA. Table 6.1 contains specific details for each shot, and example traces
of Ip, radio frequency power Prf , and Dα light for a single shot are shown in Figure 6.1.
This database of shots was originally designed to probe the RF power threshold for the low
to high confinement mode transition in NSTX. The radio frequency (RF) heating system used is
a twelve antenna high harmonic fast wave (HHFW) heating system that can deliver up to 6 MW
of power at 30 MHz [117]. The database of shots considered in this study are RF heated H-mode
plasmas with 0.5-1.2 MW of input RF heating power operated at 180◦ phasing to optimize core
heating efficiency [62, 144].
The classification of ELM type is not straightforward in this operational regime. Low heating
powers and proximity to the L-H transition suggest that Type-III ELMs would be most likely,
98
Table 6.1: Shot database for this study including the shot number, timeframe of interest, toroidalmagnetic field, plasma current, and RF heating power
Shot # Timeframe [s] BT [T] IP [kA] PRF [kW]
141917 0.23-0.25 0.45 900 625141917 0.25-0.27 0.45 900 1200141918 0.22-0.24 0.45 900 625141919 0.23-0.25 0.45 900 625141920 0.24-0.25 0.45 900 625141922 0.24-0.27 0.45 900 700142000 0.22-0.25 0.45 900 1100142001 0.22-0.25 0.45 900 1100142002 0.22-0.25 0.45 900 1100142003 0.22-0.25 0.45 900 1100142006 0.25-0.27 0.45 900 1100
Figure 6.1: Time traces of RF heating power PRF , plasma current IP , Dα light, and line-integrateddensity N for typical shot from this study. Neutral beams are also used early in the shot for plasmaconditioning. The shaded region indicates the time period for traces plotted in Figure 6.3.
however a distinguishable ELM frequency is not observed. Additionally, the presence of L-H-L
dithers further complicates identification. Alternatively, ELM precursors observed in this regime
are most similar to precursors observed in Type-I ELMing regimes on NSTX, namely, precursors are
short-lived and exhibit intermediate toroidal mode numbers. Regardless of ELM type, the events
studied here are small, and the stored energy drop in most events is below the measurable limit.
99
However, the stored energy is observed to drop 3-5% during events which trigger H-L transition.
The primary observations used for this study are from the GPI diagnostic. For this campaign,
the GPI diagnostic used a Phantom v710 fast-framing camera to capture images at 400,000 frames
per second (2.5 µs per frame). Deuterium was used for the neutral gas species, and Dα (656
nm) line-emission was imaged using collection optics and optical filters. Images were recorded
at 64x80 pixel resolution covering roughly 25 cm by 30 cm of the edge region. The GPI view
is aligned with the magnetic field such that the horizontal (x) direction of the camera image is
approximately codirectional with the radial direction, and the vertical (y) direction is then the
generalized poloidal coordinate, perpendicular to both the magnetic field and the radial direction.
Therefore, the vertical direction is approximately the projection of the machine poloidal direction
into the plane perpendicular to the magnetic field.
6.5 Precursor Oscillations in GPI Intensity
Analysis of GPI observations from the shots listed in Table 6.1 reveals a periodic edge intensity
fluctuation seen to precede ELMs and ELM-induced H-L back transitions during RF heated H-mode
operation. These edge oscillations have a distinct, elongated mode structure and are visible up to
200 µs preceding the ELM event. While these fluctuations are seen to precede ELM events, low
amplitude oscillations are also seen intermittently throughout RF H-mode operation. In this section
we detail the GPI observations of the ELM precursors.
A series of image stills from the GPI diagnostic is displayed in Figure 6.2 to illustrate the
general features of the precursor phase leading to the ELM crash. Edge intensity oscillations with
a distinct elongated structure are visible preceding the ejection of plasma captured in frames (i)-
(j). During the precursor phase, frames (a)-(h), intensity peaks travel in the positive y direction
(in the electron diamagnetic direction), and as one peak leaves the camera view a second peak
enters from the bottom. The intensity structures appear to drift radially outward and become
increasingly deformed as they propagate through the camera view. This deformation is especially
evident in frame (h) directly preceding the filamentation. Eventually, an unidentified processes is
100
Figure 6.2: Multiframe image stills of an ELM event with precursor intensity fluctuations from shot141918. The time between frames is ∼7.5 µs. Distinct mode structure can be seen in precursoroscillations leading to the ejection of the filament in the last two frames. The approximate locationof the separatrix is indicated by the dashed line.
triggered precipitating the explosive ejection of plasma filaments into the SOL. Once in the SOL,
the filaments travel in the reverse direction (-y, ion diamagnetic direction). This reversal of the
propagation direction is commonly attributed to a change in the radial electric field from inside to
outside the separatrix. In many observed ELM events clear filamentation is evident such as in Fig.
6.2(j). However, the ELM crash is a complicated nonlinear process, thus the ejection of plasma
into the SOL does not always follow the clear time evolution shown in Fig. 6.2.
Two quantities may be defined from the GPI image sequence which will be of use in the
following analysis. The first of these is the integrated edge intensity Iedge, which is defined here
as the sum of the intensity ±2 cm around the radial location of the peak time-averaged intensity
profile. The second quantity is the ratio of light in the SOL to the images total intensity, FSOL.
This quantity is a measure of the relative intensity in the SOL, thus it is a good indicator of plasma
loss into the SOL such as ELM events, intermittent blobs, and turbulent losses during L-mode
101
Figure 6.3: Time traces of (a) scrape-off layer fraction FSOL and (b) integrated edge intensity Iedgecorresponding to shaded timeperiod in Fig. 6.1. Traces show edge intensity fluctuations precedingan ELM at 0.2425s, and an ELM-induced back transition at 0.245s. Low level fluctuations can alsobe seen near 0.244s.
operation. During an H-L transition the mean value of Fsol can change by .15-.20, while ELMs can
cause deviations greater than .40 for a large event. The precursor oscillations are easily observable
in time traces of either of these quantities as seen in Figure 6.3. Iedge fluctuations preceding an ELM
event can be seen at 0.2427s, and an ELM induced H-L transition is triggered at 0.245s. Low level
periodic fluctuations can be seen near 0.244s as well. These intermittent, low-amplitude fluctuations
are found throughout RF H-mode operation, and are similar in frequency and wavenumber to the
precursor oscillations, as will be discussed in the following section.
Figure 6.4 shows the 1D plus time dynamics of the precursor event, and it illustrates the
coherent nature of the precursor oscillations. The figure shows two-dimensional “slices” of the GPI
data with (a) one cut in x (radial) versus time at y=15.50 cm and (b) one cut in y (perpendicular)
versus time at x=10.5 cm. The fluctuating intensity pattern is easily discernible in (a) at time
0.2592s, and the intensity peaks appear to drift radially outward as the mode grows. A similar
pattern is evident in (b) where the tilt in the intensity pattern, typically referred to as streaks,
indicates propagation in the positive y (electron diamagnetic direction). A simple linear fit to the
102
Figure 6.4: Two-dimensional slices through GPI data with one cut at y=15.5 cm (a), and one atx=10.5 cm (b). Precursor fluctuations are very distinct brightness pulses which appear to moveupward, or in the electron diamagnetic direction indicated by the tilt of structures in (b). Structuresare also seen to drift radially outward as indicated by +x motion in (a).
intensity streaks gives an estimate of the perpendicular velocity of ∼13 km/s.
Figure 6.5: Traces of (a) Average perpendicular velocities, and (b) FSOL for many ELM eventsfrom the shots database. Velocities are measured ∼2 cm inside the separatrix. Timings are relativeto the peak FSOL for each event.
For a more refined velocity estimate, we employ the HOP-V code, a digital image velocimetry
103
algorithm based on pattern-matching techniques [105], to estimate the two-dimensional flow field
of the precursor intensity fluctuations. Poloidal velocities averaged in the vertical direction and
measured just inside the separatrix are plotted in figure 6.5 along with FSOL traces. The figure
shows the flow behavior during several observed ELM events with precursor fluctuations. Times
for figure 6.5 are relative to the peak in FSOL, so that the ejection of plasma occurs near t=0.0 ms
and precursor activity can be seen between -0.2 ms and 0.0 ms. Average perpendicular velocities
measured inside the separatrix during the precursor period typically range between +2-8 km/s,
but velocity estimates measured at the intensity maximum are in agreement with the streak fit
estimates. At the time of the ELM crash, measured radial velocities of intensity structures can
peak as high as 8 km/s as filaments are ejected into the SOL. Additionally, average perpendicular
velocities measured inside the separatrix are seen to briefly reverse direction during the crash, and
reach flows up to -4 km/s.
6.6 Wavenumber and Frequency Characterization of the Precursor Mode
Time-Frequency analysis of the integrated edge intensity Iedge is performed using the contin-
uous wavelet transform (CWT) [150]. This method is similar to the familiar technique of windowed
Fourier transforms, however the CWT uses a ’mother’ wavelet function that is scaled and translated
to measure the power contained in a signal at a given location in time-scale (frequency) space, thus
one obtains a time-frequency power spectrum analogous to the spectrogram of windowed Fourier
transforms. The advantage of the CWT is the inherent ability of the scaling operation to yield an
optimal product of the time and frequency resolutions at each frequency value. The variable reso-
lution allows the CWT to capture the fine details of a signal near singularities while still providing
efficient measurement of lower-frequency behavior.
The 6th order Morlet CWT of the Iedge signal from Fig. 6.3 is presented in Figure 6.6 along
with the accompanying FSOL trace. The ELM and H-L back-transition events are directly preceded
by significant increases in power in the 20-30 kHz band, and the low amplitude oscillations between
0.243 and 0.244 s also show significant levels of power near 25 kHz. Other shots in this collection
104
Figure 6.6: Wavelet scalogram of integrated edge intensity, Iedge accompanied by the time trace ofthe scrape-off layer fraction for shot 141919. The power spectrum shows significant power at the20 kHz scale during the ELM precursor fluctuations. The shaded region in the wavelet scalogramindicates where edge effects become important, and the white contour indicates the 95% significancelevel.
exhibit similar activity in the 20-30 kHz range, both preceding ELM events and at low levels
intermittently throughout H-mode operation. Power increases in the 3 kHz range are associated
with the timescale of the full ELM event including the period of increased fluctuations following
the crash.
Figure 6.7: Scatter plot of frequency of precursor fluctuations against the perpendicular wavenum-ber derived from GPI intensity fluctuations. Red triangles are events that lead to an ELM orback-transition, while blue squares are edge intensity fluctuations that do not lead to an ELM.
105
Short time window FFT analysis has also been used to corroborate the results of the CWT
analysis, and a linear fit to the cross-spectral phase versus poloidal position provides an estimate
of poloidal wavenumbers, ky for the precursor. Several short time periods from the shots listed in
Table 6.1 which exhibit periodic edge intensity fluctuations have been analyzed, and the results
are presented in Figure 6.7. In agreement with the CWT analysis, typical frequencies are in the
range of 20-30 kHz, and perpendicular wavenumbers are found to be between 0.05 and 0.21 cm−1.
Periods preceding ELM events and periods of low amplitude fluctuations were both analyzed and
no significant difference in wavenumber or frequency was found between the two sets of events.
Additionally, the points cluster about a phase velocity of ∼13 km/s, which is consistent with the
image velocimetry estimates. In this case, the pattern-matching velocimetry is tracking the motion
of individual intensity peaks and troughs because of the size of the wave relative to the size of the
image subframe. Therefore, it is expected that the velocimetry estimate would be consistent with
the phase velocity.
6.7 Quantification of Edge Deformation During Precursor Evolution
Figure 6.8: Example (a) image frame from shot 141917 and (b) xedge function overlayed on intensitycontours. Maximum radial excursion relative to EFIT separatrix location is plotted in (c), and theedge curvature, κ corresponding to this point is plotted in (d). The Dashed line indicates the timepoint of image (a).
−4−2024
0.2441 0.2442 0.2443 0.2444 0.2445t [s]
−0.4
−0.2
0.0
(a) (b)
(c)rmax−rsep
(d)
κ
106
During the evolution of the precursor the edge intensity profile becomes deformed, and por-
tions of the intensity profile balloon into the SOL as they propagate through the GPI view. The
radial excursion of the intensity fluctuations and the curved edge deformation are both observable
in Figure 6.2 frames (a)-(g). To quantify these features of the mode, we first define the plasma edge,
xedge(y) by taking a weighted average of the x coordinate at each y location using the cube of the
intensity as the weight for each point. This weighting insures that the edge location closely follows
the peak of the radial profile of the intensity, and this yields a definition for the “edge” which is
consistent with a visual estimate. When turbulence fluctuations are low, such as in H-mode, the
GPI neutral gas puff penetrates an approximately poloidally uniform distance in from the plasma
edge, so that xedge, to a rough approximation, traces out a flux surface. Thus we will consider
perturbations of xedge to be a proxy for perturbations of the outer flux surface. Figures 6.8(a)-(b)
illustrate an example of xedge. The edge curvature and maximum radial excursion may then be
derived from xedge.
The maximum radial excursion at any given time corresponds to the right-most point, or
point with largest x value, on the xedge curve, and the time evolution of this position relative to
the EFIT separatrix is plotted in figure 6.8(c). The EFIT separatrix location is typically uncertain
to ±1 cm at the location of the GPI view, and MHD activity generally associated with ELMs will
perturb the location of the separatrix. Additionally, the EFIT time resolution for this analysis is
4 ms, but over a 0.1 s period during H-mode the reconstructed position of the separatrix is slowly
varying and changes by no more than 3 cm. So, we believe it to be a reasonable estimate of the
equilibrium separatrix location about which the fast time-scale behavior fluctuates. Furthermore,
it is the most physically meaningful reference location available, so we choose to use it for relative
measurements of the radial position of the edge. It should be noted that the choice of reference
does not significantly change any of the conclusions.
The curvature of the edge is calculated using the mathematical definition: κ = x′′/(1+x′2)3/2,
and the necessary derivates are calculated from a quintic spline fit of xedge using two internal knots.
In this work, negative curvature indicates that the convex side is directed radially outward. Figure
107
6.8(d) shows a time trace of the curvature measured at the point of maximum radial excursion for
that time point.
Precursor activity can be observed in figures 6.8(c)-(d) between 0.2442 s and 0.2444 s, and
modulations of the radial excursion and the edge curvature are evident. These modulations coincide
with intensity peaks passing through the camera view as seen in figure 6.2. As the precursor mode
evolves, the radial excursion increase, and the curvature becomes increasingly negative indicating
significant deformation of the edge. Shortly following the image in 6.8(a), a filament forms and
is explosively ejected into the SOL near t=0.2444 s. After this time, a well-defined edge does not
exist until the turbulence is quenched.
Figure 6.9: Scatterplot of minimum edge curvature, κmin and maximum radial excursion relativeto the EFIT separatrix location for several ELM precursor events (diamond). Intensity fluctuationsnot leading to an ELM are also included (square).
Several precursor events are compared in Figure 6.9, and low amplitude events (such as in
Figure 6.3 at 0.2440) that do not precipitate an ELM are also included. The figure compares the
minimum (i.e. most negative) curvature κmin and the maximum radial excursion during each event.
Events directly preceding ELMs are found to reach closer to the EFIT separatrix location, and
curvature values for these events are more negative than events not preceding ELMs. Therefore, the
curvature appears to be an important characteristic of the underlying instability, and the increased
108
curvature for ELM unstable events suggests that this feature is important for the triggering of the
ELM crash. Still, it is unclear why the mode sometimes saturates at a low amplitude, and does
not always precipitate an ELM.
6.8 Magnetic Fluctuations
Signals from magnetic pick-up coils distributed in toroidal angle around the device have also
been analyzed, and precursor activity in the 20 kHz region is observed in the time derivative of
the magnetic field (B). Time traces of Iedge and a lowpass filtered B signal are plotted in Figures
6.10(a) and (b), respectively. These traces illustrate the similarities in the precursor behavior, and
indeed the B signal filtered below 200 kHz and Iedge signal are highly correlated as shown in Figure
6.10(c). The figure shows the time-lagged correlations for 200 µs long time segments preceding
the ELM crash, and the coil with the highest correlation is chosen for figures 6.10(b) and (c).
This particular coil is separated 30◦ in toroidal angle from the GPI view, and is located below the
midplane. Additionally, this coil is located near field lines that pass through the GPI view. The
absolute value of the correlation coefficient exceeds 0.8 at -10 µs, thus the magnetic fluctuations
measured at this position are delayed with respect to the intensity fluctuations. The high correlation
suggests that the precursor edge intensity fluctuations seen by GPI are electromagnetic in nature
as one would expect for typical MHD instabilities believed to produce ELMs.
Toroidal mode numbers are estimated from a toroidally distributed array of magnetic pick-up
coils. Time traces of B 0.3 ms in length and directly preceding ELM events are Fourier transformed,
and precursor modes are identified as peaks in the power spectrum. Toroidal mode numbers are then
determined from the phase of the precursor Fourier mode as a function of toroidal position of the
magnetic coil. This simple method assumes a global mode, and it does not account for the possibility
of a frequency modulation during the mode evolution. We find that toroidal mode numbers between
n = 5− 10 are typical for precursor behavior similar to what has been plotted in Figure 6.10. For
comparison, it is possible to estimate toroidal mode numbers from the perpendicular wavenumbers
estimated from the GPI data. If the structures are assumed to be aligned with the magnetic
109
Figure 6.10: Time traces of (a) Iedge and (b) low-pass filtered magnetic signals from shot 141917.The black trace has been lowpass filtered at 200 kHz, and the red trace has been bandpass filteredaround 20 kHz. Magnetics traces are strongly correlated with fluctuations in edge intensity asshown in the lagged correlation plot (c). Absolute values of the correlation coefficient reach 0.8 forperiods during the precursor activity.
field, then the perpendicular wavelength can be mapped to an equivalent toroidal wavelength given
the magnetic pitch angle. Using this mapping, toroidal mode number estimates from imaging are
between n = 4− 15 which agrees with the magnetic coil estimate.
6.9 Pedestal Characteristics
As discussed in 6.2, steep pressure gradients at the edge provide a source of free energy for
Ballooning type MHD instabilities. In H-mode, the pressure profile exhibits a pedestal structure
at the edge characterized by a steep rise to a pressure of hped over a width ∆ped. Figure 6.11 shows
a typical edge electron pressure profile for the near-threshold RF heated plasmas, and a modified
Tanh fit [55, 29] to the data. Here, ψn is a normalized flux coordinate defined as:
ψn =ψc − ψψc − ψs
,
where ψc is the flux at the core and ψs is the flux at the separatrix. Pedestal width ∆ped and
pedestal height hped are parameters of the fit function, and a comparison of these parameters
for different shots is included in Figure 6.11. Parameter uncertainties are estimated by manually
varying one parameter about the best fit, and then performing a reduced fit with that parameter
110
fixed. The error bars indicate the variation in the parameter that yields a change in the reduced
χ2 of the fit of ∆χ2 = 1.
Figure 6.11: Pedestal parameters for height hped and width ∆ped are extracted from a modified tanhfit to the electron pressure profile (inset) and a comparison of several shots is presented. Squaresare 0.6-1.0 MW RF shots, triangles are Ohmic H-mode shots, stars are ∼ 1.0 MW NBI shots, andthe diamond is a 4 MW NBI heated shot.
The comparison illustrated by Figure 6.11 indicates that Ohmic H-modes (triangles) have
similar edge electron pressure profiles as the near-threshold RF H-modes (squares). In fact, ELM
precursors quantitatively similar to those observed in RF heated plasmas were observed in several
Ohmic H-mode discharges. However, 34 events with the precursor magnetic signature were ob-
served in the 9 RF shots analyzed, while only 6 events were observed in 8 Ohmic shots. Pedestal
characteristics of low-power NBI heated shots were also similar to the ohmic and RF heated cases
within the uncertainties, and GPI data from 3 NBI heated shots near L-H threshold power were
analyzed for precursor behavior. Additionally, 10 NBI heated shots in the Type-I ELMing regime
were analyzed, but no short-lived coherent precursors were observed in the GPI data for any of the
NBI heated shots. It is possible, however, that in these cases GPI neutrals may not penetrate far
enough into the plasma to allow GPI to observe the precursor modes. To improve this comparison,
more GPI data from near-threshold NBI heated plasma discharges is needed
111
6.10 Concluding Remarks on the Nature of the Precursor Mode
The question of what underlying instability drives the precursor mode still remains. There is
some indication that micro-instabilities (such as drift-type instabilities or kinetic ballooning modes)
can play a role in the evolution of the pedestal during the inter-ELM period, but these modes are
significantly smaller than the precursors observed here. As discussed above, ideal MHD modes are
likely candidates because of the observed magnetic fluctuations and fast timescales associated with
the ELM crash. These are the low-n peeling mode, and the high-n ballooning mode. The ideal
ballooning mode appears at high edge pressure gradient which is incompatible with the pedestal
height and width measured during precursor modes. The peeling mode is destabilized by edge
currents, and so could be unstable at the low pressures seen during the precursor if high edge current
is present. These modes are low-n though, which is incompatible with the magnetic observations.
Coupled peeling-ballooning modes could appear at intermediate-n in agreement with the magnetic
fluctuations, but these modes are generally seen at high edge pressure and high edge current. Some
studies indicate that non-ideal effects, such as resistivity and diamagnetic stabilization of high-n
modes, could lead to the destabilization of an intermediate-n resistive ballooning mode [102, 79].
Precise assessment of the stability requires a detailed numerical calculation. This was pursued
for this study, but was eventually abandoned because the edge density and temperature profiles
could not be adequately constrained for the models. Hypothetically, a focused experiment could
operate in a regime where these ELM events could be observed repeatedly for longer time periods.
Then, multiple Thomson scattering profiles could be combined to help constrain the plasma profiles.
In addition, an improved radial resolution for Thomson scattering in the gradient region of the edge
would be beneficial.
Chapter 7
Directions for Future Work
This dissertation has presented a number of tokamak edge physics studies using the Gas Puff
Imaging diagnostic on NSTX. Chapter 3 compared edge turbulence and fluctuation measurements
between BES and GPI. Chapter 5 characterized flows in the edge preceding the L-H transition and
explored the similarities between the observations and the limit cycle oscillation. Finally, Chapter
6 presented observations of a coherent mode preceding ELMs. In light of these results, we now
consider avenues for further study.
One critical area of research for next generation tokamak reactors is ELMs, ELM mitiga-
tion, and ELM suppression. GPI, with its high time resolution, good spatial resolution and two-
dimensional coverage of the edge gradient region, is well positioned to play an important role in this
area. Studying the evolution of edge turbulence and fluctuations in between ELMs could improve
our understanding of pedestal physics and the physical mechanisms that determine the edge gradi-
ent, pedestal height, and width. Furthermore, GPI is well suited for studying changes in the edge
turbulence during ELM suppression with Resonant Magnetic Perturbations or inter-shot Lithium
evaporation, or ELM triggering via pellet injection.
Another important research thrust is in the comparison and validation of numerical codes
with GPI observations. Currently, ELM models that are used to predict stability and ELM onset
in various devices are predominantly linear models. The nonlinear stages of ELM development
are important to understand ELM triggering and the resulting Plasma-material interactions from
ELM filaments crossing the SOL and contacting the wall or divertor. This is especially important
113
for small ELM regimes, as these are candidate operating regimes for reactor scenarios. This is an
area where GPI observations in conjunction with numerical simulations, such as BOUT++, could
have a large impact. Other possibilities include validation of edge turbulence simulations in both
L-mode and H-mode plasmas. For almost every application, the close collaboration of numerical
simulation and experimental observation will be important for continued progress.
Many questions still surround the L-H transition and the formation of the edge transport
barrier. As discussed, recent studies have pointed to the importance of zonal flows, and the nonlinear
transfer of turbulent energy at the L-H transition is identified as a possible trigger mechanism.
There is still work to be done by experiment and theory to solidify this hypothesis. It is difficult
to see how GPI can play a role here because the quantity of primary importance is the potential.
Image-based velocity estimates can be used as a proxy, but more work needs to be done to verify
these measurements. It’s not yet clear if zonal flows play a role in all L-H transitions or only those
that exhibit a limit cycle oscillation.
In conclusion, many physical phenomena of great importance for next step fusion devices
exist in the plasma edge. For example, ELMs will continue to be a problem as plasma pressure and
confinement is pushed toward reactor relevant values. The physical mechanisms that determine
the H-mode pedestal directly impact the plasma performance and achievable fusion output. Access
to the H-mode is likely necessary to reach a burning fusion reactor, thus understanding the L-H
transition is crucial. GPI is well suited to address these concerns, and therefore can be an important
diagnostic for the next decade of fusion research.
Bibliography
[1] M. Agostini, R. Cavazzana, P. Scarin, and G. Serianni. Operation of the gas-puff imagingdiagnostic in the rfx-mod device. Rev. Sci. Instrum., 77(10), 2006.
[2] M. Agostini, S. J. Zweben, R. Cavazzana, P. Scarin, G. Serianni, R. J. Maqueda, and D. P.Stotler. Study of statistical properties of edge turbulence in the national spherical torusexperiment with the gas puff imaging diagnostic. Phys. Plasmas, 14(10), 2007.
[3] P. Anandan. A computational framework and an algorithm for the measurement of visual-motion. International Journal of Computer Vision, 2(3):283–310, 1989.
[4] J. L. Barron, D. J. Fleet, and S. S. Beauchemin. Performance of optical-flow techniques.International Journal of Computer Vision, 12(1):43–77, 1994.
[5] D. R. Bates and A. E. Kingston. Properties of a decaying plasma. Planetary and SpaceScience, 11(1):1–22, 1963.
[6] D. R. Bates, A. E. Kingston, and R. W. P. McWhirter. Recombination between electronsand atomic ions .1. optically thin plasmas. Proceedings of the Royal Society of London Seriesa-Mathematical and Physical Sciences, 267(1330):297, 1962.
[7] M Becoulet, G Huysmans, Y Sarazin, X Garbet, P Ghendrih, F Rimini, E Joffrin, X Litaudon,P Monier-Garbet, J.M. Ane, P Thomas, A Grosman, V Parail, H Wilson, P Lomas, P De-Vries, KD Zastrow, GF Matthews, J Lonnroth, S Gerasimov, S Sharapov, M Gryaznevich,G Counsell, A Kirk, M Valovic, R Buttery, A Loarte, G Saibene, R Sartori, A Leonard, P Sny-der, LL Lao, P Gohil, TE Evans, RA Moyer, Y Kamada, A Chankin, N Oyama, T Hatae,N Asakura, O Tudisco, E Giovannozzi, F Crisanti, CP Perez, HR Koslowski, T Eich, A Sips,L Horton, A Hermann, P Lang, J Stober, W Suttrop, P Beyer, S Saarelma, and JET-EFDAWorkprogramme. Edge localized mode physics and operational aspects in tokamaks. PlasmaPhys. Control. Fusion, 45(A93-A113), 2003.
[8] P. Bellan. Fundamentals of Plasma Physics. Cambridge University Press, 2006.
[9] J. S. Bendat and A. G. Piersol. Random Data: Analysis and Measurement Procedures. JohnWiley and Sons, INC., 2000.
[10] H. Biglari, P.H. Diamond, and P.W. Terry. Phys. Fluids B, 2:1, 1990.
[11] J. A. Boedo, N. Crocker, L. Chousal, R. Hernandez, J. Chalfant, H. Kugel, P. Roney,J. Wertenbaker, and Nstx Team. Fast scanning probe for the nstx spherical tokamak. Rev.Sci. Instrum., 80(12), 2009.
115
[12] J. A. Boedo, J. R. Myra, S. Zweben, R. Maingi, R. J. Maqueda, V. A. Soukhanovskii, J. W.Ahn, J. Canik, N. Crocker, D. A. D’Ippolito, R. Bell, H. Kugel, B. Leblanc, L. A. Roquemore,D. L. Rudakov, and Nstx Team. Edge transport studies in the edge and scrape-off layer of thenational spherical torus experiment with langmuir probes. Phys. Plasmas, 21(4):42309–42309,2014.
[13] J. A. Boedo, D. L. Rudakov, R. J. Colchin, R. A. Moyer, S. Krasheninnikov, D. G. Whyte,G. R. McKee, G. Porter, M. J. Schaffer, P. C. Stangeby, W. P. West, S. L. Allen, and A. W.Leonard. Intermittent convection in the boundary of diii-d. Journal of Nuclear Materials,313:813–819, 2003. 15th International Conference on Plasma-Surface Interactions in Con-trolled Fusion Devices (PSI-15) May 26-31, 2002 Gifu, japan.
[14] J. A. Boedo, D. L. Rudakov, R. A. Moyer, G. R. McKee, R. J. Colchin, M. J. Schaffer, P. G.Stangeby, W. P. West, S. L. Allen, T. E. Evans, R. J. Fonck, E. M. Hollmann, S. Krashenin-nikov, A. W. Leonard, W. Nevins, M. A. Mahdavi, G. D. Porter, G. R. Tynan, D. G. Whyte,and X. Xu. Transport by intermittency in the boundary of the diii-d tokamak. Phys. Plasmas,10(5):1670–1677, 2003. 2 44th Annual Meeting of the Division of Plasma of the American-Physical-Society Nov 11-15, 2002 Orlando, florida Amer Phys Soc, Div Plasma.
[15] A. Burgess and H. P. Summers. Recombination and level populations of ions .1. hydrogenand hydrogenic ions. Monthly Notices of the Royal Astronomical Society, 174(2):345–391,1976.
[16] K. H. Burrell. Tests of causality: Experimental evidence that sheared exb flow alters turbu-lence and transport in tokamaks. Phys. Plasmas, 6(12):4418–4435, 1999. American-Physical-Society Centennial Meeting 1999 Atlanta, georgia Amer Phys Soc.
[17] K. H. Burrell, T. N. Carlstrom, E. J. Doyle, D. Finkenthal, P. Gohil, R. J. Groebner, D. L.Hillis, J. Kim, H. Matsumoto, R. A. Moyer, T. H. Osborne, C. L. Rettig, W. A. Peebles, T. L.Rhodes, H. Stjohn, R. D. Stambaugh, M. R. Wade, and J. G. Watkins. Physics of the l-modeto h-mode transition in tokamaks. Plasma Physics and Controlled Fusion, 34(13):1859–1869,1992. 9th kiev international conf on plasma theory / 9th international congress on waves andinstabilities in plasmas / 19th eps conf on controlled fusion and plasma physics Jun 29-jul 03,1992 Innsbruck, austria European phys soc; int union pure and appl phys; austrian ministsci and res; cty tyrol; city innsbruck; kongresshaus innsbruck.
[18] F. H. Busse. Convection driven zonal flows and vortices in the major planets. Chaos: AnInterdisciplinary Journal of Nonlinear Science, 4:123, 1994.
[19] J. M. Canik, R. Maingi, S. Kubota, Y. Ren, R. E. Bell, J. D. Callen, W. Guttenfelder,H. W. Kugel, B. P. LeBlanc, T. H. Osborne, and V. A. Soukhanovskii. Edge transport andturbulence reduction with lithium coated plasma facing components in the national sphericaltorus experiment (vol 18, 056118, 2011). Phys. Plasmas, 18(9), 2011.
[20] B. Cao, D. P. Stotler, S. J. Zweben, M. Bell, A. Diallo, and B. LeBlanc. Comparison ofgas puff imaging data in nstx with degas 2 simulations. Fusion Science and Technology,64(1):29–38, 2013.
[21] Steven Chu and Arun Majumdar. Opportunities and challenges for a sustainable energyfuture. Nature, 488(7411):294–303, 2012.
116
[22] J. W. Connor, R. J. Hastie, and J. B. Taylor. Shear, periodicity, and plasma ballooningmodes. Physical Review Letters, 40(6):396–399, 1978.
[23] J. W. Connor and H. R. Wilson. A review of theories of the l-h transition. Plasma Phys.Control. Fusion, 42(1):R1–R74, 2000.
[24] JW Connor, RJ Hastie, HR Wilson, and RL Miller. Magnetohydrodynamic stability oftokamak edge plasmas. Phys. Plasmas, 5(2687-2700), 1998.
[25] G. D. Conway, C. Angioni, F. Ryter, P. Sauter, J. Vicente, and Asdex Upgrade Team. Meanand oscillating plasma flows and turbulence interactions across the l-h confinement transition.Physical Review Letters, 106(6), 2011.
[26] I. Cziegler, P. H. Diamond, N. Fedorczak, P. Manz, G. R. Tynan, M. Xu, R. M. Churchill,A. E. Hubbard, B. Lipschultz, J. M. Sierchio, J. L. Terry, and C. Theiler. Fluctuating zonalflows in the i-mode regime in alcator c-mod. Phys. Plasmas, 20(5), 2013.
[27] I. Cziegler, J. L. Terry, J. W. Hughes, and B. LaBombard. Experimental studies of edgeturbulence and confinement in alcator c-mod. Phys. Plasmas, 17(5), 2010. 51st AnnualMeeting of the Division-of-Plasma-Physics of the American-Physics-Society Nov 02-06, 2009Atlanta, GA Amer Phys Soc, Div Plasma Phys.
[28] I. Cziegler, G. R. Tynan, P. H. Diamond, A. E. Hubbard, J. W. Hughes, J. Irby, and J. L.Terry. Zonal flow production in the l-h transition in alcator c-mod. Plasma Phys. Control.Fusion, 56(7), 2014.
[29] A. Diallo, R. Maingi, S. Kubota, A. Sontag, T. Osborne, M. Podesta, R. E. Bell, B. P.LeBlanc, J. Menard, and S. Sabbagh. Nucl. Fusion, 51(103031), 2011.
[30] P. H. Diamond, S.-I. Itoh, K. Itoh, and T. S. Hahm. Plasma Phys. Control. Fusion, 47:R35,2005.
[31] Diamond, P. H. and Kim, Y.-B. Phys. Fluids B, 3:1626, 1991.
[32] Ashish Doshi and Adrian G. Bors. Robust processing of optical flow of fluids. IeeeTransactions on Image Processing, 19(9):2332–2344, 2010. Doshi, Ashish/A-6438-2013 Doshi,Ashish/0000-0002-5831-5844.
[33] Ashish Doshi and Adrian G. Bors. Smoothing of optical flow using robustified diffusionkernels. Image and Vision Computing, 28(12):1575–1589, 2010. Doshi, Ashish/A-6438-2013Doshi, Ashish/0000-0002-5831-5844.
[34] H. W. Drawin. Collisional-radiative ionization and recombination coefficients for quasi-stationary homogeneous hydrogen and hydrogenic ion plasmas. Zeitschrift Fur Physik,225(5):470, 1969.
[35] H. W. Drawin. Influence of atom-atom collisions on collisional-radiative ionization and re-combination coefficients of hydrogen plasmas. Zeitschrift Fur Physik, 225(5):483, 1969.
[36] B. D. Dudson, X. Q. Xu, M. V. Umansky, H. R. Wilson, and P. B. Snyder. Simulation ofedge localized modes using bout plus. Plasma Phys. Control. Fusion, 53(5), 2011. JointVarenna-Lausanne Workshop on the Theory of Fusion Plasmas Aug 30-sep 03, 2010 Varenna,ITALY.
117
[37] Editors of progress in ITER physics basis. Progress in the iter physics basis - preface. Nucl.Fusion, 47(S1), 2007.
[38] M Endler, I Garcia-Cortes, C Hidalgo, GF Matthews, ASDEX Team, and JET Team. Thefine structure of ELMs in the scrape-off layer. Plasma Phys. Control. Fusion, 47(219-240),2005.
[39] T. Estrada, T. Happel, L. Eliseev, D. Lopez-Bruna, E. Ascasibar, E. Blanco, L. Cupido,J. M. Fontdecaba, C. Hidalgo, R. Jimenez-Gomez, L. Krupnik, M. Liniers, M. E. Manso,K. J. McCarthy, F. Medina, A. Melnikov, B. van Milligen, M. A. Ochando, I. Pastor, M. A.Pedrosa, F. L. Tabares, D. Tafalla, and Tj-Ii Team. Sheared flows and transition to improvedconfinement regime in the tj-ii stellarator. Plasma Phys. Control. Fusion, 51(12), 2009.Happel, Tim/A-2307-2009 36th European-Physical-Society Conference on Plasma PhysicsJun 29-jul 03, 2009 Natl Palace Culture, Sofia, BULGARIA European Phys Soc; UnionPhysicists; Sofia Univ St Kliment Ohrids, Fac Phys.
[40] T. Estrada, C. Hidalgo, T. Happel, and P. H. Diamond. Spatiotemporal structure of theinteraction between turbulence and flows at the l-h transition in a toroidal plasma. PhysicalReview Letters, 107(24), 2011.
[41] Dean Fantazzini, Mikael Hook, and Andre Angelantoni. Global oil risks in the early 21stcentury. Energy Policy, 39(12):7865–7873, 2011. Hook, Mikael/F-5366-2011.
[42] G. Federici, A. Loarte, and G. Strohmayer. Assessment of erosion of the iter divertor targetsduring type i elms. Plasma Phys. Control. Fusion, 45(9):1523–1547, 2003.
[43] N. Fedorczak, P. Manz, S. C. Thakur, M. Xu, G. R. Tynan, G. S. Xu, and S. C. Liu. Onphysical interpretation of two dimensional time-correlations regarding time delay velocitiesand eddy shaping. Physics of Plasmas, 19(12), 2012. Xu, Guosheng/B-4857-2013.
[44] R. J. Fonck, P. A. Duperrex, and S. F. Paul. Plasma fluctuation measurements in tokamaksusing beam-plasma interactions. Rev. Sci. Instrum., 61(10):3070–3070, 1990. 2.
[45] T. Fujimoto. A collisional-radiative model for helium and its application to a dischargeplasma. J. Quant. Spectro. Radia. Transfer, 21:439–455, 1979.
[46] T. Fujimoto, I. Sugiyama, Tachiban.K, and Y. Ogata. Population density and lte of excitedatoms in a positive-column plasma .1. calculation on hydrogen. Japanese Journal of AppliedPhysics, 11(5):718, 1972.
[47] A. Fujisawa. Nucl. Fusion, 49(1):013001, 2007.
[48] T. Fujita, Y. Kamada, S. Ishida, Y. Neyatani, T. Oikawa, S. Ide, S. Takeji, Y. Koide,A. Isayama, T. Fukuda, T. Hatae, Y. Ishii, T. Ozeki, H. Shirai, and J. T. Team. Highperformance experiments in jt-60u reversed shear discharges. Nuclear Fusion, 39(11Y):1627–1636, 1999. 2 17th IAEA Fusion Energy Conference Oct 19-24, 1998 Yokohama, japan Iaea.
[49] Punit Gohil. Edge transport barriers in magnetic fusion plasmas. COMPTES RENDUSPHYSIQUE, 7(606-621), 2006.
[50] M. Goto. Collisional-radiative model for neutral helium in plasma revisited. J. Quant. Spectro.Radia. Transfer, 76:331–344, 2003.
118
[51] P. T. Greenland and D. Reiter. Collisional-radiative models re-examined. Contributions toPlasma Physics, 38(1-2):302–306, 1998. 6th International Workshop on Plasma Edge Theoryin Fusion Devices Sep 15-17, 1997 Exeter coll, oxford, england Commiss European Union, JETJoint Undertaking, Abingdon, UK; DOE, Washington DC, US; Max Planck Inst Plasmaphys,Garching, Germany.
[52] R. J. Groebner. An emerging understanding of h-mode discharges in tokamaks. Physics ofFluids B-Plasma Physics, 5(7):2343–2354, 1993. 2.
[53] R. J. Groebner, K. H. Burrell, and R. P. Seraydarian. Role of edge electric-field and poloidalrotation in the l-h transition. Physical Review Letters, 64(25):3015–3018, 1990.
[54] R. J. Groebner, C. S. Chang, J. W. Hughes, R. Maingi, P. B. Snyder, X. Q. Xu, J. A. Boedo,D. P. Boyle, J. D. Callen, J. M. Canik, I. Cziegler, E. M. Davis, A. Diallo, P. H. Diamond,J. D. Elder, D. P. Eldon, D. R. Ernst, D. P. Fulton, M. Landreman, A. W. Leonard, J. D.Lore, T. H. Osborne, A. Y. Pankin, S. E. Parker, T. L. Rhodes, S. P. Smith, A. C. Sontag,W. M. Stacey, J. Walk, W. Wan, E. H. J. Wang, J. G. Watkins, A. E. White, D. G. Whyte,Z. Yan, E. A. Belli, B. D. Bray, J. Candy, R. M. Churchill, T. M. Deterly, E. J. Doyle,M. E. Fenstermacher, N. M. Ferraro, A. E. Hubbard, I. Joseph, J. E. Kinsey, B. LaBombard,C. J. Lasnier, Z. Lin, B. L. Lipschultz, C. Liu, Y. Ma, G. R. McKee, D. M. Ponce, J. C.Rost, L. Schmitz, G. M. Staebler, L. E. Sugiyama, J. L. Terry, M. V. Umansky, R. E. Waltz,S. M. Wolfe, L. Zeng, and S. J. Zweben. Improved understanding of physics processes inpedestal structure, leading to improved predictive capability for iter. Nucl. Fusion, 53(9),2013. Diallo, Ahmed/M-7792-2013; Lipschultz, Bruce/J-7726-2012 Lipschultz, Bruce/0000-0001-5968-3684.
[55] RJ Groebner and TH Osborne. Scaling studies of the high mode pedestal. Phys. Plasmas,5(1800-1806), 1998.
[56] D. K. Gupta, R. J. Fonck, G. R. McKee, D. J. Schlossberg, and M. W. Shafer. Phys. Rev.Lett., 97:125002, 2006.
[57] A. Hasegawa and K. Mima. Pseudo-3-dimensional turbulence in magnetized nonuniformplasma. Physics of Fluids, 21(1):87–92, 1978.
[58] A. Hasegawa and M. Wakatani. Plasma edge turbulence. Physical Review Letters, 50(9):682–686, 1983.
[59] C. C. Hegna, J. W. Connor, R. J. Hastie, and H. R. Wilson. Toroidal coupling of idealmagnetohydrodynamic instabilities in tokamak plasmas. Phys. Plasmas, 3(2):584–592, 1996.
[60] C. Holland, G.R. Tynan, G. R. Mckee, and R. J. Fonck. Rev. Sci. Instrum., 75:4278, 2004.
[61] B. K. P. Horn and B. G. Schunck. Determining optical-flow. Artificial Intelligence, 17(1-3):185–203, 1981.
[62] J. Hosea, R. E. Bell, B. P. LeBlanc, C. K. Phillips, G. Taylor, E. Valeo, J. R. Wilson, E. F.Jaeger, P. M. Ryan, J. Wilgen, H. Yuh, F. Levinton, S. Sabbagh, K. Tritz, J. Parker, P. T.Bonoli, R. Harvey, and NSTX Team. High harmonic fast wave heating efficiency enhancementand current drive at longer wavelength on the National Spherical Torus Experiment. Phys.Plasmas, 15(056104), 2008.
119
[63] J. Hugill. Edge turbulence in tokamaks and the l-mode to h-mode transition. Plasma Phys.Control. Fusion, 42(8):R75–R91, 2000.
[64] I. H. Hutchinson. Excited-state populations in neutral beam emission. Plasma Phys. Control.Fusion, 44(1):71–82, 2002.
[65] G. T. A. Huysmans, S. Pamela, E. van der Plas, and P. Ramet. Non-linear mhd simulationsof edge localized modes (elms). Plasma Phys. Control. Fusion, 51(12), 2009. 36th European-Physical-Society Conference on Plasma Physics Jun 29-jul 03, 2009 Natl Palace Culture,Sofia, BULGARIA European Phys Soc; Union Physicists; Sofia Univ St Kliment Ohrids, FacPhys.
[66] Intergovernmental Panel on Climate Change. Climate Change 2013: The Physical Science Basis.Cambridge University Press, 2013.
[67] Intergovernmental Panel on Climate Change. Climate Change 2014: Impacts, Adaptation, and Vulnerability.Cambridge University Press, 2014.
[68] K. Itoh and S. I. Itoh. The role of the electric field in confinement. Plasma Phys. Control.Fusion, 38(1):1–49, 1996.
[69] S. I. Itoh and K. Itoh. Model of l-mode to h-mode transition in tokamak. Physical ReviewLetters, 60(22):2276–2279, 1988.
[70] M. Jakubowski, R. J. Fonck, C. Fenzi, and G. R. McKee. Wavelet-based time-delay estimationfor time-resolved turbulent flow analysis. Review of Scientific Instruments, 72(1):996–999,2001. 2 13th Topical Conference on High-Temperature Plasma diagnostics Jun 18-22, 2000Tucson, arizona Los Alamos Natl Lab, Phys Div; Los Alamos Natl Lab, Inertial ConfinementFus and Radiat Phys Program; Gen Atom; Amer Phys Soc, Div Plasma Phys; US DOE, OffFus Energy Sci; US DOE, Off Def Sci.
[71] M. Jakubowski, R. J. Fonck, and G. R. McKee. Observation of coherent sheared turbulenceflows in the diii-d tokamak. Physical Review Letters, 89(26), 2002.
[72] R. K. Janev and J. J. Smith. Cross sections for collision processes of hydrogen atoms withelectrons, protons and multiply charged ions. Atomic and Plasma-Material Interaction Datafor Fusion, 4, 1993.
[73] S. C. Jardin, C. E. Kessel, J. Menard, T. K. Mau, R. Miller, F. Najmabadi, V. S. Chan,L. L. Lao, Y. R. Linliu, R. L. Miller, T. Petrie, P. A. Politzer, and A. D. Turnbull. Physicsbasis for a spherical torus power plant. Fusion Engineering and Design, 65(2):165–197, 2003.Jardin, Stephen/E-9392-2010.
[74] L. C. Johnson and E. Hinnov. Ionization, recombination, and population of excited-levels inhydrogen plasmas. Journal of Quantitative Spectroscopy and Radiative Transfer, 13(4):333–358, 1973.
[75] K. Kamiya, N. Asakura, J. Boedo, T. Eich, G. Federici, M. Fenstermacher, K. Finken, A. Her-rmann, J. Terry, A. Kirk, B. Koch, A. Loarte, R. Maingi, R. Maqueda, E. Nardon, N. Oyama,and R. Sartori. Edge localized modes: recent experimental findings and related issues. PlasmaPhys. Control. Fusion, 49(S43-S62), 2007.
120
[76] S. M. Kaye, M. G. Bell, R. E. Bell, S. Bernabei, J. Bialek, T. Biewer, W. Blanchard, J. Boedo,C. Bush, M. D. Carter, W. Choe, N. Crocker, D. S. Darrow, W. Davis, L. Delgado-Aparicio,S. Diem, J. Ferron, A. Field, J. Foley, E. D. Fredrickson, D. A. Gates, T. Gibney, R. Harvey,R. E. Hatcher, W. Heidbrink, K. Hill, J. C. Hosea, T. R. Jarboe, D. W. Johnson, R. Kaita,C. Kessel, S. Kubota, H. W. Kugel, J. Lawson, B. P. LeBlanc, K. C. Lee, F. Levinton,R. Maingi, J. Manickam, R. Maqueda, R. Marsala, D. Mastrovito, T. K. Mau, S. S. Medley,J. Menard, H. Meyer, D. R. Mikkelsen, D. Mueller, T. Munsat, B. A. Nelson, C. Neumeyer,N. Nishino, M. Ono, H. Park, W. Park, S. Paul, T. Peebles, M. Peng, C. Phillips, A. Pigarov,R. Pinsker, A. Ram, S. Ramakrishnan, R. Raman, D. Rasmussen, M. Redi, M. Rensink,G. Rewoldt, J. Robinson, P. Roney, A. L. Roquemore, E. Ruskov, P. Ryan, S. A. Sab-bagh, H. Schneider, C. H. Skinner, D. R. Smith, A. Sontag, V. Soukhanovskii, T. Steven-son, D. Stotler, B. Stratton, D. Stutman, D. Swain, E. Synakowski, Y. Takase, G. Taylor,K. Tritz, A. von Halle, M. Wade, R. White, J. Wilgen, M. Williams, J. R. Wilson, W. Zhu,S. J. Zweben, R. Akers, P. Beiersdorfer, R. Betti, T. Bigelow, et al. Progress towards highperformance plasmas in the national spherical torus experiment (nstx). Nuclear Fusion,45(10):S168–S180, 2005. Sabbagh, Steven/C-7142-2011; Nishino, Nobuhiro/D-6390-2011;Choe, Wonho/C-1556-2011; White, Roscoe/D-1773-2013; Gates, David/K-3715-2012 White,Roscoe/0000-0002-4239-2685; Gates, David/0000-0001-5679-3124 Symposium on Theory ofMagnetic Confinement held at the 20th IAEA Fusion Energy Conference Nov 01-06, 2004Vilamoura, PORTUGAL Iaea.
[77] SM Kaye, MG Bell, RE Bell, J Bialek, T Bigelow, M Bitter, P Bonoli, D Darrow, P Efthimion,J Ferron, E Fredrickson, D Gates, L Grisham, J Hosea, D Johnson, R Kaita, S Kubota,H Kugel, B LeBlanc, R Maingi, J Manickam, TK Mau, RJ Maqueda, E Mazzucato, J Menard,D Mueller, B Nelson, N Nishino, M Ono, F Paoletti, S Paul, YKM Peng, CK Phillips, R Ra-man, P Ryan, SA Sabbagh, M Schaffer, CH Skinner, D Stutman, D Swain, E Synakowski,Y Takase, J Wilgen, JR Wilson, W Zhu, S Zweben, A Bers, M Carter, B Deng, C Domier,E Doyle, M Finkenthal, K Hill, T Jarboe, S Jardin, H Ji, L Lao, KC Lee, N Luhmann, R Ma-jeski, S Medley, H Park, T Peebles, RI Pinsker, G Porter, A Ram, M Rensink, T Rognlien,D Stotler, B Stratton, G Taylor, W Wampler, GA Wurden, XQ Xu, L Zeng, and NSTXTeam. Phys. Plasmas, 8:1977, 2001.
[78] Alexander Kendl, Bruce D. Scott, and Tiago T. Ribeiro. Nonlinear gyrofluid computation ofedge localized ideal ballooning modes. Phys. Plasmas, 17(7), 2010.
[79] R. Khan, N. Mizuguchi, N. Nakajima, and T. Hayashi. Dynamics of the ballooning modeand the relation to edge-localized modes in a spherical tokamak. Phys. Plasmas, 14(6), 2007.
[80] E. J. Kim and P. H. Diamond. Mean shear flows, zonal flows, and generalized kelvin-helmholtzmodes in drift wave turbulence: A minimal model for l -¿ h transition. Physics of Plasmas,10(5):1698–1704, 2003. 2 44th Annual Meeting of the Division of Plasma of the American-Physical-Society Nov 11-15, 2002 Orlando, florida Amer Phys Soc, Div Plasma.
[81] E. J. Kim and P. H. Diamond. Zonal flows and transient dynamics of the l-h transition.Physical Review Letters, 90(18), 2003.
[82] A. Kirk, B. Koch, R. Scannell, H. R. Wilson, G. Counsell, J. Dowling, A. Herrmann, R. Mar-tin, M. Walsh, and MAST team. Evolution of filament structures during edge-localized modesin the MAST tokamak. PHYSICAL REVIEW LETTERS, 96(185001), 2006.
121
[83] A Kirk, HR Wilson, GF Counsell, R Akers, E Arends, SC Cowley, J Dowling, B Lloyd,M Price, M Walsh, and MAST Team. Spatial and temporal structure of edge-localizedmodes. PHYSICAL REVIEW LETTERS, 92(245002), 2004.
[84] C. H. Knapp and G. C. Carter. Generalized correlation method for estimation of time-delay.Ieee Transactions on Acoustics Speech and Signal Processing, 24(4):320–327, 1976.
[85] J. A. Krommes. Fundamental statistical descriptions of plasma turbulence in magnetic fields.Physics Reports-Review Section of Physics Letters, 360(1-4):1–352, 2002.
[86] John A. Krommes. The remarkable similarity between the scaling of kurtosis with squaredskewness for torpex density fluctuations and sea-surface temperature fluctuations. Phys.Plasmas, 15(3), 2008.
[87] R. Kube, O. E. Garcia, B. LaBombard, J. L. Terry, and S. J. Zweben. Blob sizes and velocitiesin the alcator c-mod scrape-off layer. Journal of Nuclear Materials, 438:S505–S508, 2013. S20th International Conference on Plasma-Surface Interactions in Controlled Fusion Devices(PSI) May 21-25, 2012 Forschungszentrum Julich, Aachen, GERMANY.
[88] B. Labit, I. Furno, A. Fasoli, A. Diallo, S. H. Mueller, G. Plyushchev, M. Podesta, andF. M. Poli. Universal statistical properties of drift-interchange turbulence in torpex plasmas.Physical Review Letters, 98(25), 2007.
[89] AW Leonard, N Asakura, JA Boedo, M Becoulet, GF Counsell, T Eich, W Fundamenski,A Herrmann, LD Hortou, Y Kamada, A Kirk, B Kurzan, A Loarte, J Neuhauser, I Nunes,N Oyama, RA Pitts, G Saibene, C Silva, PB Snyder, H Urano, MR Wade, HR Wilson, andPedestal Edge Phys ITPA Topical Gr. Survey of type i elm dynamics measurements. PlasmaPhys. Control. Fusion, 48(A149-A162), 2006.
[90] S. C. Liu, L. M. Shao, S. J. Zweben, G. S. Xu, H. Y. Guo, B. Cao, H. Q. Wang, L. Wang,N. Yan, S. B. Xia, W. Zhang, R. Chen, L. Chen, S. Y. Ding, H. Xiong, Y. Zhao, B. N. Wan,X. Z. Gong, and X. Gao. New dual gas puff imaging system with up-down symmetry onexperimental advanced superconducting tokamak. Rev. Sci. Instrum., 83(12), 2012.
[91] C. F. Maggi. Progress in understanding the physics of the h-mode pedestal and elm dynamics.Nucl. Fusion, 50(066001), 2010.
[92] R Maingi, CE Bush, ED Fredrickson, DA Gates, SM Kaye, BP LeBlanc, JE Menard,H Meyer, D Mueller, N Nishino, AL Roquemore, SA Sabbagh, K Tritz, SJ Zweben, MG Bell,T Biewer, JA Boedo, DW Johnson, R Kaita, HW Kugel, RJ Maqueda, T Munsat, R Raman,VA Soukhanovskii, T Stevenson, and D Stutman. H-mode pedestal, elm and power thresholdstudies in nstx. Nucl. Fusion, 45(1066-1077), 2005.
[93] R. Maingi, A. E. Hubbard, H. Meyer, J. W. Hughes, A. Kirk, R. Maqueda, J. L. Terry, AlcatorC-Mod Team, MAST Team, and NSTX Team. Comparison of small ELM characteristics andregimes in Alcator C-Mod, MAST and NSTX. Nucl. Fusion, 51(063036), 2011.
[94] M. A. Malkov and P. H. Diamond. Bifurcation and scaling of drift wave turbulence intensitywith collisional zonal flow damping. Physics of Plasmas, 8(9):3996–4009, 2001. Malkov,Mikhail/A-8445-2013 Malkov, Mikhail/0000-0001-6360-1987.
122
[95] P. Manz, G. S. Xu, B. N. Wan, H. Q. Wang, H. Y. Guo, I. Cziegler, N. Fedorczak, C. Holland,S. H. Mueller, S. C. Thakur, M. Xu, K. Miki, P. H. Diamond, and G. R. Tynan. Zonal flowtriggers the l-h transition in the experimental advanced superconducting tokamak. Physicsof Plasmas, 19(7), 2012. Manz, Peter/E-6992-2011; Xu, Guosheng/B-4857-2013.
[96] R. J. Maqueda, R. Maingi, and NSTX Team. Primary edge localize mode filament structurein the national spherical torus experiment. Phys. Plasmas, 16(056117), 2009.
[97] RJ Maqueda, GA Wurden, S Zweben, L Roquemore, H Kugel, D Johnson, S Kaye, S Sabbagh,and R Maingi. Edge turbulence measurements in nstx by gas puff imaging. Rev. Sci. Instrum.,72(931-934), 2001.
[98] G. R. McKee, P. Gohil, D. J. Schlossberg, J. A. Boedo, K. H. Burrell, J. S. deGrassie,R. J. Groebner, R. A. Moyer, C. C. Petty, T. L. Rhodes, L. Schmitz, M. W. Shafer, W. M.Solomon, M. Umansky, G. Wang, A. E. White, and X. Xu. Dependence of the l- to h-modepower threshold on toroidal rotation and the link to edge turbulence dynamics. NuclearFusion, 49(11), 2009. White, Anne/B-8990-2011.
[99] R. W. P. McWhirter and A. G. Hearn. A calculation of instantaneous population densitiesof excited levels of hydrogen-like ions in a plasma. Proceedings of the Physical Society ofLondon, 82(529):641, 1963.
[100] J. E. Menard, S. C. Jardin, S. M. Kaye, C. E. Kessel, and J. Manickam. Ideal mhd stabilitylimits of low aspect ratio tokamak plasmas. Nuclear Fusion, 37(5):595–610, 1997. Jardin,Stephen/E-9392-2010.
[101] K. Miki, P. H. Diamond, Oe D. Guercan, G. R. Tynan, T. Estrada, L. Schmitz, and G. S. Xu.Spatio-temporal evolution of the l -¿ i -¿ h transition. Phys. Plasmas, 19(9), 2012. Gurcan,Ozgur/A-1362-2013; Xu, Guosheng/B-4857-2013 Gurcan, Ozgur/0000-0002-2278-1544;.
[102] N. Mizuguchi, R. Khan, T. Hayashi, and N. Nakajima. Nonlinear simulation of edge-localizedmode in a spherical tokamak. Nucl. Fusion, 47(7):579–585, 2007.
[103] R. A. Moyer, K. H. Burrell, T. N. Carlstrom, S. Coda, R. W. Conn, E. J. Doyle, P. Gohil,R. J. Groebner, J. Kim, R. Lehmer, W. A. Peebles, M. Porkolab, C. L. Rettig, T. L. Rhodes,R. P. Seraydarian, R. Stockdale, D. M. Thomas, G. R. Tynan, and J. G. Watkins. Beyondparadigm - turbulence, transport, and the origin of the radial electric-field in low to highconfinement mode transitions in the diii-d to tokamak. Phys. Plasmas, 2(6):2397–2407, 1995.
[104] V Mukhovatov, M Shimada, A N Chudnovskiy, A E Costley, Y Gribov, G Federici, O Kar-daun, A S Kukushkin, A Polevoi, V D Pustovitov, Y Shimomura, T Sugie, M Sugihara, andG Vayakis. Overview of physics basis for iter. Plasma Phys. Control. Fusion, 45(12A):A235,2003.
[105] T. Munsat and S. J. Zweben. Rev. Sci. Instrum., 77:103501, 2006.
[106] J. R. Myra, D. A. D’Ippolito, D. P. Stotler, S. J. Zweben, B. P. LeBlanc, J. E. Menard, R. J.Maqueda, and J. Boedo. Blob birth and transport in the tokamak edge plasma: Analysis ofimaging data. Phys. Plasmas, 13(9), 2006.
[107] V. Naulin, O. E. Garcia, A. H. Nielsen, and J. J. Rasmussen. Statistical properties of transportin plasma turbulence. Physics Letters A, 321(5-6):355–365, 2004.
123
[108] J. Neuhauser, V. Bobkov, G. D. Conway, R. Dux, T. Eich, M. Garcia-Munoz, A. Herrmann,L. D. Horton, A. Kallenbach, S. Kalvin, G. Kocsis, B. Kurzan, P. T. Lang, M. Maraschek,H. W. Mueller, H. D. Murmann, R. Neu, A. G. Peeters, M. Reich, V. Rohde, A. Schmid,W. Suttrop, M. Tsalas, E. Wolfrum, and ASDEX Upgrade Team. Structure and dynamicsof spontaneous and induced elms on asdex upgrade. Nucl. Fusion, 48(045005), 2008.
[109] M. Ono, S. M. Kaye, Y. K. M. Peng, G. Barnes, W. Blanchard, M. D. Carter, J. Chrzanowski,L. Dudek, R. Ewig, D. Gates, R. E. Hatcher, T. Jarboe, S. C. Jardin, D. Johnson, R. Kaita,M. Kalish, C. E. Kessel, H. W. Kugel, R. Maingi, R. Majeski, J. Manickam, B. McCormack,J. Menard, D. Mueller, B. A. Nelson, B. E. Nelson, C. Neumeyer, G. Oliaro, F. Paoletti,R. Parsells, E. Perry, N. Pomphrey, S. Ramakrishnan, R. Raman, G. Rewoldt, J. Robinson,A. L. Roquemore, P. Ryan, S. Sabbagh, D. Swain, E. J. Synakowski, M. Viola, M. Williams,J. R. Wilson, and Nstx Team. Exploration of spherical torus physics in the nstx device.Nucl. Fusion, 40(3Y):557–561, 2000. Jardin, Stephen/E-9392-2010; pomphrey, neil/G-4405-2010; Sabbagh, Steven/C-7142-2011; Gates, David/K-3715-2012 Gates, David/0000-0001-5679-3124 3 17th IAEA Fusion Energy Conference Oct 19-24, 1998 Yokohama, japan Iaea.
[110] N Oyama. Progress and issues in understanding the physics of elm dynamics, elm mitigation,and elm control. Journal of Physics: Conference Series, 123(012002), 2008.
[111] N Oyama, N Asakura, AV Chankin, T Oikawa, M Sugihara, H Takenaga, K Itami, Y Miura,Y Kamada, K Shinohara, and JT-60 Team. Fast dynamics of type i elms and transport ofthe elm pulse in jt-60u. Nucl. Fusion, 44(582-592), 2004.
[112] S. J. P. Pamela, G. T. A. Huysmans, M. N. A. Beurskens, S. Devaux, T. Eich, S. Benkadda,and Jet Efda Contributors. Nonlinear mhd simulations of edge-localized-modes in jet. PlasmaPhys. Control. Fusion, 53(5), 2011. Joint Varenna-Lausanne Workshop on the Theory ofFusion Plasmas Aug 30-sep 03, 2010 Varenna, ITALY.
[113] Y. K. M. Peng and D. J. Strickler. Features of spherical torus plasmas. Nucl. Fusion,26(6):769–777, 1986.
[114] R. B. Rood. Numerical advection algorithms and their role in atmospheric transport andchemistry models. Reviews of Geophysics, 25(1):71–100, 1987. Rood, Richard/C-5611-2008Rood, Richard/0000-0002-2310-4262.
[115] D. W. Ross. On standard forms for transport-equations and quasi-linear fluxes. PlasmaPhysics and Controlled Fusion, 34(2):137–146, 1992.
[116] D. A. Russell, J. R. Myra, and D. A. D’Ippolito. Phys. Plasmas, 16:122304, 2009.
[117] PM Ryan, JR Wilson, DW Swain, RI Pinsker, MD Carter, D Gates, JC Hosea, TK Mau,JE Menard, D Mueller, SA Sabbagh, and JB Wilgen. Initial operation of the nstx phasedarray for launching high harmonic fast waves. FUSION ENGINEERING AND DESIGN,56-57(569-573), 2001.
[118] S. A. Sabbagh, A. C. Sontag, J. M. Bialek, D. A. Gates, A. H. Glasser, J. E. Menard,W. Zhu, M. G. Bell, R. E. Bell, A. Bondeson, C. E. Bush, J. D. Callen, M. S. Chu, C. C.Hegna, S. M. Kaye, L. L. Lao, B. P. Leblanc, Y. Q. Liu, R. Maingi, D. Mueller, K. C. Shaing,D. Stutman, K. Tritzs, and C. Zhang. Resistive wall stabilized operation in rotating high
124
beta nstx plasmas. Nucl. Fusion, 46(5):635–644, 2006. Sabbagh, Steven/C-7142-2011; Gates,David/K-3715-2012 Gates, David/0000-0001-5679-3124.
[119] O. Sakai, Y. Yasaka, and R. Itatani. High radial confinement mode induced by dc limiterbiasing in the hiei tandem mirror. Physical Review Letters, 70(26):4071–4074, 1993.
[120] F. Sattin, M. Agostini, P. Scarin, N. Vianello, R. Cavazzana, L. Marrelli, G. Serianni, S. J.Zweben, R. J. Maqueda, Y. Yagi, H. Sakakita, H. Koguchi, S. Kiyama, Y. Hirano, andJ. L. Terry. On the statistics of edge fluctuations: comparative study between various fusiondevices. Plasma Phys. Control. Fusion, 51(5), 2009.
[121] L. Schmitz, L. Zeng, T. L. Rhodes, J. C. Hillesheim, E. J. Doyle, R. J. Groebner, W. A.Peebles, K. H. Burrell, and G. Wang. Role of zonal flow predator-prey oscillations in triggeringthe transition to h-mode confinement. Physical Review Letters, 108(15), 2012.
[122] Y. Sechrest, T. Munsat, D. J. Battaglia, and S. J. Zweben. Two-dimensional characterizationof elm precursors in nstx. Nucl. Fusion, 52(12), 2012.
[123] Y. Sechrest, T. Munsat, D. A. D’Ippolito, R. J. Maqueda, J. R. Myra, D. Russell, and S. J.Zweben. Flow and shear behavior in the edge and scrape-off layer of l-mode plasmas innational spherical torus experiment. Phys. Plasmas, 18(1), 2011.
[124] Y. Seki, I. Aoki, N. Yamano, and T. Tabara. Preliminary comparison of radioactive wastedisposal cost for fusion and fission reactors. Journal of Fusion Energy, 16(3):205–210, 1997.
[125] Shahriar Shafiee and Erkan Topal. When will fossil fuel reserves be diminished? EnergyPolicy, 37(1):181–189, 2009.
[126] K. C. Shaing and E. C. Crume. Bifurcation-theory of poloidal rotation in tokamaks - a modelfor the l-h transition. Physical Review Letters, 63(21):2369–2372, 1989.
[127] I. Shesterikov, Y. Xu, M. Berte, P. Dumortier, M. Van Schoor, M. Vergote, B. Schweer, andG. Van Oost. Development of the gas-puff imaging diagnostic in the textor tokamak. Reviewof Scientific Instruments, 84(5), 2013.
[128] D. R. Smith, H. Feder, R. Feder, R. J. Fonck, G. Labik, G. R. McKee, N. Schoenbeck,B. C. Stratton, I. Uzun-Kaymak, and G. Winz. Overview of the beam emission spectroscopydiagnostic system on the national spherical torus experiment. Rev. Sci. Instrum., 81(10),2010. 18th Topical Conference on High-Temperature Plasma Diagnostics May 16-20, 2010Wildwood, NJ.
[129] D. R. Smith, R. J. Fonck, G. R. McKee, and D. S. Thompson. Diagnostic performance ofthe beam emission spectroscopy system on the national spherical torus experiment. Rev. Sci.Instrum., 83(10), 2012. 2 19th Topical Conference on High-Temperature Plasma DiagnosticsMay 06-10, 2012 Monterey, CA.
[130] D. R. Smith, R. J. Fonck, G. R. McKee, D. S. Thompson, R. E. Bell, A. Diallo, W. Gut-tenfelder, S. M. Kaye, B. P. LeBlanc, and M. Podesta. Characterization and parametricdependencies of low wavenumber pedestal turbulence in the national spherical torus experi-ment. Phys. Plasmas, 20(5), 2013.
125
[131] D. R. Smith, S. E. Parker, W. Wan, Y. Chen, A. Diallo, B. D. Dudson, R. J. Fonck,W. Guttenfelder, G. R. McKee, S. M. Kaye, D. S. Thompson, R. E. Bell, B. P. LeBlanc,and M. Podesta. Measurements and simulations of low-wavenumber pedestal turbulence inthe national spherical torus experiment. Nucl. Fusion, 53(11), 2013.
[132] P. B. Snyder, N. Aiba, M. Beurskens, R. J. Groebner, L. D. Horton, A. E. Hubbard, J. W.Hughes, G. T. A. Huysmans, Y. Kamada, A. Kirk, C. Konz, A. W. Leonard, J. Loennroth,C. F. Maggi, R. Maingi, T. H. Osborne, N. Oyama, A. Pankin, S. Saarelma, G. Saibene,J. L. Terry, H. Urano, and H. R. Wilson. Pedestal stability comparison and iter pedestalprediction. Nuclear Fusion, 49(8), 2009.
[133] P. B. Snyder, H. R. Wilson, and X. Q. Xu. Progress in the peeling-ballooning model of edgelocalized modes: Numerical studies of nonlinear dynamics. Phys. Plasmas, 12(5), 2005. 46thAnnual Meeting of the Division of Plasma Physics of the American-Physical-Society Nov15-19, 2004 Savannah, GA Amer Phys Soc.
[134] PB Snyder, HR Wilson, JR Ferron, LL Lao, AW Leonard, D Mossessian, M Murakami,TH Osborne, AD Turnbull, and XQ Xu. Elms and constraints on the h-mode pedestal:peeling-ballooning stability calculation and comparison with experiment. Nucl. Fusion,44(320-328), 2004.
[135] PB Snyder, HR Wilson, JR Ferron, LL Lao, AW Leonard, TH Osborne, AD Turnbull,D Mossessian, M Murakami, and XQ Xu. Edge localized modes and the pedestal: A modelbased on coupled peeling-ballooning modes. Phys. Plasmas, 9(2037-2043), 2002.
[136] S. Srinivasan and R. Chellappa. Noise-resilient estimation of optical flow by use of overlappedbasis functions. Journal of the Optical Society of America a-Optics Image Science and Vision,16(3):493–507, 1999.
[137] R. D. Stambaugh, S. M. Wolfe, R. J. Hawryluk, J. H. Harris, H. Biglari, S. C. Prager, R. J.Goldston, R. J. Fonck, T. Ohkawa, B. G. Logan, and E. Oktay. Enhanced confinement intokamaks. Physics of Fluids B-Plasma Physics, 2(12):2941–2960, 1990.
[138] D. P. Stotler, D. A. D’Ippolito, B. LeBlanc, R. J. Maqueda, J. R. Myra, S. A. Sabbagh, andS. J. Zweben. Three-dimensional neutral transport simulations of gas puff imaging experi-ments. Contributions to Plasma Physics, 44(1-3):294–300, 2004. 9th International Workshopon Plasma Edge Theory in Fusion Devices Sep 03-05, 2003 Univ Calif, San Diego, San Diego,CA UCSD, Engn Dept; Lawrence Livetmore Natl Lab; Off Fus Enegy, Dept Energy; Max-Planck Inst Plasma Phys.
[139] D. P. Stotler, C. Karney, R. Kanzleiter, and S. Jaishankar. Degas 2 User Manual. PrincetonPlasma Physics Laboratory, 2013.
[140] D. P. Stotler, C. H. Skinner, R. V. Budny, A. T. Ramsey, D. N. Ruzic, and R. B. Turkot.Modeling of neutral hydrogen velocities in the tokamak fusion test reactor. Phys. Plasmas,3(11):4084–4094, 1996.
[141] D.P. Stotler, J. Boedo, B. LeBlanc, R. J. Maqueda, and Z. J. Zweben. J. Nucl. Mater.,363:686, 2007.
[142] DP Stotler, B LaBombard, JL Terry, and SJ Zweben. J. Nucl. Mater., 313(1066), 2003.
126
[143] W Suttrop. The physics of large and small edge localized modes. Plasma Phys. Control.Fusion, 42(A1-A14), 2000.
[144] G. Taylor, R. E. Bell, J. C. Hosea, B. P. LeBlanc, C. K. Phillips, M. Podesta, E. J. Valeo,J. R. Wilson, J-W. Ahn, G. Chen, D. L. Green, E. F. Jaeger, R. Maingi, P. M. Ryan, J. B.Wilgen, W. W. Heidbrink, D. Liu, P. T. Bonoli, T. Brecht, M. Choi, and R. W. Harvey.Advances in high-harmonic fast wave physics in the National Spherical Torus Experiment.Phys. Plasmas, 17(056114), 2010.
[145] R. J. Taylor, M. L. Brown, B. D. Fried, H. Grote, J. R. Liberati, G. J. Morales, P. Pribyl,D. Darrow, and M. Ono. H-mode behavior induced by cross-field currents in a tokamak.Physical Review Letters, 63(21):2365–2368, 1989.
[146] J. L. Terry, I. Cziegler, A. E. Hubbard, J. A. Snipes, J. W. Hughes, M. J. Greenwald,B. LaBombard, Y. Lin, P. Phillips, and S. Wukitch. The dynamics and structure of edge-localized-modes in alcator c-mod. JOURNAL OF NUCLEAR MATERIALS, 363(994-999),2007.
[147] J. L. Terry, S. J. Zweben, O. Grulke, M. J. Greenwald, and B. LaBombard. Velocity fields ofedge/scrape-off-layer turbulence in alcator c-mod. Journal of Nuclear Materials, 337(1-3):322–326, 2005. 16th International Conference on Plasma Surface Interactions in Controlled FusionDevices May 24-28, 2004 Portland, ME MIT, Plasma Sci and Fus Ctr; US DOE.
[148] P. W. Terry. Rev. Mod. Phys., 72:109, 2000.
[149] The ASDEX team. THE H-MODE OF ASDEX. Nucl. Fusion, 29(1959-2040), 1989.
[150] C Torrence and GP Compo. A practical guide to wavelet analysis. BULLETIN OF THEAMERICAN METEOROLOGICAL SOCIETY, 79(61-78), 1998.
[151] G. R. Tynan, A. Fujisawa, and G. McKee. A review of experimental drift turbulence studies.Plasma Physics and Controlled Fusion, 51(11), 2009.
[152] G. R. Tynan, L. Schmidtz, L. Blush, J. A. Boedo, R. W. Conn, R. Doerner, R. Lehmer,R. Moyer, H. Kugel, R. Bell, S. Kaye, M. Okabayashi, S. Sesnic, and Y. Sun. Turbulentedge transport in the princeton beta-experiment-modified high confinement mode. Physicsof Plasmas, 1(10):3301–3307, 1994.
[153] G. R. Tynan, M. Xu, P. H. Diamond, J. A. Boedo, I. Cziegler, N. Fedorczak, P. Manz,K. Miki, S. Thakur, L. Schmitz, L. Zeng, E. J. Doyle, G. M. McKee, Z. Yan, G. S. Xu, B. N.Wan, H. Q. Wang, H. Y. Guo, J. Dong, K. Zhao, J. Cheng, W. Y. Hong, and L. W. Yan.Turbulent-driven low-frequency sheared e x b flows as the trigger for the h-mode transition.Nuclear Fusion, 53(7), 2013. Xu, Guosheng/B-4857-2013.
[154] M. R. Wade. Physics and engineering issues associated with edge localized mode control initer. FUSION ENGINEERING AND DESIGN, 84(178-185), 2009.
[155] F. Wagner. A quarter-century of h-mode studies. Plasma Phys. Control. Fusion, 49(12B):B1–B33, 2007. 34th European-Physical-Society Conference on Plasma Physics Jul 02-06, 2007Palace Culture and Sci, Warsaw, POLAND European Phys Soc; Inst Plasma Phys and LaserMicrofusion; Assoc EURATOM-IPPLM.
127
[156] F. Wagner, G. Becker, K. Behringer, D. Campbell, A. Eberhagen, W. Engelhardt, G. Fuss-mann, O. Gehre, J. Gernhardt, G. Vongierke, G. Haas, M. Huang, F. Karger, M. Keilhacker,O. Kluber, M. Kornherr, K. Lackner, G. Lisitano, G. G. Lister, H. M. Mayer, D. Meisel, E. R.Muller, H. Murmann, H. Niedermeyer, W. Poschenrieder, H. Rapp, H. Rohr, F. Schneider,G. Siller, E. Speth, A. Stabler, K. H. Steuer, G. Venus, O. Vollmer, and Z. Yu. Regime ofimproved confinement and high-beta in neutral-beam-heated divertor discharges of the asdextokamak. Physical Review Letters, 49(19):1408–1412, 1982.
[157] R. E. Waltz, G. D. Kerbel, and J. Milovich. Toroidal gyro-landau fluid model turbulencesimulations in a nonlinear ballooning mode representation with radial modes. Physics ofPlasmas, 1(7):2229–2244, 1994.
[158] J. Wesson. Tokamaks. Oxford University Press, 2011.
[159] R. P. Wildes, M. J. Amabile, A. M. Lanzillotto, and T. S. Leu. Recovering estimates of fluidflow from image sequence data. Computer Vision and Image Understanding, 80(2):246–266,2000.
[160] H. R. Wilson, S. C. Cowley, A. Kirk, and P. B. Snyder. Magneto-hydrodynamic stabilityof the h-mode transport barrier as a model for edge localized modes: an overview. PlasmaPhys. Control. Fusion, 48(A71-A84), 2006.
[161] G. S. Xu, L. M. Shao, S. C. Liu, H. Q. Wang, B. N. Wan, H. Y. Guo, P. H. Diamond, G. R.Tynan, M. Xu, S. J. Zweben, V. Naulin, A. H. Nielsen, J. Juul Rasmussen, N. Fedorczak,P. Manz, K. Miki, N. Yan, R. Chen, B. Cao, L. Chen, L. Wang, W. Zhang, and X. Z.Gong. Study of the l-i-h transition with a new dual gas puff imaging system in the eastsuperconducting tokamak. Nucl. Fusion, 54(1), 2014.
[162] G. S. Xu, B. N. Wan, H. Q. Wang, H. Y. Guo, H. L. Zhao, A. D. Liu, V. Naulin, P. H.Diamond, G. R. Tynan, M. Xu, R. Chen, M. Jiang, P. Liu, N. Yan, W. Zhang, L. Wang,S. C. Liu, and S. Y. Ding. First evidence of the role of zonal flows for the l-h transition atmarginal input power in the east tokamak. Physical Review Letters, 107(12), 2011. Naulin ,Volker/A-2419-2012; Liu, Ah Di/N-7124-2013; Xu, Guosheng/B-4857-2013.
[163] X. Q. Xu, B. D. Dudson, P. B. Snyder, M. V. Umansky, H. R. Wilson, and T. Casper.Nonlinear elm simulations based on a nonideal peeling-ballooning model using the bout plusplus code. Nucl. Fusion, 51(10), 2011.
[164] J. H. Yu, C. Holland, G. R. Tynan, G. Antar, and Z. Yan. Examination of the velocity time-delay-estimation technique. Journal of Nuclear Materials, 363:728–732, 2007. Yan, Zheng/E-7005-2011 17th International Conference on Plasma-Surface Interactions in Controlled FusionDevices May 22-26, 2006 Hefei, PEOPLES R CHINA Inst Plasma Phys; Chinese Acad Sci,Bur Int Cooperat; Natl Nat Sci Fdn China.
[165] A. Zhitlukhin, N. Klimov, I. Landman, J. Linke, A. Loarte, M. Merola, V. Podkovyrov,G. Federici, B. Bazylev, S. Pestchanyi, V. Safronov, T. Hirai, V. Maynashev, V. Levashov,and A. Muzichenko. Effects of elms on iter divertor armour materials. JOURNAL OFNUCLEAR MATERIALS, 363(301-307), 2007.
[166] H Zohm. Edge localized modes (elms). Plasma Phys. Control. Fusion, 38(105-128), 1996.
128
[167] S. J. Zweben, R. J. Maqueda, R. Hager, K. Hallatschek, S.M. Kaye, T. Munsat, F. M. Poli,A.L. Roquemore, Y. Sechrest, and D.P. Stotler. Phys. Plasmas, 17:102502, 2010.
[168] S. J. Zweben, D. P. Stotler, R. E. Bell, W. M. Davis, S. M. Kaye, B. P. LeBlanc, R. J.Maqueda, E. T. Meier, T. Munsat, Y. Ren, S. A. Sabbagh, Y. Sechrest, D. R. Smith, andV. Soukhanovskii. Effect of a deuterium gas puff on the edge plasma in nstx. Plasma Phys.Control. Fusion, 56(9), 2014.
[169] S. J. Zweben, D. P. Stotler, J. L. Terry, B. LaBombard, M. Greenwald, M. Muterspaugh, C. S.Pitcher, K. Hallatschek, R. J. Maqueda, B. Rogers, J. L. Lowrance, V. J. Mastrocola, G. F.Renda, and C. Mod Grp Alcator. Edge turbulence imaging in the alcator c-mod tokamak.Phys. Plasmas, 9(5):1981–1989, 2002. 2 43rd Annual Meeting of the Division of PlasmaPhysics of the American-Physical-Society Oct 29-nov 02, 2001 Long beach, california AmerPhys Soc, Div Plasma Phys.
[170] SJ Zweben, RJ Maqueda, DP Stotler, A Keesee, J Boedo, CE Bush, SM Kaye, B LeBlanc,JL Lowrance, VJ Mastrocola, R Maingi, N Nishino, G Renda, DW Swain, JB Wilgen, andNSTX Team. Nucl. Fusion, 44:134, 2004.