Dynamics of Plasma Structures Interactin

Dynamics of Plasma Structures Interacting with External and Self-Generated Magnetic Fields - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Dynamics of Plasma Structures Interacting with External and Self-Generated Magnetic Fields
Citation
Yun, Gunsu Soonshin
(2008)
Dynamics of Plasma Structures Interacting with External and Self-Generated Magnetic Fields.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/9CMD-C377.
Abstract
Plasmas interacting with external and self-generated magnetic fields often develop a long tubular structure of nearly uniform cross section. Such long collimated plasma tubes have been observed in a variety of contexts ranging from astrophysical plasma jets (10
15
–10
22
m) to solar coronal loops (10
–10
m). Remarkably, much smaller-sized plasmas (0.1–1 m) produced by the Caltech planar spheromak gun develop collimated structures bearing a striking resemblance to these natural plasma tubes. This thesis presents experimental observations of gun-produced plasma tubes that support a recently-proposed magnetohydrodynamic (MHD) pumping model as a universal collimation mechanism. For any flared flux tube carrying a finite axial current, the model predicts (i) magnetic pumping of plasma particles from a constricted region into a bulged region and (ii) tube collimation if the flow slows down at the bulged region, leading to accumulation of mass and thus concentrating the azimuthal magnetic flux frozen in the mass flow (i.e., increasing the pinch force). Time- and space-resolved spectroscopic measurements of gun-produced plasmas show (i) suprathermal Alfvenic flow (30–50 km/s), (ii) large density amplification from ~10
17
to ~10
22
-3
in an Alfvenic time scale (5–10 µs), and (iii) flow slowing down and mass accumulation at the flow front, the place where the tube collimation occurs according to high-speed camera imaging. These observations are consistent with the predictions of the MHD pumping model, and thus the model offers valuable insight into the formation mechanism of laboratory, solar, and astrophysical plasma structures.
Item Type:
Thesis (Dissertation (Ph.D.))
Subject Keywords:
MHD; plasma diagnostics; plasma gun; plasma jet; spheromak; Stark broadening
Degree Grantor:
California Institute of Technology
Division:
Engineering and Applied Science
Major Option:
Applied Physics
Thesis Availability:
Public (worldwide access)
Research Advisor(s):
Bellan, Paul Murray
Thesis Committee:
Bellan, Paul Murray (chair)
Meier, David L.
Polk, James E.
Shepherd, Joseph E.
Gould, Roy Walter
Defense Date:
12 July 2007
Record Number:
CaltechETD:etd-07242007-162442
Persistent URL:
DOI:
10.7907/9CMD-C377
ORCID:
Author
ORCID
Yun, Gunsu Soonshin
0000-0002-1880-5865
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
2979
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
07 Aug 2007
Last Modified:
05 Feb 2020 00:55
Thesis Files
Preview
PDF
- Final Version
See Usage Policy.
3MB
Repository Staff Only:
item control page
CaltechTHESIS is powered by
EPrints 3.3
which is developed by the
School of Electronics and Computer Science
at the University of Southampton.
More information and software credits
Dynamics of Plasma Structures Interacting with
External and Self-Generated Magnetic Fields

Thesis by

Gunsu Soonshin Yun
In Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy

California Institute of Technology
Pasadena, California

2008
(Defended July 12, 2007)

c 2008
Gunsu Soonshin Yun

iii

This thesis is dedicated in loving memory to my father,
who devoted his life to family.

Acknowledgements
My first encounter with Caltech was a pleasing scent surrounding the campus in a
night of March, 2002. Although it aggravated my allergy in the next few years, my
memories of Caltech are filled with gratefulness as pleasant as the scent thanks to the
people who have been with me.
I am deeply grateful to my advisor, Prof. Paul M. Bellan. In classroom, laboratory,
and office, his thoughtful mentorship for my academic progress and genuine interest
in science always encouraged me to endeavor in research. I find it still amazing
and grateful at the same time that, whenever I made a mistake in experiment, Paul
always encouraged me by telling a story about himself or someone else making a
similar mistake. I also enjoyed his many entertaining stories during lunch.
I am indebted to two post-docs, Setthivoine You and Shreekrishna Tripathi. Their
mentorship, collaboration, and numerous efforts to improve the experiment have
greatly benefited this thesis work. Working with and learning from Sett were as
exciting as watching the 2006 World Cup soccer games on his projection screen. We
were both allergic to our office and I thank Sett for sharing his allergy pills at the
time when I was afraid of my wife’s acupuncture. Collaborating with Shreekrishna
was always as entertaining as listening to his amazing stories about his village in the
northern India. I thank Shreekrishna and his wife Rashmi for inviting our group to
their home with warmth and authentic Indian foods.
I enjoyed lively companionship with my fellow graduate students including Carlos
Romero-Talamás, Steven Pracko, Eli Jorne, Rory Perkins, Auna Moser, Eve Stenson,
and Deepak Kumar. Carlos saved me from my first mishap in the lab by removing
a broken vacuum bolt stuck in a thread using a vise-grip, which became my favorite

tool afterwards. Steve always provided me a helping hand for finding anything in
the lab. Steve’s humorous anecdotes about predecessors, lab equipments, and past
accidents made the lab a lively workplace. Eli entertained us with his witty songs,
often intended to mock one of us, and has continued to do so even after leaving Caltech
to pursue his talent at the film industry. Rory entertained us in a very different way
with his electric guitar and his choice heavy-metallic music. He must have found
a beauty in the often disturbing music, considering his thorough understanding of
physics. Auna amazed me by her agility to move around a maze of probes inside the
vacuum chamber. I enjoyed playing Tak-zi with Auna although her Tak-zi skill is not
as good as her agility in the chamber. Eve invigorated our workplace with her plants
and her acrobatic stunt in the hallway. I thank Eve for tolerating my golf swing
practice in the office. Deepak collaborated with me in numerous experiments and
inspired me to enjoy the life at Caltech beyond academia. Deepak almost transformed
me into an Indian who loves Bollywood more than Hollywood.
I owe my safety to Mr. Dave Felt in the machine room since he taught me to
use machine tools properly. Dave single-handedly built numerous gadgets for the
experiment and fixed the power supplies we often blew up. I thank Mses. Connie
Rodriguez, Eleonora Vorobieff, Irene Loera, Mary Metz, and Cierina Marks for helping me order lab supplies and organize paperworks. I appreciate Connie’s enthusiasm
evidenced by her omnipresence in the Watson building.
I am deeply indebted to Dr. Seung-soon Jang for his brotherly love and words
of wisdom that strengthened my inner being and to my friends Chihoon Ahn and
Byung-joon Yoon for their friendship and numerous help throughout the years. I’d
also like to thank John Choi, Frederick Balagadde, and many others for their prayers
during the difficult time of the loss of my father Jong-wook.
Finally, my wife Hyun-woo, my mother Ki-jeon, my sisters Young-mi and Youngsun, and my parents-in-law Pil-soo and Myung-in deserve my sincere gratitude for
their tremendous support and sacrifice. I am breathing the pleasing scent again
thanks to my wife’s cure of my allergy and will breath the pleasing memories of
Caltech throughout my entire life.

Abstract
Plasmas interacting with external and self-generated magnetic fields often develop a
long tubular structure of nearly uniform cross section. Such long collimated plasma
tubes have been observed in a variety of contexts ranging from astrophysical plasma
jets (1015 –1022 m) to solar coronal loops (107 –108 m). Remarkably, much smaller-sized
plasmas (0.1–1 m) produced by the Caltech planar spheromak gun develop collimated
structures bearing a striking resemblance to these natural plasma tubes. This thesis presents experimental observations of gun-produced plasma tubes that support a
recently-proposed magnetohydrodynamic (MHD) pumping model as a universal collimation mechanism. For any flared flux tube carrying a finite axial current, the model
predicts (i) magnetic pumping of plasma particles from a constricted region into a
bulged region and (ii) tube collimation if the flow slows down at the bulged region,
leading to accumulation of mass and thus concentrating the azimuthal magnetic flux
frozen in the mass flow (i.e., increasing the pinch force). Time- and space-resolved
spectroscopic measurements of gun-produced plasmas show (i) suprathermal Alfvénic
flow (30–50 km/s), (ii) large density amplification from ∼1017 to ∼1022 m−3 in an
Alfvénic time scale (5–10 µs), and (iii) flow slowing down and mass accumulation
at the flow front, the place where the tube collimation occurs according to highspeed camera imaging. These observations are consistent with the predictions of the
MHD pumping model, and thus the model offers valuable insight into the formation
mechanism of laboratory, solar, and astrophysical plasma structures.

vii

Contents
Acknowledgements

Abstract

1 Background

1.1

1.2

Spheromaks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.1

Magnetic confinement fusion . . . . . . . . . . . . . . . . . . .

1.1.2

Relaxed state of magnetized plasma . . . . . . . . . . . . . . .

1.1.3

Formation schemes and role of flux conserver . . . . . . . . . .

Coplanar coaxial plasma gun – spheromak formation without flux conserver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.3

Collimated plasma structures . . . . . . . . . . . . . . . . . . . . . .

1.4

MHD pumping mechanism . . . . . . . . . . . . . . . . . . . . . . . .

1.4.1

Frozen-in flux – magnetic Reynolds number . . . . . . . . . . .

1.4.2

MHD pumping and collimation model . . . . . . . . . . . . .

1.5

Overview of the following chapters

. . . . . . . . . . . . . . . . . . .

2 Experimental setup – boundary conditions

16
17

2.1

Vacuum chamber . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

Spheromak plasma gun . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2.1

Coplanar coaxial electrodes . . . . . . . . . . . . . . . . . . .

2.2.2

Stuffing flux system . . . . . . . . . . . . . . . . . . . . . . . .

Gas injection system . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.1

2.3

Flexible plumbing . . . . . . . . . . . . . . . . . . . . . . . . .

viii
2.3.2

Fast gas valve . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.4

Main timing – gas breakdown sequence . . . . . . . . . . . . . . . . .

2.5

Plasma formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 Diagnostics

3.1

VME digitizer system . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2

High-voltage probe . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.3

High current transducer – Rogowski coil . . . . . . . . . . . . . . . . .

3.4

Multichannel magnetic probe . . . . . . . . . . . . . . . . . . . . . .

3.5

Timing of diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.6

Fast digital cameras . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.7

Spectroscopic system . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.7.1

Entrance optics . . . . . . . . . . . . . . . . . . . . . . . . . .

3.7.2

Fiber array . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.7.3

Visualizing lines of sight . . . . . . . . . . . . . . . . . . . . .

4 Observations

4.1

Evolution of plasma structures . . . . . . . . . . . . . . . . . . . . . .

4.2

Magnetic flux tube collimation – spider legs . . . . . . . . . . . . . .

4.2.1

Main observations . . . . . . . . . . . . . . . . . . . . . . . . .

4.2.2

Doppler shift measurements . . . . . . . . . . . . . . . . . . .

4.2.3

Flow velocity profile along the spider leg . . . . . . . . . . . .

4.2.4

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Large density amplification in the plasma jets . . . . . . . . . . . . .

4.3.1

Large Stark broadening of hydrogen spectral lines . . . . . . .

4.3.2

Density measurements . . . . . . . . . . . . . . . . . . . . . .

4.3.3

Density and velocity profile along the jet axis . . . . . . . . .

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.4.1

MHD pumping . . . . . . . . . . . . . . . . . . . . . . . . . .

4.4.2

MHD collimation . . . . . . . . . . . . . . . . . . . . . . . . .

4.4.3

Total particle flux by the MHD pumping . . . . . . . . . . . .

4.3

4.4

ix
4.4.4

Application – tokamak fueling . . . . . . . . . . . . . . . . . .

5 Stark broadening

63
64

5.1

Quadratic Stark effect . . . . . . . . . . . . . . . . . . . . . . . . . .

5.2

Linear Stark effect . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.3

Line broadening by charged particles . . . . . . . . . . . . . . . . . .

5.3.1

Quasi-static approximation – ion microfield . . . . . . . . . . .

5.3.2

Impact approximation – electron microfield . . . . . . . . . . .

Density diagnosis using the Stark broadened profiles . . . . . . . . . .

5.4

6 Analysis of spectral line profiles
6.1

Density estimation from spectral line profiles . . . . . . . . . . . . . .

6.1.1

Uncertainty in Stark parameter . . . . . . . . . . . . . . . . .

6.1.2

Competing broadening effects . . . . . . . . . . . . . . . . . .

6.1.3

Plasma inhomogeneity . . . . . . . . . . . . . . . . . . . . . .

6.1.4

Self-absorption, continuum background, and blending with nearby
spectral lines . . . . . . . . . . . . . . . . . . . . . . . . . . .

6.2

Deconvolution of Stark broadening – simple formula . . . . . . . . . .

6.3

Maximum entropy deconvolution . . . . . . . . . . . . . . . . . . . .

6.3.1

Undoing the instrumental broadening – an inverse problem . .

6.3.2

Constrained inverse problem . . . . . . . . . . . . . . . . . . .

6.3.3

Numerical test

. . . . . . . . . . . . . . . . . . . . . . . . . .

Theoretical fitting of Stark profile . . . . . . . . . . . . . . . . . . . .

6.4

7 Summary and conclusion

Appendices

A Leak analysis

A.1 Equilibrium pressure; pumping, leak, and wall desorption . . . . . . .

A.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

B Measurement of gas cloud profile

B.1 Fast ionization gauge (FIG) . . . . . . . . . . . . . . . . . . . . . . .

B.2 Gas profile modeling . . . . . . . . . . . . . . . . . . . . . . . . . . .

C Linear dispersion of Czerny-Turner spectrometer

101

D Large vacuum viewport

104

E Selection rules and commutation relations

107

E.1 Particle in a perturbed central potential . . . . . . . . . . . . . . . . 107
E.2 Selection rule for the quantum number m . . . . . . . . . . . . . . . . 108
E.3 Selection rule for the quantum number l . . . . . . . . . . . . . . . . 109
F Concavity of the maximum entropy potential function Z

111

Bibliography

113

List of Figures
1.1

Flux surfaces of an ideal spheromak

. . . . . . . . . . . . . . . . . . .

1.2

Spheromak formation scheme . . . . . . . . . . . . . . . . . . . . . . .

1.3

Coaxial magnetized plasma gun . . . . . . . . . . . . . . . . . . . . . .

1.4

Solar coronal loops and astrophysical jet . . . . . . . . . . . . . . . . .

1.5

MHD pumping and collimation model . . . . . . . . . . . . . . . . . .

2.1

Vacuum chamber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

Planar spheromak plasma gun . . . . . . . . . . . . . . . . . . . . . . .

2.3

Gas line plumbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.4

Fast gas valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.5

Timing configuration for gas breakdown and diagnotics . . . . . . . . .

2.6

Neutral gas density profile and bias field lines at breakdown . . . . . .

2.7

Spider legs – eight arch-shaped plasma-filled flux tubes . . . . . . . . .

3.1

Typical voltage and current traces of the plasma gun . . . . . . . . . .

3.2

Passive integrator circuit for the gun current Rogowski coil . . . . . . .

3.3

Rogowski coil calibration . . . . . . . . . . . . . . . . . . . . . . . . . .

3.4

Magnetic field structure . . . . . . . . . . . . . . . . . . . . . . . . . .

3.5

High-resolution multichannel spectroscopic system . . . . . . . . . . .

3.6

Resolution of the spectroscopic system . . . . . . . . . . . . . . . . . .

3.7

Fiber array and fiber bundle . . . . . . . . . . . . . . . . . . . . . . . .

3.8

Lines of sight of the spectroscopic system . . . . . . . . . . . . . . . .

4.1

Evolving plasma structure (hydrogen plasma) . . . . . . . . . . . . . .

4.2

Current and voltage traces . . . . . . . . . . . . . . . . . . . . . . . . .

xii
4.3

Image of collimated spider legs . . . . . . . . . . . . . . . . . . . . . .

4.4

NII spectra showing blue Doppler shifts . . . . . . . . . . . . . . . . . .

4.5

Flow velocities along the spider leg . . . . . . . . . . . . . . . . . . . .

4.6

Doppler shifts of Hβ lines . . . . . . . . . . . . . . . . . . . . . . . . .

4.7

Hβ profile showing the characteristic central dip . . . . . . . . . . . . .

4.8

Electron density of a hydrogen plasma jet vs. time . . . . . . . . . . .

4.9

Stark broadened NII spectral lines . . . . . . . . . . . . . . . . . . . . .

4.10

Electron density of a straight column nitrogen plasma jet vs. time . . .

4.11

Density and velocity along the axis of a straight column nitrogen plasma
jet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.12

Detailed view of the density and velocity profiles . . . . . . . . . . . .

4.13

Image of a nitrogen plasma jet . . . . . . . . . . . . . . . . . . . . . .

4.14

Sketch of jet collimation . . . . . . . . . . . . . . . . . . . . . . . . . .

4.15

Sketch of spider leg collimation . . . . . . . . . . . . . . . . . . . . . .

5.1

Nearest neighbor approximation of the ionic electric field . . . . . . . .

5.2

Holtsmark profile vs. nearest neighbor profile . . . . . . . . . . . . . .

6.1

Reduced half-width α1/2 . . . . . . . . . . . . . . . . . . . . . . . . . .

6.2

Distortion of spectral line profiles due to inhomogeneous plasma density

6.3

Impurity CIII (229.687 nm) line width compared to a Stark broadened
NII line profile (424.178 nm) . . . . . . . . . . . . . . . . . . . . . . . .

6.4

Steepest decent method . . . . . . . . . . . . . . . . . . . . . . . . . .

6.5

Maximum entropy deconvolution (1D) applied to spectral lines

. . . .

6.6

Maximum entropy deconvolution (2D) applied to camera images . . . .

B.1

Fast ion gauge (FIG) pressure measurement . . . . . . . . . . . . . . . 100

C.1

Czerny-Turner spectrometer configuration . . . . . . . . . . . . . . . . 103

D.1

Large vacuum viewport . . . . . . . . . . . . . . . . . . . . . . . . . . 105

Chapter 1
Background
This chapter provides background information on the spheromak formation experiment and the laboratory simulation of astrophysical jets at the Caltech Bellan plasma
group. In section 1, the concept of magnetic confinement of plasmas for thermonuclear
fusion is introduced and the concept of spheromaks is reviewed as a natural magnetic
confinement scheme. In section 2, a novel planar plasma gun used in the spheromak formation experiment is introduced. In section 3, the observation of collimated
plasma structures produced by the plasma gun, which has led to the laboratory simulation of astrophysical jets, is described. In the following section, a universal MHD
pumping model which explains the collimation process of plasma-filled magnetic flux
tubes is summarized. In the last section, an overview of the following chapters is
given.

1.1

Spheromaks

1.1.1

Magnetic confinement fusion

Nuclear fusion is the merging of two atomic nuclei to form a heavier nucleus. Sustained
nuclear fusion can release a huge amount of energy as in the Sun. The fusion reaction
rate f per volume is
f = n1 n2 hσvi,

(1.1)

where n1 , n2 are the number densities of the reactant nuclei, σ is the fusion cross
section, and v is the thermal velocity of the reactants. The term hσvi becomes
significant only at very high temperatures of 10–100 keV. Sustainable fusion reaction
requires confinement of the reactant particles at these high temperatures and thus
necessarily occurs in a plasma state. Furthermore, a net power production can be
achieved only if the total fusion power Pfusion exceeds the total power loss Ploss , yielding
the Lawson criterion:
Pfusion = f Ef V ∼ n2e hσviEf V,

=⇒ ne τE &

Ploss ∼ Esystem /τE = V ne kT /τE
kT
hσviEf

(1.2)
(1.3)

where τE is the confinement time of the high temperature plasma, Esystem is the total
energy content of the system of volume V , Ef is the energy released by a single
fusion reaction, ne is the electron density, and kT is the average thermal energy of
the particles. The critical ne τE is ∼1.5 × 1020 sec·m−3 for the deuterium–tritium
reaction.
In the Sun, the gravitational force provides enough confinement time τE to satisfy
the Lawson criterion. However, in terrestrial environment, gravitational confinement
is impossible and other confinement mechanisms are required. Since plasmas are a
good electrical conductor, plasmas can be confined in a magnetic field. A variety of
magnetic confinement devices such as tokamaks, stellarators, and magnetic mirrors
have been constructed since 1950s and much effort has been devoted to solve many
engineering challenges such as construction of large magnetic coils. Relatively simple designs which require less engineering have been developed later as alternatives.
Examples are reverse field pinches, field reversed configurations, and spheromaks [1].
These are collectively called compact toroids [2] because of their compact dimensions
compared to tokamaks.
Spheromaks [1] are a naturally occurring plasma confinement structure such as
solar prominences [3]. Spheromaks are defined as a plasma structure confined in a
magnetic field configuration characterized by toroidal magnetic flux surfaces bounded

by a spherical separatix surface as illustrated in figure 1.1. The bounding surface is
simply connected and has no toroidal magnetic field. In contrast to tokamaks and
other magnetic confinement configurations that require large external coils, spheromaks are sustained entirely by the magnetic field produced by the internal currents.
In particular, the toroidal magnetic field of spheromaks necessary for particle confinement is produced by the large axial current flowing through the axis of symmetry.

1.1.2

Relaxed state of magnetized plasma

Magnetic helicity [1, chap. 3] is a most important quantity in the theory of magnetic
confinement because the plasma decays into a self-organized relaxed state after undergoing complex instabilities while approximately preserving the total helicity content.
The total magnetic helicity K is defined as
K=

A · Bd3 r,

(1.4)

where B is the magnetic field strength, A is the magnetic vector potential, and the
integration is over the system volume V .
For an open system linked to the environment by two bounding surfaces, the
helicity can be injected into the system by applying an electrical potential ∆ϕ across
the surfaces:
dK
= 2ψ∆ϕ − 2
dt

ηJ · Bd3 r,

(1.5)

where ψ is the magnetic flux intercepting the surfaces, η is the electrical resistivity
of the plasma, and J is the current density. The term 2ψ∆ϕ describes the rate of
helicity injection and the volume integral describes the resistive decay of helicity.
The resistive helicity decay rate is small compared to the magnetic energy decay
rate [1, chap. 4]. Thus, the system will evolve into a minimum energy state while
preserving the injected helicity. Using the method of Lagrange multipliers, the minR
imum energy state is obtained by minimizing W − λK, where W = d3 rB 2 /2µ0 is

the magnetic energy and λ is the Lagrange multiplier [1, chap. 4]:

0 = δW − λδK = d rB · δB − λ d3 r(A · δB + B · δA)
= d rB · ∇ × δA − λ d3 r(A · ∇ × δA + B · δA).

(1.6)

Using the vector identities
B · ∇ × δA = ∇ · (δA × B) + δA · ∇ × B,
A · ∇ × δA = ∇ · (δA × A) + δA · ∇ × A = ∇ · (δA × A) + δA · B,
and then integrating by parts yields
0=

dr3 δA · (∇ × B − 2λB),

(1.7)

where the surface integral terms ( (δA × B) · ds and (δA × A) · ds) vanish due to
the boundary condition. Thus, after redefining λ, the problem reduces to
∇ × B = λB.

(1.8)

The minimum energy states satisfying this condition are called Taylor states. Since
the current density J = ∇ × B/µ0 k B, the J × B force vanishes and so the Taylor
states are force-free.
For axis-symmetric systems, the solution of equation (1.8) can be expressed in
terms of the poloidal flux ψ as [1, chap. 4]
B=

(∇ψ × ∇φ + λψ∇φ),
2π

(1.9)

where φ indicates the azimuthal (toroidal) angle. If the system is confined in a
conducting spherical volume of radius a, the poloidal flux function is
ψ(r, θ) = 2πaB0 rj1 (λr) sin2 θ,

(1.10)

1.0

0.5

0.0

-0.5

-1.0
-1.0

-0.5

0.0

0.5

1.0

Figure 1.1: Flux surfaces of an ideal spheromak. Thick solid circle represents the
boundary of the conducting spherical wall which corresponds to the bounding separatix surface ψ = 0. Vertical dashed line is the axis of symmetry. Length scales are
normalized by the radius a of the sphere.
where r is the distance from the origin, θ is the angle from the axis of symmetry, and
j1 is the first-order spherical Bessel function (j1 (x) = x−2 sin x − x−1 cos x).
Spheromaks are represented by equations (1.9) and (1.10) in its most idealized
form (i.e., spherical shape) and are one of the force-free Taylor states. The flux
surfaces of an idealized spheromak are illustrated in figure 1.1.

1.1.3

Formation schemes and role of flux conserver

Helicity must be injected into the plasma before it can become a spheromak. A
variety of injection schemes have been successful in producing spheromak plasmas
such as magnetized coaxial gun, z-θ pinch, conical θ pinch, and flux core [4]. The
success of these many injection methods implies that spheromaks are one of the most
natural plasma structures. All the injection methods share a common feature and

can be summarized in figure 1.2. In the figure, each electrode intercepts vacuum
magnetic field lines and thus has a finite magnetic flux ψ. The vacuum magnetic
field is generated by external coils. When an electric potential ∆ϕ is applied across
the electrodes, helicity will be injected according to equation (1.5), dK/dt = 2ψ∆ϕ.
Since the plasma is conducting, the helicity injection induces a current flowing along
the field lines. After the helicity injection, the plasma dissipates its magnetic energy
via various instabilities and relaxes into the force-free spheromak state.
In practice, most spheromak formation schemes also include a conducting surface
called flux conserver to stabilize spheromaks against instabilities such as tilt and shift
modes [5] [1, chap. 10]. The wall image currents of the flux conserver interact with
the spheromak and confine it within the flux conserver volume.
The eigenvalue λ in the force-free equation (1.8) is determined by the geometrical
shape of the wall confining the spheromak, typically the flux conserver. For example,
λ = y11 /a, for a spherical volume of radius a,
λ = (x11 /a)2 + (π/h)2 , for a cylinder of radius a and height h,

(1.11)
(1.12)

where y11 is the first zero of the spherical Bessel function j1 , x11 is the first zero of
the Bessel function J1 , and the aspect ratio of the cylinder h/a < 1.7.

1.2

Coplanar coaxial plasma gun – spheromak formation without flux conserver

The most common and least complex method to generate spheromaks is to use the
magnetized coaxial gun [6]. The device is composed of a pair of concentric cylindrical
electrodes (inner and outer) linked by a vacuum magnetic field produced by an external coil as shown schematically in figure 1.3a. The eigenvalue λ of the spheromak
produced by the coaxial gun is related to the total gun current Igun and the total

magnetic
field lines

current
plasma

anode (+)

cathode (-)

Figure 1.2: Generic spheromak formation scheme (adapted from [1, chap. 7]). The
externally generated vacuum magnetic field lines are indicated by dashed lines. An
electric potential applied along the field injects helicity according to equation (1.5).
Since the plasma is conducting, a field-aligned current is also generated.
magnetic flux intercepting the inner electrode as
λ=

k∇ × Bk
µ0 J
µ0 JSgun
µ0 Igun
BSgun
ψgun

(1.13)

where Sgun is the surface area of the gun. Thus, it is convenient to define this gun
source property as
λgun ≡

µ0 Igun
ψgun

(1.14)

For clarification, the spheromak λ will be denoted as λsph .
A modified coaxial gun design composed of planar electrodes instead of cylindrical
electrodes has been proposed by Bellan and constructed by Hsu and Bellan [7] [8] [9]
(see figure 2.2). The planar configuration has three main advantages.
(1) The planar gun has larger area and thus requires smaller magnetic field to
provide the same ψgun .
(2) Improved λ-matching [1, chap. 8 and 12]: The eigenvalue λ is an intensive property of the plasma like the temperature in thermodynamics. The eigenvalue λ
comes about by minimizing the magnetic energy under a constant helicity and
the temperature by maximizing the entropy under a constant internal energy.
Helicity flows along λ gradients to minimize the magnetic energy much like

heat (energy) flows along temperature gradients to maximize the entropy. Conventional coaxial guns are much smaller in volume than the spheromaks they
produce, and so λgun poorly matches λsph since λ is a geometrical property of a
volume. The larger planar gun has a better matching λgun , resulting in less λ
gradient and thus less plasma instability.
(3) Absence of a flux conserver and the planar geometry of the electrodes provide
better diagnostic access to the entire plasma evolution including the source
gun region. In particular, direct imaging of the gun-produced plasmas became
possible and has revealed the unique dynamics of the plasma structures (see
figure 4.1 for example).
Spheromaks have been successfully produced by the planar gun and the spheromak
formation mechanism has been studied in detail [9]. The kink instability [10, chap. 10]
of the gun-produced plasma jets has been identified as a poloidal flux amplification
mechanism, which is necessary for spheromak formation. Onset of the kink instability
observed experimentally was shown to agree with the Kruskal-Shafranov limit, i.e.,
the plasma becomes kink-unstable when its axial length becomes greater than 1/λgun .

1.3

Collimated plasma structures

In addition to spheromak applications of the planar gun, images of the gun-produced
plasmas showed remarkable similarities with naturally occurring plasmas such as solar prominences and astrophysical jets [8]. In particular, the collimated tube-like
structure (i.e., long tube with a constant area cross-section) is the most conspicuous
common feature. The laboratory plasmas produced by the planar gun are compared
with the naturally occurring plasmas in figures 1.4a and 1.4b.
Solar prominences are arch-shaped magnetized plasma structures protruding from
the surface of the Sun [11]. Typical prominences extend over many thousands of kilometers and are stable in general lasting for days or even months. The sub-category
called coronal loops are especially known for the highly collimated and elongated

flux
conserver

vacuum
magnetic fields

solenoid
gas puff
(a) Coaxial gun

gas puff

large
vacuum chamber

coil

vacuum
magnetic fields

(b) Planar coaxial gun

Figure 1.3: Schematic drawings of (a) conventional cylindrical coaxial magnetized
gun and (b) the planar coaxial magnetized gun.

10
magnetic flux tubes. Plasma particles are localized inside the collimated solar prominences, and so there exists a large density gradient across their boundaries. The
stability of the solar prominences despite the large density gradient has not been
understood yet.
Astrophysical jets are also highly collimated and emanate from massive astronomical objects such as active galactic nuclei, young stellar objects, and high-mass X-ray
binaries. The axial extent of the jets (1015 –1022 m) are many orders of magnitude
greater than the source objects and the jet velocities can be relativistic. Magnetic
fields observed in some astrophysical jets suggest that magnetohydrodynamics (MHD)
plays an important role governing jet formation and collimation [12].

1.4

MHD pumping mechanism

Motivated by the observations of collimated plasma structures both in nature and
laboratory, Bellan proposed a universal MHD pumping process [13] which explains
why the collimated plasma-filled magnetic flux tubes are ubiquitous. The model is
based on the concept of frozen-in magnetic flux and is summarized below.

1.4.1

Frozen-in flux – magnetic Reynolds number

Dynamics of magnetized plasmas is described by the Maxwell’s equations and the
Ohm’s law,
∇ × B = µ0 J ,

(1.15)

∇ · B = 0,

(1.16)

∂B
∂t

(1.17)

J = σ(E + U × B),

(1.18)

∇×E =−

where σ is the electrical conductivity and U is the plasma flow velocity. The displacement current term is ignored in the ∇ × B equation provided that the characteristic

Laboratory plasma arches
(L ~ 20 cm)

Coronal loops
(L ~ 100,000 km)

Sun

(a) Solar coronal loops
Centaurus A
galaxy core

Laboratory plasma jet
(L ~ 50 cm)

Astrophysical
plasma jet
(L ~ 30,000 ly)

(b) Astrophysical jet

Figure 1.4: (a) Solar coronal loops (image from the Transition Region and Coronal
Explorer) compared with the laboratory plasma arches. (b) Astrophysical jet (Centaurus A galaxy; Chandra X-ray/Radio image) compared with the laboratory plasma
jet. The laboratory plasmas are produced by the planar coaxial magnetized gun.

12
speed is much less than the speed of light. From these equations, the rate of growth
of magnetic field strength can be derived:
∂B
−1
= −∇ × (J /σ) − U × B =
∇ × ∇ × B + ∇ × (U × B)
∂t
µ0 σ
1 2
∇ B + ∇ × (U × B).
µ0 σ

(1.19)

The first term (diffusion term) on the right-hand side represents the rate of diffusion
of magnetic fields through plasma due to the finite electrical resistivity η = 1/σ. The
second term (convection term) represents the rate of change of the magnetic field
strength due to the plasma flow. In ideal MHD (E + U × B = 0), the diffusion term
vanishes and the magnetic flux is completely frozen into the plasma flow. In order
to see this, consider the magnetic flux Φ(t) enclosed by a closed surface S(t) in the
plasma at time t,
B · ds.

Φ(t) =

(1.20)

S(t)

Taking the convective derivative of the flux Φ(t) using the Leibniz integral rule yields,
DΦ(t)
Dt

∂B
· ds
∂S(t)
S(t) ∂t
∂B
B × U · dl +
· ds
S(t) ∂t
∂S(t)
∂B
∇ × (B × U ) · ds +
· ds
S(t)
S(t) ∂t
Z
∂B
∇ × (B × U ) +
· ds
∂t
S(t)
[∇ × (B × U ) − ∇ × E] · ds
S(t)
−∇ × (E + U × B) · ds = 0.
B · U × dl +

S(t)

Thus, the flux Φ(t) is invariant in the frame of the moving plasma, i.e., the magnetic
flux is frozen into the plasma.
The magnitude of the convection term relative to the diffusion term determines
whether the magnetic field at a point in the fluid may build up fast enough before it

13
diffuses into the surroundings. The condition for magnetic field building up by fluid
motion is that the frozen-in flux term is much greater than the diffusion term:
k∇ × (U × B)k ∼

1 2
UB

∇B ∼
µ0 σ
µ0 σL2

µ0 σU L
1,

(1.21)

where L is the characteristic length over which magnetic field varies.
Hence, the quantity µ0 σU L, called the magnetic Reynolds number (Rm ) or Lundquist number, is of fundamental importance in magnetized plasma. For plasmas with
a small magnetic Reynolds number, there is no large gradient of magnetic fields in
the plasma and the action of magnetic fields can be considered like a viscous drag on
the fluid due to induced eddy currents. For plasmas with a large magnetic Reynolds
number, the magnetic flux is carried by the fluid. The action of magnetic fields can
now be thought as if a lateral pressure B 2 /2µ0 is applied normal to the lines of force
and a longitudinal tension B 2 /2µ0 is applied along the lines of force. The magnetic
Reynolds number is often obtained from the ratio of two characteristic time scales:
Rm = τr /τA ,

(1.22)

where the resistive diffusion time scale τr = µ0 σL2 and the Alfvén time scale τA =
L/U . The diffusion time scale is obtained by considering

∂B
∂t

B/τr ∼

diffusion

1 2
∇ B,
µ0 σ

=⇒ τr ∼ µ0 σL2 .
µ0 σL

(1.23)

Rm
1 for our plasmas and so the magnetic flux is frozen into the plasma motion.

1.4.2

MHD pumping and collimation model

Consider a cylindrical magnetic flux tube as illustrated in figure 1.5. Initially, the
flux tube is flared in the middle and has no current, and the plasma particles are

J×B
force
initial
later

z = +h

z = -h
mass
flow
z=0

Figure 1.5: Collimation of initially flared flux tube (adapted from [13]). Solid lines
are initial magnetic flux surfaces and dashed lines are the same flux surfaces at a later
time.
concentrated near the foot points. An electric potential is applied along the field
lines and subsequently an axial current starts to ramp up. Note that this situation
coincides with the generic configuration for helicity injection discussed in section 1.2.
The J ×B force has a net axial component due to the flared geometry and creates
axial flows.

dUz
∂P
∂P
= (J × B)z −
= Jr Bφ − Jφ Br −
dt
∂z
2 ∂z
Bφ2
∂P
− Jφ Br = −
=−
+ P − Jφ Br
∂z 2µ0
∂z
∂z 2µ0
∂Bφ
due to the cylindrical symmetry),
(∵ µ0 Jr = −
∂z

(1.24)

where Uz is the z-directed flow velocity, ρ is the mass density, and B, J, and P are
the magnetic field strength, current density, and pressure, respectively.
The radial component of the J × B describes the pinch force. The pinch force
is greater near the foot points since both the current density J and the magnetic
field strength are greater there than in the middle. Thus, without a radial force
balance, the plasma will become more constricted at the foot points and eventually
disrupt (sausage instability [10, chap. 10]). To exclude this unstable situation from

15
consideration, assume that the pressure buildup due to the pinching quickly balances
the pinch force:
∂P
= −Jz Bφ .
∂r

(1.25)

Also, assume that the current density Jz is uniform inside a cross section of the flux
tube, i.e., Jz (z, r) = Jz (z) = I0 /πa(z)2 , where I0 is the total current flowing in the
flux tube and the local tube radius a(z) describes the flaring of the flux tube. Then,
µ0 (Jz πr2 )
µ0 J z r
2πr
∂P
µ0 Jz2 r
= −Jz Bφ = −
∂r
Bφ2 a2
µ0 Jz2 2
(r − a ) =
−1 .
P =−
µ0 r 2
Bφ =

(1.26)
(1.27)
(1.28)

Thus, equation (1.24) becomes

dUz
∂ Bφ2 a2 1
=−
− Jφ Br
dt
∂z µ0 r2 2
2

r2
∂ Bφ,a
1− 2
− Jφ Br ,
=−
∂z µ0
2a

(1.29)

where Bφ,a = µ0 I0 /2πa is the azimuthal magnetic field at the flux tube radius a. Near
the axis, Jφ Br and r2 /a2 are very small, yielding an approximate expression
2
dUz
∂ Bφ,a
≈−
dt
∂z µ0

(1.30)

Thus, the magnetic energy density term Bφ,a
/µ0 acts like an effective potential and

so the plasma particles will move axially falling down this potential. Equation (1.30)
can also be expressed in terms of the flux tube flaring (∂a/∂z):
∂ µ0 I02
µ0 I 2 ∂a
dUz
≈−
dt
∂z 4π 2 a2
2π 2 a3 ∂z

(1.31)

The potential Bφ,a
/µ0 has the minimum at the middle of the flux tube due to

the flared geometry of the flux tube. Therefore, counter-streaming axial flows will

16
be generated from both foot points toward the middle. The counter-streaming flows
collide with each other and so stagnate in the middle, resulting in accumulation of
plasma particles there. By the frozen-in flux condition, the azimuthal magnetic flux
carried by the flows also accumulates in the middle, which increases the azimuthal
magnetic field strength Bφ . As a result, the pinch force Jz Bφ increases and squeezes
the magnetic flux tube in the middle. Thus, the magnetic flux radius becomes uniform
axially, i.e., the magnetic flux tube becomes collimated.

1.5

Overview of the following chapters

The present thesis work addresses the experimental verification of the MHD pumping
model as a universal collimation mechanism. The experimental setup to produce collimated plasma tubes is described in chapter 2 and the diagnostic devices to measure
the plasma properties are described in chapter 3. In particular, construction of a
high-resolution spectroscopic system for the plasma density and flow velocity measurements is described in detail. In chapter 4, experimental results are presented and
discussed in detail, showing that the observations are consistent with the MHD pumping model. A theoretical background for the plasma density diagnostics is provided
in chapter 5 and a simple method for obtaining the plasma density from spectral line
profiles is explained in chapter 6. A summary of this thesis work is given in the final
chapter.

Chapter 2
Experimental setup – boundary
conditions
The experimental setup used in this work provides the physical boundary conditions
on the evolution of plasma structures. The experimental setup has been developed
and improved over a decade and will be summarized in this chapter. Knowledge of the
experimental setup and hence the boundary conditions of the experiment is necessary
to comprehend the motivations behind the experiment as well as to properly interpret
experimental observations.
The experimental setup comprises five major components: vacuum chamber,
spheromak plasma gun, gas injection system, timing system, and diagnostics. The
vacuum chamber provides a free space into which plasma can evolve. The spheromak
plasma gun produces magnetized plasma structures. The plasma gun is mounted on
the north end dome of the chamber as shown in figure 2.1. The plasma gun comprises
cathode and anode electrodes, gun capacitor bank, stuffing flux system, and gas feed
lines. The gas injection system delivers high pressure gas into the vacuum on a msec
time scale. The timing system consists of two independent sub-timing modules: main
timing and diagnostic timing. The main timing module triggers various power supplies for the plasma gun and the gas injection. The diagnostic timing module triggers
diagnostic devices in sync with plasma breakdown. The diagnostic devices will be
discussed in the next chapter.

2.1

Vacuum chamber

The vacuum chamber is shown schematically in figure 2.1. The vacuum chamber
provides a free boundary condition for plasmas produced in the present experiment;
the plasma structures evolve without much interaction with the chamber wall because the plasma dimensions (1–10 cm in diameter and 10–50 cm in length) are much
smaller than the chamber dimensions. This is in contrast to other spheromak experiments where the plasma geometry and the magnetic field configuration are essentially
constrained by the chamber wall.
The vacuum is maintained at a background pressure Pbkg = 1–2×10−7 torr (corresponding to a particle density ∼5 × 1015 m−3 at room temperature) by the cryopump
(APD-12S; ∼1000 liters/sec pumping speed for air). The pressure is monitored by a
Bayart-Alpert-type ionization gauge. Maintaining a good vacuum is important since
it is observed that the plasma evolution is hampered in a poor vacuum condition
Pbkg & 10−5 torr. The vacuum quality is compromised primarily by two factors; (1)
desorption of gas molecules (especially water) absorbed by the chamber wall when
exposed to atmosphere and (2) leaks through O-rings and gaskets. These two factors
are indistinguishable in practice and observed as a single effective leak. The effective
leak of the chamber can be estimated by a simple procedure described in appendix A.
In case of a large leak, this procedure can be applied to quantify the leak and thus
facilitate the location of the leak.

2.2

Spheromak plasma gun

2.2.1

Coplanar coaxial electrodes

The electrodes of the plasma gun are planar (figure 2.2a) in contrast to cylindrical
electrodes used in other spheromak experiments. The planar configuration permits
direct observation of the entire plasma formation process. The inner electrode (cathode) is a disc copper plate and has eight gas orifices evenly spaced in a circle. The
outer electrode (anode) is an annular copper plate and is coaxial and coplanar with

electrodes
1.6 m
1.4 m

plasma jet

8" viewport

Figure 2.1: Vacuum chamber. Total volume is ∼2.4 m3 and the base pressure is
1–2 × 10−7 torr. The chamber dimensions are much greater than plasma. An image
of a kinked plasma jet is inset for size comparison. The plasma expands freely without
interaction with the chamber wall.
the cathode. The anode also has eight orifices at locations corresponding to the cathode orifices. The gas orifices are channeled to the gas injection system explained in
the section 2.3. The small annular gap between the two electrodes is 6 mm wide.
The condition for gas breakdown by the electrodes can be qualitatively understood
by the empirical law known as Paschen’s law [14]; the breakdown voltage Vs is a
function of pd, gas pressure(p) times discharge path length(d). The breakdown voltage
Vs has a minimum at some pd = (pd)min called Paschen minimum, increases slowly
for pd > (pd)min , and increases rapidly for pd < (pd)min . The Paschen minimum
(pd)min for common gases such as N2 , H2 , and Ar used in this work is ∼1 cm·torr. A
path connecting an inner gas orifice and an outer gas orifice gives pd ∼ (pd)min since
p ∼ 0.5 torr near the orifices (cf. section 2.3) and the path length d is a few cm.
Hence, a sufficiently high voltage applied to the electrodes will break down the gas
along such paths. However, pd
(pd)min in the small gap between the electrodes
because of low p ∼ 10−7 torr and small d = 0.6 cm, and so there will be no gas
breakdown across the gap. No breakdown will occur in the tight space (figure 2.2b)

20
between the re-entrant port (at high voltage) and the chamber wall (at ground) for
the same reason.
High voltage (3–8 kV) is applied across the electrodes using a 120 µF capacitor
bank (high voltage (HV) gun bank) switched by an ignitron. The gun bank supplies
a current of the order of 60–150 kA flowing through the plasma.

2.2.2

Stuffing flux system

The bias coil is mounted at the end-plate of the re-entrant port flush against the inner
electrode. It produces an axisymmetric bias poloidal field as illustrated in figure 2.2b.
The coil current is supplied by a 14.4 mF power supply and peaks at ∼6.5 msec after
the power supply is triggered as traced by a current monitor. The time scale of
the coil current is much greater than the ∼20 µsec plasma lifetime so that the bias
poloidal field is constant over the duration of the plasma.
The poloidal flux is attenuated by the skin effect (eddy current) of both the copper
electrode (3.2 mm thick) and the stainless steel end-plate (9.5 mm thick). The skin
depth is the distance δ through which the amplitude of a plane wave decreases by a
factor e−1 and is given by

δ=

 9.3 mm for copper
τ /πσµ ≈
 62 mm for stainless steel,

(2.1)

where τ ≈ 20 msec is the pulse length of the coil current, σ is the electrical conductivity, and µ is the permeability (≈ µ0 for non-magnetic materials). Hence, the
magnetic field strength is attenuated by a factor of e−3.2/9.3 × e−9.5/62 ≈ 0.6 and so
is the poloidal flux. The peak timing of the poloidal flux is also delayed by the skin
effect and is different from the peak timing of the coil current. The measured poloidal
flux peaks at ∼10 msec with the peak flux of 1.4 mWb per 100 V, linear with the
power supply voltage.
The nominal LCR parameters of the stuffing flux system are 2.8 mH coil inductance, 14.4 mF power supply capacitance, and 0.4 Ω DC resistance. However, the

51 cm O.D.

cathode
anode

gas feed lines
gas orifices

fast ion gauge
magnetic probe
(a) front view
re-entrant port
chamber wall
field lines

gas feeds

bias coil

ceramic
break

copper tube

(b) side view (schematic)

Figure 2.2: Spheromak plasma gun (designed by S. C. Hsu [7]). Inner electrode is an
oxygen-free high-conductivity (OFHC) copper disc plate of 19.1 cm (7.500 ) diameter.
Outer electrode is an OFHC copper annulus plate of 50.8 cm (2000 ) outer diameter
and 20.3 cm (800 ) inner diameter. The electrodes are 3.2 mm (1/800 ) thick. The inner
electrode is mounted on the re-entrant port, which is insulated from the chamber by
a ceramic break. The outer electrode is mounted on four hollow copper tubes (two at
the bottom and the other two at the top of the electrode), each of which is anchored
to the chamber wall by a stainless steel threaded rod. Axisymmetric poloidal field is
produced by the bias coil of 110 turns located behind the inner electrode.

22
ball valve
two-way

needle
fast gas
valve
six-way

gas line A

gas line B
OL

Figure 2.3: Gas line plumbing (designed by Setthivoine You). There are two independent gas lines labeled by gas line A and B. Each gas line has a six-way valve
connected to six high-purity gas cylinders. The six-way valve selects a gas to feed the
gas line. Each of the four fast gas valves can be connected to either the gas line A or
B by a two-way valve.
effective inductance (∼1.6 mH) as estimated from current traces is smaller than the
nominal inductance due to the eddy currents induced on the inner electrode [15]. The
eddy currents exert a large repulsive force on the bias coil. The measured effective
capacitance (∼19 mF) is somewhat larger than the nominal capacitance.

2.3

Gas injection system

2.3.1

Flexible plumbing

A flexible plumbing system shown schematically in figure 2.3 has been developed by
S. You to make possible a variety of gas injection scenarios. The plumbing system
was later reconfigured to the current setup by S. You and G. S. Yun and extended
to the solar prominence simulation experiment [11] by S. K. P. Tripathi and E. V.
Stenson. In each experiment, a single gas species can be injected into the chamber or
two gas species can be injected in eight different ways.
The plumbing system is divided into two independent gas lines (labeled as A and
B in the figure). A single gas species is selected among six gas species and fed into
each gas line. The gas is injected into the vacuum chamber by four fast gas valves.
Each fast gas valve is connected to a manifold, which splits the gas flow into four
gas feed lines. The fast valves are labeled as inner left (IL), inner right (IR), outer
left (OL), and outer right (OR) according to their locations. The IL valve injects gas
through the four orifices on the left half plane of the inner electrode and the IR valve
injects gas through the other four orifices on the right. The OL and OR valves inject
gas through orifices on the outer electrode in the same manner.

2.3.2

Fast gas valve

A pressurized gas (60–100 psi) fills the 2.7 cm3 plenum of the fast gas valve (figure 2.4).
The diaphragm loaded by a spring presses an O-ring against the gas line, sealing off
the plenum from vacuum. A coil is wound underneath the diaphragm. When a pulsed
current flows in the coil, a current of the opposite sense (i.e., eddy current) is induced
on the diaphragm and the diaphragm is pushed up against the spring and the back
pressure. This action temporarily relieves the O-ring seal, allowing the gas to flow
from the plenum into the vacuum by a large pressure gradient.
The total throughput of the fast gas valve depends on the back pressure and the
coil current. The throughput can be measured by the average pressure rise per puff.

high pressure
gas inlet

diaphragm

plenum

to chamber

Figure 2.4: Fast gas valve (designed by P. M. Bellan). A high pressure gas fills
the valve body including the small plenum (2.3 cm3 ) underneath the diaphragm.
When a current pulse flows through the coil, an eddy current is induced on the
diaphragm. The repulsive force between the coil current and the eddy current opens
the diaphragm for a few msecs against the spring force (spring constant = 1600 N/m)
and the back pressure force (1–10 N). About 20% of the gas in the plenum is injected
into the chamber during the opening. The total gas throughput is controlled by the
coil current and the back pressure.

25
For a typical operation of 70 psi nitrogen in the gas line, a single fast valve gives a
pressure rise ∆P ∼ 0.8 mtorr ≈ 0.1 Pa per puff in the vacuum chamber. Since the
chamber volume V is 2.4 m3 , the total number of puffed particles is
NT = ∆P · V /kT = 5 × 1019 ,

(2.2)

which is about 20% of the total number of particles available in the plenum. It will
be shown later that about 20% of the puffed particles are used in making a plasma
jet (see section 4.4). Most of the gas particles are pumped into the chamber within
∼5 msec (cf. appendix B), giving the average particle flux rate
dNT /dt ≈ 1019 msec−1 .

(2.3)

Since there are four gas lines connected to each fast gas valve, the flux rate for a
single gas line is
dN1
1 dNT
≈ 2 × 1018 msec−1 .
dt
4 dt

(2.4)

It is interesting to note that this flux rate is comparable to the theoretical flux rate
for steady-state flow of rarefied gases through tubes [16],
n1 u1
n1
= (8a/3L) · πa2 ·
K =W ·S·

8kT
πm

(2.5)

where W is transmission probability (also known as Clausing factor), S, a, and L
are the cross-sectional area, radius, and length of the tube, respectively, n1 is the
upstream particle density, u1 is the upstream average particle speed, and m is the
particle mass. Upstream properties are denoted by the subscript 1 and downstream
properties by the subscript 2. It is assumed that n1
n2 and the gas temperature
T is constant. Using a ≈ 2 mm and L ≈ 1 m for our gas lines, the flux rate for
P1 = 70 psi nitrogen gas puffing (n1 = 1.2 × 1026 m−3 and u1 = 470 m/sec at
T = 300 K) is
K ≈ 1018 msec−1 .

(2.6)

26
The downstream density n2 at the orifice may be estimated by assuming u1 = u2
and using the continuity of the flux rate K:
n1 u 1
= πa2 · n2 u2
=⇒ n2 = (2a/3L)n1 ∼ 1023 m−3
K = (8a/3L) · πa2 ·

(2.7)

(P2 ≈ (2a/3L)P1 ∼ 0.1 torr).
The gas expands freely after leaving the orifice so the density n will decrease as
n ∼ 1/z 2 , where z is the axial distance from the orifice:
n(z) ∼ n2 × (a/z)2

∼ 1019 m−3 at z = 100 mm.

(2.8)

The plasma density measured at the same location 5–10 µsec after breakdown is
∼1022–23 m−3 , showing that the measured density is not a result of ionization of
the neutral gas cloud but due to an active pumping of plasma particles from the
source region. The MHD pumping mechanism described in section 1.4 drives plasma
particles out of the orifice.

2.4

Main timing – gas breakdown sequence

The main timing system triggers various power supplies for the stuffing flux coil, the
fast gas valves, and the HV gun bank. The timing system is optically isolated from
the power supplies to break ground loops [17] as well as to protect it from accidental
electrical shocks. All the power supplies are designed to be optically triggered. The
timings are configured such that the HV gun bank is switched at the maximum of
the bias poloidal flux and the optimal density (i.e., Paschen minimum) of the gas
cloud. Referenced to the HV gun bank trigger timing (0 µsec), the stuffing flux coil is
energized at −10 msec and then the fast gas valves are triggered at between −6 and
−1 msec depending on the injected gas species. An example configuration of timings
is illustrated in figure 2.5.

27
Bias Field

Side Gas

Gas OutL Gas InnL
Gas OutR Gas InnR

Main HV Bank

PI ICCD
(upper)
PI ICCD
(lower)

breakdown
PI Stop Clean

Imacon
OMA Ext.

-8

-6

-4

-2
msec

OMA Gate

trigger
lag

jitter

-10

VME
Dataq

800

μsec

Figure 2.5: An example of timing configuration. The main timing system controls
trigger pulses indicated by dashed lines (left to the main HV bank, except the VME
data acquisition) in msec time scale. The breakdown jitter is 0–10 µsec. The diagnostic timing system controls trigger pulses indicated by solid lines (right to the main
HV bank) in µsec time scale. The trigger lag is about 0.45 µsec (see section 3.5). The
legends are self-explanatory except PI ICCD – triggers Princeton Instruments ICCD
camera, PI Stop Clean – stops the continuous flushing of the PI ICCD pixels, OMA
Ext. – triggers the timing module of the Andor ICCD detector of the spectroscopic
system, and OMA Gate – gates the Andor ICCD.
The timings are uploaded into an 8-channel delay generator (Berkeley Nucleonics,
Model 565) using a LabView program. The delay generator initiates generation of
output pulses upon an external trigger. The output pulses are fed into a home-built
8-channel electro-optical pulse doubler (built by D. Felt). The pulse doubler generates
two optical trigger signals, one at the rising edge and the other at the falling edge of
each delay generator pulse, making total of 16 optical trigger signals.
The gas breaks down within 10 µsec but at random after a high voltage is applied
across the electrodes. This randomness or jitter of the breakdown timing is greater
than the time scale of plasma evolution (∼1 µsec), so the HV gun bank timing cannot
be used to synchronize diagnostics. A separate timing scheme for diagnostics has been
developed to trigger diagnostics at the actual gas breakdown. The diagnostic timing
system is described in section 3.5.

2.5

Plasma formation

A distinctive plasma structure forms as the gas breaks down by triggering the stuffing
flux system, the gas injection system, and the HV gun bank in sequence as described
in the previous section. Figure 2.6 illustrates a typical profile of gas cloud and bias
field at the instant of the breakdown. The initial plasma structure is defined by eight
flux tubes indicated by red solid field lines in the figure. Each flux tube spans a gas
orifice on the cathode and the corresponding orifice on the anode. The breakdown
condition is optimal along the flux tubes where gas density is highest. The ionized
gas particles are pumped into the flux tubes from the source gas orifices by MHD
force, forming a distinctive structure reminiscent of spider legs as shown in figure 2.7.
The eight plasma-filled current-carrying flux tubes will be referred to as spider legs.
Subsequent evolution of the spider legs is characterized by MHD force and will be
discussed in chapter 4.

Figure 2.6: Neutral gas density profile and bias field lines (dashed lines) at the gas
breakdown. Each field line is labeled by the amount of the poloidal flux (in mWb)
enclosed by the circle whose radius is from the axis of symmetry to the field line.
Colors represent the normalized density in a logarithmic scale as shown in the scale
bar. Red solid lines define the flux tubes where inside the gas will break down. The
density profile in the flux tubes are highlighted.

Figure 2.7: Eight plasma-filled current-carrying flux tubes (shot# 6529: N2 inner +
Ne outer). This distinctive plasma structure reminiscent of spider legs is formed by
breaking down the gas cloud shown in figure 2.6.

Chapter 3
Diagnostics
The spheromak experiment has a set of diagnostics to measure plasma properties such
as current, magnetic field, density, flow velocity, and geometry of plasma structures.
Each diagnostic corresponds to an observable quantity of the plasma. An observable
quantity may be a result of complex underlying physics and can give a valuable
information about the plasma once the underlying physics is known. For example,
electron density can be obtained if the Stark broadening of spectral lines due to
charged particles is understood (ref. chapter 5).
Plasma properties are time-varying on the scale of µsec, and so every diagnostic
must be recorded and synchronized to each other within an uncertainty much less
than this plasma time scale.

3.1

VME digitizer system

Many diagnostic signals are recorded by a multichannel fast digitizer system and such
signals are automatically synchronized with each other. The digitizer system comprises twelve data acquisition (DAQ) boards (SiS GmbH SIS3300; 8-channel 100 MHz
12-bit, 50 Ω input impedance, 256K samples/channel memory) and one counter board
(SIS3820; 32-channel 50 MHz) installed on a VME crate. Initiated by an IDL computer code, each DAQ board continuously samples until it receives a stop signal from
the counter board. The counter board is triggered by the main timing system (ref.
section 2.4), typically somewhere between 0.5 and 2 msec after the HV gun bank

31
discharge. The IDL computer code retrieves the sampled data from the DAQ boards
later on.
The VME digitizer system records the voltage and current traces of the plasma
gun, magnetic probe signals, etc. There are diagnostic data not recorded by the VME
system such as camera images and spectroscopic signals. Those diagnostics are synchronized with the VME-recorded data by a timing scheme described in section 3.5.

3.2

High-voltage probe

The plasma gun voltage (Vgun ) is measured by a 1000:1 high-voltage probe (Tektronix
P6015; 20 kV Max, 75 MHz). The Vgun signal goes through a line driver to match
the VME system’s 50 Ω input impedance. A typical Vgun trace is characterized by
four distinctive stages (figure 3.1a): (1) constant Vgun held at the applied voltage for
less than 1 µsec up to 10 µsec before gas breakdown. (2) sudden voltage drop at
the moment of gas breakdown. (3) a plateau at about half the applied gun voltage.
The plateau is sustained as long as the plasma carries the gun current. (4) sinusoidal
decay phase as the plasma is detached from the electrodes and the gun current flows
through dump resistors.
The duration of the first stage is purely random and so the delay of gas breakdown
cannot be controlled precisely, making it difficult to synchronize diagnostic signals (see
also discussions in section 2.4 and section 3.5). The breakdown delay is longer for
heavier gases in general.

3.3

High current transducer – Rogowski coil

The gun current is of the order of 100 kA and approximately sinusoidal (figure 3.1b).
A home-made Rogowski coil is wrapped around the 7.500 diameter re-entrant port
of the plasma gun (figure 2.2b) to measure the time-varying gun current (dIgun /dt).
The Rogowski coil signal is integrated by a passive integrator circuit (figure 3.2). The
current Igun measured by the Rogowski coil is calibrated using a current monitor as

Current [kA]

Voltage [kV]

-1
-2
-3

plateau

-4

breakdown

-5

breakdown

10
15 20
Time [Μsec]

(a) Voltage trace

-20
-40
-60
-80

-5

10
15 20
Time [Μsec]

(b) Current trace

Figure 3.1: Typical voltage and current traces of the plasma gun (shot# 7220: N2
70 psi, Vgun = 5 kV).
explained in figure 3.3.

3.4

Multichannel magnetic probe

A 20-unit magnetic probe array constructed by Romero-Talamás [18] has been used
to investigate the magnetic field structure of the plasma. Each unit is a cluster of
three mutually orthogonal miniature commercial chip inductors. The probe array
measures three-dimensional magnetic field at twenty locations, 20 mm apart. It is
mounted on a right-angle arm so that both its axial and radial locations can be
adjusted as shown in figure 2.2a. A measured magnetic field structure is shown in
figure 3.4a. A method known as Single Shot Propagation Inference method [19] can be
used to obtain a qualitative picture of the spatial configuration (B(r, z)) of magnetic
field from a single magnetic probe measurement (B(r, t)). The method assumes that
the magnetic field structure embedded in the plasma merely translates along the axial
(z) direction:
B(r, z, t) = B(r, zprobe , t − (z − zprobe )/v),

(3.1)

where zprobe is the location of the probe and v is the translation velocity. A plasma image is overlaid with the corresponding magnetic field structure drawn by this method
in figure 3.4b.

33
metal enclosure

150

2 μF

150
to VME
Input
(50 Ω)

triax
shield
Rogowski
coil

Figure 3.2: Passive integrator circuit for Rogowski coil (designed by P. M. Bellan).
The effective resistance of the circuit is Reff = 150k(150 + 50) ≈ 86 Ω, giving the
RC time constant of 86 · 2 × 10−6 sec = 170 µsec. The Rogowski coil is made of 72
turns of a semi-rigid coax cable (Micro-coax UT 85) with break in shield. The major
diameter of the coil is 11.500 , the minor diameter is 0.37500 , and the space between
turns is 0.500 . The coil is isolated from the VME digitizer by a transformer. A triaxial
cable connects the Rogowski coil to the integrator circuit. The outermost shield of
the triaxial cable and the metal enclosure as a whole acts like a Faraday cage blocking
electrical noises produced by the plasma gun.

Current [kA]

-1
-2
-3
-4
-5

40
60
Time [Μs]

Figure 3.3: Rogowski coil calibration. The Rogowski coil was wrapped around a
wire carrying a current pulse produced by discharging a capacitor. The wire wasR also
passed through a current monitor (Ion Physics CM-01-L; 1 V/kA, saturation Idt
= 2 A·sec). Dashed line is the reference current measured by the current monitor.
Solid gray line is the calibrated Rogowski coil signal. The calibrated signal correctly
reproduces the reference signal for the first 40 µsec, but it drifts a little bit from the
reference afterward.

radius [cm]

30
25
20
15
10

10
time [Μsec]

(a) Magnetic field structure

(b) Plasma overlaid with its magnetic field
structure

Figure 3.4: Magnetic field structure measured by the magnetic probe array (shot#
7353; H2 plasma). Contours of magnetic field energy density (B 2 /2µ0 ) are shown.

3.5

Timing of diagnostics

A separate timing system similar to the main timing system (ref. section 2.4) has
been developed in order to synchronize or trigger diagnostics with respect to the
gas breakdown. The diagnostic timing system comprises an external trigger source,
an 8-channel delay generator, and a pulse doubler. The external trigger source for
generating a pulse at the instant of breakdown is necessary since the main timing
system cannot provide such a pulse due to the breakdown jitter (ref. section 3.2).
The delay generator is triggered by the external trigger and subsequently produces
eight independent pulses with adjustable delay and pulse length. To prevent electrical
shocks and ground loops, each delay generator pulse (except the OMA gate pulse;
see figure 2.5) is fed into the pulse doubler instead of going directly to a diagnostic
device through an electrical cable. Each optical pulse from the pulse doubler goes to a
designated diagnostic device via an optical cable. The optical pulse is converted back
to an electrical pulse by a small home-built battery-operated optoelectric transducer
(OET; built by D. Felt) before entering the diagnostic device.
Two types of external trigger method have been developed and are described here.
(1) The abrupt change in Vgun trace as indicated in figure 3.1a corresponds to the
instant of breakdown. A voltage comparator circuit was built to detect the
abrupt voltage change and generate a subsequent trigger signal. The trigger

35
lags behind the breakdown by ∼0.1 µsec due to the finite rise time of the
voltage. However, a better method to detect breakdown had been sought since
the voltage threshold of the comparator had to be re-adjusted for different
settings of the gun voltage and the noise in the voltage signal near breakdown
made false triggers occasionally.
(2) An optical method to detect the breakdown was motivated by the fact that
the breakdown is always accompanied by substantial optical radiation from the
resultant plasma, especially from inside the gas orifices. The emission light
from one of the eight inner gas orifices is collected by a collimator lens focused
on that orifice and fed into an optical fiber. The optical signal is converted
to an electrical signal by an OET to trigger the delay generator. This optical
scheme provides a reliable detection of breakdown free from the problems in the
previous method with about the same trigger lag of ∼0.15 µsec.
Although both of the methods are able to detect the breakdown within 0.15 µsec,
the actual trigger pulses arriving at diagnostic devices are further lagged as much as
0.3 µsec due to the internal delays of the delay generator, the pulse doubler, OETs,
and optical/electrical cables. The total trigger lag is 0.45 ± 0.05 µsec.
This trigger lag is taken into account when comparing the diagnostic data triggered
by the diagnostic timing system (fast digital cameras, spectrometer, etc) with the
VME-digitized signals (Vgun , Igun , magnetic data, etc) which are not triggered by the
diagnostic timing system. An example configuration of diagnostic timings (together
with main timings) is illustrated in figure 2.5.

3.6

Fast digital cameras

The gun-produced plasma is rapidly evolving and highly radiative. Images of the
plasma are taken by a multiframe high-speed intensified-CCD camera (DRS Hadland,
Imacon 200; 10 bpp dynamic range, 1200 × 980 pixels, 6.7 µm square pixel). The
Imacon camera can take up to 16 frames per shot. The exposure and delay of each

36
frame can be configured in 5 nsec precision, a sufficient time resolution for capturing
the µsec-scale plasma evolution. In a normal configuration, 16 frames are taken at
an equally spaced time interval. Each frame can also be multiply exposed to capture
multiple plasma images in the frame, for example to see the evolution of the plasma
in finer time scale without reducing the total time span.
A system of dual single-frame CCD cameras (Princeton Instruments, ICCD-576G/RB-E; 16 bpp dynamic range, 576 × 384 pixels, 22.5 µm square pixel) is sometimes
used for sharper images (because of the better dynamic range). The exposure and
delay of the Princeton ICCD cameras can be configured to a 10 nsec precision.
Optical filters can be put on the camera lenses to measure the plasma radiation
within a specific optical bandpass. Filtered plasma images are used to identify particle
species in plasmas made of multiple gas species.

3.7

Spectroscopic system

Shifts and broadenings of spectral lines emitted by plasma reflect the physical condition of the plasma such as density, temperature, and flow velocity. A low-resolution
spectroscopic system (consisted of an optical fiber, a 0.22 m spectrometer, and an
ICCD detector) was used to infer plasma conditions from spectral line profiles in a
previous attempt by C. Yang [20]. Broadenings of Hα and Hβ lines as large as 1.3 nm
were observed and attributed to the Stark effect and the thermal Doppler effect. The
Stark broadening is a density effect caused by local electric fields due to ions and
electrons in the plasma and is summarized in chapter 5. The Hα line width was used
to estimate the temperature assuming it is strictly Doppler broadened and the Hβ
line width subtracted by the Hα line width was used to estimate the density. This
procedure would give a reasonable density in case of a large broadening much greater
than the 0.2 nm instrumental broadening of the spectrometer since the Stark effect is
about 5 times stronger for the Hβ line than the Hα line. A peak density of the order
1022 m−3 was reported in the study.
The large broadening of the Hβ line was routinely observed in later experiments

37
by the present author using the same spectroscopic system. However, the spectral
resolution of the spectrometer was not sufficient enough to estimate density less than
1021 m−3 nor to see any Doppler shift for flow velocity measurement. It was also
difficult to estimate densities of non-hydrogen plasmas since Stark broadenings of
non-hydrogenic lines are at least an order of magnitude smaller than hydrogen lines.
A multichannel high-resolution spectroscopic system has been constructed in order to improve the spectral resolution. The ability to measure spectra at multiple
locations of the plasma with definite lines of sight has also been implemented into the
new spectroscopic system since the plasma is inhomogeneous and its detail structure
is not reproducible. The spectroscopic system, illustrated in figure 3.5, comprises a
camera lens, a 12-channel fiber array, entrance optics, a high resolution spectrometer,
and an intensified CCD detector. The camera lens forms an image of plasma on the
plane of the fiber array input. The plasma emission is intercepted by the fiber array
at 12 different locations and fed into the spectrometer. The emission spectra analyzed
by the spectrometer are then recorded by the ICCD detector. The ICCD detector
is triggered and gated by the diagnostic timing system as indicated by “OMA Ext.”
and “OMA Gate” respectively in figure 2.5.
The specifications of the spectroscopic system are summarized below:
(1) The wavelength range is 200–500 nm, where the lower end is limited by the
camera lens and the upper end by the spectrometer.
(2) The spectral resolution is calculated from the spectrometer and ICCD parameters according to
s

R(λ) =

∆x 
2L

2d cos α0

− λ2 + λ tan α0  ,

(3.2)

where R(λ) is the pixel resolution (wavelength per pixel), ∆x is the ICCD pixel
size, L is the focal length of the spectrometer, α0 is half the angle between
the incident and the diffracted light on the grating, d is the groove spacing
of the grating, m is the diffraction order, and λ is the selected wavelength.

(g)
Diagnostic
Timing
System

(f) OET

(e) collimator

(a) spectrometer

(d) lens

Image formed
by the lens

(h) ICCD
(b) entrance
optics

Figure 3.5: Schematic drawing of the spectroscopic system. The system consists of
(a) Czerny-Turner spectrometer (JY Horiba 1000M: 1 meter focal length, f/8 aperture, 3600 grooves/mm grating), (b) F/# matching entrance optics (c) 12-channel
linear fiber assembly (RoMack custom design; UV/VIS grade, 0.22 NA, 100 µm core
size, 10 meters long) (d) camera lens (e) collimator (Fiberguide MACRO collimator;
22.2 mm aperture, 0.22 NA), (f) optoelectric transducer (OET), (g) diagnostic timing system, and (h) intensified CCD detector (Andor ICCD DH520-25F-03; 16 bpp
dynamic range, 800 × 256 active pixels, 26 µm square pixel).

39
The derivation of equation (3.2) is given in appendix C. The accuracy of this
calculation has been verified over a wide range of reference wavelengths using
spectrum tubes (Hg, D2 , O2 , and Ar).
(3) The temporal resolution, i.e., minimum gate width of the ICCD detector with
reasonable signal-to-noise ratio, is about 0.3 µsec. The temporal resolution
is mainly limited by optical throughput, which is maximized by matching fnumbers between optical components.
(4) The spatial resolution is determined by the fiber diameter and the magnification
of the imaging optics. Each fiber in the fiber array intercepts a line of sight
volume of diameter ∼3–5 mm. The interspace between the lines of sight is
approximately 10 times the line of sight diameter, which is ∼30–50 mm, when
the focal length of the camera lens (d) in figure 3.5 is 50 mm. The spatial
resolution is also affected by the temporal resolution because the plasma jets
move very fast. For example, if the ICCD gate width is 1 µsec and the jet
velocity is 30 km/sec, the detector signal is integrated over the jet travel distance
of 30 mm.
(5) Twelve spectra corresponding to twelve different locations of the plasma are
simultaneously recorded per shot. The 256 rows of the ICCD are divided into
12 tracks. The plasma emission light delivered by each fiber is spread onto the
corresponding track of the ICCD by the spectrometer.
The knowledge of spectral resolution is required to measure spectral line widths
and shifts. The spectral resolution, bandpass (pixel resolution × number of pixels),
and the Doppler shift corresponding to one pixel shift are calculated for the entire
wavelength range using equation 3.2 and are shown in figures 3.6.

Doppler shift [km/s]

Bandpass [nm]

Dispersion [pm/pixel]

40
10

200

250

300
350
400
Wavelength [nm]

450

200

500

(a) Pixel resolution

250

300
350
400
Wavelength [nm]

450

500

(b) Doppler shift per pixel

Figure 3.6: (a) Pixel resolution and bandpass of the spectroscopic system and (b)
Doppler shift corresponding to one pixel shift, as a function of the selected wavelength λ.

3.7.1

Entrance optics

The cone angle over which an optical component can accept or emit light is specified
by either f-number (F/#) or numerical aperture (NA) defined as
F/# = f /D,

(3.3)

NA = n sin(θ),

(3.4)

where f is the focal length, D is the aperture diameter, n is the index of refraction,
and θ is half the cone angle. The optical throughput, i.e., the fraction of light which
will pass through, from an optical component with a half-cone angle θ1 (F/#1 ) to
another optical component with a smaller half-cone angle θ2 (a larger F/#2 ) is given
by
throughput =

sin(θ2 /2)
sin(θ1 /2)

F/#1
F/#2

(3.5)

since the solid angle of a cone is 4π sin2 (θ/2). The approximation of the throughput
by the square of the f-number ratio is accurate for small cone angles and is preferred
for its convenience.
A typical multimode fiber has NA = 0.22 or F/# = F/2.3, corresponding to
sin−1 (0.22) ≈ 13 degrees of half-cone angle. The spectrometer’s f-number is F/8,
corresponding to tan−1 (1/8/2) ≈ 3.6 degrees of half-cone angle. Because of this
discrepancy in f-numbers, the spectrometer can accept only a tiny fraction ((2.3/8)2 ≈

41
8%) of the input light from the fiber array unless an appropriate f-number matching
optics is employed. This is like trying to connect pipes with different diameters
without a coupler.
A commercial f-number matcher (Oriel 77529) was used initially to increase the
f-number of the fiber (or decrease the cone angle of the fiber) by a factor of 2 so that
the optical throughput will increase by a factor of 4 (from 8% to 30%). However, it
was abandoned immediately since the small mirror inside the f-number matcher was
found to block a significant fraction of light. Instead, an elaborate entrance optics to
the spectrometer has been constructed in order to maximize the optical throughput.
The entrance optics consists of a spherical mirror (diameter = 300 , focal length
= 300 , protective aluminum coating), a y-z precision translation stage, and a linear
(z) stage (y refers to the direction from the fiber to the mirror and z refers to the
vertical direction). The fiber array mounted on the y-z stage and the mirror are raised
up together to the level of the spectrometer’s entrance slit by the linear stage. The
position of the fiber array is further adjusted on a µm precision by the y-z stage so that
the fiber output is focused onto the spectrometer’s entrance slit by the mirror. This
arrangement increases the fiber’s f-number by a factor of 3, achieving a high optical
throughput = (3 × 2.3/8)2 ≈ 75%. A mirror was favored over a lens configuration for
better UV transmission. A UV-grade fused silica lens has a good UV transmittance
but its index of refraction changes as a function of wavelength, making it difficult to
focus the fiber output over a wide wavelength range. Mirrors do not suffer from this
problem.

3.7.2

Fiber array

The 12-channel linear fiber array has been constructed to observe plasma emissions
at 12 different spatial locations simultaneously. The individual fibers are 100 µm core
diameter, UV/VIS grade fused silica fibers. The fibers are aligned in a single evenly
spaced row, the interspace being 1 mm at the input end of the array and 150 µm at
the output end.

42
(a) Fiber array

(b) Fiber bundle

Output end
(11 mm OD ferrule)

10 m

Figure 3.7: (a) Fiber array and (b) fiber bundle
The interspace between fibers at the input end was decided by the distance the
plasma jet would travel on the image plane of the camera lens during a typical exposure time (∼1 µsec). The optimal interspace was found to be 1 mm by considering
the following typical case: for 50 mm focal length and 1000 mm object distance (distance from the lens to the plasma), 1 mm on the image plane corresponds to ∼20 mm
on the object plane, a distance that a plasma jet of 20 km/s velocity would traverse
during 1 µsec. The interspace between fibers at the output end was decided for the
best coupling to the spectrometer. Since the image of the fiber array is magnified by
a factor of 3 due to the entrance optics, the 12 fiber images span a vertical distance
3 × (11 × 150) µm = 4.95 mm at the spectrometer’s exit plane. This vertical span
fits all the 12 fiber images in the 6.7 mm height of the ICCD’s active area.
In some experiments, a multitrack fiber bundle has been used instead of the linear
fiber array. The multitrack fiber bundle consists of 9 independent fibers (200 µm
core). The fibers are aligned only at the output end and are freely moving at the
input end as shown in figure 3.7b. Several fibers are connected to collimators at the
input end and are used to observe the plasma with multiple view angles as illustrated
in figure 3.8b.

~5 cm

(a) Fiber array

(b) Fiber bundle

Figure 3.8: (a) Lines of sight of the fiber array. The twelve lines of sight are approximately equally spaced. (b) Lines of sight of the multitrack fiber bundle. In this
example, the dashed line is perpendicular to the axis of the plasma and the solid line
is off-axis toward the electrode. The circle represents the size of the lines of sight
(∼10 mm in diameter).

3.7.3

Visualizing lines of sight

The lines of sight of the spectroscopic system can be visualized by sending a HeNe
laser beam backward through the fibers and observing the red spots this laser beam
makes on the electrodes. A holographic light shaping diffuser is used to vertically
elongate and distribute the single laser beam into all the fibers. Lines of sight are
illustrated in figure 3.8a for the fiber array and figure 3.8b for the fiber bundle.

Chapter 4
Observations
4.1

Evolution of plasma structures

Following the gas breakdown sequence described in section 2.4, the plasma expands
into vacuum undergoing several structural changes as shown in the Imacon camera
image (figure 4.1). The evolution of the plasma has four distinctive stages: (I) formation and collimation of spider legs, (II) coalescing of spider legs into a central plasma
jet, (III) expansion and collimation of plasma jet, and (IV) instability of plasma jet
and spheromak formation. It is interesting to note that, in contrast with the substantial structural changes of the plasma, the current and voltage traces are almost
featureless as shown in figure 4.2.
The first two frames of the camera image show the plasma-filled arch-shaped flux
tubes formed in a structure reminiscent of spider legs. The spider legs are initially
flared but become collimated within a very fast time scale less than 0.5 µsec. The
spider legs stretch out by the hoop force due to the gun current flowing through
them. The flux tube collimation process has been studied in detail by You, Yun, and
Bellan [21].
In the second stage (frames 3–6 in the camera image), the collimated spider legs
become diffusive and start moving toward the center as the gun current ramps up.
The spider legs attract each other since each carries an electric current in the same
direction. They eventually coalesce into a single axially expanding plasma jet.
In the third stage (frames 7–10), the central plasma jet becomes collimated and

45
expands into the vacuum. The jet velocity is Alfvénic (∼30 km/sec). The plasma jet
is very bright compared to the spider legs suggesting a jet density much greater than
the spider leg density. Large density amplification has been observed in the plasma
jet and will be discussed in section 4.3.
In the last stage (frames 11–16), the lengthened plasma jet may keep expanding
or undergo MHD instability depending on the ratio λgun = µ0 Igun /ψgun , where Igun is
the gun current and ψgun is the initial bias poloidal magnetic flux. Hsu and Bellan [9]
showed that low λgun values result in a straight plasma jet, intermediate values lead
to kinking of the jet axis (kink instability), and high values lead to a detachment of
the plasma from the electrodes. The kinking has been identified as a precursor to
spheromak formation.
The present thesis work is focused on the study of the plasma evolution at the
first and the third stages.

4.2

Magnetic flux tube collimation – spider legs

4.2.1

Main observations

The spider leg magnetic flux tubes are initially very faint and flared, i.e., the tube
diameter increases toward the outer electrode. However, the spider legs become collimated and brighter as more particles are ingested from the gas orifices as shown in
figure 4.3.
S. You [21] visualized the plasma flow in the spider legs using the high-speed
high-resolution Princeton camera. Since hydrogen spider legs are less collimated than
spider legs of heavier gas species, N2 and Ne were used in the study. He observed that
the plasma particles are ingested mainly from the inner gas orifice and the direction
of the flow is from the inner orifice toward the outer orifice. The flow velocity is of
the order of ∼100 km/sec, measured by tracing a bright front propagating along the
flux tube axis in the camera images.

0.05 μs

0.675 μs

1.30 μs

1.925 μs

2.55 μs

3.175 μs

3.80 μs

4.425 μs

5.05 μs

5.675 μs

6.30 μs

6.925 μs

7.55 μs

8.175 μs

8.80 μs

9.425 μs

Figure 4.1: Evolving plasma structure (shot# 7217, hydrogen plasma). The frame
sequence is from left to right and top to bottom. The delay of the first frame is
0.05 µsec with respect to breakdown and the inter-frame delay is 0.625 µsec. In the
last few frames, notice the presence of ghosting from previous frames.

III

-20

-1

-40

-2

-60

-3

-80

-4

-100

-5

Voltage [kV]

Current [kA]

Time [Μsec]

Figure 4.2: Gun current trace (solid curve) and voltage trace (dashed curve). Data
correspond to the same plasma discharge (shot# 7217) as in figure 4.1. Times are
with respect to breakdown. Roman numerals indicate the evolution stages.

(a) 2.5 μs

(b) 2.75 μs

Figure 4.3: Collimation of spider legs (image courtesy of S. You). Images are taken
from identical plasma discharges (shot# (a) 4345, (b) 4346, and (c) 4343). Nitrogen
is injected from the inner (cathode) orifices and neon from the outer (anode) orifices.
Times correspond to the delay of the 10 nsec camera exposure with respect to the
main bank discharge.

48
250

Counts

150

100

399.4

399.5
Wavelength [nm]

(a) Doppler shifts

Velocity [km/s]

200

10
5 ×

1.5

2.0

2.5
3.0
3.5
Time [Μsec]
(b) Doppler velocity

Figure 4.4: (a) NII spectra showing blue Doppler shifts (nitrogen discharge; shot
range 6811–6830; spectrometer ICCD exposure = 1 µsec). Dashed line indicates the
rest frame wavelength 399.5 nm. Solid lines are Lorentzian curve fits. The exposure
delays of the spectra, starting from the bottom to the top spectrum, are 1.7, 2.2, 2.6,
3.1, and 3.8 µsec, respectively, with respect to breakdown. The peaks of the spectra
are connected to show the trend of increasing blue shifts. The line of sight intercepted
one of the spider legs near its inner foot point perpendicular to the electrode. (b)
Line of sight velocities estimated from the Doppler shift measurements.

4.2.2

Doppler shift measurements

To verify that the propagation of the bright front is indeed a plasma flow and not
an ionization front, the Doppler shifts of emission spectra from a single spider leg
were measured by the present author [21] using the spectroscopic system with the
multitrack fiber bundle (section 3.7.2). The measured spectra showed blue shifts
corresponding to ∼20 km/sec line of sight velocities as shown in figures 4.4a and
4.4b. The blue Doppler shifts confirm that the propagation of the bright front is
indeed the flow of plasma ions.

4.2.3

Flow velocity profile along the spider leg

Flow velocities U (z) along the spider leg axis can be expressed in terms of the plasma
density n(z) assuming that the length of the spider leg (L) remains constant during measurement and the flow stagnates at the outer orifice (U (L) = 0). These
assumptions seem reasonable based on the camera images (figure 4.3). Integrating

49
the continuity equation yields
Z L

∂n(z 0 )
(n(z )U (z )) +
dz 0 = 0,
∂z
∂t
Z L
n(L)U (L) − n(z)U (z) +
n(z 0 )dz 0 = 0,
∂t z
1 ∂
U (z) =
n(z) ∂t

Z L

n(z 0 )dz 0

(∵ U (L) = 0 by assumption),

(4.1)

where z is the axial distance from the inner orifice.
Since line radiation in plasma results primarily from collisional excitations due to
electron impact, the radiation intensity I will be proportional to the electron density
(ne ) times the emitter density (nemitter ). Assuming that the emitter density is some
constant fraction of the ion density (ni ) and invoking charge neutrality (ni = ne )
yield
I ∝ ne nemitter ∝ ne ni ∝ n2e .

(4.2)

Thus, the flow velocity U (z) can be estimated using the image brightness I(z) as
U (z) = p
I(z) ∂t

Z Lp

I(z 0 )dz 0 .

(4.3)

The estimated velocities are shown in figure 4.5b and are consistent with the propagation velocity of the bright front.

4.2.4

Discussion

The direction of the flow is counter-intuitive since one would normally think that ions
would move toward the cathode (inner electrode) not away from it. However, this
intuitive notion is incorrect because the the static electric field vanishes inside the
plasma. According to the MHD pumping model (section 1.4), the plasma (ion) flow
in a magnetic flux tube is driven by the MHD force (equation (1.31)),
mn

dU
µ0 I 2 ∂a
= 2 3 ,
dt
2π a ∂z

(4.4)

50
1.0

0.6
0.4
0.2
0.0

10
Axial length [cm]

ó óó ã
15

(a) Brightness

Velocity [km/sec]

0.8
Image intensity

150

éé
ó = 2.5 Μs
éã
óóó
ã = 2.75 Μs
óéã
éãã
éã ã ã
ééãã ã ãã
é = 3.0 Μs
óóé ã
ééé é
óóóéééã
éééé éã
ó ã
é é ééé é é
óóó é ããããã ééééé ééé éééé
éé é
ãã
é ééé
óóóóóó ãã ã
ãããã é é é é
ó ó ó ã ãã
é é é ãã
ã ãã é ééããé éó
ãã
ó ó
ã ó
ãã ãó
é é éééã
óóó ó ó óó ãó ããó
éó
ãééó
ó ó ó ó ã ó ã óãó
ããó
óó óóóó óã ó
óóéóãã
ó ó óó ó
óó
óé
20

100

10
15
Axial length [cm]

(b) Flow velocity

Figure 4.5: (a) Brightness along the spider leg that is shown in figures 4.3a–c (data
courtesy of S. You). Triangle, square, and circle correspond to the timings 2.5, 2.75,
and 3.0 µsec, respectively. Data points at z < 2 cm are excluded due to a relatively large measurement error. (b) Flow velocity profile at 2.75 µsec obtained from
equation (4.3).
where m is the ion mass, a is the local flux tube radius, and ∂a/∂z is the flaring of the
flux tube. For the spider legs, ∂a/∂z > 0 and so an axial flow from the inner cathode
orifice to the outer anode orifice will be driven by this MHD force. The spider legs
become collimated due to the flow stagnation at the anode orifice as explained by the
MHD pumping model in section 1.4.
S. You estimated a spider leg density n ∼ 2 × 1021 m−3 using a ∼ 4 mm, ∂a/∂z ∼
0.02, and dU/dt ∼ 1011 m/sec2 measured from the camera images and assuming
that a single leg carries one-eighth of the total gun current 120 kA (that is, I =
120/8 = 15 kA). Since the formation and collimation of the spider legs take place
within 0.5 µsec, the pumping speed of the MHD force through a single gas orifice is

dN1
dt

MHD

∆n · V
2 × 1021 × 10−5
= 4 × 1016 particles/µsec,
∆t
0.5

(4.5)

where V ∼ 10−5 m3 is the volume of a single spider leg. This MHD pumping speed is
much faster than the measured particle flux rate for a single gas line by gas puffing,
dN1 /dt ∼ 2×1015 µsec−1 (equation (2.4)). Thus, the ionized particles are dominantly
ingested by the MHD force.
The Alfvén velocity UA associated with the azimuthal magnetic field due to the

51
spider leg current I is
UA =

Bφ2
∼ 100 km/sec,
µ0 nm

(4.6)

where Bφ ∼ µ0 I/2πa ∼ 1 Tesla and m is the mass of the nitrogen atom. Thus, the
measured jet velocity (∼100 km/sec) is Alfvénic as expected by the model and is
driven by the MHD pumping force. In the hydrogen spider legs, the flow velocity
(& 100 km/sec) is faster due to the smaller ion mass and the density (. 1021 m−3 ) is
lower due to less collimation.

4.3

Large density amplification in the plasma jets

4.3.1

Large Stark broadening of hydrogen spectral lines

After the spider legs coalesce into a central plasma jet, the jet expands axially with
its foot point attached to the inner electrode (cathode) until the jet undergoes instabilities at a later time. Jet velocities of ∼20–50 km/sec were measured by following
the forefront of the jet in the camera images. To rule out the possibility that these
camera images would result from propagation of either a plasma wave or an ionization front rather than an actual flow, Doppler shifts of spectral lines emitted from
the plasma jet were measured [22] using the spectroscopic system with the multitrack
fiber bundle. The observed Doppler shifts (figure 4.6) confirmed the flow velocity
deduced from the camera images.
However, some spectral lines showed unexpectedly large broadening which greatly
exceeds the instrumental broadening and the thermal Doppler effect for a credible ion
temperature [22] (see the discussion of section 6.1.2). For instance, some measured
Hβ line profiles showed a full width at half maximum (FWHM) & 1 nm, which
would correspond to a Doppler ion temperature & 1000 eV. Such large broadenings
suggest a strong Stark effect as discussed in section 6.1.2. The characteristic central
dip observed in some Hβ line profiles as shown in figure 4.7 is also an indisputable
evidence for a strong Stark effect (ref. chapter 5 and [23]).

ë ×
ë × × ×
ëë× ×
0.8
ë ë ×
ë ×
ëë ëë
ë ×× ×
0.6
× ëë ë ë ë×
××
ëë
0.4
ëë×××
ë×
ëëë
××
ë ëë ××
ë × ×
ëë ×
ëëë
0.2
×××
ëë×
ë ë
ë×
ë ëë
ë×
ë×
ë×
ëëëëë ë × ëëë ë ë
ëë ë ëëëëë ë ×××××××
ëë×ë ëë×
ëë ××
ë ××
×ëë×
ë×
×ëë ë×
ë××××
ë××
×××××
ë×
ë×
ëë×
ëë
ëëë
×ë ×
×ëë××
ë××
×××
0.×× ë

Intensity (normalized)

486.

486.1
486.2
Wavelength [nm]

486.3

Figure 4.6: Doppler shifts of Hβ lines (spectrometer ICCD exposure = 1 µsec). Line
of sight was arranged such that the plasma jet moved toward the optical probe in one
case (solid blue curve; shot# oma11.12) and away from it in the other case (dashed
red curve; shot# oma11.43). The peak of the line profile is determined by Lorentzian
curve fitting. As expected, a blue shift is observed in the first case and a red shift in
the latter case. The shifts are ∼0.017 nm corresponding to ∼11 km/sec line of sight
velocity. Rest-frame Hβ (486.133 nm) and Dβ (486.0029 nm) lines from a deuterium
lamp are shown at the bottom.

53
2000

Counts

1500

á á
áá áá
áá

1000

500

á0
484

485

486
Wavelength [nm]

487

488

Figure 4.7: Hβ profile showing the characteristic central dip (shot# oma11.28; spectrometer ICCD exposure = 1 µsec). The square points are the theoretical Stark
profile for ne = 1.3 × 1022 m−3 and T = 2 eV according to Stehlé [24]. A Doppler
profile (dashed line) for Ti = 5 eV at the Hβ rest-frame wavelength (486.133 nm) is
also shown for comparison with the much broader Stark profile. The estimated Stark
FWHM according to equation (6.2) is ∼1.1 nm, corresponding to ne ∼ 1.2×1022 m−3 .
Note that the data points near the profile tail ends are flat since they are outside the
bandpass of the spectroscopic system.

4.3.2

Density measurements

Electron densities were measured from the Stark broadened spectral lines by the
method described in section 6.2. The time dependence of the electron density of a
hydrogen plasma jet is plotted in figure 4.8. Peak density of the order 1022 m−3 is
observed when the spider legs merge completely. The density is ∼1020 m−3 when
the plasma starts to form the central jet and increases by a factor of 100 when the
jet fully forms. Merging of the spider legs does not decrease the total volume of the
plasma and thus cannot account for this large density amplification. On the other
hand, the observed density amplification is consistent with the magnetic flux tube
becoming filled with plasma by the ingestion/collimation process described in the
MHD pumping model.
Similar density amplification is observed in non-hydrogen plasmas. For nitrogen
plasma jets, Stark broadened NII spectral lines are used for density estimation and
a typical emission spectrum is shown in figure 4.9. Nitrogen plasma jets show even
greater peak density of ∼1 × 1023 m−3 (see figure 4.10).

Density [1022 m-3]

2.0

1.5

1.0

0.5

0.0 á

10
Time [Μsec]

Figure 4.8: Electron density of a hydrogen plasma jet vs. time (shot range:
oma11.16–69). Density is estimated from the Hβ line profiles. Squares correspond to
the solid (oblique) line of sight and triangles to the dashed (perpendicular) line of
sight in figure 3.8b. For the perpendicular line of sight, the density is zero initially and
appears suddenly at t = 9 µsec as expected because the jet takes time to reach the
line of sight volume. This is consistent with the measured jet velocity (∼30 km/sec).

3000

2500

Counts

2000

1500

1000

500

423.5

424.
Wavelength [nm]

424.5

Figure 4.9: Stark broadened NII spectral lines from a nitrogen plasma jet (shot#
7245; spectrometer ICCD exposure = 0.1 µsec). The rest-frame wavelengths of the
two spectral lines are 423.705 nm and 424.178 nm, respectively. Instrumental function
(dashed line) is shown at the two rest-frame wavelengths for width comparison. Solid
line is a Lorentzian curve fit to the spectrum. Estimated density is ∼8 × 1022 m−3 .

Density [1022m-3]

à à

10
Time [Μsec]

Figure 4.10: Electron density of a straight column nitrogen plasma jet vs. time (shot
range: 6736.01–30). Density is estimated from the NII 424.178 nm line profiles as
illustrated in figure 4.9. Solid curve is a polynomial fit to the data.

4.3.3

Density and velocity profile along the jet axis

Density profiles along the axis of a straight nitrogen plasma jet were measured at
several different timings using the spectroscopic system with the fiber array. The
axial density profile has a sharp peak initially but becomes uniform as the jet expands
axially as can be seen in figure 4.11a.
Velocity profiles were measured from the Doppler shifts of the same spectra used in
the density measurement and are shown in figure 4.11b. The velocity peak is always
ahead of the density peak by 3–5 cm in space and by ∼1 µsec in time. Negative
velocity gradient (i.e., slowing down of the flow) is clearly seen near the jet front.
A detailed comparison with the camera images (figure 4.13) shows that the bright
blob in the jet coincides with the region between the density peak and the velocity
peak. The blob propagates along the jet axis with an Alfvénic velocity of ∼40 km/sec.
The blob is faster than the expanding outskirts of the jet and catches up with it later
as seen in the last frame of figure 4.13. This observation is consistent with the negative
velocity gradient near the jet front.
For the detailed analysis in the next section, three density profiles are selected
from figure 4.11a and shown overlaid in figure 4.12a. The corresponding velocity
profiles are shown overlaid in figure 4.12b.

Density [1022m-3]

20
15

6.5Μs

6.0Μs

5.5Μs

5.0Μs

4.5Μs
25

6.5Μs

6.0Μs

5.5Μs

5.0Μs

4.5Μs
25

15
z [cm]

(a) Density gradient

120

Velocity [km/sec]

100

80
60

15
z [cm]

(b) Velocity gradient

Figure 4.11: (a) Density and (b) velocity along the axis of a straight column nitrogen
plasma jet (shot range: 8209–8227). Densities were estimated from Stark broadened
NII 404.131 nm lines. Velocities were estimated from the Doppler shifts of the same
NII lines. The line of sight angle (∼60◦ ) and the blue Stark shifts [25] (∼0.8 pm
per 1022 m−3 ) are taken into account in the velocity measurement. Solid curves are
polynomial fits to the data and dashed horizontal lines indicate zero lines. Each data
point is an average of three measurements. Measurement errors are the larger of
±1 × 1022 m−3 or ±15% for the density and ±5 km/sec for the velocity. Spectrometer
ICCD exposure was 0.5 µsec.

Density [1022m-3]

= 5.0 Μs
= 5.5 Μs
= 6.0Μs

15
20
25
z [cm]

(a) Density gradient

Velocity [km/sec]

ó ó
40
30
20
10

ó é ã ã

é é

15
z [cm]

ó = 5.0 Μs
ã = 5.5 Μs
é = 6.0Μs

ó é
ã é
ã ó
é ó
20
25

(b) Velocity gradient

Figure 4.12: Detailed view of the density and velocity profiles. The density increases
near the jet front and becomes uniform in the main body of the jet as indicated by
arrows.

4.5 μs

5.5 μs

6.5 μs

7.5 μs

Figure 4.13: Nitrogen plasma jet (shot# 8209). The bright blob propagates with an
Alfvénic velocity of ∼40 km/sec catching up the expanding outskirts of the jet.

4.4

Discussion

4.4.1

MHD pumping

The Stark broadening density measurements show that the plasma jets become very
dense on a microsecond time scale. This result is supported by an independent density
measurement from a laser interferometer [26]. The density of the pre-breakdown
neutral gas, as measured by a fast ion gauge, is only 1017 m−3 [21]. The measured
peak density of the nitrogen plasma jet is of the order 1023 m−3 at the same location
6 µsec later, so the particle density has increased by a factor of 106 . This steep density
rise cannot be explained by a pinch effect associated with the gun current flowing in
the jet flux tube. The merging of the spider legs cannot account for the peak density
of the jet, either, since the spider leg density (∼1021 m−3 ) is 100 times less than the
jet density and the total volume of the spider legs is smaller than the volume of the
central jet. Plasma particles must be pumped into the flux tubes from the source gas
orifices.
According to the MHD pumping model (ref. section 1.4), the driving force of the
plasma jet dynamics is the axial gradient of the azimuthal magnetic energy associated
with the axial current (see equation (1.30)),
2
dUz
∂ Bφ,a
≈−
dt
∂z µ0

(4.7)

where Bφ,a = µ0 I0 /2πa is the azimuthal magnetic field at the flux tube radius a.
The effective potential Bφ,a
/µ0 has negative slope everywhere because Bφ,a decreases

along the axis due to the flared geometry of the jet flux tube. Thus, the plasma will
fall down the slope and gain kinetic energy as it moves. The order of magnitude of
the kinetic energy gain will be
ρUz2 ∼ −∆

Bφ,a
µ0

(4.8)

z=0

Using ρ = mi ne yields ne ∼ Bφ,a
/µ0 z=0 /mi Uz2 , where mi is the ion mass. For

60
nitrogen plasma jets, the observed Uz ∼ 40 km/sec and Bφ,a ∼ 1.0 T give an electron
density ne of the order 1022 m−3 , which is consistent with the Stark broadening density
measurement.

4.4.2

MHD collimation

However, the collimation of the axially expanding jet cannot be explained by this
consistency check alone. Since the collimation is associated with slowing down of flow
and accumulation of particles according to the MHD pumping model, the observed
features of the density and velocity profiles in figure 4.12 are now examined using the
continuity equation:
∂ρ
∂ρ
∂Uz
= −∇ · (ρU ) = −Uz
−ρ
− ∇⊥ · (ρU⊥ ),
∂t
∂z
∂z

(4.9)

where Uz ∂ρ/∂z represents the density change due to the axial convection, negative
ρ∂Uz /∂z corresponds to an axial compression, and negative ∇⊥ · (ρU⊥ ) corresponds
to a radial pinching. Since the diameter of the jet remains approximately constant
during the axial expansion, the radial pinching effect can be ignored for the moment,
yielding
∂ρ
∂Uz
∂ρ
≈ −Uz
−ρ
∂t
∂z
∂z

(4.10)

Suppose there is no flow gradient, i.e., ∂Uz /∂z = 0. Then, ∂ρ/∂t ≈ −Uz ∂ρ/∂z,
which can be integrated to give ρ(z, t) = ρ(z − Uz t). Thus, the axial density profile
would just translate without any deformation if there is no flow gradient. Our plasma
jets show a large negative flow gradient (∂Uz /∂z
0) at the jet front and so the
density increases in that region because
∂ρ
∂Uz
≈ −ρ

∂t
∂z

(at the jet front).

(4.11)

The observed density increase at the jet front is indicated by an arrow in figure 4.12a.
The observation that the density peak always appears ∼1 µsec after the velocity peak
also conforms with the continuity equation because (1) at the location of the velocity

61
peak, ∂Uz /∂z = 0 and ∂ρ/∂z < 0 (see figure 4.12), giving
∂ρ
∂ρ
≈ −Uz
>0
∂t
∂z

(near the velocity peak),

(4.12)

and (2) on the other hand, at the density peak, ∂ρ/∂z = 0 and ∂Uz /∂z > 0, giving
∂ρ
∂Uz
≈ −ρ
<0
∂t
∂z

(near the density peak).

(4.13)

The collimation of the axially expanding jet can be summarized as follows according to the MHD pumping model. The axial component of the J × B force accelerates
the jet into the flared magnetic flux tube produced by the bias coil. The flow velocity
slows down near the jet front where the flux tube radius is large. The slowing down
of the flow results in accumulation of mass (∂ρ/∂t > 0) as well as the azimuthal
magnetic flux carried by the mass flow. As the magnetic flux accumulates, the magnetic field intensity increases and so does the pinch force. The amplified pinch force
subsequently squeezes the flared jet front (∇⊥ · U⊥ < 0). This squeezing action at the
jet front continues while the jet expands axially into the flared flux tube, resulting in
an elongated collimated jet. The density becomes uniform in the main body of the
jet as it becomes collimated. This collimation scenario is sketched in figure 4.14. Essentially the same scenario accounts for the collimation of the spider legs [21], except
that the collimation will be more efficient in the spider leg flux tubes because of the
flow stagnation at the anode gas orifices (figure 4.15).
The cause of the flow deceleration has not yet been investigated but could be the
tension of the curved bias magnetic field lines or buildup of neutral particles swept
by the jet at the flow front.

4.4.3

Total particle flux by the MHD pumping

The total particle flux ingested into the jet is approximately

dNT
dt

MHD

= πa2 ne Uz ∼ 1018 particles/µsec,

(4.14)

jet front

cathode
Figure 4.14: Collimation of the axially expanding jet. The initial flux surface is
indicated by solid curves (#1) and the later flux surfaces by dashed curves (#2,3).
Elliptic discs represent the jet front at different timings. The flared flux surface becomes collimated as the jet expands axially. Block arrows indicate the flow direction.

flow stagnation
( outer orifice)
Figure 4.15: Collimation of the spider leg. Plasma flow stagnates at the outer anode
orifice. Since the foot points of the spider leg are fixed, its axial length remains
approximately constant during collimation. The initial flux surface is indicated by
solid curves and the later flux surface by dashed curves.

63
using a ∼ 3 cm, ne ∼ 1022 m−3 , and Uz ∼ 30 km/sec. Again, this MHD pumping
speed is much faster than the gas puffing, dNT /dt ≈ 1016 µsec−1 (cf. equation (2.3)).
Integrating over the plasma duration of ∼10 µsec (which is limited by the capacitance
of the main gun bank), the total number of particles ingested by the MHD pumping
is NT, MHD ∼ 1019 . This amounts to about 20% of the total number of puffed particles
(cf. equation (2.2)).

4.4.4

Application – tokamak fueling

Recent work by Voronin et al. [27] suggests that our coplanar plasma gun producing high density plasma jets by the MHD pumping mechanism might be capable
of fuelling a magnetized plasma of fusion interest. In the Voronin et al. experiment, high kinetic energy plasma jets were produced by a double-stage Marshall
gun, which first ionizes a hydrogen gas released from titanium grains and then accelerates the resulting hydrogen plasma. These plasma jets were successfully injected into the Globus-M spherical tokamak and had parameters comparable to
our experiment (Globus-M: ne ∼ 1022 m−3 , total number of accelerated particles
N ∼ 1–5 × 1019 , U ∼ 50–100 km/sec, and our experiment: n ∼ 1022–23 m−3 ,
tot

Ntot ∼ 1019 , Uz ∼ 30 km/sec). This suggests that our plasma jets may be suitable for
tokamak fuelling. Higher velocity for deep penetration [1, chap. 16] can be achieved
by increasing the gun current and the minimal contact with electrodes in our plasma
gun ensures purity of the plasma jets.

Chapter 5
Stark broadening
The Stark effect is the splitting and shifting of a spectral line in the presence of an
electric field. The Stark effect differs significantly among spectral lines and can be
divided into two categories according to its dependence on the applied electric field,
namely, the linear Stark effect and the quadratic Stark effect. The Stark effect is
linear for hydrogenic lines and quadratic for non-hydrogenic lines in general. The
linear Stark effect is usually much stronger than the quadratic Stark effect.
In a plasma environment, a radiating atom or ion (emitter) is under the influence
of the local electric fields produced by the surrounding ions and electrons within the
Debye sphere of the emitter. The collective Stark effect due to these local electric fields
results in broadening of spectral lines, i.e., Stark broadening. The emitter experiences
two very different electric fields, the quasi-static electric field due to slow moving ions
and the transient electric field due to fast moving electrons. The quasi-static electric
field splits a degenerate atomic level of the emitter into distinct individual components
called Stark components. The Stark components are then broadened by the transient
electric field due to electron impacts. If the atomic level is not degenerate, the level
undergoes only the electron impact broadening. The resulting line profile will be
the sum of all possible transitions from the Stark components of an upper level to
the Stark components of a lower level. The profile width serves as the first-order
description for the Stark broadened line profile although the profile can have other
features such as shifts and asymmetries. The profile width has a strong dependence
on charged particle density and so the Stark broadening provides a convenient and

65
inexpensive method for determining plasma density [23].
Stark broadening has been studied most extensively for hydrogen lines both in
theory and experiment because the linear Stark effect is much stronger than the
quadratic Stark effect and the hydrogen atom is the simplest quantum-mechanical
system. Hydrogen Balmer lines have been the most popular choice for plasma density diagnosis because of well-established theoretical and experimental studies on the
Stark broadening of Balmer lines. In particular, the Hβ line is the best density indicator because its Stark width is nearly independent of plasma temperature. When
Hβ Stark broadening is small compared to other broadening effects such as Doppler
broadening and Zeeman splitting, higher-n Balmer lines can be used since the Stark
broadening is generally larger for higher principal quantum number n. For example,
merging of high-n Balmer lines (i.e., Inglis-Teller limit) has been utilized to determine the density in edge regions (n ∼ 1020–21 m−3 , B ∼ several Teslas) of tokamak

devices [28] [29] [24]. The present experiment lies in the opposite situation, namely
where Hβ Stark broadening is large compared to Zeeman and Doppler effects.
The following sections summarize Stark broadening from basic principles. More
thorough theoretical account of Stark broadening can be found in reviews by Margenau and Lewis [30], Lisitsa [31], and Luque, Calzada, and Sáez [32].

5.1

Quadratic Stark effect

Consider an atom having a single electron or a single valence electron in its outermost
shell subject to a uniform electric field F in the positive z-direction. The Hamiltonian
of the system is the sum of the unperturbed Hamiltonian H0 = p2 /2me + V (r) and
the perturbing Hamiltonian H1 = eF Z:
H = H0 + H1 =

p2
+ V (r) + eF Z,
2me

(5.1)

where p, me , and e are the momentum, mass, and charge of the electron, and V (r) is
a central potential.

66
The effect of the perturbing electric field on the energy levels can be evaluated
using the perturbation theory. According to the perturbation theory [33, chap. 16],
the change in the energy level Ek is given by (up to the second order in H1 )
∆Ek ≃ hk|H1 |ki +

X |hk|H1 |k 0 i|2
k0 6=k

Ek − Ek 0

(5.2)

where |ki’s and Ek ’s are the eigenstates and the corresponding energy levels of the
unperturbed Hamiltonian H0 . The eigenstate |ki and the energy level Ek are commonly represented as |n, l, mi and Enlm by the triple quantum number (n, l, m), i.e.,
principal quantum number n, azimuthal quantum number l, and magnetic quantum
number m. It is assumed that the unperturbed energy levels are completely known
and non-degenerate (i.e., Ek 6= Ek0 unless k = k 0 ) so that the perturbation theory can
be applied without difficulty.
Equation (5.2) is rewritten by replacing |ki with |n, l, mi,

∆Enlm ≃ eF hn, l, m|Z|n, l, mi + e F

(n0 ,l0 ,m0 )6=(n,l,m)

|hn, l, m|Z|n0 , l0 , m0 i|2
Enlm − En0 l0 m0

(5.3)

This expression can be simplified by applying two selection rules derived in appendix E, i.e., hn, l, m|Z|n0 , l0 , m0 i 6= 0 only if (1) m0 = m and (2) l = l0 = 0 or
l0 = l ± 1. The first order term eF hn, l, m|Z|n, l, mi vanishes unless l = 0 by the
second selection rule. However, hn, 0, 0|Z|n, 0, 0i = 0 since |n, 0, 0i is spherically symmetric. Thus, the first order term always vanishes leaving only the second order terms
and hence ∆Enlm is quadratic in the applied electric field F .
∆Enlm ≃ e2 F 2

X X |hn, l, m|Z|n0 , l0 , mi|2
n0 l0 =l±1

Enlm − En0 l0 m

≡ − αnlm F 2 ,

(5.4)

where αnlm is the polarizability of the eigenstate |nlmi. For example, the polarizability of the ground state (n = 1) of the hydrogen atom (this is the only hydrogen state
that the perturbation theory can handle for now since the n > 1 hydrogen states are

67
degenerate) is
α100 = −2e2

X |h1, 0, 0|Z|n0 , 1, 0i|2
n0 >1

E100 − En0 00

= 18π0 a30 ,

(5.5)

where a0 is the Bohr radius.
In general, non-hydrogenic energy levels are non-degenerate and so they experience
the quadratic Stark effect. Strictly speaking, most energy levels of the Hamiltonian
H0 are degenerate with respect to m (eigenstates of the same n, l but different m have
the same energy) due to the spherical symmetry of the potential V (r). However, this
degeneracy becomes inconsequential in the application of the perturbation theory
because the states with the same l will be excluded by the second selection rule. In
contrast, the n > 1 energy levels of the hydrogen atom are intrinsically degenerate,
i.e., degenerate with respect to l (eigenstates of the same n but different l, m have the
same energy) due to the pure Coulomb potential V (r) ∝ 1/r. The n > 1 hydrogen
energy levels experience a different Stark effect due to this special degeneracy, which is
considered separately in the following section. Non-hydrogenic atoms have an effective
potential differing from the pure Coulomb potential (∝ 1/r) due to the presence of
the inner electrons shielding the nucleus.

5.2

Linear Stark effect

Consider an Nn -fold degenerate energy level En
H0 |n, gi = En |n, gi

(g = 1, . . . , Nn ),

where the eigenstates |n, gi are specified by a pair of quantum numbers (n, g) instead
of the triple (n, l, m) for notational simplicity. The perturbation equation (5.3) cannot
be applied to the degenerate eigenstate |n, gi because the second order terms involving
the eigenstates |n, g 0 i (g0 6= g) are ill-defined as the denominator Eng −Eng0 = En −En =
0. This difficulty can be avoided by choosing a new orthonormal linear combination
of the original degenerate eigenstates (i.e., a new basis spanning the degenerate space
of the energy level En ) to be simultaneous eigenstates of the perturbation H1 , which

68
is always possible:

|n, gi =

hn, g00 |n, gi |n, g0 i

and

(5.6)

g00 =1

H1 |n, gi = (eF Z)|n, gi = eF λng |n, gi,

(5.7)

where |n, gi’s are the new basis vectors, |n, g0 i’s are the original basis vectors, and
λng ’s are eigenvalues of the operator Z. Then, all the singular terms in equation (5.3)
vanish since hn, g 0 |H1 |n, gi = eF λng hn, g 0 |n, gi = 0 (g0 6= g), yielding

∆Eng ≃ hn, g|H1 |n, gi +

(n0 ,g 0 )6=(n,g)

|hn, g|H1 |n0 , g 0 i|2
Eng − En0 g0

X |hn, g|H1 |n0 , g 0 i|2
= eF λng +
En − En0 g0
n0 6=n,g 0
≈ eλng F.

(5.8)

Thus, for a degenerate state, ∆Eng is linear in the applied electric field F .
As an example, consider the n = 2 hydrogen states, i.e., the 2s state |2, 0, 0i and
the three 2p states |2, 1, 0i, |2, 1, 1i, and |2, 1, −1i. The operator Z can represented
in a matrix form using these eigenstate vectors in order to find the eigenvalues λ2g ’s:

h2, 0, 0|Z|2, 1, 0i
 h2, 0, 0|Z|2, 1, 0i
Z=

0 0

0 3a0
 
 
0 0   3a0 0
=
 
0 0   0
 
0 0

0 0

0 0 
 , (5.9)
0 0 
0 0

where the selection rules (ref. appendix E) are utilized to evaluate the matrix element Zij = h2, gi |Z|2, gj i. The four eigenvalues corresponding to this matrix are
3a0 , −3a0 , 0, and 0. Thus, the four-fold degenerate energy level E2 splits into three
levels by the applied electric field F .
∆E2 = 0, ±3a0 eF.

(5.10)

69
The linear Stark effect is much stronger than the quadratic Stark effect in general.
For instance, compare the linear Stark effect of the hydrogen n = 2 state (∆E2 )
calculated above to the quadratic Stark effect of the hydrogen n = 1 state (∆E1 )
discussed in the previous section.
∆E2
3a0 eF
e/(3π0 a20 )
3a0 eF

1,
∆E1
α100 F 2 /2
9π0 a30 F 2
because e/(3π0 a20 ) ≈ (electric field strength at the Bohr radius from a charge e)
≈ 5 × 1011 V/m would be much larger than the applied electric field F in most cases.
The hydrogen spectral lines show a very large Stark effect compared to any other
non-hydrogenic spectral lines because of the strong linear Stark effect.

5.3

Line broadening by charged particles

Imagine a charged particle with charge q moves past a radiating atom (emitter).
The charged particle exerts a time-varying Coulombic electric field F (r(t)) ∼ q/r2 (t)
on the emitter, where r(t), the distance from the charged particle to the emitter,
depends on the initial location and velocity of the charged particle. Such an electric
field is commonly called as a microfield in the Stark theory literature. The microfield
perturbs the energy states of the emitter via the Stark effect:
∆ω(t) = Ca F k (r(t)),

(5.11)

where the Stark effect is represented by the frequency shift ∆ω instead of the energy
level shift ∆E = ~∆ω, Ca is the Stark effect constant of an energy state a, and k = 1
for the linear Stark effect and k = 2 for the quadratic Stark effect.
The critical impact parameter for this Coulombic collision to induce a total loss of
coherence of the energy state is called the Weisskopf radius, ρW . The Weisskopf radius
is roughly proportional to the Stark effect constant Ca and inversely proportional to

70
the velocity v of the charged particle.
ρW ∼

(5.12)

A collision with the impact parameter much smaller than ρW will not cause a significant Stark effect. The Weisskopf radius defines the effective cross section of the Stark
effect, also known as the optical cross section,
σW = πρ2W ∼ π

Ca2
v2

(5.13)

The Weisskopf radius also determines the effective duration of the critical collision
called as Weisskopf collision time
τW =

Ca
ρw
∼ 2.

(5.14)

For tW much longer than the mean lifetime of an excited state, the microfield is
effectively static during the collision. On the other hand, for tW much shorter than the
mean lifetime of the excited state, the microfield can be considered as an instantaneous
perturbation. The former assumption (quasi-static approximation) is usually valid for
ion collisions and the latter (impact approximation) for electron collisions since the
electron thermal velocity is much faster than the ion thermal velocity.
The Weisskopf radius ρW defines another important parameter, the effective number of particles interacting simultaneously with the emitter.
g ≡ nρ3W ,

(5.15)

where n is the particle density. For example, if g
1, only the nearest particle
interacts with the emitter, but, on the other hand, if g
1, a large number of
particles simultaneously affect the emitter.
The ensemble average of the Coulombic collisions of the emitter with charged
particles results in broadening of the energy levels, a collective Stark effect called as

71
ion
emitter
dr

Figure 5.1: Nearest neighbor approximation of the ionic electric field.
Stark broadening. The average microfield F (r(t)) depends on the velocity and density
distributions of the charged particles and so does the Stark broadening.

5.3.1

Quasi-static approximation – ion microfield

The microfield produced by an ion can be assumed to be static since the collision of the
ion with the emitter is a slow process (due to the slow ion thermal velocity) compared
to the atomic line transitions (lifetime τ0 ∼ 10 nsec for strong lines), i.e., τW
τ0 . It
is also assumed that the collisions are binary meaning that the emitter is under the
influence of only one ion at a time. This assumption will be valid if g = ni ρ3W . 1,
where ni is the ion number density. It follows from these two assumptions that the
profile of the frequency shift ∆ω will depend on the distribution of the individual
static ion microfield F (r(t)) = F (r) ∼ Ze/r2 , where Z is the charge number of the
ion.
Consider only the nearest ion to get an approximate microfield distribution. Let
p(r)dr be the probability that the nearest neighbor is in the shell (r, r + dr), fi (r)dr
be the probability to find an ion in the shell, and P ∗ (r) be the probability that no
ion is inside the radius r. Note that fi (r) = ni 4πr2 dr. Then,
p(r)dr = P (r) × fi (r)dr 
Rr
P ∗ (r) = 1 − 0 p(r)dr 

=⇒

p(r)
fi (r)

Z r

1−

p(r)dr .

(5.16)

72
Differentiate and then integrate the above equation to solve for p(r)
dr

p(r)
p(r)
= −p(r) = −
× fi (r) =⇒ log
= − fi (r)dr
fi (r)
fi (r)
=⇒ p(r) = fi (r) exp(− fi (r)dr) = ni 4πr2 exp(−ni 4πr3 /3).
p(r)
fi (r)

(5.17)

The ions in the shell (r, r + dr) will induce Stark shifts in the range (∆ω, ∆ω + dω)
according to equation (5.11), ∆ω = Ca F k (r). It is convenient to normalize the
radius by the inter-particle distance r0 = (3/4πni )1/3 and the Stark frequency shift
by ∆ω0 = Ca F k (r = r0 ):
k
∆ω
F (r)
β≡
= (r/r0 )−2k (∵ F ∼ 2 ),
∆ω0
F (r0 )
−2k−1
dr
dr
dβ = −2k
= −2kβ .
r0
r0

(5.18)
(5.19)

Equations (5.17)–(5.19) yield the profile I(β) of the Stark frequency shift,
I(β) = p(r)

dr
ni 4πr3
(r/r0 )3
exp(−ni 4πr3 /3) = 3
exp(−(r/r0 )3 )
dβ
2kβ
2kβ

3 −(1+3/2k)
exp(−β −3/2k ).
2k

(5.20)

A more involved treatment taking into account all ions, not just the nearest one,
leads to a profile known as Holtsmark function [34].
I(β) = β

Z ∞

x sin(βx) exp(−x3/2k ) dx

(5.21)

The real profile of the Stark frequency shift by ions will be in-between the Holtsmark
function and the nearest neighbor profile (see figure 5.2) because of the Debye screening. The separation of the two peaks occurring at β ∼ ±1.5 may serve as the width

73
I(Β)
0.6
0.5
0.4
0.3
0.2
0.1
-4

-2

Figure 5.2: The Holtsmark profile (solid line) and the nearest neighbor profile (dashed
line) for the linear Stark effect (k = 1).
of the profile. Thus, the width of the un-scaled profile I(∆ω) is
width of I(∆ω) ∼ 3∆ω0 ∼ 3Ca F k (r0 ) ∝ Ca r0−2k = Ca (3/4πni )−2k/3
2k/3

∝ C a ni

∝ Ca n2k/3

(5.22)

where ne = Zni is the electron density of the plasma.

5.3.2

Impact approximation – electron microfield

In contrast to ions, fast moving electrons perturb the emitter only momentarily, i.e.,
the duration of electron impact is much shorter than the lifetime of the excited states
of the emitter. In this limit, the Stark effect depends only on the frequency of the
electron impact, whose inverse, the average interval between impacts, shall not be
confused with the impact duration. The impact frequency νc is related to the Stark
effect cross section σW (equation (5.13)) as
νc = ne σW ve ,

(5.23)

where ne and ve are the electron number density and velocity, respectively.
An excited state of the emitter can be viewed as a quantum oscillator with its
frequency ω0 related to its energy E as E = ~ω0 . The excited state makes a transition
into a lower energy state with a natural lifetime τ0 . The ensemble average of the

74
excited state can be represented by a wavefunction ψ(t) as
ψ(t) ∼ eiω0 t e−t/τ0 ,

(5.24)

where the exponential decay term appears because the transition is a random process
(Poisson process) with the characteristic frequency ν0 = 1/τ0 . As a result of this
natural decay, a band of frequencies will be observed instead of a single frequency ω0
in the frequency space of ψ(t):
Z ∞
ψ̃(ω) ∼

eiω0 t e−t/τ0 e−iωt dt =

i(ω0 − ω) − ν0

(5.25)

where ψ̃(ω) is the Fourier transform of ψ(t). Hence, the amplitude of the frequency
component ω is

I(ω) = ψ̃(ω)

(ω − ω0 )2 + ν02

(5.26)

This frequency dispersion is called as natural broadening and its shape is known as
Lorentzian profile. The full width at half maximum (FWHM) of the profile is 2ν0 .
The natural broadening (2ν0 ∼ 108 Hz for strong atomic lines) is negligible compared
to the optical frequency ∼ 1015 Hz.
The electron impact perturbs the excited state and causes the state to decay into
a lower energy state, effectively shortening the lifetime of the excited state. Thus, the
frequency spectrum of the state will further spread beyond the natural broadening,
since the electron impact is essentially the same statistical event (Poisson process)
as the natural decay. The impact frequency νc is typically greater than the natural
decay frequency ν0 . Thus, the intensity profile of the frequency shift ∆ω = ω − ω0
will be primarily broadened by the electron impact:
I(∆ω) ∼

∆ω 2 + νc2

(5.27)

75
The profile width of I(∆ω) scales as
width of I(∆ω) = 2νc ∼ 2ne σW ve ∼ ne (Ca /ve )2 ve = Ca2 ne /ve .

5.4

(5.28)

Density diagnosis using the Stark broadened
profiles

The ultimate profile of the frequency shift of an energy level, I(∆ω), is the convolution of the quasi-static and the electron impact broadenings. The transition from
a broadened upper level to a broadened lower level results in a broadened spectral
line. In a real plasma, the line profile is further modified by secondary effects such as
Debye screening [35] and ion motion [36] [37] [24]. Theories that take into account
these secondary effects have predicted Stark broadening of hydrogen Balmer lines in
good agreement (within 1%–10%) with experiments.
For hydrogen lines, the quasi-static microfield splits degenerate energy levels via
the strong linear Stark effect. The line profile is dominantly broadened as a result of
2/3

this splitting, and so the profile width scales with ne

according to equation (5.22)

with k = 1. It is customary to express the profile width by a parameter α1/2 , the so
called reduced half-width that scales the line shape [23]:
wS = 2.5 × 10−14 α1/2 n2/3
e ,

(5.29)

where wS is the Stark FWHM in nm and ne is the electron density in m−3 . The
half-width α1/2 has been tabulated for many hydrogen lines based on both theoretical
calculations and experimental data for the temperature range 0.5–4 eV and density
range 1020 –1024 m−3 [35] [36] [24].
For lines of non-hydrogenic atoms, the electron impact broadening via the quadratic Stark effect is the dominant broadening mechanism [23]. Thus, to a first approximation, the line profile is Lorentzian and linearly proportional to the electron

76
density according to equations (5.27) and (5.28):
w S ≈ w m ne ,

(5.30)

where wm is the proportionality parameter. For an extensive collection of N II lines,
this property has been substantiated by the experimental work of Mar et al. [38].
Measurements of Stark broadening of non-hydrogenic spectral lines have been tabulated for well-defined and independently measured plasma conditions [25].

Chapter 6
Analysis of spectral line profiles
6.1

Density estimation from spectral line profiles

The profile of a Stark broadened line is a complicated function of plasma density and
temperature as described in chapter 5. However, for electron density measurement,
the detail of the Stark broadened profile is not necessarily required. Instead, the full
width at half maximum (FWHM) of the profile is used because the electron density
is a simple function of the Stark FWHM for many spectral lines. Even though this
simple approximation is applicable for most spectral lines, extraction of the pure
Stark effect from a line profile is generally not straightforward. To clarify difficulties
in reliable density estimation from Stark broadened line profiles, issues pertinent to
our plasma source are discussed [23].

6.1.1

Uncertainty in Stark parameter

The reduced half-width α1/2 for hydrogen lines (equation (5.29)) has a weak dependence on density and on temperature. In addition, it differs slightly between
calculations. The half-width α1/2 of the Hβ line (approximately 0.085) is plotted as a
function of temperature (density) for several representative densities (temperatures)
in figure 6.1a(b) according to the Gigosos-Cardeñoso (GC) theory [36] and according
to the Kepple-Griem (KG) theory [35], showing a slight discrepancy between the two
theories. The uncertainty in α1/2 introduces an error < 20% in density estimation for

78
0.1
Half-width, Α1/2

Half-width, Α1/2

0.1
0.09
0.08
0.07
0.06
0.5

1. 1.5 2. 2.5 3.
Temperature [eV]

3.5

0.09
0.08
0.07
0.06
1020

(a) α1/2 vs. temperature

1021
1022
Density [m-3]

1023

(b) α1/2 vs. density

Figure 6.1: Reduced half-width α1/2 of theoretical Hβ Stark profiles according to
the GC theory (solid lines) and according to the KG theory (dashed lines). (a) α1/2
versus temperature for densities 5 × 1020 , 5 × 1021 , 5 × 1022 m−3 . Lines are drawn
thicker for increasing density. (b) α1/2 versus density for temperatures 1, 2, and 3 eV.
Lines are drawn thicker for increasing temperature.
the hydrogen plasma jets.
Non-hydrogenic spectral lines are used to measure the density of non-hydrogen
(nitrogen, argon, etc) plasma jets. The uncertainties are about 15%–30% for the
Stark width parameters wm (see equation (5.30)) of the spectral lines used in the
present study. For instance, the Stark width parameter for N II 424.178 nm is wm =
152.6 pm/1023 m−3 within 23% error [25].

6.1.2

Competing broadening effects

Besides Stark broadening, a spectral line can be broadened by other effects such as
Doppler effect, instrumental broadening, and Zeeman splitting.
The Doppler broadening is due to the thermal motion of emitters and the profile
is a Gaussian function with FWHM wD ,
wD = 7.7 × 10−5 λ Ti /M ,

(6.1)

where λ is the wavelength of the spectral line, Ti is the emitter temperature (≃ the
ion temperature) in eV, and M is the emitter mass in amu. The ion temperature Ti
appears to be less than 10 eV in the plasma jets. Therefore, 20 pm ≤ wD ≤ 100 pm

79
for hydrogen lines and 6 pm ≤ wD ≤ 30 pm for nitrogen lines in the wavelength range
300–500 nm.
The presumption of low Ti < 10 eV is supported by an impurity line measurement
(C III 229.687 nm) showing broadening ≤ 10 pm corresponding to Ti ≤ 4 eV (see
the inset profile in figure 6.3). It had been assumed that Ti ∼ Te in the previous
work [9, 21], where Te was either measured by a triple Langmuir probe (5–15 eV) or
inferred from spectral line ratios (1–5 eV). For nitrogen plasma jets, the assumption is
supported by a good agreement between the measured line ratios and the calculated
line ratios using the Saha-Boltzmann equation, suggesting that the plasma jets are
at local thermodynamic equilibrium and so Ti ∼ Te . 10 eV. Significant ion heating
(Ti
Te ) via magnetic reconnection has been observed in some other gun-produced
spheromak plasmas [1, chap. 13]. However, magnetic reconnection is not a dominant
mechanism in the plasma jets under consideration here as the magnetic topology is not
undergoing changes. There is mainly a stretching of field lines, but not reconnection
(except for the minimal reconnection associated with merging of spider legs).
The instrumental function describes the spreading of a monochromatic light source
recorded by an spectroscopic system. The instrumental function is mainly determined
by the fiber core size, the slit width, and the entrance optics in the present spectroscopic system (section 3.7). The instrumental function is measured by recording
a known spectral line emitted from a spectrum tube within or close to the spectral range of interest. The measured instrumental broadening is ∼7 ICCD pixels
at FWHM, which corresponds to ∼30 pm according to the pixel resolution equation (3.2). The Zeeman splitting of the energy levels [33, chap. 17] is ∆E ∼ µB B,
where µB = e~/2me is the Bohr magneton. In terms of wavelength, the Zeeman effect
is given by ∆λ[pm] ∼ 10−4 λ2[nm] B[Tesla] < 8 pm for B = 0.3 Tesla, and so the Zeeman
effect is negligible. Spectral lines can also split via the Doppler effect if there exist
relative bulk flow motions within the line of sight volume. However, such a splitting
will be minimal because it is less than 50 pm even for unlikely 30 km/sec relative flow
motions.
If a line width is measured greater than 200 pm, it can simply be taken as the

80
Stark width because the other effects cannot account for a line width > 100 pm, and
so the Stark effect must be the dominant broadening effect in such a case. For Hβ
line, the Stark broadening becomes dominant for ne & 1021 m−3 according to equation (5.29). If the measured line width is comparable to the instrumental broadening
or the expected Doppler broadening, the Stark width can be estimated using one of
the deconvolution procedures described in the next sections 6.2, 6.3, and 6.4.

6.1.3

Plasma inhomogeneity

The plasma jets in our experiments have internal structures (see figure. 3.8b) and so
may have a density gradient along the line of sight. A significant distortion may occur
at the center of a line profile as illustrated in figure 6.2a because the lower density
region concentrates its emission energy in a narrow band while the higher density
region spreads the emission energy over a broad band.
In particular, special care must be applied to the interpretation of the characteristic central dip of the Hβ line. The density gradient effect becomes more appreciable
in the central dip region as the narrow Hβ emission from a low density region can
stand out in that region. The Hβ central dip can appear in high density plasma,
where the ion field discussed in chapter 5 is strong enough to make the line splitting
(via the linear Stark effect) larger than the electron impact broadening. The other
even-numbered hydrogen Balmer lines can also have a central dip [29].
A narrow peak overlaid over a broad envelope similar to figure 6.2a was indeed
observed in some measured Hβ line profiles as shown figure 6.2b. Note that the
theoretical fit according to Stehlé and Hutcheon [24] lacks such a narrow peak because
the theory assumes a homogeneous plasma density. The electron density may be
underestimated if the peak value is used to measure the profile FWHM. For nonhydrogenic lines and hydrogen lines with no central dip, the FWHM is obtained by
fitting the profile to a Lorentzian shape, since this fitting process puts more weight
on the side bands than on the center and so avoids errors associated with a density
gradient along the line of sight. For hydrogen lines showing the central dip, the

1500

Counts

Intensity [a. u.]

20
-15 -10

-5
Pixel [a. u.]

1000
500
484

485
486
487
Wavelength [nm]

(a)

488

(b)

Figure 6.2: (a) Sum of a wide and a narrow Lorentzian profile. Dashed line is the
wide profile with total intensity = 1000 and FWHM = 10, dotted line is the narrow
profile with total intensity = 20 and FWHM = 0.5, and solid line is the sum of
the two profiles. (b) Measured Hβ line profile showing the characteristic central dip
distorted by a peak from a low density contribution (data# oma11.61, perpendicular
line of sight). Measured data points are represented by dots with a fitting curve (thin
solid line). Bold solid line connecting the square points is a theoretical fit according
to Stehlé [24] with ne = 1.05 × 1022 m−3 and Ti = Te = 2 eV. Vertical dashed line
indicates the Hβ rest-frame wavelength 486.133 nm.
FWHM is not well-defined but the same Lorentizan fitting process may be used to
define the FWHM. The electron density obtained via equation (5.29) using such
FWHM is about the same with the electron density estimated by manually finding
the best theoretical fit to the line profile (see sections 6.2 and 6.4).

6.1.4

Self-absorption, continuum background, and blending
with nearby spectral lines

Self-absorption in an optically thick plasma can lead to an overestimation of electron
density. Overlaps between the internal structure of the plasma jet and the objects in
the background are observed in camera images when the camera is operated within
its dynamic range, i.e., not saturated (see figure 3.8b). This suggests that our plasma
is optically thin and so self-absorption will be insignificant.
Blackbody radiation, the background noise of the ICCD detector, Balmer continuum, etc. contribute to the background level of spectral data. Spectral profiles are
corrected by subtracting the measured background level taken from spectral regions

Intensity

CIII 229.687

2000
8 μs

1000

6 μs
4 μs
2 μs

424.

424.2
Wavelength [nm]

424.4

Figure 6.3: Observed line profiles (NII , 424.178 nm) from a nitrogen plasma jet at t =
2, 4, 6, and 8 µs after breakdown (shot#: 6736–3, 7, 11, and 15, respectively). Solid
lines are Lorentzian fits. The estimated densities are 0.1, 1.7, 7.8, and 4.0 × 1022 m−3 ,
respectively. A measured impurity line profilep(CIII , 229.687 nm) is inset at the top for
width comparison. Its Doppler width (wD = wT2 − wI2 ) is ∼0.007 nm, corresponding
to Ti = 2 eV. Vertical dashed line indicates rest-frame wavelength.
close to the spectral line of interest but separated by several times its width. Additionally, every spectral line used in the experiment is isolated from other lines by at
least its line width to avoid mixing.

6.2

Deconvolution of Stark broadening
– simple formula

The Stark broadening and hence the electron density can be extracted from measured
line profiles using a simple and computationally inexpensive procedure. The total
FWHM (wT ) of a spectral line profile is measured by fitting the data to a Lorentzian
profile. The Stark FWHM (wS ) is then obtained by subtracting the Doppler effect
(wD ) and the instrumental function (wI ) from the total width wT according to the
conventional method [39]
wS =

wT2 − wD
− wI2

if wT / wD
+ wI2 > 1.04.

(6.2)

83
This formula assumes that the instrumental function is Gaussian, the Stark broadening is Lorentzian, and so the total broadening is a Voigt profile [40], i.e., the convolution of a Gaussian profile and a Lorentzian profile. The electron density is then calculated using equation (5.29) for hydrogen lines and equation (5.30) for non-hydrogenic
lines. The reliability of this simple method is substantiated by comparing the density
obtained by this method with the density obtained by the theoretical fitting according
to Stehlé [24] (see figure 4.7). The error in the density estimation using Hβ profiles
is less than 15% for ne ≥ 1022 m−3 taking into account the issues discussed in the
previous section. The error gets larger for lower density (25% for ne ∼ 1021 m−3 and
50% for ne ∼ 1020 m−3 ).

6.3

Maximum entropy deconvolution

The simple formula (6.2) becomes inaccurate as the ratio wT / wD
+ wI2 approaches
unity. However, if the instrumental effect can be removed, the ratio wT /wD will
become significantly greater than unity since wD . wI in our experiment, and so
the simple formula will become applicable once again. The instrumental function wI
of the spectroscopic system can be removed computationally since the instrumental
function is invariant and can be measured before taking spectrum. A deconvolution
method based on the maximum entropy principle [41] has been implemented to remove
the instrumental function and thereby achieve an effective resolution higher than the
nominal spectroscopic resolution. The deconvolution method has been applied to
some spectral lines for the Stark density measurement in case of wT ≈ wD
+ wI2 .
The deconvolution method has also been used to separate overlapping Doppler shift
components of a same spectral line and thereby resolve velocity components of plasma
jets in the solar prominence simulation experiment [42, Fig. 5].

6.3.1

Undoing the instrumental broadening
– an inverse problem

Removing the instrumental effect corresponds to the following ill-posed inverse problem.
gi =

Hik fk

(i = 1, . . . , M ),

(6.3)

k=1

where gi is the measured signal at the ith pixel, fk is the true signal to be recovered,
and Hik is the instrumental function which blurs the true signal f = (f1 , . . . , fN )t .
Since the instrumental function preserves the total intensity, the matrix H is normalized as

Hik = 1.

i=1

Hence,
i=1

gi =

M X

Hik fk =

i=1 k=1

k=1

Hik

i=1

fk =

fk .

k=1

It is also convenient to normalize f so that

fk =

k=1

gi = 1.

(6.4)

i=1

The inverse problem (6.3) is ill-posed since the matrix H is not invertible or, even if
it is, the inverting process is numerically unstable. This comes at no surprise because
the instrumental function smooths out fluctuations of the original signal f and so the
inverse process (deconvolution) will amplify the noise in the measured signal g. To
examine the noise-amplifying property of the inverse process, suppose the matrix H
is a square matrix (M = N ) whose maximum and minimum eigenvalues are λM and
λ1 , respectively, and consider a true signal f in the direction of the eigenvector of
λM . Then, the measured signal will be
g = Hf = λM f .

85
Now imagine that the measurement introduced a small error ∆g in the direction of
the eigenvector of λ1 . The corresponding error ∆f will appear in the true signal
recovered by the inverse process as
∆g = H∆f = λ1 ∆f .
Thus, the relative error
λM k∆gk
k∆f k
kf k
λ1 kgk
which implies that a small error in the measured signal can appear amplified by the
factor of |λM /λ1 | in the recovered signal f . This factor is called the condition number
of the matrix H and is a measure of the instability of the inverse process [43, sec. 5.3].

6.3.2

Constrained inverse problem

In earlier work [44], a deconvolution algorithm to solve the inverse problem (6.3)
has been implemented based on the maximum entropy principle [41] and a steepest
descent method devised by Bellan [45] for finding the minimum of a function. The
maximum entropy principle modifies the original problem by imposing the condition
that the entropy S(f ) defined as

S(f ) = −

fk log(fk )

(6.5)

k=1

be maximized. Suppose that fk corresponds to the number of photons hit on the
pixel k. Then, the entropy S(f ) is a measure for the total number of ways that the
photons redistribute themselves among the n pixels but preserving the configuration
{fk }. Thus, the configuration {fk } which maximizes the entropy S(f ) under the
M +1 constraints (6.3) and (6.4) is the least committal solution among all the possible
solutions satisfying the constraints.
The maximum entropy solution f can be found using the method of Lagrange

86
multipliers. Consider the Lagrangian

L(f ; λ) = −

fk log(fk ) − λ0

k=1

fk − 1

λi

i=1

k=1

Hik fk − gi

(6.6)

k=1

Maximization of the Lagrangian L(f ; λ) with respect to each fk gives the solution f
as a function of the Lagrange multipliers λ’s:

fk (λ) = exp −

λi Hik

/Q(λ),

(6.7)

i=1

Q(λ) =

exp −

k=1

λi Hik

(6.8)

i=1

where the partition function Q(λ) comes about from the normalization condition of
f (equation (6.4)). The final step is to find the set of the Lagrange multipliers λ’s
which satisfy the system of equations (6.3):

gi −

Hik fk (λ) = 0

(i = 1, . . . , M ).

(6.9)

k=1

Finding the roots of the above equations is equivalent to minimizing the potential
function defined as
Z(λ) = log Q(λ) +

λi gi

(6.10)

i=1

because ∂Z/∂λi = gi −

k=1 Hik fk (λ).

In case of M = N (i.e., the matrix Hik is a square matrix), it can be shown that
the potential function Z(λ) is strictly concave [46] (see appendix F for the proof), and
thus, the minimization process will always lead to the single global minimum. The
gradient of Z(λ) becomes progressively weaker as one approaches the global minimum
since Z(λ) is a smooth function of λ. A steepest descent algorithm utilizing this strict
concavity was devised by Bellan [45] to find the minimum. The strategy is to take
iterative steps in variable size along the direction of the gradient. A large step ε∇Z
is taken initially from a starting point λ0 and then the potential at the new point

87
λ0 = λ0 − ε∇Z is compared with the starting value Z0 = Z(λ0 ). The parameter ε is
a small positive number which determines the step size. If Z(λ0 ) > Z0 , the step size
ε is decreased and Z is re-evaluated until it becomes smaller than Z0 . If Z(λ0 ) < Z0 ,
another search starts from the new point in the new direction ∇Z(λ0 ). To avoid
taking too small steps, a step size ε slightly greater than the last step size is taken at
the start of each search. This steepest descent algorithm is illustrated in figure 6.4.
The variance Ω at each step is defined as
1 X
Ω=
gi −
Hik fk
N i

1 X ∂Z
N i ∂λi

k∇Zk2

(6.11)

and is a measure of goodness of the signal recovery. The iteration stops if Ω becomes
smaller than an expected variance Ω0 or the total number of steps taken exceeds a
given maximum number of steps. The parameter Ω has a practical importance [47].
Ω0 indicates the recovered signal (f ) is a poor fit to the measurement (g). On
the other hand, Ω
Ω0 indicates the recovered signal is an over fit and contains
structures arising from the measurement error. The variance of a measured background spectrum g0i = c + ei , where c is the constant background level and ei is the
noise, may be taken as the expected variance Ω0 .

1 X 2
Ω0 =
e.
N i=1 i

6.3.3

(6.12)

Numerical test

The maximum entropy deconvolution scheme was applied to a set of synthetic spectra
with noise broadened by a given instrumental function to test its reliability. It was
seen that the deconvolution scheme is robust against measurement noise. Figure 6.5a
shows that the deconvolution completely recovers the double peaks in the original
spectrum from the noisy data broadened by the instrumental function. In practice,
the instrumental function is measured by taking a reference spectrum and so contains
measurement errors. The deconvolution scheme was also tested against uncertainty

Figure 6.4: Example of the steepest decent iterations. Green dot indicates a successful step toward the minimum and red dot indicates too large a step size. The given
example is a somewhat worst-case scenario where the initial step size is too large.
in the instrumental function. No noticeable change in the recovered spectrum was observed until the signal-to-noise ratio of the instrumental function was degraded below
∼20 (figure 6.5b). The signal-to-noise ratio of the measured instrument function for
our spectroscopic system is very high (& 100) and so the corresponding deconvolution
error would be small.
The two-dimensional version of the maximum entropy deconvolution scheme has
also been implemented to enhance the Imacon camera images (figure 6.6). In the
thesis work of Romero-Talamás [48], a similar deconvolution scheme based on the
work of Wilczek and Drapatz [49] was used to enhance the resolution of the blurred
SSPX plasma images.

6.4

Theoretical fitting of Stark profile

The profiles of hydrogen lines emitted from hydrogen plasma of large density (ne &
1022 m−3 ) are significantly different from the Lorentzian function. The Hβ line profile
characterized by the central dip as shown in figure 4.7 is a good example. For such
line profiles, it is not straightforward to define the profile width. The width may be
defined as the FWHM of the Lorentzian function best fitting the profile or the FWHM

89
1.5

Intensity

100

1.0

0.5

0.0

100

150

Pixel

(a) Maximum entropy deconvolution of a synthetic spectrum

Intensity

1.5

1.0

0.5

0.0

100

150

Pixel

(b) with degraded instrumental function

Figure 6.5: (a) Maximum entropy deconvolution of a synthetic spectrum. The original synthetic spectrum (red line) contains two peaks that are merged together in
the measured spectrum (connected dots) by the instrumental broadening. Random
noise has been added to the measured spectrum so that the signal-to-noise ratio is
20 at the peak. Thick solid line represents the recovered spectrum by the maximum
entropy deconvolution. (b) Maximum entropy deconvolution with degraded instrumental function (red solid line). The original instrument function (dashed line) is
overlaid with the degraded one. The signal-to-noise ratio of the degraded instrumental function is 20 at the peak. No noticeable change except small details in the peaks
is seen in the recovered spectrum.

(a) Original Imacon image

(b) Reconstructed image

Figure 6.6: Example of the image reconstruction by the maximum entropy deconvolution method.
of the raw profile. However, the validity of applying such definitions of profile width
to equation (5.29) for density estimation needs to be checked.
Instead of introducing an arbitrary definition of profile width, the density of hydrogen plasma can be estimated by finding the best theoretical fit to observed line
profiles. Stehlé and Hutcheon [24] calculated Stark broadened hydrogen line profiles for a pure hydrogen plasma and tabulated the result according to line species,
density, and temperature. The Stark broadening tabulation spans a density range
between 1016 and 1025 m−3 (20 densities) and a temperature range between 2500 and
1.3 × 106 K (10 temperatures). Local thermodynamic equilibrium is assumed in the
calculation and so Ti = Te .
The best theoretical fit to an observed line profile can be found as follows. A rough
density estimate is made using the FWHM of the raw profile in equation (5.29). The
plasma temperature is chosen between 1 and 10 eV (section 6.1.2). The choice of
temperature is not critical for Hβ line profiles because the Stark broadening of the
Hβ line is nearly independent of temperature (chapter 5). The computed line profile
corresponding to the trial density and temperature is retrieved by a computer code

91
supplied in the distribution of the Stark broadening tabulation. The computer code
returns interpolated data if necessary. The retrieved profile is then convolved with
the instrumental function to simulate the same intrumental broadening present in the
observed line profile. The trial density is varied until the computed profile fits well to
the observed profile. The temperature may be further adjusted to obtained a better
fit.
The resulting best-fit density is compared to the density estimated by the simple
method described in section 6.2, where the total width is taken from a Lorentzian fit
to the observed profile. The difference between the two estimates of density is less
than 10%, ensuring the validity of the simple density estimation method.

Chapter 7
Summary and conclusion
The Caltech spheromak plasma gun generates highly collimated rapidly evolving
plasma structures. The planar configuration of the gun electrodes has permitted
direct observation of the entire plasma dynamics and four distinct evolution stages
have been identified in previous work by others. The four evolution stages are (1)
eight well-collimated arched flux tubes reminiscent of spider legs form across the gun
electrodes, (2) the spider leg flux tubes coalesce into a single central plasma jet, (3)
the plasma jet expands axially and becomes collimated, and (4) the elongated plasma
jet may undergo MHD instabilities. The present work has reported new experimental
observations of the first and the third stages showing that the observed collimation
is a result of the MHD pumping of plasma particles into flux tubes.
In chapter 3, the construction of a high-resolution spectroscopic system is described. The spectroscopic system has 12 channels with flexible lines of sight and can
measure emission spectra radiated from multiple locations of the plasma simultaneously. Jet velocities are measured from Doppler shifts of spectral lines and electron
densities are measured from Stark broadened spectral lines. The Stark broadening
mechanism is summarized in chapter 5. A simple method to extract densities from
Stark broadened lines is described in chapter 6. The method provides reliable plasma
density diagnostics for densities & 1021 m−3 .
In chapter 4, the experimental results are presented, showing in detail that the
observed densities and velocities are consistent with the MHD pumping model. The
measured jet velocities (∼40 km/sec) are the Alfvénic velocity corresponding to the

93
measured density (∼1022 m−3 ) and azimuthal magnetic field strength (∼1 T) as predicted by the MHD pumping model. It is demonstrated that the flow slowing down
observed at the forefront of the jet leads to the observed mass accumulation and
subsequent jet collimation according to the MHD pumping model. In addition, it is
clearly shown that the measured density amplification (from ∼1017 up to ∼1022 m−3 )
is entirely due to the MHD pumping and cannot be accounted by the gas puffing.
We conclude that the MHD pumping and collimation model has provided a consistent explanation for the observed collimation and density amplification of the gunproduced plasmas. Thus, the MHD pumping model provides valuable insight into the
ubiquitous collimation of laboratory, solar, and astrophysical plasmas.

Appendix A
Leak analysis
A.1

Equilibrium pressure; pumping, leak, and wall
desorption

The change in number of particles in the vacuum chamber with volume V is
∆N = V ∆n = −nRp ∆t + nA Rl ∆t,

(A.1)

where n is the number density of air inside the chamber, nA is the number density of
the air outside the chamber, Rp is the pumping rate (volume per unit time), and Rl
is the effective leak (including wall desorption) rate.
dn
= −nRp + nA Rl
dt
dp
= −pRp + pA Rl (expressed in terms of pressures)
dt

=⇒ p = (p0 − pA Rl /Rp ) exp(−tRp /V ) + pA Rl /Rp .

(A.2)
(A.3)
(A.4)

Thus, the pressure (p) inside the chamber exponentially approaches the equilibrium
pressure pe given by
pe = pA Rl /Rp .

(A.5)

95
If there is no pumping,

dp
dt

= pA Rl /V ; linear pressure rise

dp
=⇒ Rl =
V /PA
dt no pump

dp
Rp = Rl pA /pe =
V /Pe .
dt no pump

(A.6)

no pump

(A.7)
(A.8)

Hence, both the effective leak Rl and the pumping speed Rp can be estimated from
the linear pressure rise dp
by shutting off pumping for a moment.
dt no pump

A.2

Example

When the gate valve of the cryopump was closed, the chamber pressure increased
from 1.4 × 10−7 torr to 9.3 × 10−6 torr in two minutes.

dp
dt

= (9.3 × 10−6 − 1.4 × 10−7 )/120 = 7.6 × 10−8 torr/sec.

no pump

Since pA = 760 torr, pe = 1.4 × 10−7 torr, and V = 2.4 m3 ,
Rl =
Rp =

dp
dt

V /PA = 2.4 × 10−10 m3 /sec = 2.4 × 10−4 cc/sec,

no pump

V /Pe = 1.3 m3 /sec = 1300 liters/sec.

no pump

The cryopump is working more or less with the specified pumping speed ∼1000
liters/sec. The effective leak (∼10−4 cc/sec) is reasonable considering tens of Orings (∼10−6 cc/sec per O-ring) installed on the chamber and desorption from the
wall.
In the event of a leak much greater than the normal effective leak (∼10−4 cc/sec),
determining the size of leak aperature may help locate the leak.
Rl ≈ vth × Dleak

(A.9)

96
where vth = 340 m/sec is the sound speed of air and Dleak is the leak aperature
diameter. For example, consider the case when the equilibrium chamber pressure is
as high as 1.0 × 10−5 due to an unknown leak. The leak rate is
Rl = Rp pe /pA = 1300 · 1.0 × 10−5 /760 liters/sec ≈ 0.017 cc/sec,
and the corresponding leak aperature size is
Dleak =

1.7 × 10−8 /340 m ≈ 7 µm.

Appendix B
Measurement of gas cloud profile
B.1

Fast ionization gauge (FIG)

A home-built fast ionization gauge (FIG) has been constructed by P. M. Bellan to
measure the output of the fast gas valves [21]. The FIG measurement detail is reproduced here for completeness by the kind courtesy of S. You and a modified gas cloud
model based on his original model is presented.
The FIG is mounted on a right-angle arm as shown in figure 2.2 so that both
its axial and radial locations can be adjusted. The FIG is operated in the same
way as standard hot-cathode ionization gauges. Thermionic electrons emitted from
a heated filament (cathode; negatively biased) are accelerated toward a positively
biased electrode (anode). The electrons collide with gas molecules in the background
and ionize them. The distance between the cathode and the anode is designed to
be smaller than the electron’s ionization mean free path to prevent multiple ionizing
collisions. The ionized atoms are attracted to a grounded electrode called collector.
The total number of ionizing collisions per unit time is
Ṅi = nσi,g lṄe = (P/kT )σi,g lṄe ,

(B.1)

where n is the gas number density, σi,g is the ionization cross-section for gas “g”, l is
the distance between the cathode and anode, Ṅe is the number of emitted electrons
per unit time. In terms of the collector current Ic = Ṅi e and the emission current

98
Ie = Ṅe e,
Ic = P Sg Ie =⇒ P = Ic /(Sg Ie ) = Ic /(Rg SN 2 Ie ),

(B.2)

where Sg = σi,g l/kT is called as gauge sensitivity factor [torr−1 ]. Relative gauge
sensitivity factors (Rg ) commonly calibrated for nitrogen are available from the public
vacuum literature [50]. The FIG’s sensitivity factors are obtained by comparing
with a standard Bayard-Alpert type gauge. However, the pressure calculated by
equation (B.2) has only a relative meaning since the pressure near the gas orifices is
presumably of the order of mtorr and Sg becomes nonlinear for pressure > 1 mtorr.
FIG measurements are shown in figure B.1. Figure B.1b shows that the particle
flux from the fast gas valve lasts for ∼5 msec, much longer than our plasma lifetime
of ∼20 µsec.

B.2

Gas profile modeling

Consider the gas density profile fi (r, z) produced by the orifice i (i = 1, 2, . . . , 16
is the index number of each orifice). The radial velocity distribution is presumably
Gaussian at the orifice since gas particles will be thermalized in radial direction as
they flow along the long gas line. The particle source at the orifice can be considered
point-like at distance ( r2 + z 2 ) much greater than the orifice diameter. In this limit,
the Gaussian radial velocity distribution produces a Gaussian radial density profile
because
Z (r+∆r)/t
fi ∆r ∝

exp(−v 2 /vT2 )dv

r/t

∝ exp(−v 2 /vT2 )∆r/t
∝ exp(−r2 /(tvT )2 )∆r/t (∗ v = r/t),

(B.3)

where fi ∆r is the number of particles in the ring (r, r + ∆r) at time t and vT is the
radial thermal velocity of the gas.
The density drops quadratically in the axial direction assuming the axial expansion

99
velocity is constant (an exponential drop was assumed in the original model of S. You).
Hence, the gas blooms out radially at the speed of sound (vT ) and expands axially at
some Mach number, forming a cylindrical cone of gas cloud. The width of the radial
profile gives the Mach number since the Mach number (M ) is essentially the ratio of
the axial velocity (vz ) to the sonic radial velocity (vT ):
z p
log 2 = tvz (vT /vz ) log 2 =
log 2,
z p
M=
log 2 = tan α log 2,
rHM

rHM = tvT

(B.4)
(B.5)

where rHM is the radius of half maximum density at a given axial position z and α
is half the cone angle measured at half maximum density. The density profile can be
written in terms of Mach number M
r2
r2
fi (r, z) = f0i 2 exp −
= f0i 2 exp −
(tvz /M )2
(z/M )2

(B.6)

The total particle density fT at a point (r, φ, z) is the sum of the 16 gas clouds

fT (r, φ, z) =

fi (ri , z),

(B.7)

i=1

where ri is the radial distance from the orifice i.
The FIG measurements fit well with the gas cloud model calculated by equation (B.7) as shown in figure B.1.

100

3.5
Pressure [arb. unit]

Pressure [arb.unit]

2.5
2.0
1.5
1.0
0.5
0.0

3.0
2.5
2.0
1.5
1.0
0.5
0.0

-30

-20

-10

r [cm]

(a) Radial pressure profile

10
15
Time [msec]

(b) Pressure trace

Figure B.1: Fast ion gauge (FIG) pressure measurement (data courtesy of S. You).
Nitrogen was injected by the four fast gas valves. The FIG’s axial location was z = 18
cm. (a) Radial pressure profile. The data points are taken at 7 msec (after the fast
gas valves triggered) and the error bars indicate signal variations in ±0.5 msec. The
solid curve is the calculated profile according to equation (B.7). The shape of the
radial profile becomes stationary after about 5 msec and starts to fit with calculated
profiles. (b) Pressure trace at r = 0 cm. The pressure peaks at about 4 msec. The
gas flow speed is (L + 0.18m)/4msec = 300 m/sec ≈ the sound speed of nitrogen,
where L ≈ 1 m is the length of gas lines (distance from fast gas valves to orifices).

101

Appendix C
Linear dispersion of Czerny-Turner
spectrometer
A Czerny-Turner spectrometer is shown schematically in figure C.1. Knowledge of the
linear dispersion of the spectrometer at the exit plane is desired to calibrate a spectral
window for which no reference spectral lamps are available and also to circumvent
repetitive use of spectral lamps for each spectral window. The grating equation is
d (sin α + sin β) = mλ,

(C.1)

where α is the angle of incidence, β is the angle of diffraction, d is the groove spacing,
m is the diffraction order, and λ is the wavelength of light. The sign of an angle
is positive (negative) if drawn counterclockwise (clockwise) in figure C.1. Consider
the angular dispersion, i.e., variation of λ as a function of β for a fixed α. The
angular dispersion at the focusing mirror (M 2) can be obtained by differentiating
equation (C.1) with respect to β:
d cos β = m

∂λ
∂β

=⇒

∂λ
d cos β
∂β

(C.2)

Thus, the linear dispersion at the exit focal plane is
1 ∂λ
d cos β
∂λ
∂x
L ∂β
Lm
where L is the focal length of the focusing mirror.

(C.3)

102
A rotation of the grating by an angle θ determines the incidence/diffraction angles
and selects a wavelength λ, which will be focused at the center of the exit focal plane.
The selected wavelength λ is related to θ according to the grating equation (C.1):
α = α0 + θ,

(C.4)

β = −α0 + θ,

(C.5)

d (sin(α0 + θ) + sin(−α0 + θ)) = 2d sin θ cos α0 = mλ
=⇒

sin θ =

mλ
2d cos α0

(C.6)

The linear dispersion at the selected wavelength λ can be expressed as a function
of λ by inserting equations (C.4)–(C.6) into equation (C.3):
∂λ
cos β =
cos(−α0 + θ) =
(cos α0 cos θ + sin α0 sin θ)
∂x
Lm
Lm
Lm
(2d cos α0 )2 − (mλ)2
mλ
cos α0
+ sin α0
Lm
2d cos α0
2d cos α0
s
2
1 
2d cos α0
− λ2 + λ tan α0 )
(C.7)
2L
The pixel resolution R(λ) of a CCD detector placed at the exit focal plane is R(λ) =
∆x ∂λ/∂x, where ∆x is the detector pixel size.
The linear dispersion will be nearly constant across the detector plane (i.e., the exit
focal plane) if the detector size is small compared to the focal length of the spectrometer since the bandpass of the detector ∼ (detector size)×(∂λ/∂x) ∝ (detector size)/L.
Consider a wavelength λ0 = λ + ∆λ close to the selected wavelength λ. The wavelength λ0 will have a different diffraction angle β 0 = β + δ than β and subsequently
be focused at a point off the center of the exit focal plane. Because ∂λ/∂x ∝ cos β
(equation (C.3)), the relative error ∆R/R of approximating (∂λ/∂x)λ=λ0 by expression (C.7) is
∆R
cos β 0 − cos β
−δ sin β
= −δ tan β.
cos β
cos β

(C.8)

103
M2

M1
D

exit

entrance

Figure C.1: Czerny-Turner configuration. G is the grating with groove spacing d.
M 1 is the collimating mirror and M 2 is the focusing mirror with focal length L. α is
the angle of incidence, β is the angle of diffraction, α0 is half the angle between the
incident and the diffracted light, and θ is the rotation angle of the grating.
The ∆λ is related to δ as
mλ0
− sin α = sin(β + δ) ≈ sin β + δ cos β
sin β =
d 0
mλ0 mλ
m∆λ
mλ
− sin α − sin β / cos β =
/ cos β =
=⇒ δ ≈
d cos β

Thus, the relative error is
∆R
m∆λ
tan β/ cos β . O

∆λ

(C.9)

For the spectroscopic system used in the present thesis work, ∆λ across the detector
plane is ∼2 nm and d = 1/3600 mm ≈ 280 nm. Thus, the relative variation of the
linear dispersion across the detector plane is less than 1%.

104

Appendix D
Large vacuum viewport
Two large custom-made viewports (800 diameter window welded on a 10” flange) and a
commercial viewport (600 view) installed on the east side of the vacuum chamber have
been utilized for taking plasma pictures and spectra. More vacuum viewports with
800 or greater view were desired for the cameras and the spectroscopic system to gain
a better access to the plasma. An economic viewport design has been devised since
commercial vacuum viewports are available only up to ∼6” view and a custom-made
viewport with larger view size would be expensive prohibiting multiple installations
of such viewports. The viewport design (figure D.1 consists of a modified 1000 Conflat
flange, a viewport glass (BorofloatTM , 8.500 diameter, 11 mm thick), an O-ring to
form a vacuum seal (Viton, AS568A no. 172), and several clamps to hold the glass in
place. The clamps are to give only an initial vacuum seal. Once the chamber starts
being pumped out, the pressure difference will build up across the glass window, press
the O-ring against the flange, and form a vacuum seal. Several viewports have been
constructed according to the design and successfully installed on the chamber.
The window material and thickness of the home-built viewport were chosen to
meet the optical transmission and mechanical strength required for the experiment
as well as the budget. Borosilicate (BorofloatTM ) was chosen as the window material
for its excellent optical transmission (50% at 300 nm and 90% at 350–2000 nm) and
resistance to thermal shock (due to a very low thermal expansion about one third of
ordinary glass). For better UV transmission, the window can be replaced with more
expensive fused silica or quartz window (80% at 180 nm and 90% at 300–2000 nm).

105
clamp
glass

O-ring
modified
10″ Conflat

Figure D.1: Coss-sectional view of the large viewport (800 view diameter). An 8.500
diameter 11 mm thick borosilicate glass is installed on a modified 1000 Conflat flange.
The O-ring forms a vacuum seal as the glass is pressed against the flange by the
atmosphere.
The window needs to be thick enough to support the stress created by a large pressure
difference across the window. A circular window under a uniform loading (pressure)
W experiences the maximum stress σ at the center given by [51, p.175]:
σ=

3(3 + ν)
W r02 /t2 ≈ W r02 /t2 ,

(D.1)

where r0 and t are the radius and thickness of the window, respectively, and ν ∼ 0.2
is the Poisson’s ratio of borosilicate (ν ∼ 0.2–0.4 for most materials). The maximum
tensile strength (critical stress) of borosilicate is ∼25 MPa, so the minimum thickness
of the window should be
σ ≈ W r02 /t2 < 25 MPa
=⇒ t > r0 0.1/25 ≈ 6 mm,

(D.2)

where the atmospheric pressure W ∼ 100 kPa = 0.1 MPa and r0 = 400 ≈ 100 mm.
This minimum thickness is consistent with an empirical practice for determining load
resistance of glass in buildings (ASTM Standard E 1300-04) [52]. The standard
lists the critical load under which the glass of a given size and thickness will fail.
For example, the critical loads are 7.2 and 2.2 kPa for annealed monolithic glasses
of dimensions 3000 × 4000 × 8 mm and 6000 × 9600 × 10 mm, respectively. Since the

106
critical thickness t ∝

W × (window area) for a given maximum tensile strength,

the minimum thickness for the 800 borofloat window can be extrapolated as

W × πr02
W∗ × (window area)∗
100×π42
 8×
7.2·(30×40)
≈ 6 mm,
100×π42
 10 ×
2.2·(60×96)

t > t∗

(D.3)

where the subscript ∗ denotes the values listed in the ASTM standard. The thickness
of the borofloat window is chosen to be 11 mm for a good margin of safety.

107

Appendix E
Selection rules and commutation
relations
E.1

Particle in a perturbed central potential

Consider a particle in a central potential V (r). The Hamiltonian of the system is H0 =
p2 /2m+V (r), where p, m are the momentum and mass of the particle. The eigenstate
of the system is characterized by the principal quantum number n, azimuthal quantum
number l, and magnetic quantum number m, and is commonly denoted as |n, l, mi.
The quantum number l and m are related to the angular momentum operators as
Lz |n, l, mi = m~|n, l, mi,

(E.1)

L2 |n, l, mi = (L2x + L2y + L2z )|n, l, mi = l(l + 1)~2 |n, l, mi.

(E.2)

The angular momentum operator Li (i = x, y, z) is defined as
Li = ijk Rj Pk or
L = R × P,

(E.3)

where Rj (Rx,y,z ≡ X, Y, Z) is the position operator and Pj is the momentum operator.
Suppose a perturbing potential H1 = F z (F : constant) is introduced to the
original system H0 . The eigenstates |n, l, mi’s mutually orthogonal in the original
system may become coupled with each other by the perturbation, i.e., hn, l, m|(H0 +

108
H1 )|n0 , l0 , m0 i = hn, l, m|F Z|n0 , l0 , m0 i may become non-zero for (n, l, m) 6= (n0 , l0 , m0 ).
The existence of a coupling between a pair of eigenstates can be determined by simply comparing the quantum numbers (n, l, m) and (n0 , l0 , m0 ) instead of evaluating
hn, l, m|Z|n0 , l0 , m0 i. Such rules called as selection rules can be found by considering
commutation relations between the position operators and the angular momentum
operators. The basic commutation relations among R, P, and L operators are
[Pi , Rj ] = −i~δij ,

(E.4)

[Li , Rj ] = [ikl Rk Pl , Rj ] = ikl Rk [Pl , Rj ] = −i~ikl Rk δlj
= i~ijk Rk ,

(E.5)

[Li , Lj ] = i~ijk Lk ,

(E.6)

[L2 , Li ] = 0,

(E.7)

L · R = Li Ri = ijk Rj Pk Ri = ijk Rj ([Pk , Ri ] + Ri Pk )
= ijk Rj (−i~δki + Ri Pk ) = ijk Rj Ri Pk = 0.

(E.8)

The following selection rules are derived using these commutation relations in the
following sections.
hn, l, m|Z|n0 , l0 , m0 i 6= 0 only if (1) m = m0 and (2) l = l0 = 0 or l = l0 ± 1.

E.2

Selection rule for the quantum number m

The selection rule for the magnetic quantum number m is easily derived from the
commutation relation (equation (E.5)), [Lz , Z] = 0.
hn, l, m|[Lz , Z]|n0 , l0 , m0 i = (m − m0 )~hn, l, m|Z|n0 , l0 , m0 i = 0.
Thus, hn, l, m|Z|n0 , l0 , m0 i = 0 unless m0 = m.

(E.9)

109

E.3

Selection rule for the quantum number l

The selection rule for the angular quantum number l can be found by considering the
commutation relation between R and L2 operators.
[L2 , Ri ] = [Lj Lj , Ri ] = Lj [Lj , Ri ] + [Lj , Ri ]Lj = i~jik (Lj Rk + Rk Lj )

(E.10)

[L2 , R] = −i~(L × R − R × L)

(E.11)

(in vector form).

Applying L2 on the commutation relation above leads to
[L2 , [L2 , R]] = −i~[L2 , L × R − R × L] = −i~(L × [L2 , R] − [L2 , R] × L)
= −~2 {L × (L × R − R × L) − (L × R − R × L) × L}
= −~2 {L × (L × R) + (R × L) × L − L × (R × L) − (L × R) × L}
= −~2 {L(L · R) − (L · L)R + (R · L)L − R(L · L)}
+ ~2 {L × (R × L) + (L × R) × L}
= ~2 L2 R + RL2 + L × (R × L) + (L × R) × L

(using (E.8)).

The last two terms can be combined to give
L × (R × L) + (L × R) × L = L2 R + RL2

(E.12)

=⇒ [L2 , [L2 , R]] = 2~2 (L2 R + RL2 ).

(E.13)

Equation (E.12) is derived as follows:
L × (R × L) = Li RLi + (L · R)L = Li RLi
(L × R) × L = Li RLi + L(R · L) = Li RLi
Li Rj Li = Li ([Rj , Li ] + Li Rj ]) = Li (i~jik Rk + Li Rj ) = L2 Rj + i~jik Li Rk
Li Rj Li = ([Li , Rj ] + Rj Li ])Li = (i~ijk Rk + Rj Li )Li = Rj L2 + i~ijk Rk Li
=⇒ 2Li Rj Li = L2 Rj + Rj L2 + i~(jik Li Rk + ijk Rk Li ) = L2 Rj + Rj L2 ,

(E.14)

110
where the term (jik Li Rk + ijk Rk Li ) vanishes because
jik Li Rk + ijk Rk Li = (L × R)j + (R × L)j
= (L × R + R × L)j = ((R × P) × R + R × (R × P))j
= (Rk PRk − R(P · R) + R(R · P) − (R · R)P)j
= Rk Pj Rk − Rj Pk Rk + Rj Rk Pk − Rk Rk Pj
= Rk (Pj Rk − Rk Pj ) − Rj (Pk Rk − Rk Pk ) = Rk [Pj , Rk ] − Rj [Pk , Rk ]
= −Rk i~δjk + Rj i~ = −Rj i~ + Rj i~ = 0.
The selection rule for l is obtained by applying [L2 , [L2 , Z]] on the eigenstates:
hn, l, m|[L2 , [L2 , Z]]|n0 , l0 , m0 i = (l(l + 1) − l0 (l0 + 1))~2 hn, l, m|[L2 , Z]|n0 , l0 , m0 i
= (l(l + 1) − l0 (l0 + 1))2 ~4 hn, l, m|Z|n0 , l0 , m0 i.

(E.15)

(E.16)

Subtracting these two equivalent expressions leads to the selection rule for l:

(l(l + 1) − l0 (l0 + 1))2 − 2(l(l + 1) + l0 (l0 + 1)) hn, l, m|Z|n0 , l0 , m0 i = 0
(l + l0 + 2)(l + l0 )(l − l0 + 1)(l − l0 − 1)hn, l, m|Z|n0 , l0 , m0 i = 0.

Therefore, hn, l, m|Z|n0 , l0 , m0 i = 0 unless l = l0 = 0 or l = l0 ± 1.

(E.17)

111

Appendix F
Concavity of the maximum
entropy potential function Z
The potential function Z defined in section 6.3 is strictly concave. The original proof
is given by Alhassid et al. [46] and a succinct version of the proof consistent with the
notations in the present manuscript is given here for completeness. Equations (6.7),
(6.8), and (6.10) are rewritten in a form convenient for the proof,
fk = wk /Q,
Q=
wk ,

(F.1)
(F.2)

Z = log Q +

λk gk ,

(F.3)

P
and H is a square matrix (i.e., M = N ).
where wk = exp − N
ik
i=1
The potential Z is strictly concave if and only if the Hessian matrix ∂ 2 Z/∂λi ∂λj
is positive definite. Consider the following equations to find an expression for the
Hessian matrix:

112

∂wk
= −Hik wk
∂λi
∂Q
=−
Hik wk
∂λi

(F.4)
(F.5)

∂Z
1 ∂Q
+ gi = −
Hik fk + gi
∂λi
Q ∂λi
Pk
Hjm wm
−Hjk wk
∂fk
+ wk m 2
= −Hjk fk + fk
Hjm fm
∂λj
= −fk Hjk −
Hjm fm

(F.6)

(F.7)

The Hessian matrix is obtained from the last two equations:
∂fk X
∂ 2Z
Hik
Hik fk
=−
∂λi ∂λj
∂λ

Hjk −

Hjm fm .

(F.8)

The expression above can be simplified by defining ĝj ≡

m Hjm fm

and Ĥjk ≡

Hjk − ĝj :
∂ 2Z
Hik fk (Hjk − ĝj )
∂λi ∂λj
ĝi fk (Hjk − ĝj )
(Hik − ĝi )fk (Hjk − ĝj ) +

Ĥik fk Ĥjk ,

(F.9)

where

k ĝi fk (Hjk − ĝj ) = ĝi (ĝj −

k fk ĝj ) = ĝi (ĝj − ĝj ) = 0 is used at the last step.

Now, the positive definiteness of the Hessian matrix is easily seen as
i,j

i2
X h
∂ 2Z
xj =
(xi Ĥik )fk (xj Ĥjk ) =
fk (xT Ĥ)k > 0
∂λi ∂λj
i,j,k

for any vector x 6= 0.

(F.10)

113

Bibliography
[1] P. M. Bellan. Spheromaks. Imperial College Press, 2000.
[2] H. Bruhns. Recent compact toroid research. Plasma Phys. Control. Fusion,
28:1389–1400, 1986.
[3] P. M. Bellan and J. F. Hansen. Laboratory simulations of solar prominence
eruptions. Phys. Plasmas, 5(5):1991–2000, 1998.
[4] T. R. Jarboe. Review of spheromak research. Plasma Phys. Control. Fusion,
36:945–990, 1994.
[5] T. R. Jarboe. The spheromak confinement device. Phys. Plasmas, 12:058103,
2005.
[6] H. Alfvén, L. Lindberg, and P. Mitlid. Experiments with plasma rings. J. Nucl.
Energy, Part C, 1:116–120, 1960.
[7] S. C. Hsu. Studies of magnetic helicity injection and dynamo fluctuations in a
spheromak formed by a comparably sized planar coaxial gun. Proposal for the
Fusion Energy Postdoctoral Research Program, 2000.
[8] S. C. Hsu and P. M. Bellan. A laboratory plasma experiment for studying
magnetic dynamics of accretion disks and jets. Mon. Not. Royal Astron. Soc.,
334(2):257–261, 2002.
[9] S. C. Hsu and P. M. Bellan. Experimental identification of the kink instability
as a poloidal flux amplification mechanism for coaxial gun spheromak formation.
Phys. Rev. Lett., 90(21):215002, 2003.

114
[10] P. M. Bellan. Fundamentals of Plasma Physics. Cambridge University Press,
2006.
[11] J. F. Hansen. Laboratory simulations of solar prominences. PhD thesis, California Institute of Technology, 2001.
[12] W. H. T. Vlemmings, P. J. Diamond, and H. Imai. A magnetically collimated
jet from an evolved star. Nature, 440:58–60, 2006.
[13] P. M. Bellan. Why current-carrying magnetic flux tubes gobble up plasma and
become thin as a result. Phys. Plasmas, 10(5):1999–2008, 2003.
[14] L. B. Loeb and J. M. Meek. The Mechanism of the Electric Spark. Stanford
University Press, 1941.
[15] N. Bowler and Y. Huang. Electrical conductivity measurement of metal plates
using broadband eddy-current and four-point methods. Meas. Sci. Technol.,
16(11):2913–2200, 2005.
[16] P. Clausing. The flow of highly rarefied gases through tubes of arbitrary length.
J. Vac. Sci. Technol., 8(5):636–646, 1971.
[17] P. M. Bellan. A simple system for locating ground loops. Rev. Sci. Instrum.,
78:065104, 2007.
[18] C. A. Romero-Talamás, P. M. Bellan, and S. C. Hsu. Multielement magnetic
probe using commercial chip inductors. Rev. Sci. Instrum., 75(8):2664–2667,
2004.
[19] J. Yee and P. M. Bellan. Taylor relaxation and λ decay of unbounded, freely
expanding spheromaks. Phys. Plasmas, 7(9):3625–3640, 2000.
[20] C. Yang. Observations of spheromaks using an optical multichannel analyzer.
Caltech Summer Undergraduate Research Fellowship, unpublished, 2000.

115
[21] S. You, G. S. Yun, and P. M. Bellan. Dynamic and stagnating plasma flow
leading to magnetic-flux-tube collimation. Phys. Rev. Lett., 95(4):045002, 2005.
[22] G. S. Yun, S. You, and P. M. Bellan. Large density amplification measured on
jets ejected from a magnetized plasma gun. Nuclear Fusion, 47:181–188, 2007.
[23] W. L. Wiese. Line broadening. In R. H. Huddlestone and S. L. Leonard, editors,
Plasma Diagnostic Techniques. Academic Press, 1965.
[24] C. Stehlé and R. Hutcheon. Extensive tabulations of Stark broadened hydrogen
line profiles. Astron. Astrophys. Suppl. Ser., 140:93–97, 1999.
[25] N. Konjević, A. Lesage, J. R. Fuhr, and W. L. Wiese. Experimental Stark widths
and shifts for spectral lines of neutral and ionized atoms. J. Phys. Chem. Ref.
Data, 31(3):819–927, 2002.
[26] D. Kumar and P. M. Bellan. Heterodyne interferometer with unequal path
lengths. Rev. Sci. Instrum., 77(8):083503, 2006.
[27] A. V. Voronin, V. K. Gusev, Yu. V. Petrov, N. V. Sakharov, K. B. Abramova,
E.M. Sklyarova, and S. Yu. Tolstyakov. High kinetic energy plasma jet generation
and its injection into the Globus-M spherical tokamak. Nucl. Fusion, 45(9):1039–
1045, 2005.
[28] Y. Marandet, P. Genesio, M. Koubiti, L. Godbert-Mouret, B. Felts, R. Stamm,
H. Capes, and R. Guirlet. Characterization of tokamak edge plasmas using
spectroscopic line profiles. Nucl. Fusion, 44:118–122, 2004.
[29] B. L. Welch, H. R. Griem, J. Terry, C. Kurz, B. LaBombard, B. Lipschultz,
E. Marmar, and G. McCracken. Density measurements in the edge, divertor and
X-point regions of Alcator C-Mod from Balmer series emission. Phys. Plasmas,
2(11):4246–4251, 1995.
[30] H. Margenau and M. Lewis. Structure of spectral lines from plasmas. Rev. Mod.
Phys., 31(3):569–615, 1959.

116
[31] V. S. Lisitsa. Stark broadening of hydrogen lines in plasmas. Sov. Phys. Usp.,
20(7):603–630, 1977.
[32] J. M. Luque, M. D. Calzada, and M. Sáez. Experimental research into the
influence of ion dynamics when measuring the electron density from the Stark
broadening of the Hα , and Hβ lines. J. Phys. B: At. Mol. Opt. Phys., 36:1573–
1584, 2003.
[33] S. Gasiorowicz. Quantum Physics. John Wiley & Sons, 2nd edition, 1996.
[34] S. Chandrasekhar. Stochastic problems in physics and astronomy. Rev. Mod.
Phys., 15(1):1–89, 1943.
[35] P. Kepple and H. R. Griem. Improved Stark profile calculations for the hydrogen
lines Hα , Hβ , Hγ , and Hδ . Phys. Rev., 173(1):317–325, 1968.
[36] M. A. Gigosos and V. Cardeñoso. New plasma diagnosis tables of hydrogen Stark
broadening including ion dynamics. J. Phys. B: At. Mol. Opt. Phys., 29:4795–
4838, 1996.
[37] C. Stehlé. Stark broadening of hydrogen Lyman and Balmer in the conditions
of stellar envelopes. Astron. Astrophys. Suppl. Ser., 104:509–527, 1994.
[38] S. Mar, J. A. Aparicio, M. I. de la Rosa, J. A. del Val, M. A. Gigosos, V. R.
González, and C. Pérez. Measurement of Stark broadening and shift of visible
N II lines. J. Phys. B: At. Mol. Opt. Phys., 33:1169–1184, 2000.
[39] M. Mimura and K. Sato. Convolution-fitting method for Doppler-broadening
plasma ion temperature measurement. Jpn. J. Appl. Phys., 34:1668–1672, 1995.
[40] Y. Liu, J. Lin, G. Huang, Y. Guo, and C. Duan. Simple empirical analytical
approximation to the Voigt profile. J. Opt. Soc. Am. B, 18(5):666–672, 2001.
[41] N. Agmon, Y. Alhassid, and R. D. Levine. An algorithm for finding the distribution of maximal entropy. J. Comp. Phys., 30:250–258, 1979.

117
[42] S. K. P. Tripathi, P. M. Bellan, and G. S. Yun. Observation of kinetic plasma
jets in a coronal-loop simulation experiment. Phys. Rev. Lett., 98:135002, 2007.
[43] C. W. Groetsch. Inverse Problems. Mathematical Association of America, 1999.
[44] G. S. Yun. Analysis on stability, reliability and efficiency of maximum entropy
deconvolution. Caltech Graduate Science Symposium, 2003.
[45] P. M. Bellan. private communication, 2003.
[46] Y. Alhassid, N. Agmon, and R. D. Levine. An upper bound for the entropy and
its applications to the maximal entropy problem. Chem. Phys. Lett, 53:22–26,
1978.
[47] S. F. Gull and G. J. Daniell. Image reconstruction from incomplete and noisy
data. Nature, 272(20):686–690, 1978.
[48] C. A. Romero-Talamás. Investigations of spheromak plasma dynamics: highspeed imaging at the Sustained Spheromak Physics Experiment and magnetic diagnostics at the CalTech spheromak experiment. PhD thesis, California Institute
of Technology, 2005.
[49] R. Wilczek and S. Drapatz. A high accuracy algorithm for maximum entropy
image restoration in the case of small data sets. Astron. Astrophys., 142:9–12,
1985.
[50] R. Holanda. Sensitivity of hot-cathode ionization vacuum gages in several gases.
Technical Report E-6759; NASA-TN-D-6815, NASA, 1972.
[51] D. W. A. Rees. Mechanics of Solids and Structures. Imperial College Press,
2000.
[52] J. E. Minor and H. S. Norville. Design of window glass for lateral pressures. J.
Archit. Eng., 12(3):116–121, 2006.