Untitled

University of California
Los Angeles
Generation and Amplification of Coherent

Radiation with Optical Orbital Angular
Momentum in a Free-Electron Laser
A dissertation submitted in partial satisfaction

of the requirements for the degree
Doctor of Philosophy in Physics
by
Erik Willard Hemsing
2011

c Copyright by
2011
The dissertation of Erik Willard Hemsing is approved.
Jianwei (John) Miao
Mateo Pellegrini
Claudio Pellegrini
James B. Rosenzweig, Committee Chair
University of California, Los Angeles

2011
ii
To my family
iii
Table of Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Outline of Dissertation . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Principles of FEL operation . . . . . . . . . . . . . . . . . . . . . 7
1.3 Principles of OAM light . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.1 Laguerre-Gaussian Paraxial Modes . . . . . . . . . . . . . 23
1.3.2 Understanding optical OAM . . . . . . . . . . . . . . . . . 28
2 Dielectric Eigenmode Expansion of High-Gain 3D FEL Equa-

tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 Outline of Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2 Dielectric Waveguide Expansion Modes in the Presence of a Local

Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Cold Electron Beam Fluid Description . . . . . . . . . . . . . . . 38
2.4 Coupled Excitation Equations . . . . . . . . . . . . . . . . . . . . 43
2.5 Bunching Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6 Initial Values for Modal Amplitudes due to Shot Noise . . . . . . 50
2.7 Supermode Matrix Solutions . . . . . . . . . . . . . . . . . . . . . 55
2.7.1 Single Transverse Mode . . . . . . . . . . . . . . . . . . . 58
2.8 Uncorrelated Energy Spread Effects . . . . . . . . . . . . . . . . . 62
2.8.1 Integro-Differential Energy Spread Equations . . . . . . . . 63
2.8.2 Limitations on Energy Spread . . . . . . . . . . . . . . . . 68
iv
2.9 Transverse Dynamical Effects . . . . . . . . . . . . . . . . . . . . 69
2.9.1 Integro-Differential FEL Emittance Equations . . . . . . . 73
2.10 Recovering the 1D regime . . . . . . . . . . . . . . . . . . . . . . 79
2.11 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3 Description of FEL Interaction in terms of Eigenmodes of Orbital

Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.1 Guided Laguerre-Gaussian Modes . . . . . . . . . . . . . . . . . . 86
3.2 Dielectric Mode Coupling Factor . . . . . . . . . . . . . . . . . . . 89
3.3 Scaled Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4 Bunching Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.5 Negligible Emittance . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.5.1 Relevant equations and coupling factor . . . . . . . . . . . 94
3.5.2 Modeling of FEL with η , ηγ → 0 . . . . . . . . . . . . . . 94
3.5.3 OAM mode amplification in η , ηγ → 0 system . . . . . . . 100
3.5.4 Longitudinal Dispersion . . . . . . . . . . . . . . . . . . . 112
3.5.5 Single Mode Limit . . . . . . . . . . . . . . . . . . . . . . 119
3.5.6 Fitting Formula for 3D Gain Length with Space Charge . . 126
3.6 Emittance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.6.1 Single mode limit . . . . . . . . . . . . . . . . . . . . . . . 131
3.6.2 Fitting Formula for 3D Gain Length with Emittance . . . 133
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4 Harmonic Interaction in a Helical Undulator . . . . . . . . . . . 140
v
4.0.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.1 Harmonic Interaction . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.1.1 Density Modulation at Harmonics in the Modal Expansion 142
4.1.2 Emittance and Energy Spread Effects . . . . . . . . . . . . 152
4.2 High-Gain High Mode Generation (HGHMG) Scheme . . . . . . . 159
4.2.1 Unwanted mode coupling through misalignment . . . . . . 162
4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5 Helical Microbunching Experiment at the Neptune Laboratory 166
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.2 Design of the Experiment . . . . . . . . . . . . . . . . . . . . . . 166
5.3 Neptune Laboratory . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.3.1 CO2 Laser . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.3.2 Laser pulse measurements . . . . . . . . . . . . . . . . . . 172
5.3.3 Electron Beamline . . . . . . . . . . . . . . . . . . . . . . 174
5.3.4 Helical Undulator . . . . . . . . . . . . . . . . . . . . . . . 179
5.4 Analytic and numerical modeling of helical microbunching . . . . 187
5.4.1 Excitation of adjacent modes . . . . . . . . . . . . . . . . 192
5.5 Coherent Transition Radiation from helical beam . . . . . . . . . 196
5.5.1 CTR from ideal beam . . . . . . . . . . . . . . . . . . . . 197
5.5.2 Helical Phase Detection . . . . . . . . . . . . . . . . . . . 201
5.6 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . 209
5.7 Discussion: Evidence for helical microbunching . . . . . . . . . . . 213
vi
5.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
A List of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
B Coupling Coefficients and Supplementary Calculations . . . . . 227
C Mode Expansion of Integro-Differential FEL Equations . . . . 231
D Helical Undulator Focusing and Beam Transport . . . . . . . . 234
D.0.1 Simple Helical Undulator Transport . . . . . . . . . . . . . 235
E Coherent Transition Radiation . . . . . . . . . . . . . . . . . . . . 240
E.1 Single particle emission . . . . . . . . . . . . . . . . . . . . . . . . 240
E.2 Coherent emission from multiple particles . . . . . . . . . . . . . . 244
E.2.1 CTR from Entire Beam . . . . . . . . . . . . . . . . . . . 246
E.2.2 CTR from Microbunching . . . . . . . . . . . . . . . . . . 247
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
vii
List of Figures
1.1 Helical FEL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Ponderomotive phase bucket . . . . . . . . . . . . . . . . . . . . . 12
1.3 Spontaneous emission distribution . . . . . . . . . . . . . . . . . . 15
1.4 Off-axis harmonic emission . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Helical phase and Poynting vector of OAM mode. . . . . . . . . . 22
1.6 Phase contours for azimuthal modes. . . . . . . . . . . . . . . . . 24
1.7 LG mode intensity distributions. . . . . . . . . . . . . . . . . . . . 25
1.8 LG mode phase distributions. . . . . . . . . . . . . . . . . . . . . 26
1.9 Electric field amplitude contours for several LG modes in the x − z

plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1 Comparison of modal shot noise contribution to power and gain

with numerical SASE simulations . . . . . . . . . . . . . . . . . . 54
2.2 Comparison between Gaussian and Cauchy-Lorentz energy distri-

butions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.3 Matched beam envelope . . . . . . . . . . . . . . . . . . . . . . . 72
2.4 Space-charge and energy spread effects on a 1D FEL. . . . . . . . 82
3.1 Predicted gain length versus number of expansion modes and ex-
pansion waist size . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2 Gain and profile evolution for the seeded VISA FEL compared
with Genesis 1.3 (parallel beam limit) . . . . . . . . . . . . . . . . 96
viii
3.3 Single mode profile vs Multimode comparison for large diffraction
parameter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.4 Gain length versus detuning . . . . . . . . . . . . . . . . . . . . . 99
3.5 OAM theory vs simulation comparison. . . . . . . . . . . . . . . . 101
3.6 OAM mode gain in large diffraction parameter system. . . . . . . 102
3.7 Single OAM mode profile vs Multimode comparison for large diffrac-
tion parameter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.8 OAM Intensity and Phase after FEL amplification. . . . . . . . . 104
3.9 OAM mode amplification in diffraction dominated FEL . . . . . . 105
3.10 OAM supermode excitation efficiency versus seed spot size and
longitudinal position of optical beam waist . . . . . . . . . . . . . 106
3.11 Modeling of SASE FEL . . . . . . . . . . . . . . . . . . . . . . . . 107
3.12 Predicted region of OAM dominance versus longitudinal position

for VISA FEL. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.13 Relative OAM mode power versus initial helical bunching factor. . 111
3.14 Dispersion of l=1 and l = 0 modes in cold beam . . . . . . . . . . 113
3.15 OAM mode competition in a helically micro-bunched e-beam . . . 114
3.16 Dispersion of l=2 and l = 0 modes in cold beam . . . . . . . . . . 115
3.17 Real part of the scaled complex wavenumber for azimuthal modes
versus detuning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.18 Real part of the scaled complex wavenumber for azimuthal modes
as a function of energy spread. . . . . . . . . . . . . . . . . . . . . 117
3.19 3D modal dispersion. . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.20 Effect of energy spread on OAM gain. . . . . . . . . . . . . . . . . 121
ix
3.21 Effect of space charge on OAM gain. . . . . . . . . . . . . . . . . 122
3.22 Re{α} in single-mode description. . . . . . . . . . . . . . . . . . . 124
3.23 Im{α} in single-mode description. . . . . . . . . . . . . . . . . . . 125
3.24 Fit function surface for T EM00 mode. . . . . . . . . . . . . . . . . 127
3.25 Fit function surface for T EM01 mode. . . . . . . . . . . . . . . . . 128
3.26 Effect of space charge and diffraction on the T EM00 and T EM01
mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.27 Effect of large emittance on gain in quasi-1D regime. . . . . . . . 134
3.28 Effect of emittance on OAM gain for fixed betatron phase advance. 134
3.29 Effect of transverse mixing on gain of T EM00 and T EM01 modes. 135
3.30 Power fit with emittance for lowest OAM modes. ηγ vs η . . . . . 137
3.31 Power fit with emittance for lowest OAM modes. ηd vs η . . . . . 138
4.1 Helical buncher cartoon. . . . . . . . . . . . . . . . . . . . . . . . 147
4.2 Helical bunching growth in cold beam at second harmonic. . . . . 150
4.3 Helically microbunched beam from TREDI simulations. . . . . . . 151
4.4 Longitudinal space-charge effects in cold beam helical microbunch-

ing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.5 Evolution of helical density distribution during matched betatron

focusing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.6 Bunching factor fluctuations during beta-matched transport . . . 156
4.7 Helical beam modulation evolution . . . . . . . . . . . . . . . . . 157
4.8 Optical klystron arrangement for OAM generation at high-energies. 160
4.9 Simulated OAM mode power gain at LCLS. . . . . . . . . . . . . 162
x
4.10 Simulated transverse OAM intensity and phase at saturation at
LCLS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.11 Simulated transverse OAM intensity and phase at saturation at

LCLS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.1 HELiX concept. . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.2 Neptune laser system. . . . . . . . . . . . . . . . . . . . . . . . . 169
5.3 Measured temporal synchronization of the 100 ps CO2 laser seed

with the e-beam. . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.4 CO2 laser seed spot size transport in HELiX modulator. . . . . . 174
5.5 Neptune beamline . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.6 Cathode to undulator Neptune transport modeled with UCLA-

Parmela. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.7 Example beam distribution at undulator entrance from Parmela

simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5.8 HELiX undulator magnets from RADIA . . . . . . . . . . . . . . 181
5.9 E-beam position and divergence in undulator . . . . . . . . . . . . 181
5.10 Electron trajectory in undulator. . . . . . . . . . . . . . . . . . . 182
5.11 CAD undulator model. . . . . . . . . . . . . . . . . . . . . . . . . 183
5.12 Magnet in holder . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.13 Modeled and measured on-axis undulator magnetic fields . . . . . 185
5.14 Modeled and measured undulator K̄ . . . . . . . . . . . . . . . . 185
5.15 Arrangement of undulator and CTR screen in interaction box. . . 187
xi
5.16 Calculated bunching factor inside HELiX undulator for a matched
beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.17 Calculated and simulated bunching factor versus energy at the

undulator exit for HELiX . . . . . . . . . . . . . . . . . . . . . . 189
5.18 Calculated bunching factor at the undulator exit for HELiX versus
energy and laser spot size w0 . . . . . . . . . . . . . . . . . . . . . 190
5.19 Calculated bunching factor versus energy and laser waist position 190
5.20 Calculated bunching factor versus spot size and laser waist position.191
5.21 Calculated bunching factor versus energy and emittance. . . . . . 191
5.22 Comparison between helical bunching factors in the HELiX and

in a comparable undulator with on-axis e-beam transport. . . . . 194
5.23 Proper transport in HELiX undulator. . . . . . . . . . . . . . . . 195
5.24 Calculated angular and spectral CTR emission. . . . . . . . . . . 198
5.25 Simulated CTR output from QUINDI particle simulations. . . . . 199
5.26 Calculated angular CTR emission for different l-modes. . . . . . . 200
5.27 Simulated CTR intensity distribution comparison between e-beam

with on-axis transport and one with off-axis transort. . . . . . . . 201
5.28 Modeled ideal CTR profile viewed through a triangular aperture. 203
5.29 Modeled profile of CTR that is emitted from beam microbunched

in Tredi simulations and sent through a triangular aperture. . . . 204
5.30 Picture of helical phase interferometer. . . . . . . . . . . . . . . . 206
5.31 Idealized l = 0 CTR interferometer signal. . . . . . . . . . . . . . 207
5.32 Idealized l = 1 CTR interferometer signal. . . . . . . . . . . . . . 207
xii
5.33 Calculated normalized total interferometer output power as a func-
tion of delay for different modes. . . . . . . . . . . . . . . . . . . . 208
5.34 Measured CTR signal at HCT detector as a function of laser power.211
5.35 Measured CTR signal at HCT detector as a function of e-beam

energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
5.36 Comparison of measured CTR signal with Tredi simulations a a

function of e-beam energy. . . . . . . . . . . . . . . . . . . . . . . 213
5.37 Coupling to other microbunching modes through parallel misalign-

ment of e-beam and laser. . . . . . . . . . . . . . . . . . . . . . . 215
5.38 Coupling to other microbunching modes through angular misalign-

ment of e-beam and laser. . . . . . . . . . . . . . . . . . . . . . . 216
E.1 Schematic of particle motion and location of field measurement

position of the CTR emitted at the foil. . . . . . . . . . . . . . . . 241
xiii
List of Tables
3.1 Scaled FEL parameters . . . . . . . . . . . . . . . . . . . . . . . . 90
3.2 Model FEL parameters . . . . . . . . . . . . . . . . . . . . . . . . 95
5.1 Neptune CO2 laser parameters . . . . . . . . . . . . . . . . . . . . 173
5.2 Neptune beam parameters . . . . . . . . . . . . . . . . . . . . . . 179
5.3 HELiX undulator parameters . . . . . . . . . . . . . . . . . . . . 180
xiv
Acknowledgments
Completion of this work would not have been possible without the generous
contribution of many people.
My thesis advisor Professor James Rosenzweig is continuous source of fresh

ideas and encouragement. His insight, diplomacy, boundless energy, unselfish-
ness, and contagious enthusiasm made the whole of this work possible. I am
also grateful to Professor Avraham Gover who exposed me to, and ushered me
through, the initial modal formalism that forms the basis of the theoretical de-
scription in this work. My conversations with Professor Pietro Musumeci are
always inspiring and fruitful, and his open office door is emblematic of his will-
ingness to entertain new ideas and to provide thoughtful and welcomed advice.
Big thanks go likewise to Dr. Sven Reiche who also provided me with astute
counsel on FEL theory and with Genesis simulations.
I am deeply indebted to my friend and colleague Dr. Agostino Marinelli, with

whom I have enjoyed numerous discussions on physics and philosophy, and with
whom I worked closely to develop some of the advanced modal theory here. I
also thank him for bravely proofreading this dissertation and thus assuming tacit
responsibility for all errors (of which I absolve him as a dutiful ersatz Godfather).
I am also grateful to my friends and officemates both present and former, Dr.
Gerard Andonian, Dr. Michael Dunning, and (soon-to-be Dr.) Gabriel Marcus,
who made working in the office and the labs a pleasure, and who are owed due
commendation for surviving the local entropic maximum around my desk – the
natural result of turning coffee into ““theory””. To my friend Oliver Williams I
am profoundly indebted, for his help in the trenches, for his experimental prowess,
and good wine. Thanks also to the PBPL computational virtuoso David Schiller
whose technical help and contributions to the QUINDI simulations helped eluci-
xv
date the complicated helical e-beam CTR field profile.
Much of the experimental work was shared by several fellow PBPLers: An-
drey Kynazik, Finn O’Shea, Dr. Atsushi Fukasawa, Brendan O’Shea and more,
to whom I am thankful for their efforts. The machining of the undulator assem-
bly and of numerous other parts was done expertly by Harry Lockhart and his
fantastic crew in the Physics and Astronomy department machine shop. I would
also like to thank Dr. Sergei Tochitsky, Dan Haberberger, Dr. Alan Cook, and
the rest of the Neptunian fraternity of past and present, whose work yielded the
experimentally fertile accelerator and laser systems used in this experiment.
Finally, I am eternally grateful to my family - to my amazing wife Lea for

her love, support, and contagious joie de vivre; to my siblings, of whom I am
ceaselessly amazed and proud, and to my extraordinary father and mother, for
everything.
xvi
Vita
1979 Born, Los Alamos, New Mexico, USA.
2001 B.S. (Physics) and B.S. (Mathematics), University of New Mex-

ico.
2002 – 2004 Post Baccelaureate Student Researcher, Los Alamos National

Laboratory
2002 – 2004 Plasma Physics Summer School, Los Alamos National Labora-
tory
2004 Outstanding Student Award, Los Alamos National Laboratory
2004 – 2005 M.S. (Physics), UCLA.
2005 Top Score, Comprehensive Qualifying Exam in Physics, UCLA.
2004 – 2005 Teaching Assistant, Physics and Astronomy Department,

UCLA.
2005 US Particle Accelerator School, Cornell University, NY.
2006 US Particle Accelerator School, Boston University, MA
2005 – 2011 Graduate Student Researcher, Physics and Astronomy Depart-

ment, UCLA.
xvii
Publications
E. Hemsing, A. Marinelli, and J. B. Rosenzweig. Generation of optical orbital an-

gular momentum in a high-gain free-electron laser at the first harmonic. Physical
Review Letters, 106(16):164803, 2011.
E. Hemsing, et al. Helical electron-beam microbunching by harmonic coupling

in a helical undulator. Physical Review Letters, 102(17):174801, 2009.
E. Hemsing and J. B. Rosenzweig. Coherent transition radiation from a he-

lically microbunched electron beam. Journal of Applied Physics, 105(9):093101,
2009.
Erik Hemsing, Avraham Gover, and James Rosenzweig. Virtual dielectric

waveguide mode description of a high-gain free-electron laser. I. Theory. Physical
Review A (Atomic, Molecular, and Optical Physics), 77(6):063830, 2008.
Erik Hemsing, Avraham Gover, and James Rosenzweig. Virtual dielectric

waveguide mode description of a high-gain free-electron laser. II. Modeling
and numerical simulations. Physical Review A (Atomic, Molecular, and Opti-
cal Physics), 77(6):063831, 2008.
Erik Hemsing, Agostino Marinelli, Sven Reiche, and James Rosenzweig. Lon-
gitudinal dispersion of orbital angular momentum modes in high-gain free-electron
lasers. Phys. Rev. ST Accel. Beams, 11(7):070704, Jul 2008.
xviii
E. Hemsing, J. Rosenzweig, and A. Gover. A virtual dielectric eigenmode
expansion of high-gain fel radiation for study of paraxial wave mode coupling.
Nuclear Instruments and Methods in Physics Research Section A: Accelerators,
Spectrometers, Detectors and Associated Equipment, 593(1-2):98 – 102, 2008.
E. Hemsing, et al. Analysis of visible light images from a fast-gated inten-

sified charge coupled device camera during flux rope interaction and magnetic
reconnection. Review of Scientific Instruments, 75(10):4106–4108, 2004.
G. Andonian, et al. Longitudinal profile diagnostic scheme with sub-femtosecond

resolution for high-brightness beams. Submitted to Phys. Rev. ST-AB. 2011
G. Andonian, et al. Resonant Excitation of Coherent Cerenkov Radiation in

Dielectric Lined Waveguides. Appl. Phys. Lett. 98(20): 202901, 2011.
A. Marinelli, E. Hemsing, and J. B. Rosenzweig. Three dimensional anal-

ysis of longitudinal plasma oscillations in a thermal relativistic electron beam.
Submitted to Phys Rev E, 2011.
Ivo Furno, et al. Coalescence of two magnetic flux ropes via collisional mag-
netic reconnection. Physics of Plasmas, 12(5):055702, 2005.
E. Torbert, et al. A plasma-shielded, miniature Rogowski probe. Review of

Scientific Instruments, 74(12):5097–5100, 2003.
E.W. Hemsing, I. Furno, and T.P. Intrator. Fast camera images of flux ropes
xix
during plasma relaxation. Plasma Science, IEEE Transactions on, 33(2):448–
449, April 2005.
G. Andonian, et al. Observation of Coherent Edge Radiation Emitted by a

100 Femtosecond Compressed Electron Beam. International Journal of Modern
Physics A, 22:4101–4114, 2007.
M. Dunning, et al. Seeded visa: A 1064nm laser-seeded FEL amplifier at the

BNL ATF. Particle Accelerator Conference, 2007. PAC. IEEE, pages 1257–1259,
June 2007.
E. Hemsing, et al. Characterization of orbital angular momentum modes in

FEL radiation. Particle Accelerator Conference, 2007. PAC. IEEE, pages 1263–
1265, June 2007.
G. Andonian, et al. Chicane radiation measurements with a compressed elec-

tron beam at the BNL ATF. Particle Accelerator Conference, 2007. PAC. IEEE,
pages 1254–1256, June 2007.
A. Fukasawa, et al. The FINDER photoinjector. Particle Accelerator Con-

ference, 2007. PAC. IEEE, pages 1260–1262, June 2007.
J.B. Rosenzweig, et al. Generation of ultra-short, high brightness electron

beams for single-spike sase FEL operation. Nuclear Instruments and Methods in
Physics Research Section A: Accelerators, Spectrometers, Detectors and Associ-
ated Equipment, 593(1-2):39 – 44, 2008.
xx
G. Andonian, et al. Advanced studies at the visa FEL in the SASE and
seeded modes. Nuclear Instruments and Methods in Physics Research Section A:
Accelerators, Spectrometers, Detectors and Associated Equipment, 593(1-2):11 –
13, 2008.
xxi
Abstract of the Dissertation
Generation and Amplification of Coherent

Radiation with Optical Orbital Angular
Momentum in a Free-Electron Laser
by

Doctor of Philosophy in Physics
University of California, Los Angeles, 2011
Professor James B. Rosenzweig, Chair
The object of this work is to examine how coherent light that carries orbital angu-
lar momentum (OAM) can be generated and amplified in a single pass, high-gain
free-electron laser (FEL) at the fundamental operating frequency. This concept
unites two rapidly expanding, but at present largely non-overlapping fields of
study: high-order OAM light modes, which interact in new ways with matter,
and FELs, in which a relativistically energetic electron beam emits coherent, ul-
tra high-brightness, highly frequency-tunable light. The ability to generate OAM
light in an FEL enables new regimes of laser interaction physics to be explored
at wavelengths down to hard x-rays.
The theoretical portion of this dissertation attempts to provide a new pre-

dictive mathematical framework. It builds on existing work, and describes the
three-dimensional electromagnetic field of the high-gain FEL as a sum of OAM
modes such that the amplification properties of individual modes can be char-
acterized. The effects of uncorrelated energy spread, longitudinal space charge,
energy detuning, and transverse emittance in the electron beam are included,
xxii
as is the diffraction of the laser light. Theoretical predictions are corroborated
by detailed numerical Genesis 1.3 simulations. When the theory is extended to
frequency harmonics, a novel interaction is uncovered that generates a helical
electron beam density distribution. These predictions are also supported by nu-
merical Tredi simulations. This type of highly correlated structure is shown to
naturally emit OAM light, and forms the basis of a new high-gain, high-mode
generation (HGHMG) scheme proposed in its entirety here.
The experimental section examines the helical microbunching concept in a

proof-of-principle experiment dubbed HELiX, performed at the UCLA Neptune
laboratory. We present detailed measurement of the coherent transition radia-
tion emitted by the 12.5 MeV electron beam that is microbunched in a second
harmonic interaction with an input laser and helical undulator. The predicted
dependence of the CTR signal on the input laser polarization is observed, and
is consistent with microbunching that has a periodicity near the 10.6 µm wave-
length of the 30 MW CO2 laser pulse. Scans of the interaction energy bandwidth
are consistent with predictions that indicate a dominant azimuthal density mode
with a bunching factor of 10%, and thus provide indication of the first experi-
mental evidence of helical microbunching. This result offers support for future
successful realization of the proposed HGHMG scheme to generate OAM modes
in high-gain FELs.
xxiii
CHAPTER 1
Introduction
Electron beam accelerators have emerged in the last several decades as reliable,
efficient and tunable sources of electromagnetic (EM) radiation. Numerous labs
and facilities worldwide use electron beams (e-beams) to generate radiation that
spans many orders of magnitude in wavelength; from millimeters (10−3 m) to
Ångstroms (10−10 m), for applications that run the gamut from basic research on
fundamental physics to medical treatment.
With the development of ever more advanced accelerator facilities has also
come increased scientific progress on accelerator-based light source technologies.
At the forefront is the free-electron laser, or FEL. In an FEL the highly ener-
getic electron beam emits light in the forward direction as it traverses a periodic
magnetic field, a process sometimes referred to as stimulated bremsstrahlung ra-
diation [1]. When the light is amplified, the FEL operates as a laser, which stands
for light amplification by stimulated emission of radiation. In this sense, the FEL
is akin to conventional laser sources in that it generates light that has a high de-
gree of spatial coherence (the wavefronts of the light are in phase), low divergence
(the beam is well-collimated) and narrow bandwidth (the laser emits primarily
at a single frequency, or at several well-separated frequencies). In contrast to
conventional sources however, light is generated in an FEL by the free – that is,
unbound – electrons. Accordingly, FELs belong to a class of vacuum tube-like
devices in which the electrons interact with the radiation field in vacuum. This
1
distinction uniquely provides the FEL enormous versatility; the wavelength of
the light in the FEL is determined by the electron beam energy, and is thus not
restricted to specific wavelengths dictated by energy transitions in the active las-
ing medium. FELs can therefore generate high power coherent radiation over a
extremely large range of wavelengths. They can also obtain high average power
levels since FELs do not suffer from thermal lensing, birefringence, or other heat
dissipation issues that tend to place constraints on conventional lasers. By op-
erating in vacuum, FELs also generate essentially diffraction-limited radiation,
which is ideal for many transport and imaging applications.
With all the unique features and advantages of the FEL, it has come to
occupy its own niche among modern coherent light sources. The initial concept
dates back to the work of Hans Motz who, in 1951, examined the radiation
emitted from an e-beam passing a periodic array of electric or magnetic fields
[2]. Motz even suggested that such a device may be a useful source of narrow
band tunable radiation at wavelengths all the way down to hard x-rays. Some
twenty years later, the first working FEL operating in the infrared regime was
constructed by John Madey and his colleagues at Stanford University. Now, over
the sixty years since its inception, the FEL concept has evolved and the field
has undergone tremendous progress. Indeed, the modern age of the FEL as a
“fourth generation” coherent radiation source has arrived with the recent high-
gain lasing of the Linac Coherent Light Source (LCLS) FEL at x-ray wavelengths
(1.5Å) and at unprecedented peak spectral brightness (1034 photons s−1 mm−2
mrad−2 0.1%BW −1 ) [3]. And there is much more on the horizon.
With the continuing development and refinement of FEL technologies, as well

as the push to operate FELs at ever shorter wavelengths and higher peak and av-
erage brightnesses, there is also interest in operating FELs in more exotic regimes
2
to further extend both their versatility and applicability. One new regime, which
drives the central theme of this dissertation, is the precise, controlled manip-
ulation of the three-dimensional spatial distribution of the FEL radiation field
intensity and phase across the entire range of accessible wavelengths. Consider
that, in most modern high-gain FELs, the light is purposefully generated to have
a simple, round transverse distribution. That is, the intensity of the emitted light
is typically localized to a small axisymmetric spot where the intensity is peaked
on-axis and falls off towards the edge. The phase is likewise axisymmetric, having
no azimuthal variation. In the simplest case, the optical intensity profile on a
screen is described by Gaussian intensity distribution as a function of the radius
(see, for example, the simple round intensity spot shown in the upper left corner
of Figure 1.7). Such simple Gaussian distributions are highly desirable for many
applications in modern laser science, and are ubiquitous because they arise as
the lowest order mathematical solutions to the wave equation that describes the
propagation of a finite-sized laser beam in free-space. However, there are also
realizable higher-order solutions that describe optical modes with more elaborate
transverse intensity and phase distributions. These modes have different phys-
ical characteristics than the simple Gaussian, so they propagate differently and
interact in different ways with matter. Accordingly, they can provide access to a
different class of laser-based experiments and applications. The ability to gener-
ate such modes in an FEL, particularly in situ, thus provides a compelling level
of added flexibility to the already considerable suite of useful features specific to
FEL light.
This thesis examines a particular subset of higher-order transverse optical

modes in the FEL: light modes that carry orbital angular momentum (OAM).
These modes are identified in part by their characteristic doughnut-shaped trans-
verse intensity profiles in which the peak intensity is localized to an annular region
3
off-axis and is zero in the middle (see again Figure 1.7). This hollow structure
is also naturally accompanied by an on-axis phase singularity (or vortex) as well
as a helical off-axis phase dependence that describes the spiral-like instantaneous
field amplitude as a function of position around the ring (Figure 1.8). Because
of this unique structure each of the photons carries l~ units of orbital angular
momentum, where l is the integer azimuthal mode number and ~ is Planck’s
constant [4]. This orbital component of the angular momentum comes in addi-
tion to any spin angular momentum (SAM) the modes may possess due to the
polarization of the field. It is this basic feature that distinguishes OAM modes
from other higher-order modes, and is the mechanism by which they manifest
their unique properties. The intrinsic helical phase is realized in a physical sense
as an azimuthal component of the linear photon momentum, so the field energy
corkscrews about the axis.
It is easy to see how this additional momentum component presents a myr-

iad of new research opportunities. OAM modes generated by conventional lasers
have already demonstrated light-driven micro-mechanical rotational devices in
which micron sized particles are driven to rotate with optical tweezers[5]. OAM
light also exhibits rotationally induced frequency shifts much like circularly po-
larized light[6]. Experiments with rotating cylindrical lenses have shown that
the frequency of the light can be adjusted via an azimuthal Doppler effect [7],
and later work has shown that an OAM light beam that is circularly polar-
ized can undergo a rotation frequency shift proportional to the total angular
momentum[8]. This confirmed predictions of such shifts in atomic[9] and quan-
tum systems[10] and provides an additional experimental knobs in interactions
with matter whereby ‘rotating’ states are subject to different frequencies. OAM
light modes have also opened the door to applications in sub-diffraction limit
microscopy[11], information encoding[12], quantum entanglement schemes[13],
4
Bose-Einstein condensates[14], and molecular transitions[15, 16]. The additional
degree of freedom provided by the OAM is also predicted to provide experimental
access to higher-order processes at short wavelengths. X-rays with OAM have
recently been proposed for research in scattering and spectroscopy[17], chirality
in biological materials[18], and quadrupolar transitions in materials due to strong
dichroic properties[19]. Indeed, the additional ability to generate OAM light in
modern FELs provides a further opportunity to probe matter with OAM photons
at Å length and attosecond time scales not currently obtainable in conventional
sources. Not only would this serve to extend the versatility of FELs into new
regimes, but could open and expand avenues of research in fundamental physics
and light-source based application technologies.
The present challenge comes in the methods by which the OAM modes are
typically generated. OAM modes can be produced in numerous ways with con-
ventional sources. Spiral waveplates cause an azimuthally dependent phase delay
in a transmitted beam to produce a helical phase[20], laser cavities can be opti-
mized to emit optical vortices[21], mode converters composed of pairs of cylin-
drical lenses can be used to convert higher-order Hermite-Gaussian modes into
cylindrically symmetric OAM modes[22], and phase masks such as forked holo-
grams are often employed like a diffraction grating to generate different orders of
OAM modes as the light passes through a dislocation[23]. Many of these meth-
ods are being explored both in the radio wave regime[24] and at x-rays[25]. The
difficultly for modern high-brightness, single-pass, high peak power sources like
the LCLS FEL is that techniques that rely on cavity tuning or external optical
elements are either impractical or unavailable due to losses, material thresholds,
or other systematic constraints. Further, at present the broad range of opera-
tional frequencies accessible to individual FEL devices does not also extend to
the materials required to perform such direct manipulations on the optical beam.
5
The generation of OAM light in FELs therefore requires the development of al-
ternate methods which, in turn, demand an understanding how the OAM light
interacts with, and is amplified by, the source electron beam. Such is the focus
of this dissertation.
1.1 Outline of Dissertation
The outline of this work is as follows. The principles of FEL operation are
presented briefly in the next section, followed by an introduction to the physics
of light that carries OAM. To set the stage for how these two subjects may be
merged, Chapter 2 presents an in-depth analytic formulation of the high-gain
3D FEL in terms of a generalized mode expansion formalism. In this approach,
both the optical beam and the electron beam are described by a common set
of functions (modes) that enables one to investigate how the amplified optical
modes in the FEL couple to the modes that describe the density and velocity
distribution of electrons in the e-beam. This model is is dually inspired by features
of OAM light and by features of the high-gain FEL modes. Thus it lays the
general foundation for examining OAM mode amplification presented in Chapter
3, wherein the effects of diffraction in the optical beam and energy spread, space-
charge and emittance in the e-beam are all included. Chapter 4 extends the
analysis to the higher-harmonics, where a newly developed High-Gain High Mode
Generation (HGHMG) scheme that generates OAM in and FEL is described.
This sets the stage for the proof-of-principle harmonic microbunching experiment
described in Chapter 5.
6
1.2 Principles of FEL operation
The details of high-gain FELs are explored in Chapter 2. Here, however, a basic
outline of the physical process that governs emission from a charged particle in
an FEL undulator is presented.
FELs operate on the principle that accelerating charges emit radiation. One
of the simplest ways to accelerate a charged particle like an electron is to send
it through a transverse magnetic field. The field changes the trajectory of the
particle, accelerating it in a direction transverse to the direction of initial motion.
By traversing a successive series of magnets that periodically alternate in their
polarity, an electron propagating through an undulator will follow a oscillatory
path, emitting radiation as a consequence of the transverse acceleration. For
a relativistic electron traveling close to the speed of light v0 ' c, most of the
radiation is emitted in the forward (ie, downstream) direction, producing a narrow
angular light distribution. The periodic structure also provides a mechanism by
which radiation at specific frequencies can be naturally selected and amplified.
Consider a helical undulator magnetic field (Figure 1.1) with periodicity λw ,
B⊥w = B̄w (êx cos kw z + êy sin kw z) . (1.1)
where B̄w is the root-mean-square (rms) magnetic field, kw = 2π/λw is the un-
dulator wavenumber and êx and êy are unit vectors pointing in the x and y
directions, respectively. In this form, the transverse magnetic field evolves in a
right-handed sense along the z axis, pointing in the x direction at z = 0, and
then in the y direction one quarter period later. An electron traveling in the
z-direction at velocity v0 will have a transverse helical velocity component due to
the Lorentz force on the particle given by the equation γmdv⊥ /dt = −ev0 êz ×Bw .
7
Λw
x
z
Figure 1.1: The electron in a helical undulator traces out a corkscrew path with
period λw . The relative axial shift of λw /4 between the magnets in the two planes
generates the helical field geometry.
8
The dominant transverse velocity is then,
K̄c
v⊥ = − (êx cos kw z + êy sin kw z) (1.2)
γ
where
|eB̄w |
K̄ = (1.3)
mckw
is the rms undulator parameter, c is the speed of light, m is the mass of an

electron, −e is the electron charge, γ = E0 /mc2 is the relativistic energy factor
and E0 is the total energy of the electron. The electron traces out a helical path
in the undulator, completing one orbit every undulator period λw .
We also consider a plane EM field that is helically polarized and propagating

co-linearly with the electron,
E⊥ = Ē0 [êx sin(kz − ωt) + êy cos(kz − ωt)] (1.4)
where Ē0 is the rms field amplitude (which is the peak field in a helical geometry),
λ = 2π/k is the wavelength, and ω = kc is the frequency. Propagating together,
the field can transfer energy to the electron at a rate given by
dE0 eĒ0 K̄c

= −eE⊥ · v⊥ = sin [(k + kw )z − ωt] . (1.5)
dt γ
If the energy exchange rate is positive (dE0 /dt > 0), the electron receives energy
from the field and is accelerated longitudinally. If the rate is negative (dE0 /dt <
0), the field gains energy from the electron and grows in amplitude.
Since the free EM wave propagates at velocity c and the electron travels slower
(at dz/dt = v0 < c), the electron slips backward with respect to the radiation
field. At a certain radiation frequency, however, the electron slips backward
9
exactly one wavelength as it goes through one undulator period λw , thus main-
taining the same position in phase. The cycle continues, due to the periodicity of
the system, and the electron interacts continuously with the resonant wavelength
λR . The value of this wavelength is found by examining the rate of change of the
phase ψ = (k + kw )z − ωt in (1.5);
dψ
= (k + kw )v0 − kc (1.6)
dt
For the electron to remain stationary in phase this quantity must be zero. The
resonant longitudinal electron velocity is then
kR c c
v0 = = . (1.7)
kR + kw 1 + λR /λw
This can be written in terms of the relativistic factor γ, which is given by the
−1 v02 +|v⊥ |2
total velocity of the electron as γ 2 = (1 − β 2 ) where β 2 = βz2 + β⊥2 = c2
.
Combining this with (1.7) for a relativistic beam (γ 1) we obtain the resonant
wavelength condition for an FEL,

1 λw
− 1 ' 2 1 + K̄ 2 .

λR = λ w (1.8)
βz 2γ
This expression highlights two essential features of the FEL. The first is that
with highly relativistic electrons one can obtain wavelengths much shorter than
the undulator period. The second feature is the FEL tunability; for a given
undulator defined by λw and K̄, the resonant radiation wavelength is determined
by the electron energy, so changing the wavelength is done simply by changing
the energy (ie, velocity) of the electron.
During the FEL interaction, the electron energy deviates by only small values
from the resonant energy, to first order. We therefore define a scaled energy
10
parameter for the electron near the resonant energy γR :
γ − γR
η= 1, (1.9)
γR
such that the rate of energy change in (1.5) is dγ/dt ' cγR dη/dz. Together,
equations (1.5) and (1.6) become
dη eĒ0 K̄
= 2 2 sin ψ
dz γR mc
(1.10)
dψ
= 2kw η
dz
These are the familiar pendulum equations that describe the evolution of the
electron energy η and phase ψ along the longitudinal axis during the interaction1 .
Figure 1.2 displays the behavior of a collection of electrons in the pondero-

motive phase bucket as governed by the pendulum equations. An initially un-
modulated beam enters the undulator with the electrons randomly distributed in
phase. It then becomes sinusoidally modulated in energy after interacting with
the combined fields of the radiation and undulator. The small energy modula-
tion is well-described by a simple sinusoid at a single frequency. The modulation
grows as the interaction proceeds and the electrons continue along their phase-
space trajectories; trapped electrons rotate about the center of the bucket while
untrapped electrons (not shown) follow the oscillatory trajectories outside the
separatrix boundary (thick blue lines surrounding shaded region). Note that the
trapped electrons at phase positions for which the energy transfer is positive
sin ψ > 0 move up in energy, but that the total energy exchange of the beam is
zero since an equivalent number of electrons also lose energy. This identifies the
1
The field is assumed to have a constant amplitude E¯0 in this low-gain approximation. High-
gain FEL operations where the field amplitude grows slowly are the subject of the next chapter,
but the relevant dynamics of the electrons in the bucket are similar for present purposes.
11
0.004
0.002
Η 0.000
-0.002
-0.004
-2 Π -Π 0 Π 2Π
Ψ
Figure 1.2: Snapshots of electrons (dots) in the ponderomotive phase bucket at

three different positions along the undulator. The initially unmodulated beam
(blue dots) acquires a simple sinusoidal energy modulation (red) and eventually
develops a strong density modulation as electrons pile-up in the center of the
phase bucket (yellow).
low-gain regime where field is constant. As the interaction continues, the energy
modulation becomes more pronounced and eventually develops into a density
modulation as the electrons begin to collect in the center of each phase bucket.
In a high-gain FEL, the density modulation – known as microbunching – has

a feedback effect on the phase of the amplified radiation field. During this process
known as the FEL instability the ponderomotive bucket shifts to the left and the
microbunched electrons radiate coherently, depositing energy into the EM field
which subsequently grows exponentially. When the microbunching approaches
its maximum value, the energy spread of the e-beam covers the FEL bandwidth
and the high-gain process begins to saturate. At this point neither the energy
nor density modulations are simple sinusoids and the phase space distribution
becomes highly non-linear.
Let us examine the single electron description of the FEL to inspect some
12
features of the emission. Equation (1.8) gives the fundamental wavelength of the
spontaneous radiation emitted by the electron along the axis of the undulator.
In reality, the emission is not at only a single frequency, but instead has an
associated bandwidth centered about the resonant frequency as well as each of
the harmonics. The non-zero bandwidth is the natural result of the finite time
the electron radiates as it traverses the undulator with length L = Nw λw , where
Nw is the number of periods. Increasing the number of periods decreases the
spectral bandwidth of the emission, so one can in principle obtain very narrow
band radiation from an undulator with many periods. The radiation also has
finite spatial distribution that occupies a narrow forward cone about the forward
angle θ. Both of these features are found in the dominant terms of the forward
spectral and angular energy distribution of the spontaneous radiation. This is
given for a helical undulator as
d2 U 2 g2
= RE
dωdΩ µ0 c
2
µ0 eωNw K̄ 2 ω(1 − βz cos θ) 2 ω K̄ sin θ
' sinc Nw π − h Jh−1 .
8πc kw γ kw c γkw c
(1.11)
Here, U is the energy of the radiation field, RE
g is product of the distance from the
undulator R and the electric field E evaluated in the frequency domain, J(x) is a
Bessel function of the first kind, and h is the harmonic number of the frequency,
h = 1 being the fundamental. The function sinc(x) = sin(x)/x is peaked about
x = 0, which gives a strong peak in the angular spectral energy distribution for
frequencies ω = hkw c/(1 − βz cos θ). On-axis (θ = 0) and at the fundamental,
this is precisely the resonance relation in (1.7). Near θ = 0 we expand to reveal
the quadratic relationship between the angular radiation distribution and the
13
wavelength;
λw
1 + K̄ 2 + θ2 γ 2 .

λ' 2
(1.12)
2hγ
The half maximum of the sinc2 (x) function is found when x ' ±1.39. Thus one
can determine the FWHM of the spectral energy distribution both in angle and
in frequency. On-axis the spectral FWHM is
0.88
∆ωF W HM = ωR (1.13)
Nw
where ωR = hkw c/(1 − βz ) ' 2hγz2 kw c is the harmonic frequency peak, and γz2 =
(1 − βz2 )−1 is the longitudinal relativistic energy factor. Clearly, increasing the
number of undulator periods narrows the frequency bandwidth. At the harmonic
frequency ωR the angular distribution has a spatial FWHM given by
r
1 0.88
∆θF W HM ' (1.14)
γz Nw
which is also narrowed by increasing the number of periods, as well as by increas-

ing the beam energy. Note that since γz depends on K̄ by Eq. (1.8), an increase
in K̄ also leads to an increase in the angular FWHM. Physically, this is part of
the ‘searchlight’ effect where the electron exhibits larger transverse displacements
in the undulator for large K̄ that cause the downstream radiation to sweep out
larger angles.
Taking a look back at Eq. (1.11) we see two important aspects of the radiation
at harmonics, h > 1. The first is that the harmonic emission vanishes on axis by
virtue of the Bessel function Jh−1 (x), which is zero for x = 0 except when h = 1.
Thus, only the fundamental frequency is emitted along the undulator axis at
θ = 0. The second feature, which will play an important role in this dissertation,
is revealed by looking at the field components that make up the emitted energy
14
Θ=0
d2 U
dΩd FWHM ~ ΩR/Nw
ΩR Ω
ΩΩR
d2 U
FWHM ~ 1/γz Nw
dΩd
 Θ
Figure 1.3: Spontaneous emission distribution at the resonant frequency and

on-axis
15
distribution. Explicit evaluation of the transverse electric field component terms
yields:

iωeµ0 Nw K̄ ω(1 − βz cos θ)
REx = √
g sinc Nw π −h
4 2πkw γ kw c

ω K̄ sin θ
× Jh−1 (±1)h exp [iωR/c ± i(h − 1)φ]
γkw c
= ∓iRE
gy
(1.15)
where the ± sign indicates a right (+) or left (-) handed undulator field. From
these expressions we see something that was not shown in the total energy spec-
trum in Eq. (1.11), namely that the phase of the field carries an azimuthal de-
pendence for harmonics h > 1. This appears in the term exp(iωR/c ± i(h − 1)φ)
where φ is the azimuthal coordinate of the radiation, and describes contours of
constant phase that are helical along the axis. This azimuthal dependence is not
reflected in the angular energy distribution since it is in the phase of the radiation
field and therefore vanishes when taking the complex-conjugated amplitude. But
here in the explicit representation it becomes readily apparent. It is independent
of the polarization of the fields2 , which is a fixed π/2 phase lag between the com-
ponents and is independent of φ. The helical phase at the harmonics of a helical
FEL was first explored by Sasaki and McNulty [26] because, as mentioned, light
that exhibits this property carries a non-zero value of OAM about the axis of
propagation. Sasaki and McNulty’s idea was thus to use the harmonic emission
in the helical FEL as a tunable source of OAM light. Though this method of
OAM generation differs significantly from the method of high-brightness OAM
generation at the first harmonic explored in this thesis, the concept of coupling to
2
The handedness of the polarization along z is, interestingly, opposite to the polarization of
the undulator. See Appendix B.
16
Figure 1.4: Emission at harmonics in a helical undulator is off-axis.
the azimuthal structures in the radiation field (and by extension, in the e-beam)
through the harmonics will play an important role in the HGHMG scheme to
generate dominant OAM in an FEL. This is explored in Chapter 4.
1.3 Principles of OAM light
The notion that light carries angular momentum - of any kind - dates back to
John Henry Poynting in 1909 [27]. In his paper he reasoned, by analogy of a
wave propagating on a rotating cylindrical shaft, that circularly polarized light
waves must carry angular momentum by virtue of the energy flux carried by the
rotating electromagnetic fields. Upon interacting with matter, he suggested, the
torque associated with the transfer of angular momentum could be measured. He
proposed a conceptual experiment composed of a series of quarter wave plates
suspended by a thin torsion fiber, each normal to an incident circularly polarized
17
beam of light. By measuring the optical power and the twist on the waveplates
as they converted the circular polarization into linear polarization, he posited
that the angular momentum of the light could be revealed, at least in principle.
Despite his insightful proposition, Poynting hedged his bets at the closing of his
paper, suggesting that, “... even with such multiplications [from multiple wave
plates], my present experience of light forces does not give me much hope that the
effect could be detected, if it has the value suggested by the mechanical model”.
In 1936, however, twenty-two years after Poynting’s death, the effect was in-
deed detected by Beth [28] (and independently confirmed by Holbourn later that
year [29]) in an experiment remarkably similar to what Poynting had proposed.
Beth’s landmark experiment quantitatively confirmed the angular momentum
flux Poynting had predicted. Further, the measured angular momentum was the
same as what would be expected from the quantum theory of light in which the
photon carries ~ units of angular momentum. This experiment is generally con-
sidered to be the first direct measurement of the spin angular momentum (SAM)
of the photon.
Inspection of the angular momentum of the EM fields, however, illustrates

that the spin does not necessarily account for all of the angular momentum of the
fields. Indeed, the notion that light can carry an associated component of orbital
angular momentum has received considerable attention since 1992 after Allen et
al [4] recognized that certain modes of light can carry discrete values of orbital
angular momentum per photon, and that such modes could be generated readily
in the lab. Since then, interest and research in the field has grow tremendously.
A beautiful and straightforward review of the recent progress is given by Franke-
Arnold et al [30].
To see how both types of angular momenta arise in the EM field, first consider
18
the general expression for angular momentum density,
j = x × p, (1.16)
where p is the linear momentum density (momentum per unit volume). Electro-
magnetic fields in a vacuum have a linear momentum density given by
p = 0 E × B (1.17)
where E is the electric field and B is the magnetic field. Note that, for infinite
transverse plane waves, the cross-product E × B is strictly in the direction of
propagation and so according to (1.16) there cannot be a component of angular
momentum x×(E×B) in the same direction. In reality, however, the fields are not
infinite plane waves, but are localized to a finite region of space (ie, wavepackets).
They therefore have small components in the direction of propagation, which gives
rise to an angular momentum along the same axis.
To illustrate, lets us consider simple monochromatic waves with spatial vari-

ation, h i
E(x, t) =Re Ẽ(x)e−iωt ,
h i (1.18)
−iωt
B(x, t) =Re B̃(x)e .
In free-space, the magnetic fields are related to the electric fields by ∇× Ẽ = iω B̃.
The total angular momentum of the fields is the integral of j over all space [31],
Z Z
3
Jrad = j d x = 0 x × (E × B)d3 x. (1.19)
The time-averaged total angular momentum of the fields can be written in terms
19
of the electric fields as
Z Z
0 X ∗ 0
Jrad = 3
Ẽj (x × ∇)Ẽj d x + Ẽ∗ × Ẽ d3 x
2iω j 2iω
(1.20)
=Lrad + Srad .
The first term Lrad is often interpreted as the orbital component of the angular
momentum by way of the operator −ix × ∇, while the spin component Srad
is given by the second term. This separation appears throughout the literature,
but should be taken with caution since there conceptual difficulties [32] associated
with this identification (see Sec 1.3.2). Nevertheless, it appears that the linear
momentum density of a finite beam can have components that give rise to an
orbital component associated with variation in the spatial field distribution.
The separation of the total angular momentum into components in (1.17) has
been known for some time. Allen’s insightful work, however, provided a simple
way to understand discrete values of OAM arise in specific (but not uncommon)
modes. To see how, let us examine the specific case of light propagating in the z
direction with the electric field written in terms of a spatial distribution function
u(x) and unit transverse polarization vector ê⊥ as
Ẽ(x) = E0 ê⊥ u(x)eikz . (1.21)
For any field distribution mode amplitude of the form
u(x) = û(r, z)eilφ , (1.22)
where φ is the azimuthal component, the time-averaged linear momentum density
20
is
" ! #
0 E02 ∗ 2il 2 ∂ |û|2 2
p= û ∇û − û∇û∗ + |û| − iσ êφ + 2ik |û| êz . (1.23)
4iω r ∂r
Note the first two terms which resemble the quantum-mechanical expression for
the expectation value of the linear momentum of a wavefunction [4]. The spread-
ing of the beam as it propagates is given by the radial momentum components,
while the momentum in the êz direction gives the longitudinal component along
the z-axis. Two terms appear in the azimuthal component of the momentum:
the first depends on the integer azimuthal mode number of the field, l, and the
second depends on the polarization σ. The former can be interpreted as the
orbital contribution, while the latter is the spin contribution. Right and left
circularly polarized light is given by σ = ∓1 while linear polarization is σ = 0.
It is now relatively easy to visualize how the angular momentum is manifested
for a finite field distribution: in combination with the longitudinal momentum,
the azimuthal component of the linear momentum describes a trajectory that is
helical about the axis of propagation. The wave front spirals like a corkscrew
(Figure 1.5). It is this azimuthal momentum component that can be imparted as
a torque upon interaction with matter. Several experiments based on this result
have demonstrated that optically power micro-machines are important applica-
tions of optical OAM.
With regard to the forthcoming examination of OAM light in FELs, let us

consider angular momentum in the context of paraxial waves. Paraxial waves
are waves for which the associated transverse dimensions are small compared to
the longitudinal distances over which the wave amplitudes change. Even in the
presence of high-gain field amplification, the light from an FEL is a prime example
of a paraxial wave as the light is well-directed and propagates primarily along
21
Figure 1.5: Helical phase and Poynting vector near the axis of propagation.
the longitudinal dimension. The field distribution amplitude u(x) is thus said
to be slowly-growing, and satisfies the required paraxial conditions |∂ 2 u/∂z 2 |
|k 2 u| , |2k∂u/∂z| , |∇2⊥ u|. The EM wave equation in the paraxial limit is then
∂
∇2⊥ u + 2ik u = 0. (1.24)
∂z
In this approximation, the total energy of the field is mostly due to the longitu-
R
dinal momentum, U ' c pz d3 x. From Eqs. (1.16) and (1.23), the z-component
of the angular momentum density measured from the axis is,
!
0 E02 ∂ |û|2
jz = 2l |û|2 − σr . (1.25)
4ω ∂r
Integrating over all space and dividing by the total energy of the paraxial fields,
we obtain the ratio Z
Jz 1
= jz d3 x
U U
R 2

2l |û|2 − σr ∂|û|
∂r
d2 x⊥
' (1.26)
2ω |û|2 d2 x⊥
R
1
= (l + σ) .
ω
22
By recognizing the total energy of a single photon as U = ~ω, we identify the
distinct contributions of the orbital and spin components of angular momen-
tum, with the former component appearing in discrete values according to the
azimuthal mode number; Lrad = l~. The latter component is given by Srad = ∓~
for circular polarization [4].
Note that a similar, exact expression for non-paraxial beams has been ob-
tained elsewhere [33, 34], but does not display the simple separation of the orbital
and spin angular momentum components as obtained above. The non-paraxial
result does show, however, that the manifestation of orbital angular momentum
is not an artifact of the paraxial approximation, even though in the non-paraxial
case the simple separation between the orbital and the spin parts is obscured.
1.3.1 Laguerre-Gaussian Paraxial Modes
The fact that cylindrically symmetric paraxial waves appear to have well-defined
values of both OAM and SAM was first pointed out in the context of Laguerre-
Gaussian (LG) optical modes by Allen et al [4]. Laguerre-Gaussian modes rep-
resent a class of exact higher-order solutions to the paraxial wave equation, and
are given by
s √ !|l|
2r2

2p! w0 r 2 |l|
up,l (r, φ, z) = (−1)p Lp
(p + |l|)! w(z) w(z) w(z)2
r2 ikr2

−1 z
× exp ilφ − + − i(2p + l + 1) tan
w(z)2 2Rc (z) zR
(1.27)
where zR = kw02 /2 is the Rayleigh length, w0 is the waist size, Rc (z) = (zR2 +z 2 )/z
p
is the radius of wavefront curvature and w(z) = w0 1 + z 2 /zR2 is the spot size
along z. The radial mode number is given by the integer p, and the functions Llp
23
l=1 l=2 l=3
Figure 1.6: Phase contours for OAM modes at the waist show the multivalued
helical geometry.
are Associated Laguerre polynomials. Note that for p, l=0, the simple Gaussian
distribution is retrieved. FIgures 1.7 and 1.8 display the intensity and phase of
the lowest order modes at the waist. The modes are mutually orthogonal and
R
here are normalized such that up,l up0 ,l0 d2 x⊥ = δp,p0 δl,l0 πw02 . The well-defined
angular momentum carried by the LG modes is a direct result of the fact that
they have an azimuthal dependence of the form (1.22) and satisfy the paraxial
wave equation in (1.24). The angular momentum/energy ratio for LG modes is
therefore given by Eq. (1.26).
The ubiquity of the LG modes in the OAM literature, and in this dissertation,
stems from the fact that they are a complete and orthogonal set of functions. Any
arbitrary light beam can be expressed as a complete sum of LG modes. Indeed,
that the amplified light emitted from an FEL can be decomposed into a set
of LG modes motivates the modal description of the 3D FEL presented in the
following chapters. This provides a direct method by which the OAM content of
the high-gain FEL can be directly examined.
24
Figure 1.7: LG mode intensity distributions.
25
Figure 1.8: LG mode phase distributions.
26
Figure 1.9: Electric field amplitude contours for several LG modes in the x − z
plane.
27
1.3.2 Understanding optical OAM
Before moving forward, it is important to comment on certain issues associated

with orbital angular momentum in the electromagnetic fields. One of the most
obvious problems is the assignment of an orbital angular momentum component
to a single photon via the classical analysis. Quantum mechanically, the mass-
less photon carries only spin angular momentum with value ±~ as demanded
by Lorentz invariance. Accordingly, the separation of the total angular momen-
tum into the components Lrad and Srad may seem unphysical, (and has been
regarded as such in some literature). However, while it is true that the separa-
tion is not Lorentz invariant, the notion that it is unphysical it is predicated on
the assumption that both Lrad and Srad are precisely the quantum mechanical
operators of angular momentum for a spin 1 particle, which they are not [32].
A fully quantum-mechanical treatment of the EM field shows that Lrad and Srad
do not commute, nor do they generate orbital and polarization rotations of the
transverse EM field. They are in fact well-defined and separately measurable,
but they do not have the precise meaning of orbital and spin angular momentum
in the familiar quantum-mechanical sense (hence the italics here).
The question remains, then, what Lrad and Srad physically represent, partic-
ularly in paraxial instance shown in Eq. (1.26) where they appear well-defined
and distinct. Their respective meanings become clearer when one considers the
difference between OAM carried by an optical mode and OAM carried by a single
photon. Clearly, the component of Lrad along the direction of propagation of a
single photon is zero, as it must be when measured from a position in which the
transverse momentum is zero. As we see in equations (1.23) and (1.25), however,
there is clearly an azimuthal component of the linear momentum of the field that
gives rise to two distinct orbital and spin components of angular momentum.
28
They arise because both quantities are calculated from the axis of the optical
beam – the ‘center of mass’ of the system in the transverse plane. This optical
axis does not correspond to the propagation axis of a single photon measured on
a detector screen. Note in Figure 1.7 that all modes with |l| > 0 have zero inten-
sity on-axis. The intensity is the quantum-mechanical probability for the photon
to be at a certain place, so there are no photons on-axis. So if the OAM value
of a single photon were to be measured about its off-axis position at the detector
screen, the value would be zero. But not all the photons arrive at the same posi-
tion - they are distributed about the annular ring that encircles the propagation
axis of the entire beam. Therefore, if the OAM content of the |l| > 0 mode is
measured a single photon at a time, one would expect to measure a non-zero value
of OAM about the mode propagation axis due to the azimuthal momentum con-
tent of the constituent off-axis photons. In this sense, calling these pieces orbital
and spin makes some sense from the standpoint of the optical modes described
by the classical wave theory, wherein it is the sum of all the photons in the mode
that conspire to yield a finite value of OAM about the beam axis.
From this it follows that the OAM of an optical mode is fundamentally dis-
tinct from the SAM, even though they share common features. The total spin
component is always independent of the choice of axis, and is therefore said to
be intrinsic to the light beam [35]. The same can be said of the orbital angular
momentum component, as long as the direction of the axis is taken so that the
total transverse momentum of the light beam is zero [36]. Otherwise, the OAM is
an extrinsic feature of the light and depends on the measurement axis. Because
of this dual nature OAM is sometimes referred to as quasi-intrinsic. To further
illustrate, consider a scenario where the beam interacts with, or passes through,
an element that presents a finite aperture. If the aperture is symmetric about
the beam axis, the total transverse angular momentum is zero and the ‘intrinsic’
29
aspect of both angular momenta discussed above is recovered – neither compo-
nent depends on the chosen axis. If the aperture is asymmetric about the beam
however (off to one side), the transverse momentum is non-zero when integrated
over the aperture. While in this case the spin component is still independent of
the axis, the orbital component is not. Only about the original beam axis does
calculation of the orbital component still yield the intrinsic value, but at any
other arbitrary position this value depends on the axis, and hence the OAM is
considered extrinsic. This scenario is analogous to the OAM measurement with
a single photon.
This distinction between SAM and OAM can be made manifest in their re-
spective interactions with matter. For instance, a dielectric particle trapped in
an optical beam that carries both SAM and OAM is moved dynamically in two
different ways. The SAM causes the particle to spin about its own axis, while the
OAM causes the particle (which is optically trapped off-axis in the intensity ring
of the OAM mode) to orbit about the optical beam axis [35]. This illustrates not
only that the OAM is a physically observable quantity, but that it can be used
via micro-mechanical transduction to perform work [37].
30
CHAPTER 2
Dielectric Eigenmode Expansion of High-Gain

3D FEL Equations
In Section 1.2 the basic principles of spontaneous emission from an electron in

an undulator are presented. While the analysis demonstrates essential features
of the FEL concept, (such as the tunable wavelength) it is only part of the story.
High-gain FELs uphold their namesake by operating as powerful lasers in which
the radiation emitted by the electrons is amplified by a feedback instability. In
contrast to purely spontaneous emission, the radiation emitted by the electrons
in an FEL feeds back onto the electrons in the beam, which then re-radiate the
light. This cycle repeats, and during the process the e-beam can actively affect
the radiation distribution, and vice versa. The collective evolution of both the
e-beam and the FEL light (also called the signal field) must be therefore be taken
into account in the description.
In the 3D FELs, the feedback mechanism responsible serves not only as an

amplification process, but also creates a phenomena known as gain-guiding in
which the coherent interaction introduces an inward curvature in the phase front
of the light that refracts it back towards the lasing core of the e-beam [38, 39, 40].
During the gain process the e-beam can behave like a guiding structure that
suppresses diffraction, reduces transverse power losses and enhances the EM field
amplification. For a sufficiently long interaction length, the guided EM field
31
eventually settles into a propagating, self-similar eigenmode of the FEL system
which has a fixed transverse profile and grows exponentially in amplitude as a
function of longitudinal coordinate z.
Several different approaches currently exist for describing the 3D FEL lin-
ear gain process. Analytic derivations of guided FEL eigenmodes have been
performed previously by direct derivation of the eigenmode equations from the
coupled Maxwell-Vlasov equations [41, 42, 43]; from solutions to the Maxwell
equations using Green’s functions [44]; or from expansion of the FEL signal field
in terms of eigenmodes of a hollow, conducting-boundary waveguide [45], eigen-
modes of a step-index fiber [40], and free-space paraxial waves [46]. Such analytic
descriptions of the 3-D FEL equations, particularly those that exploit radiation
mode expansions to find the FEL eigenmodes, have certain utility for providing
physical insight into the character of the radiation. Since the e-beam operates
simultaneously as an optical source and as a waveguide structure in a high-gain
FEL, an expansion mode description of the FEL light permits investigation of
the amplification, guiding and coupling efficiency of the individual basis modes.
In general, the choice of analytical model and expansion basis depends on

the FEL geometry and operational characteristics. The fields inside the e-beam
are optically guided during high-gain, so a guided expansion basis is a natural
choice for the optical field. But the expansion may be plagued by inefficiency in
describing the EM field structure if the guiding characteristics of the e-beam are
markedly different from those of the virtual waveguide that yields the particular
basis set. (The waveguide is referred to as ‘virtual’ because no such waveguide is
assumed to be present in the physical system.) In Ref [45] an analysis is presented
for calculating the gain-guided FEL supermode by means of a field expansion
using eigenmodes of a hollow waveguide with conducting walls. This method can
32
be useful even when no such physical boundary exists and the optical beam is
guided only by the e-beam. Clearly, in this example analysis and in general, this
approach is legitimate if the presence of the artificial boundary does not affect the
physical result. The virtual waveguide dimensions must be taken wide enough so
that the fields in the actual problem vanish at the virtual boundaries, but not so
wide that many modes are required to describe the radiation field such that the
calculation is inefficient or does not converge.
An alternate but still general approach, which is central to this work and
described in the next few sections, uses eigenmodes of a smooth, transversely in-
homogeneous dielectric waveguide to model the field and thus avoids the undesir-
able influence of artificial boundary conditions [47]. This approach also provides
flexibility in the form of the expansion basis, which is determined by the refrac-
tive index distribution. Of particular interest in any expansion mode approach
(particularly with respect to the discussion in the last chapter) is the coupling
to the naturally occurring Hermite-Gaussian (HG) or Laguerre-Gaussian (LG)
modes that describe free-space wave propagation in the paraxial limit. This con-
nection is useful for relating the FEL modes to free-space propagating modes,
which are also present before and after the undulator. With this connection one
may, for example, compactly and clearly describe the input radiation needed for
efficiently seeding the FEL. One may also robustly describe the propagation of
the FEL light after the light exits the undulator, allowing a clear understanding
of the demands made on downstream optics and on the expected mode structure
in the FEL, or on the experiments themselves. More importantly, however, this
basis can be used to examine precisely how such individual modes couple and are
amplified in an FEL, thereby providing insight into how one may select specific
modes (ie, OAM modes) for amplification. That being said, the details of LG
modes in a high-gain FEL are left for the next chapter. Here, a general formalism
33
is provided that describes the FEL in terms of an arbitrary basis set. The effects
of diffraction, space-charge, emittance and energy spread are all included, where
applicable.
Strong optical guiding in an FEL occurs in the exponential gain regime, when
the focusing effects in the source e-beam tend to balance the natural diffraction
of the generated EM radiation. Obviously, an expansion set consisting of guided
modes can be used to accurately describe the field during long sections of gain-
guiding, over which the otherwise free-space fields have many Rayleigh lengths to
diffract but are instead guided. Such an expansion can also accurately describe
short sections of dynamic profile evolution (like the FEL start-up period in a
high-gain amplifier). But guided modes may not efficiently capture (in the sense
of providing a compact description) the field behavior during long sections of
weak guiding when diffraction is dominant over guiding, like during saturation or
during long lengths of low-gain. The regimes of validity for a description using
guided modes may be estimated by inspection of the relative values of the optical
Rayleigh length zR = 2kσx2 (assuming for the moment that the field has the same
characteristic radius as the e-beam, σx ), the e-folding radiation power gain length
LG , and the overall length of the FEL interaction length L. In the guided mode
expansion here, we restrict our attention to high-gain (L1D L) FEL systems,
prior to the onset of saturation, for which the weakly diffracting (LG L < zR )
or strong-guiding (LG < zR < L) conditions are valid, and the guided mode
expansion description can be used efficiently.
2.1 Outline of Chapter
In this chapter, a virtual dielectric expansion description of high-gain FEL radi-

ation is presented. The excitation of the slowly-growing mode amplitude coeffi-
34
cients in the presence of a local source current is derived. The e-beam is linearized
and in the cold-beam fluid approximation, a set of coupled excitation equations,
similar to Ref [47], is obtained. The coupled equations are then recast by a simple
transformation into a matrix equation for solutions of the FEL supermode as a
function of parameters for diffraction, energy detuning, longitudinal space-charge.
The effects of energy spread and emittance in the e-beam are also included in
subsequent sections, by simultaneous solution of Vlasov and Maxwell’s equations.
The theory presented here provides a general framework for more detailed ana-
lytic and numerical studies performed in the following chapters. There, results
that highlight the advantages of the dielectric eigenmode expansion approach are
presented and analyzed.
2.2 Dielectric Waveguide Expansion Modes in the Pres-

ence of a Local Source
The radiation fields E(x, t) and B(x, t) emitted by the transverse currents in
the FEL are localized to a narrow forward cone near the axis. They can there-
fore be expanded in terms of transverse radiation modes of a guiding structure,
with slowly-growing amplitudes that vary only as a function of the axis and
e-beam propagation coordinate, z. The general paraxial field expansion for time-
h i
harmonic fields E(x, t) =Re Ẽ(x)e−iωt is [47],
X
Ẽ⊥ (x) = cq (z)Ẽ ⊥q (x⊥ )eikzq z
q
(2.1)
X
B̃⊥ (x) = cq (z)B̃⊥q (x⊥ )eikzq z
q
35
where B̃⊥q = (µ0 /Zq )êz × Ẽ ⊥q is the magnetic field expansion mode, kzq is the
q th mode axial wavenumber, k = ω/c is the free-space wavenumber, and the
p
impedance is Zq = (k/kzq ) µ0 /0 for TE modes. In general, the sum extends
over both the guided and the cutoff modes, and the modes form a complete set
[48]. For the purposes of investigating the coupling of different mode structures,
both in the radiation field and in the electron beam, it is convenient to rewrite
the basis modes Ẽ ⊥q and B̃⊥q in terms of amplitudes that carry the proper units
and a unitless expansion function, designated by uq . This will facilitate a simpler
description of the e-beam later, in order that we will not have to concern ourselves
with mixing units in the expansion. We define:
Ẽ ⊥q (x⊥ ) ≡ E0 uq (x⊥ ) = E0 uq (x⊥ )êq , (2.2)
where êq is the polarization unit vector of the E-field and E0 is the amplitude,
taken to be real. This definition will be used throughout. The mode expansions
are then written as
X
Ẽ⊥ (x) = E0 cq (z)uq (x⊥ )eikzq z
q
(2.3)
X 1
B̃⊥ (x) = E0 µ0 cq (z) êz × uq (x⊥ )eikzq z .
q
Zq
The equation for the time-harmonic fields in the presence of a source of

h i
charged particles with current density J(x, t) =Re J̃(x)e−iωt is
∇2 Ẽ + k 2 Ẽ = −iωµ0 J̃. (2.4)
The transverse charge density gradient ∇ρe of the e-beam in an FEL is typically
small and is neglected in the approximation that |c2 ∇ρe | |∂J/∂t|. Plugging in
36
the electric field from Eq. (2.3), we can further simplify with the paraxial approx-
2
imation on the amplitudes, namely that |d2 cq (z)/dz 2 | kzq cq (z). Maxwell’s
equation for the field (2.4) in terms of the mode set becomes

X
ikzq z 2 2 2 ∂
E0 e ∇⊥ + k − kzq + 2ikzq cq (z)uq (x⊥ ) = −iωµ0 J̃. (2.5)
q
∂z
For a given choice of expansion basis, the field expansions in Eqs. (2.3) are
inserted into Eq. (2.4) to obtain a differential equation for the amplitude co-
efficents cq (z) [45]. Here, we explore the case where the expansion mode uq is
an eigenmode of a dielectric waveguide with arbitrary refractive index nI (x⊥ ).
It is straightforward then to apply the analysis to eigenmodes have the same
transverse functional form as free-space paraxial modes evaluated at the waist,
thus establishing a direct link between the expansion modes of the FEL system
and modes that have a well-defined value of OAM (see Chapter 3). Assum-
ing small transverse variation in the refractive index of the virtual waveguide,
(∇n2I )⊥ k, the eigenmodes can be regarded as dominantly transverse, and the
dielectric eigenmode equation is
∇2⊥ uq (x⊥ ) + [nI (x⊥ )2 k 2 − kzq

2
]uq (x⊥ ) = 0. (2.6)
This dielectric eigenmode equation is used to eliminate the transverse Laplacian

term ∇2⊥ . The excitation equation for the mode q in the presence of a local source
current is given by,
kµ0 ce−ikzq z
ZZ
d X
cq (z) = − J̃⊥ (x) · u∗q (x⊥ )d2 x⊥ − i Kq,q0 cq0 (z)eiz(kzq0 −kzq )
dz 2kzq E0 huq |uq i q0
(2.7)
37
where
k 2 huq |nI (x⊥ )2 − 1|uq0 i
Kq,q0 = . (2.8)
2kzq huq |uq i
RR
Here the bra-ket notation is borrowed for brevity, short for the integral uq ·
u∗q0 d2 x⊥ = δq,q0 huq |uq i, where the modes are taken to be orthogonal. The term
Kq,q0 quantifies the extent to which the expansion modes “overlap” in the imag-
inary dielectric. It can be thought of as representing the virtual polarization
currents that are necessarily subtracted when using eigenmodes of a dielectric
waveguide, but practically speaking, it is a useful mathematical adjustment that
enables one to solve directly for the mode amplitudes cq (z) given a choice of
modes and refractive index that satisfy Eq. (2.6). The expansion modes uq are
normalized such that the mode power given by
Z Z
1 ∗ 2
Pq δq,q0 = Re [Ẽ ⊥q (x⊥ ) × B̃⊥q0 (x⊥ )] · êz d x⊥
2µ0
(2.9)
E02

huq0 |uq i
= Re .
2 Zq 0
|cq (z)|2 Pq .
P
The total power in the field is P(z) = q
Equation 2.7 describes the response of the dielectric mode amplitudes to a

harmonically varying current J̃⊥ . What remains is to define this current in the
context of the FEL e-beam.
2.3 Cold Electron Beam Fluid Description
Several different approaches exist for describing the electron beam in its inter-
action with the undulator and electromagnetic fields during the FEL process.
The choice of model is guided by the particulars of the FEL system. In the case
38
of a cold e-beam (negligible energy spread1 ) with vanishing transverse betatron
motion2 , a linear plasma fluid model [45, 47] provides insight into the physical
3D system through transparent coupled evolution equations which are developed
in this section. This description provides a straightforward way of modeling the
simultaneous evolution of both the FEL signal field and the electron beam from
the onset of the interaction throughout exponential gain. In the context of the
modal expansion pursued here, it also provides a direct and simple way to specify
the initial conditions on the amplitudes of the coherent EM fields and the density
and velocity modulations for each mode. This includes the effective initial ampli-
tudes due to shot noise that can be amplified, (found explicitly for each mode in
Section 2.6), which serves as an appealing alternative solution to the well-studied
grand initial value problem [49] in the cold, parallel beam limit. This way, one
can investigate how individual modes of interest can develop and evolve.
The relativistic e-beam in an FEL experiences transverse oscillations, or wig-

gling, driven by the interaction with the undulator and electromagnetic fields.
This periodic motion leads to an axial ponderomotive force that modulates the
longitudinal electron velocity which, in a high-gain FEL, then leads to longitudi-
nal density bunching that significantly amplifies the field. In general, the change
in velocity of charged particles in the presence of electric and magnetic fields is
given by the Lorentz force equation:
d
γmv = −e [E + v × B] (2.10)
dt
This vector expression can be used to calculate both the transverse wiggling
velocity and the longitudinal velocity modulation due to the field interaction.
1
Investigated in Section 2.8
2
Investigated in Section 2.9
39
The transverse magnetic field that drives the transverse beam motion is the
sum of the undulator field and the magnetic field of the EM wave:
" #
h i X
B⊥ = Re B̃⊥w e−ikw z + Re cq (z)B̃⊥q (x⊥ )ei(kzq z−ωt) (2.11)
q
where B̃⊥w is the transverse magnetic field amplitude vector of the undulator.
Under the justified assumption that the change in the transverse velocity is large
compared to the change in the relativistic energy γ, the lowest order components
of the transverse electron velocity driven by these magnetic fields can be written
from Eq. 2.10 similarly as the sum,
" #
X
−ikw z i(kzq z−ωt)

v⊥ = Re ṽ⊥w e + Re cq (z)ṽ⊥q (x⊥ )e (2.12)
q
where ṽ⊥q is the transverse electron velocity due to the Lorentz force of the q th
mode of the signal field and ṽ⊥w = (−icK/γ)êz × êw is the transverse electron
velocity due to the magnetic undulator field, where êw is the unit polarization
vector of the undulator field (See Appendix B) and
e|B̃⊥w | √
K= = 2K̄ (2.13)
mckw
is the so-called undulator parameter, here also written in terms of the rms pa-
rameter K̄ which is also used interchangeably throughout.
The interaction between the transverse electron beam motion and the fields of
both the undulator and the EM signal results in an axial ponderomotive field that
drives the velocity modulation. Let us assume that, to first-order, the z-velocity
40
distribution can be written in general as
vz (x, t) = v0 1 + Re δv (x)e−iωt

(2.14)
where v0 = βz c is the average longitudinal velocity of the entire beam and δv is

the small velocity modulation at the signal frequency ω. Longitudinal variations
in the velocity like those found in planar undulator systems are ignored for the
moment. The time rate of change of the relativistic momentum in the z direction
from (2.10) is mc d(γβ z)
= mc(γ + βz2 γz2 γ) dβ = γγz2 mc dβ
p
dt dt
z
dt
z
where γz = 1/ 1 − βz2
is the axial relativistic energy factor. The relativistic force equation for the axial
velocity is then γm dvdtz = − γe2 [E+v×B]z . The ponderomotive field near resonance
z hP i
i(kzq +kw )z−ωt
is therefore the sum (v × B)z = Re c
q q (z)Ẽpm,q (x⊥ )e where the
∗ ∗
ponderomotive mode is Ẽpm,q = 21 [ṽ⊥q × B̃⊥w + ṽ⊥w × B̃⊥q ] · êz . With (2.14) the
evolution of the longitudinal velocity modulation is given by,
∂ ω e J˜z X
δv (x) − i δv (x) = − 2 2 [ + cq (z)Ẽpm,q (x⊥ )ei(kzq +kw )z ] (2.15)
∂z v0 γγz mv0 iω0 q
Longitudinal space-charge modulations are captured in the longitudinal current

density term, J˜z /iω0 . It has been assumed that the axisymmetric e-beam size
is large compared to the modulation wavelength in the e-beam frame, σx >
γz λ/2π, so that the fringing fields and the transverse space-charge effects are
neglected. The effects of fringing fields due to a transverse variation can be
included for a more complete description [50, 51], but will be neglected here
since we are concerned primarily with scenarios where space-charge is dominantly
longitudinal.
In addition to the velocity modulation δv , the electrons are also microbunched
41
in density. Similar to (2.14), the density is described in a linear model as
n(x, t) = n0 f (x⊥ ) + Re δn (x)eiω(z/v0 −t)

(2.16)
where n0 is the electron density, f (x⊥ ) is the unmodulated transverse density

profile of the e-beam and δn (x) is the small, slowly-varying 3D density modula-
tion. The total vectorial e-beam current density depends on the particle density
as
J(x, t) = −en(x, t)v(x, t) (2.17)
Thus, to lowest order, the time-harmonic modulated component of the longitu-

dinal current density from (2.14) and (2.16) is therefore,
J˜z (x) = −ev0 n0 f (x⊥ )δv (x) + δn (x)eiωz/v0 .

(2.18)
Now, if the transverse divergence of the current density modulation is assumed

small ∇⊥ · J˜⊥ ∂ J˜z /∂z, the continuity equation ∇ · J = e∂n(x, t)/∂t can be
written as
∂ ˜
Jz = −iωen0 δn (x)eiωz/v0 , (2.19)
∂z
which gives f (x⊥ )∂δv /∂z = −eiωz/v0 ∂δn /∂z. The continuity equation shows that
a change in the density modulation relates directly to a change in the velocity
modulation, and vice versa. Now, by combining Eqs. (2.19), (2.18) and (2.15)
the density bunching evolution is a second order differential equation. We obtain
a result similar to that in Refs. [45, 47] with transverse fields for the density
modulation evolution during the FEL interaction:
∂2

2 0 f (x⊥ )
X
+ θ 2
p f (x⊥ ) δn (x) = iθ p (kzq + kw )cq (z)Ẽpm,q (x⊥ )e−iθq z (2.20)
∂z 2 n0 e q
42
where the longitudinal relativistic plasma wavenumber on axis is
ωp2 e2 n0
θp2 = = . (2.21)
v02 γγz2 0 mv02
The parameter θq = ω/v0 −(kzq +kw ) is the characteristic detuning for a given
mode q. The resonant condition is obtained when θq = 0 which, for a relativistic
k
e-beam γz 1 gives kw = 2
2γzR
+ k − kzq . In the 1D scenario the wavenumbers of
the modes are degenerate kzq = k and the detuning parameter is θ0 = k/2γz2 −kw .
The standard 1D FEL resonance condition is thus recovered with θ0 = 0:
2
k = 2γzR kw (1D resonance) (2.22)
with γ 2 = γz2 (1 + K 2 /2).
With Eqs. (2.7) and (2.20), the FEL instability can be summarized as follows:
the transverse e-beam motion, in combination with the signal and undulator
fields, generates a ponderomotive force that longitudinally modulates the e-beam
(2.20). This modulation, wiggling back and forth through the undulator as a
transverse current, excites the signal fields (2.7), which then act back on the
longitudinal modulation via the ponderomotive force.
2.4 Coupled Excitation Equations
Equation (2.20) describes the evolution of the 3D density modulation δn (x) in the
presence of the ponderomotive fields which arise due to the resonant interaction
of the e-beam with the fields in the undulator. The question of how the e-beam
density modulation is related to the FEL signal fields is answered by inspection
of the transverse currents. With Eq. (2.17), a little algebra shows that the
43
transverse component of the current density that excites the signal wave in (2.7)
near resonance is written in terms of the longitudinal density modulation in (2.16)
as
1
J̃⊥ (x) = − en0 δn (x)ṽ⊥w eiz(ω/v0 −kw ) . (2.23)
2
With this definition both Eq. (2.20) and (2.7) can be simplified by expressing
the density modulation as a sum over the expansion basis functions and slowly-
varying amplitudes,
X
δn (x) = aq (z)uq (x⊥ ). (2.24)
q
Plugging this expression into Eqs. (2.23), (2.20), and (2.7), and integrating over
the transverse dimensions in Eq. (2.20), the orthogonality of the eigenfunctions
can be used to simplify the equations and to write both coupled equations in
terms of the slowly growing amplitudes cq (z) and aq (z) of the signal field and
the density modulation, respectively. This yields the coupled FEL excitation
equations for a cold beam:
d X
cq (z) = −iξq ĝq∗ eiθq z aq (z) − i Kq,q0 cq0 (z)ei(θq −θq0 )z
dz q0
(2.25)
2
d X 1 X
2
aq (z) + θp2 Fq,j aj (z) = − ∗ Qq,q0 cq0 (z)e−iθq0 z ,
dz j
ĝq ξq q0
where ξq = Kn0 ek/4γ0 E0 kzq . The first equation describes the excitation and
evolution of the mode amplitude cq of a dielectric waveguide eigenmode due
to the transverse wiggling motion of the bunching current throughout the FEL
interaction. The second equation describes the excitation and evolution of the
density bunching amplitude. The 3D coupling between the e-beam and the signal
44
field is given by the mode coupling coefficient for TE modes (Appendix B):
(kzq0 + kw )2 K 2
Qq,q0 = JJ ĝq0 ĝq∗ θp2 Fq,q0 . (2.26)
8kzq γ
where JJ= [J0 (kK 2 /8γ 2 kw ) − J1 (kK 2 /8γ 2 kw )]2 is now included for a planar un-
dulator (JJ=1 for a helical undulator geometry) where J0 and J1 are the first
and second Bessel functions. The factor ĝq∗ = (êz × êw ) · ê∗q is the polarization
alignment factor that measures the relative alignment of the transverse electron
motion in the undulator (êz × êw ) with the E-field mode polarization direction,
êq . Maximum coupling from polarization is obtained when ĝq = 1 (See Appendix
B). This parameter is somewhat superfluous for the high-gain FEL since the sig-
nal field generated by the beam will by definition be polarization matched to the
direction of electron motion. But tuning this parameter may be useful in model-
ing modulator scenarios like those described in Chapter 5 where the high-power
input seed drives microbunching in the e-beam and its power remains fixed.
The term under the sums, Fq,q0 , is referred to as the beam profile overlap coef-
ficient and quantifies the spatial overlap of the e-beam profile with the expansion
modes q, q 0 :
huq |f (x⊥ )|uq0 i
Fq,q0 = . (2.27)
huq |uq i
This term determines the magnitude of the e-beam coupling to the radiation
modes, and the density mode coupling in the longitudinal plasma oscillations.
The e-beam mode coupling parameter in (2.26) is closely approximated in the
limit k ' kzq0 kw for dominantly transverse modes by,
kK 2
Qq,q0 'JJθp2 Fq,q0
8γ 2
(2.28)
'QFq,q0
45
Where Q is the degenerate value of Qq,q0 in the 1D limit. Near resonance it is
directly related to ρ, the well-known Pierce parameter often used in FEL theory:
2
3 Q JJ Kθp γz
ρ = = . (2.29)
(2kw )3 32 γkw
The effect of the longitudinal space-charge in the beam is calculated in the

second term of the second equation, and takes into account the effects of longi-
tudinal plasma oscillations (Langmuir waves) in the FEL interaction through the
longitudinal plasma wavenumber θp = ωp /v0 of a uniformly distributed electron
beam profile used in a 1D model. If the quarter-plasma oscillation wavelength
is much longer than the total interaction length L this quantity can be safely
neglected. Otherwise, collective effects can affect the dynamics and this term
should be included for θp L > π/2.
The energy modulation in the e-beam can be modeled by the modal expansion
in much the same way as the density modulation in (2.24);
X
δv (x) = avq (z)uq (x⊥ )eiωz/v0 . (2.30)
q
where, like aq (z), the velocity mode amplitude avq (z) is slowly growing along z. In
the absence of space-charge effects this definition results in a simple expression
for the excitation of a velocity modulation due to the signal field, from (2.25):
d v Q
aq (z) = −iĝq cq (z)e−iθq z . (2.31)
dz ξq k
Note that there is no sum over modes in (2.31); the signal field modes couple
directly to their velocity mode counterparts. This is not necessarily the case for
the excitation of the density modes, many of which may be excited simultaneously,
46
depending on the coupling. The velocity expansion in Eq. (2.30) results in a
relation between the density modulation amplitudes and the velocity modulation
amplitudes via (2.19),
v
daq0 (z)

d X ω v
aq (z) = − Fq,q0 + i aq0 (z)
dz q 0
dz v0
(2.32)
ω X
' −i Fq,q0 avq0 (z).
v0 q 0
Finally, we see that the velocity modulation generated during the FEL inter-
action translates to an effective energy spread on the e-beam. By time-averaging
the longitudinal relativistic factor (1 − |vz (x, t)|2 /c2 )−1 for γ 1, the energy
spread is written in terms of of the velocity modulation as
∆(γ 2 )

' 2γz4 |δv (x)|2 . (2.33)
γ2 t
Averaging over the beam (see Eq. 2.37) gives the result in terms of modal am-
plitudes;
∆(γ 2 ) 2γz4 X v v

∗
= aq (z) aq0 (z) Fq,q0 huq |uq i (2.34)
γ2 Ab q,q0
R
where Ab = f (x⊥ )d2 x⊥ is the transverse e-beam area.
The initial conditions for the coupled equations in Eqs. (2.25) specify the
operating characteristic of the FEL. For example, when operating as a single-pass
amplifier (seeded FEL) there is negligible initial density and velocity modulation
(aq (0), avq (0) = 0) and the initial seed field is non-zero (cq (0) 6= 0). For a self-
amplified spontaneous emission (SASE) FEL, the amplified shot noise can be
related to the pre-bunching conditions (aq (0) 6= 0, avq (0) 6= 0) and the input
signal field vanishes (cq (0) = 0). Initial density bunching due to shot noise is
47
explored in Section 2.6.
In the preceding sections, the linear cold e-beam fluid model has been in used
in conjunction with an expansion of the EM radiation fields to obtain a 3D de-
scription of an FEL system. During high-gain operations the virtual dielectric
waveguide EM modes provide a convenient working basis to describe the ampli-
fication and self-guiding of the the radiation field over many diffracting Rayleigh
lengths of an otherwise free-space field. The basis modes satisfy (2.6) and can be
freely chosen to yield different functional forms, allowing flexibility to choose a
basis optimized for a given FEL system. The closer that the choice of the virtual
dielectric waveguide distribution is to the real solution to the given FEL system,
the fewer the number of modes that will be required to converge the equations
in Eqs. (2.25) to the correct solution for the FEL. This can be found either
iteratively or by an educated guess given the systematic constraints.
2.5 Bunching Factor
One of the most useful parameters used to characterize both the e-beam and the
evolution of the FEL interaction is the electron beam bunching factor. It is the
normalized density bunching amplitude at a given frequency and quantifies the
extent to which the electrons have been collectively bunched in the ponderomotive
phase bucket. Density bunching is induced by the energy modulation through
the continuity equation and by longitudinal dispersion. Since the induced energy
modulation is also accompanied by an increased energy spread that determines
the available gain bandwidth, the bunching factor is a telling measure of the in-
teraction status, providing a guidepost for marking the end of the linear high-gain
regime when the bunching approaches unity. It can also be measured indirectly
experimentally from the signatures of radiation at, or near, the microbunching
48
frequency [52, 53, 54, 55].
A beam composed of Ne discrete particles has bunching factor at the wavenum-

ber k defined as
Ne
1 X
b= e−iksj (2.35)
Ne j=1
where s = z − v0 t is the co-moving coordinate. In the discrete model, if all the

electrons have the same position in the phase bucket, the bunching factor ampli-
tude is unity and the beam is maximally bunched. In practice this limit is never
reached; the electrons do not occupy precisely the same position in phase. They
do become strongly bunched during high-gain, however. The bunching factor
exponentially approaches unity until tapering off at saturation at a maximum of
|b| ' 60 − 70%.
In the linear plasma fluid model the bunching factor at a single frequency is
maximum at |b| = 1/2 in order to satisfy the physical constraint that the density
distribution stays positive everywhere. Nonlinear and higher harmonic terms in
the evolution equations must be included near saturation, but in the regime |b|
1/2, the linear equations adequately model the exponential evolution. The overall
bunching factor is obtained by averaging over the continuous density distribution
49
in (2.16):
* Ne
+

2 1 X
−ik(sj −sj 0 )
|b(z)| = 2 e
Ne j,j 0 =1
" Ne
#
1 X
e−iksj eiksj0

= Ne +
Ne2 j6=j 0
(2.36)

Ne (Ne − 1)
Z Z
1 ∗
= 2 Ne + δn (x)dx⊥ δn (x)dx⊥
Ne 4A2b
Z Z
1 X
' 2
aq (z)aq0 (z) uq (x⊥ )dx⊥ u∗q0 (x⊥ )dx⊥ ,
∗
4Ab q,q0
where Ne 1 is assumed. The averaging over is beam is defined as a normalized

integral
1
Ra R
lima→∞ −a
2a
(. . . )F (x1 , x2 , . . . , xn , s)dx1 dx2 . . . dxn ds
h. . . i = 1
Ra R , (2.37)
lima→∞ 2a −a F (x1 , x2 , . . . , xn , s)dx1 dx2 . . . dxn ds
where F (x1 , x2 , . . . , xn , s) is the e-beam distribution function of independent vari-

ables x1 , x2 , . . . , xn , and s. In a cold, parallel beam F = n(x, s).
2.6 Initial Values for Modal Amplitudes due to Shot Noise
The discussion so far has concerned only continuous electron beam distributions.
This works well as an idealized model for describing the collective behavior of
the large number of electrons in the beam during the linear FEL regime. In
any realistic e-beam, however, the smooth fluid distribution from Eq. (2.16) is
an approximation to a discrete distribution of a large number of Ne electrons.
Macroscopically the electrons are positioned according to the continuous distri-
bution functions, but microscopically they are positioned randomly within the
50
beam both in space and in time. This is referred to as shot noise, and results in
random, broadband fluctuations in the beam density. A beam with shot noise
will statistically have a non-zero value of the initial e-beam modulation within the
FEL bandwidth that will act as a seed for the amplification of radiation. An FEL
that starts purely from shot noise is referred to as a self-amplified spontaneous
emission (SASE) FEL.
The initial density modulation due to shot noise is contained in all of the
transverse beam modes (decreasing, in general, with higher mode numbers), in-
cluding the lowest order fundamental. These modulations act as pre-bunched
seeds to excite radiation in the corresponding optical modes. Therefore, to excite
and amplify a specific dominant higher-order optical mode, one must take care
to overcome the effective shot noise power in the adjacent modes that will be
amplified and may dominate3 . The density modulation amplitudes due to shot
noise, at the frequency ω, are calculated as follows: The distribution in (2.16) at
the undulator entrance z = 0 is

1 −iωt 1 ∗ iωt
n(x⊥ , z = 0, t) = n0 f (x⊥ ) + δn (x⊥ , z = 0)e + δn (x⊥ , z = 0)e (2.38)
2 2
where the density mode amplitudes at the entrance are given by rearranging
(2.24)
δn (x⊥ , 0)u∗q (x⊥ )d2 x⊥
R
aq (0) = . (2.39)
huq |uq i
We observe the electron beam as it traverses the undulator entrance for a time
period T , during which time Ne = |T I/e| electrons go by where I = |e|v0 n0 Ab is
the beam current. For ωT 1, the initial density modulation function can be
3
Alternatively, suppression of the intrinsic shot noise at the frequency of interest has been
suggested [56] and may relax the constraints on the initial conditions required for dominant
higher-order mode amplification, provided that the accompanying increase in energy modulation
does not offset the effect.
51
expressed as
Z T /2
2
δn (x⊥ , 0) = n(x⊥ , z = 0, t)eiωt dt. (2.40)
n0 T −T /2
In a discrete model, the total density distributions is a sum over delta functions
that represent the location of each electron. For each electron j at transverse
position x⊥j and time tj traveling at the average longitudinal velocity v0 we have
Ne
1 X
n(x⊥ , t) = δ 2 (x⊥ − x⊥j )δ(t − tj ). (2.41)
v0 j=1
2
PNe
This is inserted into (2.40), and the integral over time gives δn (x⊥ , 0) = n0 T v0 j=1
δ 2 (x⊥ − x⊥j )e−iωtj . The density mode amplitude is then given directly in terms
of the discrete sum over all the electrons:
Ne
Ab 1 X
aSN
q (0) =2 u∗q (x⊥j )eiωtj . (2.42)
huq |uq i Ne j=1
The mean-square value of the initial mode amplitudes averaged over the beam
gives an approximation of the initial bunching due to shot noise that leads to
gain and SASE. At the frequency ω, the distribution of the particles in time
is assumed to be uncorrelated, such that the statistical average over the beam
52
retains only the incoherent portion.
* Ne
+
2 E 2Ab 2
D X
SN
aq (0) =
Ne huq |uq i u∗q (x⊥j )uq (x⊥j 0 )eiω(tj −tj0 )
j,j 0 =1
"* Ne + *N +#
2Ab 2 e

X X
= |uq (x⊥j )|2 + u∗q (x⊥j )uq (x⊥j 0 )eiω(tj −tj0 )
Ne huq |uq i j=j 0 j6=j 0
" Ne Ne
#
2Ab 2 X

X
|uq (x⊥j )|2 + u∗q (x⊥j )eiωtj uq (x⊥j 0 )e−iωtj0

=
Ne huq |uq i j=j 0 j6=j 0
4Ab
= Fq,q .
Ne huq |uq i∗
(2.43)
This gives the density mode amplitudes due to shot noise in the beam at the
undulator entrance, for a given mode, at a given frequency. However, since all of
the shot noise modes within the FEL bandwidth will be amplified, the applicable
shot noise amplitudes are obtained by integration over the FEL bandwidth:
D
aSN
2 E 8Ab
q (0)
R ' Fq,q (2.44)
dω Nc huq |uq i∗
√
where Nc = I 2π/|e|ckσω,q is the number of electrons in a coherence length
and σω,q is the mode-dependent relative rms frequency bandwidth (calculated
in Section 2.7.1). Since Nc Ne , this value for the mode amplitudes due to
shot noise is much larger than that in (2.43). Equation (2.44) describes the
initial modal density modulation values due to shot-noise in the beam, which can
be used as a reasonable staring point to model a SASE FEL using the modal
formalism (Figure 2.1), provided that the e-beam is much longer than the total
slippage. It also provides a baseline value for the initial density modulation of
a given mode in a prebunched e-beam to overcome the intrinsic shot noise in
53
25
1010 20
106 15
P @WD
dPPdz
10
100
5
0.01 0
10-6 -5
0 1 2 3 4 0 1 2 3 4
z@mD z@mD
Figure 2.1: Comparison of modal theory (blue) including density mode shot noise
as an initial condition with time-dependent numerical Genesis simulations with
shot noise (red). FEL parameters are given in Table (3.2).
that mode. Because the shot noise density bunching in the fundamental mode is
the typically the largest (because F0,0 is generally largest), and because many of
the modes can couple to each other, the modal density modulation amplitude of
the desired optical mode must be greater than the sum total of all the effective
D 2 E
2 SN
shot noise mode amplitudes, |aq (0)| ≥ aq (0) R . Effective in this sense

dω
refers to the modes with the largest gain within the FEL bandwidth which will
complete for dominance due to the self-tuning nature of the SASE process. This
adds an additional constraint that the initial conditions for amplifying higher-
order modes. The amplitude must be such that the desired mode will not only
dominate at the beginning, but will also dominate throughout the linear gain
regime until either the interaction stops or the process saturates. This will be
addressed later after we take a look at the exponential gain regime. For now
we note qualitatively that, because the direct coupling between the e-beam and
higher optical modes is generally weaker, in order to amplify higher order modes
with OAM one must contend both with reduced coupling during amplification
(ie, less incremental gain), as well as a larger effective shot noise power in the
fundamental mode.
54
2.7 Supermode Matrix Solutions
In the high-gain regime of an FEL with finite e-beam width, the radiation field
tends to concentrate near the beam producing an amplified self-similar trans-
verse field distribution that propagates along the interaction length. This specific
complex-valued combination of the expansion modes is referred to as a super-
mode, or eigenmode of the high-gain FEL. In a sufficiently long undulator, the
fundamental supermode evolves spontaneously due to shot noise in the beam, and
can eventually dominate the other eigenmodes unless it is actively suppressed or
the system saturates. The initial conditions (i.e. the transverse profile and phase
distribution of the injected radiation field or density pre-bunching structure) af-
fect the mode excitation composition and evolution and thus have an effect on
the supermode establishment length.
To find the optimal injection parameters that match and expedite the estab-
lishment of the desired supermode, one must solve Eqs. (2.25) with the appro-
priate initial conditions. To simply find the system supermodes however, it is
enough to find the eigensolutions to Eqs. (2.25). These are the combinations
of the expansion mode profiles that propagate self-similarly, i.e., with constant
amplitude coefficients and with distinct complex wavenumbers [44]. In the pres-
ence of gain, each supermode wavenumber will be different from the wavenumber
of free-space and can be written with a perturbative complex term δk, or gain
parameter, that is due to the FEL interaction:
kSM = k + δk. (2.45)
The complex gain parameter has two parts: The real part δkr = Re{δk} shows up
as an effective refractive index in the electron beam, and the imaginary part δki =
55
Im{δk} is directly related to the gain of the optical field. Since the supermodes
evolve after the initial startup period and have fixed transverse profiles along z,
one can substitute
cq (z) = sq ei(kSM −kzq )z (2.46)
for the mode amplitude coefficients in Eq. (2.1). The supermode amplitude co-
efficients sq are constants, and the z-dependence is contained solely in the mode-
independent exponential term. The time-harmonic electric field of the supermode
is then
hX i
ẼSM (x) = sq Ẽ q (x⊥ ) eikSM z (2.47)
q
where the exponential is outside the sum. The effect of the real and imaginary
components of the gain parameter now become clear. The supermode field is
fixed in transverse profile (since the mode coefficients are independent of the
longitudinal coordinate), and has both an oscillatory component due to k + δkr
and an exponential component due to δki which amplifies (or attenuates) the
field amplitude along z. This transformation converts the coupled second-order
differential equations in Eq. (2.25) into a set of coupled algebraic equations:
h X i
2
(δk − θ0 ) (δk − ∆kq )sq + Kq,q0 sq0
q0
( ) (2.48)
X kzj h X i X
−θp2 Fq,j (δk − ∆kj )sj + Kj,q00 sq00 + Qq,q0 sq0 = 0,
j
kzq q 00 q0
where ∆kq = kzq − k and again θ0 = ω/v0 − k − kw is the detuning parameter for
a 1D model.
Solutions to the coupled equations in (2.48) yield values for the supermode
coefficients sq in terms of the gain parameter δk. The indexed parameters are
56
simply matrix elements and it is convenient to write Eqs (2.48) in a simplified
matrix determinant form

h ih i
2 2
I(δk − θ ) − θ Iδk + − ∆k + = 0. (2.49)

0 p M K Q

The matrix elements of M are given by Mq,q0 = (kzq0 /kzq )Fq,q0 , and similarly for
K = {Kq,q0 }, Q = {Qq,q0 }, and ∆k = {∆kq δq,q0 }. The matrix I is the identity.
Since Eq. (2.49) is cubic in δk for each expansion mode, solutions yield 3Nm
values for δk, where Nm is the total number of expansion modes. Each solution
for δk is inserted in Eq. (2.48) to find the solutions for each mode amplitude
coefficient sq . For a given solution δk, whatever the gain characteristics, the
corresponding set of coefficients sq inserted into the supermode field expression
(2.47) yield an eigenmode of the FEL system. It is not necessary to find all
the solutions, however, since only one of the three solutions for δk will lead to
exponential gain for a each supermode. From Eq. (2.47) and the definition of
kSM in Eq. (2.45), it is clear the dominant complex valued solution for the gain
parameter is of the form,
δk ≡ δkr − iδki (2.50)
where δkr , δki > 0. This will yield the coefficients of the dominant supermode in
Eq. (2.48) where the field is ẼSM ∝ exp[iz(k + δkr ) + zδki ]. The 3D power gain
length, or e-folding length, is therefore given by,
1
LG = . (2.51)
2δki
This is the longitudinal distance over which the optical power of the supermode
increases by e. The effective 1D refractive index of the e-beam is given by the
57
real part δkr as
δkr
nef f = 1 + . (2.52)
k
The phase velocity of the guided FEL signal field is thus vph = ω/knef f . This
is distinct from the phase velocity of the ponderomotive wave, which travels at
velocity ω/(k + kw ) and is the same as that of the electron beam velocity v0 at
resonance. The phase velocity of the FEL wave is faster than the ponderomotive
wave and the electron beam slips backward in phase. In addition there are, of
course, 3D guiding effects which arise in the full supermode solutions as curvature
of the optical wavefront which show up mathematically as a phase that depends
on the transverse coordinates.
2.7.1 Single Transverse Mode
It is useful to obtain an approximate solution to the full supermode equations

by assuming that the system is dominated by a single mode in the expansion
basis. This assumes one chooses an expansion mode very close in character to the
supermode, which can be done by first solving the full equations and readjusting
the expansion mode for the best fit. Space-charge is ignored in the assumption
that θp L π/2. To second order in detuning, the dominant solution to the
matrix equation (2.49) in the single mode limit is:
! !
1/3
Qq,q Kq,q − ∆kq Kq,q − ∆kq θ0 1 Kq,q − ∆kq θ02
δk(q) ' − + 2+ 1/3
+ + 1/3 1/3
2 3 3Qq,q 3 2 3Qq,q 9Qq,q
" ! #
√ 1/3
Qq,q Kq,q − ∆kq θ0 1 Kq,q − ∆kq θ02
−i 3 − 1/3
− − 1/3 1/3
2 3Qq,q 3 2 3Qq,q 9Qq,q
(2.53)
58

1/3
where |Kq,q − ∆kq | Qq,q has been assumed, valid in the limit where the single
2 2
mode is weakly diffracting over the gain length. From ẼSM,q ∝ exp(iδk(q) z)

we obtain the power gain length for the single mode:
1 1 L1D
LG,q ' √ 1/3
' √ 1/3
= 1/3 (resonance) (2.54)
3Qq,q 2 3kw ρFq,q Fq,q
and the resonant Gaussian-like rms detuning bandwidth as a function of distance

along the undulator,
v
1/3
u
u Qq,q
σθ,q (z) ' 3u
t √ h i. (2.55)
2 3z 1 − 23 Kq,q −∆k
1/3
q
Qq,q
1/3
Characterized by the term (Kq,q − ∆kq )/Qq,q > 0, diffraction effects would oth-
erwise have the effect of increasing the overall bandwidth, were it not for the
1/3
larger contribution of the 3D coupling term Qq,q which, by itself, drives the
bandwidth down. Either way, these effects are typically small compared to the
dominant bandwidth narrowing obtained during exponential gain along the un-
√
dulator given by 1/ z. The quantity σω (z) = σθ,q (z)/kw gives either the single
mode FEL relative energy bandwidth or frequency bandwidth, as the detuning
can be written either in terms of a shift in energy, θ0 = kw (γR2 − γ 2 )/γ 2 , or as
a shift in frequency, θ0 = kw (k − kR )/kR where kR is the resonant wavenumber.
Taking (2.55) to the 1D limit, the relative rms bandwidth at the end of an un-
q √
dulator of length L = λw Nw is σθ (L)/kw = 9ρ/2π 3Nw in agreement with
1/6 Kq,q −∆kq
Kim [57]. The bandwidth of the 3D modes are thus only ' Fq,q (1 + 1/3 )
6kw ρFq,q
times smaller than the 1D case, which is a weakly dependent function of the 3D
coupling. With (2.53), the detuning to obtain the shortest available gain length
59
is given by
Kq,q − ∆kq
θ0,max ' − 2 Kq,q −∆kq
. (2.56)
1− 3 1/3
Qq,q
The value of the FEL bandwidth for a single dominant mode provides an
indicator of the end of the linear gain regime for that mode, and thus enables
an estimation of the saturation power and saturation length. Saturation occurs
when the dynamics of the electrons become non-linear and exponential gain of the
radiation abates. To examine saturation completely, one typically needs numer-
ical simulations, or much more involved analytic models (see [58], for example).
But the onset of saturation can be roughly gauged by examining the increase of
the electron beam energy spread during high-gain in the following way; because
the phase space density is conserved, the development of coherent microbunching
in the beam leads to a complementary increase in the energy spread. The process
saturates when the energy spread covers the FEL bandwidth, so the saturation
point can be approximated as the point where h∆γ 2 /γ 2 i ' σθ,q
2
/kw2 from (2.34).
The velocity mode amplitudes can be written in terms of the density mode am-
plitudes via the continuity equation (2.32), and the density modulation is related
to the overall bunching factor via (2.36). Since saturation also corresponds to the
maximum of the bunching factor |b| = 1/2 in the fluid model, these expressions
can be combined to obtain
2/3 R
2
9L1D Fq,q uq (x⊥ )dx⊥
zsat ∼ (2.57)
Ab huq |uq i
where zsat is the approximate distance it takes during exponential gain (ignoring
lethargy) to increase the energy spread from zero until it covers the bandwidth at
resonance. The total length to saturation depends on the initial conditions and
must include the length of the initial startup regime which can last for several
60
gain lengths, depending on the input. It is worth reiterating that this is a rough
estimate – saturation is a highly non-linear process that depends on several factors
not included in this linear model. The length to saturate the FEL is also highly
sensitive to the startup and evolution process which includes several transverse
modes and happens over a range of frequencies in the FEL bandwidth. This
affects the proportionality between the saturation and the gain length in (2.57).
Numerical coefficients aside, it is interesting to note the scaling dependence of zsat
on the mode coupling Fq,q , which itself generally decreases with increasing mode
number for simple e-beam geometries. This suggests that higher-order modes
will tend to saturate earlier.
With (2.57) the relative bandwidth at saturation is approximately,

s
σθ,q (zsat ) ρ Ab huq |uq i
σω (zsat ) = ∼ 1/6 . (2.58)
kw uq (x⊥ )dx⊥ 2
R
Fq,q
Note that, aside from a weak dependence on 3D factors, the relative bandwidth
at saturation is essentially ρ as expected from 1D theory [59].
We can now address the issue of optical mode amplification from initial con-
ditions on the FEL system. In Section 2.6 we calculated the equivalent density
bunching amplitudes for each mode due to shot noise. This identifies the statis-
tical minimal values on the initial conditions required for the amplified optical
mode to exceed the radiation amplified starting from shot noise. For an FEL op-
erating as an amplifier of a narrow band external seed, the injected mode power
must at least be higher than the equivalent shot noise power,
1 1/3
Pin,q ρPb Fq,q (2.59)
18Ne
for a single mode, at the seed frequency ω. The quantity Pb = γmc2 I/e is the
61
e-beam power.
Equation (2.59) is a minimal constraint that specifies the lower power limit at
a single frequency. In reality the system is more complicated. To insure that the
input is above the broadband SASE power across the entire range of frequencies in
the FEL bandwidth, the 1D effective shot noise power for a cold beam, integrated
over the bandwidth, provides a more robust baseline [57]:
1
Pin,q > PSN = ρPb (2.60)
9Nc
where Nc = ILλ /|e|c is the number of electrons in a coherence length4 Lλ =

√ √
λ/ 2πσω (zsat ) ' λ/ 2πρ at saturation.
2.8 Uncorrelated Energy Spread Effects
So far, the focus has been on the dynamics of the FEL system in the cold e-beam
limit in which all the electrons in the beam have the same energy and travel with
the same longitudinal velocity v0 . The e-beam as a whole has no correlated energy
spread other than the projected spread it obtains from the velocity modulation
during the FEL interaction. For optical or longer wavelength FELs the cold beam
analysis is adequate, as the effects of slice energy spread at the level obtained
in modern high-brightness beams can often be neglected. In general however,
it is necessary to include energy spread effects, particularly in the discussion of
shorter radiation wavelengths and higher-order modes since an energy spread in
the beam can have a negative impact on such small scale structures. Preservation
of higher-order structures in the beam is required, at least over a few gain lengths,
4
√
Not to be confused with the oft-quoted cooperation length, Lcoop = λ/4π 3ρ, which is the
e-beam slippage over a 1D gain length, and is Lcoop ' 0.12Lλ .
62
for amplification of higher-order radiation modes in the FEL.
Energy spread effects have been studied in detail by several authors, since
the earliest work on high-gain FEL theory [60, 61, 57, 44]. The most common
(and perhaps straightforward) approach is to incorporate the solutions of the
Vlasov equation for a linearized e-beam distribution into Maxwell’s equations. In
this section, a brief derivation using the Vlasov approach is presented with minor
modifications from previous work in order to accommodate the present eigenmode
expansion formalism for the optical field. This provides a direct way to include
the energy spread distribution of the e-beam into the modal excitation equations.
The resulting equations are particularly useful in the steady state limit of the FEL
where the supermode matrix solutions from the previous sections are retrieved,
but with the effects of energy spread included via a simple integral. In the cold
beam limit, the results from the last section are recovered. Further details on the
derivation can be found in the cited literature, with the most relevant treatment
presented in [62]. Certain elements of the derivation specific to the present work
are discussed in greater detail in Appendix C.
2.8.1 Integro-Differential Energy Spread Equations
Let us assume that the FEL EM field is described by a simple transverse field
h i
(which will be expanded into a modal set shortly) of the form E⊥ = Re Ẽs ei(kz−ωt) ,
with Ẽs = Ẽs ê⊥ which satisfies the paraxial wave equation in the presence of a
source current
∂ Ẽs
∇2⊥ Ẽs + 2ik = −iωµ0 J̃s (2.61)
∂z
h i
where J⊥ = Re J̃s ei(kz−ωt) .
To include an energy distribution we describe the e-beam by a distribution
63
function F = F (x⊥ , z, η, ψ) that satisfies the Vlasov continuity equation
∂F ∂F dψ ∂F dη
+ + =0 (2.62)
∂z ∂ψ dz ∂η dz
where via the parallel-beam stipulation dx⊥ /dz = 0 it has been assumed that the
transverse e-beam profile is fixed along the undulator length, ie., no envelope scal-
loping or active transverse rearrangement of the electrons. The ponderomotive
phase and the relative energy variables are given respectively as
γ1 − γ
ψ = (k + kw )z − ωt, η= . (2.63)
γ
The variable ψ specifies the position of an electron in the phase bucket of the
ponderomotive field, and η quantifies it’s relative energy deviation from the nom-
inal beam energy γ, which itself may be different from the resonant energy γR .
The total derivative of the phase is thus
dψ k
= 2 η − θ0 . (2.64)
dz γz
Assuming that the wiggling velocity due to the undulator field dominates
the transverse motion (that is, taking only the first term in Eq. 2.12), it is
straightforward to also verify that the change in relative energy during the FEL
interaction in the presence of a longitudinal field due to space-charge, Ẽz , is
dη −e h iψ
i
= (Ẽz − iK Ẽs /2γ)e + c.c. . (2.65)
dz 2γmc2
The field polarization direction is assumed parallel to the direction of electron

motion in the undulator (ĝ = 1) and JJ = 1 is taken for a helical undulator.
For small density perturbations we can expand the distribution function as
64
F = F0 (x⊥ , η) + 21 F1 (x⊥ , z, η)eiψ + c.c. and obtain the a linear expression for the
perturbed distribution amplitude
!
∂F1 k e K Ẽs ∂F0
+i 2
η − θ0 F1 − Ẽz − i = 0, (2.66)
∂z γz γmc2 2γ ∂η
with solutions
!
z
Z
−iz k
2 η−θ0
e 0 K Ẽs ∂F0 i k
2 η−θ0 (z 0 −z)
F1 = F1 (0)e γz + dz Ẽz − i e γz ,
γmc2 0 2γ ∂η
(2.67)
where F1 (0) is the initial modulation at the undulator entrance.
In terms of the distribution function, the current density vector is
Z
J = −ecβ F dη, (2.68)
which yields J˜⊥ = − iecK

R
2γ
F1 dη for the transverse current density amplitude
in (2.61) and J˜z = −ev0 F1 dη for the longitudinal modulation. This gives
R
Ẽz = iω1 0 J˜z = − iω

ev0
R
0
F1 dη for the space-charge field in (2.67), assuming as
before that the transverse beam size is large compared to the microbunching
wavelength in the co-moving frame, σx > γz λ/2π.
By assuming zero initial modulation and integrating over η, Eq. (2.67) can
be written as an integro-differential equation for the slowly varying FEL fields
(in SI units) [44],
( " # )
z
eK 2
Z
∂ Ẽs e ∂
∇2⊥ Ẽs + 2ik = iωµ0 ec dz 0 Ẽs + 2 ∇2⊥ + 2ik 0 Ẽs
∂z 0 4γ 2 k ∂z
(2.69)
∞
Z
1 ∂F0 i k
2 η−θ0 (z 0 −z)
× dη e γz .
γmv02 −∞ ∂η
65
This equation describes the self-consistent excitation and evolution of the ampli-
fied radiation fields during the FEL interaction, including the effects of longitudi-
nal space-charge and energy spread in the e-beam. The fields are slowly-varying
functions of z, and can be expressed as the product of independent functions of
x⊥ and z for solutions of the high-gain self-similar FEL supermode. Since ini-
tial modulations have been ignored, this describes an FEL system for which the
growing instability is started by an input EM seed.
As in [63] the procedure is to mode-expand integro-differential equation (2.69)

and find the supermodes in order to establish an analytic connection with the
modal description, and to pick up an energy spread term enroute. The paraxial
field Ẽs is related to the mode-expanded electric fields in (2.3):
X
Ẽs (x⊥ , z) = Ẽ⊥ (x)e−ikz = E0 cq (z)uq (x⊥ )ei(kzq −k)z (2.70)
q
where again, kzq is the axial wavenumber of the q th mode and cq (z) is the slowly-
growing amplitude coefficient. The modes uq are eigenfunctions of the dielectric
equation in (2.6). Thus, in order to express (2.69) in terms of waveguide modes,
we simply plug in (2.70). Many of the details are worked out in Appendix C.
For simplicity, we assume that the unmodulated distribution F0 can be writ-

ten as the product of two uncorrelated functions of the transverse coordinate
and energy spread, F0 = n0 f (x⊥ )fη (η), where energy spread function satisfies
R
fη dη = 1. After some simple manipulation we obtain an expression for the
66
evolution of mode amplitude coefficients in terms of familiar coupling terms,
θp2 γz2 z
Z
d Xh
i(kzq0 −kzq )z 0
cq (z) = −icq0 (z)Kq,q0 e + dz 0 ei(kzq0 z −kzq z)
dz q0
k 0
( )
k2K 2 X kzq00 kzq0 d
× 2
Fq,q0 − Fq,q00 Kq00 ,q0 + i Fq,q0 0 cq0 (z 0 )
8kzq γ q 00
kzq kzq dz
∞
Z
dfη i k
2 η−θ0 −k (z 0 −z)
i
× dη e γz .
−∞ dη
(2.71)
Just like the coupled fluid equations in (2.25), this equation solves for the evolu-
tion of individual mode amplitudes, both inside the undulator and during propa-
gation before and after the interaction where K = 0. When k ' kzq we recognize
θp2 k2 K 2
the coupling Qq,q0 ' F 0
8γ 2 kzq q,q
above.
The fixed-profile solutions are exponentially amplified supermodes from Equa-

tion (2.46). Inserting them into Eq. (2.71) transforms it into an algebraic rela-
tionship for a direct calculation of the supermode coefficients sq . The results is
an algebraic matrix equation similar to (2.49),
" #
h 1 ih i
I − θp2 M Iδk + K − ∆k + Q s = 0 , (2.72)
S1 (δk, θ0 )
but where here the effects of uncorrelated energy spread and detuning are ex-
pressed as a single function S1 (δk, θ0 ), defined by the integral
∞
γ2
Z
dfη /dη
S1 (δk, θ0 ) = z dη . (2.73)
k −∞ δk − θ0 + γk2 η
z
Note that for a Gaussian distribution in η, from this one retrieves the derivative of
the plasma dispersion function. As before, the supermode coefficients sq are found
from (2.72) and the dominant, or highest gain supermode coefficients correspond
67
to the solution that yields the largest value of δki > 0 in the gain parameter
δk = δkr − iδki .
There are only a handful of sensible forms for fη (η) that yield simple ana-
lytic solutions to the energy spread dispersion relation in Eq. (2.73). One is a
Cauchy-Lorentz distribution, which has a simple analytic form and is a reason-
able approximation to an actual beam (Figure 2.2). With a relative full-width
half maximum (FWHM) energy spread of δη the Cauchy-Lorentz form is,
1 δη
fη (η) = . (2.74)
2π η 2 + (δη/2)2
Plugging this distribution in the energy spread function in Eq. (2.73) the super-
mode determinant equation from (2.72) is,
" 2 #
k
− θp2 M Iδk + K − ∆k + Q = 0.

I δk − θ0 − iδη 2 (2.75)

2γz
For a monochromatic (cold) beam δη → 0 the limit of the energy distribution

function in (2.74) is fη (η) = δ(η) and equation (2.72) becomes precisely the
supermode equation in (2.49) derived using the linear e-beam fluid model. The
equivalence between the Vlasov formalism in Ref. [44] for a cold beam and
the linear plasma fluid model for the supermodes of a high-gain FEL is thus
established [63].
2.8.2 Limitations on Energy Spread
The preceding analysis is limited to the linear operating regime of the FEL.
At saturation the exponential gain stops once the induced energy spread covers
the FEL bandwidth. With an additional correlated energy spread on the beam
described in this section, not only is the gain length increased, but the time it
68
a)
b)
FWHM
Figure 2.2: Comparison between a) Gaussian and b) Cauchy-Lorentz energy

distributions
√ with the same FWHM. The FWHM of a Gaussian energy spread is
2σγ ln 4 where σγ is the rms.
takes for the interaction to reach saturation is reduced since the beam already
occupies a larger area in phase space a priori. Clearly, then, the constraint on the
uncorrelated energy spread is that it be much less than the bandwidth otherwise
the exponential gain process cannot take place. For a relative rms energy spread
of σγ , the requirement that it not exceed the rms bandwidth from Eq. (2.58) is
written simply as
σγ σω . (2.76)
2.9 Transverse Dynamical Effects
The preceding analysis has considered only electron beams in which the trans-
verse motion of the electrons is due only to the undulator and signal fields (2.12).
The electrons are assumed to maintain a fixed transverse position with respect
to one another in the e-beam (parallel beam limit). This approximation holds
for FELs where any transverse mixing of the electrons and resulting longitudinal
69
energy spread occurs on long scales such that the transverse motion does not have
an appreciable effect on the FEL process. The parallel beam limit permits a sim-
plified look at the FEL process in terms of the cold-fluid equations which describe
the evolution from startup to exponential gain in terms of differential equations
(2.25), which in turn, provide a simple framework to model initial conditions.
In real systems, however, the electrons are not strictly frozen transversely; they
can move as dictated either by intrinsic beam properties such as emittance or
through an external influence like transport through the fields of the beamline
optical elements.
Simple transport through the undulator causes transverse motion additional

to the simple transverse wiggling motion. This is because the magnetic field
in undulators is not purely transverse as in Eq. (2.11) but has longitudinal
field components that lead to periodic transverse beam focusing for the off-axis
electrons (See Appendix D). In this type of averaged, periodic focusing, whatever
the source, an electron will undergo sinusoidal transverse oscillations of the form
x0 β (0)
xβ (z) = xβ (0) cos (kβ z) + sin (kβ z)
kβ
(2.77)
dxβ
x0 β (z) = = x0 β (0) cos (kβ z) − xβ (0)kβ sin (kβ z).
dz
where kβ = 2π/λβ is the betatron wavenumber of the focusing channel. The
transverse motion reduces the longitudinal velocity and reverses the position of
the particle in the beam every half focusing period. Betatron motion of this sort,
combined with a distribution in the transverse electron velocities characterized
by the beam emittance, affects the resonant FEL interaction process, as we will
soon see.
The beam emittance quantifies the amount of relative transverse motion of
70
the electrons. The emittance is the area of an ellipse in the transverse phase space
that contains some fraction of the beam5 . For a given normalized distribution
in the (x, x0 ) phase space F = F (x, x0 ) of a beam on-axis, we can define the
mean-squared beam size, divergence and correlation respectively as
Z Z Z
0 02 0
σx2 = 2
x F dxdx , σx20 = x F dxdx , σ
xx0 = xx0 F dxdx0 (2.78)
The beam rms emittance is then given by
q
x = σx2 σx20 − σxx
2
0 (2.79)
which defines an area in the (x, x0 ) plane which contains the same fraction of
electrons at each longitudinal position along the beamline provided no forces
act on the beam. The more useful oft-quoted figure of merit is the normalized
emittance, nx = γβx which is invariant under acceleration and a constant of
motion in the absence of energy dissipation by virtue of Liouville’s Theorem.
The emittance is an essential beam parameter as it dictates, among other things,
the transport characteristics of the beam and the gain characteristics of the FEL.
With the position and velocity of the particles along the focusing channel
given by (2.77) and the associated rms beam properties in (2.78), the electron
beam size during periodic focusing is (Appendix D),
2σxx0 (0) σx20 (0) 2

σx2 (z) = σx2 (0) cos2 (kβ z) + sin (kβ z) cos (kβ z) + sin (kβ z). (2.80)
kβ kβ2
Clearly, if one wants to assure that the beam size remains at a fixed value through-
out transport then σx2 (z) = σx2 (0) for all z, which demands there be no correlation;
σxx0 = 0. This defines the e-beam waist which is preserved if the beam is beta-
5
Which fraction depends on the convention. Here we use the rms emittance.
71
beam envelope
ΛΒ
Figure 2.3: Matched beam envelope.
matched :
x
r p
σx = , σx0 = x kβ (2.81)
kβ
where x = σx σx0 . For low energy machines the natural focusing of the undulator,
given by the wavenumber
K̄kw
kf = √ , (2.82)
2βz γ
may be sufficient to preserve a beta-matched beam throughout the undulator.
In this case, kβ = kf . Otherwise an external focusing channel such as a FODO6
lattice can be used. For this analysis it is important to note that the betatron
phase advance per FODO cell is taken to be small (and symmetric between x and
y) compared to the average beta function of the beam to avoid large beam size
variation during transport. External FODO focusing is generally used for high
energy machines due to the kf ∝ 1/γ dependence in natural undulator focusing.
6
FODO: focus, drift, defocus and drift
72
2.9.1 Integro-Differential FEL Emittance Equations
Transverse emittance and focusing effects are crucial for a fully applicable analytic
model of FELs. They have been investigated by numerous authors [57, 64, 65, 42,
66, 67, 68], though not exactly in the context of eigenmode expansions pursued
here7 . For a deeper analysis see Huang [67] which follows directly from Kim [66].
Here we cover relevant modifications to accommodate the dielectric expansion
basis modes.
Similar to a beam with an energy spread in 2.8, the effects of emittance

are described through the evolution of the beam’s distribution function F =
F (x⊥ , x0⊥ , η, ψ, z) along z, which satisfies the continuity equation
∂F ∂F dψ ∂F dη ∂F dx⊥ ∂F dx0⊥
+ + + + 0 =0 (2.83)
∂z ∂ψ dz ∂η dz ∂x⊥ dz ∂x⊥ dz
where x⊥ and x0⊥ are the averaged transverse variables and dx0⊥ /dz = −kβ2 x⊥ .
There are two transverse dynamical effects that contribute to the FEL interaction
process. The first is the transverse motion of the electrons as they oscillate back
and forth across the beam profile, characterized explicitly by the last two terms
in (2.83). Since the 3D radiation field profile is not necessarily constant over the
e-beam, the particles that traverse the e-beam profile necessarily interact with
EM fields of varying amplitude. This effect to the FEL emission can be partic-
ularly relevant for higher-order modes since these field profiles have additional
transverse variation as well as a phase that depends on transverse position. The
ponderomotive phase bucket of these modes has a well-defined 3D geometry (like
the helical microbunching geometries considered later) which can be washed out
7
Ref. [68] comes the closest to our approach, using an expansion basis that closely resembles
the LG modes examined in the following chapter, but only in the case of zero or large (x
1/2k) geometric emittance, and without mention of the orbital angular momentum content of
the modes.
73
by uncompensated transverse mixing. This motion can affect both the establish-
ment length and gain length of the OAM modes, but is negligible if the betatron
period is much larger than a 1D gain length, kβ L1D 1, or if the e-beam is small
compared to the radiation profile [62].
The second effect is the effective reduction in longitudinal velocity that results
from the transverse betatron motion. Electrons that have the same total energy
have a spread in longitudinal velocities in the undulator due to the distribution
of betatron amplitudes. This leads to an effective energy spread and contributes
to the evolution of the electron phase via the second term in (2.83), which now
includes both the relative energy and the transverse motion variables: dψ/dz =
kη/γz2 − θ0 − k(kβ2 x2⊥ + x02
⊥ )/2. For a symmetric Gaussian transverse velocity
distribution, the effective longitudinal energy spread is h(∆γ)2 /γ 2 i = γz4 2x kβ2 ,
and can be added in quadrature to the intrinsic uncorrelated component to yield
a total effective energy spread [62, 69],
q
σγ(ef f ) = σγ2 + γz4 2x kβ2 (2.84)
If γz2 x kβ σγ , the contribution of emittance to the longitudinal velocity spread

can be neglected.
The overall emittance contributions are calculated by solving (2.83). Lineariz-

ing the distribution function, we again expand for small modulation amplitudes,
F = F0 (x⊥ , x0⊥ , η) + 21 F1 (x⊥ , x0⊥ , z, η)eiψ + c.c., the linear evolution of the modu-
lated component is given by

∂F1 k k 2 2 ∂F1 ∂F1 eK Ẽs ∂F0
+i 2
η − θ0 − (kβ x⊥ + x⊥ ) F1 +x0⊥
02
−kβ2 x⊥ 0 +i 2 2 = 0.
∂z γz 2 ∂x⊥ ∂x⊥ 2γ mc ∂η
(2.85)
Longitudinal space charge is neglected. By the method of integration over the
74
unperturbed trajectory [66] via the method of characteristics, solutions for F1
with zero initial modulation are
„ «
Z z
eK ∂ i k k 2 2 02
2 η−θ0 − 2 (kβ x⊥ +x⊥ ) (z 0 −z)
F1 = −i 2 2 dz 0 Ẽs (x⊥0 , z 0 ) F0 (x⊥ , x0⊥ , η)e γz
,
2γ mc 0 ∂η
(2.86)
with the unperturbed trajectory
x0 ⊥
x⊥0 = x⊥ cos [kβ (z 0 − z)] + sin [kβ (z 0 − z)] . (2.87)
kβ
The current density vector of a beam characterized by a finite transverse

velocity distribution tis J = −ec βF dηdx0⊥ , which gives J˜⊥ = − iecK F1 dηdx0⊥
R R
2γ
for the transverse current density amplitude in (2.61). Thus, by integrating (2.86)
over the energy η and velocity x0⊥ variables and using the paraxial field relation
in (2.61) we obtain
Z z Z ∞
∂ 0
2
∇⊥ + 2ik 2
Ẽs (x⊥ , z) =2iQγz dz dx0⊥ Ẽs (x⊥0 , z 0 )
∂z 0 −∞
„ « (2.88)
Z ∞
1 ∂F0 i k k 2 2 02
2 η−θ0 − 2 (kβ x⊥ +x⊥ ) (z 0 −z)
× dη e γz
.
n0 −∞ ∂η
This describes the self-consistent evolution of the 3D FEL EM field from startup
with an EM input through the high-gain regime. The effects of diffraction, energy
spread, betatron motion, and transverse emittance are all included.
2.9.1.1 Modal Solutions in a Matched Beam
In order to calculate the evolution of the individual waveguide eigenmodes as they

interact in the FEL process, we use the expansion for Ẽs in Eq. (2.70) and insert
it into (2.88). For simplicity, we further assume that the e-beam is beta-matched
75
to the transport inside the undulator; that is, there is no correlation on average
between the transverse velocity and the transverse position of an electron in the
beam and F0 is stationary. We further posit that the uncorrelated energy spread
is also decoupled from the transverse variables and that the distribution function
is written as a product of independent functions: F0 = n0 f (x⊥ )fp (x0⊥ )fη (η).
The integro-differential equation (2.88) is then written in terms of the modal
expansion amplitudes as,
z
γ2
Z Z Z
d Xh dfη
cq (z) = −icq0 (z)Kq,q0 ei(kzq0 −kzq )z + z d2 x0⊥ fp (x0⊥ ) dη dz 0
dz q0
k dη 0
i
0 kη/γz2 −θ0 −k− k2 x02 0
⊥ (z −z)
×cq0 (z 0 )ei(kzq0 z −kzq z) Qq,q0 (x0⊥ , z, z 0 )ei .
(2.89)
The coupling Qq,q0 between the e-beam and the EM field modes is given by:
0

Qk (z − z)k
Qq,q0 (x0⊥ , z, z 0 )
2 2

= uq (x⊥ ) f (x⊥ )exp −i kβ x⊥ uq0 (x⊥0 )
kzq huq |uq i 2
(2.90)
In the steady-state supermode limit we write τ = z 0 − z and get,
"
0
γ2
Z Z Z
X
2 0 0 dfη
(δk − ∆kq )sq = − sq0 Kq,q0 +i z d x ⊥ fp (x ⊥ ) dη dτ
q0
k dη −∞
#
δk−θ0 +kη/γz2 − k2 x0 2⊥
×Qq,q0 (x0⊥ , τ )eiτ ( ) .
(2.91)
Note that when the transverse momentum distribution tends toward a zero-
emittance beam fp (x0 ⊥ ) = δ(x0 ⊥ ) and the focusing effects vanish (kβ → 0), then
the coupling goes to Qq,q0 → Qq,q0 as given by (2.28) and the above expression
becomes exactly Eq. (2.72) with space-charge effects neglected and the energy
spread integral given by S1 defined in Eq.(2.73).
76
Equation (2.91) describes the evolution of the supermode coefficients sq in the
presence of diffraction, energy spread, emittance and period-averaged betatron
motion for a fixed profile e-beam. The coupling Qq,q0 between the optical mode
fields and the e-beam differs from the cold, emittance-free coupling factor Qq,q0
through the presence of the transverse motion (given by the transverse velocity
spread in fp ) and through the periodic focusing characterized by the wavenumber
kβ . Both effects are manifest in extra phase terms in the integral and in the
functional dependence of the expansion mode on the betatron-weighted orbit
coordinate x⊥0 . The transverse energy spread and betatron motion both lead
to an increase in the gain length and to an enhanced coupling between the non-
orthogonal transverse modes. This behavior is to be expected physically, as
the transverse electron motion communicates phase information across the beam
profile. This mechanism can expedite the evolution of the system towards the
fundamental transverse mode, as well as increase the transverse coherence up to
x ' k −1 [70].
2.9.1.2 Gaussian Beam Solutions
The presence transverse motion in a general sense prevents a straightforward

mathematical simplification of the supermode emittance equation (2.91) into a
matrix form as in (2.72) because the modes are functions of τ via the integral over
the unperturbed trajectory. The emittance equations can, however, be written
in a somewhat more condensed and useful form by assuming specific energy,
momentum and density distributions (closely mimicking Xie [42]):
1 δη 1 −x0 2⊥ /2σ20 2 2
fη (η) = , fp (x0 ⊥ ) = e x , f (x⊥ ) = e−x⊥ /2σx .
2π η + (δη/2)2
2 2πσx20
(2.92)
77
The e-beam distribution is assumed to be axisymmetric both spatially and in
divergence. For our azimuthal modes of the form uq (x⊥ ) = ûq (r)eilφ , a coordinate
transformation enables evaluation of the transverse divergence integral and we
obtain
" Z ∞
X 2πδl,l0 Qk τ dτ 2
(δk − ∆kq )sq = − s q0 Kq,q 0 − 2 e−iτ (δk−θ0 −iδηk/2γz )
q0
kq σx2 huq |uq i 0 sin (kβ τ )
#
Z ∞ Z ∞
rdr ûq (r)ûq0 (r̃)Il 2ζ rr̃ cos(kβ τ ) exp −ζ (r + r̃2 ) .
2 2 2
r̃dr̃
0 0
(2.93)
Here, ζ 2 = (1 − iτ kσx2 kβ2 )/2σx2 sin2 (kβ τ ) and Il (x) is a modified Bessel function.
This form for the steady-state equations is virtually identical to that presented
in Refs [65, 42, 67, 68] but is specifically tailored for solutions with optical modes
that satisfy the general dielectric waveguide in (2.6). One can therefore choose a
given mode basis of interest and use (2.93) to calculate the modal amplitudes dur-
ing high-gain lasing. This is done in the next chapter. Note that for a Gaussian
energy spread distribution fη (η) = (2πσγ2 )−1/2 exp(−η 2 /2σγ2 ) with rms relative en-
ergy spread σγ , one simply replaces the term −τ δηk/2γz2 in the exponential with
−τ 2 k 2 σγ2 /2γz4 .
It is worth pointing out that even with the effects of emittance and smoothed
focusing included, no direct cross-couping exists between the modes with different
azimuthal mode numbers, though there is still coupling between radial modes.
This is shown via the Kronecker delta function δl,l0 multiplying the coupling
integral in (2.93), and comes about because the e-beam distribution is, on average,
axisymmetric. As a result, in contrast to the well-known non-orthogonality of
the radial modes, the azimuthal modes are orthogonal, even in the presence of the
transverse mixing terms that appear in the Vlasov equation in (2.83). This result
78
will have important consequences for the amplification of OAM modes discussed
in subsequent chapters, as it indicates that excited OAM modes will not couple
directly to the fundamental which tends to have the highest growth.
The exact solutions to (2.93) can be obtained numerically for each of the
modes q. One method pursued in the literature is to discretize the radial coordi-
nates r̃ and r to transform the integrals into finite sums [42]. One then obtains
a matrix equation that can be solved iteratively for specified values of detuning,
energy spread, betatron wavelength and beam size. This can be a powerful tool
to find the exact solutions for a given FEL system. However, since the numerical
solutions obtained in this way do not provide a general insight into the physi-
cal scaling characteristics of the system, they will not be elaborated upon here.
Explicit integral solutions will be obtained for our purposes using the LG mode
basis in Chapter 3.
2.10 Recovering the 1D regime
In the 1D FEL limit the contribution of the transverse effects vanishes. The opti-
cal field is treated as an infinite plane wave in the assumption that the Rayleigh
length is much larger than the gain length, zR L1D such that the diffraction
effects are negligible. Since the light does not diffract, the transverse dependence
of the refractive index of the virtual dielectric, which is used to model the effective
3D optical guiding of the signal field goes to zero, so we set Kq,q0 = 0 in the cou-
pled evolution equations (2.25). The axial modal wavenumbers are all degenerate
in the plane-wave limit, so kzq = k. In the additional limit of a large, transverse
beam profile f (x⊥ ) → 1 in the parallel-beam limit (kβ = 0) the beam profile
overlap coefficient is diagonal F = I, as is the coupling: Qq,q0 = Qq,q0 → Qδq,q0 .
The cold-beam coupled equations (2.25) are then written as the evolution of 1D
79
amplitudes as
d d2 1
c(z) = −iξeiθ0 a(z), 2
a(z) + θp2 a(z) = − Qc(z)e−iθ0 z , (2.94)
dz dz ξ
where ξ = Kn0 e/4γ0 E0 . Ignoring space charge effects, the amplitude of the field
at resonance is reduced to simply,
d3
c(z) = iQc(z), (2.95)
dz 3
with general solutions

3
X
c(z) = Cj eiδkj z . (2.96)
j=1
where
√ √
1/3 1 + i 3 1/3 1 − i 3 1/3
δk1 = −Q , δk2 = Q , δk3 = Q . (2.97)
2 2
These are the three roots of the well-known cubic equation. The constants Cj in
(2.96) can be re-expressed in terms of the initial conditions on the fields c(0), the
density modulation a(0), and the velocity modulation av (0):
3
X
c(0) = Cj
j=1
3
dc(z) X
z=0 = −iξa(0) = i δkj Cj (2.98)
dz j=1
3
d2 c(z) k v X
z=0 = −ξ a (0) = − (δkj )2 Cj
dz 2 βz j=1
80
The field amplitude in (2.96) in terms of the initial conditions thus has the form
3
1X ξ kξ
c(z) = c(0) − a(0) + 2
a (0) eiδkj z .
v
(2.99)
3 j=1 δkj βz (δkj )
When the interaction begins, all of the terms contribute to the evolution. This
is referred to as the startup period. Shortly thereafter the exponentially growing
terms associated with the root δk3 begin to dominate over the oscillatory δk1 and
decaying δk2 terms. Let’s take, for example, a scenario where the initial density
and energy modulation is zero a(0), av (0) = 0, and the interaction is started by
an input field c(0) 6= 0. The power in the field goes like:
2
3 √ 1/3
1 X 1
|c(z)|2 = c(0) eiδkj z ' |c(0)|2 e 3Q z (2.100)

9 j=1
9
The startup period ends when the high-gain terms dominate and the field power
grows exponentially. This occurs after approximately one gain-length, given by,
1 1 1
L1D = − =√ = √ , (2.101)
2 Im{δk3 } 3Q 1/3 2 3kw ρ
which is the length over which the field intensity will grow by e = 2.718. Since
effects like space charge and emittance modify the gain length, L1D is a convenient
parameter used to gauge the deviation of more complex systems from the 1D
model.
The real part of the dominant root δk3 gives the contribution to the prop-
agating wavenumber of the signal field kf el = k+Re{δk3 } = knef f where the
corresponding effective 1D refractive index is
kw ρ
nef f = 1 + . (2.102)
k
81
∆Η = 0 Θp = 0
1.0 1.0
0.8 a) 0.8
f)
0.6 0.6
∆ki ∆ki g)
0.4 e) d) 0.4
c) b)
i) h)
0.2 0.2
0.0 0.0
-4 -2 0 2 4 6 8 10 -4 -2 0 2 4 6 8 10
θ θ
Figure 2.4: Left plot: 1D gain versus detuning for different values of space-charge
with δη = 0. Labeled curves are: a) θp = 0, b) θp = 0.5, c) θp = 1, d) θp = 2,
and e) θp = 3. Right plot: Gain versus detuning for different values of the scaled
energy spread with θp = 0. f) δη = 0, g) δη = 0.5, h) δη = 1, i) δη = 2.
The resonant exponential gain process thus has the additional effect of slowing
the phase velocity of the otherwise free-space plane wave which travels at c. The
resonant phase velocity of the amplified radiation is vph = ω/kf el = ω/(k + kw ρ),
which is greater than that of the ponderomotive wave given by vpm = ω/(k + kw )
since ρ 1. The phase velocity of the emitted FEL signal field thus travels
faster than the electron beam in the resonant phase bucket, but slower than c.
For a finite length e-beam the FEL light can then slip out of the front of the
bunch, given enough time8 . Also, since the gain depends on the frequency of the
wave, the propagation through the e-beam is dispersive. An interesting property
of the high-gain regime is that the resonant group velocity vg = ∂ω/∂kf el =
c/(1 + 1/3γz2 ) is also slower than c.
The effects of energy spread, space-charge and detuning can be investigated for
the 1D limit by examining the 1D form of the the steady-state matrix equation in
(2.49). Since all the modes are degenerate in 1D, equation (2.75) is fully diagonal.
8
This effect is not considered in the present analysis where the electron bunch length is
considered much longer than the slippage during the interaction, Nw λ.
82
It reduces to a simple cubic equation in terms of detuning θ0 , beam coupling Q,
the longitudinal plasma wavenumber θp and fwhm energy spread δη,
h k i
δk (δk − θ0 − iδη 2 )2 − θp2 + Q = 0. (2.103)
2γz
It is useful at this point to define the following scaled parameters9 : δk = 2δkL1D ,

θp = 2θp L1D , θ = 2θ0 L1D and δη = δηkL1D /γz2 . With these the 1D cubic equation
(2.103) becomes:
3
h
2 2
i 2
δk (δk − θ − iδη) − θp =− √ . (2.104)
3
Figure 2.4 plots the scaled gain parameter δki = L1D /LG of the dominant root as
a function of detuning for several values of longitudinal space-charge and energy
spread. Both effects increase the overall gain length LG and push the gain curve
peak into the region θ < 0 where the e-beam energy is above the resonant energy.
In a cold beam subject to strong space charge effects where |δk| θp , detuning
the system to θ = −θp where the space charge wave is synchronous with the EM
wave maximizes the available gain per unit length[61]. In the absence of both
space-charge effects and energy spread, the 1D detuning threshold to still obtain
complex-valued solutions that yield exponential gain is θ ≥ −2.182.
|δk| θp , it is straightforward to show that the shortest gain length (the

maximum of δk i ) is obtained at a detuning of θ0 = −θp where the space charge
wave is synchronous with the EM wave[61]
9
More scaled variables will be added in the following chapters. See Table 3.1 for a complete
list.
83
2.11 Conclusions
In this chapter we have developed a detailed description of the 3D high-gain FEL

in terms of an arbitrary expansion basis for the optical modes. In the cold, parallel
beam limit, the effects of detuning, diffraction, and longitudinal space charge are
included in a set of coupled differential equations that provide a straightforward
way to examine the evolution of the system in terms of the initial conditions on
the e-beam and on the field. The effective density modulation amplitudes due to
shot noise are also derived in the modal basis for examination of SASE processes.
The bandwidth, optimal detuning, saturation length, and equivalent shot noise
power have been derived in the single mode limit. The effects of energy spread
and emittance have also been included using the Vlasov formalism, and like the
parallel beam case, are shown to preserve the orthogonality of azimuthal modes
in optical beam. This will have important consequences for the amplification of
OAM modes, examined in the next chapter.
84
CHAPTER 3
Description of FEL Interaction in terms of

Eigenmodes of Orbital Angular Momentum
In the previous chapter, a set of coupled excitation equations was derived for the
slowly-growing mode coefficients of the EM signal field of a high-gain FEL. In
this chapter, a complete orthogonal guided basis set of Laguerre-Gaussian (LG)
modes is examined as a particular set of basis functions in the waveguide expan-
sion for FEL. They are obtained in the waveguide mode description by choosing
a refractive index for the virtual dielectric that is parabolic in the transverse de-
pendence. This is known as a quadratic index medium or QIM. The LG modes
are not necessarily eigenmodes of the FEL (though we will show they are often
quite close), but they are of particular interest for several key reasons. The first
is that they have the same transverse profile as their ubiquitous 3D counterparts
in descriptions of paraxial wave propagation. LG modes are also related by a
simple transformation to Hermite-Gaussian modes, which are also common, par-
ticularly for systems with substantial rectilinear symmetry. This establishes a
useful connection between the guided modes of the FEL and the guided modes of
an optical fiber. The fact that the guided LG modes are intrinsically related to
free-space LG modes enables a straightforward examination of the propagation
of radiation entering and exiting the undulator, as well as investigation of the
coupling characteristics of individual paraxial modes in the FEL interaction.
85
The second benefit of using LG modes as a FEL basis, which is central to the
theme of this work, is that they are eigenmodes of orbital angular momentum in
free-space propagation, as discussed in Chapter 1. Higher-order LG azimuthal
modes possess l~ units of OAM per photon, where l is the azimuthal mode num-
ber. The modal expansion technique provides a direct way to examine the OAM
mode content of the FEL, since the radiation field is expressed explicitly in terms
of the LG modes.
3.1 Guided Laguerre-Gaussian Modes
The basis eigenmodes used in the expansions of the fields and the density mod-
ulation satisfy the dielectric field equation
∇2⊥ uq (x⊥ ) + [nI (x⊥ )2 k 2 − kzq

2
]uq (x⊥ ) = 0. (3.1)
The spatial dependence of nI (x⊥ )2 in Eq. (3.1) fully determines the functional
form of the expansion basis in the excitation equations. To obtain the desired
guided Laguerre-Gaussian modes we choose the refractive index of a quadratic
index medium, with a specified radial dependence of the form:
2r 2
n2I (r) = n2I0 − , (3.2)
kw2
where nI0 is the refractive index on-axis and w is the mode expansion parameter.
By inserting Eq. (3.2) into Eq. (3.1) we obtain guided LG modes which have the
form [71, 72, 73],
s √ !|l|
r2
2
2p! p r 2 |l| 2r
uq (r, φ) = (−1) exp ilφ − 2 Lp (3.3)
(p + |l|)! w w w2
86
Pp
where Llp (x) = j=0 (p + l)!(−x)j /j!(p − j)!(l + j)! is an associated Laguerre
polynomial. The mode index now takes on two values; q = (p, l), corresponding
to the radial (p) and azimuthal (l) mode indices.1 For the modes to be mutually
orthogonal as required by the modal expansion, the expansion parameter w must
be real.2 In this case, we define it to be w = w0 where w0 is the waist size. The
normalization is then hup0 ,l0 |up,l i = δp,p0 δl,l0 πw02 . The mode power from (2.9) is
then given by
E02
Pq = πw02 , (3.4)
2µ0 c
assuming dominantly transverse modes.
The LG modes in (3.3) provide a convenient working basis to model the

FEL radiation for geometries that are largely axisymmetric over the interaction
length. They are identical in the transverse dependence to free-space LG fields
that satisfy the paraxial wave equation when the free-space modes are evaluated
at their optical beam waist. Thus, provided the paraxial approximation remains
valid (which it does in high-gain FEL), the guided LG modes can be used to
model the properties of gain-guided modes with OAM in the FEL system.
In this parameterization, the z-component of the time-averaged mechanical

torque (angular momentum per second) of the fields about the axis of propagation
due to the OAM is given by the sum over the mode amplitudes [74],
1 XX
τz (z) = l |cp,l (z)|2 Pp,l (3.5)
ω p l
|cq (z)|2 Pq . If the total field is domi-

P
where again the total power is P(z) = q
1
NOTE: The single index q will be used throughout this work to represent the pair of indices
p, l when the multiple indexes are cumbersome.
2
Later, in the case of only a single mode in the system, the orthogonality constraint is not
necessary and we permit w to be complex.
87
nated by a single mode with index l, the torque is:
l
τz (z) = P(z). (3.6)
ω
It should be noted that assuming the LG modes are dominantly transverse

does not preclude using them to explore the features of OAM amplification, even
though it is precisely the on-axis longitudinal fields that give rise to orbital an-
gular momentum. One reason is that the axial fields of OAM modes scale as
p
λ |l| + 1/w relative to the transverse fields and are therefore negligible in the
paraxial regime of interest here. This is particularly true for the lowest order
modes which are the only modes that can achieve reasonable coupling and ampli-
fication. The second reason they can be neglected is that the longitudinal fields
are not resonant with the ponderomotive fields that drive the interaction at the
fundamental frequency, and thus have no net contribution to the dynamics over
distances much longer than a single undulator period.
Inserting Eq. (3.2) and (3.3) into (3.1) we obtain the axial wavenumber
associated with each guided mode:
2 4
kzq = k 2 n2I0 − (2p + |l| + 1), (3.7)
w2
The axial wavenumbers of the guided LG modes differ from the wavenumbers of
vacuum paraxial modes [73, 75]. The waves are modified in the total wavenumber
both by the presence of the homogeneous dielectric contribution knI0 , and by a
transverse wavenumber factor attributed to the guided focusing of the paraxial
wave. Much like in the high-gain FEL, the modes are dispersive in the virtual
dielectric. They propagate with axial phase velocity ω/kz;p,l . Mode dispersion
will become an important (and, in principle, measurable!) feature of the high-
88
gain amplification process when multiple modes are simultaneously present and
competing in the e-beam (Section 3.5.4).
3.2 Dielectric Mode Coupling Factor
The dielectric mode coupling parameter Kq,q0 is defined in general in Eq. (2.8) and
can now be evaluated explicitly using the LG basis and the quadratic refractive
index from (3.2):
k2 4 huq |r2 |uq0 i

2
Kq,q0 = (nI0 − 1)δl,l0 δp,p0 − 4 2
2kzq w k huq |uq i
(3.8)
k2

2 2
= (nI0 − 1)δl,l0 δp,p0 − 2 2 hq,q0 ,
2kzq w k
where
p p 0
0 0
p X X (−1)j+j +p+p (j + j 0 + |l| + 1)!
hq,q0 = δl,l0 p!(p + |l|)!p0 !(p0 + |l|)! 0 !(p − j)!(p0 − j 0 )!(|l| + j)!(|l| + j 0 )!
.
0
j=0 j =0
j!j
(3.9)
For q = q 0 then hq,q = 2p + |l| + 1. Note that since n2I (r) in (3.2) is axisym-
metric, the mode coupling Kq,q0 vanishes between modes with different azimuthal
dependence (l 6= l0 ).
3.3 Scaled Parameters
Before moving forward, it is useful to write the relevant FEL parameters as

unitless scaled variables as listed in Table 3.1. By inspection we see that:
• The scaled gain parameter is simply the ratio δk i = L1D /L3D ≤ 1 which
quantifies the deviation of the calculated 3D power gain length from that
89
Table 3.1: Scaled FEL parameters
Parameter Symbol
Complex Wavenumber δk = 2δkL1D
Mode Expansion Parameter α = w2 /4σx2
Spot Size α0 = w02 /4σx2
Phase Curvature αR = L1D /Rc
Space Charge θp = 2θp L1D
Detuning θ = 2θ0 L1D
Diffraction ηd = L1D /2kσx2
FWHM Energy Spread δη = δηkL1D /γz2
RMS Energy Spread ηγ = σγ kL1D /γz2
Emittance η = 2kx kβ L1D
90
of the 1D model.
• The scaled expansion parameter α is general and can be complex in the

single mode description. For multiple modes it must be real and α = α0 .
• The real-valued spot size parameter α0 gives the relative overlap of the
optical mode intensity with the e-beam. If α0 ≤ 1, the rms spot size of the
optical intensity profile is less than or equal to the rms e-beam profile. If α
is complex then α0 = |α|2 /Re[α∗ ].
ηd ∗
• The radius of phasefront curvature is given by the parameter αR =− |α| 2 Im[α ].
It characterizes the transverse bending in phase of the guided mode wave.
• Longitudinal plasma oscillations are given by θp and are negligible over a

gain length if θp 1. When θp = 4π the beam undergoes a full plasma
oscillation over a 1D gain length.
• Detuning of the electron beam energy (or the wavelength) away from res-
√
onance is θ = (γR2 − γ 2 )/( 3γ 2 ρ) where γR is the resonant energy. When
θ < 0 the e-beam energy is above resonance.
• Diffraction effects can be gauged by ηd which is the inverse of the Fresnel

number of the e-beam over the distance of a gain length. The limit η → 0
corresponds to a 1D system.
• The uncorrelated energy spread is given either by δη corresponding to a

Cauchy-Lorentz distribution with a relative FWHM energy spread of δη,
or ηγ corresponding to a Gaussian distribution with a relative rms energy
spread of δη, depending on which distribution better models the system.
• Beam emittance and betatron focusing in a matched beam leads to a spread
91
in longitudinal velocities and transverse mixing characterized by the scaled
emittance variable η .
3.4 Bunching Factor
For the purposes of OAM mode amplification, we are interested in electron beam
bunching into discrete azimuthal modes l, since the coupling between the optical
field and the e-beam gives a direct link between the l−modes in the field and
the l−modes in the beam via δl,l0 (see Eqs. 3.15 or 2.93). We therefore define a
bunching factor for the azimuthal modes as
Ne
1 X
bl = e−iksj −ilϕj (3.10)
Ne j=1
where ϕj is the azimuthal position of the j th electron. In the continuous beam

limit from (2.36) we define
Z
1 X
bl (z) = ap,l (z) up,l (x⊥ )e−ilϕ dx⊥ (3.11)
2Ab p
where by inspection the overall bunching factor in (2.36) is just the bunching into
0
the l = 0 mode: b(z) = (2π)−1 l bl (z) eilϕ dϕ0 = b0 (z).
P R
Solving directly with the Gaussian beam distribution for f (x⊥ ) and the LG
mode expansion (3.3), we obtain a general expression in terms of the mode mod-
ulation amplitudes ap,l (z):
s
2|l|+1 (p + |l|)!

X (|l| /2)! |l|
bl (z) = α ap,l (z) 2 F1 −p; + 1; |l| + 1; 2 (3.12)
p
(−1)p |l|! p! 2
P∞ n
where 2 F1 (a; b; c; x) = n=0 (a)n (b)n x /(c)n n! is the hypergeometric series and
92
(a)n = Γ(a + n)/Γ(a) is the rising factorial or Pochhammer symbol. If only
a single radial mode is dominant, the bunching is directly proportional to the
density modulation amplitude for that mode. Considering only the fundamental
radial mode the simplification is straightforward,
s
2|l|+1
bl (z) = αa0,l (z)(|l| /2)! , (3.13)
|l|!
and is useful for calculating the limiting regimes of OAM mode amplification.
We note the physical requirement that bl (z) ≤ 1/2 for all z, which gives an upper
limit on the density modulation amplitudes.
3.5 Negligible Emittance
Transverse emittance in a matched beam can be essentially neglected if two con-

ditions are satisfied: the first condition is that the matched betatron oscillation
period be much longer than the gain length kβ L1D 1. This condition is written
√
equivalently as ηd η 1, and if satisfied, says that the transverse mixing of the
electrons in terms of the betatron phase advance over a gain length is negligible.
In this limit the electrons are assumed to be frozen transversely, ie, this is the
‘parallel-beam’ limit. Diffraction dominated FELs are characterized by ηd ≥ 1,
so this limit demands that for these FELs the scaled emittance be much smaller
than unity, η 1.
The second condition states that the transverse emittance can be ignored
if the longitudinal energy spread that results from the uncorrelated spread in
transverse electron velocities is negligible, i.e., η → 1. The contribution of the
emittance to the energy spread is given by Equation (2.84), which is written in
93
scaled variables as
2 2 2
ηγ(ef f ) = ηγ + η /4. (3.14)
Thus, in the presence of a non-zero uncorrelated energy spread, the thermal

motion due to emittance can be neglected if η 2ηγ . The requirement that the
effective relative energy spread be less than the bandwidth ρ gives an additional
upper limit on the scaled variables ηγ2 + η2 /4 < 1/3.
3.5.1 Relevant equations and coupling factor
In the absence of emittance effects, the coupled differential equations derived in

Sections (2.4)-(2.8) accurately describe the FEL interaction through the startup
and linear regimes. With a Gaussian transverse e-beam profile (2.92), the beam
profile overlap factor in (2.27) solved using the LG mode basis is [76],
0 0
δl,l0 (−1)p+p (p + p0 + |l|)! αp+p
Fq,q0 =p 0
p!p0 !(p + |l|)!(p0 + |l0 |)! (α + 1)p+p +|l|+1

0 0 1
× 2 F1 −p ; −p; −p − p − |l|; 1 − 2 .
α
(3.15)
3.5.2 Modeling of FEL with η , ηγ → 0
To get a sense of the general character of an FEL system with negligible emittance
described by the full evolution and supermode equations (2.25,2.49), we examine a
simplified FEL system described by the parameters in Table 3.2 and compare the
results of the theory to numerical simulations from Genesis 1.3. The model FEL is
similar in most aspects to the Visible to Infrared Spontaneous or Seeded Amplifier
(VISA) FEL at Brookhaven National Laboratory [77, 78], which operates either
in self-amplified spontaneous emission (SASE) mode or as an amplifier of input
94
Table 3.2: Model FEL parameters
Parameter Symbol Value
Undulator period λw 1.8 cm
Undulator length L 4m
Undulator Parameter K (K̄) 1.26 (0.89)
e-beam relativistic factor γ 123.18
Optical wavelength λ 1064 nm
Matched e-beam size σx 65 µm
Peak e-beam current I 300 A
Diffraction parameter ηd 1.35
Space charge θp 0.38
radiation in seeding mode. It provides a convenient FEL model for examination

of the cold-beam model in a diffraction-dominated system where emittance and
energy spread are both taken to have a negligible effect on the FEL process (at
VISA η , ηγ =0.05).
First we examine the FEL as a pure signal amplifier (no shot noise) of a
Gaussian input laser field at resonance, which is introduced coaxially to the un-
dulator and co-propagates with the e-beam. The frequency bandwidth of the
seed is assumed to be much smaller than the FEL bandwidth, so we consider
the behavior only at a single frequency (ie, time-independent simulations). This
simple scenario provides a starting point to examine the behavior of the input
seed mode as it initiates the FEL interaction and evolves toward the supermode.
Figure (3.2) shows the evolution of the radiation spot size and output power
as the seed is amplified from the initial P =0.1 µW level3 . Excellent agreement
3
This power is much less than the PSN ∼0.1W of equivalent 1D shot noise power in the
system, so the seed would be buried in the dominant SASE emission. But in the absence of
SASE for examination purposes, this small input prevents the system from saturating before
95
13.4
Α0=4
13.2
13.0 Α =3
L 3D [cm]
0
12.8
12.6
12.4 Α0=2
12.2
0 2 4 6 8 10
p
Figure 3.1: Predicted gain length at resonance for the model FEL as a function
of the maximum value of the radial mode p included in the excitation equa-
tions. Solid line is from Genesis. Different values of the expansion waist size
α0 = w02 /4σx2 generate a range of values for L3D with only the lowest radial
modes included in the expansion equations, but quickly converge to the correct
value as more higher-order modes are added.
Power Profile Evolution

105
12
1000
10
10
8
P @WD
0.1 Α0
6
0.001
4
10-5
2
10-7
0 1 2 3 4 0 1 2 3 4
z@mD z@mD
Figure 3.2: Evolution of the FEL power and optical mode spot size modeled for
the VISA FEL operating as a single-pass amplifier in the parallel e-beam limit.
A free-space, transversely Gaussian input seed is injected coaxially to the e-beam
with the waist positioned at the undulator entrance. The supermode evolves after
the fluctuations settle ∼1.5 m downstream of injection. Solid lines are solutions
to theory from Eq. (2.25) and points are from Genesis simulations.
96
is obtained between the solution to the full expansion equations and simulations
performed with Genesis 1.3 [79]. The system shows clear dynamic evolution to-
wards the supermode. The free-space input field comes to a waist at the undulator
entrance and undergoes a short period of diffraction during initial startup before
high-gain amplification develops. The dynamic behavior is also recorded in the
power where small fluctuations occur before the field eventually settles into fixed
gain, fixed profile propagation with a power gain length of 13.1 m. The effects of
longitudinal space-charge are naturally included. (This effect increases the gain
length by 11% from the case when the space-charge is ignored.) Note in Fig. 3.2
that the evolution calculated from the coupled excitation equations in Eqs. (2.25)
closely matches Genesis simulations which also include the effects of transverse
fringing in the space-charge fields by a Fourier decomposition. This suggests that
the “parallel plate capacitor” assumption still provides reliable results in model-
ing only the longitudinal space charge contribution, which is included in the full
mode expansion. This is an improvement from alternate approaches in which the
longitudinal space-charge is included as merely a single parameter for all modes
[80].
The mode expansion spot size in the equations is taken to be w0 = 4σx

(α0 = 4), in accordance with the best results obtained for efficiency in Fig. 3.1
for a cold, parallel beam. (Any value of w0 can be chosen, of course, since it is a
free parameter in the expansion. But for a finite number of modes, the greatest
efficiency is generally obtained by choosing a value close to that of the eventual
FEL eigenmode.) The rms e-beam size is fixed throughout the interaction length,
and is set to σx =65 µm as determined by the n =2 mm-mrad normalized trans-
verse emittance and external focusing lattice. This is motivated by the VISA
the end of the 4 m undulator, thus allowing useful comparisons of the numerical simulations
with the modal expansion in the linear regime
97
undulator transport which uses strong focusing in the undulator to reduce the
beam size from the larger matched beam value dictated by the vertical natural
undulator focusing alone. This smaller beam size pushes the FEL deeper into
a diffraction dominated regime and slightly increases the effects of space-charge,
but all in order to obtain a shorter gain length for the fundamental mode. (It is
an approximate 10% improvement overall, and assists in allowing VISA to reach
saturation during SASE within the four meter total length [78].) The strong fo-
cusing, however, adversely affects the gain of the higher-order modes by adding
significant transverse motion of the electrons during transport. As a result, the
actual VISA FEL works well for lasing at the fundamental with a short gain
length, but not so well for higher-order modes. See Section 3.6.1.1 for more on
the effects of focusing on the growth of higher-order OAM modes.
The output field distribution at the undulator exit is shown in Figure 3.3. Note
that, even though the diffraction parameter is larger than unity, the single mode
theory closely matches the exact solutions given by the full multimode description
for the fundamental mode profile. In this diffraction-dominated regime the optical
beam is larger than the electron beam.
Figure 3.4 shows the dependence of the relative power gain on detuning.
Again, excellent agreement with Genesis is observed. As long as the input power
of the EM seed is larger than the effective shot-noise power, SASE effects can be
ignored and detuning the system away from resonance provides an experimental
knob by which the gain length can be shortened and power coupling efficiency
between the e-beam and the input fields is enhanced [59]. Experimentally, the
simplest method to maximize the available efficiency is to detune by increasing
the e-beam energy above the resonant energy set by the input laser frequency
[81]. Here the minimum gain length is predicted at a 0.3% increase in the beam
98
1.0
0.8
0.6
0.4
0.2
0.0
0 1 2 3 4 5 6
rΣx
Figure 3.3: Fundamental mode profile; a comparison of the single mode theory
(thick tan line) with profile from the full multiple mode solutions (thin red line).
Thick dashed blue line is the e-beam.
Γ
124.14 123.59 123.04 122.5 121.95
0.56 7.62
0.54 7.28
@m-1 D
0.52
6.94
d Log HPL
∆k
dz
0.50
6.6
0.48
0.46 6.26
-1.0 -0.5 0.0 0.5 1.0

Θ
Figure 3.4: Detuning gain curve for seeded VISA, cold parallel beam limit. The
solid line is from the mode expansion theory, points are from Genesis simulations.
99
energy above resonance.
3.5.3 OAM mode amplification in η , ηγ → 0 system
The good agreement between the theory and the numerical simulations (which
have also been benchmarked to experiments) in the parallel-beam limit supports
the use of the modal description to explore the FEL system for different scenarios.
In this section we examine the more exotic operation of OAM mode amplification.
Just like Gaussian seed modes, OAM modes can be amplified in an FEL provided
that the input power of the mode is higher than the effective shot-noise power,
and that the mode is injected so that it can couple efficiently.
Figure 3.5 shows the evolution of the spot size and the differential power for an
l = 1 OAM optical seed on a Gaussian e-beam for the diffraction-dominated VISA
FEL in Table 3.2 without SASE. Much like the seeding of a simple Gaussian mode,
there is a brief period of startup followed by the emergence of the steady-state
OAM mode that grows exponentially and with a fixed profile. There is again
excellent agreement between the theory and Genesis simulations, both which
assume zero emittance, energy spread, and shot noise.
Figure 3.6 gives a flavor for the relative gain lengths one may expect for the
first few steady-state OAM modes, calculated using the variational principle in
the single mode approximation. Clearly the fundamental has the highest relative
gain across most of the FEL band. The wide separation between modes is char-
acteristic of diffraction dominated, three-dimensional systems, and grows with
the parameter ηd . This can be understood in terms of the longitudinal phase
velocity of each steady-state mode. Higher-order modes have larger transverse
wavenumbers (they diffract more), and thus smaller longitudinal wavenumbers.
The axial phase velocity is thus larger for these modes, so the electron beam must
100
4 12
11
(1/P)dP/dz
3
<r2>/Σx
10
9
2
8
1 7
6
0 5
0 1 2 3 4 0 1 2 3 4
z [m] z [m]
Figure 3.5: Evolution of an amplified l = 1 EM seed mode predicted from mode

expansion theory (solid line) compared with numerical simulations from Genesis
(points).
also move faster to keep up. Higher e-beam energies for a fixed seed wavelength
correspond to θ < 0, and thus, detuning to θ = −1.3 maximizes the gain for the
l = 1 mode in this case.
Figure 3.9 shows the comparative evolution of three different optical seed
modes. The input seeds are a Gaussian (fundamental) mode (p = 0, l = 0), the
first order radial mode (p = 1, l = 0) and the first OAM mode (p = 0, |l| = 1),
with the input optical waist located zw = 30 cm inside the undulator entrance
in each case. The parameters of the system are otherwise identical. The reduced
coupling for OAM modes demands that the system be detuned to the maximum
gain, (θ = −1.3 from Fig. 3.6), to achieve significant gain-guiding of the OAM
mode. This value of detuning is away from the peak growth rate value for the
radial modes, and therefore results in their slightly weaker overall coupling and,
accordingly, a larger steady-state spot size than what it would be at their peak
growth rate. The evolution of the power and spot size in this plot demonstrates
an essential feature of the coupling between transverse optical modes, and their
respective coupling to the e-beam. In each case, the development of an even-
tual FEL steady state mode is apparent. Only the OAM mode has a different
growth rate and spot size by the end of the undulator. This is because, for an
101
0.6
0.5
0.4
∆k
0.3
0.2
-8 -6 -4 -2 0
Figure 3.6: OAM mode gain curves predicted by single mode theory. Blue curve
is l = 0 fundamental, red is l = 1 and tan is l = 2.
axisymmetric e-beam profile f (x⊥ ) = f (r), the modes with different p values
couple directly to each other, but the different l modes do not. Thus, an ex-
cited higher-order radial mode can couple to all the other radial modes, with
the power in the higher modes cascading down towards the fundamental radial
mode of the system (usually the Gaussian). But the higher-order l modes do not
couple to each other. This, combined with the reduced gain length, demonstrates
that the character of the final output OAM mode is specifically determined by
the azimuthal number of the input seed field. The l-mode of the seed is the lone
azimuthal mode excited in the FEL interaction for the axisymmetric e-beam, and
thus defines the fundamental azimuthal mode of the final EM signal field. The
system evolves to steady-state, generating a dominant output mode defined by
the azimuthal mode number of the input. That is, the steady-state supermode
will have a unique spot size, gain length and helical phase, all defined by the
orbital angular momentum content of the seed and the parameters of the FEL
system. As a result, all of the gain delivered to the radiation field is deposited
102
1.0
0.8
0.6
0.4
0.2
0.0
0 1 2 3 4 5 6
rΣx
Figure 3.7: OAM mode profile comparison. Single mode theory (thick tan line)
and full multiple mode solutions (thin red line) predict different structure near
the beam edges, due mostly to the richer structure in the optical beam.
only into the initial l-mode. This demonstrates a simple scenario of pure OAM
light amplification. Again, in the presence of SASE, in order to ignore it’s effects,
one must seed with a power well in excess of SASE startup power, and also take
care that the total (potentially higher) gain in the fundamental does not allow
this SASE mode to compete with the final power in the desired l 6= 0 LG mode
(Section 3.5.3.2).
The reduced gain for OAM modes shown in Figure 3.6 is the result of two
main causes. First, OAM modes do not couple as well as the fundamental to
the Gaussian e-beam because the transverse optical fields vanish on axis. This
issue is even more important for higher-order OAM modes because the central null
becomes increasingly wider for larger values of l (see Fig. 1.7). Thus, the electron
beam centroid sits within a null in the field profile, as seen in Fig. 3.7. This
translates to a smaller value of the overlap factor Fq,q that describes the coupling
to that mode. The modes therefore cannot exchange energy as efficiently with
103
Intensity Phase
6 6
4 4
2 2
yΣx
yΣx
0 0
-2 -2
-4 -4
-6 -6
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
xΣx xΣx
Figure 3.8: Intensity and phase of l = 1 OAM mode amplified from seed at
undulator exit in Figure 3.5.
the e-beam, especially if the wiggle amplitude of the beam is small compared
to the optical mode size. The second reason the gain is reduced is that the
OAM modes tend to diffract more than the fundamental. Physically this is the
result of the larger transverse momentum for higher-order modes. With stronger
diffraction, the modes need even better coupling to the beam to be gain-guided
during the FEL process. Thus, the OAM modes have reduced gain due both to
poorer coupling and weaker gain-guiding. They can, nevertheless, be amplified.
3.5.3.1 Seeding Efficiency
In an optically seeded high-gain FEL, the longitudinal position and waist size of
the input seed may have a significant effect on the excitation efficiency of the su-
permode, especially for strongly-diffracting systems. Excitation efficiency in this
sense refers to the ability to couple and amplify an input. Maximum power output
is obtained for a proper balance between a small injection spot size to maximize
power coupling to the e-beam, and a large spot size to minimize power loss from
104
8 9
6 8
dLogPdz
7
rΣx
4 6
2 5
4
0 3
2
z @mD z @mD
0 1 2 3 4 0 1 2 3 4
Figure 3.9: Comparison of signal field propagation for FEL seeding with a pure
Gaussian mode (blue), first order radial mode p = 1 (tan), and an l = 1 OAM
mode (red) in a diffraction dominated FEL. Coupling between radial modes is
evident, while the azimuthal OAM mode establishes it’s own characteristic growth
rate and spot size.
diffraction. Once the supermode is established, it will grow exponentially ac-

cording to the specifics of the FEL system, independent of the initial conditions.
But the total amount of gain in the system for a given longitudinal position is a
function of how long it takes to establish exponential gain. The longer it takes to
reach, the lower the gain and thus the lower the FEL amplification efficiency. The
relationship between the efficiency and the gain characteristics is also of practical
interest for prediction of the optimal spot size and axial waist position of the
injected seed mode.
√
3Q1/3 z
In a 1D FEL, the exponential gain regime satisfies G1D (z) = (1/9)e ,
where Q is the 1D coupling parameter (2.29). In the 3D scenario the gain is
expected to differ from the 1D model, both in the exponential dependence and in
the 1/9 proportionality. To quantify this effect, the exponential gain is expressed
as
1 √ 1/3
G3D (z) = gef f e 3Q z . (3.16)
9
where gef f is the 3D supermode excitation efficiency and Q is the 3D coupling;

a generic version of (2.26) that is the total coupling for the FEL supermode,
105
a) b)
7 7 1.2
6 6
1.0
5 5
zw / L1 D
zw / L1 D
4 4 0.8
3 3 0.6
2 2 0.4
1 1
0.2
0 0
2 3 4 5 6 7 2 3 4 5 6 7
geff
ws / Σx ws / Σx
Figure 3.10: The supermode excitation efficiency gef f for a) Gaussian EM seed
injection at an optimized detuning value corresponding to the minimum gain
length θ = −0.5 and for b) an l = 1 OAM EM seed mode injected at optimized
detuning, θ = −1.3
and may include the contribution from several expansion modes q. Scanning the
efficiency for different FEL mode amplification schemes allows assessment of the
optimal running conditions.
Figure 3.10 shows the dependence of gef f on the longitudinal position and
waist size of both a Gaussian input seed and an |l| = 1 OAM seed for the model
FEL parameters in 3.2. In both cases the system is detuned to maximize the
growth rate for that particular mode. Results for both modes indicate that the
peak efficiency is obtained when the seed waist is approximately three times the
rms e-beam size, and roughly two 1D power gain lengths inside the undulator.
The OAM mode has a lower overall efficiency, as one might expect, as well as a
slightly narrower injection window for the optical beam size. That is, the effi-
ciency drops rapidly to zero if the optical beam is too big, while if it is too small,
the optical beam essentially just diffracts away, and there is little amplification.
106
1010
106
P [W]
100
0.01
10 -6
0 1 2 3 4
z [m]
Figure 3.11: Without an input seed, the FEL instability will still develop from
intrinsic shot noise in the beam. Modal theory (blue) including density mode shot
noise as an initial condition compared with time-dependent numerical Genesis
simulations with shot noise (red).
3.5.3.2 Including effective shot noise contributions
Up to this point we have considered FELs acting as amplifiers of pure input seeds,
and have ignored the effects of shot noise in the electron beam. This has served
to illuminate some of the primary features of the FEL process; the evolution from
the initial startup period through the high gain regime and the accompanying
guided modes that define the system according to the idealized inputs. In such
systems the azimuthal modes are fixed by the input fields, and pure OAM modes
can be amplified without pollution from the fundamental, or other modes.
In reality, however, noise in the beam (or in the seed in general) provides a
source to amplify many of the modes in the FEL bandwidth. As an example,
Figure 3.11 shows how, without an external EM seed, the VISA-like FEL will still
evolve into the exponential regime via amplification of the shot noise in the e-
beam. Here, the modal expansion solutions are compared with time-independent
107
Genesis simulations. Excellent agreement is shown through the build-up and lin-
ear growth regimes until the system saturates at P ' |b|2 ρPb , where b = 0.7.
Zero initial input field cq (0) = 0 and velocity modulation avq (0) = 0 are assumed,
while the initial density modulation is taken from the shot noise equation (2.44)
rD 2 E
with aq (0) = aSN
q (0) R . The FEL process starts from statistical noise in
dω
the e-beam, eventually developing into high-gain where the fundamental trans-
verse mode is amplified and dominates the system. This behavior is typical of
modern SASE FELs and is in many respects enormously useful since it provides
a mechanism by which no external seed is required to achieve high-brightness
emission at the resonant FEL wavelength.
In our present study of OAM amplification, however, SASE and other noise
can disturb our attempts to operate the FEL as a ‘pure’ OAM amplifier (Section
3.5.3). The OAM seed must contend with the amplification of the other modes
during startup that tend to spoil the desired mode purity, especially if the optical
power in the other modes is comparable. For SASE this issue can be practically
overcome, as mentioned previously, by seeding with an input (either EM or mi-
crobunching on the e-beam) with power well above the effective shot-noise power
so that the SASE signal is negligible amid the OAM signal. In general this works
for a while, particularly if the input seed power is just a few gain lengths short
of the saturation power. But the fundamental has the shortest gain length and,
for smaller input OAM power levels, can eventually catch up unless the system
saturates or the interaction is stopped. It is thus useful to calculate the approxi-
mate minimum input power needed, as a function of length along the undulator,
to obtain a dominant OAM mode before other modes excited by SASE take over.
Strictly speaking, the minimum power required to seed the system with an
OAM mode and then extract it as the dominant mode depends on the details of
108
Figure 3.12: The region in which a seeded l = 1 mode is predicted to dominate
over 1D SASE for the VISA-like FEL is shaded. The curve is calculated according
to Eq. (3.18), and the optical modes are calculated with the full modal theory
with P0 (0) = PSN .
109
the system, and specifically how well the mode couples during injection. Nev-
ertheless, we can provide a heuristic power baseline in a simplified scenario.
Consider a system where only two modes are amplified; one is the fundamen-
tal mode and the other is an OAM mode. Let us further assume that the two
modes are independent and uncoupled. This is valid in the linear regime and
necessarily true for differing azimuthal modes due to the orthogonality between
different l. Ignoring the startup period, the output power in the mode q grows
as Pq (z) ' Pq (0)exp(z/LG,q ) where Pq (0) is the initial power in the mode at the
start of exponential gain. If the OAM mode is to dominate its power must be
larger than the fundamental, Pl (z) > P0 (z), for all z up to the end of the high-
gain interaction. The constraint on the OAM input power at the position zmax
is thus
1 1
Pl (0) > P0 (0)exp zmax − . (3.17)
LG,0 LG,l
Of course, the input seed power has an upper bound set by the saturation power
ρPb in order to achieve gain. Equation (3.17) is a general but somewhat weak
inequality, since we want the OAM mode to truly dominate not just be slightly
larger. We can thus place an even stronger estimate the minimum OAM input
power by asserting that it must compete with a mode with the shortest possible
gain length; one that grows according to the 1D limit. The stronger constraint
is then written in unitless form as

P̂l exp ẑ 1 − δk i . (3.18)
where P̂l = Pl (0)/P0 (0) is the ratio of the OAM and 1D input mode powers and
ẑ = zmax /L1D is the position from the beginning of exponential gain where the
interaction is stopped, in units of the 1D gain length. The scaled gain length is
δk i = L1D /LG,l , and can be found by iteration or as a power-law function of the
110
1000
10
P1 P0
0.1
0.001
0.01 0.05 0.10 0.50 1.00 5.00 10.00

b1 %
Figure 3.13: Relative OAM mode power versus initial helical bunching factor for
VISA-like FEL. The dashed line is at resonance, while the solid line is detuned
to minimize the length in the OAM mode.
scaled parameters of the system in the single mode approximation as in Sections

3.5.6 and 3.6.2.
An illustration of the two regions defined by Eq. (3.18) is shown in Figure

3.12 in which an input l = 1 OAM mode competes with SASE for dominance
in the VISA-like FEL (parallel beam limit). The scaled OAM mode gain length
is calculated from the fitting parameters in Eq. (3.29). The shaded region is in
qualitative agreement with simulations, provided the electromagnetic OAM seed
is approximately optimized to couple effectively to the e-beam, as illustrated by
Fig. 3.10.
The FEL OAM mode can also be seeded by a helical structure in the e-beam
(see Chapter 4). In this case it is more useful to consider optimization of the initial
helical bunching factor |bl |2 . By analogy with the power, it should be much larger
than the effective bunching factor due to shot noise (2.60), but smaller than the
111
bunching factor at saturation,
1 1
|bl |2 < . (3.19)
9Nc 2
Since |bl |2 grows exponentially with the power, and it is usually ideal to allow a
handful of gain-lengths of amplification before the interaction saturates, an initial
bunching factor of |bl | ∼ 1 − 5% is typically a reasonable starting point for most
modern FELs, but may be smaller if other schemes are used to further suppress
the shot noise in undesirable modes [82].
Figure 3.13 shows how the power in the l = 1 OAM mode P1 compares to
the power in the fundamental mode P0 at saturation, as a function of the initial
helical bunching factor. The FEL is seeded both by the helical beam and by shot
noise according to (2.44). Again, detuning to maximize the growth of the OAM
mode clearly delivers far more power into the OAM mode at saturation for the
same initial helical bunching factor.
3.5.4 Longitudinal Dispersion
Unless the seed power of the OAM modes is large enough to dominate for all
z, multiple modes can simultaneously compete for dominance. In this case the
system will dynamically evolve, with the seeded OAM mode eventually being
over-powered. For illustration, again consider the simple case of only two modes
being simultaneously amplified; the OAM mode and the fundamental. In addition
to the different characteristic gain-lengths of the modes, they also have different
values for the effective index of refraction. This is determined by the associated
112
Intensity Phase
2 2
1 1
y/r 0
y/r 0
0 0 z=0
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
x/r 0 x/r 0
2 2
1 1
x
0 0 z=L/4
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
1 1
0 x 0 z=L/2
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
1 1
0 0 z=3L/4
x
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
1 1
0 0 z=L
x
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
Figure 3.14: Intensity and phase along the undulator for an l = 1 seed on a cold
beam, prebunched at the fundamental. The fundamental mode grows faster than
the OAM mode, appearing as a bright spot (x) on the annular intensity pattern
and eventually pushing the phase singularity, located at the central field null, off
to the side. The different effective phase velocities of the modes is manifest in
the rotation of the bright spot along the undulator.
113
λ
l=0
l=1
Figure 3.15: Side by side schematic comparison of geometric bunching due to

different modes. The fundamental mode (top) tends to generate separate bunches
along the z-axis. The OAM modes (bottom) tend to bunch the beam in helices
due to the azimuthal component of the ponderomotive phase. If both modes are
simultaneously present, the electrons on one side of the e-beam (thick arrows)
are resonant (solid vertical line), while those in a different transverse position
(slender arrows) are bunched out of phase (dashed vertical line). The intensity
therefore grows off-axis, where the electrons are in-phase.
real part of the complex wavenumber, δkr , and is defined for each mode as,
δkr
nef f = 1 + . (3.20)
k
The different modes each have different phase velocities, resulting in longitudinal
modal dispersion in the e-beam [63]. In optical modes with a significant transverse
variation in intensity and phase like OAM modes, this can be an experimentally
observable effect as the two modes shift in overall phase with respect to each
other, causing portions of the transverse profile to change in intensity along the
undulator.
Figure 3.14 shows this effect for an OAM mode that is seeded and amplified
simultaneously with the fundamental mode, which itself grows due to a small, but
non-zero value of pre-bunching as in the case of SASE. Parameters are similar to
the SPARC FEL [83], with ηd = 0.5 and θp = 0.25 (assuming ηγ , η = 0). The
114
Intensity Phase
2 2
1 1
y/r 0
y/r 0
0 0 z=0
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
x/r 0 x/r 0
2 2
1 1
0 x x 0 z=L/4
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
1 1
x z=L/2
0 0
x
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
1 1
0 xx 0 z=3L/4
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
1 1
0 0 z=L
-1 -1
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
Figure 3.16: Intensity and phase along the undulator for an l = 2 seed on a cold
beam prebunched at the fundamental. The axial phase singularity is broken into
two l = 1 vortices as the fundamental grows.
115
OAM mode in Fig. 3.14 is injected with enough power such that both modes
have the same power at the undulator exit. Evident from the plots in Figure
3.14 are both the off-axis growth of the intensity which pushes the axial field
null aside, and the rotation of the peak intensity spot along z as both modes
are amplified. Both effects are explained by the helical structure of phase of
the OAM field as it is manifested in the e-beam microbunching distribution. In
general, the phase is such that amplification of the pure OAM mode tends to
bunch the e-beam into a continuous helix (or multiply twisted helices for |l| > 1)
longitudinally. The fundamental mode also tends to bunch the beam, but into
longitudinally separated microbunches (Fig. 3.15). With amplification of both an
OAM mode with index l and the fundamental mode in the system, the electrons
are resonant on one side of the e-beam, but are lπ out of phase on the opposite
side. This means that there is higher growth of the signal field in the transverse
position(s) where the electrons are mutually bunched, and the intensity then
grows off-axis. This also explains why the evolution toward the fundamental in
an idealized system does not begin with the peak intensity growing directly from
the axial center of the beam, as would be the case if the initially dominant OAM
mode were not present. The off-axis intensity bulge grows along the undulator
both transversely and in amplitude due to the higher gain of the dominant mode,
rotating about the axis as the two modes shift out of phase by ψ0 = |∆nef f |k∆z,
where ∆nef f is the difference in the effective refractive index between the modes
in Eq. (3.20), and ∆z is the length traveled along the undulator during high-
gain. Thus, OAM modes with azimuthal mode numbers differing by ∆l and with
comparable amplitudes interfere constructively off-axis to generate an intensity
peak which rotates about the axis by ψ0 /∆l radians over the length ∆z. In
addition, since the different pure modes also have different intrinsic radial profiles,
the azimuthal motion of the intensity peak is accompanied by motion in the radial
116
HaL HbL
1 0.5
0 0.4
-1
∆kr D∆kr 0.3
-2
0.2
-3
0.1
-4 -3 -2 -1 0 1 -4 -3 -2 -1 0 1
Θ Θ
Figure 3.17: The real part of the scaled complex wavenumber is shown in (a)
for the l=0,1 and 2 (blue, red, and tan respectively) azimuthal modes. The
differences in δk r between modes are shown in (b), where the dark line is the
difference between the fundamental and the l = 1 mode, and the light line is the
difference between the fundamental and the l = 2 mode. In both plots ηd = 0.5,
θp = 0.25 and δη, η = 0.
direction as the initially off-axis peak moves toward the radial maximum of the
dominant mode of the system – typically the axial center for an axi-symmetric
beam.
For the cold, parallel beam scenario in Fig. 3.14 the calculated phase shift
between modes is ψ0 ' 2π/3. This is calculated using the values of the refractive
index for the OAM seed and the fundamental given by the l = 1 and l = 0
supermode solutions. This phase shift is confirmed by visual inspection from
HaL HbL
0.30
0.25 0.28
0.20 0.26
0.24
∆kr 0.15
D∆kr 0.22
0.10
0.20
0.05 0.18
0.00 0.16
0.00 0.05 0.10 0.15 0.20 0.25 0.00 0.05 0.10 0.15 0.20 0.25
∆Η ∆Η
Figure 3.18: Same content as Figure 3.17 at resonance θ = 0 for different values
of energy spread.
117
the motion of the peak intensity spot which travels approximately 1/3 of the
way around the axis over an axial propagation distance of ∆z = 3L/4. Detun-
ing, energy spread and space charge all affect the amount of observed rotation.
This is because the phase difference between the modes is a function of ∆nef f ,
which varies for different values of of the associated real part of the complex
wavenumber. For example, it is found here that |∆nef f | is larger (faster rota-
tion) with increasing energy spread (keeping other parameters fixed) between the
fundamental mode l = 0 and OAM modes |l| > 0.
In Fig. 3.16, an l = 2 OAM mode is seeded onto the same pre-bunched beam
as for Fig. 3.14, so ∆l = 2 between the amplified modes. The initial phase
at z = 0 is π rotationally symmetric (versus the 2π rotational symmetry for
|l| = 1), so the e-beam that is bunched into a double-helix is resonant with the
fundamental at two off-axis locations, which in turn drives two intensity peaks
that appear in opposition at z = L/4. The peaks rotate along the undulator axis
according to the ψ0 ' 2π × (0.31) phase slippage between modes over ∆z = L/2
meters. This results in a rotation of ψ0 /∆l ' 0.98 radians (∼ 56◦ ) before the
peaks appear to merge. The direction of rotation is also in the same clockwise
sense as in the seeding shown in Fig. 3.14, since both OAM seeds have positive
values of l. The gain of the l = 2 mode is less than the l = 1 mode so the
fundamental clearly dominates at z = L. This contrasts Fig. 3.14, where the
influence of the OAM mode is still recognizable at the end. It is interesting to
note that the field null for the l = 2 mode is overtaken dominant fundamental
from two sides, and that the single phase singularity at the undulator entrance
is broken into two distinct phase vortices, each with topological charge 1. This
occurs in the interaction between the fundamental and all modes with |l| > 1 due
to the symmetry in phase about the axis; the fundamental peak evolves toward
the center, fragmenting the phase singularities and taking their place on axis.
118
Future investigations could shed light on this issue in detail, particularly local
and global evolution of the optical angular momentum.
3.5.5 Single Mode Limit
Having examined the full solutions described by the multiple-mode equations in

(2.25), it is useful to understand the general behavior of the FEL in the limit of a
single dominant mode. After startup, a lone mode dominates the optical field in
an undepleted system. This mode evolves in the system naturally, as in the case
of SASE, or can be imposed by arranging the initial conditions properly. Either
way, the mode is distinct from the other modes in the system.
In the single mode limit we consider only a solitary optical mode q = q 0 ≡ q0

in the beam at a time. Each is one of the steady state TEM modes (T EM00 ,
T EM01 , T EM10 and so on) of the FEL system. In this parallel beam limit, the
single modes are found with (2.75),
3
h
2 2
ih ηd i 2
(δk − θ − iδη) − θp Fq0 δk + (2p + |l| + 1) = − √ Fq 0 . (3.21)
α 3
Each mode is defined by a unique value of the mode expansion parameter α

(hence the non-orthogonality between steady state modes), which is found by
applying the variational condition δ(δk)/δα = 0. The scaled complex wavenum-
ber δk and expansion parameter α can then be determined by solving the two
resulting equations [41, 42]. Figures 3.19, 3.20 and 3.21 plot the behavior of the
fundamental and l = ±1 OAM modes of the FEL system described by (3.21) for
various regimes.
Figure 3.19 illustrates a basic feature of the 3D regime for a single mode for
different values of ηd . The degeneracy between the modal gain curves is lifted
119
from the 1D case, and the gain is peaked at different values of detuning according
to the mode number. Again, this is a direct result of the modal dispersion, and
is in agreement with the trend shown in Fig. 3.6 for the VISA FEL. To be at the
peak in the gain curve, the electron beam energy is different for different modes
which can be selected for dominance by detuning with respect to the external
seed. Figures 3.19, 3.20 and 3.21 show the gain curves for the l = 0, ±1 modes
in the presence of energy spread and space charge effects, both of which serve
to reduce the scaled gain, but not dramatically affect the overall separation in
detuning space. Thus, as before, since there is no cross-coupling between different
azimuthal modes, once the OAM mode is excited it is amplified without exciting
any other l-modes in the system. Note that because the gain curves depend only
on the magnitude of l, the azimuthal modes l = ±1 are degenerate. Therefore, in
combination with detuning, generation of a dominant OAM mode is achieved by
seeding the system either with an EM OAM seed with a helical phase, or with a
helically microbunched e-beam that will preferentially excite an OAM mode. We
have examined the former in the previous sections, while the latter forms the basis
of the proof-of-principle experiment described at the end of this dissertation.
Returning to the parallel-beam single mode solutions given by (3.21), let us

assume for the sake of simplification that the effects of energy spread and space-
charge can be neglected (δη, θp → 0) and derive some simple analytic relations for
the parameter δk and the spot size parameter α. In the limit that the diffraction
parameter is small ηd < 1 the radiation field diffracts little over the interaction
gain length and the optical mode size is assumed to be small such that to lowest
order in α the mode overlap parameter is Fq0 ' 1−α(2p+|l|+1). The variational
condition δ(δk)/δα = 0 then yields the following two equations for the unknown
120
1.0
0.8 a)
0.6
∆k b)
0.4
0.2
0.0
-4 -3 -2 -1 0 1
Figure 3.19: 3D effects on mode gain. The dashed line is the degenerate 1D limit
of ηd → 0 where all the modal gain curves overlap. For ηd = 0.5, the 3D effects
separate the modes into distinctive gain curves: a) T EM00 mode and b) T EM01
mode. Here η , θp , δη = 0.
0.7
0.6 a)
0.5
∆k
0.4
0.3 b)
0.2
-4 -3 -2 -1 0 1
Figure 3.20: Effect of energy spread on the a) T EM00 mode and the b) T EM01
modes with ηd = 1, θp , η = 0. Dashed curves are each mode in the absence of
energy spread, solid lines are each mode calculated with δη = 0.1.
121
0.7
0.6
a)
0.5
∆k
0.4
b)
0.3
-4 -3 -2 -1 0 1
Figure 3.21: Space charge effect on the a) T EM00 mode and the b) T EM01 modes
with ηd = 1, δη, η = 0. Dashed curves are each mode with zero space charge.
Solid lines are the labeled modal gain curves calculated with θp = 0.5.
quantities δk and α:
3
2
h ηd i 2
(δk − θ) δk + (2p + |l| + 1) = − √ [1 − α(2p + |l| + 1)]
α 3
3 (3.22)
2
ηd (δk − θ)2 = − √ α2 .
3
In the limit ηd → 0 we recover the 1D gain solution given in (2.97) which has a
maximum growth at θ = 0:
1
δk = √ − i. (3.23)
3
To the next orders in ηd , the 3D gain parameter and complex expansion
122
parameter at resonance are given approximately by
" s #
1 1 2ηd
δk ' √ −i 1− √ (2p + |l| + 1)
3 3 3
(3.24)
1
q √ √
α' 2ηd 3( 3 + i).
4
These expressions are in good agreement with the full numeric solutions for ηd <
1, and accordingly for small values of l (See Figures 3.22 and 3.23). The scaled
q √ q √
spot size is α0 = 2ηd / 3 and the scaled phase curvature is αR = − ηd 3/6,
which gives;
√ 1/4
w0 /σx =2(2ηd 3/3)1/4 = 2.07ηd ,
s (3.25)
6
Rc /L1D = − √ = −1.86ηd−1/2 .
ηd 3
It is useful to note that the single Gaussian mode description works very well
for FELs that satisfy the small diffraction parameter condition ηd < 1. (It can
also be a good approximation for slightly larger values of ηd , but the accuracy
is diminished for modes larger than the fundamental.) In the weakly diffracting
regime, the guided optical beam is contained mostly within the electron beam
and is therefore adequately modeled by a single, simple mode structure such as
a Gaussian. In contrast, a large diffraction parameter ηd 1 tends to describe
systems in which the optical mode has richer structure, and may have peripheral
structure outside the e-beam. The diffracting optical fields do not couple well near
the edge of the electron bunch, so the profiles tend to be somewhat pinched in the
center or distorted near the edges. These profiles differ enough from the presumed
Gaussian shape that the present single mode description becomes untrustworthy4 .
4
The full multi-mode solutions, however, correctly describe diffraction-dominated FELs
where the steady-state optical beam is large compared to the e-beam (See Section 3.5.2)
123
1.0
0.8
c)
0.6
Re{Α} b)
0.4
0.2 a)
0.0
-4 -3 -2 -1 0 1
Figure 3.22: Real part of the single Gaussian mode expansion parameter for
different values of diffraction parameter. a) ηd = 0.1, b) ηd = 0.5 and c) ηd = 1.
Here η , θp , δη = 0.
In this regime one should consider using a more suitable basis with modes that
share some of the same features of the eventual steady-state modes.
The transverse e-beam distribution affects both the structure of the guided
modes and the modal detuning curves. Of course, the guided optical mode cannot
be very much larger than the finite e-beam profile since the gain-guiding process
cannot take place in regions where there are no electrons. This places a sharp
constraint on the maximum optical mode size guided in hard-edge e-beam profiles
such as the flat-top or uniformly-filled ellipsoidal distributions where the electrons
are enclosed within a well-defined boundary. The hard-edge boundary leads to
sharper cutoffs in detuning where the gain peaks of the individual modes are
distinctly separated [44]. Here, however, in assuming a Gaussian model of the
electron beam profile which extends smoothly to infinity, the constraint on the
maximal size of the optical modes is relaxed and the detuning curves are smoothed
like in Fig. 3.19.
Before concluding this section it is worth mentioning a mathematical subtlety
124
1.0
c)
0.8
b)
0.6
Im{Α}
0.4
a)
0.2
0.0
-4 -3 -2 -1 0 1
Θ
Figure 3.23: Imaginary part of the single Gaussian mode expansion parameter
for different values of diffraction parameter. a) ηd = 0.1, b) ηd = 0.5 and c)
ηd = 1. Here ˜, θp , δη = 0.
in the treatment of the expansion parameter α. In the mode expansion formal-

ism, more than one mode is included to describe the complete solutions. The
√
parameter α = α0 is a real number in this case and 2σx α0 = w0 is the spot size
of the expansion mode. Insisting that α be real in the expansion insured that the
LG modes were orthogonal to each other and that the virtual refractive index
was real, i.e., that it had no gain or attenuation properties. This is not a problem
for describing the guiding nature of the FEL gain process since collectively, the
sum of different LG modes with complex amplitudes cq (z) or sq gave rise to the
characteristic phase front curvature in the optical beam which is amplified and
guided by the electron bunch. The sum of all the modes results in a phasefront
that depends on the transverse coordinate (ie, inwardly curved towards the lasing
√
center of the beam), and a spot size that may vary significantly from 2σx α0 ,
depending on the choice of α0 .
In the single mode scenario, things are a bit different. The curved phasefront
cannot arise using only a single mode unless α is a complex number, given the
125
definition of the expansion modes in (3.3). In fact, the single mode solutions
to (3.33) demand that it be complex since each of the single modes is an ap-
proximation of an FEL supermode, which necessarily has a curved phasefront.
A complex valued α violates the complex-conjugated orthogonality condition of
the LG modes (which strictly speaking, become biorthogonal to a related but
distinct set of functions), but this does not prevent use of a complex-valued spot
size parameter in the single mode description since only a single mode is being
considered. The fact that orthogonality was used to derive the full mode expan-
sion equations which were then reduced to find the single mode solutions doesn’t
matter since we could have simply started with the assumption that there is only
a single mode in the system to begin with, and that mode need not be part of
an orthogonal set5 . Thus, orthogonality requires α to be real in the multimode
description given by (2.75) or (3.30). But the analysis for the single modes allows
for α (and hence, w) to be complex-valued.
3.5.6 Fitting Formula for 3D Gain Length with Space Charge
In the absence of emittance effects, the gain length of each optical mode in the
FEL is determined by four scaled parameters of the system: the diffraction pa-
rameter ηd , the fwhm energy spread δη, the space charge θp and the detuning θ.
The complicated dependence precludes writing a simple analytic expression for
the FEL gain that includes all contributions. Here however, in order to facilitate
quick determination of the gain length, we generate a handy fitting formula cal-
culated numerically from the full solutions to Eq. (3.21). This technique follows
the example set by Xie [84], but here includes the effect of space-charge (and
5
This is bolstered by the fact that the FEL supermodes are themselves not orthogonal
to each other since each one has different gain which, due to the gain-guiding, leads to a
supermode-dependent spot size that thus renders the modes non-orthogonal.
126
Figure 3.24: Fit function surface for T EM00 mode. Points correspond to numer-
ical solutions to (3.21).
neglects emittance). The scaled gain parameter δk i = L1D /L3D of a given mode
q0 = (p, l) is written as,
1
δk i = (3.26)
1 + Λp,l
where
Λp,l = a1 ηda2 + a3 δη a4 + a5 θp a6 +a7 ηda8 δη a9 + a10 δη a11 θp a12

(3.27)
+ a13 ηda14 θp a15 + a16 ηda17 δη a18 θp a19 .
With the detuning adjusted to yield the shortest gain length given approx-
imately by the fitting formula θ ' −0.52δη 1.03 + 0.25δη 1.08 θp 1.76 − 0.74θp 1.85 +
0.3δη 0.83 θp 1.39 ηd0.19 − 0.33ηd0.31 + 0.14θp 3.61 ηd0.42 + 0.47δη 1.09 ηd0.76 , the mode fitting
127
Figure 3.25: Fit function surface for T EM01 mode.
coefficients for the relative gain of the T EM00 mode are
Λ0,0 = 0.45ηd0.56 + 0.99δη 1.15 + 0.21θp 1.78 +2.83ηd0.81 δη 1.46 + 0.21δη 0.63 θp 2.14
+ 0.08ηd0.53 θp 1.6 + 18ηd1.17 δη 5.38 θp 0.97 ,

(3.28)
where the range of the calculated variables is given by 0 ≤ ηd ≤ 1, 0 ≤ δη ≤ 0.5
and 0 ≤ θp ≤ 1. Note that a similar formula is given in [85] for FELs in which
the energy spread distribution is Gaussian, rather than Cauchy-Lorentz.
For the first order OAM mode – a T EM01 mode – the coefficients are
Λ0,1 = 1.1ηd0.57 + 0.94δη 1.02 + 0.23θp 2.04 +26.58ηd1.29 δη 1.97 + 28.91δη 5.62 θp 2.14
+ 0.16ηd2.44 θp 0.05 + 1.35ηd0.63 δη 0.61 θp 1.13 ,

(3.29)
for an optimized detuning of θ = −0.52δη 1.02 − 0.76θp 1.54 + 0.69δη 1.50 θp 2.71 +
0.38δη 0.16 θp 0.80 ηd0.27 − 0.98ηd0.48 + 1.16δη 1.25 ηd0.58 − 0.24θp 0.27 ηd2.37 . Here the range
128
on the relative energy spread is reduced to 0 ≤ δη ≤ 0.35.
For both the T EM01 and T EM01 modes the error between the calculated
numerical points and the value of the fitting function is < 5% in the specified
ranges. Between the three variables, the diffraction parameter is overwhelm-
ingly responsible for generating the largest difference in gain lengths between the
T EM00 and T EM01 modes compared with equivalent incremental changes in the
other two parameters. Since the fundamental optical mode driven by SASE is
intrinsically self-tuned to the maximal gain, in order to optimize a given system
for T EM01 mode production it is advantageous to minimize ηd at the expense of
a complementary increase in the energy spread or longitudinal space charge.
3.6 Emittance
The effects of transverse emittance can play an important role in the dynamics of
the FEL system. This is particularly true for short-wavelength (VUV and x-ray)
FELs in which 2kx is generally greater than unity. With the LG modes in hand,
the exact steady-state solutions with emittance are computed from (2.91) for a
given e-beam profile. Assuming a β-matched Gaussian e-beam distribution of
the form shown in (2.92) the FEL supermodes calculated from Eq. (2.93) with
the LG basis satisfy the integral equation:
" 3 Z #
∞
X 2 −iτ (δk−θ)−2τ 2 ηγ2
(δk − ∆k q )sq = − sq 0 Kq,q0 − √ τ dτ e Fq,q0 (τ ) (3.30)
q0
3 0
where x = 2L1D x for the parameter x if it has units of inverse length and x =
x/2L1D if it has units of length. The effects of space-charge are ignored since
the short wavelength systems of interest typically rely on high energy e-beams
for which the space-charge effects can be neglected over the entire interaction
129
Figure 3.26: Comparison of the effect of space charge and diffraction on the scaled
gain length of the T EM00 (unfilled mesh) and T EM01 modes for δη = 0
130
length.
The coupling parameter Fq,q0 (τ ) is given in complete form for all LG modes
as
h √ i|l|
4αδl,l0 (−1) p+p0
(p + p + |l|)!χp−
0
0
4α(1 − iτ η ) cos (2τ η ηd )
Fq,q0 (τ ) =p i−p
p+p0 +|l|+1
p!p0 !(p + |l|)!(p0 + |l|)!χ+
h √
4α2 (1 − iτ η )2 − sin2 (2τ η ηd )

0 0 χ+ (χ− − χ0 )
× 2 F1 −p ; −p; −p − p − |l|;
χ− (χ+ − χ0 )
(3.31)
with
√
χ± =4α2 (1 − iτ η )2 + 2α(1 − iτ η )(1 ± 1) ± sin2 (2τ η ηd )
√ (3.32)
8α2 (1 − iτ η )2 cos2 (2τ η ηd )
χ0 = √ .
2α(1 − iτ η ) − sin2 (2τ η ηd )
Together, equations (3.8), (3.30) and (3.31) determine the steady state solutions
for a 3D FEL with the effects of diffraction, energy spread and emittance for LG
modes. Note that Fq,q0 → Fq,q0 from equation (3.15) in the limit that η = 0. Due
to the dependence of the coupling on the scaled longitudinal distance τ , solutions
for the supermodes are best obtained numerically.
3.6.1 Single mode limit
Individual steady-state modes can be found from the single mode version of (3.30):
3 Z ∞
ηd 2 2 η2
δk + (2p + |l| + 1) = √ τ dτ e−iτ (δk−θ)−2τ γ Fq0 (τ ). (3.33)
α 3 0
Note again the presence of the δl,l0 term in Fq0 (τ ) that prohibits cross-coupling
between OAM modes. As before, the variational condition δ(δk)/δα = 0 applied
131
here provides solutions for the scaled complex wavenumber δk. For modes q0 =
(0, 0), (1, 0) and (0, 1), this expression is equivalent in form to that obtained in
[86], but here is expanded to include all higher-order radial and azimuthal modes.
Even in the steady-state, exact solutions to (3.33) are unwieldy analytically

because the integral over the coupling is not a simple function for arbitrary LG
modes. Numerical solutions for the gain and mode profiles can be obtained for
specific modes [42], or approximate analytic solutions can be found in extreme
limiting cases, such as for the lowest order modes in x-ray FELs [68].
3.6.1.1 Single OAM modes
The scaling characteristics of the steady state OAM modes are found directly
from (3.33) with l 6= 0. At the fundamental radial mode p = 0 the coupling
simplifies to
h √ i|l|
(4α)|l|+1 (1 − iτ η ) cos (2τ η ηd )
Fl (τ ) = (3.34)
√ |l|+1
4α2 (1 − iτ η )2 + 4α(1 − iτ η ) + sin2 (2τ η ηd )
In this form, the effect of transverse motion of the electrons in the beam for
OAM modes now becomes more apparent. Aside from the effective energy spread
given by the secular terms, the betatron motion contributes to the coupling by
way of the trigonometric functions. The cosine term in the numerator typically
dominates and produces a sinusoidal variation in the coupling as a function of τ .
It therefore cannot be generally disregarded as in [68]. While this term vanishes
when l = 0, it can play an important role for the amplification of OAM modes
since it serves to reduce the coupling. This is to be expected, since the stability
of the correlated helical microbunching structure in the e-beam that necessarily
develops during OAM amplification is particularly vulnerable to this type of
132
transverse mixing. It is noted that when the energy spread is large enough, this
term be approximately neglected, by virtue of the gaussian suppression factor
p
τ 2 ηγ2 in the integrand. The suppression is most pronounced when ηγ 4 |l|η ηd ,
p
or equivalently when σγ 2 |l|kβ /kw . However, the requirement that the
energy spread be small compared to the FEL bandwidth generally renders this
condition moot for high-gain conditions, so this transverse mixing term should
be included in general for an accurate description. This fact is evidenced clearly
in Figure 3.29 where the discrepancy between the full solutions and the solutions
obtained by neglecting these terms is conspicuous. The error is especially large
for the OAM modes which shows that the effects of emittance, particularly in the
form of transverse mixing across the e-beam during matched transport, have a
marked effect on the growth rate of the OAM modes. Accordingly it is important
to note that, while our previous examinations of OAM mode amplification in the
η → 0 limit were instructive to assess basic coupling and propagation aspects,
the sensitivity of OAM modes to transverse mixing highlights the importance of
mitigating emittance effects.
3.6.2 Fitting Formula for 3D Gain Length with Emittance
Just as in Section 3.5.6, numerical solutions to the full variational equations taken
from Eq. (3.33) can be fit to a power formula over the constituent parameter
space. The scaled gain length of the fundamental gaussian mode, given by δk i =
1/(1 + Λ0,0 ), is approximated by [42]
Λ0,0 = 0.45ηd0.57 + 3.0ηγ2 +0.55η1.6 + 51ηd0.95 ηγ3 + 5.4ηd0.7 η1.9

(3.35)
+ 0.35ηγ2.4 η2.9 + 1140ηd2.2 ηγ3.2 η2.9
133
0.55
0.50
∆ki 0.45
0.40
0.35
-2.0 -1.5 -1.0 -0.5 0.0
Figure 3.27: Effect of large emittance in the quasi-1D regime on the relative gain
of the T EM00 mode (upper blue curve) and the T EM01 mode with OAM (lower
red). Here ηd = 0.01, η = 1 and ηγ = 0.01.
0.35
0.30
∆ki 0.25
0.20
0.15
-2.0 -1.5 -1.0 -0.5 0.0
Figure 3.28: Emittance effect on growth rate for T EM01 mode for two different
√
cases with identical betatron phase advance η ηd , but different effective energy
spread due to emittance. The upper blue curve corresponds to a 3D FEL with
ηd = 0.5, η = 0.1 and ηγ = 0.01, while the lower red curve is quasi-1D with
ηd = 0.1, η = 0.5 and ηγ = 0.01.
134
0.7
0.6
∆ki 0.5
0.4
0.3
-2.0 -1.5 -1.0 -0.5 0.0
Figure 3.29: The contribution of the transverse mixing terms in the coupling is
revealed when the full, correct numeric solutions (solid lines) are compared to
the incorrect solutions (dashed lines) where the trigonometric terms are ignored
√ √
(cos (2τ η ηd ) = 1, sin (2τ η ηd ) = 0) in the coupling integral. Blue lines iden-
tify the T EM00 mode, red lines the T EM01 OAM mode. Here ηd = 0.5, η = 0.1
and ηγ = 0.01.
at the detuning that minimizes the gain length. Likewise, the scaled gain length
δk i = 1/(1 + Λ0,1 ) of the |l| = 1 OAM mode is given analogously by the numerical
fit,
Λ0,1 = 1.1ηd0.57 + 3.0ηγ2 +0.60η1.56 + 950ηd1.5 ηγ3.7 + (5.5ηd1.10 η0.5 + 11ηd0.7 η1.2 )
+ 1.14ηγ5.1 η1.6 + 20300ηd2.3 ηγ1.75 η2.1 .

(3.36)
Here, the absolute difference between the fit value of δk i,f it and the numerical
value δk i,num is < 0.05 for all values of the parameters less than unity. The rms
of δk i,f it /δk i,num − 1 is 4.7%, with the largest errors occurring when all of the
parameters are large such that the scaled gain length is small (< 0.1). It is worth
pointing out that there is nothing fundamental about the power fit with respect
to the FEL system; it is simply a useful shortcut for calculating the optimized
135
gain length. It is also not unique. The numerical fit values can vary significantly
depending on the parameter space of interest since the manifold described by the
parameters is not exactly described a single, simple power law.
3.7 Conclusions
The virtual dielectric mode expansion developed in Chapter 2 has been imple-
mented with a set of Laguerre-Gaussian to examine optical mode propagation
in high-gain FELs. The LG mode basis provides a natural description of guided
FEL radiation by virtue of its connection to the paraxial modes of free-space, and
allows the investigation of coupling to modes that contain orbital angular mo-
mentum. This technique has allowed explorations of the amplification, seeding,
growth, guiding and dispersive characteristics of OAM modes for both long wave-
length (diffraction dominated) and short wavelength (emittance, energy spread
dominated) FEL systems. Simple analytic scaling for the growth of OAM modes
has been derived in limiting cases, and power-fit functions for both the funda-
mental and the l = 1 OAM mode have been obtained in two regimes: long wave-
length devices where the longitudinal space charge plays an important role, and
short-wavelength devices where the effects of emittance are pronounced. Overall,
results suggest that due to the dominance of the fundamental modes of the typi-
cal FEL system, the excitation of a dominant OAM mode requires an OAM seed.
Seeding with an EM OAM seed has been explored, and has been shown to be
an effective avenue of OAM amplification for the proper values detuning, input
power and spot size for optimizing efficiency and output. This is predicated on
the existence of a high-quality EM OAM seed, which may not be available at
all wavelengths. Seeding with a helical e-beam distribution is an equally viable
option and was briefly examined in Section 3.5.3.2 in the context of overcoming
136
Figure 3.30: Scaled gain length from fitting formulas for the T EM00 mode (un-
filled) and the T EM01 mode with ηd = 0.1 (above) and ηd = 1 (below).
137
Figure 3.31: Scaled gain length from fitting formulas for the T EM00 mode (un-
filled) and the T EM01 mode with ηγ = 0.
138
the fundamental due to SASE. The question of how to generate such an initial
distribution remains, however, and is investigated in the next chapter.
139
CHAPTER 4
Harmonic Interaction in a Helical Undulator
4.0.1 Introduction
From the results of the last chapter, it is clear that amplification of a dominant
OAM mode is indeed possible, and in fact experimentally feasible, in a high-gain
FEL at the fundamental frequency. The electron beam in the undulator provides
a suitable medium to amplify such higher-order mode structures, provided an
adequate OAM seed is introduced to the system.
There are three main ways to seed an FEL to stimulate gain, and they are set
by the initial conditions. One is to provide an electromagnetic seed, as analyzed
earlier on. This technique works well in general for seeding OAM modes since
there are numerous ways to generate the OAM seed using external optical ele-
ments [16]. But what if no such electromagnetic seed is available? What if, at the
wavelengths or power levels of interest, it is difficult, impractical, or unfeasible to
supply an electromagnetic seed OAM mode? The seed must then come from one
or both of the other two initial conditions: the energy and density modulations in
the electron beam. Because of the particular phase structure of the OAM modes,
both of these options require precise manipulation of the e-beam structure at
the level of an optical wavelength. The e-beam distribution must be rearranged
so that the EM emission in the undulator has the helical phase structure of the
desired OAM mode, thereby initiating the FEL process. Here we examine how
140
this microscopic e-beam rearrangement is generated.
4.1 Harmonic Interaction
Let us start by taking a quick look back at the factor Fq,q0 which first appears
in Eq. 2.27 and is used in various forms throughout to describe the coupling.
Recall that this factor quantifies the ‘overlap’ of the electron beam profile with
the different modes in the expansion basis. Put another way, it gives the coupling
between the optical mode q 0 and the density modulation mode q, as mediated
by the transverse e-beam profile, f (x⊥ ). Thus, it provides a correspondence
between the density modes excited by the optical modes, and vice-versa. For the
azimuthal modes in particular, the correspondence is simple for an axisymmetric
e-beam (See 3.15 or 3.31):
F(p,l),(p0 ,l0 ) ∝ δl0 ,l . (4.1)
From this perspective, it is easy to see what type of micro-manipulation on the e-

beam structure should be performed in order to excite optical OAM modes. The
δl,l0 selection rule shows that the azimuthal mode l0 in the optical beam couples
directly to the azimuthal mode l in the e-beam density modulation. Thus, to
emit OAM light with index l, the density modulation in the e-beam must be
that of an l-twist helix. This structure is illustrated schematically in Fig. 3.15
where the depicted microbunching distribution results from the interaction with
the optical OAM mode. The emergence of the helical structures in the e-beam
and in the optical field are therefore complementary: a helically microbunched
beam emits OAM, and an OAM mode helically microbunches.
The question still remains how such a helical structure can be created in lieu of
141
an OAM seed. Clearly an EM OAM seed generates it, but that begs the question.
A clue to the solution was supplied by Sasaki and McNulty [26] who, as mentioned
in Chapter 1, found that the higher-harmonic emission in a helical undulator
carries a helical phase, and thus an associated optical OAM about the undulator
axis1 . This connection is shown by the emitted fields in Eq. (1.15) where the
azimuthal number of the optical phase l is given by the harmonic number h − 1,
with an overall sign that depends on the polarization of the undulator. Though
these expressions (as well as the ones given by Sasaki and McNulty) describe the
emission from only a single electron, the relationship between the phase and the
harmonic number suggest a new way to understand and manipulate the harmonic
coupling. As such, the concept was extended for a distribution of electrons in [88],
wherein the modal expansion formalism was broadened to include interactions at
harmonic frequencies. This work ultimately provided a straightforward solution
to the question of how a helical e-beam distribution can be generated for seeding
dominant OAM modes.
4.1.1 Density Modulation at Harmonics in the Modal Expansion
To investigate the harmonic interaction in a helical undulator we revisit the simple

case of a cold, parallel beam described by the differential equations derived in
Sections 2.3 and 2.42 . As before, the electric field is given by the modal expansion
1
It is worth pointing out that, while the phase of the harmonic emission in a helical undulator
possesses an associated OAM, this method of generating OAM light is typically inefficient. The
coupling to, and thus amplification of, the harmonics in a helical undulator are relatively
weak since, as opposed to planar undulators, the resonant interaction does not contain any
higher harmonic components to lowest order. Further, because radiation at the fundamental
frequency dominates, harmonic emission methods rely either on external spectral filters or on
other techniques to suppress the fundamental interaction [87] in order to isolate the OAM light.
2
The simplistic cold, parallel beam scenario serves to illustrate the primary dynamics. Emit-
tance and energy spread in a matched beam are included in Section (4.1.2).
142
" #
X
E⊥ (x, t) = E0 Re cq (z)uq (x⊥ )ê⊥ ei[kzq (ω)z−ωt] (4.2)
q
where now the dependence of the axial wavenumber kzq (ω) on the frequency ω is
made explicit. With the guided LG mode basis, it varies with the frequency as
4
kq2 = (ω 2 /c2 )n2I0 − w2
(2p + |l| + 1).
We are more interested in the coupling between the harmonic field and the
e-beam than on the amplification of the harmonic field, so we will concentrate
on the how the electron beam evolves. Also as before, the e-beam density is
described in a linear fluid model as,
n(x, t) = n0 f (x⊥ ) + Re δn (x)eiω(z/v0 −t) ,

(4.3)
which yields the expression for the density evolution given in Eq. (2.20):
∂2

2 2 0 cK̄E0 X 2
(1)
−iθq0 z
+ θ p f (x⊥ ) δn (x) = −ĝθ p f (x⊥ ) √ c q 0 (z)[kzq 0 + kw ] uq 0 (x⊥ )e .
∂z 2 2γeω q0
(4.4)
(h)
The detuning is now superscripted to designate the harmonic number, θq0 =
ω/v0 − (kzq0 (ω) + hkw ), with the resonant harmonic frequencies given by
ωR = 2hkw cγz2 . (4.5)
Polarization alignment (which is taken to be mode-independent) between in-

put EM field and the electron motion in the undulator is again quantified by
ĝ = ê⊥ · (êz × ê∗w ), where the unit polarization vector of the helical undulator is
√
êw = (êx ± iêy )/ 2 (4.6)
143
corresponding to either right (+) or left (-) circular polarization along z. Max-
imal coupling from polarization alignment (ĝ = 1) is obtained when the field
polarization matches the direction of motion of the electrons ê⊥ = êz × êw which
describes a left-circularly polarized wave for a right-circularly polarized (RH)
undulator3 .
The e-beam is coupled to the input field modes through the ponderomotive
fields on the right hand side of Eq. (4.4). Since the resonant interaction is
(1)
sustained near the synchronous condition only for the first harmonic (θq0 ' 0),
there is no higher-harmonic coupling of the electron beam to first order. As
shown in [88], however, coupling at higher harmonics can be excited through the
higher-order resonant interaction between the electrons and the gradients of the
EM fields. This contribution is calculated by Taylor expansion of the field modes
about the average centroid trajectory of an electron x̄⊥ ,
∞
X 1h −ikw z

¯
in
uq (x⊥ ) = Re x̃⊥w e · ∇ uq (x̄⊥ ), (4.7)
n=0
n!
where the transverse coordinate of an electron in the bunch is x⊥ = x̄⊥ +

√
Re x̃⊥w e−ikw z , the electron wiggling amplitude is x̃⊥w = (K̄ 2/kw γβz )êz × êw ,

¯ is the standard gradient operator which acts on x̄⊥ .

and ∇
The interaction of the electron with the different regions of the field that vary
in intensity introduces additional oscillatory terms in the motion, some of which
have the right frequency to couple to resonant harmonics of the FEL interac-
tion. Accordingly, Equation (4.7) can be simplified for practical cases near these
resonances. In the regime where |x̃⊥w | is much smaller than the characteristic
transverse e-beam size σx , we make the approximation x̄⊥ ' x⊥ in Eq. (4.7)
and obtain an expression for the Taylor expanded harmonic field modes that are
3
See Appendix B for explanation of this apparent oddity.
144
resonant with the integer harmonics h,
h−1 X∞
(−1)m

∓iφ ∂ i ∂
u(h)
q (x⊥ ) = e ∓
∂r r ∂φ m=0
m!(m + h − 1)!
(4.8)
2m+h−1
±iK̄
× ∇2m
⊥ uq (x⊥ ),
2kw γβz
where ∇2⊥ is the transverse Laplacian operator. The higher-order terms (m > 0)
in the Taylor series are typically negligible and do not affect the results, but are
temporarily included for completeness. With Eq. (4.8) harmonic interactions
that occur as a result of transverse variation in the fields are described in Eq.
(1) (h)
−iθq0 z (h) −iθq0 z
(4.4) by replacing u⊥q0 (x⊥ )e with u⊥q0 (x⊥ )e .
Like before, it is convenient to write the density perturbation in (4.4) as a

sum over the expansion eigenmodes,
X
δn (x) = aq (z)uq (x⊥ ), (4.9)
q
such that the orthogonality of the basis can be used to compactly write the
density evolution equation (4.4) in terms of spatial modulation amplitudes. The
result is an expression similar to (2.25), but now incorporates frequencies near
the resonant harmonics,
d2 2
X 1 X (h) (h)
−iθq0 z
a q (z) + θ p F q,j a j (z) = − Q 0 c q 0 (z)e , (4.10)
dz 2 j
ξq q0 q,q
√
where ξq = K̄ 2n0 eω/4cγ0 E0 kzq . The coupling of the e-beam with arbitrary
transverse density distribution f (x⊥ ) to the harmonic fields is given by the har-
145
monic coupling coefficient
2
+ kw )2

(h) (kzq0 K̄ (h)
Qq,q0 = ĝ θp2 Fq,q0 (4.11)
4kzq γ
and the overlap of the e-beam and the input fields is
(h)
f (x⊥ )uq0 (x⊥ )u∗q (x⊥ )d2 x⊥
R
(h)
Fq,q0 = R , (4.12)
|uq (x⊥ )|2 d2 x⊥
(1)
where Fq,q0 = Fq,q0 with m = 0 in Eq. (4.8). Note the similarity with Eq. (2.27),
but where here the harmonic field expansion is within the integral.
Much like before, the coupling selection rules between the azimuthal modes
in the field and those in the e-beam are elucidated by a mode basis of the form
(1.22, 3.3),
uq (x⊥ ) = ûp,l (r)eilφ . (4.13)
With an axisymmetric e-beam profile f (x⊥ ) = f (r), the integral over φ in Eq.
(4.12) is straightforward with respect to the harmonic number and gives an im-
portant result;
(h)
F(p,l),(p0 ,l0 ) ∝ δl0 ,l±(h−1) . (4.14)
Clearly the coupling between azimuthal density modulation modes in the e-beam,
l, and the azimuthal modes in the EM field, l0 , now depend on the harmonic
number h and the direction of undulator polarization (±). This is an extension
of (4.1) and demonstrates that at harmonics, the azimuthal modes differ by h−1.
At the second harmonic h = 2, for example, the lowest order EM field mode l0 = 0
will generate a helically bunched e-beam with azimuthal mode number l = −1
in a right-handed undulator and l = 1 in a left-handed undulator. At the third
harmonic, the same input mode will excite the double-helix l = ±2 density mode,
146
helical
undulator
unmodulated helically
e-beam bunched e-beam
Gaussian input field

at harmonic of undulator
Figure 4.1: Depiction of the buncher device in which the axi-symmetric input
field imprints a helical density and velocity modulation on the e-beam through
the harmonic interaction in a helical undulator.
and so on. The harmonic interaction ‘selection rules’ between modes therefore
provide a way to generate a helically microbunched e-beam, without an EM
OAM seed (Figure 4.1). This concept forms the basis of the present discussion of
OAM mode generation in an FEL at the first harmonic without mode-converting
the EM field. That is: a simple axisymmetric optical field generates a helically
microbunched electron beam at harmonics in a helical undulator.
What we therefore obtain is a scenario that utilizes harmonic coupling in a

helical undulator to generate a helically micro-bunched e-beam for seeding OAM
emission in a downstream high-gain FEL. The proposed arrangement is that of
an optical klystron [89], composed of a modulator section followed by a radia-
tor. The technique may be broadly dubbed as high-gain high-mode generation
(HGHMG), wherein the source e-beam distribution is prearranged so that it ra-
diates transverse higher-order optical modes the FEL4 . Since no electromagnetic
OAM seed is required, this technique generates OAM modes without the use of
external optical mode conversion elements (such as cylindrical lenses and spiral
phaseplates) and therefore can work for OAM production across the wide range
4
Note that the same technique can be applied, for example, to generating lobed T EM01 or
T EM10 Hermite-Gaussian modes in a radiator by seeding a planar modulator with a Gaussian
EM beam at the second harmonic.
147
of wavelengths and power levels accessible to FELs. One only needs to seed a
helical modulator with a Gaussian beam at a frequency harmonic of the inter-
action and allow the resulting helical e-beam distribution to radiate coherently
in a downstream undulator. This concept is examined in more detail in Section
4.1.2.2 for short wavelengths, but can obviously be applied to longer wavelengths
as well.
To complete the discussion of the modal formalism we note that modeling the
harmonic modulator system also requires a set of equations to describe the seed
field that performs the modulation. The harmonic density evolution equation
(4.10), together with the EM field evolution equation in (2.25),
d X
cq (z) = −i Kq,q0 cq0 (z)ei(kzq0 −kzq )z , (4.15)
dz q0
model an electron beam modulator where an input field, given by the sum of
modes with amplitudes cq (z), drive the harmonic helical microbunching process.
Here the density modal amplitude aq (z) that describes the feedback of the e-
beam density amplitudes has been ignored because we are not interested in the
amplification of the seed field. Equations (4.10) and (4.15) properly model the
harmonic microbunching process (within the aforementioned constraints) in the
modal expansion formalism.
We can follow the growth of the helical bunching factor, defined in (3.12) for
the LG mode basis, for a low-energy system as is presented in [88]. Bunching
curves at the second harmonic are shown in Figure 4.2 for three different values
of the total input power of an l0 = 0 gaussian free-space mode at 2πc/ω ' λ=10.6
µm. In each case, there is good agreement between the solutions to Eq. (4.10)
with an LG basis and 3D numerical particle tracking simulations performed with
148
TREDI [90] for dominant bunching at the l = −1 density mode. Significant
bunching is observed near the exit of the Lm =35 cm RH undulator with only 15
MW of input power. Bunching into modes l 6= −1 is negligible. Note that the
expected departure of our linear theory from particle simulations is evident for
very large bunching (|bl | > 35%), as in the case of 125 MW of input power.
The effects of space-charge are ignored in Figure 4.2 (θp → 0), but can play a
crucial role in the evolution of the bunching at low energies. Figure 4.4 includes
space-charge effects in the case of injection of l0 = −1, l0 = 0, and l0 = 1 free-
space LG modes with a Gaussian e-beam in a RH undulator. The corresponding
axial velocity modulation modes (l = −2, l = −1, and l = 0 modes excited
in the e-beam, respectively) are manifest in the growth of the bunching factors
downstream of the 20 cm undulator, which reach a maximum and then decay
due to space-charge oscillations. Maximum peak bunching is achieved with the
injection of the “natural” l0 = 1 field, which bunches the beam into longitudinally
separated micro-bunches (b0 ). It is interesting to note that this optical mode
excites larger bunching than the other modes at the second harmonic because it
has larger resonant coupling to the e-beam. Since the exponential gain process in
an FEL is given by the same coupling mechanism as bunching, this optical input
mode therefore also achieves the highest relative gain as an output mode in an
FEL at the second harmonic, in agreement with Sasaki [26]. This result is true in
general for this interaction: the dominant harmonic radiation mode amplified by
coupling to transverse EM field gradients in a helical undulator with right (left)
circular magnetic field polarization is an l0 = h − 1 (l0 = 1 − h) azimuthal mode
with left (right) circular polarization. It is interesting to notice that the spin and
orbital components of the angular momentum in the emitted photons in the FEL
therefore add, with each photon possessing |h~| units of total angular momentum
in the classical limit. These |l0 | > 0 EM modes described here are characterized
149
60 -1.0
-0.5 z/λ
0.0
0.5
50 1.0 125MW
1.0
40
0.0 y/r0
|b -1 | % 30 l=-1 -1.0 30MW

1.0
0.0
-1.0 x/r0
20
10 15MW
0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
z (m)
Figure 4.2: Injection of a gaussian input mode onto a gaussian e-beam for a RH
undulator excites l = −1 density bunching at the second harmonic (λ=10.6 µm).
The associated helical bunching curves for a w0 =350 µm spot size gaussian mode
with waist at z = 0.15 m on a σx =250 µm e-beam are in good agreement with
TREDI simulations (points) with γ=26, K̄=0.65, λw =2 cm.
by a null intensity on-axis, as expected [91, 92, 93].
It is noted that the non-linear contribution of the input field to the interaction
has been justifiably neglected in our analytic description under the small signal
approximation. The effect of this term on the e-beam dynamics is negligible
even for the maximum input laser power specified here since the normalized field
amplitude of the input field Kf = e|E⊥ |/ωme c, is small (Kf /K)2 < 10−6 . Large
fields (Kf ∼ K) alter the dynamics (and the resonance condition) and should
be included to properly describe several aspects of the coherent process. The
linear contribution of Kf , however, is important as it establishes the width of
the ponderomotive phase bucket for linear bunching to occur (since no resonant
interaction is possible for Kf = 0). The e-beam is well inside the bucket for the
parameters given here, with the energy well within the range 25 < γ < 26.7 even
for the minimum 15 MW case.
150
zΛ
4 2 0
-2
-4
-0.5
0.0
x
0.5 0.5
0.0
y
-0.5
-1.0
Figure 4.3: Typical section of microbunched beam from numerical TREDI simu-
lations.
30 l‘=1 (b 0 )
25
20 l‘=0 (b -1 )
|b| % 15
10
l‘=-1 (b-2 )
5
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
z (m)
Figure 4.4: Longitudinal space-charge effects included in bunching evolution for

the injection of l0 = −1, l0 = 0, and l0 = 1 modes onto a gaussian e-beam for a
RH undulator at h=2. The optical spot size is 500 µm for each mode, with the
waist positioned halfway through the undulator. The vertical line denotes the
end of the 20 cm undulator, after which the beam drifts and displays the onset
of plasma oscillations in the bunching factor. Input power is 15 MW and the
e-beam current is 300 A with γ=26, K̄=0.65, λw =2 cm.
151
4.1.2 Emittance and Energy Spread Effects
The modal description in the last section clearly displays the correspondence
between the azimuthal modes and the harmonic number via the coupling in (4.14)
and can be useful for modeling modulator systems in which the energy spread,
emittance and focusing can be ignored. As usual, however, emittance and energy
spread in the beam can affect the microbunching process in the modulator.
We have already done most of the analytic work to examine these effects,
in fact, when we derived the expression for the modulated component of the
(h)
normalized distribution function F = F0 (x⊥ , x0⊥ , η) + 12 F1 (x⊥ , x0⊥ , z, η)eiψ + c.c.
in Eq. (2.86). First we note that the change in energy of an electron in the
beam as it interacts with the fields of the modulator and laser near the harmonic
resonance h is given by
dη eK h (h)
i
=√ Re iĝ Ẽ (h) (x)eiψ0 (4.16)
dz 2γ 2 mc2
(h)
where the harmonic ponderomotive phase is ψ0 = kz +hkw z −ckt. It is straight-
forward to now account for the interaction with the Taylor expanded harmonic
(h)
field Ẽs (x⊥0 , z 0 ):
„ «
Z z (h) k
ĝeK̄ ∂ i k 2 2 02
2 η−θ0 − 2 (kβ x⊥ +x⊥ ) (z 0 −z)
F1 = −i √ dz 0 Ẽs(h) (x⊥0 , z 0 ) F0 (x⊥ , x0⊥ , η)e γz
,
2γ 2 mc2 0 ∂η
(4.17)
h i
where as before, the nominal field is given by E⊥ = Re Ẽs ei(kz−ωt) . The har-
(h)
monic field Ẽs (x⊥0 , z 0 ) is thus expanded in precisely the same way as (4.8). Here,
however, the fields are assumed to not be affected by the electron beam so they
can be specified independently. Here we assume that the drive laser has a pulse
length cστ much longer than the slippage through the modulator, cστ > Nw λ.
152
It is therefore taken to be monochromatic and constant longitudinally over the
e-beam throughout the interaction.
The measure of the density modulation is the helical bunching factor (2.37)
which, as a function of z along the undulator, is given simply as
Z
0
bl (k , z) = h F (x⊥ , x0⊥ , η, s; z)e−ik s−ilφ dx⊥ dx0⊥ dηi.
0
(4.18)
Brackets denote averaging over the beam coordinate s = z − βz ct. In electron

beams much longer than the wavelength, the bunching factor is maximized at
the frequency k 0 = k/βz .
Let us consider the specific case of seeding the modulator at the second har-
monic with a free-space Gaussian (p, l = 0) paraxial wave from (1.27). Let the
unmodulated e-beam distribution also be Gaussian and matched to the focus-
ing channel through the modulator, F0 = (4π 2 σx2 σx20 2πσγ2 )−1 exp(−x2⊥ /2σx2 −
p
x0 2⊥ /2σx20 −η 2 /2σγ2 ). The bunching factor along the interaction length is calculated
from the integral

K̄ 2 p2πP/P Z z dz 0 (z 0 − z) cos [k (z 0 − z)]
A β
|b∓1 (z)| = ĝ

3 4
2γ σx w(z 0 )ν(z 0 )ζ 2 sin2 [k (z 0 − z)]
0 β
(2)
−iz 0 θ0 −
k 2 σγ2
(z 0 −z)2 −i tan−1 [(z 0 −zw )/zR ]
(4.19)
4
e 2γz
×h

i1/2 3/2
ν(z 0 )2 ζ 2 sin2 [kβ (z 0 −z)]
1+ ζ2
1+ ν(z 0 )2
where ζ 2 = (1 + i (z 0 − z) kkβ2 σx2 )/2σx2 sin2 [kβ (z 0 − z)], P is the input seed power,
PA = IA mc2 /e = 8.7 GW where IA = 4π0 mc3 /e = 17 kA is the Alfvén current,
and ν(z)2 = 1/w(z)2 − ik/2R(z). The Rayleigh length is zR = kw02 /2, w0 is
the waist size, Rc (z) = (zR2 + (z − zw )2 )/(z − zw ) is the radius of wavefront
p
curvature, w(z) = w0 1 + (z − zw )2 /zR2 is the spot size along z, and zw is the
153
position of the waist. Equation (4.19) is completely general and is straightforward
to solve numerically. Section 5.4 explores the results of this expression, and
compares them with numerical simulations for specific parameters of the HELiX
microbunching experiment described in the next chapter.
4.1.2.1 Effect of betatron motion and energy spread on helical distri-

bution
As in the case of the FEL, the evolution and stability of the helical micro-
distribution generated at the second harmonic is subject to the transverse motion
due to focusing and the energy spread. In comparing the analytic form of Eq.
(4.19) with the coupling to OAM modes in (3.34) we find that the helical e-beam
structure is again affected by a cosine term that suppresses the l = ±1 helical
structure roughly every half betatron oscillation period, λβ /2. The exact period-
icity and amplitude of this fluctuation depends on the detuning, but the presence
of the overall effect is to be expected as electrons in the correlated helical struc-
ture are transported across the profile during focusing. The transverse positions
are mirrored in x and y across the axis, and the helical structure replicates itself
every half-betatron period, but shifted by ∼ π in phase. See Figure 4.5. (The
phase shift is ∼ lπ for an l-mode helix, and is therefore unique to this type of cor-
related microbunching structure.) Figure 4.6 depicts the effect on the bunching
and on the phase along the modulator. Here, the modeled seed laser is focused
near the undulator entrance to initiate the helical interaction, and then diffracts
away, contributing little to sustaining the interaction. The bunching factor has
initial growth that is periodically punctuated by the betatron focusing until fi-
nally the energy spread washes out the structure completely. Though this effect
adds an additional level of dynamic complexity to the transport, the periodic-
154
Figure 4.5: Evolution of helical density distribution during matched betatron
focusing.
ity provides a mechanism by which the system can be gamed to optimize the
bunching at the modulator exit. Either the length of the modulator is designed
so that it is much shorter than λβ /4 to avoid the first dip, or the system is tuned
in accordance with one of the periodic λβ /2 bunching maxima so that the helical
bunching is at a peak by the end.
155
30 Π
20 Π
2
10
Èb-1 È% 0 0 argHb-1 L
-10 Π
-2
-20
-Π
0.0 0.5 1.0 1.5 2.0
zΛ Β
Figure 4.6: The bunching factor of the helical beam fluctuates during
beta-matched transport, changing in phase by ∼ π every half-betatron period.
4.1.2.2 Helical bunching at short wavelengths/high energies
At x-ray wavelengths the seed field diffracts weakly over the length of the mod-
ulator Lm , so the Rayleigh length of the Gaussian input seed can be taken to
zR → ∞ and the seed stays at a fixed size w0 . Further, for high energy e-beams
the quarter-betatron period is frequently much larger than Lm , so the emittance
effects on the microbuching can be ignored as kβ → 0. In these limits the bunch-
ing factor is dramatically simplified to,

p 2 2

ĝ K̄ 2 2πP/P Z z 0 θ (2) − k σγ (z 0 −z)2

A 0 0 −iz 0 4
|b∓1 (z)| = dz (z − z) e 2γz
. (4.20)

3 2h 2
i3/2
w0 0
γ σx 1 + 2σ2

x
The FWHM detuning bandwidth in the limit of vanishing energy spread is

(2) (2)
θ0 (F W HM ) ' 10/Lm , peaked about θ0 = 0. At resonance and in the cold
beam limit, the bunching factor grows quadratically with distance.
For many such high-energy devices, the density bunching amplitude obtained
in the modulator is simply too small unless impractically long modulator sec-
156
-2 Π -Π 0 Π 2Π
-2 Π -Π 0 Π 2Π
Ψ0 ¡l Φ
Figure 4.7: Initially unmodulated e-beam (blue) becomes helically modulated in

energy during the harmonic interaction (red), and is helically density bunched
(tan) after transit through the dispersive section.
tions are used. The induced energy modulation, however, can be significant even
over a short section. Accordingly, a longitudinally dispersive section placed im-
mediately downstream of the modulator is commonly used to convert the energy
modulation into a density modulation. This works by providing the higher energy
electrons in the bunch a different path length than the lower energy electrons.
For the correct magnitude of dispersion, a periodic energy modulation will there-
fore be transformed into a periodic density modulation as the electrons clump
together longitudinally (see Figure 4.7). Mathematically, the dispersive section
is a transformation on the longitudinal coordinate that sends the electrons from
the relative position s to s + R56 η, where R56 = ∂s/∂η is the first order trans-
port matrix element. For an modulator of length Lm the equivalent transform is
157
Lm + R56 γz2 , and the bunching factor at the end of the dispersive section is

ĝ K̄ 2 p2πP/P
A
|b∓1 | =

3 2h 2
i3/2
w
γ σx 1 + 2σ02
x
(4.21)
2 2

Z Lm (2) k σ 2
−iz 0 θ0 − 4γ (z0 −Lm −R56 γz2 )
dz 0 z 0 − Lm − R56 γz2 e

2γz
,
0
(2)
which simplifies at the second harmonic resonance θ0 = 0 to
p 4 " 2 2 2 2 )2 σ 2
k2 (Lm +R56 γz
#
|ĝ| K̄ 2 2πP/PA γz k R56 σγ
− γ
|b∓1 | = i3/2 2σ2
e− 2 − e 4
2γz . (4.22)
k
h 2
3 2 w γ
γ σx 1 + 2σ2 0
x
If the modulator is short Lm R56 γz2 , essentially all of the density bunching
occurs in the dispersive section and to first order in Lm we retrieve the expression
from [94]; p
|ĝ| K̄ 2 R56 Lm 2πP/PA − k2 R56
2 σ2
γ
|b∓1 | = h i 3/2
e 2 . (4.23)
2 2 w02
γσx (1 + K̄ ) 1 + 2σ2
x
The bunching is maximized in this limit for R56 ' (kσγ )−1 . Further, it is easy to
show that the bunching is also maximized when σx /w0 = 1/2. At these optimal
values, the maximum bunching in this regime is expressed as
p
|ĝ| K̄ 2 Lm P/PA
|b∓1 | = 0.59 (maximum). (4.24)
γσγ zR (1 + K̄ 2 )
Finally, in the limit of vanishing energy spread σγ → 0 the bunching factor in

(4.22) is p
|ĝ| K̄ 2 2πP/PA
Lm Lm + 2R56 γz2 .

|b∓1 | = h i 3/2
(4.25)
w2
2γ 3 σx2 1 + 2σ02
x
We see by comparison of Eq. (4.23) with the expression from [94] that in this
158
σγ → 0 limit there is an extra term, the Lm in the parentheses, which identifies the
density bunching contribution from the modulator section. This is usually small
compared to 2R56 γz2 and is typically ignored in linearized analyses that assume
the phase position of an electron in the bucket does not change in the modulator.
While this approximation is well-justified for high-energies, the contribution from
the modulator should be included in general for improved accuracy.
4.2 High-Gain High Mode Generation (HGHMG) Scheme
Imprinted with the helical density distribution at the chicane exit, the e-beam
then enters the undulator (with any polarization) with period λw,r tuned to emit
light at the fundamental wavelength of the microbunching period: λb = 2π/kb =
λw,r 2
2γ 2
(1 + Kr ), where Kr is the rms undulator parameter of the radiator. The
OAM emission is at the same wavelength as that of the seed laser in the modulator
section, so the entire setup in Fig. 4.8 acts as a ‘mode converter’ which transforms
the initially transversely Gaussian T EM00 laser pulse into an T EM01 OAM mode
by virtue of the natural manipulation of the e-beam microbunching distribution.
As noted before, since the other modes can by amplified spontaneously due to shot
noise in the beam, and because mode competition usually favors the fundamental,
the helical bunching factor must be greater than the effective shot noise bunching
q√
factor in the radiator, bSN ' 2πecρ/9Iλb to allow for the OAM mode to
dominate from the outset.
Figure 4.9 shows the power as a function of the azimuthal mode number
in the radiation field calculated with time-independent numerical simulations
from Genesis [79] for an LCLS-type e-beam [3] at λb = 1.5 Å. Virtually all of
the power is emitted into the dominant l = 1 mode, with less than 1% in the
l = 0, −1 modes at any given z position. The characteristic hollow profile and
159
Chicane
Helical OAM light
Seed laser Undulator
Undulator
+
Harmonic
e-beam Optical dump
interaction
Figure 4.8: Arrangement for generating OAM light in a short wave-

length/high-energy FEL. The e-beam is helically modulated via the 2nd harmonic
interaction in the initial modulator stage.
helical phase of the l = 1 mode are shown Fig. 4.10. The dominance of the l = 1
mode over the fundamental is also indicated in the bunching factor of the l = 0
beam mode, which stays below 2% throughout. Relevant parameters used are
γ = 26700, σx = 23 µm, ση = 0.01%, giving a b1 = 1.9% initial bunching factor5
using a P = 1 GW, w0 = 31 µm Gaussian input laser in a K = 3, Lw = 7 m
modulator followed by a small R56 = 0.24 µm chicane. At this level the bunching
factor is much larger than the effective shot-noise bunching bSN ' 1.2 × 10−4 ,
but is small enough that the coherent OAM emission develops into high-gain
amplification with an average gain length of 4.5 m before saturating after 50 m.
The e-beam emits Pout =10 GW of OAM light through a linearly polarized LCLS-
type undulator with Kr = 2.47, λw,r = 3 cm at a beam current of I = 3 kA and
normalized emittance nx = 0.5 µm. Note that with Pout /P = 10, the HGHMG
scheme acts as a combination mode-converter and amplifier of the Gaussian seed,
delivering ten times as much OAM output power as it took in the Gaussain seed
to perform the modulation. The output OAM mode is describable in mechanical
5
The bunching is b1 = 1.47% if calculated straight from Eq. 4.23 as in [94] where the small
density modulation in the modulator is ignored. In simulating the performance of the FEL this
difference is negligible.
160
terms as having an available integrated torque about the propagation axis3.6 of
τz = lPout /kc = 0.8 × 10−9 N-m.
The HGHMG scheme in this example is seeded by an attenuated x-ray seed

similar in character to the output of the LCLS [3] itself. The timing between the
seed x-ray FEL laser pulse and the e-beam in the helical modulator section sug-
gest a self-seeded running scenario to modulate the e-beam; using either a single
e-beam [95], or better a two-beam self seeded scheme [96] in which two e-beams
are separated with a precise delay by correct excitation of the photocathode elec-
tron source. The two-beam approach has the advantage of drastically reducing
the footprint of the chicane required to temporally synchronize the seed pulse
(emitted by the first beam in the FEL) with the second e-beam in the modula-
tor. This scheme can also can also be employed in tandem with a kicker to impart
a large betatron amplitude to the second e-beam so that it does not lase in the
seed undulator. In this way, the energy spread and microstructure of the e-beam
is preserved for the downstream helical bunching and OAM lasing stations. In ei-
ther seeing scenario, the seed FEL laser pulse passes through an optical transport
line that can also be equipped with either a reflective or transmissive monochro-
mator [97] to lengthen the seed pulse and increase the temporal overlap with the
e-beam in the modulator. When the length of the coherent seed is greater than
that of the e-beam bunch, the helical modulation is in phase across the entire
bunch and full longitudinal coherence in the emitted OAM mode is obtained.
Monochromatization comes at the expense of the seed power, so optimization of
this type of setup depends on the available FEL seed power and bandwidth. This
is a topic left for future studies. Note also that for pump-probe-type experiments
that rely on different optical modes, the optical dump that blocks the seed light
can be removed such that the initial Gaussian laser also exits through the second
radiator. Since it exits prior to the OAM light pulse at a fixed delay, the HGHMG
161
10
10
8
10
Power (W) 40
RMS radiation size (Μm)

6
10
30
4 l=1 20
10
l = -1
l=0 10
2 Total Power
10
0
0 10 20 30 40 50 60
z (m)
Figure 4.9: Optical power in different l−modes emitted from helical e-beam from
time-independent numerical Genesis simulations.
setup is that of a two-pulse, two-mode system of a Gaussian beam followed by a

OAM beam.
4.2.1 Unwanted mode coupling through misalignment
The dominance of the OAM mode in the HGHMG FEL scheme is predicated on
the fact that the microbunching structure is dominated by the required helical
e-beam mode. While this is readily generated in an ideal system as described
above, this type of highly-correlated structure relies on significant symmetries in
the e-beam, laser, and transport, which may be otherwise compromised by ex-
perimental asymmetries. Such non-ideal conditions lead to undesirable coupling
to other microbunching modes which can then be amplified in the downstream
FEL, so it is useful to examine the available tolerances.
There are many non-ideal effects that result in coupling to other modes, many
of which are best analyzed numerically. Here, however, to get a flavor for the
contribution from a non-ideal scenario, we analytically examine a simple parallel
162
Amplitude (A.U.) Phase (radians)
−100 1
3
−80 0.9
−60 0.8 2
−40 0.7
1
y (microns) −20 0.6
0 0.5 0
20 0.4
−1
40 0.3
60 0.2 −2
80 0.1
−3
100
−100 −80 −60 −40 −20 0 20 40 60 80 100 −100 −80 −60 −40 −20 0 20 40 60 80 100
x (microns) x (microns)
Figure 4.10: Transverse profile of the light reveals a dominant l = 1 OAM mode
at saturation.
misalignment of the laser and the e-beam. Consider an e-beam with its centroid
laterally displaced from the laser axis by an amount y0 . In order to see how this
excites other azimuthal modes, the transverse beam distribution is written as an
expansion in polar coordinates;
− x2 (y−y0 ) 2 2r 2 +y02
X (−i)l0 (−1)p0 eil0 φ y0 2p0 +|l0 | r |l0 | 0 r2
2 − 2σ 2 −
e 2σx x =e 4σx 2
0 + |l 0 |)!
L|lp | .
p0 ,l0
(p 2σ x σ x σx2
(4.26)
Clearly if y0 = 0 only the p0 , l0 = 0 terms are nonzero and the distribution is a
r2
− 2
simple on-axis Gaussian, e 2σx , as before.
For simplicity, we ignore emittance and assume that the Gaussian laser spot
size is unchanged along the interaction (zR → ∞). At the second harmonic,
the relative magnitude of the bunching factor in (4.18) into the mode l at a
163
Σx /w0 = 1 Σx /w0 = 0.5
1.0 1.0
1=-1 1=-1
0.8 0.8
bl (y0 )/b-1 (0)
bl (y0 )/b-1 (0)

0.6 0.6 1=-2,0
0.4 1=-2,0 0.4
0.2 1=-3,1 0.2 1=-3,1
0.0 0.0
0 1 2 3 4 0 1 2 3 4
y0 /Σx y0 /Σx
Figure 4.11: Transverse profile of the light reveals a dominant l = 1 OAM mode
at saturation.
displacement y0 is given by
0 0 2p0 +|l0 | 0
bl (y0 ) − y022 X (−i)l (−1)p

= e 4σx y0 |l | + 1
δl0 ,l±1 0 0 !
b∓1 (0)
0
p ,l 0
p !|l |!(1/2)! 2σ x 2
 
−|l0 |/2 0
σx2

1 0 |l | + 3 0 1 
+ 2 F1
 −p , , |l | + 1, σ2 .
w02 2 2 x
+ 21
w2 0
(4.27)
At y0 = 0 the bunching is purely in the l = ∓1 mode, so the magnitude of the
displacement effect is weighed against this value.
Figure 4.11 depicts the effect of the misalignment for a RH undulator. The
displaced e-beam shows equal coupling to the modes symmetric about the primary
mode l = −1. For example, the bunching factor into the l = −2 mode is excited
with equal magnitude as the fundamental l = 0 mode for equal displacements.
The modes closest to the primary mode have the largest effective excitation,
which diminishes for all the modes as y0 grows due to the Gaussian suppression
factor in (4.27). For a given displacement, the extent to which the adjacent
modes are excited depends on the size of the e-beam compared with the laser
spot. As the e-beam grows large compared to w0 , coupling into adjacent modes
is more strongly suppressed, suggesting it is advantageous for tolerances to make
164
the ratio σx /w0 large.
As a final remark, it is instructive to provide a baseline condition on the

displacement in order that bunching into the fundamental mode is minimized.
Although the l = −2 mode is equally excited in a RH undulator, the l = 0
provides a seed for the fundamental optical mode in the FEL which can dominate.
We thus posit that the l = 0 bunching factor be less than 1% that of the l = −1,
which approximately gives
√
0.01 2πσx
y0 ≤ q < 0.025σx . (4.28)
2σx2
1 + w2
0
Alignment in the modulator must thersfore be within a few percent of the e-beam
size insure OAM mode dominance in the HGHMG scheme.
4.3 Conclusions
By virtue of the harmonic interaction in an helical modulator, it has been shown

that a correlated helical modulation can be imprinted on an electron beam for
seeding high-gain OAM emission in a downstream FEL at the dominant funda-
mental frequency. The natural complementarity between the azimuthal modes in
the optical field and in the e-beam provides an avenue by which the highly corre-
lated heical structure can be imparted on the e-beam on the scale of the desired
optical wavelength. This HGHMG technique requires only a simple Gaussian in-
put seed for l = ±1 OAM mode generation, so the combination of the modulator
and radiator (plus, at high energies, dispersive section) acts like an optical mode
convertor using the electron beam as the conversion medium.
165
CHAPTER 5
Helical Microbunching Experiment at the

Neptune Laboratory
5.1 Introduction
The previous chapters have examined the conditions required to generate dom-
inant optical OAM modes in a high-gain FEL. Analysis and simulations clearly
show that a dominant OAM mode can be amplified from the proper seed which,
given the selective azimuthal coupling between optical and e-beam modes at har-
monics in a helical undulator, can be supplied by helically microbunched electron
beam. This chapter describes efforts to investigate the helical microbunching con-
cept in a proof-of-principle experiment dubbed the helical interaction experiment,
or HELiX, performed at the Neptune Laboratory at UCLA.
5.2 Design of the Experiment
The primary goal of the HELiX project is to generate and measure a helically
microbunched e-beam in the laboratory. Successful measurement of such a struc-
ture would serve not only to support theoretical and numerical predictions that
illuminate the highly-correlated 3D harmonic effects, but more importantly would
demonstrate that the helical microbunching technique may be applied towards
166
thin
detector
iris metal foil
helically
laser helical bunched e-beam
undulator
unmodulated CTR
e-beam
mirror
Figure 5.1: HELiX concept.
the in situ OAM production in a high-gain FEL. The basic experimental arrange-
ment is shown in Fig. 5.1. The primary components are the Gaussian seed laser,
the helical undulator, the electron beam and finally a thin metal foil to act as
the radiator (of coherent transition radiation in this case) to provide information
on the microbunching structure. The following sections describe each of these
components in detail, all of which are housed at the Neptune lab.
5.3 Neptune Laboratory
The Neptune lab is, and has been, home to numerous experiments. It is a state-
of-the art facility dedicated to investigating the interaction and dynamics between
relativistic electron beams, high-power lasers and plasmas. It is jointly operated
by the Particle Beam Physics laboratory (PBPL) headed by Professor James
Rosenzweig of the UCLA Department of Physics and Astronomy, and by the
Laser-Plasma Group directed by Professor Chan Joshi of the UCLA Department
of Engineering. The lab consists of two main systems; the high-power CO2 laser
system(s) and the electron beam linear accelerator (LINAC) driven by an RF
photoinjector. References [98, 99, 100], and many of the references therein contain
167
in-depth reports of the Neptune capabilities and infrastructure used for previous
experiments.
5.3.1 CO2 Laser
Depending on the operational setup, the 10.6 µm wavelength CO2 laser system is
capable of delivering upwards of 15 TW of peak laser power (45 Joules, 3 ps) at
a repetition rate of one pulse per five minutes [101]. However, for the purposes of
the less-demanding HELiX and past related microbunching experiments [102],
the system can be configured to run reliably at 20-80 MW of peak power at a
pulse repetition of 0.5Hz, with FWHM pulse lengths of 100 ps.
The primary components of the laser system are shown in Figure 5.2. The
10.6 µm seed laser for the HELiX is a gas CO2 laser system that starts with a
500 ns, 30 mJ linearly polarized pulse and ultimately ends with a 100 ps, 3-6 mJ
circularly polarized pulse at the undulator to drive the microbunching process.
The laser chain begins with a hybrid TEA CO2 oscillator which, with a low
pressure (30 Torr) amplifier in the cavity, generates a 10.6 µm single longitudinal
mode output (Fig. 5.2(d)). This produces a 30 mJ, 500 ns output pulse that
is then temporally sliced using a 100 ps long 1.064 µm beam from the Nd:YAG
laser system (a-c). The 100 ps CO2 pulse then propagates into a multipass 8 atm
regenerative amplifier (e), resulting in an pulse train of roughly a dozen 10mJ,
100 ps pulses, separated by 12 ns. This amplifier has a gain volume of 1 × 1 × 60
cm3 and is filled with a gas mix of equal parts CO2 and N2 to 14 parts H3 . A
single pulse from the train is switched out with an external Pockels cell positioned
between crossed polarizers. Finally, this pulse is delivered through single-pass 8
atm booster amplifier (f) which doubles the pulse energy to obtain 20-40 MW.
This is then delivered to the experiment. With two thin film polarizers and a
168
(b)
GE 100 1.064 µm Nd:Glass
To Photocathode
Oscillator Regen & Booster
(a)
Optical Delay
CS2
Kerr cell (f)
Pockels cells (c)
169
8 atm
Booster
8 atm
TEA Regen
10.6 µm (e) Amp
Master Osc
Figure 5.2: Neptune laser system.

(d) Low Pressure
CO2 Amp
Nd:YAG Regenerative Amplifier
& Booster
quarter-waveplate, the booster stage can be modified to become a double-pass
device capable of generating up to 80 MW.
Temporal slicing of the initial 500 ns CO2 oscillator output pulse is achieved
via optical gating in a single carbon disulfide (CS2 ) Kerr cell with the 1.064
µm Nd:YAG laser pulse. The 1.064 µm beam is obtained by beam splitting the
output of the actively mode-locked CW GE100 oscillator (Fig. 5.2(a)) with a
50% mirror. One half is delivered to the Nd:YAG amplifier of the slicing system
and the other half sent to the Nd:Glass regenerative amplifier system (b) which is
frequency quadrupled to drive the photocathode of the e-beam linac. This way,
both the linac and CO2 slicing systems are based on the same timing source,
which ensures adequate synchronization of the e-beam and CO2 laser pulse on
the picosecond time scale as required by the microbunching experiment. The
light split from the GE100 oscillator output for temporal slicing is seeded into a
two-stage regenerative amplifier system (b). A lens is used to focus the seed beam
on an intracavity iris in the first multipass amplifier stage to establish a high-
quality transverse mode. The sequential triggering of two intracavity Pockels
cells ultimately delivers a single 500 µJ, 100 ps pulse from the pulse train (7 ns
pulse separation) to the adjacent booster stage which brings the energy up to 15
mJ. This 1.064 µm pulse propagates through a mechanized delay stage (optical
trombone for ps synchronization with the e-beam) and is then co-propagated
with the 500 ns CO2 pulse at a small angle (∼ 2◦ ) through the Kerr cell, which is
transmissive to the 10.6 µm light only for the duration of the 100 ps pulse. The
slicing setup is described in detail in Tochitsky et al [103], the only difference
being that here the Kerr medium is used to perform the optical slicing rather
than two germanium (Ge) semiconductor slabs.
A Ge slab is used in the present setup (not shown) downstream of the CS2
170
1.4 0.05
1.2
0.04
1.0
signal @a.u.D
0.03
I @a.u.D
0.8
0.6 0.02
0.4
0.01
0.2
0.0 0.00
0 2 4 6 8 10 12 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Dz @cmD Dt @nsD
Figure 5.3: Measured temporal cross-correlation of the 100 ps CO2 laser seed
with the plasma formed by the e-beam on Germanium at the undulator exit.
The left plot shows the CO2 signal transmitted through a Ge slab as a function
of position on the optical slicing delay stage. The right plot shows the 100 ps
FWHM laser pulse, extracted from the derivative of the smoothed left curve.
Kerr cell as an optically gated switch to improve the temporal contrast of the
sliced 10.6µm pulse before seeding it into the 8 atm regenerative amplifier. This
chopping method is based on modulating the reflective and transmissive prop-
erties of the Ge slab by controlling the surface free-carrier charge density. The
1.064 µm pulse forms a reflective plasma on the surface a few picoseconds before
the arrival of the single 100 ps pulse. Pre-pulses pass through, while the desired
main pulse is reflected and delivered to the high-pressure multipass CO2 regen-
erative amplifier (e) through a 50% ZnSe output coupler cavity mirror. There,
the single pulse is amplified in the multipass cavity which converts the nJ seed
into a train of 10 mJ pulses. The pulse train then propagates back out the ZnSe
coupler to the Ge switch and, because it arrives well after the recombination time
of the reflecting surface plasma, is transmitted through the slab and is thus de-
coupled from the system. This injection mode-locking arrangement also prevents
the amplified pulse train from re-entering the upstream optics and lasers. After
switching through the final Pockels cell, the 40% efficiency of this arrangement
results in the 4 mJ pulse for the final booster amplifier stage (f).
171
5.3.2 Laser pulse measurements
Precise temporal synchronization of the 100ps CO2 laser pulse with the e-beam
in the undulator is measured via gating of a Ge slab at the undulator exit. The
principle is the same as that described earlier for improving the contrast of the
CO2 pulse; the plasma formed on the surface of the slab – this time by the co-
propagating e-beam – is opaque to the CO2 light. The surface plasma is formed
within ∼ 10 ps of the arrival of the e-beam and lasts for several nanoseconds. It
is therefore capable of completely blocking the entire 100 ps CO2 pulse, which
is much longer than the 3.3 ps e-beam. By measuring the light transmitted
through the slab as a function of the temporal delay, one can determine the
position of the e-beam with respect to the CO2 laser pulse. Figure 5.3 shows the
experimental results of the scan performed by moving the optical delay stage of
the 1.064 µm slicing pulse in the laser room, which shifts the timing of the CO2
pulse with respect to the e-beam at the undulator. The Ge slab was positioned
on a movable in-vacuum probe 2 cm past the undulator exit. The e-beam and
the CO2 were co-propagated through a 1mm iris positioned just before the slab,
which helped to improve the transverse overlap of the e-beam and the laser spot.
A downstream mirror directed the transmitted 10.6 µm light out of the vacuum
chamber through an NaCl vacuum window to an mercury cadmium telluride
(HCT) detector. The HCT detector signal is measured, along with a reference
signal taken far upstream, and the ratio of the two signals is taken to remove
common mode fluctuations in laser power. When the e-beam arrives at the Ge
slab before the CO2 pulse, the opaque plasma blocks most of the light and the
HCT signal ratio is strongly attenuated (left plot in Fig. 5.3 at ∆z = 0). As
the CO2 pulse is moved earlier in time, the pulse emerges from the front of
the e-beam, eventually escaping completely. The measured ramping signal is
172
Table 5.1: Neptune CO2 laser parameters
Wavelength λ 10.6 µm
Pulse duration (FWHM) – 100 ps
Waist size (in undulator) w0 400 µm
Waist position (in undulator) zw 10 cm
Polarization σ 1 (LH)
Peak power P 30-80 MW
a temporal cross-correlation between the plasma created by the e-beam and the
laser pulse. Since the formation time of the plasma is short compared to the laser
pulse duration and lasts for a longer time, the plasma is essentially a step function,
so the temporal profile of the laser can be determined by differentiation of the
cross-correlation signal. A 100 ps FWHM laser profile is revealed, in quantitative
agreement with streak camera measurements of the laser pulse length. Finally,
for running the experiment, the timing is adjusted so that the e-beam sits at the
peak intensity of the laser pulse.
Maximizing the cylindrical symmetry of the transverse laser profile is key for
optimization of the microbunching into the dominant azimuthal mode, so care
was taken to preserve a high-quality Gaussian laser pulse throughout the optical
transport line. Further, both the spot size and waist position both have to be
carefully positioned near the middle of the undulator to maximize the interaction
with the e-beam. Transport from the laser system to the beamline bunker was
first modeled by the simple ABCD solver and was designed to place the 400
173
2.0
1.5
w0 x ,w0 y @mmD 1.0
0.5
0.0
z @mmD
100 150 200 250 300
Figure 5.4: Spot size of CO2 laser seed in the undulator in the HELiX exper-
iment. In x (blue) from fit (line): w0 = 393 µm, zw = 13.4 cm, zR = 4.6 cm
with M 2 = 1.04. In y (red): w0 = 403 µm, zw = 13.3 cm, zR = 4.8 cm with
M 2 = 1.21.
µm waist as close as possible to the longitudinal midpoint of the undulator.

Experimentally, the laser spot size and waist position was then measured as a
function of distance through the undulator using a Pyrocam with a mirror placed
at the undulator entrance. Results of the scan are shown in Fig. 5.4. The spot size
p
in each dimension is fitted to the formula w(z) = w0 1 + M 2 (λ(z − zw )/πw02 )2
as a function z for the parameters M 2 , zw and w0 . The measured results are in
good agreement with the transport model. Excellent symmetry is found between
the x and y beam dimensions, with a round optical waist of w0 = 400 µm a
distance 13 cm from the undulator entrance.
5.3.3 Electron Beamline
The Neptune photoinjector and linac systems together produce a high-brightness

beam of electrons at energies up to 14.5 MeV. The complete beamline is shown
in Fig. 5.5. Electrons are born in the RF photoinjector gun, which is a 1.625 cell
174
π-mode standing wave cavity. The coupling to the RF waveguide is through the
full-cell with a symmetrizing port on the opposing side to cancel dipole fields.
The peak on-axis electric field is 115 MV/m at maximum with the available
power, but is typically run at 100 MV/m (3.7 MW gun forward RF power)
to produce an e-beam energy of 4.6 MeV at the gun exit from a 30◦ launch
phase (60◦ in front of the 90◦ phase that corresponds to maximum field.) A
magnesium cathode was chosen for HELiX due to the 2-3 times better quantum
efficiency over the copper alternative. Emittance compensation at the gun port
exit is performed with a magnetic solenoid, the fringing fields of which generate
focusing to remove phase-space correlations that arise as space-charge and RF
defocusing fields act on the beam. At the proper tuning of the solenoid field
strength, the transverse emittance can be minimized for injection into the linac.
The normalized emittance can vary depending on the beam charge (up to a few
nC), but is nominally about nx = 4 mm mrad for 300 pC.
The linac is a 9 cell (7 full cells, 2 half cells, 42 cm long), plane wave trans-
former (PWT) π-mode standing wave structure. An S-band (2.856 GHz) SLAC
model XK-5 klystron provides up to 20 MW of RF power for both the gun (typ-
ically receiving 4 MW) and the PWT (about 10 MW), with an adjustable RF
phase shifter in between. Synchronization of the RF system with the laser sys-
tem is provided by using the 38.08 MHz signal from a quartz crystal oscillator.
A portion of the signal is split and sent to a phase-lock loop in the GE100 drive
oscillator, while the remaining 1W is frequency multiplied by 75 and amplified
by a pulsed 1kW solid-state amplifier to serve as a seed for amplification in the
klystron. Pulsing of the klystron is performed with a modulator that has a 3.5 µs
pulse length set by a 10 capacitor pulse forming network that is charged to 38.5
kV and discharged via thyratron switch into the pulse transformer, where the
voltage is multiplied by 12. Power is delivered to the RF structures in the beam-
175
Gun
Energy spectrometer
and chicane LINAC
Laser Quads
Folding
Box
Undulator
Box
Dogleg
(unused
for HELiX)
Quads
Figure 5.5: Neptune beamline
176
2 3 4 5 7 8 9 10
2.0
1.5
Σx , Σy HmmL
1.0
0.5
0.0
z HcmL
0 100 200 300 400 500 600
2 3 4 5 7 8 9 10
7
6
5
Εnx , Εny HΜmL
4
3
2
1
0
z HcmL
0 100 200 300 400 500 600
Figure 5.6: Cathode to undulator Neptune transport modeled with

UCLA-Parmela.
line through waveguides filled to 28 psi with sulfur hexafluoride, with forward
and revered power levels measured at different waveguide positions and recorded
in the control room.
The laser pulse for the photoinjector originates in the diode-pumped GE100
oscillator operating at a wavelength of 1.064 µm. The 650 mW (average power),
80MHz output train of 10 ps pulses is split; half of the 650 mW output is sent
to the amplifier system for CO2 slicing, and the other half goes to the Nd-doped
glass amplifier system which outputs a single 1mJ, 7ps pulse at a pulse repetition
frequency of 5 Hz. Each of the two subsequent external potassium dihydrogen
phosphate (KDP) crystals frequency-doubles the pulse, finally delivering a 266
nm wavelength pulse onto the photoinjector cathode. The work function of the
Mg cathode is 3.7 eV; less than the 4.6 eV photon energy. The phase of the RF
can be changed with respect to the arrival time of the laser pulse, permitting
precise control of the e-beam position within the RF cycle.
177
0.6 0.04
0.4
0.02
y HmmL 0.0
0.2
ΓΒz y' 0.00
-0.2 -0.02
-0.4
-0.04
x HmmL y HmmL
-0.4-0.2 0.0 0.2 0.4 -0.4-0.2 0.0 0.2 0.4 0.6
0.04 24.95
24.90
0.02 24.85
ΓΒz x' 0.00 ΓΒz 24.80
-0.02 24.75
24.70
-0.04 24.65
x HmmL z HmmL
-0.4-0.2 0.0 0.2 0.4 - 1.0
-1.5 - 0.50.0 0.5 1.0 1.5
Figure 5.7: Example beam distribution at undulator entrance from Parmela sim-
ulations.
Simulation of the Neptune beamline from the cathode to the undulator en-
trance was performed with the UCLA-Parmela particle tracking code [104]. About
105 -106 marcoparticles representing the ∼ 109 electrons are tracked by the code
in 6D phase space. Precise tuning of the laser, gun, solenoid, linac, and beamline
magnets is available in simulation, allowing optimization of the e-beam transport
line for proper injection into the undulator. A tune for the HELiX experiment
is shown in Fig. 5.6, and an example e-beam distribution in Fig. 5.7. The tune
is verified experimentally using known beamline optic calibrations and e-beam
spot size, emittance and charge measurements. The final beam distribution from
Parmela is then used as an input for numerical Tredi simulations of the mi-
crobunching process.
178
Table 5.2: Neptune beam parameters
Energy γ 25
Normalized emittance n,x 4 µm
Charge Ne e 300 pC
Bunch length σz (σt ) 1 mm (3.3 ps)
Relative energy spread σγ 0.5 %
5.3.4 Helical Undulator
A custom permanent magnetic helical undulator was designed and built for HE-
LiX to perform the helical microbunching. The baseline tuning of the undulator
period λw and field strength K̄ are in general set by the wavelength λ = 10.6 µm
of the seed laser and the nominal Neptune beam energy γ ' 12 − 13MeV. At the
second harmonic resonance, the undulator parameters must be such that
λw
1 + K̄ 2 .

λ' 2
(5.1)
4γ
From conception to construction, the undulator project was conducted in

three main stages. First, a virtual undulator was first modeled and tuned using
3D field solver software Radia [105]. Second, precise orthographic drawings of all
the individual pieces – from the supporting structure to the individual magnets
– were generated in SolidWorks and machined (big thanks to Harry Lockart
and crew at the UCLA physics machine shop). Lastly, the components were
all assembled in a Knudsen building lab at UCLA, and the undulator field was
179
Table 5.3: HELiX undulator parameters
Undulator parameter K 0.578
RMS on-axis field Bw 3.25 kG
Period λw 1.90 cm
Number of periods Nw 12
Gap – 1.0 cm
Focusing wavelength λβ 1.16 m

√
Polarization êw (êx + iêy )/ 2 (RH)
measured and tuned.
Various field strengths, polarizations and geometries of the individual un-

dulator magnets (Fig. 5.8) were first examined using the Mathematica add-on
package Radia [105]. The design follows the model of a similar undulator built
for a different experiment at Neptune [106]. Radia is a 3D magnetostatics solver
optimized for undulators. Different magnetic and non-magnetic materials can
be modeled, and the trajectory of a charged particle through the 3D field can
be computed (see Figs. 5.9 and 5.10). Conceptually, this permits evaluation of
both the e-beam trajectory through the undulator as well as the 3D effects on
the finite beam distribution, such as focusing. In practice, however, the solver
was too slow to efficiently calculate the transport of more than a couple particles,
so Radia was used primarily for optimizing the overall field strength (Figs. 5.13
and 5.14) and centroid trajectory, rather than for 3D beam transport.
180
z [mm] 200
150
100
50
0
10
0
x [mm]
!10
!10
0
10
y [mm]
Figure 5.8: HELiX undulator magnets in Halbach configuration, modeled by

RADIA. Note that the half magnets at the ends are pushed slightly in by a
calculated 330 µm to assure that the e-beam enters and exits along the central
axis.
0.5
30
0.4 20
Hx',y'L @mradD
10
H-x,yL @mmD
0.3
0
0.2 -10
0.1 -20
-30
0.0
z @mmD z @mmD
-50 0 50 100 150 200 250 -50 0 50 100 150 200 250
Figure 5.9: Modeled e-beam transport in RADIA. Note that the x deflection is
negative.
181
z @mmD 200
100
0
0.0
x @mmD
-0.2
-0.4
0.0
y @mmD
0.2
0.4
Figure 5.10: Right-handed helical trajectory calculated in RADIA for on-axis

electron.
Under the design constraints set in part by the available space in the inter-
action box, the spot size of the laser through the undulator, and the maximum
field obtained with most conventional permanent magnets, the chosen period was
λw = 1.9 cm with a corresponding scaled field of K̄ = 0.578. Other relevant pa-
rameters are listed in Table 5.3. After the optimal tuning of the undulator fields
was chosen from the Radia model, a mechanical design of the full undulator
assembly was built with SolidWorks 3D CAD software. A partially assembled
model is shown in 5.11, where the front structural support is removed to reveal
the design theme. The 6061 aluminum chassis supports and maintains alignment
of the four bronze rails (machined from 954 alloy) at each end. Each bronze rail
is machined with rows of slots on two adjacent sides that hold the magnets in
place which are mounted in holders. It is noted that this alloy of bronze is slightly
magnetic (contains 3-5 % iron), but the Radia model shows no difference in the
182
Figure 5.11: SolidWorks model of the HELiX undulator with endplate and tuning
plate sections removed to reveal the position of magnets in holders and bronze
rails.
field experienced by the beam compared with a non-magnetic material, and no

ill-effects were observed in the experiment.
The magnets are permanent Neodymium Iron Boron (NeFeB) magnets, each
with nominal measured characteristics: Br = 11,940 G, iHc = 2241 kA/m, bHc
= 918 kA/m, (BH)max = 274 kJ/m3 . One is pictured in Fig. 5.12 in the holder.
They are arranged in a standard Halbach configuration to maximize the available
field strength near the axis. In this configuration, the λw /4 thick dipole magnets
that create the dominant helical field are inter-spliced with axially polarized λw /4
thick magnets that enhance and smooth the field and push flux towards the axis.
Each period of each polarization plane has eight magnets, four per side. Two of
the four are dipoles and two are axial magnets. All told, the complement of 196
magnets is made up of only six different versions: two dipoles that as a pair define
183
Figure 5.12: Magnet glued in aluminium holder.
a north-south dipole field across the gap, two axially polarized magnets for the
Halbach configuration, and two complementary thin dipoles (λw /8 thick) that are
used for trajectory matching at the entrance and exit. Each magnet is glued with
high-vacuum TorrSeal into its own snug, lightweight mounting bracket machined
from 6061 aluminum by Mt. Ida Machining. The mounting bracket slides into the
precision grooves machined into the adjacent bronze structural rails, which hold
the magnet rigidly in two dimensions but allow the radial distance (the undulator
gap) to be adjusted for each magnet individually. Precision adjustment is made
with a 0-80 5/16” long socket cap tuning bolt that screws into the back of the
holder. The head of the bolt is sandwiched between two external mounting
plates, which are themselves rigidly mounted to the outside of the undulator
rails. This configuration allows the bolt to be turned in place (there is a small
hole in the outer plate that enables access to the hex socket cap) while moving
the magnet/holder assembly. This way precise tuning of each magnet externally
allows adjustments to be made when the undulator is fully assembled. Each full
turn of the bolt corresponds to a radial movement of 300 µm.
184
1 2 3 4 5 6 7 8 9 10 11 12 13
0.3
0.2
0.1
B [T] 0.0
-0.1
-0.2
-0.3
-0.4
0 50 100 150 200 250 300
z [mm]
1 2 3 4 5 6 7 8 9 10 11 12 13
400
Bmeas / Bradia [G]
200
-200
-400
0 50 100 150 200 250 300
z [mm]
Figure 5.13: Upper plot: On-axis magnetic fields modeled from Radia (solid lines)
and measured with a Hall probe (points) for the horizontal (red) and vertical
(blue) polarizations. Lower plot: Difference interpolated between points between
the measured and modeled fields. RMS difference is 61 G for the horizontal, 65
G for the vertical polarization.
1 2 3 4 5 6 7 8 9 10 11 12 13 1 2 3 4 5 6 7 8 9 10 11 12 13
0.03
0.6
0.02
0.5
0.01
0.4
K ∆K/K 0.00
0.3
-0.01
0.2
-0.02
0.1
-0.03
0.0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
z [mm] z [mm]
Figure 5.14: Left plot: modeled (red) and measured (blue) undulator K̄ = 0.578.
Right: Error ∆K̄/K̄, the rms value of which is 0.67%.
185
Measurements of the undulator field were made with a computer assisted
motorized scan with a LakeShore 8” long single-axis transverse Hall probe and
421 Gaussmeter. The probe was mounted on a Daedal motion stage controlled
by LabView 8.2 software that moved it along the z-axis and recorded the field
amplitude at each step. Both the horizontally and the vertically polarized fields
were independently measured and compared with the Radia model, as shown in
Figure 5.13. The 0.18” (0.46 cm) transverse probe dimension covered a large are
compared to the region of field experienced by the full e-beam during transport,
and thus precluded a fine-grained scan of the field geometry in the transverse
dimensions. Each measurement is therefore primarily the on-axis field, with the
difference between the measured and the modeled fields being less than 200 Gauss
throughout (∼ 60 Gauss rms).
In lieu of a detailed 3D field measurement, a sensitive measure of the proper

undulator tuning is the transport of an electron calculated by integration of the
on-axis field. The trajectory and velocity of the electrons as calculated from
the on-axis Radia field were compared with those calculated from the measured
fields, and the differences minimized by tuning each magnet individually. Because
the integrated motion is highly sensitive to field errors, particularly upstream,
this tuning method was repeated for trajectories calculated both forwards and
backwards until agreement was reached between model and measurement. The
resulting difference in the fields and the undulator parameter after tuning are
shown in 5.13 and 5.14, respectively. Qualitative confirmation of the electron
trajectory was also made with a subsequent pulse wire measurement. Though
the results showed a proper trajectory consistent with the model, the pulse wire
test mimics the transport of a particle with much higher energy (∼ GeV), and
is therefore not as sensitive to small errors that become pronounced at the much
lower 13 MeV design energy.
186
NaCl
window
CTR Foil
Pop-in Undulator
CTR
e-beam &
CO2 laser
3D Movable
Movable CTR mirror
probe
Figure 5.15: Arrangement of undulator and CTR screen in interaction box.
5.4 Analytic and numerical modeling of helical microbunch-

ing
The right-handed undulator is predicted to produce strong bunching into the

l = −1 mode for a left-circularly polarized laser seed, where the rotation of the
E-field vector is counter-clockwise along positive z for a fixed time (See Eq. 1.4).
Figure 5.16 shows the evolution of the bunching factor and the seed laser spot
size along the modulator for design parameters, as calculated from the general
expressions in Eq. 4.19 and 4.10. The bunching factor reaches a maximum of just
under 10% at 30 MW in the presence of betatron focusing, energy spread and
space charge effects. The contribution from each effect is shown independently,
as is their combined contribution (solid black line) which ultimately produces an
attenuated but not completely suppressed helical structure (See Section 4.1.2.1).
187
It is clear that the longitudinal space charge reduces the final bunching factor by
working against the microbunching process, though it is not the dominant effect.
At 300 pC, the quarter-plasma wavelength is π/2θp =16 cm, which is shorter
than the interaction length, but since the energy and density modulations develop
slowly, the effective reduction in the bunching factor is calculated to be only about
17% from Eq. 4.10. In contrast, the π/2kβ =30 cm quarter-betatron wavelength
due to focusing affects the distribution constantly throughout, both through the
active transport of particles and in the effective energy spread. Together with the
uncorrelated energy spread, these contributions dominate. With all the effects
included, the maximum bunching is predicted to occur not at the γ = 24.46
harmonic 1D resonance, but rather at the higher γ = 25 energy as shown in
(2)
Figure 5.17. This direction of detuning, θ0 < 0, is consistent with that of a 3D
FEL in order to maximize the coupling with the higher-order interaction mode.
With Eq. (4.19) it is straightforward to analytically explore the parametric

optimization in terms of multiple experimental knobs. The laser spot size can be
adjusted, for example, in conjunction with the detuning to maximize the bunch-
ing (Figure 5.18), as can the axial position of the waist (Figures 5.19 and 5.20).
In a matched beam, the normalized emittance determines the e-beam size, which
can also be optimized slightly (Figure 5.21), depending on the emittance com-
pensation upstream. Inspection of the parameter space represented in all these
figures shows that the set of parameters that generates the maximal bunching
factor is almost exactly the measured e-beam and laser parameters, save for the
slightly oversized laser spot size which would ideally be about 250 µm.
To verify the analytic results (from 4.19) that include the effects of e-beam
focusing, energy spread, and seed laser transport, numerical simulations were
performed with the particle code Tredi [90]. Tredi, in its latest incarnation, can
188
20 1.0
15 0.8
wHzL @mmD
Èb-1 È %
0.6
10
0.4
5
0.2
0 0.0
z @mD z @mD
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Figure 5.16: Calculated bunching factor inside HELiX undulator for a matched
beam at γ = 25, w0 = 400 µm, zw = 10 cm, P = 30 MW, σx = 180 µm,
σγ = 0.5%. The largest final bunching factor is obtained in the parallel, cold
beam limit. Longitudinal space charge alone reduces the factor by 17 % (red
line), whereas energy spread and emittance have stronger effects (overlapping
green and yellow lines). With all effects, the calculated bunching factor is shown
as the solid black line.
10
6
Èb-1 È%
4
0
23 24 25 26 27
Γ
Figure 5.17: Calculated and simulated bunching factor versus energy at the un-
dulator exit for HELiX. Resonance is γR = 24.46 for the λ = 10.6 µm laser seed.
The solid line is from theory (Eq. 4.19) show excellent agreement with points
obtained from Tredi numerical simulations.
189
Figure 5.18: Calculated bunching factor at the undulator exit for HELiX versus
energy and laser spot size. The waist position is fixed at zw = 10 cm.
Figure 5.19: Calculated bunching factor versus energy and laser waist position.
The spot size is w0 = 400 µm.
190
Figure 5.20: Calculated bunching factor versus spot size and laser waist position
at γ =25.
Figure 5.21: Calculated bunching factor versus energy and emittance at w0 = 400
µm and zw = 10 cm.
191
simulate the microbunching interaction with a user-specified 3D undulator field
file as well as an externally specified 6D input particle distribution. Accordingly,
the 3D field map from the Radia design model and the particle distribution
from the UCLA-Parmela simulations were used for a variety of assessments and
parametric scans. Excellent agreement was found between the prediction of the
analytic model and Tredi simulations for the evolution of the l = −1 bunching
factor, even when energy spread and focusing effects are included. This is shown
most clearly in Fig. 5.17 for b−1 at the undulator exit as a function of energy
detuning. Additional effects were made evident from simulations. These are
describe in the next section.
5.4.1 Excitation of adjacent modes
The excellent agreement between Tredi simulations and the analytic predictions
reinforces the use of either approach to model the baseline experimental condi-
tions for the l = −1 mode. But Tredi can be used to examine additional effects
not included in the analytic theory presented here. For example, the e-beam is
assumed beta-matched during undulator transport in the theory, whereas parti-
cle simulations allow the beam envelope to vary. Simulations can also be used to
examine non-ideal effects, such as electron beam and laser spot ellipticity, and
parallel and angular misalignments. While the effects of these errors can be mini-
mized experimentally, Tredi also provides important insight into high order effects
that can couple at harmonics, such as coupling to the axial EM field and, in par-
ticular, e-beam emittance, some of which are difficult to remove experimentally
due to systematic constraints.
One such feature that couples to adjacent l modes in the e-beam is the off-axis
e-beam orbit in the undulator, which is a result of the design transport. Shown
192
in 5.9 and 5.10, the e-beam centroid for injection on-axis is displaced inside
the undulator an average of ∼ 400 µm off-axis, and has a mild curvature due
to focusing. While the seed laser axis can be parallel shifted to compensate the
secular offset, the displaced orbit and slight curvature lead to additional harmonic
coupling to adjacent modes for a variety of reasons. First, as described by Colson
[107] and Xie [108], imperfect trajectories (angles in the electron velocity) lead
to harmonic coupling which provides a mechanism by which several l modes can
be excited. From the standpoint of cylindrical symmetries, it is relatively easy
to visualize how the transverse motion of the e-beam during a curved trajectory
against the static laser profile can excite other modes. The e-beam samples
different portions of the laser field during the arc, which couples to adjacent
l-modes in the microbunching distribution. Figure 5.22, generated from Tredi
results, illuminates the extent to which this motion excites the undesirable l =
0 and 1 density modes in addition to the design l = −1 mode. To quantify
the contribution due to the trajectory, two results are shown that compare two
undulator transport geometries, both tuned to the same interaction conditions.
The undulator on the left is the HELiX design, while the other is tuned to force
the e-beam trajectory to be on-axis. Between the two, the bunching factor into
the targeted l = −1 mode is identical, but bunching into the l = 0 mode appears
in the HELiX undulator with the same amplitude as the target l = −1 mode
due partly to the curvature in the trajectory.
There is a second non-ideal feature evident in the two plots in Fig. 5.22.
While bunching into modes other than l = −1, 0, 1 is negligible and within the
statistical noise, excitation of the distant l = 1 bunching mode is clearly observed
and is of a similar magnitude in both cases. This mode is likewise not expected to
be excited according the the analytic theory presented previously, but it’s source
is again explicable. Physically, this additional coupling is due to two issues. The
193
a) b)
12 12
10 10
8 8
b  6 b  6
4 4
2 2
0 0
10.6 10.7 10.8 10.9 10.6 10.7 10.8 10.9
ΛΜm] ΛΜm]
400 150
300 100
x,y Μm
x,y Μm
50
200
0
100
50
0
100
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
z cm
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
z cm
Figure 5.22: Numerical simulation of 10000 particles. Comparison between helical

bunching factors into the l=1 (red), 0 (blue) and -1 (purple) modes in the a)
HELiX undulator and in b) an identically tuned undulator with on-axis e-beam
transport.
first issue, which is common to both the on-axis and off-axis transport geometries,
is the fact that the matched e-beam size, σx = 180 µm, is not much larger than
the ∼ 140 µm helical orbit amplitude. It was assumed that σx K/γkw in the
simplification of the Taylor expanded fields in 4.8, so when this constraint is not
strongly satisfied, the purity of the excited l = −1 mode is compromised. The
reason is essentially the same as in the arcing trajectory case; the e-beam profile
moves about the laser axis and samples different field amplitudes and transverse
gradients. Consider for a moment the opposite case where the e-beam profile is
large compared to the wiggle motion; the electrons on one side of the e-beam
make small orbits within a field gradient that is essentially unchanging over an
undulator period and has the opposite sign as the gradient on the other side of
the e-beam. The electrons on opposite sides are therefore modulated according
to the ∓(h − 1)φ phase term, and the helical structure thus emerges. In this
194
0.3
10
0.2
8
ÈbÈ % 6
0.1
x,y @mmD
0.0
4 -0.1
2 -0.2
0 -0.3
Λ @ΜmD z @cmD
10.6 10.7 10.8 10.9 0 5 10 15 20 25 30 35
Figure 5.23: Transport of the e-beam through the HELiX undulator can be
made on-axis by adjusting initial centroid displacement and angle. The resulting
simulated bunching factors into the l=1 (red), 0 (blue) and -1 (purple) modes
are identical to a purely on-axis undulator.
σx ∼ K/γkw case however, the large relative orbit samples large variations in the
fields and gradients which tend to smear out the local phase structure after an
undulator period. This also contributes to the multi-mode coupling in the case of
the curved trajectory wherein the e-beam samples different portions of the laser
peak, only in this case, the smearing of the structure occurs primarily along the
axis in the plane of curvature. It is distinct in this way in that it breaks the
cylindrical symmetry and couples strongly into the fundamental.
The second issue appears to be the harmonic coupling due to the collective
transverse electron momenta, as described [108]. This effect exactly vanishes for a
zero-emittance e-beam but is not properly treated by the matched beam equations
in 4.19. It is important to note that the relative value of the l = −1, 0 and 1
modes is independent of the seed laser power for Neptune values. In light of both
of these effects, to maximize the purity of the helical microbunching geometry
it is advantageous to both minimize the e-beam emittance and to maintain the
σx K/γkw constraint. Both of these are easily satisfied at the high energies
associated with short-wavelength FELs for modern injectors, but here result in
non-removable coupling to other modes.
It is desirable, of course, to minimize bunching into adjacent modes. While
195
the bunching into the undesirable l = 1 mode persists with the fixed Neptune
emittance, it is clear from Fig. 5.22 that the more crucial l = 0 fundamental
can be mitigated to some extent by transport. One solution to this issue is to
retune the undulator to support axial transport. While this is advisable for future
designs, the better, simpler solution to the problem (though slightly challenging
in practice) is to alter the e-beam centroid and trajectory angle at the HELiX
undulator entrance so that the primary helical orbit ends up being about the
central axis. As shown in Fig. 5.23, the e-beam entering with a displacement
of ∼ 270 µm (opposite of the kick side) and at an angle (away from the axis)
of γβx , γβy = 0.035 has an intra-undulator trajectory and exit bunching factor
virtually identical to that of the on-axis undulator in Fig. 5.22.
5.5 Coherent Transition Radiation from helical beam
In order to properly measure and verify the bunching of the electron beam into
a helix, one must be able to either resolve the correlated e-beam structure di-
rectly (perhaps with a high-resolution deflector scheme as proposed in [109]) or
otherwise be able to resolve a secondary effect such as the phase structure of
the radiation emitted by the beam in some process. Since direct high-resolution
e-beam measurements are not currently available at Neptune, the latter indirect
methods are pursued. One of the simplest techniques to measure microbunching
(in lieu of the downstream HGHMG radiator) is with coherent transition radia-
tion, or CTR [53, 110, 111, 112, 113]. In such setups, the microbunched e-beam
strikes a dielectric or metal surface and emits radiation that is strongly peaked
at the microbunching frequency. The radiation also carries signature features
of the e-beam microbunching structure [102]. Analytic studies have shown that
the CTR emitted from the helically microbunched beam will carry an embedded
196
helical phase structure that corresponds to the azimuthal modes in the beam
[55]. Due to its relative experimental simplicity, CTR is thus the measurement
of choice for HELiX.
5.5.1 CTR from ideal beam
The first experimental assessment of an interaction is simply a measurement of

the total emitted CTR in order to establish that microbunching is taking place.
Many details of CTR emission from a helical beam distribution are worked out
analytically in the Appendix E. Under the assumption of a single dominant l
mode in the e-beam, the total CTR energy is given in general by E.24:
(1) Ne2 e2 2 γ 4 (|l| + 1)!

(1)
UC,l ' √ bl . (5.2)
3
16 π 0 σz σx kb [(|l|/2)!]2 2|l|
The baseline Neptune beam values in Table 5.2 at 30 MW laser power return
an expected CTR energy of 5 pJ for a lone l = −1 mode, above the smallest CTR
energies (∼ 0.8 pJ) recorded at a previous Neptune experiment [102]. With the
microbunching in the adjacent l modes as indicated by Tredi simulations in Fig.
5.23, the total energy becomes 9 pJ, since the other modes also emit coherently.
With sufficient CTR signal levels, the far field transverse distribution can be
measured to provide information about the intensity structure (and ultimately the
phase structure) of the radiation. The angular spectral energy is given in Eq. E.23
and is plotted along with the angular distribution in Fig. 5.24 for the Neptune
parameters, assuming a single l = −1 e-beam mode. As expected, the spectral
emission is sharply peaked at the microbunching frequency, and is emitted into
an angular cone that subtends less than the 2/γ = 80 mrad window. These
analytic results are supported by numerical particle simulations from QUINDI
197
Figure 5.24: Left: The calculated angular CTR emission spectrum shows strong
peaks both at the microbunching frequency kb =2π/10.6 µm and at the opening
angle θ = 13 mrad. Right: Angular distribution at the frequency peak at a
distance of 30 cm from the foil. The units of d2 U/dkdΩ are pJ - s/ rad.
[114], a particle tracking code that calculates radiation using an algorithm based
on the Liénard-Wiechert potentials. An output CTR distribution from QUINDI
is shown in Fig. 5.25 where the total intensity and phase are shown, as well
as the intensity and phase of each polarization. The particle distribution used
for the input was an output distribution from Tredi for a simple setup with
no energy spread, emittance or focusing effects. The microbunching into the
single azimuthal mode is thus highly pure, a fact attested to by the purity of
the simulated CTR signal. A clear helical phase structure and highly symmetric
intensity ring are evident in the far field plots. It is interesting to also note the
contribution of the radial polarization to the phase structure which produces an
additional phase shift across the central axis. The radial polarization is naturally
accompanied by a z polarization component (far right plots, Fig. 5.25), the phase
of which maps directly to the helical e-beam structure since the z-component
198
Intensity x!pol Intensity y!pol Intensity z!pol Intensity
14 14 14 14
12 12 12 12
10 10 10 10
8 8 8 8
6 6 6 6
4 4 4 4
2 2 2 2
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14
phase x!pol phase y!pol phase z!pol phase
14 14 14 14
12 12 12 12
10 10 10 10
8 8 8 8
6 6 6 6
4 4 4 4
2 2 2 2
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Figure 5.25: Simulated CTR output from QUINDI particle simulations of pure
l-mode microbunching.
otherwise has the same sign across the beam. These features provide a guidepost
for the ideal CTR emission one can expect from a helically bunched e-beam.
While the ringed intensity distribution of the CTR is reminiscent of the hollow
OAM intensity profile, it is important to note that CTR in the far field always
has a ringlike intensity structure, even in the absence of a helical phase. Thus
the presence of a helical e-beam structure is difficult to ascertain by simple ob-
servation of the CTR profile alone. Figure 5.26 shows the angular emission for
different l-modes, where it is clear that all of them have an axial null due to the
radially polarized light that results from the CTR emission process. This is in
direct contrast to FEL emission for example, where the presence of an annular
profile is good evidence of a helical phase discontinuity at the first harmonic of
the microbunching wavelength. Thus, the CTR method of examining the e-beam
structure, while dramatically simpler experimentally than an FEL radiator, is
complicated by the inherent hollow profile.
The issue of how to resolve the helical structure persists. One may note that
199
2.0
d 2 UdkdW @pJ - sradD

1.5
1.0
0.5
0.0
Θ @mradD
-40 -20 0 20 40
Figure 5.26: The calculated angular CTR emission for different l-modes all share
an on-axis null due to the nature of the CTR emission process.
the angular peak of the distributions in 5.26 varies with the mode number. At
k = kb the peak is given by the angle [55]
s
|l| + 1
θ= , (5.3)
2γ 2 + σx2 kb2
which is 9, 13, and 15 mrad for the 0, 1, and 2 modes, respectively. Thus in
principle, the existence of a dominant l-mode could therefore be determined by
measurement of the intensity ring size, all other things being equal. In practice
this approach is not so simple. Figure 5.22, shows that more than a single l-
mode is present in the beam. Each of these modes will emit, producing the
CTR distribution patterns displayed in Fig. 5.27. No matter which undulator
transport is used, no single, dominant intensity ring in the CTR distribution
stands out by which one could hope to quickly discern the dominant mode from
the angular peak. The intensity profile is thus not a reliable indication of the
modal content by itself for the HELiX. It would therefore be useful to also obtain
200
off!axis on!axis
1.0 1.0
0.5 0.5
y !cm"
y !cm"
0.0 0.0
!0.5 !0.5
!1.0 !1.0
!1.0 !0.5 0.0 0.5 1.0 !1.0 !0.5 0.0 0.5 1.0
x !cm" x !cm"
Figure 5.27: Simulated CTR intensity distribution comparison between

beta-matched e-beam with on-axis transport and one with off-axis transport.
some phase information, which is in principle available experimentally with a few

clever tricks.
5.5.2 Helical Phase Detection
Certain techniques are required for unambiguous determination of the radiation

phase distribution, and therefore by extension, the helical structure of the e-beam.
There are numerous methods by which the helical phase can be determined. A
knife-edge or wire can be inserted into the light path near the phase singularity
to produce a forked diffraction pattern that can be imaged [25]. The number of
prongs in the pattern is directly proportional to the dominant l-mode, which can
be thus inferred simply by counting. Optical mode conversion elements such as
cylindrical lenses can also be used to convert the cylindrically symmetric OAM
mode into a higher-order Hermite-Gaussian mode, again allowing determination
by simple counting of intensity lobes [22, 115]. Along these same lines, several
techniques that work for generating OAM modes have also been shown to work
in reverse; forked diffraction gratings [23], spiral phase plates [20] and non-spiral
phase plates [116] can be employed. Direct visualization of the transverse phase is
201
perhaps most straightforward with a Shack-Hartmann-type [117] wavefront sensor
that uses a micro-lens array (or something analogous) to reveal the wavefront
curvature. Interferometric techniques are also used in which the input OAM beam
is sorted according to the l number [118], in principle allowing determination of
the OAM content of the mode at the single photon level.
5.5.2.1 Triangular aperture
As long as there is enough CTR signal to measure the transverse profile, there
is a simple method that can provide direct phase information. It comes from
a scheme described by Hickmann et al [119] in which a dominant helical phase
structure is inferred by simple examination of the radiation profile after it diffracts
through a triangular aperture. The technique is simple and is illustrated for two
ideal cases in Figure 5.28. In the first image, the radiation passing through the
aperture is regular, radially polarized CTR with no helical phase, such as would
be emitted from an e-beam with l = 0 microbunching. The resulting detector
image is similar in profile to the annular CTR intensity profile itself, namely,
a ring with a null on-axis. In contrast, CTR emission that carries a dominant
helical imbedded phase produces a dramatically different detector profile after
passing through the triangular region. As seen in the second (lower) plot in 5.28,
the diffracted image has an on-axis peak that is surrounded by triangular lattice
of intensity nulls1 . The presence of the on-axis peak provides clear evidence of
the azimuthal phase variation, the l− number of which is related directly to the
number of peaks and nulls in the surrounding triangular lattice. For an ideal
beam the modal distinction is clear.
This technique is modeled for the CTR emission of the non-ideal helical beams
1
This is distinct from Hickmann [119] in which there is instead a central null surrounded by
a triangular lattice of peaks. The difference arises because CTR is radially polarized.
202
Aperture Detector
y y
x x
Aperture Detector
y y
x x
Figure 5.28: Modeled ideal CTR profile viewed through a triangular aperture.
Top: CTR emission from an l = 0 e-beam. Bottom: CTR emission from an
l = −1 e-beam.
203
Aperture at 10. cm Detector at 30. cm
4 2
2 1
y!mm"
y!mm"
0 0
!2 !1
!4 !2
!4 !2 0 2 4 !2 !1 0 1 2
x!mm" x!mm"
Aperture at 10. cm Detector at 30. cm

4 2
2 1
y!mm"
y!mm"
0 0
!2 !1
!4 !2
!4 !2 0 2 4 !2 !1 0 1 2
x!mm" x!mm"
Figure 5.29: Modeled profile of CTR sent through a triangular aperture (3.8 mm
on a side) positioned 10 cm from the foil. The CTR distribution is calculated
from the modal bunching factors and beam parameters obtained from numerical
Tredi simulations. Upper plot: The CTR fields generated by a beam with an
off-axis undulator contain significant l = 0 content, which is clearly seen in the
diffracted CTR image (right upper plot) as seen on a screen 30 cm from the foil as
a donut. Lower plot: An on-axis undulator orbit preserves the dominant l = −1
mode as seen in the diffracted image (lower right) as a distinct on-axis peak in
intensity.
204
generated in Tredi simulations to provide insight into the experimental scenario.
When sent through the triangular aperture, the fields used to generate the in-
tensity plots in Figure 5.27 for the on-axis and off-axis undulators generate the
diffracted intensity patterns shown in Figure 5.29. The patterns reveal the modal
content of the CTR. The upper plot is generated from a beam that experienced
the off-axis orbit of the HELiX undulator. It is clear that, even though the l = 0
and l = −1 modes have a similar microbunching amplitude, the CTR signal ap-
pears to be dominated by the l = 0 mode, as evidence by the hollow diffraction
profile. This is in stark contrast to the case of the on-axis beam, which has a
50% smaller l = 0 bunching factor. The pattern shows a clearly dominant l = −1
helical phase characterized by the pronounced central peak (bottom right plot
in Figure 5.29). The distinction between these is readily apparent provided that
the CTR signal is large enough to measure the transverse profile.
5.5.2.2 Interferometer
For CTR signals that are too small to be measured with a 2D profile monitor but
large enough that a total time-integrated CTR signal can be detected, interfero-
metric techniques may be used to infer the transverse phase structure. A modified
Mach-Zehnder interferometer much like that described in [118] was constructed
for this purpose. Shown in Figure 5.30, it is designed to interfere the input CTR
signal co-linearly with itself flipped about one axis. It functions on the principle
that an |l| = 1 phase in the CTR is distinguishable from |l| = 0 by the total
power of the output. As shown, the CTR signal passes through two alignment
irises, and is split by a beam splitter (BS at a)). Half the signal intensity travels
through a delay stage c), and the other half is reflected off of two mirrors before
the two recombine at the second beam splitter b). The signal is measured by
205
Figure 5.30: Helical phase interferometer.
a detector at position d). Similar to the scheme in [118], the extra reflection in
one leg interferes the beam with a flipped image of itself, generating lobed inten-
sity patterns that, in principle, identify the helical phase content. Figures 5.31
and 5.32 demonstrate the effect for ideal scenarios with a single azimuthal mode.
Due to the radial polarization, the l = 0 mode produces two-lobed patterns (or a
donut, depending on the delay), while the |l| = 1 mode produces a pattern with
four lobes.
The actual e-beam modes are not pure, of course, and this again limits the
utility of counting intensity lobes. The interferometer provides a further benefit,
however, particularly if the transverse profile cannot be measured. The integrated
output signal, obtained by focusing the output profile into a single detector,
distinguishes between modes as a function of interferometer delay. In Figure 5.31,
a lone l = 0 CTR mode has a perfectly flat output power signature, whereas the
ideal |l| = 1 mode in 5.32 has a strong variation. Therefore, in principle, a strong
l = ±1 bunching mode can be directly inferred from the sinusoidal interferometer
signal.
206
Input CTR signal
Integrated detector signal

1.0
0.8
Udet Uin
0.6
mode=0 CTR (ideal)
0.4
0.2
0.0
interferometer delay @degD

0 50 100 150 200 250 300 350
Output delay=0 Output delay=90 Output delay=180
Figure 5.31: Idealized l = 0 CTR interferometer signal.
Input CTR signal
Integrated detector signal

1.0
0.8
Udet Uin
0.6
mode=1 CTR (ideal)
0.4
0.2
0.0

0 50 100 150 200 250 300 350
Output delay=0 Output delay=90 Output delay=180
Figure 5.32: Idealized l = 1 CTR interferometer signal.
207
0.7
0.6
Udet Uin
0.5
0.4
0.3

0 50 100 150 200 250 300 350
Figure 5.33: Calculated normalized total interferometer output power as a func-

tion of delay for different modes, using measured σ and π polarization reflection
coefficients for the beam splitters. The CTR signal from on-axis beam (solid blue
line) is almost identical to the signal from a pure l = −1 mode. The off-axis
beam (solid red) contains more l = 0 mode content and approaches the signal of
a single l = 0 mode beam.
To assess the efficacy of this scheme in experimental conditions, it is neces-

sary to first determine the reflection coefficients of the mirrors and beam splitters
in the interferometer. This was done using a small tabletop CO2 laser pulse
for calibration. The laser pulse was sent through each optical element individ-
ually, for both σ and π laser polarizations obtained with a λ/2 waveplate. The
anti-reflection (AR) coated mirrors showed no discernible difference in reflectiv-
ity between polarizations, each about 50% for a wide range around 45 degree
incidence. Two detectors measured the reflected and transmitted signals for the
beam splitters, and the signal at each recorded. The beam splitters are AR
coated on only one side, optimized for 50/50 reflection with σ polarization at
45◦ incidence. Measurement of the AR side reflectivity gave coefficients Rσ =45%
and Rπ =22% for each BS, and Rσ =53% and Rπ =26% for light incident from the
backside (no AR coating). The splitters are thus effectively symmetric with re-
spect the light incident on the front or back, which is relevant only for the second
BS at b). Using these numbers, the calculated interferometer signal is shown in
208
Figure 5.33 for different CTR inputs for comparison. The interferometer power
signal is normalized to the power of the incoming CTR to also ascertain the losses
in the system. As shown, the e-beam that experiences on-axis transport through
the undulator and has a dominant l = −1 mode has a signal virtually identical to
that of a purely |l| = 1 mode beam. Interestingly, it has a larger predicted peak
to peak power fluctuation; the minimum value is 58% of the maximum value,
compared with 53% for the pure |l| = 1 mode. The off-axis beam is not much
different in fact, showing a strongly modulated interferometer signal (46%) even
with a large l = 0 bunching factor. Due to the asymmetric reflectivity between
polarizations, the ideal l = 0 beam also has a small modulation of 21% of the
maximum value.
5.6 Experimental Results
Three different data scans of the total CTR signal measured at the LN-cooled
HCT were performed. The first was a scan of the CTR signal as a function of the
seed CO2 laser input power. This scan happens automatically due to inherent
shot-to-shot laser fluctuations. The second scan recorded the CTR signal as a
function of two different input laser polarizations, circular and linear. Third, the
e-beam energy was varied across most of the expected interaction bandwidth.
For each shot in all the scans, three signals were recorded electronically from
oscilloscopes in the control room and saved to a data file: the e-beam charge
at the undulator entrance as measured by a calibrated integrated current trans-
former (ICT) positioned at the entrance of the undulator; the laser power as
measured by a reference energy meter; and the CTR signal level measured at the
HCT detector positioned outside the salt window on the vacuum chamber. The
reference detector for the laser power was positioned to non-invasively collect a
209
small portion of the CO2 pulse reflected off a tilted salt window placed in the
upstream laser transport line, after the booster amplifier. The reflected portion
was focused into the detector by a lens. This detector signal was calibrated using
a separate Gentec energy meter (125 mV/mJ) placed directly in the downstream
laser path which collected the full of the transmitted laser pulse, and both signals
were measured simultaneously. The reference detector calibration was found to
be 1.5 V/mJ at this position without any attenuation on a small (about 80 µJ)
input signal, but during runs with up to 30 mW of laser power the signal had to
be attenuated by a 1mm thick slab of teflon, which itself was recorded to reduce
the signal by a factor of 120.
A 2 inch off-axis parabolic mirror directed and focused the CTR signal into the
HCT detector, the peak signal of which was recorded. The background 10.6 µm
emission from the seed laser was also recorded separately over many shots with no
e-beam, and the resulting data file used for a baseline noise measurement that is
subtracted from the HCT signal in the analysis. This background was minimized
to a mean value of 23 mV by placing the 1.5 cm square, 10 µm thick aluminum
CTR foil to the front side of the probe assembly (Fig. 5.15) and flush against
the 1 cm diameter undulator exit port. Any CO2 light the could leak around
the foil edges was further blocked by adding an additional 1 cm outcropping of
thicker aluminum foil about the periphery of the CTR foil, and also by placing
an 8 mm iris aperture 7 cm downstream of the foil. This way, in principle, only
the CTR signal that was emitted by the e-beam from the back of the foil passed
cleanly through the iris and out the salt window to the HCT. Without these
measures, the pJ-level CTR signal would have been swamped by the 3 mJ CO2
pulse background.
The scans of the laser power and the polarization were done in tandem. The
210
8
CTR @pJD
4
C02 Power @MWD

0 5 10 15 20 25 30
Figure 5.34: Measured CTR signal at HCT detector as a function of laser power,
for LH (blue) and linear (red) laser polarizations. Solid lines are linear fits to
data.
two polarizations were generated by a λ/4 waveplate positioned 1 m downstream

of the booster amplifier. Figure 5.34 shows the measured CTR signal emitted
from the bunched e-beam, normalized at each shot to the charge squared, as a
function of the laser power for the 100 ps CO2 laser pulse. During the scans the
charge stayed steady at a mean value of 240 pC, with rms fluctuations of 9 pC,
and the e-beam energy was 12.5 MeV with a FWHM relative energy spread of
1.3%. The transverse e-beam size at the undulator exit, measured on the YAG
screen on the movable probe, was found to be near the matched beam value of 180
µm. As seen in the figure, the correct circular polarization (left-handed for this
right-handed undulator) was observed to consistently generate an approximately
49% larger signal at the HCT than linear polarization. This is very close to the
expected value of 50% given by the dependence of the CTR energy on |b|2 , which
itself scales according to the polarization coupling factor |ĝ|2 , which is |ĝ|2 = 1 for
LH laser polarization and |ĝ|2 = 1/2 for linear. CTR energies up to 9 pJ (5×108
photons at 10.6 µm) were observed for shots with circularly polarized light at 30
MW of power, very close the levels expected from theory and simulation. This
CTR energy generated a raw signal of 95 mV on the HCT, corresponding to a
211
8
UCTR @pJD
X
4
X
X X
X X
24.2 24.4 24.6 24.8 25.0

Γ
Figure 5.35: Measured CTR signal at HCT detector as a function of e-beam

energy for shots with 25±2 MW of laser power.
signal-to-noise ratio of about 3. This is discussed in greater depth in Section 5.7.
The third scan, in which the CTR signal was measured as a function of e-
beam energy, was done by adjusting the RF attenuation in the linac to change the
energy. Results of this scan are shown in Fig. 5.35, where a clear increase in the
CTR signal is observed as the e-beam energy approaches the 12.5 MeV interaction
peak. Both the shape and the peak level are consistent with expectations.
It is important to note that a series of null tests confirmed that the signal
measured at the HCT was indeed generated by e-beam as a result of its interaction
with the laser and undulator fields. The signal was observed to vanish in several
instances: when the CO2 laser pulse was blocked, when the e-beam was removed
(by blocking the photocathode drive laser), when the timing synchronization
between the e-beam and the 100 ps CO2 pulse was lost (by intentionally moving
the CO2 pulse too early or late), when the e-beam energy was well outside the
0.5 MeV bandwidth of the interaction (peaked at 12.5 MeV), and when spatial
alignment between the e-beam and laser was destroyed by steering the e-beam so
that it still hit the foil, but so that it was grossly misaligned to the laser in the
undulator.
212
10 8
X
8 6
UCTR @pJD
X
ÈbÈ%
6 4
X
X X
4 2
X X
2 0
24.2 24.4 24.6 24.8 25.0 25.2 25.4 24.2 24.4 24.6 24.8 25.0 25.2 25.4
Γ Γ
Figure 5.36: Left plot: Bunching factors for the l = −1 (blue), l = 0 (red), and
l = 1 (yellow) modes from Tredi simulations for 25 MW of laser power. Right
plot: Comparison of measured CTR signals (denoted by X’s) with CTR energies
calculated from bunching factors from Tredi simulations in the left plot. The
blue line is the CTR calculated from the l = −1 alone, while the thicker black
line is the sum of the CTR energies from all the modes.
5.7 Discussion: Evidence for helical microbunching
The data provide several compelling indications that helical microbunching of

the sort described in this dissertation was observed for the first time in this ex-
periment. While the running conditions provided ample total CTR energy levels,
the low signal-to-noise ratio precluded implementation of the more telling trans-
verse profile diagnostics such as the interferometer and the triangular aperatures
(See Section 5.5.2) to directly determine the microbunching geometry. There
are, however, numerous indications that the helical structure was generated and
emitted CTR with a helical phase. These are elucidated by comparison of the
data with Tredi simulations for the same running conditions.
First, the scaling of the CTR energy with the different laser polarizations
matches expectations in terms of the relative CTR amplitude, and clearly shows
that microbunching was generated as the helical e-beam motion coupled to the
input laser. This alone does not indicate helical microbunching, but shows that
the second harmonic interaction was indeed taking place.
213
A strong clue to the modal content of the microbunching structure is found
from the energy scan data in Fig. 5.35, where both the shape and the maximum
signal match closely with the total CTR emission energy calculated from Tredi
simulations at 25 MW. This is shown in Fig. 5.36. In the left plot, the bunching
factors for the l = −1, 0 and l = 1 modes is shown2 . The desired l = −1 has the
largest bunching factor across most of the bandwidth, peaking at 10% at 12.5
MeV as expected. The CTR energy emitted from this l = −1 mode alone can be
calculated directly from Eq. 5.2, and is shown as the blue line in the right plot.
Also shown are the CTR energies from experimental measurements, indicated by
X’s. The CTR signal from the l = −1 mode alone peaks at 3 pJ, which is below
the measured 6.1 pJ peak at 12.5 MeV. Of course, the l = −1 mode is not the
only mode in the e-beam that radiates. If we also add the calculated emission
from the other modes, indicated by the thick black line in the right plot in Fig.
5.36, we find values very close to those that are experimentally measured. This
agreement between simulations and measurements provides strong evidence that
the bunching factors in the e-beam are closely represented by the curves in the
left plot, and thus that the l = −1 mode was the dominant one. This further
indicates that the desired on-axis transport of the e-beam through the undulator,
as shown in Fig. 5.23, was achieved experimentally, as the laser transport axis
was carefully aligned to the undulator axis.
While the measured CTR energies in Fig. 5.36 below γ = 25 are in good
agreement with simulations that indicate a dominant l = −1 mode, the measured
value at γ = 25 is significantly larger. This could be due to a few things, such
as fluctuations in e-beam size or energy spread. At this level, neither of these
dramatically affect the relative dominance of the l = −1 mode and thus do
not offer a contradiction to the notion of a helically bunched beam. But it is
2
Excitation of adjacent modes is discussed in Section 5.4.1
214
14
6
12
5
10
UCTR @pJD
8 4
ÈbÈ%
6 3
4 2
2 1
0 0
Parallel misalignment @ΜmD Parallel misalignment @ΜmD

0 200 400 600 800 1000 0 200 400 600 800 1000
Figure 5.37: Coupling to other microbunching modes through parallel misalign-

ment of e-beam and laser. Left plot: Bunching factors for the l = −1 (blue),
l = 0 (red), and l = 1 (yellow) modes from Tredi simulations for 25 MW of laser
power and at γ = 25 as a function of a lateral displacement of the laser beam
with the axis of the e-beam motion. Right plot: CTR energy from the l = −1
(blue) mode and from all the modes (black). The horizontal line indicates the
measured CTR value.
useful to explore other non-ideal conditions that can adversely affect the mode
purity so we can quantify the limitations on the experiment. One of the simplest
experimental design conditions to violate is the alignment of the laser with the
e-beam trajectory in the undulator, either via a parallel offset3 (where the beams
propagate in parallel but not coaxially) or via an angular misalignment (where
the beams cross in the undulator). Simple steering of the e-beam can lead to
both, and both of these lead to a reduction in the l = −1 bunching factor and
an increase in the l = 0 mode, so it is instructive to examine their effect in this
case.
Figure 5.37 illustrates the effect of a parallel misalignment on the microbunch-

ing and the resulting CTR emission from Tredi simulations at 25 MW and γ = 25.
The left plot shows how the bunching factors of the l = −1, 0 and l = 1 modes
vary with the magnitude of the misalignment. With the σx =180 µm e-beam, a
displacement of 100 µm is sufficient to make the l = 0 mode the dominant one,
3
See Section 4.2.1 in the case of short wavelength modulators.
215
14 8
12
6
10
UCTR @pJD
8
ÈbÈ%
4
6
4 2
2
0 0
Angular misalignment @mradD Angular misalignment @mradD

0 2 4 6 8 10 0 2 4 6 8 10
Figure 5.38: Coupling to other microbunching modes through angular misalign-

ment of e-beam and laser for the same conditions as Fig. 5.37.
and it continues to grow to it’s maximum until the shift is σx . In the right plot,
the CTR signal of the lone l = −1 mode quickly reduces to zero, while the sum
total of the CTR signals reaches a peak at a shift of σx where the microbunching
distribution is dominated by the l = 0 mode. The horizontal line is given by the
experimental measurement; this is the 6.1 pJ CTR signal observed at 25 MW
and γ = 25. If there is any actual experimental misalignment of this type, the
amount is not precisely known (hence the representation as a line rather than a
point), but is estimated to be less than 200 µm. In any case, it is clear that this
value could be generated at misalignments of either 150 or 250 µm, the l = 0
mode strongly dominating in the latter case.
Similarly, Figure 5.38 shoes the effect of an angular misalignment on the

bunching. The beams are simulated to cross each other at the z0 = 10 cm laser
waist position. The trend is akin to that of the parallel misalignment; a 2 mrad
crossing angle leads to the dominance of the l = 0 mode, and the measured 6.1
pJ CTR signal is retrieved for crossing angles of 1 and 5 µm.
With these points in mind, it is worth commenting that large errors of this
type were unlikely the cause of the large signal seen at only the γ = 25 energy
in the HELiX experiment. Firstly, the other points in the energy scan were in
216
agreement with expectations from simulations and theory. Secondly, due care was
taken to define both the e-beam and the laser transport through the undulator
axis. Two 1 mm diameter adjustable irises were positioned 50 cm apart, one in
front of the undulator, one 7 cm behind. Two counter-propagating alignment
lasers (one pointing upstream, the other downstream) were used to define the
electron beam transport axis on several of the beamline pop-ins, and the e-beam
was within 100 µm of these values during the experiment. The undulator was
likewise aligned, using custom machined cylindrical inserts with 1 mm diameter
on-axis holes both front and back, 30 cm apart. The CO2 laser pulse too was
carefully aligned through the system of irises with a Pyrocam. Altogether, the
parallel alignment of the beams to each other is estimated to be better than 200
µm, and the angular alignment better than 2 mrad. Nevertheless, these examples
demonstrate the sensitivity of the helical structure to the cylindrical symmetry
in the system, indicating that it is important to maintain good alignment of the
laser and e-beam throughout undulator transport. They further indicate a more
subtle point; that it is not advantageous to simply maximize the total CTR signal
by steering the e-beam or laser. Careful alignment must be maintained to create
and maintain the complex helical microbunching structure.
5.8 Conclusions
Results of the HELiX experiment provide compelling first evidence that heli-
cal microbunching of an electron beam was generated and measured in a second
harmonic interaction in a helical undulator. This provides an encouraging first
step for future successful realization of the proposed HGHMG scheme. While
bunching into other azimuthal modes evidently also occurs, it was both antici-
pated and in excellent quantitative agreement with simulations. Since the purity
217
of the desired azimuthal mode is expected to be significantly enhanced at higher
e-beam energies where transverse effects are less pronounced, this method of gen-
erating a helically pre-bunched beam therefore shows considerable promise for
the purposes of OAM mode amplification in short-wavelength high-gain FELs.
218
APPENDIX A
List of Variables
α Scaled mode expansion parameter
α0 Scaled mode spot size
αR Scaled radius of phasefront curvature
β, βz , β⊥ Scaled relativistic velocity
γ, (γR ) Scaled relativistic energy (resonant)
γz , (γzR ) Scaled relativistic longitudinal energy (resonant)
δn Density modulation
δv Longitudinal velocity modulation
δη, (δη) FWHM of Lorentzian energy distribution (scaled)
δk, (δk) Complex wavenumber (scaled)
δkr , (δk r ) Real part of complex wavenumber (scaled)
δki , (δk i ) Gain parameter (scaled)
0 Permittivity of free-space
x Geometric emittance
219
nx Normalized emittance
η Relative e-beam energy
ηγ Scaled RMS energy spread
ηd Diffraction parameter
η Scaled RMS emittance
θ Forward observation angle in spherical coordinates
θ0 , (θ) 1D detuning (scaled)
θp , (θp ) Longitudinal plasma wavenumber (scaled)
θq Mode q detuning
Λp,l Power fit function
λ, (λR ) Radiation wavelength (resonant)
λβ Betatron wavelength
λb Microbunching wavelength
λw Undulator period
µ0 Permeability of free-space
ρ Pierce parameter
ρe Charge density
σ Polarization
σγ RMS energy spread
220
σx , σz , σt RMS e-beam size
σx0 RMS divergence
στ RMS Laser pulse length
σθ Detuning RMS bandwidth
σω Relative RMS bandwidth
τz Torque
φ Azimuthal radiation angle
ϕ Azimuthal e-beam angle
ψ Ponderomotive phase
ω, (ωR ) Angular frequency (resonant)
ωp Angular plasma frequency
Ω Solid angle
Ab Transverse e-beam area
aq Density modulation mode amplitude coefficient
aSN
q Effective shot noise mode amplitude coefficient
avq Velocity modulation mode amplitude coefficient
B Magnetic field vector
B̃⊥q , (B̃⊥q ) Transverse magnetic field of waveguide eigenmode, (vector)
B̃⊥w , (B̃⊥w ) Transverse magnetic field of undulator, (vector)
221
b, (bl ) Bunching factor (l-mode)
bSN Equivalent shot noise bunching factor
c Speed of light
cq Waveguide eigenmode amplitude coefficient
−e Charge of electron
ê Unit vector
E Electric field vector
Ẽs Time-harmonic paraxial electric field vector
Ẽpm,q Ponderomotive electric field mode
Ẽ⊥q , (Ẽ ⊥q ) Transverse electric field of waveguide eigenmode, (vector)
E0 , (E0 ) Electric field amplitude (RMS)
E0 E-beam energy
F E-beam distribution function
F0 Unperturbed e-beam distribution function
F1 Perturbed e-beam distribution function
Fq,q0 E-beam/waveguide eigenmode overlap factor
f Transverse e-beam profile
f⊥ Normalized transverse e-beam profile
fη Energy spread function
222
fp Transverse momentum distribution function
fz Normalized longitudinal e-beam profile
gef f 3D supermode excitation efficiency
ĝq Mode polarization alignment factor
G Power gain
G3D 3D power gain
h Harmonic number
~ Planck’s constant
I Current
J Current density vector
Jrad Total angular momentum vector
j Total angular momentum density vector
k Free-space wavenumber
kβ Betatron wavenumber
kb Microbunching wavenumber
kSM Supermode wavenumber
kw Undulator wavenumber
kzq Longitudinal waveguide eigenmode wavenumber
K, (K) Undulator parameter (RMS)
223
Kq,q0 Dielectric waveguide mode coupling factor
l Azimuthal mode number
L Length of undulator
Lλ Coherence length
L1D 1D power gain length
LG , (LG,l ) 3D Power gain length (for mode l)
Lm Modulator length
Lrad Orbital angular momentum vector
m Mass of electron
Nc Number of electrons in a coherence length
Ne Number of electrons
Nw Number of undulator periods
nI Refractive index distribution of virtual dielectric
nI0 On-axis index of refraction of virtual dielectric
nef f Effective index of refraction of e-beam
n Beam density distribution
n0 Beam density from 1D model
P Total EM power
Pq Mode power
224
PSN Shot noise power
p Radial mode number
Q 1D e-beam/waveguide eigenmode coupling factor
Qq,q0 E-beam/waveguide eigenmode coupling factor
q Mode index
r Radial coordinate
r0 RMS radius of Gaussian e-beam
S1 Energy spread and detuning function
Srad Spin angular momentum vector
s Longitudinal coordinate in moving frame
sq Supermode amplitude coefficient
t Time
U Radiation field energy
uq , (uq ) Waveguide eigenmode, (vector)
v0 Nominal longitudinal e-beam velocity
ṽ⊥w Transverse velocity vector of e-beam due to undulator field
ṽ⊥q Transverse velocity vector of e-beam due to EM field
w Spot size of Gaussian laser beam
w0 Waist of Gaussian laser beam
225
x 3D spatial coordinate vector
x⊥ Transverse spatial coordinate vector
z Longitudinal cartesian coordinate
zR Rayleigh length
Zq Waveguide mode impedance
226
APPENDIX B
Coupling Coefficients and Supplementary

Calculations
The overall 3D coupling between the spatial overlap of the e-beam and the signal
field is given by the mode coupling coefficient[47]:
θp2 0
ZZ
∗
Qq,q0 = JJ (kzq0 + kw ) f (x⊥ )Ẽpm,q0 (x⊥ )ṽ⊥w · Ẽ ⊥q (x⊥ )d2 x⊥ (B.1)
8Pq
A generalized, explicit expression in any spatial expansion basis can be obtained

starting from the ponderomotive field
1 ∗ ∗
Ẽpm,q (x⊥ ) = [ṽ⊥q × B̃⊥w + ṽ⊥w × B̃⊥q ] · êz . (B.2)
2
From the relativistic Lorentz force equation we can write the transverse velocities
as[50],
e
ṽ⊥w = −i êz × B̃w
γmkw
e h i e
ṽ⊥q (x⊥ ) = −i Ẽ ⊥q (x⊥ ) + v0 êz × B̃⊥q (x⊥ ) = −i Ẽ ⊥q (x⊥ )
γm(ω − kzq v0 ) γmω
(B.3)
where in the last equality for ṽ⊥q it has been assumed that Ẽ ⊥q = −(ω/kzq )êz ×
B̃⊥q for TE modes, in keeping with the dominantly TEM mode expansion sets of
227
the present work. With the equations in (B.3) the expression for Ẽpm,q becomes
e(kzq + kw )
Ẽpm,q (x⊥ ) = i (êz × B̃w )∗ · Ẽ ⊥q (x⊥ ). (B.4)
2γmωkw
The complex phasor notation for the undulator field Bw =Re{B̃w e−ikw z } al-
lows a general description for any type of undulator polarization. Defining a
general transverse field polarization unit vector as êw = B̃w /|B̃w |, for a linear
undulator êw = êy (with vertical polarization), the field is given as
Bw = |B̃w |êy cos kw z. (B.5)
In a helical undulator the magnetic field is given by the unit vector êw = (êx ±
√
iêy )/ 2 where the (+) sign describes a field that evolves in a right-handed sense
moving along z, and (–) is left-handed. The field is then
|B̃w |
Bw = √ (êx cos kw z ± êy sin kw z). (B.6)
2
The undulator parameter is defined as K = e|B̃w |/(kw mc). In terms of the rms
√
magnetic field, which is given by B̄w = |B̃w |/ 2 for both planar and helical
undulators, the undulator parameter can be written simply as
√
K= 2K̄ (B.7)
where K̄ = eB̄w /(kw mc) is the rms undulator parameter.
In terms of K, Eq. (B.4) can be written as
kzq + kw K
Ẽpm,q (x⊥ ) = iE0 ĝq uq (x⊥ ). (B.8)
2k γ
228
Substitution of this expression and the mode power normalization from Eq. (2.9)
into Qq,q0 results in a general expression for the mode coupling coefficents:
2
+ kw )2

(kzq0 K
Qq,q0 = JJ θp2 ĝq∗ ĝq0 Fq,q0 (B.9)
8kzq γ
where
ĝq = (êz × ê∗w ) · êq (B.10)
is the polarization alignment factor, and the spatial overlap factor Fq,q0 is defined
as
∗
RR
f (x⊥ )Ẽ⊥q0 (x⊥ )Ẽ⊥q (x⊥ )d2 x⊥ huq |f (x⊥ )|uq0 i
Fq,q0 = = . (B.11)
huq |uq i
RR
2 2
|Ẽ⊥q (x⊥ )| d x⊥
The general expression for the mode coupling coefficient in Eq. (B.9) permits
analysis of mode coupling for radiation modes that vary transversely in their
complex field amplitude and in their polarization relative to the undulator in the
cold beam limit.
In order to obtain maximal coupling, the optical field must have finite spa-
tial overlap with the e-beam distribution and be “polarization matched” to the
direction of electron motion in the undulator. This is obtained when
êq = êz × êw (B.12)
which yields ĝq = 1 in Eq. (B.10) for any undulator polarization. In the linear
case, the optical field is linearly polarized, orthogonal to the polarization axis of
the undulator magnetic field. In a helical undulator, (B.12) describes an electric
field of the form
1
Ẽ⊥ ∝ Re{êq ei(kz−ωt) } = √ [êx sin(kz − ωt) ± êy cos(kz − ωt)] . (B.13)
2
229
For a right-handed undulator, we take the (+) sign. At a fixed point in time,
this defines an electric field vector that evolves in a left-handed sense along z,
opposite to the spatial field evolution of the undulator. Such a wave is called
left circularly polarized in the parlance of optics, but has a positive projection of
angular momentum onto the propagation axis and thus a positive helicity [120].
Therefore, maximal polarization coupling between the optical field and the e-
beam is obtained when the spatial field evolution has the opposite helical sense
as that of the undulator.
This seemingly contradictory result is understood in terms of a boosted frame

of reference. Electrons follow helical trajectories that correspond to the same
handedness as the magnetic field of the undulator (as do positrons, incidently,
though their orbits are π out of phase). For example, in the lab frame of a
right-handed undulator, electrons propagating along z trace out a right-handed
helix. If one were to boost to a moving frame in z that travels faster than the
electrons, the same electrons would appear to move in the negative z direction
from the perspective of the observer but in a left-handed helix since the direction
of rotation about the axis does not change. One cannot change the helicity of
the optical mode through a boost, since one cannot move to a frame that travels
faster than c. So the optical wave, which always travels faster than the beam, sees
a similar transformation of the e-beam’s helical trajectory, and therefore couples
best to the electron orbit that shares the same handedness as the optical E-field
polarization in the boosted frame.
230
APPENDIX C
Mode Expansion of Integro-Differential FEL

Equations
With the dielectric waveguide eigenmodes in Eq. (2.70), and assuming the un-
modulated beam distribution function can be written as F0 = n0 f (x⊥ )fη (η), the
modified integro-differential equation in (2.69) becomes
Z z
X d X 0
ikzq z
dz 0
2
e ∇⊥ +k 2 − kzq
2
+ 2ikzq cq (z)uq (x⊥ ) = iωµ0 ecn0 f (x⊥ ) eikzq z
q
dz 0 q
( " #)
2
eK e d
× JJ 2
+ 2 ∇2⊥ + k 2 − kzq
2
+ 2ikzq 0 cq (z 0 )uq (x⊥ )
4γ k dz
∞
Z
1 dfη i −θ−k+ k
2η (z 0 −z)
× dη e γz
E0 −∞ dη
(C.1)
The dielectric eigenmode relation in Eq. (2.6) is inserted to remove the trans-
verse Laplacian term and simplify using orthogonality of the basis functions. This
condenses the transverse spatial dependence of the fields and the beam into cou-
pling coefficients and converts the 3D equations into evolution equations for the
mode amplitudes along z. The longitudinal evolution of the mode amplitudes is
231
then (2.71)
θp2 γz2 z
Z
d Xh 0
cq (z) = −icq0 (z)Kq,q0 ei(kzq0 −kzq )z + dz 0 ei(kzq0 z −kzq z)
dz q0
k 0
( )
k2K 2 X kzq00 kzq0 d
× JJ 2
Fq,q0 − Fq,q00 Kq00 ,q0 + i Fq,q0 0 cq0 (z 0 )
8kzq γ q 00
kzq kzq dz
∞
Z
dfη i −θ−k+ k
2η (z 0 −z)
i
× dη e γz .
−∞ dη
(C.2)
p
The relativistic plasma wavenumber is θp = n0 e2 /0 mγγz2 v02 . Supermodes are
then found with the transformation
cq (z) = sq eiz(k−kzq +δk) (C.3)
and (C.1) becomes
X h
(δk − ∆kq )sq = − sq0 Kq,q0 + θp2 S1 (δk, θ)
q0
( )
k2K 2 X kzq00 kzq0 i
× JJ Fq,q0 − Fq,q00 Kq00 ,q0 − (δk − ∆kq )Fq,q0 ,
8kzq γ 2 q 00
kzq kzq
(C.4)
where
∞
γz2
Z
dfη /dη
S1 (δk, θ0 ) = dη . (C.5)
k −∞ δk − θ0 + γk2 η
z
The sum over the product of the elements of Fq,q00 and Kq00 ,q in the second
232
term in the z-integrand results from the integral
k 2 huq |f (x⊥ ) [nI (x⊥ )2 − 1] |uq0 i X k 2 huq |f (x⊥ )|uq00 ihuq00 |nI (x⊥ )2 − 1|uq0 i
=
2kzq huq |uq i q 00
2kzq huq |uq i2
X kzq00
= Fq,q00 Kq00 ,q0 .
q 00
kzq
(C.6)
P
In the second step the identity operator q 00 |uq00 ihuq00 | has been inserted with
the condition that the eigenmode basis forms a complete set.
In matrix form (C.4) becomes Eq. (2.72)

" #
h 1 ih i
I − θp2 M Iδk + K − ∆k + Q s = 0 . (C.7)
S1 (δk, θ)
233
APPENDIX D
Helical Undulator Focusing and Beam Transport
In a addition to the sinusoidal motion of the electrons through the periodic un-
dulator fields, there is additional motion of the particles due to the magnetic
fringing fields, or other non-ideal components of the magnetic fields.
One of the most significant contributions to the transport, provided the non-
ideal components are minimized, is the natural focusing of the e-beam due to
the growth of the transverse fields off-axis. In the region between the individual
magnetic dipoles that form the Bx field, for example, the strength of the Bx field
grows in the regions closer to the pole faces, particularly if the pole faces are flat
(as in the HELiX design) or bent towards each other. In the following section
a simple model of the helical undulator field is described which provides insight
into the transport characteristics of the beam.
234
D.0.1 Simple Helical Undulator Transport
In a helical undulator the simplest form of the contribution that leads to the
dominant focusing mechanism is modeled with the fields
Bx = B̄w cosh (kw x) cos (kw z)
By = B̄w cosh (kw y) sin (kw z) (D.1)
Bz = B̄w [sinh (kw y) cos (kw z) − sinh (kw x) sin (kw z)]
√
for a right-handed undulator. The rms magnetic field value B̄w = |B̃w |/ 2 is
used. Note that on-axis the longitudinal field vanishes and the fields become
exactly those of (B.6). The particle trajectories are given by
e 0
x00 = − [y Bz − By ]
pz
(D.2)
e
y 00 = [x0 Bz − Bx ]
pz
with x0 = dx/dz, and pz = γmv0 . The full equations are difficult to solve and
not particularly enlightening, so it is convenient to divide the trajectories into
two parts: The portion due to the fast wiggling motion (xw ) and the slower
component due to focusing (xf ),
x = xw + xf
(D.3)
y = yw + yf .
235
The fields are also expanded about the axis (x, y) = (0, 0) to give
(kw x)2

Bx ' B̄w cos (kw z) 1 + = Bx(0) + Bx(2)
2
(kw y)2

By ' B̄w sin (kw z) 1 + = By(0) + By(2) (D.4)
2
Bz ' B̄w kw [y cos (kw z) − x sin (kw z)] = Bz(1)
The wiggling components satisfy the first-order equations
e (0) K̄
x00w = B → xw (z) = − sin (kw z)
pz y γβz kw
(D.5)
e K̄
yw00 = − Bx(0) → yw (z) = cos (kw z)
pz γβz kw
with the rms undulator parameter K̄ = eB̄w /(kw mc). These equations describe
the sinusoidal trajectory through the undulator. Using them, we can derive the
relationship between the nominal total energy γ and the longitudinal energy
factor γz as mentioned in Chapter 2 with the effects of focusing excluded. The
total energy is γ 2 = (1 − βx2 − βy2 − βz2 )−1 where βx = βz x0w and βy = βz yw0
for constant longitudinal velocity. Defining γz2 = (1 − βz2 )−1 and using (D.5) we
obtain γ 2 = γz2 (1 + K̄ 2 ), qed.
With (D.2) the focusing trajectories are found from
e 0
x00f = − [(yw + yf0 )Bz(1) − By(2) ]
pz
(D.6)
e
yf00 = [(x0w + x0f )Bz(1) − Bx(2) ].
pz
The focusing trajectory is clearly a second order effect that results from the trans-
verse dependence of the fields. It is assumed to be slowly-varying compared to the
236
wiggling orbits such that x0f x0w . The focusing motion can also be considered
constant over an undulator period which permits several terms to be dropped
R λw
by averaging over a wiggling period via h.i = λ−1
w 0
(.)dz. This simplification
yields hBx i, hBy i = 0 such that the transverse fields do not contribute on average.
Focusing thus comes directly from the axial field and we obtain
2
K̄k
x00f =− √ w xf → xf (z) = xf1 cos (kf z) + xf2 sin (kf z)
2βz γ
(D.7)
2
K̄k
yf00 = − √ w yf → yf (z) = yf1 cos (kf z) + yf2 sin (kf z)
2βz γ
where
K̄kw
kf = √ (D.8)
2βz γ
is the natural undulator focusing wavenumber. The focusing wavelength is λf =
√
2π/kf = 2βz γλw /2π K̄. The initial conditions on the undulator entrance set the
matching constants xf1 , xf2 , yf1 and yf2 . We specify that, at the entrance to the
periodic undulator fields the phase space coordinates are x(0) = x0 , x0 (0) = x00 ,
y(0) = y0 , and y 0 (0) = y00 . The resulting trajectory through the undulator is,

K̄ 1 0 K̄
x(z) = − sin (kw z) + x0 cos (kf z) + x0 + sin (kf z)
γkw kf γ
(D.9)
y0

K̄ K̄
y(z) = cos (kw z) + y0 − cos (kf z) + 0 sin (kf z).
γkw γkw kf
Aside from the sinusoidal wiggle motion, the trajectories are the same as those
due to a linear focusing element of length Ld with focal length f = (kf2 Ld )−1 in
both dimensions.
With (D.9) we can compute the evolution of the first and second moments
of the e-beam distribution during transport through the undulator. For a beam
237
composed of discrete particles Ne the first moments refer to the displacement in
position and transverse velocity of the particle ensemble (considering only the x
direction since the y behavior follows trivially for a helical undulator):
Ne Ne
1 X 0 1 X
x̄ = xi , x̄ = x0 . (D.10)
Ne i=1 Ne i=1 i
These quantities are generally taken to be small enough so that the expansion
about the magnetic axis in (D.4) is satisfied for the entire beam, for the entire
undulator length. The second moments yield the rms (root mean squared) values
and correlation:
Ne Ne Ne
1 X 1 X 1 X
σx2 = (xi −x̄)2 , σx20 = (x0i −x̄0 )2 σxx0 = (xi −x̄)(x0i −x̄0 ).
Ne i=1 Ne i=1 Ne i=1
(D.11)
Combining equations (D.11) and (D.9) by summing over each particle coordinate
(xi (z), x0i (z)) we obtain an expression for the evolution of the beam size along z
in terms of the initial values,
2σxx0 (0) σ 20 (0)

σx2 (z) = σx2 (0) cos2 (kf z) + sin (kf z) cos (kf z) + x 2 sin2 (kf z). (D.12)
kf kf
The natural focusing of the helical undulator can be used to suppress the nat-
ural tendency of the e-beam to diverge. For the proper tuning, ie matching, the e-
beam size remains fixed throughout the undulator length. This condition is satis-
fied if σx2 (z) = σx2 (0) for all z. Equation (D.12) then gives (σx2 −σx20 /kf2 ) sin (kf z) =
2σxx0 cos (kf z)/kf which can only be satisfied if σxx0 = 0 which identifies an elec-
tron beam waist. In terms of the geometric emittance 2x = σx2 σx20 − σxx
2
0 the
238
matching conditions are thus
x
r p
σx = , σx0 = x kf . (D.13)
kf
For low-energy systems the natural undulator focusing can be sufficient to main-
tain a matched beam size. Otherwise external focusing is sometimes used.
239
APPENDIX E
Coherent Transition Radiation
Charged particles that cross a boundary separating media with differing electro-
magnetic properties emit transition radiation (TR). Calculating the emitted fields
can be difficult for the general case, but simplifications can be made for certain
conditions. First, we assume that the radiation is measured in the far-field, be-
yond the radiation formation length 2γ 2 λ. Second, at the frequencies of interest
(usually visible to THz), we can assume that the TR is emitted as a result of
the charged particles hitting a perfectly conducting foil such that the boundary
is one between a perfect vacuum and a perfect conductor. In this approximation
all of the radiation is reflected, and can be modeled from a kinematic standpoint
as a result of a sudden change in the particle’s velocity at the boundary interface.
E.1 Single particle emission
The simplest way to calculate the far-field TR emission spectrum for an electron
bunch is to find the single particle emission distribution and then add all the
individual contributions. This can be done either in a discrete fashion or by
integration, depending on the number of particles and the available computation
method.
In the single particle model, a single electron (or any negatively charged parti-
cle) is envisioned to be traveling at a constant velocity v0 = βc in the z-direction
240
x
foil
φ
e R P
x(t)
θ z
y
z0
Figure E.1: Schematic of particle motion and location of field measurement po-
sition of the CTR emitted at the foil.
with a time-dependent position vector
x(t) = x⊥ ê⊥ + z(t)êz . (E.1)
The particle strikes and stops instantly1 at the thin foil at z = z0 at time t = t0 .
The z-position as a function of time is then z(t) = βc(t − t0 )Θ(t0 − t) + z0 where
the unit step function Θ(t)=1 for t ≥ 0 and Θ(t)=0 for t < 0. The normalized
velocity is
1d
β(t) = x(t) = βΘ(t0 − t)êz . (E.2)
c dt
The radiation is measured in the far-field downstream of the foil at position P in

Fig. E.1 given by the vector R from the origin, in the unit direction n̂ = R/R =
sin θ cos φ êx + sin θ sin φ êy + cos θ êz . It is important to distinguish between
the coordinates that specify the position P where the radiation is measured:
xf = (R, θ, φ), from the coordinates that describe the position of the particle:
x = (x(t), z). The fields nominally depend on both, but the particle coordinates
1
Stopping instantaneously is not required for this model, and is certainly non-physical,
but the approximation makes the calculation simpler and does not change the effective result
provided that the stopping distance is much shorter than the radiation formation length 2γ 2 λ.
241
will be temporarily suppressed for brevity. The vector from the particle position
to P is R0 = R − x(t). In the far-field we assume |x| << R so we can make the
simplification |R0 | ' R − n̂ · x(t).
From [120], the radiation field from the charge is given by

" #
e n̂ × [(n̂ − β) × β̇]
E1 (xf , t) = − (E.3)
4π0 c (1 − n̂ · β)3 |R0 |
tr
where −e < 0 for an electron and the subscript tr indicates evaluation of the
quantities inside the brackets at the retarded time tr = t − |R0 |/c, which is the
time when the fields were emitted from the particle, prior to their being detected
at P. Since the particle stops abruptly it is convenient to work in the frequency
domain so we can write
Z ∞
1
Ẽ1 (xf , ω) = √ E1 (x, t)eiωt dt
2π −∞
∞ 0
n̂ × [n̂ × β] eiω(tr +|R |/c)

cn̂ · β
Z
e
= √ + iω(1 − n̂ · β) dtr .
4π0 c 2π −∞ |R0 | 1 − n̂ · β |R0 |
(E.4)
Integration by parts has been performed in the last step assuming the fields vanish
at infinity. From the definitions we find that n̂ · β = β cos θ Θ(t0 − t) and that
n̂ × [n̂ × β] = βΘ(t0 − t)a(θ, φ) where the polarization of the electric field is given
by the vector
a(θ, φ) = cos θ sin θ cos φ êx + cos θ sin θ sin φ êy − sin2 θ êz . (E.5)
In the far-field (n̂ · x/R 1) several terms can be neglected and integration over
the retarded time yields an expression for the electric field of a single electron
242
impacting the foil at time t0 :
eβ eiω(t0 +R/c) e−iωn̂·x(t0 )/c

Ẽ1 (xf , ω) = √ a(θ, φ). (E.6)
4π0 c 2π R 1 − β cos θ
If the foil is a perfect conductor, there is an image charge that also adds to the
far-field radiation and modifies the angular distribution. It has the opposite sign
as the particle, and travels in the -z direction also impacting the foil at z0 and
time t0 . The image charge moves longitudinally as z I (t) = βc(t0 − t)Θ(t0 − t) + z0
and the velocity is β I (t) = −βΘ(t0 − t)êz . Adding this contribution we obtain a
total TR field of
eβ eiω(t0 +R/c) e−iωn̂·x(t0 )/c

Ẽ(xf , ω) = √ a(θ, φ). (E.7)
2π0 c 2π R 1 − β 2 cos2 θ
The energy of the TR per unit time, per unit area is given by the Poynting vec-
tor which, with the magnetic radiation field B = n̂×E/c, gives S = (µ0 c)−1 |E|2 n̂.
The energy radiated into an angular area of R2 dΩ normal to the field propagation
vector n̂ is the time integral of the Poynting vector, which gives
Z ∞
1 dU
= S · n̂ dt
R2 dΩ −∞
(E.8)
Z ∞
2
= |Ẽ|2 dω
µ0 c 0
where we have used the Fourier transformed field expression in (E.4). It is com-
mon to integrate only the physically meaningful positive frequencies, hence the
factor of two that appears in the last expression because the field in (E.3) is real.
We can therefore write the energy per unit frequency per solid angle emitted from
243
a single charge hitting a perfect conductor as
d2 U 2R2 2
= |Ẽ|
dωdΩ µ0 c
(E.9)
e2 β 2 sin2 θ
= 3 .
4π 0 c (1 − β 2 cos2 θ)2
This expression is well-known and shows the characteristic flat-frequency response

and hollow axi-symmetric angular distribution. Integrating over the solid angle
gives [121],
e2 1 + β2 1 + β

dU
= 2 ln −1 . (E.10)
dω 4π 0 c 2β 1−β
E.2 Coherent emission from multiple particles
The total radiation field emitted from multiple particles Ne in an ensemble is

simply the sum of the fields from the individual particles that arrive at time
t0i with longitudinal velocity βi and at position xi (t0i ) = xi êx + yi êy + z0 êz .
All of the particles stop at the same longitudinal position z0 . The total Fourier
transformed field is then
Ne
X
ẼT (xf , ω) = Ẽ(xf , xi , ω)
i=1
(E.11)
Ne
iωR/c X
ea(θ, φ) e βi exp [−iω (n̂ · xi (t0 )/c + t0i )]
= .
(2π)3/2 0 c R i=1
1 − βi2 cos2 θ
The phase contribution from each particle due to its arrival time and transverse
position at the foil are therefore already contained in the field calculation. Assum-
ing all the particles have the same initial velocity, βi = β, in the continuous beam
limit the collection of discrete particles is described by a distribution function
244
f0 (x⊥ , t0 ) and the sum is converted to an integral over the beam’s transverse and
R
temporal distribution. The distribution is normalized to f0 (x⊥ , t0 )dx⊥ dt0 = 1.
The TR field for a continuous bunch is then
Z
ẼT (xf , ω) = f0 (x⊥ , t0 )Ẽ(xf , ω)dx⊥ dt0
eβa(θ, φ) eiω(R−z0 cos θ)/c

Z
= f0 (x⊥ , t0 )eiω(t0 −n̂·x⊥ /c) dx⊥ dt0 .
(2π)3/2 0 cR (1 − β 2 cos2 θ)
| {z }| √{z }
Ẽ0 (xf ,ω) F (ω)
(E.12)
In cylindrical e-beam coordinates the exponential is n̂ · x⊥ = r cos ϕ sin θ cos φ +
r sin ϕ sin θ sin φ where ϕ is the azimuthal cylindrical coordinate for the e-beam
distribution, and φ is the azimuthal spherical coordinate of the field distribution.
The first term Ẽ0 (xf , ω) is just the single particle spectral emission distribution.
p
The second term, F (ω), is the complex-valued square-root of the form factor,
sometimes called the “structure factor” which takes into account the individual
phase contributions from the particles in the bunch.
For the purposes of the helical beam structures described in this work, we
consider the beam distribution in Eq. (2.16). Since it is straightforward, we also
include a finite overall temporal distribution (which makes the FEL equations
more complicated but is simple in this case) as well as discrete harmonics of
the microbunching modulation. Evaluating at the foil position, the normalized
distribution is
" ( )#
X
f0 (x⊥ , t) = ft (t) f⊥ (x⊥ ) + Re δn(n) (x⊥ , z0 )einωb (z0 /v0 −t) (E.13)
h
where n is the harmonic number of the microbunching frequency ωb . Assuming
245
Gaussian unmodulated distributions both temporally and radially we obtain
1 −r2 /2σx2
f⊥ (x⊥ ) = e
2πσx2
(E.14)
1 −t2 /2σt2
ft (t) = p e .
2πσt2
E.2.1 CTR from Entire Beam
Coherent emission at wavelengths comparable to the bunch length (which is as-

sumed much larger than the microbunching wavelength) is dominated by the
unmodulated component of the distribution. The spectral energy distribution is
given by
d2 UC N 2 e2 sin2 θ 2 2 2 2 2
= e3 2 2 2
e−k (σz cos θ+σx sin θ) , (E.15)
dkdΩ 4π 0 (1 − β cos θ)
where σz = cσt and k = ω/c. Assuming a pancake beam distribution σx > σz ,

the spectrum is obtained by integration over the forward angle 0 ≤ θ ≤ π/2 and
is given in terms of the multivariate confluent hypergeometric functions Φ1 from
[122, 123]:
2 2
dUC Ne2 e2 e−k σx
=
dk 2π 2 0

1 3 2 2 2 2 1 3 5 2 2 2 2
Φ1 , 2, ; β , k (σx − σz ) − Φ1 , 2, ; β , k (σx − σz ) .
2 2 3 2 2
(E.16)
The total forward energy is,
 
2
 2 γσ
s
Ne2 e2 γ σx
z
−1
σx
2 γσz 
−1 σx
UC = √ 3/2 tanh 1− − .
 
2
8π π0 σx γσz

 γσ 2 γσz
−1

σx
z
−1 σx
(E.17)
246
In the limit of a strongly pancaked beam σx γσz ,
" 2 #
Ne2 e2 γ

1 γσz
UC ' √ 1− . (E.18)
16 π0 σx 2 σx
E.2.2 CTR from Microbunching
For frequencies near the beam’s characteristic harmonic microbunching frequen-

cies, ω ∼ nωb , the emission is dominated by the modulation in the beam. The
harmonic expansion of the modulated component from (2.24) is,
X (n)
δn(n) (x⊥ , z0 ) = ap,l (z0 )up,l (x⊥ ). (E.19)
p,l
Using the LG mode basis from Eq. (3.3) for up,l (x⊥ ) with expansion parameter
w0 set by the optical field that generates the modulation, the structure factor for
the emission due to the microbunching structure is
s
p w02 − (kw0 sin θ)2 X X (n) p!
F (ω) = 2 e 4 ap,l (z0 ) |l|+1 (p + |l|)!
(kw0 sin θ)|l|
2σx n p,l
2
nωb z0 σt2 π |l| (kw0 sin θ)2

2
× exp i − (ω − nωb ) + il φ − Lp .
v0 2 2 2
(E.20)
This full expression for the structure factor, though somewhat unwieldy, can be
used with to calculate the distribution of coherent transition radiation in the far
(n)
field if the modulation coefficients ap,l (z0 ) are known. For modulations at the
fundamental frequency h = 1, the mode amplitudes can be calculated directly
from the coupled FEL equations in (2.25) if the modulation is generated by the
FEL instability process.
To examine the CTR emission characteristics from helically microbunched
247
beams, the structure factor can be cast into a more transparent form by way of
the azimuthal bunching factor from Eq (3.12) generalized to harmonics:
s
w02 2|l|+1 (p + |l|)!

(n)
X (n) (|l| /2)! |l|
bl (z0 ) = ap,l (z0 ) 2 F1 −p; + 1; |l| + 1; 2
2σx2 p
(−1)p |l|! p! 2
(E.21)
where the total azimuthal bunching factor is the sum over the harmonics bl (z) =
P (n)
h=1 bl exp[inωb z/c]. Taking only the fundamental radial mode for illustration,
the structure factor at p = 0 can be written in terms of the helical microbunching

factor:
p −
(kw0 sin θ)2 X
(n) (kw0 sin θ)|l| i nωvb z0 − σ2t2 (ω−nωb )2 +il(φ− π2 )
F (ω) = e 4 bl (z0 ) |l|+1 e 0 (E.22)
n,l
2 (|l|/2)!
In tandem with Eq. E.12 this describes the far-field, frequency domain electric
field from a helically microbunched beam. The CTR emission spectrum given
by (E.20) is sharply peaked around the microbunching frequencies, is strongly
dependent on the forward angle θ for higher-order mode structures, and has a
helical azimuthal phasefront for l 6= 0 modes. In contrast to radiation emitted by
such a beam in an FEL, which would be a linear superposition of linear polar-
izations, the phasefront generated by the beam in the form of CTR is weighted
by the radial structure of the polarization. This adds an additional phase shift
across the phase front which which goes like cos(φ) in x and sin(φ) in y given by
Eq. (E.5).
Assuming that the transverse dependence of the modulation is the same as

√
that of the unmodulated component gives w0 = 2σx . The spectral energy
248
distribution at a single harmonic n and azimuthal mode l is then:
(n) |l| 2
d2 UC,l Ne2 e2 β 2 sin2 θ

(n) (kσ x sin θ) e−(kσx sin θ)2 −σt2 (ω−nωb )2
= b (z0 )
l |l|/2
dωdΩ 16π 3 0 c (1 − β 2 cos2 θ)2 2 (|l|/2)!
(E.23)
In calculating the total emission energy, the integral of Eq. (E.23) over frequency
is simplified in the limit where the emission spectrum approaches a delta function,
δ(ω − nωb ). The result is given in [55]:
(n) N 2 e2 2 γ 4 (|l| + 1)!

(n)
UC,l ' √e bl , (E.24)
16 π 3 0 σz σx nkb [(|l|/2)!]2 2|l|
where kb = ωb /c. At the fundamental spatial mode l = 0, the expression for

the emission energy of a purely longitudinally micro-bunched beam is obtained,
including the characteristic (γ/σx kb )4 scaling[124, 53]. This scaling applies in the
limit γ/σx kb 1, where the microbunching wavelength in the beam frame is
much smaller than the transverse beam size. Otherwise the scaling is somewhat
weaker than (γ/σx kb )4 and is found through the full integral solutions of Eq.
(E.23). For higher-order azimuthal modes that satisfy Equation (E.24), the total
energy in the forward direction is actually greater than for lower order modes
since the factor (|l| + 1)!/[(|l|/2)!]2 2|l| grows with |l|. This effect is the result of
the increased overlap of the form factor of these beam modes with the angular
distribution kernel of the single particle TR. Physically, the higher-order modes
naturally emit into larger opening angles, delivering a larger fraction of the total
power into the region outside the intrinsic axial null of the TR where the emission
is suppressed.
Equation (E.24) is a useful approximation for the total emission energy, in the
relevant limits. However, since the higher modes emit into the larger emission
angles, the small angle assumption used in obtaining (E.24) may not be strongly
249
satisfied, particularly for reduced transverse beams sizes, which also emit into
larger angles. In fact, when γ/σx kb ≥ 1, the small θ assumption begins to fail
dramatically where the full solution actually predicts a lower total emission en-
ergy for increasing l while the approximate solution in Eq. (E.24) erroneously
predicts the opposite. Mathematically the disagreement simply shows where the
assumptions made in deriving the simple scaling begin to break down. Physi-
cally the drop in emission energy can be understood from the perspective of the
virtual photon model of emission[125].The virtual photons are related directly
to the transverse fringing fields between microbunch packets that are ripped off
and emitted as CTR photons when the beam strikes the foil. The discrepancy is
therefore attributed, in part, to the increase in the helical pitch angle of the he-
lical density distribution. As the beam radius shrinks, the helical winding of the
density distribution becomes steeper. This changes the direction of the virtual
photon fields and eventually reduces the effective fringing fields that drive the
CTR emission. The pitch angle of the helical winding is greater for the higher l
modes, so the total emission energy is reduced for them first as the radius shrinks.
In the limit where the pitch angle approaches the π/2 maximum (a zero pitch an-
gle describes longitudinal bunching at l = 0), the beam is effectively de-bunched
and the coherent emission energy at hkb drops toward zero. There is thus a bal-
ance between the increase in emission energy for the higher-order modes which
emit off-axis, and the decrease in energy that results from the modified fringing
fields.
250
References
[1] John M. J. Madey. Stimulated emission of bremsstrahlung in a periodic

magnetic field. Journal of Applied Physics, 42(5):1906–1913, 1971.
[2] H. Motz. Applications of the radiation from fast electron beams. Journal
of Applied Physics, 22(5):527–535, 1951.
[3] Paul Emma et al. First lasing and operation of an angstrom-wavelength
free-electron laser. Nat Photon, 4(9):641–647, 2010.
[4] L. Allen, M. W. Beijersbergen, R. J. C. Spreeuw, and J. P. Woerdman.
Orbital angular momentum of light and the transformation of laguerre-
gaussian laser modes. Phys. Rev. A, 45(11):8185–8189, Jun 1992.
[5] N. B. Simpson, K. Dholakia, L. Allen, and M. J. Padgett. Mechanical
equivalence of spin and orbital angular momentum of light: an optical
spanner. Opt. Lett., 22(1):52–54, 1997.
[6] Gerard Nienhuis. Doppler effect induced by rotating lenses. Optics Com-
munications, 132(1-2):8 – 14, 1996.
[7] J. Courtial, K. Dholakia, D. A. Robertson, L. Allen, and M. J. Padgett.
Measurement of the rotational frequency shift imparted to a rotating light
beam possessing orbital angular momentum. Phys. Rev. Lett., 80(15):3217–
3219, Apr 1998.
[8] J. Courtial, D. A. Robertson, K. Dholakia, L. Allen, and M. J. Padgett.
Rotational frequency shift of a light beam. Phys. Rev. Lett., 81(22):4828–
4830, Nov 1998.
[9] L. Allen, M. Babiker, and W.L. Power. Azimuthal doppler shift in light
beams with orbital angular momentum. Optics Communications, 112(3-
4):141 – 144, 1994.
[10] Iwo Bialynicki-Birula and Zofia Bialynicka-Birula. Rotational frequency
shift. Phys. Rev. Lett., 78(13):2539–2542, Mar 1997.
[11] P. Török and P. Munro. The use of gauss-laguerre vector beams in sted
microscopy. Opt. Express, 12(15):3605–3617, 2004.
[12] Graham Gibson, Johannes Courtial, Miles Padgett, Mikhail Vasnetsov, Va-
leriy Pas’ko, Stephen Barnett, and Sonja Franke-Arnold. Free-space infor-
mation transfer using light beams carrying orbital angular momentum. Opt.
Express, 12(22):5448–5456, 2004.
251
[13] Eric Yao, Sonja Franke-Arnold, Johannes Courtial, Miles J. Padgett, and
Stephen M. Barnett. Observation of quantum entanglement using spatial
light modulators. Opt. Express, 14(26):13089–13094, 2006.
[14] M. F. Andersen, C. Ryu, Pierre Clade, Vasant Natarajan, A. Vaziri,
K. Helmerson, and W. D. Phillips. Quantized rotation of atoms from pho-
tons with orbital angular momentum. Phys. Rev. Lett., 97(17):170406,
2006.
[15] Adrian Alexandrescu, Dan Cojoc, and Enzo DiFabrizio. Mechanism of an-
gular momentum exchange between molecules and laguerre-gaussian beams.
Phys. Rev. Lett., 96(24):243001, 2006.
[16] L Allen, S. M Barnett, and Miles J Padgett. Optical angular momentum.
Institute of Physics Pub., 2003.
[17] C. Stamm et al. Femtosecond x-ray absorption spectroscopy of spin and
orbital angular momentum in photoexcited ni films during ultrafast demag-
netization. Phys. Rev. B, 81(10):104425, Mar 2010.
[18] Rasmita Raval. Chemistry: Mirrors in flatland. Nature, 425(6957):463–
464, 2003.
[19] Michel van Veenendaal and Ian McNulty. Prediction of strong dichro-
ism induced by x rays carrying orbital momentum. Phys. Rev. Lett.,
98(15):157401, Apr 2007.
[20] M. W. Beijersbergen, R. P. C. Coerwinkel, M. Kristensen, and J. P. Wo-
erdman. Helical-wavefront laser beams produced with a spiral phaseplate.
Optics Communications, 112(5-6):321 – 327, 1994.
[21] P. Coullet, L. Gil, and F. Rocca. Optical vortices. Optics Communications,
73(5):403 – 408, 1989.
[22] M. W. Beijersbergen, L. Allen, H. E. L. O Van der Veen, and J. P. Wo-
erdman. Astigmatic laser mode converters and transfer of orbital angular
momentum. Optics Communications, 96:123–132, 1993.
[23] VY Bazhenov, MV Vasnetsov, and MS Soskin. Laser-beams with screw
dislocations in their wave-fronts. JETP Letters, 52(8):429–431, OCT 25
1990.
[24] B. Thidé, H. Then, J. Sjöholm, K. Palmer, J. Bergman, T. D. Carozzi,
Ya. N. Istomin, N. H. Ibragimov, and R. Khamitova. Utilization of photon
orbital angular momentum in the low-frequency radio domain. Phys. Rev.
Lett., 99(8):087701, Aug 2007.
252
[25] Andrew G. Peele, Philip J. McMahon, David Paterson, Chanh Q. Tran,
Adrian P. Mancuso, Keith A. Nugent, Jason P. Hayes, Erol Harvey, Barry
Lai, and Ian McNulty. Observation of an x-ray vortex. Opt. Lett.,
27(20):1752–1754, 2002.
[26] Shigemi Sasaki and Ian McNulty. Proposal for generating brilliant
x-ray beams carrying orbital angular momentum. Phys. Rev. Lett.,
100(12):124801, 2008.
[27] J. H. Poynting. The wave motion of a revolving shaft, and a suggestion

as to the angular momentum in a beam of circularly polarised light. Pro-
ceedings of the Royal Society of London. Series A, Containing Papers of a
Mathematical and Physical Character, 82(557):pp. 560–567, 1909.
[28] Richard A. Beth. Mechanical detection and measurement of the angular

momentum of light. Phys. Rev., 50(2):115–125, Jul 1936.
[29] A.H.S. Holbourn. Angular momentum of circularly polarised light [4]. Na-
ture, 137(3453):31, 1936. cited By (since 1996) 6.
[30] S. Franke-Arnold, L. Allen, and M. Padgett. Advances in optical angular

momentum. Laser & Photonics Reviews, 2(4):299–313, 2008.
[31] S.J. van Enk and G. Nienhuis. Eigenfunction description of laser beams and
orbital angular momentum of light. Optics Communications, 94(1-3):147 –
158, 1992.
[32] S. J. van Enk and G. Nienhuis. Spin and orbital angular momentum of
photons. EPL (Europhysics Letters), 25(7):497, 1994.
[33] Stephen M. Barnett and L. Allen. Orbital angular momentum and non-
paraxial light beams. Optics Communications, 110(5-6):670 – 678, 1994.
[34] Chun-Fang Li. Spin and orbital angular momentum of a class of non-
paraxial light beams having a globally defined polarization. Phys. Rev. A,
80(6):063814, Dec 2009.
[35] A. T. O’Neil, I. MacVicar, L. Allen, and M. J. Padgett. Intrinsic and

extrinsic nature of the orbital angular momentum of a light beam. Phys.
Rev. Lett., 88(5):053601, Jan 2002.
[36] Michael V. Berry. Paraxial beams of spinning light. In Marat S. Soskin,

editor, International Conference on Singular Optics, volume 3487, pages
6–11. SPIE, 1998.
253
[37] Kosta Ladavac and David Grier. Microoptomechanical pumps assembled
and driven by holographic optical vortex arrays. Opt. Express, 12(6):1144–
1149, 2004.
[38] N. M. Kroll, P. L. Morton, and M. N. Rosenbluth. Free-electron lasers

with variable parameter wigglers. IEEE Journal of Quantum Electronics,
17:1436–1468, August 1981.
[39] G. T. Moore. The high-gain regime of the free electron laser. Nuclear
Instruments and Methods in Physics Research A, 239:19–28, August 1985.
[40] E. T. Scharlemann, A. M. Sessler, and J. S. Wurtele. Optical guiding in a

free-electron laser. Phys. Rev. Lett., 54(17):1925–1928, Apr 1985.
[41] M. Xie and D. A. G. Deacon. Theoretical study of FEL active guiding

in the small signal regime. Nuclear Instruments and Methods in Physics
Research A, 250:426–431, September 1986.
[42] M. Xie. Exact and variational solutions of 3D eigenmodes in high gain

FELs. Nuclear Instruments and Methods in Physics Research A, 445:59–
66, May 2000.
[43] Juhao Wu and Li Hua Yu. Eigenmodes and mode competition in a high-
gain free-electron laser including alternating-gradient focusing. Nuclear
Instruments and Methods in Physics Research Section A, 475(1-3):79–85,
2001.
[44] E. L. Saldin, E. A. Schneidmiller, and M. V. Yurkov. On a linear theory of

an FEL amplifier with an axisymmetric electron beam. Optics Communi-
cations, 97:272–290, March 1993.
[45] Y. Pinhasi and A. Gover. Three-dimensional coupled-mode theory of free-

electron lasers in the collective regime. Phys. Rev. E, 51(3):2472–2479, Mar
1995.
[46] P. Sprangle, A. Ting, and C. M. Tang. Radiation focusing and guiding

with application to the free electron laser. Phys. Rev. Lett., 59(2):202–205,
Jul 1987.
[47] Erik Hemsing, Avraham Gover, and James Rosenzweig. Virtual dielectric
waveguide mode description of a high-gain free-electron laser. i. theory.
Physical Review A, 77(6):063830, 2008.
[48] D. Marcuse. Theory of dielectric optical waveguides. New York, Academic

Press, Inc., 1974. 267 p., 1974.
254
[49] Ming Xie. Grand initial value problem of high gain free electron lasers.
Nuclear Instruments and Methods in Physics Research Section A: Acceler-
ators, Spectrometers, Detectors and Associated Equipment, 475(1-3):51 –
58, 2001.
[50] B. Z. Steinberg, A. Gover, and S. Ruschin. Three-dimensional theory of

free-electron lasers in the collective regime. Phys. Rev. A, 36(1):147–163,
Jul 1987.
[51] Gianluca Geloni, Evgeni Saldin, Evgeni Schneidmiller, and Mikhail Yurkov.
Theory of space-charge waves on gradient-profile relativistic electron beam:
an analysis in propagating eigenmodes. NUCL.INSTRUM.METH.A,
554:20, 2005.
[52] A. H. Lumpkin, R. Dejus, W. J. Berg, M. Borland, Y. C. Chae, E. Moog,

N. S. Sereno, and B. X. Yang. First observation of z-dependent electron-
beam microbunching using coherent transition radiation. Phys. Rev. Lett.,
86(1):79–82, Jan 2001.
[53] A. Tremaine, J. B. Rosenzweig, S. Anderson, P. Frigola, M. Hogan,

A. Murokh, C. Pellegrini, D. C. Nguyen, and R. L. Sheffield. Observation
of self-amplified spontaneous-emission-induced electron-beam microbunch-
ing using coherent transition radiation. Phys. Rev. Lett., 81(26):5816–5819,
Dec 1998.
[54] A. Tremaine, X. J. Wang, M. Babzien, I. Ben-Zvi, M. Cornacchia, H. D.

Nuhn, R. Malone, A. Murokh, C. Pellegrini, S. Reiche, J. Rosenzweig,
and V. Yakimenko. Experimental characterization of nonlinear harmonic
radiation from a visible self-amplified spontaneous emission free-electron
laser at saturation. Physical Review Letters, 88(20), 2002.
[55] E. Hemsing and J. B. Rosenzweig. Coherent transition radiation from

a helically microbunched electron beam. Journal of Applied Physics,
105(9):093101, 2009.
[56] A. Gover and E. Dyunin. Collective-interaction control and reduction of

optical frequency shot noise in charged-particle beams. Phys. Rev. Lett.,
102(15):154801, Apr 2009.
[57] Kwang-Je Kim. Three-dimensional analysis of coherent amplification and

self-amplified spontaneous emission in free-electron lasers. Phys. Rev. Lett.,
57(15):1871–1874, Oct 1986.
255
[58] Robert L. Gluckstern, Samuel Krinsky, and Hiromi Okamoto. Analysis of
the saturation of a high-gain free-electron laser. Phys. Rev. E, 47(6):4412–
4429, Jun 1993.
[59] J. B. Murphy and C. Pellegrini. Introduction to the physics of the free

electron laser. In W. Colson, C. Pellegrini, and A. Renieri, editors, Laser
Handbook, volume 6, chapter 5. North Holland, Amsterdam, 1990.
[60] R. Bonifacio, C. Pellegrini, and L. M. Narducci. Collective instabilities and

high-gain regime in a free electron laser. Optics Communications, 50:373–
378, July 1984.
[61] A. Gover and P. Sprangle. A unified theory of magnetic bremsstrahlung,

electrostatic bremsstrahlung, Compton-Raman scattering, and Cerenkov-
Smith-Purcell free-electron lasers. IEEE Journal of Quantum Electronics,
17:1196–1215, July 1981.
[62] E. L Saldin, E. A Schneidmiller, and M. V Yurkov. The physics of free

electron lasers. Springer, Berlin, 2000.
[63] Erik Hemsing, Agostino Marinelli, Sven Reiche, and James Rosenzweig.
Longitudinal dispersion of orbital angular momentum modes in high-gain
free-electron lasers. Phys. Rev. ST Accel. Beams, 11(7):070704, Jul 2008.
[64] S. Krinsky and L. H. Yu. Output power in guided modes for amplified
spontaneous emission in a single-pass free-electron laser. Phys. Rev. A,
35(8):3406–3423, Apr 1987.
[65] L. H. Yu, S. Krinsky, and R. L. Gluckstern. Calculation of universal scaling

function for free-electron-laser gain. Phys. Rev. Lett., 64(25):3011–3014,
Jun 1990.
[66] Kwang-Je Kim. Fel gain taking into account diffraction and electron beam
emittance; generalized madey’s theorem. Nuclear Instruments and Methods
in Physics Research Section A: Accelerators, Spectrometers, Detectors and
Associated Equipment, 318(1-3):489 – 494, 1992.
[67] Zhirong Huang. Review of x-ray free-electron laser theory. Phys. Rev. ST
Accel. Beams, 10(3):26, Mar 2007.
[68] E Saldin, E Schneidmiller, and M Yurkov. Output power and degree of

transverse coherence of x-ray free electron lasers. Optics Communications,
281(18):4727–4734, Sep 2008.
256
[69] S. Reiche. Numerical studies for a single pass high gain free-electron laser.
PhD thesis, University of Hamburg, 1999.
[70] E.L. Saldin, E.A. Schneidmiller, and M.V. Yurkov. Coherence properties
of the radiation from x-ray free electron laser. Optics Communications,
281(5):1179 – 1188, 2008.
[71] O. Georg. Use of the orthogonal system of Laguerre-Gaussian functions

in the theory of circularly symmetric optical waveguides . Applied Optics,
21(1):141–145, Jan 1982.
[72] L. Yu, W. Huang, M. Huang, Z. Zhu, X. Zeng, and W. Ji. The Laguerre-
Gaussian series representation of two-dimensional fractional Fourier trans-
form . Journal of Physics A Mathematical General, 31:9353–9357, Novem-
ber 1998.
[73] Amnon Yariv. Optical Electronics in Modern Communications (Oxford Se-

ries in Electrical and Computer Engineering). Oxford University Press,
USA, March 1997.
[74] T. A. Nieminen, T. Asavei, V. L. Y. Loke, N. R. Heckenberg, and

H. Rubinsztein-Dunlop. Symmetry and the generation and measurement of
optical torque. Journal of Quantitative Spectroscopy & Radiative Transfer,
110:1472–1482, September 2009.
[75] A.E. Siegman. Lasers. University Science Books, 1986., 1986.
[76] Erik Hemsing, Avraham Gover, and James Rosenzweig. Virtual dielectric
waveguide mode description of a high-gain free-electron laser. ii. modeling
and numerical simulations. Physical Review A, 77(6):063831, 2008.
[77] A. Murokh, R. Agustsson, M. Babzien, I. Ben-Zvi, L. Bertolini, K. van

Bibber, R. Carr, M. Cornacchia, P. Frigola, J. Hill, E. Johnson, L. Klaisner,
G. Le Sage, M. Libkind, R. Malone, H-D. Nuhn, C. Pellegrini, S. Reiche,
G. Rakowsky, J. Rosenzweig, R. Ruland, J. Skaritka, A. Toor, A. Tremaine,
X. Wang, and V. Yakimenko. Properties of the ultrashort gain length, self-
amplified spontaneous emission free-electron laser in the linear regime and
saturation. Phys. Rev. E, 67(6):066501, Jun 2003.
[78] G. Andonian, A. Murokh, J. B. Rosenzweig, R. Agustsson, M. Babzien,

I. Ben-Zvi, P. Frigola, J. Y. Huang, L. Palumbo, C. Pellegrini, S. Reiche,
G. Travish, C. Vicario, and V. Yakimenko. Observation of anomalously
large spectral bandwidth in a high-gain self-amplified spontaneous emission
free-electron laser. Physical Review Letters, 95(5):054801, 2005.
257
[79] S. Reiche. GENESIS 1.3: a fully 3D time-dependent FEL simulation code.
Nuclear Instruments and Methods in Physics Research A, 429:243–248,
June 1999.
[80] E. Hemsing, G. Andonian, J. Rosenzweig, and A. Gover. A description of

guided FEL radiation with dielectric waveguide eigenmodes. Proceedings
of FEL 2007 Conference, pages 65–68, 2007.
[81] X. J. Wang, T. Watanabe, Y. Shen, R. K. Li, J. B. Murphy, T. Tsang,

and H. P. Freund. Efficiency enhancement using electron energy detuning
in a laser seeded free electron laser amplifier. Applied Physics Letters,
91(18):181115, 2007.
[82] A. Marinelli, E. Hemsing, and J. B. Rosenzweig. Three dimensional analysis

of longitudinal plasma oscillations in a thermal relativistic electron beam.
Submitted to Phys Rev A, 2011.
[83] D. et al. Alesini. Status of the sparc project. Nuclear Instruments and
Methods in Physics Research Section A: Accelerators, Spectrometers, De-
tectors and Associated Equipment, 528(1-2):586–590, 2004.
[84] M. Xie. Design optimization for an x-ray free electron laser driven by slac
linac. In Proceedings of the 1995 Particle Accelerator Conference, 1996.
[85] G. Marcus, E. Hemsing, and J. B. Rosenzweig. Gain length fitting formula

for space-charge dominated free-electron lasers. Submitted to Phys Rev
ST-AB, 2011.
[86] M. Xie. Transverse coherence of self-amplified spontaneous emission. Nu-

clear Instruments and Methods in Physics Research A, 445:67–71, May
2000.
[87] B. W. J. McNeil, G. R. M. Robb, M. W. Poole, and N. R. Thomp-

son. Harmonic lasing in a free-electron-laser amplifier. Phys. Rev. Lett.,
96(8):084801, 2006.
[88] E. Hemsing, P. Musumeci, S. Reiche, R. Tikhoplav, A. Marinelli, J. B.

Rosenzweig, and A. Gover. Helical electron-beam microbunching by
harmonic coupling in a helical undulator. Physical Review Letters,
102(17):174801, 2009.
[89] N.A. Vinokurov. Proceedings of the 10th international conference on parti-

cle accelerators. Proceedings of the 10th International Conference on Par-
ticle Accelerators, Serpukhov, 2:454, 1977.
258
[90] L. Giannessi, P. Musumeci, and M. Quattromini. Tredi: fully 3d beam
dynamics simulation of rf guns, bendings and fels. Nucl. Instrum. Methods
Phys. Res., Sect. A, 436(3):443–444, 1999.
[91] W. Colson. The nonlinear wave equation for higher harmonics in free-
electron lasers. IEEE J. Quant. Elect., 17(8):1417–1427, Aug 1981.
[92] Gianluca Geloni, Evgeni Saldin, Evgeni Schneidmiller, and Mikhail Yurkov.
Theory of nonlinear harmonic generation in free-electron lasers with heli-
cal wigglers. Nucl. Instrum. Methods Phys. Res., Sect. A, 581(3):856–865,
2007.
[93] E. Allaria, F. Curbis, M. Coreno, M. Danailov, B. Diviacco, C. Spezzani,

M. Trovó, and G. DeNinno. Experimental characterization of nonlinear
harmonic generation in planar and helical undulators. Phys. Rev. Lett.,
100(17):174801, 2008.
[94] E. Hemsing, A. Marinelli, and J. B. Rosenzweig. Generating optical orbital

angular momentum in a high-gain free-electron laser at the first harmonic.
Phys. Rev. Lett., 106(16):164803, Apr 2011.
[95] J. Feldhaus et al. Possible application of x-ray optical elements for reducing
the spectral bandwidth of an x-ray sase fel. Optics Communications, 140(4-
6):341 – 352, 1997.
[96] Yuantao Ding, Zhirong Huang, and Ronald D. Ruth. Two-bunch self-
seeding for narrow-bandwidth hard x-ray free-electron lasers. Phys. Rev.
ST Accel. Beams, 13(6):060703, Jun 2010.
[97] Gianluca Geloni, Vitali Kocharyan, and Evgeni Saldin. A simple method
for controlling the line width of SASE X- ray FELs. arXiv: 1004.4067
[physics.acc-ph], 2010.
[98] P. Musumeci. Acceleration Of Electrons By Inverse Free Electron Laser

Interaction. PhD thesis, University of California Los Angeles, 2004.
[99] R. J. England. Longitudinal Shaping of Relativistic Bunches of Electrons

Generated by an RF Photoinjector. PhD thesis, University of California
Los Angeles, 2007.
[100] A. Cook. Generation of Narrow-Band Terahertz Coherent Cherenkov Ra-

diation in a Dielectric Wakefield Structure. PhD thesis, University of Cal-
ifornia Los Angeles, 2009.
259
[101] D. Haberberger, S. Tochitsky, and C. Joshi. Fifteen terawatt picosecond
co2 laser system. Opt. Express, 18(17):17865–17875, Aug 2010.
[102] S. Ya. Tochitsky, O. B. Williams, P. Musumeci, C. Sung, D. J. Haber-
berger, A. M. Cook, J. B. Rosenzweig, and C. Joshi. Efficient harmonic
microbunching in a 7th-order inverse-free-electron laser interaction. Phys.
Rev. ST Accel. Beams, 12(5):050703, May 2009.
[103] S. Ya. Tochitsky, R. Narang, C. Filip, B. Blue, C. E. Clayton, K. A. Marsh,
and C. Joshi. Amplification of two-wavelength co2 laser pulses to terawatt
level. Proceedings of LASERS99, Quebec, Canada, page 265, 1999.
[104] L. Young and J. Billen. Parmela. Technical report, Los Alamos National
Laboratory, 1996.
[105] P. Elleaume, O. Chubar, and J. Chavanne. Computing 3D Magnetic Fields
from Insertion Devices. APS Meeting Abstracts, pages 9–+, May 1997.
[106] A. Knyazik, R. Tikhoplav, and J. B. Rosenzweig. Status of ucla helical
permanent-magnet undulator. Proceedings of the 23rd Particle Accelerator
Conference, 4 8 May 2009 Vancouver, British Columbia, Canada, 22:2441–
2443, 2009.
[107] W. B. Colson, G. Dattoli, and F. Ciocci. Angular-gain spectrum of free-
electron lasers. Phys. Rev. A, 31(2):828–842, Feb 1985.
[108] Ming Xie. New mechanisms of interaction for even harmonic generation
in free electron lasers. Nucl. Instrum. Methods Phys. Res., Sect. A, 483(1-
2):527–530, 2002.
[109] G. Andonian, E. Hemsing, D. Xiang, P. Musumeci, A. Murokh, S. Tochit-
sky, and J. B. Rosenzweig. Longitudinal profile diagnostic scheme with
sub-femtosecond resolution for high-brightness beams. Submitted to Phys.
Rev. ST-AB.
[110] Y. Liu, X. J. Wang, D. B. Cline, M. Babzien, J. M. Fang, J. Gallardo,
K. Kusche, I. Pogorelsky, J. Skaritka, and A. van Steenbergen. Experimen-
tal observation of femtosecond electron beam microbunching by inverse
free-electron-laser acceleration. Phys. Rev. Lett., 80(20):4418–4421, May
1998.
[111] A. H. Lumpkin, R. Dejus, W. J. Berg, M. Borland, Y. C. Chae, E. Moog,
N. S. Sereno, and B. X. Yang. First observation of z-dependent electron-
beam microbunching using coherent transition radiation. Phys. Rev. Lett.,
86(1):79–82, Jan 2001.
260
[112] A. H. Lumpkin, N. D. Arnold, W. J. Berg, M. Borland, U. Happek, J. W.
Lewellen, and N. S. Sereno. Development of a coherent transition radiation-
based bunch length monitor with application to the aps rf thermionic gun
beam optimization. Nuclear Instruments and Methods in Physics Research
Section A: Accelerators, Spectrometers, Detectors and Associated Equip-
ment, 475(1-3):476–480, 2001.
[113] Mitsumi Nakamura, Makoto Takanaka, Shuichi Okuda, Takahiro Kozawa,

Ryukou Kato, Toshiharu Takahashi, and Soon-Kwon Nam. Evaluation of
electron bunch shapes using the spectra of the coherent radiation. Nu-
clear Instruments and Methods in Physics Research Section A: Acceler-
ators, Spectrometers, Detectors and Associated Equipment, 475(1-3):487–
491, 2001.
[114] D. Schiller, S. Reiche, and M. Ruelas. Amplification of two-wavelength co2

laser pulses to terawatt level. Proceedings of PAC07, Albuquerque, NM,
USA, page 3612, 2007.
[115] M. Padgett, J. Arlt, N. Simpson, and L. Allen. An experiment to observe

the intensity and phase structure of laguerre–gaussian laser modes. Amer-
ican Journal of Physics, 64(1):77–82, 1996.
[116] Guang-Hoon Kim, Jin-Ho Jeon, Kwang-Hoon Ko, Hee-Jong Moon, Jai-
Hyung Lee, and Joon-Sung Chang. Optical vortices produced with a non-
spiral phase plate. Appl. Opt., 36(33):8614–8621, Nov 1997.
[117] B. C. Platt and R. Shack. History and principles of Shack-Hartmann wave-

front sensing. Journal of refractive surgery (Thorofare, N.J. : 1995), 17(5),
2001.
[118] Jonathan Leach, Miles J. Padgett, Stephen M. Barnett, Sonja Franke-

Arnold, and Johannes Courtial. Measuring the orbital angular momentum
of a single photon. Phys. Rev. Lett., 88(25):257901, Jun 2002.
[119] J. M. Hickmann, E. J. S. Fonseca, W. C. Soares, and S. Chávez-Cerda.

Unveiling a truncated optical lattice associated with a triangular aperture
using light’s orbital angular momentum. Phys. Rev. Lett., 105(5):053904,
Jul 2010.
[120] J. D. Jackson. Classical electrodynamics. J.Wiley and Sons, New York, 3

edition, 1999.
[121] V. L. Ginzburg. Transition radiation and transition scattering. Physica

Scripta Volume T, 2:182–+, June 1982.
261
[122] I. S. Gradshteyn and I. M. Ryzhik. Table of integrals, series and products.
Academic Press, sixth edition, 2000.
[123] G. Andonian, E. Hemsing, A. Murokh, M. Dunning, G. Marcus, J. Rosen-

zweig, O. Williams, and V. Yakimenko. Transverse Beam Size Effects
on Longitudinal Profile Reconstruction. arXiv:1002.1992 [physics.acc-ph],
2010.
[124] J. Rosenzweig, G. Travish, and A. Tremaine. Coherent transition radiation

diagnosis of electron beam microbunching. Nuclear Instruments and Meth-
ods in Physics Research Section A: Accelerators, Spectrometers, Detectors
and Associated Equipment, 365(1):255–259, 1995.
[125] A. Tremaine. Coherent Radiation Diagnosis of Self Amplified Spontaneous

Emission Free Electron Laser-Derived Electron Beam Microbunching. PhD
thesis, University of California Los Angeles, 1999.
262

Untitled

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Untitled

Hochgeladen von

Copyright:

Verfügbare Formate

University of California

Generation and Amplification of Coherent

A dissertation submitted in partial satisfaction

Erik Willard Hemsing

Jianwei (John) Miao

James B. Rosenzweig, Committee Chair

University of California, Los Angeles

1.1 Outline of Dissertation . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 Principles of FEL operation . . . . . . . . . . . . . . . . . . . . . 7

1.3 Principles of OAM light . . . . . . . . . . . . . . . . . . . . . . . 17

1.3.1 Laguerre-Gaussian Paraxial Modes . . . . . . . . . . . . . 23

1.3.2 Understanding optical OAM . . . . . . . . . . . . . . . . . 28

2 Dielectric Eigenmode Expansion of High-Gain 3D FEL Equa-

2.1 Outline of Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.2 Dielectric Waveguide Expansion Modes in the Presence of a Local

2.3 Cold Electron Beam Fluid Description . . . . . . . . . . . . . . . 38

2.4 Coupled Excitation Equations . . . . . . . . . . . . . . . . . . . . 43

2.5 Bunching Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.6 Initial Values for Modal Amplitudes due to Shot Noise . . . . . . 50

2.7 Supermode Matrix Solutions . . . . . . . . . . . . . . . . . . . . . 55

2.7.1 Single Transverse Mode . . . . . . . . . . . . . . . . . . . 58

2.8 Uncorrelated Energy Spread Effects . . . . . . . . . . . . . . . . . 62

2.8.1 Integro-Differential Energy Spread Equations . . . . . . . . 63

2.8.2 Limitations on Energy Spread . . . . . . . . . . . . . . . . 68

2.9.1 Integro-Differential FEL Emittance Equations . . . . . . . 73

2.10 Recovering the 1D regime . . . . . . . . . . . . . . . . . . . . . . 79

3 Description of FEL Interaction in terms of Eigenmodes of Orbital

3.1 Guided Laguerre-Gaussian Modes . . . . . . . . . . . . . . . . . . 86

3.2 Dielectric Mode Coupling Factor . . . . . . . . . . . . . . . . . . . 89

3.3 Scaled Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 89

3.4 Bunching Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

3.5 Negligible Emittance . . . . . . . . . . . . . . . . . . . . . . . . . 93

3.5.1 Relevant equations and coupling factor . . . . . . . . . . . 94

3.5.2 Modeling of FEL with η , ηγ → 0 . . . . . . . . . . . . . . 94

3.5.3 OAM mode amplification in η , ηγ → 0 system . . . . . . . 100

3.5.4 Longitudinal Dispersion . . . . . . . . . . . . . . . . . . . 112

3.5.5 Single Mode Limit . . . . . . . . . . . . . . . . . . . . . . 119

3.6 Emittance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

3.6.1 Single mode limit . . . . . . . . . . . . . . . . . . . . . . . 131

3.6.2 Fitting Formula for 3D Gain Length with Emittance . . . 133

3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

4 Harmonic Interaction in a Helical Undulator . . . . . . . . . . . 140

4.1 Harmonic Interaction . . . . . . . . . . . . . . . . . . . . . . . . . 141

4.1.1 Density Modulation at Harmonics in the Modal Expansion 142

4.1.2 Emittance and Energy Spread Effects . . . . . . . . . . . . 152

4.2 High-Gain High Mode Generation (HGHMG) Scheme . . . . . . . 159

4.2.1 Unwanted mode coupling through misalignment . . . . . . 162

4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

5 Helical Microbunching Experiment at the Neptune Laboratory 166

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

5.2 Design of the Experiment . . . . . . . . . . . . . . . . . . . . . . 166

5.3 Neptune Laboratory . . . . . . . . . . . . . . . . . . . . . . . . . 167

5.3.1 CO2 Laser . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

5.3.2 Laser pulse measurements . . . . . . . . . . . . . . . . . . 172

5.3.3 Electron Beamline . . . . . . . . . . . . . . . . . . . . . . 174

5.3.4 Helical Undulator . . . . . . . . . . . . . . . . . . . . . . . 179

5.4 Analytic and numerical modeling of helical microbunching . . . . 187

5.4.1 Excitation of adjacent modes . . . . . . . . . . . . . . . . 192

5.5 Coherent Transition Radiation from helical beam . . . . . . . . . 196

5.5.1 CTR from ideal beam . . . . . . . . . . . . . . . . . . . . 197

5.5.2 Helical Phase Detection . . . . . . . . . . . . . . . . . . . 201

5.6 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . 209

5.7 Discussion: Evidence for helical microbunching . . . . . . . . . . . 213

A List of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

3.5.2 Modeling of FEL with η , ηγ → 0 . . . . . . . . . . . . . . 94

3.5.3 OAM mode amplification in η , ηγ → 0 system . . . . . . . 100